Multi-omics insights into Palmitoylation-mediated brain network remodeling and glioblastoma risk

Abstract

Glioblastoma (GBM) remains a lethal brain tumor with limited therapeutic progress, necessitating novel insights into its molecular drivers. This study investigates the causal role of protein palmitoylation-a post-translational modification regulating tumor metabolism and microenvironment-in GBM pathogenesis through neuroimaging-mediated pathways. Leveraging two-sample Mendelian randomization (MR) and multi-omics data from FinnGen (406 GBM cases/378,749 controls) and UK Biobank (3,935 brain imaging phenotypes), we identified protective effects of palmitoylation-related genes (ZDHHC3: OR = 0.17, 95%CI = 0.0602–0.4824; ZDHHC6: OR = 0.41, 0.2330–0.7267; ZDHHC13: OR = 0.64, 0.4745–0.8618; PPT2: OR = 0.59, 0.3568–0.9843) against GBM risk. Neuroimaging mediation analysis revealed that 15–23% of this protection operates through structural and functional brain remodeling, including reduced parahippocampal volume (mediated 15.96% for ZDHHC13), diminished default mode network connectivity ICA100 edge 295, and ICVF Body of corpus callosum(15.29% mediation for ZDHHC6). Bidirectional regulation was observed in motor pathways, where palmitoylation genes simultaneously suppressed corticospinal tract integrity (FA) and enhanced cortical plasticity. Sensitivity analyses confirmed robustness (Cochran’s Q p > 0.05; MR-PRESSO global test p > 0.05), with Steiger filtering excluding reverse causation. Our findings suggest that palmitoylation may play a key modulatory role in GBM risk through influencing brain network dynamics, highlighting neuroimaging features (e.g., DMN connectivity, parahippocampal atrophy) as potentially informative for GBM risk and warranting further investigation for their role in early detection. These results also raise the possibility of palmitoylation-targeted therapies aimed at disrupting tumor-microenvironment crosstalk, suggesting a direction for future therapeutic exploration.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-03369-3.

Glioblastoma multiforme (GBM) is the most aggressive primary malignant tumor of the adult central nervous system, with an annual incidence of about 3.2 cases per 100,000 people [1]. Over the past three decades, the survival rate of GBM patients has only seen limited improvement [2]. Patients typically undergo standard treatments, including surgical resection, radiotherapy, and chemotherapy; however, the median overall survival remains around 15 months [3]. Although some progress has been made in treatments like immunotherapy and targeted therapy, the poor prognosis of GBM is closely linked to its high heterogeneity, diffuse infiltrative growth, and treatment resistance due to glioma stem cells (GSCs) [4]. Recently, multi-omics studies have shown that protein post-translational modifications (PTMs) are crucial in the onset and progression of GBM. One notable modification is palmitoylation, which covalently attaches palmitate to cysteine residues. This modification has gained attention for its dual role in regulating tumor metabolism and remodeling the tumor microenvironment [5]. For example, palmitoylation promotes glycolysis in tumor cells by stabilizing the membrane localization of GLUT1 [6], The DHHC protein family, through palmitoylation modification of multiple key substrates, widely regulates core signaling pathways and biological processes closely related to the malignant progression of GBM, including immune evasion, stemness maintenance, drug resistance, proliferation, migration, metabolism, and DNA repair [7], suggesting its potential value as a therapeutic target. However, most current studies are limited to in vitro experiments, and the causal effect of palmitoylation in the population and its cross-omics regulatory network are still poorly understood.

Regression analysis from observational studies has long been the primary method for exploring GBM risk factors, but it is prone to confounding bias (e.g., immune cell infiltration heterogeneity) and reverse causation (e.g., tumor-induced epigenetic changes), complicating the identification of true causal relationships [7]. For example, genome-wide association studies (GWAS) have identified over 30 loci associated with GBM susceptibility, but interpreting their biological mechanisms is challenging due to issues with genetic correlation and multiple effects of genes. Mendelian randomization (MR) employs genetic variants as instrumental variables to create a framework for causal inference akin to randomized controlled trials. This approach offers a methodological foundation to address the existing challenges [8–10]. Recent studies have used MR to uncover the causal effects of gut microbiota, plasma metabolites, and cerebrospinal fluid components on GBM [11, 12], further validating the unique advantages of this method in neuro-oncology.

