The overall flow chart of the bidirectional MR study is shown in Figure 1. The genome-wide association studies (GWAS) summary statistics for GERD were obtained from a recent genome-wide association meta-analysis study (12) that included 129,080 European GERD patients and 473,524 healthy controls (20). These data are available in the IEU Open GWAS Project database (21).¹ Additionally, GWAS summary statistics for anxiety disorders and depression were extracted from the FinnGen consortium R8 release (22).² The data for anxiety disorders were obtained from 290,361 individuals (35,385 cases and 254,976 controls), while the data for depression came from 338,111 individuals (38,225 cases and 299,886 controls). Table 1 provides details of the GWAS summary-level data of exposure and outcome analyzed in this MR study. All data analyzed in this study were obtained from publicly available databases in which ethical approval was obtained for each cohort, and informed consent was obtained from all participants prior to participation.

We selected IVs based on three generally recognized assumptions: (i) IVs need to be strongly associated with exposure, (ii) IVs are independent of confounders, and (iii) IVs are solely related to outcomes through exposure without a direct association with outcomes (17). We first screened the IVs for MR analysis using a genome-wide significance threshold (P < 5e−8). In addition, to increase the confidence of our findings, we additionally screened more IVs under a locus-wide significance threshold (P < 1e−5) for a secondary MR analysis. The SNPs within a window size of 10,000 kb were pruned under the threshold of r² < 0.001 to mitigate linkage disequilibrium (LD), thus ensuring the independence of each IV. Smoking, alcohol consumption, and body mass index (BMI) might be potential confounders influencing GERD, anxiety disorders, and depression (23–27). Therefore, we retrieved SNPs associated with these confounders (P < 5e−8) from the IEU Open GWAS project database and excluded them from the IVs. The accession number of these confounders is shown in Table 2. Then, palindromic SNPs, outcome-related SNPs (P < 0.05), and SNPs not present in outcome GWAS summary data were removed from the IVs. Finally, we calculated the F-statistic of IVs to assess the degree of weak instrumental bias. Only IVs with F > 10 were retained to avoid bias caused by weak IVs (28).

The inverse-variance weighted (IVW) method was used as the primary analytical method for estimating causal effects, which is an extension of the Wald ratio estimator based on the principles of meta-analysis (29). The significance threshold was set at P < 0.05, and the results of causal associations were presented as odds ratios (OR) and 95% confidence intervals (95% CI). To further evidence the stability and directionality of the results, in addition to the IVW method, five additional MR methods [maximum likelihood, MR-Egger, weighted median, robust adjusted profile score (MR-RAPS), and mode-based estimate (MR-MBE)] were performed to assess causal associations. The maximum likelihood is a traditional means which estimates probability distribution parameters by maximizing the likelihood function with low standard errors (30). The criterion for using the weighted median method is that at least 50% of the SNPs must satisfy the premise that they are valid IVs (31). The MR-Egger method provides unbiased estimates even in the presence of horizontal pleiotropy (32). MR-RAPS provides robust estimates in the presence of systematic and idiosyncratic pleiotropy (33). MR-MBE provides robust causal estimates when horizontal pleiotropy is present in most IVs and has a stronger effect on detecting causality than MR-Egger (34).

We then performed a series of sensitivity analyses. First, the MR Steiger test was performed to ensure that causal inferences were not biased by reverse causality (35). Next, Cochran’s Q test was used to assess heterogeneity. Then, the MR-Egger intercept and MR-PRESSO global tests were used to detect horizontal pleiotropy (36, 37). Finally, the leave-one-out sensitivity analysis was performed to evaluate the robustness of the results.

For the above analysis process, we first performed the MR analysis based on the IVs screened under the genome-wide significance threshold (P < 5e−8). Subsequently, we performed a secondary MR analysis based on more IVs screened under the locus-wide significance threshold (P < 1e−5) to validate the findings. All analyses in this study were performed based on R software (version 4.2.1). The “TwoSampleMR” R package (38),³ the “MendelianRandomization” R package (39),⁴ and the “’MRPRESSO” R package⁵ (37) were used in our MR study.

