Exploring the causal relationship between serum EFNB2 levels and epilepsy: a bidirectional Mendelian randomization and co-localization analysis

Abstract

Background

Epilepsy is a severe neurological disorder characterized by persistent seizures and, in some patients, associated neurobiological, cognitive, and psychosocial consequences. It is influenced by various genetic factors, including the Ephrin-B2 (EFNB2) gene.

Methods

This study utilized bidirectional Mendelian randomization (MR) to explore the potential causal relationship between serum levels of EFNB2 and epilepsy using data from extensive genome-wide association studies (GWAS). We selected serum levels of EFNB2 and generalized epilepsy traits, applying strict criteria for instrumental variables to ensure validity and mitigate confounding influences. The analysis included sensitivity tests like the MR pleiotropy residuals and outliers test, as well as co-localization to evaluate shared genetic influences.

Results

Our results indicated a significant causal relationship between serum levels of EFNB2 and epilepsy, suggesting that EFNB2 could be involved in the pathogenesis of epilepsy through mechanisms that may not be directly linked to shared genetic pathways.

Conclusion

These results suggest a potential association between EFNB2 and epilepsy, highlighting the need for further studies to clarify its role and explore its possible relevance as a therapeutic target.

Background

Epilepsy represents a severe neurological disorder that not only affects neural function but also significantly impedes cognitive and developmental progress, presenting considerable challenges for affected individuals and families globally [1]. Characterized by persistent seizures and profound neurodevelopmental delays, epilepsy is a complex condition influenced by various biochemical and genetic factors [2]. It has been reported that disruptions in neuronal signaling, network formation, and genetic factors are often associated with epilepsy [3,4,5,6].

EFNB2 is a gene encoding a member of the Ephrin family of proteins, which are cell surface proteins involved in cell signaling and tissue organization [7]. EFNB2 plays a pivotal role in the regulation of various physiological processes, including angiogenesis, neuronal development, and the establishment of tissue boundaries [8,9,10]. Beyond its well-documented involvement in cancer biology—where its elevated expression is associated with aggressive tumor progression, such as in colorectal and breast cancers [11]. In the context of neurological disorders, especially epilepsy, EFNB2 assumes a critical role due to its involvement in neuronal connectivity and plasticity [12]. Whether EFNB2 directly causes the development of epilepsy, or if it is merely a concurrent genetic variation in the condition, remains unclear. The rationale for investigating EFNB2 in the context of epilepsy lies in its involvement in key signaling pathways, such as mTOR and MAPK/ERK, which are known to contribute to epileptogenesis by regulating neuronal excitability and synaptic remodeling. Dysregulation of these pathways has been implicated in various neurological disorders, including epilepsy, suggesting that EFNB2 may play a mechanistic role in seizure pathogenesis. Despite this, the relationship between EFNB2 and epilepsy has been seldom explored. Previous research has not addressed whether EFNB2 directly influences the development of epilepsy or whether it represents a concurrent genetic or molecular marker associated with the condition. This knowledge gap underscores the importance of further investigation to clarify the specific functions and causal roles of EFNB2 in epilepsy. Understanding EFNB2’s contribution to the pathogenesis of epilepsy could provide new insights into disease mechanisms and identify potential therapeutic targets.

MR is an innovative genomics method that leverages genetic variants as instrumental variables to establish causal relationships between exposures and outcomes using data from population-based observational studies [13]. This approach is particularly useful in neuropsychiatry for determining the directionality of associations. Unlike latent causal variable analysis, which only evaluates two phenotypes and does not test bidirectional relationships, MR can robustly assess directionality. Furthermore, the multivariable extension of MR (MVMR) enhances this capability by allowing the simultaneous evaluation of multiple exposures. This can identify the direct effects of each exposure on an outcome, thereby clarifying complex relationships, such as those between brain structure and alcohol consumption, while accounting for potential mediating or confounding factors [14].

In our research, we utilized bidirectional MR to investigate potential causal connections between EFNB2 gene expression and the incidence of epilepsy. This analysis was based on the most extensive GWASs available for risk factors in European populations. To enhance the robustness of our findings, we conducted sensitivity analyses focusing exclusively on SNP (Single nucleotide polymorphism) with potential functional implications and iteratively excluding each SNP to assess their individual effects. Moreover, we compared our causal estimates with the findings from a meta-analysis of observational studies. We performed a co-localization analysis to ascertain whether identical genetic variations influenced both the gene expression of EFNB2 and the incidence of epilepsy.