This study integrates multi-omics strategies to systematically analyze the causal mechanisms by which palmitoylation-related genes (ZDHHC3/6/13, PPT2) mediate a protective effect against GBM through neuroimaging endophenotype pathways. Based on previous evidence, we propose a two-level hypothesis: First, at the genetic regulation level, variants in palmitoylation enzyme genes influence mRNA expression, which directly lowers GBM risk. Second, at the imaging mediation level, brain structure and function phenotypes (e.g., corpus callosum axonal density and default mode network connectivity) mediate at least 15% of the genetic protective effect.

In this study, we utilize the FinnGen and UK Biobank databases within a two-sample Mendelian randomization (MR) framework, employing inverse variance weighted (IVW) methods along with multivariate mediation models. Our goal is to measure how palmitoylation-related SNPs influence the risk of GBM, expressed as an Odds Ratio per Standard Deviation, and to identify specific neuroimaging biomarkers related to spatial aspects from structural MRI, DTI, and functional connectivity. We implement the Steiger directional test and sensitivity analyses, including MR-PRESSO and Cochran’s Q, to validate the robustness of our causal inferences. This study provides genetic evidence for novel therapies targeting the palmitoylation pathway. By clarifying how palmitoylation affects brain network plasticity over time and space—such as the atrophy of the parahippocampal gyrus and the integrity of the inferior fronto-occipital fasciculus—this study establishes a theoretical basis for developing early imaging biomarkers for GBM and designing combination therapy strategies.

Methods

Study design

This study used a two-sample Mendelian randomization (MR) approach to analyze the causal relationship between palmitoylation modification and GBM, along with its potential mechanisms (Fig. 1). First, we selected instrumental variables (IVs) based on eQTL and mQTL data (p < 1 × 10⁻⁵). Next, we applied the Inverse Variance Weighted (IVW) method to estimate the direct effects of palmitoylation genes and DNA methylation sites on GBM risk. Additionally, we confirmed the correct direction of causality using the Steiger test. A subsequent two-step MR mediation analysis indicated that palmitoylation may indirectly increase GBM risk by influencing neuroimaging endophenotypes, such as cortical thickness, with this indirect effect accounting for over 15%. Sensitivity analyses, which included Cochran’s Q, MR-Egger, and MR-PRESSO, confirmed the robustness of our findings. This study reveals the potential causal role of palmitoylation modification in GBM and its multi-omics regulatory pathways, providing new insights for targeted interventions.

Fig. 1.

Presents our analytical framework for studying the role of protein palmitoylation in glioma development. It mainly includes: (1) Mendelian randomization analysis, evaluating the direct causal effects of genetically predicted palmitoylation on GBM risk; (2) Mediation analysis, examining the role of neuroimaging endophenotypes and palmitoylation as mediators; (3) Sensitivity analysis, ensuring the reliability of the results. Color-coded pathways represent different mechanisms of action

Data sources

This study combined diverse data sources to investigate the molecular mechanisms of GBM. We obtained genome-wide association analysis (GWAS) data from the Finnish FinnGen database (version R12). This public–private partnership analyzed 406 GBM patients and 378,749 control samples (dataset ID: finngen_R12_C3_GBM_EXALLC.gz) to reveal genotype–phenotype associations in the Finnish ancestral population. We integrated data on the palmitoylation gene family from studies by Chen et al. [13], Li et al. [14], and Chamberlain [15], along with 31 key genes from the eQTLGen database (https://eqtlgen.org), which include APT1/2, PPT1/2, ABHD17A/18B/19C, and DHHC family members (Supplementary Table.S1). We integrated imaging genomics data from the latest 3,935 brain imaging phenotype GWAS datasets released by the UK Biobank. This data was publicly obtained through the BIG40 platform (https://open.win.ox.ac.uk/ukbiobank/big40/) and the European Bioinformatics Institute (EBI) [16]. All original studies received ethical approval and obtained informed consent from participants. This study strictly follows relevant data usage guidelines.