We further performed a reverse MR analysis to assess whether anxiety disorders or depression causally affect GERD. Since the number of SNPs satisfying the genome-wide significance threshold (P < 5e−8) in the GWAS summary-level dataset for anxiety and depression was extremely limited, we screened SNPs satisfying the locus-wide significance threshold (P < 1e−5) threshold as IVs associated with anxiety or depression. Subsequently, (i) screening of IVs, (ii) MR analysis, and (iii) sensitivity analyses were performed as described in sections “2.2. Selection of instrumental variables” and “2.3. Statistical methods.”

Finally, a multivariable MR (MVMR) analysis was implemented for significant exposure-outcome pairs identified by univariate MR analysis. Specifically, three confounders, smoking (IEU GWAS ID: “ieu-b-4877”), alcohol consumption (IEU GWAS ID: “ukb-b-5779”), and BMI (IEU GWAS ID: “ukb-b-19953”), were included for MVMR analysis. After combining the GWAS summary level datasets of exposure and the three confounders, it should be ensured that each IV is strongly correlated (P < 5e−8) with at least one or more of the exposure or the three confounders. Then, the SNPs within a window size of 10,000 kb were pruned under the threshold of r² < 0.001 to mitigate LD. Finally, after excluding palindromic SNPs, outcome-related SNPs (P < 0.05), and SNPs not present in outcome GWAS summary data, we used the IVW method to assess causal effects after adjusting for confounders.

The MR results of this section were based on the IVs screened under the genome-wide significance threshold (P < 5e−8).

First, a total of 41 SNPs associated with confounders (smoking, alcohol consumption, and BMI) were excluded. Subsequently, after excluding SNPs not present in the outcome, outcome-related SNPs, and palindromic SNPs, we assessed the causal effects of GERD on anxiety disorders and depression based on 15 and 16 IVs, respectively. Details of the confounders/outcomes-related SNPs are available in Supplementary Table 1, and details of the IVs for MR analysis are presented in Supplementary Table 2. In addition, the F-statistics of all IVs ranged from 29.75 to 50.68.

The results of the MR analysis based on the IVs screened under the genome-wide significance threshold are presented in Table 3. The MR results suggest a causal effect of GERD on anxiety disorders and depression. Specifically, the MR results of IVW indicated that GERD significantly increased the risk of anxiety disorders (OR = 1.35, 95% CI 1.15–1.59, P = 2.25 × 10^–4) and depression (OR = 1.32, 95% CI: 1.15–1.52, P = 1.26 × 10^–4) (Table 3). In addition, similar causal estimation results were derived from five other MR methods, including maximum likelihood, MR-Egger, weighted median, MR-RAPS, and MR-MBE (Table 3 and Figure 2).

Subsequently, sensitivity analyses were performed to assess the robustness of the MR results. First, the MR Steiger test indicated that the inferred causal direction between exposure (GERD) and outcomes (anxiety disorder and depression) was “TRUE.” Then, Cochran’s Q test suggested that there was no heterogeneity among the IVs in our MR analysis (P > 0.05) (Table 4). In addition, the results of both the MR-Egger intercept test and the MR-PRESSO global test suggested that the MR analysis did not suffer from any potential influence of horizontal pleiotropy (P > 0.05) (Table 5). Finally, the leave-one-out sensitivity analysis confirmed the robustness of the MR results since there were no leading SNPs that could drastically affect the results after being eliminated (Supplementary Figure 1).

In order to increase the confidence of the MR results, a secondary MR analysis was performed using IVs screened based on the locus-wide significance threshold (P < 1e−5), for which the following section shows the results.

First, a total of 92 SNPs associated with confounders (smoking, alcohol consumption, and BMI) were excluded. Subsequently, after excluding SNPs not present in the outcome, outcome-related SNPs, and palindromic SNPs, we assessed the causal effects of GERD on anxiety disorders and depression based on 75 and 79 IVs, respectively. Details of the confounders/outcomes-related SNPs are available in Supplementary Table 3, and details of the IVs for MR analysis are presented in Supplementary Table 4. In addition, the F-statistics of all IVs ranged from 19.67 to 50.68.

