Introduction

Peptic ulcer disease (PUD) is a common gastrointestinal disorder that significantly impacts the quality of life [1, 2]. Management has become more challenging than ever before due to the increasing prevalence of PUD not caused by the use of NSAIDs or Helicobacter Pylori infection [2]. The identification of modifiable risk factors for PUD is clinically important in reducing the burden of the disease.

Obesity has been associated with an alteration in gut microbiota [3], gut inflammation, and the breakdown of the gastrointestinal mucosal epithelial barrier [4]. Epidemiological studies suggested that obesity may increase the incidence of PUD [5,6,7]. However, evidence of the association was contradictory. In another observational study, Jeung Hui Pyo et al. reported that obesity was not related to PUD [8] upon multiple adjustment. In addition, observational associations may be biased by reverse causation and residual confounders, thus distorting true relationships.

Clarifying the causal link between obesity and PUD may offer new avenues for the treatment against PUD. Mendelian randomization (MR) has been widely used as an epidemiological tool for causality inference in associations of exposures with outcomes [9]. With genetic variants leveraged as instruments, this technique can minimize the biases inherent in observational studies, thereby strengthening the causal inference. The present study sought to evaluate the potential causal effect of total obesity and abdominal obesity on PUD risk by the MR approach.

Methods

Study design

First, we performed two-sample MR analyses to assess the effect of body mass index (BMI) (total obesity) and waist-to-hip ratio (WHR) (abdominal obesity) as exposures on PUD as an outcome trait. Then, we performed reverse MR analyses, where PUD was used as exposure and obesity characteristics (BMI/WHR) were used as outcome. The following three key assumptions should be considered when performing MR analysis. First, instrumental variables (IVs) are linked to BMI or WHR at a genome-wide significance level; Second, IVs are not associated with potential confounding factors; Third, IVs should lead to PUD exclusively via the selected exposure. The current study did not require specific ethical approval or written informed consent.

Data sources and instruments variables (IVs) selection

Summary-level data for PUD was derived from a genome-wide association study (GWAS) including up to 16,666 cases and 439,661 controls from UK biobank (Table 1) [10]. Individuals with gastric ulcer, duodenal ulcer, other site peptic ulcer, or astro-jejunal were defined as PUD cases [10].

Table 1 Detailed information on data sources
Full size table

Primary analyses were performed using genome-wide significant (P < 5 × 10–8) single nucleotide polymorphisms (SNPs) identified from the largest GWAS that combined data from UK biobank and the Genetic Investigation of Anthropometric Traits (GIANT) consortium, including up to 694,649 individuals of European descent (Table 1) [11]. We set a cut-off for minor allele frequency to > 1%, leaving 85,044 and 39,427 SNPs for BMI and WHR (unadjusted for BMI), respectively. To obtain valid IVs, these SNPs were then pruned at r2 < 0.001 across a window size of 10000 kb (based on the 1000 Genomes Project population [12]) to avoid bias of linkage disequilibrium. For those SNPs not available in the outcome dataset, proxies (r2 > 0.8) were found by searching the publicly available website (http://snipa.helmholtz-muenchen.de/snipa3/) (Additional file 1: Table S1). Finally, 531 and 343 SNPs were included for BMI and WHR, respectively (Additional file 1: Table S2 and 3). In addition, we calculated F-statistics to assess whether there was a weak IV bias (\(F={R}^{2}\frac{N-2}{1-{R}^{2}}\)). R2 refers to the percentage of the variation explained by SNPs and is calculated as described by Shim et al. [13]; N represents the total sample size [14].

However, there was a sample overlap (UK Biobank mainly) in the participants included in the primary analyses (Table 1). To test the robustness of the results, we then retrieved summary-level data for BMI and WHR from other GWASs where only individuals of the GIANT consortium were included (339,224 samples for BMI and 224,459 samples for WHR), so that there would be no sample overlap with the outcome data (Table 1) [15, 16]. These SNPs underwent similar quality-control steps, leaving 78 SNPs and 29 SNPs for BMI and WHR, respectively. Considering 2 SNPs for BMI not available in the outcome datasets with no suitable proxies found, both SNPs were excluded from the study. Here, we provided the characteristics of the remaining 76 SNPs and 29 SNPs in Additional file 1: Table S4 and S5, with R2 and F-statistics calculated as well.

