Exploring therapeutic targets for hepatocellular carcinoma through druggable genes

Abstract

Hepatocellular carcinoma (HCC) ranks as the third leading cause of cancer-related mortality worldwide. Mendelian randomization (MR) has emerged as a powerful approach for therapeutic target identification in HCC. In this study, we applied two-sample MR to assess the causal effects of multiple genes on HCC risk. Replication analyses integrated summary-data-based MR, Bayesian co-localization, and protein–protein interaction networks with druggability assessment to prioritize targets. Single-cell RNA sequencing validated cell-type-specific expression patterns in T cells, endothelial cells, and hepatocytes. MR-, summary-data-based MR-, and co-localization-based approaches identified target genes associated with HCC risk, comprising 3 first-level targets (HLA-DPA1, MBTPS1, and TIMP3), 2 second-level targets (TNXB and HSD17B11), and also 4 tertiary targets (HLA-DPB1, PLD2, KLHL8, and TGFBR1). All in all, this research identifies several potential therapeutic targets relevant to the risk of HCC and provides new insights into the identification of therapeutic agents for the treatment of HCC.

1. Introduction

Primary hepatic carcinoma, the 6th most common malignant tumor and the 3rd leading cause of cancer-related deaths globally, is increasing in its rate of incidence and mortality worldwide.^[1] According to histological features, liver cancer can be classified into 2 main types: hepatocellular carcinoma (HCC) and cholangiocarcinoma. HCC is the most common type and originates from hepatocytes in the liver.^[1] As medical advancements continue, HCC remains a significant threat to public health. Despite advances in understanding its pathogenesis, HCC remains a critical health threat. Projections indicate ≈ 1 million annual HCC diagnoses globally by 2025,^[2] driven by aging populations, environmental exposures, and epidemics of chronic liver diseases (e.g., hepatitis, non-alcoholic fatty liver disease). This trajectory imposes substantial pressure on healthcare systems worldwide.

Current HCC management includes surgical resection, thermal ablation, systemic therapies, and pharmacotherapy.^[3] However, diagnosis typically occurs at advanced stages due to resource limitations and screening barriers, severely constraining therapeutic options. Surgical interventions are only viable for early-stage HCC, yet even after curative resection, 5-year recurrence rates exceed 70%, causing substantial patient distress. First-line systemic therapies (e.g., sorafenib) frequently cause burdensome adverse effects ranging from nausea/vomiting to severe complications, including dermatologic toxicity and secondary malignancies.^[4] Moreover, ablation techniques and conventional chemotherapy demonstrate limited efficacy in improving long-term prognosis. These constraints – coupled with surgical contraindications and acquired drug resistance – create significant clinical dilemmas in treatment selection.^[5] This therapeutic landscape necessitates continued discovery of novel therapeutic targets to enhance treatment efficacy and patient survival.

Genome-wide association studies (GWAS) and whole-exome sequencing studies have developed rapidly over recent years, offering decision-making assistance in the identification of drug targets and indications.^[6] GWAS has identified disease-associated single-nucleotide polymorphisms (SNPs), which can help scientists establish the association between genetic variations and specific diseases. These SNPs can be used to assist in the identification and validation of drug targets.^[7] Thus, GWAS analysis alone does not completely recognize the disease-causing genes. Mendelian randomization (MR), a statistical method of genetics for assessing causality between exposures and outcomes, greatly minimizes the interference of confounding factors such as the environment and reverse causality, as genetic variants follow a random assignment principle at conception and cannot change their genotypes in the development of disease.^[8] Cis-expression quantitative trait loci (cis-eQTL) are considered proxies for gene expression levels, and the efficacy of disease drugs is predicted by analyzing eQTL and disease GWAS datasets to mimic randomized controlled trials.^[9]

In this study, we first employed two-sample MR to evaluate causal relationships between blood cis-eQTLs and HCC risk. Second, we applied summary-data-based MR (SMR) with heterogeneity in dependent instruments (HEIDI) testing to validate gene-HCC associations and enhance robustness.^[10] Subsequently, Bayesian co-localization analyses identified shared causal variants between candidate genes and HCC.^[11] Using single-cell expression type analysis to identify cell types associated with HCC risk. Finally, protein–protein interaction (PPI) network construction and computational druggability assessment prioritized potential therapeutic targets.