Mendelian randomization analysis of the relationship between Palmitoylation and GBM

This study used the R package “TwoSampleMR” (version 0.6.0) to conduct Mendelian randomization analysis. The goal was to estimate the potential causal relationship between genetically predicted palmitoylation genes and GBM. We selected single nucleotide polymorphisms (SNPs) significantly associated with gene and methylation site expression (p < 1 × 10^–5) as instrumental variables (IVs). We required an F-statistic greater than 10 to prevent bias from weak instruments. To ensure independence, we removed SNPs that were in linkage disequilibrium with an R² value less than 0.1 over a window size of 100 kb. For palmitoylation genes with only one SNP, we used the Wald ratio method. For genes with two or more SNPs, we estimated causal effects using several methods: Inverse Variance Weighted (IVW), MR Egger, Weighted Median, simple mode, and weighted mode. IVW was the primary analysis method. In the analysis, we calculated odds ratios (OR) for GBM risk associated with each standard deviation (SD) increase in palmitoylation gene levels. Additionally, we performed Steiger filtering to validate the causal direction, ensuring that the explanatory power of genetic variants for the exposure variable was higher than for the outcome variable. If an instrumental variable met this assumption, it was labeled “TRUE”; otherwise, it was labeled “FALSE”. After removing all SNPs labeled “FALSE”, all Mendelian randomization analyses were repeated using the IVW method.

Sensitivity analysis

To confirm the reliability of our results, we conducted multiple sensitivity analyses. We used Cochran's Q test to evaluate the heterogeneity of genetic variants. A p-value greater than 0.05 indicates no significant heterogeneity. The intercept from MR-Egger regression was used to assess horizontal pleiotropy, with a p-value greater than 0.05 indicating no evidence of this issue. We employed the MR-PRESSO method to identify potential outliers affected by pleiotropy. A global test p-value > 0.05 indicated the absence of outliers due to horizontal pleiotropy. We conducted a leave-one-out analysis, sequentially excluding each instrumental variable (IV) and repeating the MR analysis to evaluate the impact of each IV on the overall effect estimate [17].

Mediation analysis

This study used an integrated framework that combines two-sample Mendelian randomization (TSMR) with mediation analysis. It systematically investigates the pathway linking palmitoylation to GBM risk and evaluates the potential mediating roles of neuroimaging endophenotypes. In the dimension of neuroimaging endophenotypes: an effect decomposition model was used to partition the total effect (β_{_all}) into a direct effect (β_{_direct}, i.e., the independent effect not dependent on the mediator) and an indirect effect (β₁ × β₂, i.e., the effect mediated by brain structural indices such as hippocampal volume and prefrontal cortex thickness through palmitoylation). A significance threshold was set where the proportion of the indirect effect relative to the total effect (β₁ × β₂/β_{_all}) exceeded 15%.

Results

Mendelian randomization analysis of the relationship between Palmitoylation and GBM

This study systematically assessed how palmitoylation-related genes affect the risk of GBM through a two-sample Mendelian randomization analysis and the inverse variance weighted (IVW) method. The results indicated that four genes demonstrated significant protective effects: ZDHHC3 (OR = 0.17, 95% CI = 0.0602–0.4824), ZDHHC6 (OR = 0.41, 95% CI = 0.2330–0.7267), ZDHHC13 (OR = 0.64, 95% CI = 0.4745–0.8618), and PPT2 (OR = 0.59, 95% CI = 0.3568–0.9843) (Fig. 2). Sensitivity analyses, including MR-Egger regression, the weighted median method, and leave-one-out analysis, confirmed the robustness of the results. Furthermore, no evidence of directional pleiotropy—where a genetic variant influences multiple traits—was detected, as all intercept p-values were greater than 0.05 (see Supplementary Table S1).

