Evidence of shared genetic factors in the etiology of gastrointestinal disorders and endometriosis and clinical implications for disease management

Summary

In clinical practice, the co-existence of endometriosis and gastrointestinal symptoms is often observed. Using large-scale datasets, we report a genetic correlation between endometriosis and irritable bowel syndrome (IBS), peptic ulcer disease (PUD), gastro-esophageal reflux disease (GORD), and a combined GORD/PUD medicated (GPM) phenotype. Mendelian randomization analyses support a causal relationship between genetic predisposition to endometriosis and IBS and GPM. Identification of shared risk loci highlights biological pathways that may contribute to the pathogenesis of both diseases, including estrogen regulation and inflammation, and potential therapeutic drug targets (CCKBR; PDE4B). Higher use of IBS, GORD, and PUD medications in women with endometriosis and higher use of hormone therapies in women with IBS, GORD, and PUD, support the co-occurrence of these conditions and highlight the potential for drug repositioning and drug contraindications. Our results provide evidence of shared disease etiology and have important clinical implications for diagnostic and treatment decisions for both diseases.

Graphical abstract

Highlights

• •Endometriosis and gastrointestinal disorders are genetically correlated
• •MR analyses support a causal relationship between endometriosis and IBS and GPM
• •Shared risk loci highlight biological pathways and therapeutic drug targets
• •Potential for drug repositioning and contraindications between diseases

Yang et al. provide multiple levels of evidence supporting shared etiological factors between endometriosis and gastrointestinal disorders and highlight target genes and pathways contributing to the shared etiology. The results suggest potential targets for treatment, considerations for disease management, and caution around drug contraindications for both diseases.

Introduction

Endometriosis is a common gynecological disease affecting around 11% of reproductive aged women, significantly impacting their quality of life, work productivity and, in some cases, fertility.^1,2 The clinical manifestations of endometriosis are diverse. Many of the symptoms are non-specific, which preclude timely diagnosis and further prognosis.^1,3 It has often been observed that many women diagnosed with endometriosis also experience symptoms associated with gastrointestinal (GI) disorders including abdominal pain, bloating, constipation, heartburn, dyspepsia, vomiting, painful bowel movements, diarrhea, and nausea.^4,5,6 Studies have shown that while these symptoms do not necessarily involve bowel lesions associated with endometriosis, symptoms such as cyclic-related bloating, constipation, and diarrhea can get worse during menstruation.^5,6,7,8

GI symptoms, similar to those described in endometriosis, are also commonly associated with irritable bowel syndrome (IBS), peptic ulcer disease (PUD), gastro-esophageal reflux disease (GORD), and inflammatory bowel disease (IBD). IBD affects around 0.84% of the population while the other three disorders affect 5%–22% of the population.^{9,10,11,12,13} IBS, a chronic disorder of bowel function, is most common in women and young people and manifests as abdominal pain or discomfort that occurs in association with a change in bowel habits.¹⁴ GORD occurs when there is a reflux of stomach acid, and/or nonacidic reflux, into the esophagus which could result in erosion of the esophagus and associated heartburn and upper abdominal pain.^15,16 PUD is often defined as a mucosal break (ulcer) greater than 3–5 mm in the stomach or upper portion of small intestine which can cause abdominal pain and bloating.¹⁷ IBD mainly describes two conditions, Crohn’s disease and ulcerative colitis, which are characterized by chronic inflammation of the GI tract associated with symptoms such as abdominal pain, diarrhea, and bloody stools.¹⁸ Inevitably, the overlap in symptomology between endometriosis and these GI diseases presents challenges for clinicians to accurately diagnose these conditions in women. Understanding shared disease etiology critically impacts both disease diagnosis and management.