2. Materials and methods

2.1. Data sources

Exposure statistics were obtained from the eQTLGen consortium (https://eqtlgen.org/). The eQTLGen database uses approximately 31,684 human blood samples of European ancestry and contains information on 16,987 genes.^[12] The GWAS summary data for HCC were obtained as published by Sakaue et al.^[13] The data consisted of 379 case counts, 475,259 controls (https://www.ebi.ac.uk/gwas, GWAS ID: GCST90018858), a combined total of 475,638 sample sizes, and 24,194,938 SNPs.

2.2. Selection of cis-eQTLs associated with druggable genes

The eQTLGen database was acquired to get 16,897 genes (FDR < 0.05) cis-eQTLs data and allele frequency information, and after removing the null and gene data on the sex chromosomes, the final number of druggable genes obtained was 4302. For these 4302 gene sets of data, firstly, we chose the distance from the gene transcription start site ± 100 kb, and also to avoid the effect of linkage disequilibrium, the SNPs were clustered in the range of r² < 0.1, kb = 10,000 using the 1000G genome program.^[14] Second, to prevent interference from confounders, we ran searches in the PhenoScanner database (https://www.ebi.ac.uk/gwas/) and censored SNPs that were associated with confounders and liver cancer.^[15] Third, when the SNP is not found in the end, we delete it and remove those palindromic SNPs with allele frequency. Lastly, to prevent the effect of bias of weak instrumental variables, the F-value statistic was calculated by the formula β²/SE², and when the F-value was <10, it was excluded.^[16]

2.3. MR analysis

In this study, causal estimates were derived using the Wald ratio for exposures instrumented by a single SNP. For exposures with ≥ 2 SNPs, inverse-variance weighted (IVW) regression served as the primary method. Sensitivity analyses employed MR-Egger, weighted median, maximum likelihood, and MR Robust Adjusted Profile Score (MR-RAPS) to confirm IVW robustness.^[9,17] We apply Cochran Q and MR-Egger intercept experiments to examine heterogeneity and horizontal pleiotropy between all instrumental variables.^[18,19] MR-Steiger analysis evaluated potential reverse causality.^[20] All associations underwent Bonferroni correction, with corrected P < .05 considered statistically significant.

2.4. SMR analysis

SMR analysis evaluates associations between gene expression and complex traits by integrating summary-level GWAS and eQTL data. Result validation employs the HEIDI method to distinguish pleiotropy from shared causal mechanisms.^[10] In the current study, we used SMR analysis and HEIDI testing to further verify the MR results. When the Bonferroni-corrected P-value was <.05 and the HEIDI test was >.05, we concluded that the gene could be considered a candidate drug target for HCC.

2.5. Co-localization analysis

For genes analyzed by MR and subjected to validation analysis, we conducted a co-localization analysis. This analysis combines information from multiple SNPs or other genetic variants to determine whether they are simultaneously associated with a specific phenotype. We employed default a priori probabilities (p1 = 1E−4, p2 = 1E−4, and p12 = 1E−5), representing the likelihood that SNPs in selected regions would be associated with significant gene expression, HCC risk, or both. The posterior probabilities were validated against 5 hypotheses: PP.H0, indicating that SNPs are not associated with either trait; PP.H1, suggesting that SNPs are correlated with gene expression but not with HCC risk; PP.H2, indicating that SNPs are associated with the risk of HCC occurrence but not with gene expression; PP.H3, suggesting that SNPs are associated with both gene expression and HCC risk but are driven by different SNPs; and PP.H4, indicating that SNPs are associated with both HCC risk and gene expression, specific to the common genetically causal variants.^[11] Due to the limited efficacy of the co-localization analysis, we set the significance threshold for PP.H4 to 0.75.