In the reverse analyses, we extracted 8 independent SNPs as IVs for PUD using the above method. SNPs not found in the outcome datasets were replaced with their proxies if any (Additional file 1: Table S6). Detailed information for these SNPs were provided in Additional file 1: Table S7.

Mendelian randomization analyses

Inverse-variance weighted (IVW) in the multiplicative random-effects model was employed as the main method given the presence of heterogeneity among SNPs [17, 18]. A series of complementary analyses were conducted: The weighted median [19] and MR-Egger [20] provided more conservative causal estimates; and the MR-pleiotropy residual sum and outlier (MR-PRESSO) methods [21] was employed to identify pleiotropic outliers that may bias the results. Heterogeneity among IVs was measured by the heterogeneity Q test and I2 statistics. Horizontal pleiotropy was assessed using the MR-Egger regression intercept [22].

Since BMI and WHR are correlated covariates to each other, collider bias is likely to affect our results. To avoid this problem, we performed the multivariable MR method [23] to adjust BMI for WHR, and likewise, to adjust WHR for BMI. In addition, we included Helicobacter Pylori (H.P.) infection, type 2 diabetes (T2D), dyslipidemia, smoking, and alcohol use in multivariable MR to determine whether the causal association between adiposity and PUD, if any, was mediated by these traditional risk factors. Summary statistic for H.P. infection was derived from UK Biobank database (http://www.nealelab.is/uk-biobank, ID: ukb-b-531), T2D from the GWAS conducted by DIAGRAM consortium (452,244 individuals; 81,412 cases) [24], circulating lipid levels (LDL-C, HDL-C, TC, and TG) from the Global Lipids Genetics Consortium (188,578 individuals) [25], and smoking and alcohol use from the GWAS & Sequencing Consortium of Alcohol and Nicotine use (1,232,091 individuals) [26].

We calculated the statistical power using a publicly available tool (https://shiny.cnsgenomics.com/mRnd/) based on a type 1 error of 5% [27]. There was over 80% power to detect an OR of 1.09 for the BMI-PUD association, and 1.12 for the WHR-PUD association in the primary analyses; an OR of 1.15 for the BMI-PUD association, and 1.25 for the WHR-PUD association in the replication analyses. The estimates were considered significant at P < 0.025 (0.05/2 exposures). MR analyses were performed using packages including TwoSampleMR [28], MendelianRandomization [29], and MR-PRESSO [21] within software R (version 4.1.0).

Results

All IVs in the present study had F-statistics above the threshold of 10, suggesting sufficient IV strength for MR analyses. These SNPs were estimated to account for 5.51% and 3.43% of the phenotypic variation of BMI and WHR, respectively (Additional file 1: Table S2 and S3). In the replication analyses, genetic variants IVs explained 2.32% and 0.78% of the variation of BMI and WHR, respectively (Additional file 1: Table S4 and S5). The selected SNPs for PUD were estimated to account for 2.32% of the variation of PUD (Additional file 1: Table S7).

For per SD increase in genetically determined BMI (~ 4.8 kg/m2) there was 34% increased odds of PUD (95% confidence interval [CI], 25–44%; P = 9.72 × 10–16; Fig. 1 and Additional file 1: Fig S1). In an analogous analysis, we observed that each 0.09 ratio higher WHR was associated with a 35% increase in risk for PUD (95% CI, 23–49%; P = 2.35 × 10–10; Fig. 1 and Additional file 1: Fig S1). The results remained broadly consistent in the complementary analyses such as weighted median and MR-Egger regression (Fig. 1). There were 3 outliers for WHR (rs1680490, rs668871, and rs6861681) identified by the MR-PRESSO method. The results did not substantially change after correcting for these outliers (Fig. 1). Importantly, little evidence for heterogeneity and horizontal pleiotropy was found (Additional file 1: Table S8).

Fig. 1
figure 1

Associations of genetically determined BMI and WHR with PUD risk. BMI: body mass index; WHR: waist-to-hip ratio; PUD, peptic ulcer disease; SNP, single nucleotide polymorphism; OR, odds ratio; CI, confidence interval; IVW (mre), multiplicative random-effects inverse-variance weighted; MR-Egger, Mendelian Randomization-Egger; MR-PRESSO, MR-Pleiotropy Residual Sum and Outlier. Datasets for BMI and WHR were extracted from GWAS conducted by Pulit et al. *Excluding the outliers for WHR (rs1680490, rs668871, and rs6861681)

Full size image

Multivariable MR demonstrated that the effect of BMI on PUD was attenuated upon adjustment for WHR (Table 2). Nevertheless, the causal association of WHR with PUD persisted after adjusting for BMI (Table 2). Besides, both BMI-PUD and WHR-PUD associations remained consistent following adjustment for genetically determined H.P. infection, T2D, circulating lipid levels, smoking, or alcohol use (Table 2). Furthermore, the results were robust under adjustment for all these lifestyle factors (Table 2).