Methods

Data sources

We selected serum levels of protein EFNB2 in the EUR population from the GWAS_Catalog database [15] (ID:GCST90090281), GWAS summary-level data was used as exposure data, the levels of EFNB2 (Serum protein concentrations were measured in nanograms per milliliter. These measurements were obtained using a high-throughput proteomic platform, which quantifies protein levels across a wide dynamic range) were measured as part of a broader set of proteins included in the study, with data collected from large cohort populations. which includes 5,365 samples, and the number of SNPs detected was 7,492,219 (Table 1). From the Finngen database [16], we selected the Generalized epilepsy (ID: finngeN_R9_GE) GWAS summary level data of the EUR population as the outcome data, which includes 366,832 samples and detected 20,170,015 SNPs (Table 1). The diagnosis of generalized epilepsy was based on participant data from large-scale genetic studies that followed standardized diagnostic criteria.To achieve 80% statistical power and a significance level of 0.05, the required sample size was calculated to be 18 based on the expected odds ratio (OR) of 0.8 from the Mendelian randomization analysis. The sample size in the database was sufficient to detect a significant causal relationship between serum EFNB2 levels and epilepsy, given the strong instrumental validity with an F statistic greater than 10.

Table 1 Details of exposure and outcome data

Regarding the validation cohort, we selected the serum levels of protein EFNB2 (ID: GCST90087807) GWAS summary-level data of the EUR population from the GWAS_Catalog database as the exposure data, which contained 5,361 samples, and the number of detected SNPs was 7,492,219 (Table 2). The Non-cancer illness code, self-reported: epilepsy (ID: ukb-b-16309) GWAS summary-level data of the EUR population was selected from the IEU database as the outcome data [17], which contained 462,933 samples and the number of detected SNPs was 4,455,394 (Table 2).

Table 2 Details of exposure and outcome data of validation cohort

Selection of instrumental variables

To ensure that the instrumental variables are valid, they must meet the three assumptions of MR, ① Correlation assumption: genetic variation needs to be strongly correlated with exposure factors; ② Independence assumption: genetic variation is not associated with any possible confounding factors; ③ Exclusivity assumption: genetic variation can only affect the outcome through the exposure factors and cannot be directly related to the outcome. We used a series of criteria to screen instrumental variables: to ensure sufficient instrumental variables, we selected SNPs with a genome-wide significance level of p-value < 5e⁻⁶, and F statistics > = 10 (F statistics = beta2/se2) to ensure that the instrumental variables are strongly correlated with the exposure factors. This choice was made to balance the trade-off between identifying a sufficient number of SNPs for robust MR analyses and minimizing the risk of weak instrument bias. Although adopting a more lenient threshold increases the potential for false positives, we mitigated this risk by conducting additional sensitivity analyses, including MR-PRESSO, MR-Egger intercept tests, and leave-one-out analyses, to identify and address any pleiotropy or bias arising from weak instruments. Secondly, the selected instrumental variables needed to satisfy the independence test. To check the independence and linkage disequilibrium (LD) effect of these variables, we set the linkage disequilibrium parameter r² to 0.001 and the genetic distance to 10000 kb. Finally, to ensure that the effect of the SNP on exposure was consistent with the effector allele direction on the outcome, we removed palindromic SNPs (e.g., with A/T or G/C alleles). The selection of the above instrumental variables ensures the reliability of our research results. The selection criteria outlined above ensure the reliability and robustness of our instrumental variables, while acknowledging and addressing the limitations inherent in the use of a less stringent significance threshold.

Mendelian randomization analysis

We conducted a two-sample MR analysis using inverse variance weighted (IVW), MR-Egger, weighted median, simple mode, and weighted mode methods, to estimate the effect of exposure factors on outcome data. In terms of algorithmic principle, the IVW method integrates the Wald ratio of the causal effect of each SNP through meta-analysis to produce the most accurate estimate. Therefore, the results of more than one instrumental variable are mainly based on the IVW method, supplemented by the other four methods.