2.6. PPI and druggability evaluation

To assess intracellular protein interactions, the GeneMANIA database (https://genemania.org/) was used to construct the protein interaction network.^[21] DGIdb (https://www.dgidb.org/) and DrugBank (https://go.drugbank.com/) bring together a large amount of detailed information about drugs and data on drug–gene interactions,^[22,23] providing important data and functionality in drug target prediction, individualized therapy, and clinical applications. The potential of identified target genes as therapeutic HCC was further determined by using DGIdb and DrugBank to find these links between proteins and drugs.

2.7. Single-cell type expression analysis

To characterize the cellular expression patterns of identified target genes, we analyzed single-cell RNA sequencing (scRNA-seq) data from human HCC tissues. Validation was performed using the HCC scRNA-seq dataset published by Lu et al, which comprises samples from 10 primary tumors, 2 portal vein tumor thromboses, 1 metastatic lymph node, and 8 normal liver tissues.^[24] The raw dataset contained 25,479 genes across 71,915 cells. We applied standard computational workflows including quality control, normalization, dimensionality reduction, and clustering. Subsequent identification of differentially expressed genes within cell clusters, followed by cluster annotation, provided deeper insight into the biological functions governed by the target genes.

2.8. Methods and software used for analysis

In this study, MR analyses were performed using the TwoSampleMR R package (v0.5.6, University of Bristol, Bristol, UK).^[25] SMR and HEIDI analyses were conducted using SMR software (University of Bristol, Bristol, UK).^[10] Co-localization analysis was implemented with the Coloc R package (v5.1.0, University of Bristol, Bristol, UK).^[11] PPI networks were constructed via the GeneMANIA database.^[21] Single-cell RNA sequencing data were processed using the Seurat R package (v4.4.0, New York Genome Center (NYGC), New York).^[26,27] All R-based analyses (TwoSampleMR, Coloc, and Seurat) utilized R version 4.2.3. Cell cluster annotation integrated automated labeling (SingleR package, The Weizmann Institute of Science, Rehovot, Israel), marker gene verification (CellMarker database: http://xteam.xbio.top/CellMarker/), and manual curation based on literature signatures.^[28,29]

3. Results

3.1. MR analysis of druggable genes identified 9 HCC-associated genes

In the current study, after a series of rigorous screening and analysis as described above, we eventually obtained 2631 pharmacological gene MR results, of which 9 genes passed the Bonferroni correction (P < 1.9E−05 [0.05/2631]) test (Figs. 1 and 2). Specifically, genes HLA-DPB1 (OR = 0.811; 95% confidence interval [CI], 0.756, 0.870; P = 4.14E−09), HSD17B11 (OR = 0.818; 95% CI, 0.759, 0.881; P = 1.38E−07), TNXB (OR = 0.777; 95% CI, 0.703, 0.858; P = 6.33E−07), TIMP3 (OR = 0.338; 95% CI, 0.212, 0.539; P = 5.04E−06), KLHL8 (OR = 0.457; 95% CI, 0.325, 0.641; P = 5.83E−06), and gene HLA-DPA1 (OR = 0.872; 95% CI, 0.819, 0.928; P = 1.78E−05) had a protective effect on HCC risk. Genetically predicted genes PLD2 (OR = 1.379; 95% CI, 1.208, 1.575; P = 2.05E−06), MBTPS1 (OR = 1.178; 95% CI, 1.098, 1.264; P = 4.58E−06), and TGFBR1 (OR = 2.160; 95% CI, 1.538, 3.034; P = 8.84E−06) associated with increased risk of HCC (Figs. 1 and 2). In addition, 4 complementary methods, MR-Egger, Maximum likelihood, MR-RAPS, and Weighted median in Table S1, Supplemental Digital Content, https://links.lww.com/MD/Q52 all support the results of the present work.

Figure 1.

MR results of druggable genes Manhattan map. MR = Mendelian randomization.

Figure 2.