Previous observational studies have provided some evidence for associations between endometriosis and digestive disorders. Meta-analyses have reported a 3-fold increase in the prevalence of IBS in women with endometriosis compared with women without endometriosis.^19,20 This was supported by a recent retrospective study using a large nationwide biobank-based cohort, the Estonian Biobank (EstBB), which reported a notable proportion of women diagnosed with endometriosis (n = 7,142) also suffered from IBS (13.6%) and conversely a proportion of women with IBS (n = 10,781) were also diagnosed with endometriosis (9.0%).²¹ A nationwide Danish cohort study found a significantly increased risk of IBD in endometriosis patients with a standardized incidence ratio of 1.5 (95% confidence interval [CI], 1.4–1.7), and the relationship became stronger when restricted to surgically confirmed endometriosis.²² Currently, few observational studies have investigated the association between endometriosis and other GI disorders. However, endometriosis symptoms overlap with other common GI disorders, including PUD and GORD.²³

Despite studies showing that endometriosis patients are more likely than people without endometriosis to present with GI symptoms or have a diagnosis of GI disorders, it is uncertain whether this is due to (1) direct effects of one disease on the other, i.e., endometriosis causing GI symptoms and disorders and vice versa, (2) shared etiological factors between endometriosis and GI diseases, i.e., confounding factors, (3) side effects of medical treatments, or (4) the bias in observational studies, such as misdiagnosis, and other sources of residual or unmeasured confounding.^24,25 For example, therapeutic use of gonadotropin-releasing hormone analogs and nonsteroidal anti-inflammatory drugs (NSAIDs) to manage symptoms of endometriosis has been reported to aggravate the severity of GI symptoms and contribute to GI disorders including PUD.^{6,26,27,28,29} Overlapping symptoms, symptoms-based diagnosis of IBS, and inclusion of self-reported illness may lead to misdiagnosis, a common source of confounding in observational studies.¹⁹

The application of genetic data in a Mendelian randomization (MR) framework provides a valuable approach to understand shared disease etiology. Genetic variants are randomly determined at conception and MR analysis uses exposure-associated single-nucleotide polymorphisms (SNPs) as instrumental variables that can minimize some forms of bias that weaken results obtained from observational approaches and can also infer causal relationships between two conditions.^24,30 Endometriosis and GI diseases are common multifactorial diseases with environmental and genetic risk factors both playing roles in the development of these diseases.^31,32 Twin and family studies have shown a heritable component to endometriosis, IBS, GORD, PUD, and IBD, with reported heritability estimates of 51%, 31%, 57%, 62%, and 70%, respectively,^{33,34,35,36,37} and genome-wide association studies (GWASs) have identified independent genetic risk loci for endometriosis and the four GI disorders. GORD/PUD medicated (GPM) is a recently constructed phenotype from a combination of disease-diagnosis for GORD and/or PUD and/or corresponding use of medications for each disease (GPM),^38,39,40 since both GORD and PUD are acid-related diseases and the medications for PUD also have a therapeutic effect on GORD in clinical practice. GWAS data are available for the GPM phenotype first described by Wu et al.,³⁸ and referred to as PG + M in that study.

Using genetic data, Adewuyi et al.⁴¹ found evidence of causal links between endometriosis and depression with gastric mucosa abnormalities. However, the relationship between endometriosis and other common GI disorders, such as IBS, PUD, and IBD, as well as the newly defined GPM phenotype have not been investigated in detail. This study presents a comprehensive evaluation of the relationship between endometriosis and GI disorders through analysis of large-scale genetic datasets, and epidemiological and pharmaceutical data in the UK Biobank (UKB) and the Australian Longitudinal Study on Women’s Health (ALSWH).^42,43

Results

Significant comorbid relationship between endometriosis and GI disorders

Data from 188,461 unrelated women in UKB was used to validate the co-occurrence of endometriosis and GI disorders. We found a comorbid relationship between endometriosis and each of the four GI disorders (IBS, GORD, PUD, and IBD) (Table 1). We also observed the co-occurrence of endometriosis with GPM, a more powerful and representative GI disorder phenotype that combined GORD and PUD and/or corresponding medication treatments due to the shared treatment therapy and high genetic correlation between the two disorders (genetic correlation [rg] = 0.65).³⁸ Women with endometriosis were two times more likely to have an IBS diagnosis (odds ratio [OR] = 2.01; 95% CI, 1.86–2.16; p value = 3.90e−68), and 1.4 times more likely to have a GORD diagnosis (OR = 1.40; 95% CI, 1.30–1.50; p = 3.54e−18) than those without a reported diagnosis of endometriosis (Table 1). Similarly, women with IBS were two times more likely to have an endometriosis diagnosis (OR = 2.13; 95% CI, 1.98–2.30; p = 7.14e−79), while women with GORD were 1.45 times more likely to have an endometriosis diagnosis (OR = 1.44; 95% CI, 1.35–1.57; p = 5.94e−22). Only IBS, GORD, and GPM remained significant after accounting for multiple tests. Results may be influenced by smaller sample sizes of unrelated women diagnosed with PUD and IBD when compared with other GI disorders in UKB. Alternatively, it might also suggest the comorbid relationship between endometriosis with PUD and IBD were not strong. A competitive comorbidity analyses further confirmed that women with IBS are more prone to be comorbid with endometriosis, followed by GORD, when compared with PUD and IBD (Figure S1).