Fig. 2.

Mendelian randomization analysis of the relationship between Palmitoylation and GBM

Mediation analysis of the association between Palmitoylation and GBM through neuroimaging endophenotypes

In the first step, we identified 282 neuroimaging endophenotypes that had a significant causal association with GBM risk using Mendelian randomization analysis (IVW method, p < 0.05) (Supplementary Fig. S1, Table S2). Among the associations that reduce risk (OR < 1), decreased functional connectivity and reduced white matter integrity were predominant. The most significant effects were observed with decreased connectivity in the default mode network (DMN) at ICA100 edge 694 (OR = 0.33, 95% CI = 0.1546–0.7105) and in the salience network at ICA100 edge 1426 (OR = 0.34, 95% CI = 0.1567–0.7385). In white matter tracts, ICVF Body of corpus callosum (OR = 0.37, 95% CI = 0.2690–0.5128), FA Splenium of corpus callosum (OR = 0.38, 95% CI = 0.2364–0.5970), and reduced FA Retrolenticular part of internal capsule L (OR = 0.46, 95% CI = 0.3089–0.6716) significantly lowered risk. Additionally, in gray matter structures, reduced ICVF Cingulum cingulate gyrus R (OR = 0.55, 95% CI = 0.3959–0.7716) and aparc-Desikan lh thickness frontalpole (OR = 0.38, 95% CI = 0.2081–0.7094) also provided protective effects. Risk-promoting associations (OR > 1) were marked by hyperconnectivity and abnormal white matter proliferation. The most significant risk-promoting effect was observed with increased connectivity in the DMN at ICA100 edge 193 (OR = 3.77, 95% CI = 1.3004–10.9122), followed by ICA100 edge 1127 (OR = 2.81, 95% CI = 1.3467–5.8511). In white matter, the abnormal increase in L2 Genu of corpus callosum(OR = 2.39, 95% CI = 1.5846–3.6020) and lh area G-front-inf-Opercular (OR = 2.01, 95% CI = 1.0187–3.9669) significantly increased risk. Multimodal analysis showed synergistic effects between structure and function. The co-localization of reduced splenium FA (OR = 0.38) and weakened DMN connectivity (e.g., ICA100 edge 694, OR = 0.33) suggests that the simultaneous degeneration of white matter and functional networks may inhibit tumor invasion along fiber tracts. In contrast, the presence of white matter damage in the motor pathway (left corticospinal tract FA: OR = 0.46) and thickened gray matter in the motor cortex (OR = 2.01) indicates that there is a bidirectional regulation of peri-tumoral neural plasticity. Fifteen results remained robust, highlighting key markers such as corpus callosum body ICVF (OR = 0.37), DMN ICA100 edge 694 (OR = 0.33), and corpus callosum genu ICVF (OR = 2.39).

We evaluated the impact of four palmitoylation genes on neuroimaging endophenotypes in our second analysis. Since PPT2 was excluded due to failing the Presso test, we revealed the causal regulatory effects of the palmitoylase gene family members ZDHHC3, ZDHHC6, and ZDHHC13 on neuroimaging endophenotypes (see Supplementary Fig. S2, Table S3). We found 46 significant associations using the Inverse Variance Weighted method (p < 0.05), with ZDHHC3 associated with 24 of these endophenotypes. Increased expression of ZDHHC3 significantly raised the risk of left fronto-occipital fasciculus optical anisotropy (OR = 1.32, 95%CI = 1.1523–1.5080) and default mode network functional connectivity (ICA100 edge 295, OR = 1.18, 95%CI = 1.0288–1.3443), while decreased expression correlated with reduced L2 Cingulum cingulate gyrus R (OR = 0.78, 95%CI = 0.6846–0.8959). ZDHHC6 positively regulated midbrain peduncle white matter microstructure (OR = 1.15, 95%CI = 1.0626–1.2479) and FA Retrolenticular part of internal capsule L (OR = 1.14, 95%CI = 1.0550–1.2389), but it significantly inhibited the volume of the left parahippocampal gyrus HATA subregion (OR = 0.84, 95%CI = 0.772–0.9088). ZDHHC13 exhibited bidirectional regulatory effects: increased expression was positively associated with enhanced ProbtrackX L2 cgc r (OR = 1.13, 95%CI = 1.0811–1.1862) and cerebellar-cortical functional connectivity (ICA100 edge 74, OR = 1.14, 95%CI = 1.0786–1.2030), while it negatively correlated with rh area parahippocampal (OR = 0.90, 95%CI = 0.8444–0.9637) and FA Splenium of corpus callosum (OR = 0.94, 95%CI = 0.8941–0.9811). These results suggest that palmitoylases may participate in the dynamic balance of brain networks by regulating synaptic plasticity (e.g., ZDHHC3's effect on white matter integrity) and gray matter atrophy (e.g., ZDHHC13's inhibitory effect on the rh area parahippocampal), with ZDHHC13's bidirectional regulatory mechanism potentially providing a potential target for early intervention in GBM.