Forest plot of MR results for significant genes. CI = confidence interval, MR = Mendelian randomization, SNP = single nucleotide polymorphism.

3.2. Verification analysis

Heterogeneity analysis revealed no outliers for the genes HLA-DPB1, HSD17B1, TNXB, TIMP3, KLHL8, PLD2, MBTPS1, and TGFBR1. However, slight heterogeneity was observed for HLA-DPA1 (P = .029; Table S1, Supplemental Digital Content, https://links.lww.com/MD/Q52). Multivariable MR-Egger intercept analysis indicated no significant pleiotropy (Table S1, Supplemental Digital Content, https://links.lww.com/MD/Q52). Furthermore, MR-Steiger directionality tests demonstrated statistically significant results (all P < .05), suggesting minimal risk of reverse causality (Table S1, Supplemental Digital Content, https://links.lww.com/MD/Q52). Although HLA-DPA1 exhibited slight heterogeneity, the absence of pleiotropy (MR-Egger) and evidence for correct causal direction (MR-Steiger), combined with the use of an IVW random-effects model inherently robust to heterogeneity, support the reliability of the finding that HLA-DPA1 is negatively associated with HCC risk. Overall, 2631 genes showed associations with HCC risk. Complete MR results and sensitivity analyses are detailed in Tables S2 to S5, Supplemental Digital Content, https://links.lww.com/MD/Q52.

3.3. SMR analysis validated 5 disease-causing genes

For the 9 significant gene targets analyzed by MR, we did an SMR analysis. Taking into account that multiple hypothesis tests were carried out to avoid the probability of type I error, we used the rather conservative Bonferroni correction with a significance threshold of P < .006 (0.05/9). For this SMR analysis, we detected that 5 of these 9 genes remained to be statistically significant, which consisted of the genes HLA-DPA1 (P_SMR = 4.64E−08; P_HEIDI = 0.091), MBTPS1 (P_SMR = 1.05E−04; P_HEIDI = 0.523), TNXB (P_SMR = 4.50E−04; P_HEIDI = 0.058), HSD17B11 (P_SMR = 1.10E−03; P_HEIDI = 0.115), and gene TIMP3 (P_SMR = 1.26E−03; P_HEIDI = 0.310), also they passed the HEIDI test and no significant heterogeneity is observed (Fig. 3, Table S6, Supplemental Digital Content, https://links.lww.com/MD/Q52). In addition, the genes PLD2 (P_SMR = 0.012; P_HEIDI = 0.118), HLA-DPB1 (P_SMR = 0.024; P_HEIDI = 0.323), TGFBR1 (P_SMR = 0.069; P_HEIDI = 0.603), and KLHL8 (P_SMR = 0.838; P_HEIDI = 0.628) didn’t pass the Bonferroni correction based on SMR analysis (Fig. 3, Table S6, Supplemental Digital Content, https://links.lww.com/MD/Q52). From these results, we concluded that genes HLA-DPA1, MBTPS1, TNXB, HSD17B11, and TIMP3 can serve as candidate drug targets for the treatment of HCC.

Figure 3.

Results of SMR analysis of significant genes. CI = confidence interval, HCC = hepatocellular carcinoma, HEIDI = heterogeneity in dependent instruments, SMR = summary-data-based MR.

3.4. Co-localization analysis supports the causal linkage of 3 genes to HCC

To assess whether SNPs associated with HCC risk and eQTLs share causal genetic variation, we performed co-localization analyses within ± 100 kb of each gene. For the 5 genes (HLA-DPA1, MBTPS1, TNXB, HSD17B11, and TIMP3) showing significance in prior SMR analyses, strong co-localization evidence was observed in blood tissue for HLA-DPA1 (PP.H4 = 0.99), MBTPS1 (PP.H4 = 0.95), and TIMP3 (PP.H4 = 0.77; Table 1; Fig. 4). This indicates a high probability of shared causal variants influencing both gene expression and HCC risk for these genes. Conversely, TNXB (PP.H4 = 0.74) and HSD17B11 (PP.H4 = 0.19) did not meet co-localization significance thresholds (Table 1; Fig. 4). Based on these combined results, HLA-DPA1, MBTPS1, and TIMP3 – having met all analytical criteria – are designated as primary therapeutic targets for HCC. Genes TNXB and HSD17B11, which showed SMR significance but failed co-localization, are classified as secondary targets. The genes HLA-DPB1, PLD2, KLHL8, and TGFBR1, which lacked SMR significance, are considered tertiary biomarkers for HCC.