Table 2 Multivariable Mendelian randomization for the associations between obesity traits and PUD adjusting for potential mediators
Full size table

In the replication analyses, we used IVs based on summary-level data for both obesity traits from individuals included in the GIANT consortium [15, 16]. IVW analysis showed that genetically-predicted higher BMI (per SD) was associated with an 17% increase in risk for PUD (95% CI = 5–31; P = 0.006; Fig. 2 and Additional file 2: Fig S2). The MR estimate for WHR-PUD association was also positive and, notably, of a greater magnitude; Genetically-predicted higher WHR was associated with 35% higher risk of PUD (95% CI = 13–61%; P = 0.001; Fig. 2 and Additional file 2: Fig S2). Complementary analyses returned broadly consistent results (Fig. 2 and Additional file 2: Fig S2). No pleiotropic outliers were detected by the MR-PRESSO test. The risk of heterogeneity and horizontal pleiotropy should be low according to multiple tests listed in Additional file 1: Table S9.

Fig. 2
figure 2

Replication analyses showing the associations of genetically determined BMI and WHR with PUD risk. BMI: body mass index; WHR: waist-to-hip ratio; PUD, peptic ulcer disease; SNP, single nucleotide polymorphism; OR, odds ratio; CI, confidence interval; IVW (mre), multiplicative random-effects inverse-variance weighted; MR-Egger, Mendelian Randomization-Egger; MR-PRESSO, MR-Pleiotropy Residual Sum and Outlier. Datasets for BMI and WHR were extracted from GWAS conducted by Locke et al. and Shungin et al., respectively

Full size image

Reverse MR analyses provided no evidence for the causal effect of PUD on BMI or WHR (Fig. 3 and Additional file 2: Fig S3). The association pattern persisted when different datasets for BMI and WHR were used (Fig. 3 and Additional file 2: Fig S4). Besides, complementary analyses yielded broadly consistent results (Additional file 1: Table S10). SNP heterogeneity was high for BMI and WHR in the original analyses but not in the replication analyses. (Additional file 1: Table S11 and S12). MR Egger regression indicated a low risk of pleiotropy in both the original and replication analyses (Additional file 1: Table S11 and S12).

Fig. 3
figure 3

Associations of genetically determined PUD with risk of BMI and WHR. BMI: body mass index; WHR: waist-to-hip ratio; PUD, peptic ulcer disease; SNP, single nucleotide polymorphism; OR, odds ratio; CI, confidence interval

Full size image

Discussion

This bidirectional MR study demonstrated that genetically predicted obesity (notably abdominal obesity) was causally associated with PUD risk. The results were consistent across complementary analyses and survived adjustment for several traditional risk factors. Besides, no evidence was found to support the causal effect of PUD on obesity.

The association between obesity and PUD remain a subject of ongoing debate. Generally speaking, PUD is divided into gastric ulcers (GU) and duodenal ulcers (DU). Our results collaborated with a series of previous observational studies. Evidence from a cross-sectional study reported a positive association between obesity and GU (but not DU) (OR = 4.15; 95% CI, 1.31–13.13) among northern Sweden individuals [7]. A retrospective cohort study enrolling 32,472 Korean individuals found a pattern of relationships of higher BMI with increased risk of GU (OR, 1.32; 95% CI, 1.16–1.49; P < 0.001), but not with the risk of DU [8]. Likewise, another large prospective cohort study with 47,120 U.S. individuals reported an increased prevalence of GU in subjects with higher BMI upon multivariate adjustment (OR, 1.83; 95% CI, 1.20–2.78; P < 0.01) [5]. Similarly, the same research group showed that BMI was not associated with a greater tendency in suffering from DU [5]. In this study, we provided evidence supporting the positive association of obesity with PUD risk in European populations and demonstrated a potential causality of this relationship. Considering that significance was observed only in the BMI-GU relationship in previous studies, we speculated that the positive results observed in this MR study may be largely attributable to the fact that BMI increased the risk of GU, but not DU. Future MR study using separate GWAS datasets for GU and DU should shed light on this important issue.