Table 1.

Comorbid relationship between endometriosis and GI disorders in unrelated European women in the UK Biobank

	IBS (n = 16,330)	GORD (n = 22,383)	PUD (n = 5,208)	GPM (n = 36,535)	IBD (n = 2,708)
Endo vs. GI	2.01 (1.86–2.15) p = 3.90e−68	1.40 (1.30–1.50) p = 3.54e−18	1.22 (1.05–1.42) p = 0.01	1.29 (1.22–1.38) p = 6.51e−16	1.25 (1.01–1.53) p = 0.04
GI vs. Endo	2.13 (1.98–2.30) p = 7.14e−79	1.45 (1.35–1.57) p = 5.94e−22	1.22 (1.05–1.42) p = 0.01	1.37 (1.28–1.46) p = 1.60e−21	1.24 (1.0–1.53) p = 0.04

Endo, endometriosis; IBS, irritable bowel syndrome; GORD, gastro-esophageal reflux disease; PUD, peptic ulcer disease; GPM, a combination of GORD and PUD and/or corresponding medication treatments; and IBD, inflammatory bowel disease. Association between diseases was tested using the Fisher’s exact test. The first row (Endo vs. GI) represents Fisher’s exact results of whether women with endometriosis are more likely to have a diagnosis of GI disorders, while the second row (GI vs. Endo) represent Fisher’s exact results in a reverse direction where women with GI disorders are more likely to have a diagnosis of endometriosis. The numbers in the parentheses after each disease represent the number of women diagnosed with this disease. There are a total number of 5,392 women diagnosed with endometriosis. Fisher’s exact results are represented by OR, 95% CI values within parentheses and p values.

Genetic correlation between endometriosis and GI disorders

To establish if the relationship between endometriosis and GI diseases was in part driven by shared genetic risk factors, we estimated the genetic correlation between the disorders using GWAS data. Compared with the comorbid relationships identified above, the linkage disequilibrium score regression (LDSC) analysis provided evidence of a significant positive genetic correlation (rg) between endometriosis and IBS (rg = 0.22, p = 5.0e−3), GORD (rg = 0.16, p = 4.0e−3), PUD (rg = 0.23, p = 3.0e−3), and GPM (rg = 0.22, p = 2.17e−06) (Figure 1A). There was no evidence of significant correlation between endometriosis and IBD. Given that there is no sample overlap between endometriosis and the GI phenotype GWAS studies, we also constrained the intercept in the LDSC analysis, resulting in a smaller standard error with little change in the genetic correlation (Figure 1). Although endometriosis is rarely observed in men, men do carry endometriosis risk alleles, and we also conducted the genetic correlation analysis separately for women and men. The results remained significant between endometriosis and both the separate female (rg = 0.28, p = 3.0e−3) and male (rg = 0.21, p = 0.01) IBS GWAS cohorts (Figure 1B).

Figure 1.

Results of genetic correlation between endometriosis and GI disorders

(A) Genetic correlation between endometriosis and each of five GI disorders that combined male and female diagnosis.

(B) Genetic correlation between endometriosis, female, male, and combined irritable bowel syndrome (IBS). GORD, gastro-esophageal reflux disease; PUD, peptic ulcer disease; GPM, a combination of GORD and PUD and/or corresponding medication treatments; IBD, inflammatory bowel disease. The x axis indicates the value of genetic correlation, and the error bar indicates its 95% confidence interval. All red lines represent the results after constraining the heritability intercept to one considering no sample overlap for each comparison. Solid lines represent correlations with a significant p-value at a false discovery rate (FDR) < 0.05.