We calculated the total effect (β_{_all}), mediation effect (β₁*β₂), and direct effect (β_{_direct}) based on the two-step results. We defined a mediation effect greater than 15% as significant, excluding the gene ZDHHC3. The mediation analysis showed that the protective effects of genes ZDHHC13 and ZDHHC6 against GBM significantly rely on the partial mediation effects of specific neuroimaging phenotypes. ZDHHC13 accounted for 15.96% and 22.68% of the protective effect by regulating structural atrophy in the right parahippocampal gyrus (β₁ =—0.10) and enhancing functional connectivity in the default mode network (ICA100 edge 74, β₁ = 0.13 → β2 = − 0.78), indicating that limbic system remodeling and high-order network integration are crucial mechanisms. ZDHHC6 mediated protection by increasing corpus callosum axon density (ICVF, β₁ = 0.14 → β2 = − 0.99, contributing 15.29%) and inhibiting fronto-parietal network connectivity (ICA100 edge 295, β₁ = − 0.15 → β₂ = 0.91, contributing 15.07%) (Table 1).

Table 1.

Mendelian randomization analysis of palmitoylation-related genes mediating GBM risk through neuroimaging phenotypes

Exposure	Mediators	Outcome	β_all	β1	β2	β1*β2	β_dir	Mediated proportion (%)
ZDHHC13	aparc-Desikan rh area parahippocampal	GBM	− 0.4471	0.1006	0.7091	0.0713	− 0.3757	15.96
ZDHHC6	IDP dMRI TBSS ICVF Body of corpus callosum	GBM	0.8880	0.1371	0.9904	0.1358	− 0.7522	15.29
ZDHHC13	rfMRI connectivity (ICA25 edge 152)	GBM	0.4471	0.0836	0.9214	0.0770	− 0.3701	17.22
ZDHHC13	rfMRI connectivity (ICA100 edge 74)	GBM	0.4471	0.1302	0.7788	0.1014	− 0.3457	22.68
ZDHHC6	rfMRI connectivity (ICA100 edge 295)	GBM	0.8880	0.1472	0.9090	0.1338	− 0.7542	15.07%
ZDHHC13	rfMRI connectivity (ICA100 edge 1426)	GBM	0.4471	0.0663	1.0784	0.0715	− 0.3756	15.99

Note Mendelian randomization analysis shows palmitoylation genes (ZDHHC13, ZDHHC6) reduce GBM risk through specific brain imaging features (all mediation effects ≥ 15%). ZDHHC13 works via hippocampal structure (15.96%) and brain network connections (22.68%) and ZDHHC6 via corpus callosum structure (15.29%). See Methods for details. (FA = fractional anisotropy, ICVF = axonal density)