Table 1.

Five gene co-localization results through SMR testing.

Gene	No SNP	PP.H0	PP.H1	PP.H2	PP.H3	PP.H4
HLA-DPA1	373	0	0	0	0.01	0.99
MBTPS1	1249	0	0.04	0	0.01	0.95
TIMP3	802	0	0.18	0	0.05	0.77
TNXB	194	0	0.11	0	0.15	0.74
HSD17B11	676	0	0.07	0	0.74	0.19

PP.H0–PP.H4 represents the posterior probabilities of different hypotheses.

PP.H4 > 0.75 represents a strong colocalization between gene expression and HCC risk.

SMR = summary data-based Mendelian randomization, SNP = single nucleotide polymorphism.

Figure 4.

Regional association map for co-localization analysis between genes and liver cancer risk. (A) Graph of the region where the genes HLA-DPA1 cis-eQTL and HCC co-localize at ± 100 kb. (B) Graph of the region where the genes MBTPS1 cis-eQTL and HCC co-localize at ± 100 kb. (C) Graph of the region where the genes TIMP3 cis-eQTL and HCC co-localize at ± 100 kb. (D) Graph of the region where the genes TNXB cis-eQTL and HCC co-localize at ± 100 kb. (E) Graph of the region where the genes HSD17B11 cis-eQTL and HCC co-localize at ± 100 kb. HCC = hepatocellular carcinoma.

3.5. PPI and drug product candidate forecasting

The PPI interaction network analysis revealed that these 15 drug targets and 20 potentially interacting genes had a total of 160 interactions with each other, including co-expression (54.43%) and physical interaction (36.36%; Fig. 5). Functional analysis of this network described the roles of drug targets and associated genes and their functions, suggesting that target genes are associated with epithelial and endothelial cell proliferation.

Figure 5.

Interaction map of PPI networks constructed using GeneMANIA. MHC = human major histocompatibility complex.

In the druggability assessment, drugs targeting TIMP3 can be combined with other drugs to cure diabetic nephropathy (Pimagedine). Drugs targeting PLD2 are anti-inflammatory pain relievers related to aspirin and are used to relieve pain caused by arthritis, mouth ulcers, and herpes orofacialis (Choline salicylate). Drugs that target TNFSF8 are approved for the treatment of Hodgkin lymphoma and systemic mesenchymal lymphoma (Brentuximab vedotin). Drugs targeting TGFBR1 have been investigated for the care and basic science of rheumatoid arthritis and immune thrombocytopenic purpura, and have recently been used as an alternative therapy for the management of acute respiratory distress syndrome in patients with severe COVID-19 (Fostamatinib).

3.6. Single-cell type expression analysis

We additionally analyzed complementary scRNA-seq data (GSE149614) from the Gene Expression Omnibus database for cell-type-specific expression profiling. Following clustering, 33 distinct cell populations were identified: 10 T/NK cell clusters, 8 myeloid cell clusters, 7 hepatobiliary epithelial cell clusters, 4 B cell clusters, 2 fibroblast clusters, and 2 endothelial cell clusters (Fig. 6). In HCC tissues, MBTPS1, KLHL8, and TGFBR1 exhibited specific enrichment in T cells, whereas HLA-DPA1 and HLA-DPB1 were enriched in both T cells and macrophages. TIMP3 showed endothelial-specific enrichment, while HSD17B11 was detected in T cells, macrophages, hepatocytes, and biliary epithelial cells (Fig. 7).

Figure 6.

Cell profile of single-cell transcriptome in HCC. HCC = hepatocellular carcinoma.