Complex causal relationship between endometriosis and GI disorders

Following the identification of shared genetic correlation, we applied a widely accepted MR method called generalized summary data-based MR (GSMR)⁴⁴ to estimate the causal relationship between GPM, IBS, and endometriosis, and results are summarized in Figure 2 and Table S1. We identified evidence of a significant association between GPM and endometriosis whereby genetic variants contributing to the risk of GPM (genetic predisposition to GPM) also increased risk of endometriosis (OR = 1.56; 95% CI, 1.35–1.76; p = 2.47e−5). The reverse MR analysis (the genetic effect of endometriosis on GPM) was not statistically significant (OR = 0.98; 95% CI, 0.94–1.02; p = 0.32). To test the reliability of this causal inference, we conducted additional MR analyses (inverse variance weighted [IVW]-MR, MR-Egger, weighted median, simple median and MR-PRESSO), and sensitivity tests (Table S2. Summary of MR results for endometriosis (Endo), irritable bowel syndrome (IBS), and GORD/PUD medicated (GPM) using MR-PRESSO, related to MR-PRESSO analysis in the STAR Methods, Table S3. Summary of MR results for endometriosis (Endo), irritable bowel syndrome (IBS), and GORD/PUD medicated (GPM) using additional MR approaches, related to IVW-MR, MR-Egger, weighted median and simple median analyses in the STAR Methods, Table S4. Summarized results of sensitivity tests for IVW-MR analysis, related to IVW-MR analysis in the STAR methods). We reported MR results generated using two GWAS thresholds, p < 5e−8 and p < 5e−6, for inclusion of variants as instrumental variables. There was evidence of pleiotropy and, following removal of outlier variants and reanalysis using MR-PRESSO, evidence for a causal relationship between GPM (exposure) and endometriosis (outcome) (OR = 1.370; 95% CI, 1.117–1.623; p = 0.029) was significant supporting results of the GSMR analysis. Results from several other MR approaches were not significant, but effect sizes were similar for MR approaches including GSMR, IVW-MR, and MR-PRESSO.

Figure 2.

Simplified causal relationship identified by GSMR

Different arrow color represents the specific direction of causal relationship. Odds ratio (OR), p value (p) and the number of SNP instruments (nSNP) from the GSMR (generalized summary data-based Mendelian randomization) analysis are shown on the arrow. Created with BioRender.com.

We found evidence of bidirectional causal relationships between IBS and endometriosis. A bidirectional causal relationship occurs when there is a significant causal relationship between two traits A and B, both when A is set as the exposure and when B is set as the exposure. Using endometriosis as the exposure there was evidence that genetic variants contributing to the risk of endometriosis had a small effect on risk for IBS (OR = 1.07; 95% CI, 1.01–1.13; p = 0.04). Due to the limited number of SNP instruments available (nSNP = 1) for IBS when using a genome-wide significant level (p < 5e−8), we used a relaxed GWAS threshold (p < 5e−6) to estimate the effect of IBS on endometriosis risk. Results showed that genetic variants that increased risk of IBS had a significant risk effect on endometriosis (OR = 1.15; 95% CI, 1.03–1.28; p = 0.03)(Figure 2). These results were supported by other MR approaches (Tables S2 and S3). Sensitivity analyses indicated that all SNP instruments were valid (F statistics >10) and there was no evidence of pleiotropy (Table S4).