The multivariable MR study suggested a great attenuation of the association between BMI and PUD after adjusting for WHR. Indeed, BMI is considered an imprecise obesity classification method since it ignores the distribution of adiposity and cannot distinguish between lean body mass and fat mass [30, 31]. Therefore, it may be a relatively poor tool for exploring associations between obesity and diseases. On the other hand, our results demonstrated that the association between WHR and PUD was not significantly attenuated after adjusting for BMI. As previously reported, WHR is an independent biological tool for abdominal adiposity and visceral fat measurement [32]. The results from the multivariable MR specifically demonstrated that the causal association between WHR and PUD was independent of BMI. In other words, individuals with higher abdominal adiposity are more likely to suffer from PUD, even if their BMI is relatively low.

The pathophysiology of PUD has been associated with the use of NSAIDs and Helicobacter Pylori infection. However, the underlying mechanisms linking obesity to PUD is not yet well-elucidated. Observationally, obesity was strongly associated with PUD in non-NSAIDs/aspirin users and Helicobacter Pylori-negative subjects [5]. Therefore, the observed causal effect of obesity on PUD is likely to be independent of anti-inflammatory drug use or Helicobacter Pylori infection. Recent studies have linked obesity to mucosal dysfunction [33, 34], which is one of the potential mechanisms associated with PUD [35]. In our study, we performed multivariable MR analyses to explore potential mediators. Obesity has been associated with metabolic diseases like T2D [36] and dyslipidemia [37], and unhealthy behaviors like smoking [38] and alcohol use [39]. However, this MR study demonstrated that the association of obesity with PUD risk persisted upon adjustment for these risk factors. In addition, obesity-induced alteration in gut microbiota [3, 4] and gastrointestinal inflammation [40] are also likely to mediate the causal associations. However, detailed research is insufficient. Further studies into how obesity is involved in the development of PUD are warranted.

There are three issues that can violate the MR assumption and lead to biased causal inferences: (1) biological mechanism; (2) genetic coinheritance; and (3) population effects. Pleiotropy represents one of the most important biological mechanisms. In the present study, MR-Egger intercept test was carried out to assess pleiotropy. As shown in Additional file 1: Table S8, S9, S11 and S12, there was no evidence for pleiotropy for all associations considered. Secondly, we ruled out the effect of non-Mendelian inheritance by performing the clumping process with R2 threshold of < 0.001 and window size of 10,000 kb. Thereby, the influence of genetic coinheritance would be minimal in this study. Third, the population stratification is unlikely to bias our results since all of the participants involved in these original GWASs were of European ancestry.

The MR design that we used is the first strength of this study, which can minimize biases such as residual confounders and reverse causation. Reverse causation describes a scenario where post-event measurement of the exposure can be affected by the outcome event. For example, one can hardly tell whether event A cause higher risk of event B or event B increased incidence of event A in the observational studies. The MR approach can avoid reverse causation because genetic variants were randomly assorted during conception and unlikely to be influenced by disease status. Besides, we here performed reverse MR analyses from PUD to obesity and no evidence was found for the causal association in this direction, which further minimized the influence of reverse causation. Second, the present study leveraged a high statistical power (above the threshold of 80%) enabled by the use of the largest GWAS meta-analyses to date. Third, all the complementary approaches returned consistent results, strengthening the causal inference. Finally, the analyses were less likely to be affected by the bias of population structure since they were restricted to individuals of European descent.

However, several limitations deserved consideration. First, we did not evaluate the associations between obesity and different subtypes of PUD (GU and DU) due to a lack of related GWASs. Second, the potential non-linear association cannot be assessed since this study relies on summary statistics. Third, sample overlap in the GWASs of obesity and PUD (UK Biobank mainly) are likely to bias the causal estimates and inflate Type 1 error rates [41] in the primary analysis. However, given that all genetic variants that we used were confirmed to be strong (F statistics > 10), we did not expect substantial bias here [41]. In addition, the results were consistent in the replication analyses where no sample overlap was present between the exposure and outcome. Finally, the limitation of participants in this study to Europeans might limit the generalizability of this study.

Conclusions

In the present study, we provided genetic evidence showing that obesity (notably abdominal obesity) is causally associated with increased PUD risk. Programs aimed at weight loss may play an important role in preventing PUD.