Figure 7.

Expression of druggable genes in each cluster of the HCC single-cell transcriptome. (A) Single-cell transcriptome expression profile of druggable genes in each cluster. (B) Bubble plot of druggable gene expression in each cluster of HCC single-cell transcriptome. HCC = hepatocellular carcinoma.

4. Discussion

For this study, we conducted a thorough investigation of 4302 genes and HCC risk. HCC risk was explored by several MR methods (Wald ratio/IVW, MR-Egger, Weighted median, Maximum likelihood, and MR-RAPS), including diverse sensitivity analyses (Cochran Q heterogeneity test, MR-Egger intercept pleiotropy test, MR-Steiger direction test) as well as the use of PhenoScanner to remove confounding SNPs to explore genes associated with HCC risk. Various tests, such as replicate MR, SMR, and HEIDI tests, were used to further validate the reliability of the results. Bayesian co-localization strengthened the evidence of genes with HCC risk. Based on MR, SMR, and co-localization methods, we identified target genes involved in HCC risk, consisting of 3 primary targets (HLA-DPA1, MBTPS1, and TIMP3), 2 secondary targets (TNXB and HSD17B11), and 4 tertiary targets (HLA-DPB1, PLD2, KLHL8, and TGFBR1).

MBTPS1 stands for a protease of the human genome that plays an essential role in cholesterol metabolism and synthesis, protein secretion, and processing.^[30] MBTPS1 cleaves and activates several transcription factors, including sterol regulatory element-binding protein, activating transcription factor 6, and brain-derived neurotrophic factor,^[31] which are associated with lipid metabolism, tumor cell proliferation, and apoptosis in cancer cells.^[30,32] Previous studies have implicated MBTPS1 as a possible therapeutic target in colon cancer by blocking fatty acid synthesis through sterol regulatory element-binding protein knockdown, inhibiting the growth of colon cancer xenografts, and regulating gene expression to control colon cancer proliferation.^[32,33] Previous experimental studies have also shown that inhibition of MBTPS1 expression reduces the activity of tumor cells such as melanoma and glioblastoma.^[34,35] Our study found that gene MBTPS1 expression may be associated with an increased risk of liver cancer. Although there is no direct evidence to confirm this yet, more epidemiological and experimental studies are needed to further confirm our findings. TIMP3 plays a crucial role in regulating the metabolism of the extracellular matrix by inhibiting various metalloproteinases, including matrix metalloproteinases and angiotensin-converting enzymes. These metalloproteinases have significant involvement in physiological and pathological processes such as tissue repair, inflammatory responses, and tumor development.^[36] Mouse model experiments confirm that TIMP3 reduces hepatic lipid and cholesterol accumulation and prevents HCC.^[37] Experimental evidence has confirmed that miR-452-5p downregulates the oncogene TIMP3, promoting apoptosis, invasion, and migration of HCC cells.^[38] Similarly, our study extends the range of evidence and confirms that the oncogene TIMP3 inhibits the growth of HCC cells.

The human major histocompatibility complex (MHC) region is a cluster of genes located on human chromosome 6 that encode numerous proteins associated with the immune system, and the human leukocyte antigen (HLA) genes are part of the MHC region that encode subunits of the MHC molecule.^[39] HLA has been shown to play a role in autoimmune, oncologic, and infectious diseases.^[40–42] In particular, HLA-DPB1 and HLA-DPA1 alleles in the MHC region are believed to play an active role in persistent chronic hepatitis B virus (HBV) infection and spontaneous clearance.^[43] In a genome-wide expression-associated SNP study involving 651 liver samples, it was demonstrated that SNPs rs3077 and rs9277535 were associated with mRNA expression of HLA-DPA1 and HLA-DPB1. These variants were found to be independent factors closely linked to chronic HBV infection.^[40] Another study by Yu et al focused on HLA family genes in liver cell carcinoma tissues and discovered that 34 variants across 8 HLA families, including HLA-DPA1, HLA-DPB1, and others, were strongly associated with Hepatitis B-associated HCC.^[41] Furthermore, analysis of data from 212 patients with HBV-associated HCC revealed that HLA-DPA1 and HLA-F could potentially serve as prognostic biomarkers for HCC.^[44] Single-cell transcriptome analysis of NK cells from liver, spleen, and peripheral blood of HCC patients and healthy liver tissues identified the gene HLA-DPA1, which was significantly overexpressed in tumor tissues and blood, as a promising noninvasive prognostic biomarker for HCC patients.^[45] Similar to these studies, our findings point to a negative correlation between gene HLA-DPA1 expression and the hazard of HCC.