To evaluate whether the causal associations identified by MR analyses were impacted by a potential confounder, a risk factor associated with both exposure and outcome through SNP instruments, we used the mtCOJO corrected GWAS summary statistics to perform the MR analysis. IBS and GPM were previously reported to be genetically correlated by Wu et al.³⁸ There was a strong bidirectional causal association with genetic predisposition to IBS having a causal effect on GPM (OR = 1.20; 95% CI, 1.15–1.24; p = 1.93e−14) and genetic predisposition to GPM having a causal effect on IBS (OR = 1.34; 95% CI, 1.28–1.40; p = 7.29e−23). In the MR analysis with endometriosis, both GPM and IBS may act as a confounder due to their causal association with each other (Figure 2). To avoid the potential effect of GPM on the relationship between endometriosis and IBS, we adjusted the GWAS data of both endometriosis and IBS for the effects of GPM using mtCOJO to identify disease-specific variant associations independent of GPM. Following this conditional analysis, evidence for the bidirectional causal relationship between IBS and endometriosis remained (Table S1; Figure S2), and was also supported by multivariable MR (MVMR), another approach to adjust for the potential confounding effects (Table S5). The causal relationship between GPM and endometriosis, following adjustment for the genetic effects of IBS, had similar effect size when using both GWAS thresholds (p < 5e−8: OR = 1.22; 95% CI, 1.00–1.44; p = 0.07) (p < 5e−6: OR = 1.16; 95% CI, 1.05–1.27; p = 0.01) (Table S1; Figure S2). However, the causal association for GPM and endometriosis was not significant after adjusting for the confounding effect from IBS when using MVMR (Table S5). One reason for the inconsistent results between mtCOJO and MVMR was the limited power of the IBS GWAS. The MVMR framework only used a single SNP instrument for IBS (Nsnp = 1) to estimate the causal effect on endometriosis and then used this predicted value to estimate the direct effect of exposure on outcome in a multivariable regression analysis. If the predicted value in the first step was not accurate, the final estimate would probably be biased.

Finally, MR was run separately for GORD and PUD to test if the relationship between GPM and endometriosis was driven by one particular phenotype. There was evidence of a significant relationship between GORD and endometriosis, which remained when conditioning both traits on PUD (OR = 1.15; 95% CI, 1.10–1.21; p = 1.69e−6); however, there was no evidence of a causal relationship between PUD and endometriosis (Table S6). This might be due to the small number of SNP instruments in the current PUD GWAS because of the smaller sample size or it may reflect a true absence of a causal association between the two conditions.

Genomic loci associated with both endometriosis and GI disorders

To identify risk loci associated with both endometriosis and IBS or GPM, we carried out a cross-trait meta-analysis using two complementary methods, MetABF and the Eskin random-effects model (RE2C). SNPs were considered associated with both diseases at a genome-wide significant level if they had a logABF > 4 in MetABF and p < 5e−8 in RE2C models and p < 0.05 in each individual GWAS analysis. As a result, a total of 477 SNPs met criteria for endometriosis and GPM, while only 32 SNPs were significant for endometriosis and IBS. Using FUMA to analyze genome-wide significant SNPs, 12 genomic risk loci (21 independent signals) were identified as significantly associated with both endometriosis and GPM and 3 with endometriosis and IBS (Table 2). Among those loci identified by the cross-trait meta-analysis, the SNP on chr2:67845739 (rs2861694) within ETAA1 was previously reported as associated with both endometriosis and GPM, another five SNPs were significantly associated with either endometriosis or GPM. The remaining nine risk loci have not been previously identified at a genome-wide level of significance for endometriosis and GPM and IBS (Table 2).

Table 2.

Significant SNP loci identified by endometriosis, IBS, and GPM cross-trait meta-analysis