Discussion

This study employed MR analysis to investigate the potential association between palmitoylation modification and the occurrence and development of GBM, along with its underlying mechanisms. The results suggest that genetic variants associated with certain palmitoylation-related genes, such as ZDHHC3, ZDHHC6, ZDHHC13, and PPT2, are linked to a reduced risk of GBM. This finding points towards a potential interplay between genetic factors and brain imaging characteristics. Furthermore, mediation analysis indicates that the observed protective association may partly operate through alterations in brain imaging features. This observation highlights the potential significance of the dynamic interplay between brain structure and function in the pathogenesis of GBM. While these findings underscore the relevance of palmitoylation modification to GBM risk, they provide evidence that could guide the identification of potential targets for future therapeutic exploration. However, it is crucial to note that MR findings represent indirect evidence and are inherently observational, thus requiring further validation through prospective studies and functional experiments before any definitive conclusions or clinical applications can be drawn.

These findings show some parallels with prior research on the association between palmitoylation and cancer. Zhou et al. [18] conducted a systematic review highlighting that palmitoylation plays a important role in various malignant tumors by regulating key biological behaviors of tumor cells, such as proliferation, invasion, and metastasis. This supports our finding that palmitoylation-related genes provide a protective effect against GBM. Zhu et al. [19] observed an association between abnormal palmitoylation regulation and poor prognosis in liver cancer patients, speculating a potential link to tumor metabolic reprogramming. This indicates that palmitoylation could influence tumor progression through diverse mechanisms. The metabolic aspect of this finding might offer a potential avenue for explaining the phenomenon observed in our study (i.e., where palmitoylation-related genes appear to lower GBM risk by altering neuroimaging phenotypes), but the specific connection between these observations requires further investigation to be clarified.

To better understand how palmitoylation influences tumorigenesis, we explored its molecular mechanisms. Research indicates that palmitoylation alters the localization and function of various oncogenes and tumor suppressor genes, providing insights into the molecular mechanisms underlying tumorigenesis [14]. Existing studies have confirmed that palmitoylation plays a crucial role in multiple growth signaling pathways, such as AKT, Wnt, and IGF-1/IGF-1R [20–23]. These findings are consistent with our experimental results, which demonstrate that the ZDHHC3 and PPT2 genes significantly inhibit the development of GBM by influencing cellular signaling and structural integrity. Some studies suggest that the dynamic interactions between palmitoylation and other post-translational modifications (such as phosphorylation, ubiquitination, acetylation, etc.) may contribute to the molecular basis for tumor heterogeneity and differences in treatment response [24, 25]. This hierarchical regulation of the “modification network” suggests that studying a single modification alone cannot fully explain tumor behavior; therefore, it is critical to investigate the synergistic effects of multiple modifications within a systems biology framework.

Based on our mechanistic discussion and Mendelian randomization analysis, we suggest that palmitoylation affects GBM risk by altering neuroimaging endophenotypes. Our analysis specifically identified four palmitoylation-related genes: ZDHHC3, ZDHHC6, ZDHHC13, and PPT2. These genes show a significant protective effect against GBM, indicating a potential causal relationship between palmitoylation and this aggressive brain tumor's development. The odds ratio analysis for these genes demonstrates that the genetic variants associated with them are related to GBM risk, underscoring the need for further investigation into the role of palmitoylation in tumorigenesis [26].