Sensitivity analysis

For sensitivity analysis, various statistical methods were applied. We first performed the MR pleiotropy residuals and outliers (MR-PRESSO) test to detect horizontal pleiotropy (p < 0.05) and remove outlier SNPs. Horizontal pleiotropy was also assessed using the MR-Egger intercept method. Then, Cochran’s Q test (heterogeneity test) was used to analyze the heterogeneity between each instrumental variable. According to the degree of heterogeneity (when Q > 0.05, the IVW calculation used the fixed effect model, when Q < 0.05 the IVW calculation used the random effect model), further analysis was conducted using the fixed effects model or random effects model. We also performed a leave-one-out analysis to detect whether significant associations between exposure factors and outcomes were dominated by a single SNP, meaning that different SNPs were removed in each iteration for MR analysis. All the above analyses (including sensitivity analysis and MR analysis) were performed in the R package TwoSampleMR [17], and the software used was RStudio version 2023.12.1.

Co-localization analysis

After determining the causal relationship between exposure factors and outcomes, the following method was used to calculate gene co-localization: each instrumental variable screened in the Mendelian analysis was used as lead SNPs, and SNPs within the upstream and downstream 500 kb range are co-localized [18, 19], with 0.75 as the threshold to determine whether there was pleiotropy at the gene level. This step of analysis was performed in the R package coloc [20].

Results

Selection of instrumental variables

By screening for SNPs with P values less than 5e⁻⁶, 27 SNPs significantly correlated with exposure factors were identified. Then, the ld_clump function in the R package ieugwasr was used to perform linkage disequilibrium analysis. The r² and kb parameters were set to 0.001 and 10,000 to remove highly correlated variables to ensure that the instrumental variables did not interfere with each other, there were 8 instrumental variables identified that were significantly correlated with the exposure factors and independent of each other. Finally, instrumental variables with F values greater than 10 were screened. F values are an indicator used to measure the effectiveness of instrumental variables. F values greater than 10 are usually considered significant enough and are used as screening criteria for instrumental variables to ensure our final selection. The 8 instrumental variables (Table S1) can be statistically effectively used for causal inference analysis.

Mendelian randomization analysis

After removing palindromic sequences and unmatched SNPs, 8 instrumental variables were included for two-sample MR analysis (Table S2). The results of the IVW model show that there is a significant causal relationship between exposure factors and outcome indicators (OR = 0.6552, 95%CI = 0.4642–0.9249, p-value = 0.0162, Fig. 1A), and there is a causal relationship between the increase in exposure factors and the decrease in outcome indicators (Table 3). The MR estimate of the causal effect was OR = 0.6552 (95% CI: 0.4642, 0.9249), indicating that for each standard deviation increase in serum EFNB2 levels, the odds of developing epilepsy decreased by approximately 34.5%.The 95% confidence interval did not cross 1, and the p-value was 0.0162, indicating that this effect is statistically significant.The pleiotropy test results of MR-Presso showed that there was no pleiotropy (Table 4), the pleiotropy test results of MR-Egger intercept showed that there was no pleiotropy (Table 4), and the Cochran’s Q test results showed that there was no heterogeneity (Table 5, Fig. 1B). The results of the leave-one-out analysis showed that the overall relationship estimate was not significantly affected by any single SNP (Fig. 1C). In the analysis of individual SNPs, it was found that some SNPs had a certain protective trend on outcome indicators (Fig. 1D).

Fig. 1

MR analysis. The results of the IVW model showed that there is a significant causal relationship between exposure factors and outcome indicators (A). The results of the Cochran’s Q test showed that there was no heterogeneity (B). The results of the leave-one-out analysis showed that the overall relationship estimate was not significantly affected by any single SNP (C). The analysis of individual SNPs showed that some had a certain protective trend on outcome indicators (D)

Table 3 MR analysis results

Table 4 Pleiotropy test

Table 5 Heterogeneity test

Co-localization analysis

Although the results of Mendelian analysis have suggested a possible causal relationship between exposure and outcome, it is unclear whether exposure is a mediator of the genetic pathogenic pathway of outcome. Therefore, we used coloc to test for shared causal SNPs between exposure and outcome. We included 8 causal SNPs screened out in the MR analysis and conducted colocalization based on the 500 kb upstream and downstream of these SNP loci. Using 0.75 as the threshold, we found that the analysis results were all negative (Table 6), and there was no significant relationship between exposure and outcome. There is a lack of evidence of colocalization, i.e., pathways of genetic influence between exposures and outcomes may not overlap or share the same genetic variation. This result may imply that the causal relationship between exposure and outcome is not through a common genetic pathway, but may involve other biological mechanisms or environmental factors.