rsID	Nearest gene	Chr	BP	Beta		p Value		Meta
				Endo	GI	Endo	GI
Endometriosis and IBS
rs12407439	LINCO1635	1	22,347,396	−0.14	−0.03	6.57e−08	2.50e−02	3.98e−08
rs6661808	MYSM1	1	59,089,534	−0.08	−0.07	2.00e−03	1.20e−06	7.67e−09
rs1250244	FN1	2	216,297,796	0.10	0.03	8.73e−08	1.00e−03	9.09e−10
Endometriosis and GPM
rs7515106	WNT4	1	22,473,410	−0.10	0.01	1.89e−09	4.50e−02	5.93e−09
rs7547294	PDE4B	1	66,351,735	−0.03	−0.03	3.50e−02	2.40e−07	2.53e−08
rs11675830	ETAA1	2	67,776,860	0.07	0.02	2.57e−06	9.60e−06	5.41e−10
rs4260227	ETAA1	2	67,843,537	0.07	0.04	2.21e−05	2.10e−09	1.20e−12
rs2861694	ETAA1	2	67,845,739	0.07	0.04	7.77e−07	1.00e−10	1.05e−15
rs13031614	SPAG16	2	215,068,514	0.03	0.03	2.20e−02	4.20e−07	3.06e−08
rs7613360	CAMKV	3	49,916,710	−0.05	−0.03	1.00e−03	7.40e−07	7.61e−09
rs2008877	SEMA3F-AS1	3	50,162,291	0.04	0.03	2.00e−03	6.50e−08	8.00e−10
rs2526388	SEMA3F-AS1	3	50,174,886	0.03	0.03	4.60e−02	1.70e−07	2.20e−08
rs1046953	SEMA3F	3	50,197,097	0.04	0.03	3.00e−03	3.00e−08	5.15e−10
rs12631337	SEMA3F	3	50,198,537	0.05	0.03	1.00e−03	1.50e−10	7.38e−13
rs71557318	HIST1H2BC	6	26,118,570	0.05	−0.03	2.40e−02	1.30e−09	1.03e−09
rs10811669	CDKN2B-AS1	9	22,172,893	−0.08	−0.01	1.68e−08	3.00e−02	2.04e−08
rs10500661	CCKBR	11	6,273,744	0.05	−0.04	6.00e−03	9.70e−12	2.63e−12
rs1479406	RERG	12	15,387,543	−0.06	−0.03	9.24e−06	8.80e−07	1.06e−10
rs11056461	PTPRO	12	15,558,466	−0.08	−0.02	2.12e−06	4.80e−04	2.66e−08
rs3803042	MIR196A2	12	54,387,947	−0.06	−0.02	4.54e−06	1.00e−04	1.65e−08
rs11170785	HOXC8/9/-AS1	12	54,407,290	−0.06	−0.02	3.46e−05	8.20e−06	3.25e−09
rs736825	HOXC4/5/6	12	54,417,576	−0.08	−0.03	1.98e−05	3.00e−06	1.53e−09
rs773111	RAB5B	12	56,375,740	0.03	0.03	2.80e−02	8.10e−08	6.80e−09
rs9807058	LINC01982	17	50,338,523	−0.04	−0.03	5.00e−02	1.50e−07	2.18e−08

Chr, chromosome; BP, base pair; Endo, endometriosis; GI, gastrointestinal; Meta, p-value from cross trait meta-analysis; IBS, irritable bowel syndrome; GPM, a combination of GORD and PUD and/or corresponding medication treatments.

Evidence of shared causal variants between endometriosis and GI disorders

A pairwise GWAS (GWAS-PW) analysis was used to perform a colocalization analysis to assess if any of the genomic regions jointly affected endometriosis and IBS. Results identified three regions likely containing the same (pleiotropic) causal variant for both traits (PPA3 > 0.5, Table S7) and three regions containing distinct and independent causal variants for endometriosis and IBS (PPA4 > 0.5). Analyzing endometriosis and GPM, we identified six regions likely containing the same (pleiotropic) causal variant for both traits (PPA3 > 0.5, Table S7) and 60 regions with independent causal variants for endometriosis and GPM (PPA4 > 0.5). None of the regions with likely pleiotropic causal variants for endometriosis and GPM overlapped with the three regions identified between endometriosis and IBS. Among these identified regions, two loci (MYSM1 and FN1) with the highest probability of a shared causal variant for endometriosis and IBS and the three loci (ETAA1, HOXC gene cluster, and RERG) with the highest probability of a shared causal variant for endometriosis and GPM were also identified by the cross-trait meta-analysis described above. Specifically, the region with the strongest pleiotropic effect on endometriosis and GPM is located near ETAA1/LINCO1812 on chromosome 2 (Figure 3), SNPs in this region are significantly associated with these two diseases and the index SNPs in the individual GWAS studies are in strong linkage disequilibrium (r² = 0.88). Other regions identified by both cross-trait meta-analysis and GWAS-PW are shown in Figure S3.

Figure 3.

Mirror plot of the first two GWAS-PW regions containing same causal variants for endometriosis and GPM

The left is the genomic locus near ETAA1 on chromosome 2, while the right is the locus around HOXC4 on chromosome 12.