While there has been no direct experimental evidence to support the involvement of these genes in HCC, there are corresponding studies indicating their association with cancer cells. For example, it has been demonstrated that obesity-related protein (fat mass and obesity-associated gene) regulates the target gene HSD17B11 to promote lipid formation in esophageal cancer cells, contributing to the progression and development of cancer cells.^[46] Xue et al analyzed and experimentally validated multi-omics data to determine that TNXB promotes the growth and spread of gastric adenocarcinoma cells and is related to poor prognosis of gastric adenocarcinoma.^[47] TNFRSF25 has also been identified as a potential target for prostate cancer treatment.^[48] Each of these studies indicates that the target genes we screened can act as oncogenes, but whether they are applicable in HCC requires further epidemiologic and experimental investigations to further explore our discovery.

The merit of the current study is that genes linked to HCC risk were successfully selected by the two-sample MR method, which greatly avoided the interference of confounding bias. Meanwhile, the experimental conclusions were further validated using replicate MR, SMR, and co-localization, which greatly enhanced the stability of the study results and reduced the possibility of false positives. The PPI protein regulatory network provided insights into the pathogenic mechanism, and the assessment of druggability offers potential for the treatment of HCC. This study has several limitations. First, as the genetic data were derived exclusively from European populations, validation in diverse ethnic groups is required. Second, despite incorporating multiple druggable targets, the absence of suitable SNPs for certain genes constrained generalizability. Third, although pleiotropy was rigorously mitigated through sensitivity analyses, residual confounding cannot be fully excluded. Specifically: genetic pleiotropy may persist (e.g., MBTPS1-associated SNPs could influence HCC risk via cholesterol metabolism pathways independent of gene expression); unmeasured gene-environment interactions (such as alcohol intake or obesity modulating HSD17B11/TIMP3 effects); and potential false co-localization signals from long-range linkage disequilibrium in complex loci like the MHC region harboring HLA-DPA1/DPB1. Future studies integrating liver-specific multi-omics data with prospective multi-ethnic cohorts will help resolve these confounders.

5. Conclusion

Collectively, this study employed MR, SMR, and co-localization analyses to identify genes associated with HCC risk. We established a hierarchical classification: 3 primary therapeutic targets (HLA-DPA1, MBTPS1, TIMP3), 2 secondary targets (TNXB, HSD17B11), and 4 tertiary biomarkers (HLA-DPB1, PLD2, KLHL8, TGFBR1). These findings highlight promising candidates for HCC biomarker screening and potential therapeutic development. Future work should prioritize functional validation and clinical trials to confirm these results.

Acknowledgments

We thank the eQTLGen Alliance and other researchers and participants for providing publicly available data for this analysis.

Author contributions

Conceptualization: Zhen Ding.

Data curation: Chao Tang.

Formal analysis: Chao Tang, Shuangjing Zhu.

Funding acquisition: Zhen Ding.

Investigation: Zhen Ding.

Project administration: Zhen Ding.

Resources: Chao Tang.

Software: Chao Tang, Shuangjing Zhu.

Supervision: Zhen Ding.

Validation: Chao Tang, Shuangjing Zhu.

Visualization: Chao Tang, Shuangjing Zhu.

Writing – original draft: Chao Tang.

Writing – review & editing: Chao Tang.

Supplementary Material

medi-104-e44627-s001.xlsx ^{(1.1MB, xlsx)}