Table 6 SNP co-localization results

Mendelian randomization analysis of the validation cohort

Selection of instrumental variables

By screening for SNPs with P-values less than 5e-06, 97 SNPs that were significantly correlated with the exposure factors were identified. Then we used the ld_clump function in the R package ieugwasr to perform linkage disequilibrium analysis. The r² and kb parameters were set to 0.001 and 10,000 to remove highly correlated variables to ensure that there was no interference among the instrumental variables. There were 18 instrumental variables identified that were significantly related to the exposure factors and independent of each other. Finally, instrumental variables with F values ​​greater than 10 are screened. F values ​​are an indicator used to measure the effectiveness of instrumental variables. F values ​​greater than 10 are usually considered significant enough and are used as screening criteria for instrumental variables to ensure our final selection. The 18 instrumental variables (Table S3) are statistically effective for causal inference analysis.

Mendelian randomization analysis

After removing palindromic sequences and unmatched SNPs, 10 instrumental variables were included for two-sample MR analysis. The results of the IVW model show that there is a statistically significant causal relationship between exposure factors and outcome indicators (OR = 0.9984, 95% CI = 0.9971–0.9996, p-value = 0.0126, Fig. 2A). This indicates a weak but significant inverse association, suggesting that an increase in EFNB2 levels is associated with a slight decrease in epilepsy risk (Table 7). The pleiotropy test results of MR-PRESSO showed no evidence of horizontal pleiotropy (Table 8), and the MR-Egger intercept also indicated no pleiotropy (Table 8). Furthermore, the Cochran’s Q test results confirmed that there was no heterogeneity among the included SNPs (Table 9, Fig. 2B). The leave-one-out analysis demonstrated that the overall causal estimate was not significantly influenced by any single SNP (Fig. 2C). Additionally, individual SNP analysis showed that certain SNPs exhibited a protective trend against outcome indicators (Fig. 2D).

Fig. 2

MR analysis of the validation cohort. The results of the IVW model showed that there is a significant causal relationship between exposure factors and outcome indicators (A). The results of the Cochran’s Q test showed that there was no heterogeneity (B). The results of the leave-one-out analysis showed that the overall relationship estimate was not significantly affected by any single SNP (C). The analysis of individual SNPs showed that some had a certain protective trend on outcome indicators (D)

Table 7 MR analysis results

Table 8 Pleiotropy test

Table 9 Heterogeneity test

However, compared to the primary cohort, the causal effect observed in the validation cohort was notably weaker. This discrepancy may be attributed to differences in data collection methods. Specifically, the validation cohort relied on self-reported epilepsy diagnoses, which are susceptible to recall bias and potential misclassification. Such bias could introduce noise into the outcome variable, thereby attenuating the observed association. Moreover, the larger sample size of the validation cohort, while increasing statistical power, may also dilute the effect size due to the inclusion of less strictly defined epilepsy cases, such as undiagnosed or subclinical forms of the disease.

Another plausible explanation for the weaker effect is the broader phenotype definition in the validation cohort compared to the primary cohort, which focused exclusively on clinically confirmed generalized epilepsy. The inclusion of more heterogenous epilepsy subtypes in the validation cohort could reduce the specificity of the association. Despite these limitations, the statistically significant results in both cohorts provide supportive evidence for a potential causal relationship between EFNB2 and epilepsy.

Co-localization analysis

Although the results of Mendelian analysis have suggested a possible causal relationship between exposure and outcome, it is unclear whether exposure is a mediator of the genetic pathogenic pathway of outcome. For this reason, we used coloc to test for shared causal SNPs between exposure and outcome. We included the 10 causal SNPs screened out in the MR analysis and conducted colocalization based on the 500 kb upstream and downstream of these SNP loci. Using 0.75 as the threshold, we found that the analysis results were all negative (Table 10), and there was no significant relationship between exposure and outcome. There is a lack of evidence of colocalization, i.e., pathways of genetic influence between exposures and outcomes may not overlap or share the same genetic variation. This result may imply that the causal relationship between exposure and outcome is not through a common genetic pathway, but may involve other biological mechanisms or environmental factors.