Gene mapping and functional annotation of shared risk loci

To understand the potential regulatory function of identified risk loci and prioritize candidate genes for endometriosis, IBS, and GPM, we conducted two functional annotation analyses including EpiMap⁴⁵ and summary data-based MR (SMR)⁴⁶ (see details and criteria in STAR Method). EpiMap used epigenome maps from both reproductive and GI tissues to perform epigenome annotations for the genome-wide significant SNPs from the cross-trait meta-analysis in this study. SMR analyses were performed to test for causal or pleiotropic associations between gene-level expression and a disease trait by integrating GWAS signals with genetically regulated gene-level expression in 12 digestive and reproductive tissues in the GTEx project⁴⁷ and endometrium.⁴⁸ As a result, for endometriosis and IBS, we identified that two genes, MYSM1 and FN1, which were the nearest genes to independent genome-wide significant SNPs, have functional evidence from either of the SMR or EpiMap analyses. SMR analysis identified that variants associated with both endometriosis and IBS modified MYSM1 expression in sigmoid colon (p_smr < 0.05 and p_HEIDI > 0.05), demonstrating that it may contribute to the risk of both endometriosis and IBS mediated through MYSM1 expression in this tissue. EpiMap identified one significant SNP in the endometriosis and IBS GWAS meta-analysis located in a predicted enhancer region in colon and esophagus with the target gene being FN1. Both FN1 and MYSM1 loci were also prioritized by GWAS-PW. Similarly, using EpiMap we identified five loci containing genome-wide significant SNPs from the endometriosis and the GPM GWAS meta-analysis located in predicted enhancers in both reproductive and/or digestive tissues and these enhancers were predicted to target the nearest genes. These five genes are SEMA3F, RAB5B, HOXC gene cluster (HOXC4/5/6/8/9/10/-AS1), RERG, and PTPRO with the last three also located in genomic regions that were identified through GWAS-PW. SMR analysis did not identify any significant colocalization of GWAS and eQTL signals in these regions.

Additional phenotypes associated with shared risk loci

We next investigated whether there were previously reported trait associations with SNPs related to risk of both endometriosis and GI diseases through PhenoScanner V2⁴⁹ and the GWAS Catalog.⁴⁹ Interestingly, our results (Table S8) identified that fat and estrogen-related traits (BMI, body fat percentage, waist circumference, hip circumference, WHR, weight, age at first birth, and age at menarche) are associated with six (WNT4, SEMA3F, HIST1H2BC, RERG, RAB5B, and HOXC4/5/6) of the 12 regions shared between endometriosis and GPM, suggesting those regions identified might contribute to the risk of both endometriosis and GPM through the dysregulation of estrogen and inflammation.

Potential for drug repositioning

Identification of potential target genes contributing to both endometriosis and GI disorders may also highlight possible drug targets and opportunities for drug repositioning. We searched the online open-target drug platform⁵⁰ for all known endometriosis and GI disorder (GORD, PUD, IBS) drug targets. GPM was not included because it is a combined GORD and PUD phenotype with limited information about drug targets in the drug platform. A total of 34 genes with encoded proteins were targets of both endometriosis and IBS/GORD/PUD drugs (Table S9). Restricting the analysis to the genes nearest to independent SNPs from the cross-trait meta-analysis (Table 2), one gene, CCKBR, encoded a protein that was targeted by two drugs Proglumide (ATC code: A02BX06) and Netzepide (NCT01298999 and NCT02597712) for the clinical treatment of GORD and PUD. In addition, PDE4B with the encoded protein cAMP-specific 3′,5′-cyclic phosphodiesterase 4B was targeted by Pentoxifylline, which acts as an inhibitor targeting the immune system and has been clinically trialled for the treatment of both endometriosis (phase III) and IBS (phase IV) separately (Table S10). Compared with IBS, which has multiple sources of evidence to support the promising treatment effect of Pentoxifylline,^51,52 there is limited evidence on whether Pentoxifylline impacts endometriosis related pain reduction.⁵³