The effect size of neuroimaging endophenotypes in mediating the relationship between palmitoylation genes and GBM risk is somewhat limited, with a contribution rate of 15.07%-22.68%. However, the consistent effect direction and spatial specificity of certain brain regions indicate significant biological importance. ZDHHC13 negatively regulates the volume of the rh area parahippocampal (OR = 0.90, 95%CI = 0.8444–0.9637) and the functional connectivity of the default mode network (ICA100 edge576, OR = 0.91, 95%CI = 0.8723–0.9567). Additionally, ZDHHC6 inhibits the rh volume Cerebral White Matter (OR = 0.92, 95%CI = 0.8507–0.9974). These findings are all consistent with a trend of reduced GBM risk. These features are significantly correlated with clinical prognostic indicators (such as ADC values, Cho/NAA ratios, FA, and rCBV) [27–29] and demonstrate cross-modal synergistic traits: structurally, reduced axon density in the corpus callosum (β = − 0.75, p < 0.05) may inhibit tumor invasion along white matter tracts; functionally, weakened default network connectivity (β = − 0.37, p < 0.05) may indicate the deactivation of peritumoral neural compensatory mechanisms [30]. Notably, ZDHHC13 negatively affects the ICA100 edge 576 and is spatially associated with atrophy of the parahippocampal gyrus. This suggests that palmitoylation may contribute to the formation of the tumor microenvironment by disrupting the stability of the limbic system-cortical circuit, which is closely related to the proliferation and invasion of glioma cells along white matter tracts [31]. In summary, despite the limited contribution rate of imaging endophenotypes, their spatial specificity and multimodal integration in the gene-tumor association offer valuable insights into how genetic factors influence GBM occurrence through brain network remodeling.

This study suggests that palmitoylation plays a significant role in influencing neuroimaging endophenotypes, which is associated with the risk of GBM. These findings provide insights into the molecular mechanisms underlying GBM and highlight the potential for exploring the use of multimodal biomarkers in risk stratification and targeted therapies [32]. Future research should aim to elucidate these pathways further. Moreover, it is crucial to integrate multimodal imaging with molecular experiments to explore the potential causal relationships involved in tumorigenesis and their possible clinical applications. The outcomes of this research are may inform the development of new personalized treatment strategies for GBM.

Conclusion

Although this study successfully established a direct association between the palmitoylation pathway and GBM, and revealed the pivotal mediating role of brain imaging features, such as reduced hippocampal volume, in the mechanism through which palmitoylation influences GBM risk via mediation analysis, thereby proposing a ‘palmitoylation-brain structural remodeling-GBM’ cascade model, several significant limitations necessitate further investigation in future work. The primary limitation lies in the fact that the inference of the mediating role of brain imaging features is predominantly based on statistical analyses of observational data, demanding rigorous experimental validation using techniques like CRISPR/Cas9 to provide stronger causal evidence. Additionally, while the overall sample size is substantial, the specific mediation analyses might be constrained by insufficient sample sizes, potentially affecting the precision of effect estimation and the generalizability of the results. Furthermore, although the integration of multi-source data broadens the analytical perspective, it also introduces heterogeneity due to variations in data acquisition and processing standards across sources, complicating the interpretation of the findings. Moreover, a notable limitation is the absence of comprehensive multiple testing correction in the MR and mediation analyses, which may increase the risk of false positive findings. Finally, and critically, the study remains at the hypothesis-generating stage, lacking clinical translational validation. The actual utility of the identified associations and proposed model as GBM risk predictors, early screening tools, or therapeutic targets awaits rigorous evaluation through prospective clinical studies, intervention trials, and more extensive clinical sample validations.

Author contributions

S. Tan: Investigation, Formal analysis, Writing-original draft; J. You: Validation, Data curation, Conceptualization; H. Liu: Conceptualization, Supervision, Writing—review and editing (Corresponding Author), Software; X. Tang: Conceptualization, Visualization; H. Wang: Investigation, Methodology; L. Zhao: Investigation, Methodology; S. Li: Investigation, Visualization.

Funding

Not applicable.

Data availability

Genome-wide association study data: Derived from the Finnish FinnGen database (version R12), specifically the dataset：finngen_R12_C3_GBM_EXALLC.gz. Imaging genomics data from 3935 brain imaging phenotype GWAS datasets. These data can be publicly accessed through the BIG40 platform (https://open.win.ox.ac.uk/ukbiobank/big40/) and the European Bioinformatics Institute (EBI). eQTLGen data: Obtained from the official website of the eQTLGen project (https://eqtlgen.org), which provides large-scale whole-blood cis-eQTL data.

Declarations

Ethical approval and consent to participate

Not available.

Consent for publication

All authors reviewed and approved the final manuscript.

Competing interests

The authors declare no competing interests.

Clinical trial number

Not applicable.