Table 10 SNP co-localization results

Reverse Mendelian analysis

We performed MR analysis by exchanging outcomes and exposures (Table 11) to further determine the direction of causality between the two related phenotypes. By screening SNPs with P-value less than 5e-06, we obtained 101 SNPs that were significantly related to the exposure factors. Then, the ld_clump function in the R package ieugwasr was used to perform linkage disequilibrium analysis. The r² and kb parameters were set to 0.001 and 10,000 to remove highly correlated variables to ensure that the instrumental variables would not interfere with each other. There were 11 instrumental variables that were significantly related to the exposure factors and independent of each other. Finally, instrumental variables with F values greater than 10 are screened. F values are an indicator used to measure the effectiveness of instrumental variables. F values greater than 10 are usually considered significant enough and are used as screening criteria for instrumental variables to ensure our final selection. The 11 instrumental variables (Table S4) are statistically effective for causal inference analysis. After removing palindromic sequences and unmatched SNPs, 10 instrumental variables were included for two-sample MR analysis (Table S5). The results of the IVW model showed that there was no significant causal relationship between exposure factors and outcome indicators (Table 12).

Table 11 Details of exposure and outcome data

Table 12 MR analysis results

Discussion

This is the first study to directly demonstrate the causal relationship between EFNB2 and epilepsy. In our study, we selected 7,492,219 SNPs from 5,365 individuals in a GWAS to form the exposure group. Additionally, we identified 27 SNPs associated with serum EFNB2 concentrations. After excluding palindromic sequences and unmatched SNPs, we included 8 instrumental variables for a two-sample Mendelian Randomization (MR) analysis. In the research cohort, we discovered that reduced serum EFNB2 levels may lead to the occurrence of epilepsy (P = 0.016; OR: 0.655; 95% CI: 0.464–0.925). Our reverse Mendelian randomization study indicated that epilepsy does not cause changes in serum EFNB2 concentrations (Table 12). However, in the validation cohort, though we also identified the causal relationship between EFNB2 and epilepsy(P = 0.012), the odds ratio is less significant(OR: 0.998; 95% CI: 0.997–0.999). That may be because we used the generalized epilepsy samples as the outcome group in the research cohort while we used the non-cancer illness code, self-reported epilepsy as the outcome group in the validation cohort which may cause recall bias. However, for the co-localization results, we found that the EFNB2-related SNPs do not share causal genetic variants with epilepsy, which further supports the causal relationship between them.

Currently, studies on EFNB2 have primarily focused on its role in angiogenesis, particularly its contribution to the development and progression of tumors, such as colorectal and breast cancer [21]. A recent clinical study demonstrated that inhibiting EFNB2 improved overall survival and objective response rates in patients with metastatic urothelial carcinoma [22]. In the central nervous system, EFNB2 has been shown to play a critical role in hippocampal neurogenesis, neural crest cell migration, and synaptic plasticity [23, 24]. These processes are highly relevant to neuronal development and network stability, both of which are critical in the context of epilepsy. Our findings suggest that reduced serum EFNB2 concentrations may contribute to the development of epilepsy. However, the specific biological mechanisms underlying this relationship remain unclear and require further exploration. EFNB2 is known to interact with key signaling pathways, including mTOR and MAPK/ERK, which have been implicated in epilepsy pathogenesis. For example, the mTOR pathway regulates neuronal excitability and synaptic plasticity, and its dysregulation has been associated with epilepsy, particularly in conditions such as tuberous sclerosis complex. Similarly, the MAPK/ERK pathway plays a critical role in neuronal survival, differentiation, and synaptic plasticity, and its hyperactivation has been linked to epileptogenesis in experimental models [25,26,27,28]. These findings suggest that EFNB2 may influence epilepsy through its modulation of these pathways, potentially affecting neuronal connectivity and excitability.