The results of the MR analysis based on the IVs screened under the locus-wide significance threshold are presented in Table 6. Interestingly, similar to the MR results in the previous section, the MR results in this section remained indicative of a causal effect of GERD on anxiety and depression. Specifically, the MR results of IVW indicated that GERD significantly increased the risk of anxiety disorders (OR = 1.22, 95% CI: 1.13–1.32, P = 1.19 × 10^–6) and depression (OR = 1.25, 95% CI: 1.16–1.34, P = 1.81 × 10^–9) (Table 6). In addition, similar causal estimation results were derived from five other MR methods, including maximum likelihood, MR-Egger, weighted median, MR-RAPS, and MR-MBE (Table 6 and Figure 3).

Subsequently, sensitivity analyses were performed to assess the robustness of the MR results. First, the MR Steiger test indicated that the inferred causal direction between exposure (GERD) and outcome (anxiety disorder and depression) was “TRUE.” Then, Cochran’s Q test suggested that there was no heterogeneity among the IVs in our MR analysis (P > 0.05) (Table 7). In addition, the results of both the MR-Egger intercept test and the MR-PRESSO global test suggested that the MR analysis did not suffer from any potential influence of horizontal pleiotropy (P > 0.05) (Table 8). Finally, the leave-one-out sensitivity analysis confirmed the robustness of the MR results since there were no leading SNPs that could drastically affect the results after being eliminated (Supplementary Figure 2).

We performed a reverse MR analysis to assess whether anxiety disorders or depression causally affect GERD. First, 12 and 9 SNPs associated with confounders (smoking, alcohol consumption, and BMI) were excluded from the IVs of anxiety disorders and depression, respectively. Subsequently, after excluding SNPs not present in the outcome, outcome-related SNPs, and palindromic SNPs, we assessed the causal effects of anxiety disorders and depression on GERD based on 16 and 16 IVs, respectively. Details of the confounders/outcomes-related SNPs are available in Supplementary Table 5, and details of the IVs for reverse MR analysis are presented in Supplementary Table 6. In addition, the F-statistics of all IVs ranged from 19.71 to 35.67.

All MR methods showed no causal relationship between anxiety or depression and GERD risk (P > 0.05) (Table 9). Cochran’s Q test showed that the reverse MR analysis was not influenced by heterogeneity (P > 0.05) (Table 10). In addition, the MR-Egger intercept test and MR-PRESSO global test indicated that the reverse MR analysis was not influenced by water product pleiotropy (P > 0.05) (Table 11). Finally, the leave-one-out sensitivity analysis confirmed the robustness of the reverse MR results (Supplementary Figure 3).

We performed an MVMR analysis to assess the causal effect of GERD on anxiety and depression after adjusting for three confounding factors (smoking, alcohol consumption, and BMI). MVMR analysis identified that: for anxiety disorders, after adjusting for smoking (OR = 1.23, 95% CI: 1.12–1.35, P = 1.06E−05), alcohol consumption (OR = 1.23, 95% CI: 1.13–1.35, P = 5.61E−06), BMI (OR = 1.35, 95% CI: 1.18–1.55, P = 2.43E−05), and all of these three confounders (OR = 1.50, 95% CI: 1.29–1.75, P = 1.65E−07), GERD remained causally related to anxiety disorders risk and had a more substantial effect than the causal relationship identified by univariate MR (Figure 4A). For depression, after adjusting for smoking (OR = 1.14, 95% CI: 1.06–1.24, P = 8.40E−04), alcohol consumption (OR = 1.23, 95% CI: 1.15–1.33, P = 3.10E−08), BMI (OR = 1.31, 95% CI: 1.16–1.48, P = 2.00E−05), and all of these three confounders (OR = 1.32, 95% CI: 1.15–1.52, P = 7.62E−05), GERD remained causally associated with depression risk, with effects remaining consistent with univariate MR results (Figure 4B). Overall, GERD is causally associated with an increased risk of anxiety disorders and depression.