From another perspective, patients with long-standing epilepsy often exhibit cognitive impairments and memory decline, which is a sign of the decline of synaptic plasticity [29]. Given that EFNB2 is associated with synaptic plasticity [30], it is plausible to consider that the decline in EFNB2 might primarily contribute to these secondary complications rather than directly causing epilepsy itself. This suggests that while the role of EFNB2 in initiating epilepsy might be complex and still under investigation, its more pronounced impact could be on the neurodegenerative aspects seen in chronic epilepsy patients. This relationship highlights a potential area of focus where EFNB2’s involvement in synaptic modulation and neuronal connectivity could be critical in managing or possibly mitigating the cognitive decline associated with long-term epilepsy. Further research into how EFNB2 affects these cognitive domains could provide valuable insights into comprehensive treatment strategies for epilepsy that address not only the seizures but also the broader neurological consequences. Therefore, whether we can mitigate the progression of the disease by administering EFNB2 remains unknown.

Our study still has several limitations. Firstly, the cohort included only participants of European ancestry, which restricts the generalizability of our findings to other populations. Given the potential for genetic heterogeneity across different ethnic groups, future studies should aim to replicate these findings in more diverse populations to validate their applicability. Secondly, our analysis relied on serum EFNB2 concentrations as a proxy for its systemic role, but we were unable to assess its concentration in cerebrospinal fluid (CSF), which is more directly relevant to the central nervous system. This limitation stems from the lack of available CSF data in the utilized databases. Investigating EFNB2 levels in CSF could provide more direct insights into its involvement in epilepsy pathogenesis. Thirdly, while Mendelian randomization is a powerful tool to infer statistical causality, it cannot establish biological causation. The absence of laboratory-based functional studies to validate the specific effects of EFNB2 variants on neurobiological processes limits the interpretability of our findings. Experimental approaches, such as in vitro studies of neuronal cell lines or in vivo animal models, are necessary to elucidate the mechanisms by which EFNB2 may influence epilepsy through biochemical pathways. Additionally, epilepsy is a heterogeneous disorder with multiple subtypes that may have distinct genetic and biological underpinnings. Our study primarily focused on generalized epilepsy, and the findings may not extend to other forms such as focal epilepsy or idiopathic epilepsy. Future studies should investigate subtype-specific effects of EFNB2 to determine whether its role varies across epilepsy types, which could provide more precise biomarkers or therapeutic targets. Lastly, the reliance on self-reported epilepsy data in the validation cohort introduces the possibility of recall bias, which may have affected the robustness of our findings. Employing more clinically verified datasets in future research would improve the reliability of these results. Moreover, the co-localization analysis in our study revealed that EFNB2-related SNPs and epilepsy-associated SNPs do not share causal genetic variants, which introduces ambiguity regarding the observed causal relationship between EFNB2 and epilepsy. This result suggests that the relationship identified through Mendelian randomization may not be mediated by direct genetic overlap but rather through alternative mechanisms. One possible explanation is the existence of indirect genetic pathways, where EFNB2 variants influence intermediate phenotypes that, in turn, contribute to epilepsy risk. Another plausible factor could involve environmental interactions that modulate the expression or function of EFNB2 in a way that affects epilepsy susceptibility. These findings underscore the need for future studies that integrate multi-omics approaches, such as transcriptomics or epigenomics, to better understand the interplay between EFNB2 and epilepsy. Additionally, exploring non-genetic factors, such as environmental triggers or epigenetic modifications, may provide further clarity on the pathways connecting EFNB2 to epilepsy.

To date, there have been no studies specifically investigating the relationship between serum EFNB2 concentrations and epilepsy. This gap underscores the need for further research to clarify the systemic and central roles of EFNB2 in epilepsy pathogenesis. While our findings suggest a potential association, they remain preliminary and exploratory. Understanding the mechanisms linking EFNB2 to epilepsy, as well as its possible role in neurological complications associated with the disease, may eventually inform biomarker development and therapeutic strategies. However, such advancements will require rigorous experimental validation and broader, more diverse population studies to strengthen the evidence base and support translational efforts.

Conclusion

In conclusion, this study has elucidated a significant causal relationship between serum levels of EFNB2 and the development of epilepsy, employing a rigorous bidirectional Mendelian randomization approach. Our findings underscore the critical role of the EFNB2 gene in neurological disorders, particularly epilepsy, and highlight its potential as a target for therapeutic interventions. Further research is required to explore the underlying mechanisms between EFNB2 and epilepsy.