Insights from medication use in both diseases

Patterns of medication usage may reflect the underlying patterns of health conditions and has been suggested as a useful tool to characterize comorbidity in a population.^54,55,56 Moreover, as mentioned previously, the use of NSAIDs for the treatment of endometriosis is also a known risk factor for PUD.²⁶ Therefore, we investigated prescription medication use in women by endometriosis status using data from the Australian Pharmaceutical Benefits Scheme (PBS) records for both the 1973–78 and the 1989–95 ALSWH cohorts.^42,43 Interestingly, medication use for PUD and GORD were within the top 10 most frequently used medications in both cohorts and the frequency of PUD and GORD medication use was significantly higher in the 1989–95 ALSWH cohort in women with endometriosis compared with those without endometriosis following correction for multiple testing (Table S11), further evidence of the likely co-occurrence of the diseases and disease symptoms. Using age-matched medication data of unrelated European women within the UK biobank revealed that, in addition to the expected hormonal therapies and NSAIDs, up to seven medications for treatments of GORD, PUD, and IBS were also significantly higher in women with endometriosis compared with women without a diagnosis for endometriosis (Tables 3 and S12). Comparing medication usage between women with and without GI disorders in the UK biobank, we also found a significantly higher use of hormone therapies among women diagnosed with IBS, GORD, and PUD, but not for IBD (Tables 3 and S13), consistent with results from the comorbidity and genetic analyses suggesting a relationship between endometriosis and IBS, GORD, and PUD. It should be noted that many other health, lifestyle, and environmental factors may influence medication usage; however, patterns observed from the pharmaceutical data were consistent with relationships highlighted by the comorbidity and genetic analyses.

Table 3.

Comparison of medication usage in UKB unrelated European women with and without a diagnosis of endometriosis or GI disorders

Medication	Diagnosis		Controls			p Value	BF p value	Description
	Med	No	Med	No
Endometriosis
Omeprazole	540	4,852	5,521	86,823		1.49e−28	1.73e−25	GI disorder treatment
Laxatives	353	5,039	3,749	88,595		1.52e−16	1.76e−13	constipation treatment
Lansoprazole	266	5,126	2,881	89,463		6.96e−12	8.05e−09	GI disorder treatment
Mebeverine	73	5,319	558	91,786		3.84e−09	4.45e−06	IBS treatment
Ranitidine	163	5,229	1,765	90,579		1.06e−07	1.00e−04	GI disorder treatment
Senna	30	5,362	176	92,168		1.25e−06	1.50e−03	constipation treatment
Esomeprazole	33	5,359	242	92,102		2.88e−05	3.34e−02	GI disorder treatment
Paracetamol	1,914	3,478	25,462	66,882		8.16e−35	9.45e−32	painkiller
Premarin	132	5,260	595	91,749		1.49e−33	1.73e−30	hormone therapy
Irritable bowel syndrome
Vagifem	200	15,679	758		126,274	2.29e−18	2.91e−15	hormone therapy
Premarin	185	15,694	891		126,141	2.31e−09	2.94e−06	hormone therapy
Estraderm	80	15,799	291		126,741	6.54e−09	8.31e−06	hormone therapy
Ovestin	51	15,828	151		126,881	1.59e−08	2.03e−05	hormone therapy
Estradiol product	99	15,780	430		126,602	2.33e−07	3.00e−04	hormone therapy
Evorel 25 patch	82	15,797	346		126,686	9.59e−07	1.20e−03	hormone therapy
Conjugated oestrogens	18	15,861	29		127,003	1.22e−06	1.60e−03	hormone therapy
Climaval 1mg tablet	65	15,814	258		126,774	2.08e−06	2.60e−03	hormone therapy
Estriol product	37	15,842	114		126,918	3.17e−06	4.00e−03	hormone therapy
Omeprazole	2 ,242	13,637	4,209		122,823	0	0	GI disorder treatment
Gastro-oesophageal reflux disease
Estradiol product	134	22,248	317		89,211	8.50e−07	1.00e−03	hormone therapy
Estraderm	94	22,288	217		89,311	1.81e−05	2.07e−02	hormone therapy
Premarin	233	22,149	672		88,856	2.44e−05	2.79e−02	hormone therapy
Lansoprazole	3,705	18,677	1,640		87,888	0	0	GI disorder treatment
Omeprazole	6,948	15,434	3,335		86,193	0	0	GI disorder treatment
Ranitidine	1,596	20,786	1,081		88,447	0	0	GI disorder treatment
Gaviscon liquid	714	21,668	237		89,291	1.01e−295	1.15e−292	GI disorder treatment
Paracetamol	7,900	14,482	21,440		68,088	1.92e−249	2.20e−246	painkiller
Esomeprazole	434	21,948	99		89,429	3.35e−205	3.83e−202	GI disorder treatment
Peptic ulcer disease
Estraderm	35	5,173	205		72,707	1.26e−05	1.35e−02	hormone therapy
Omeprazole	1,375	3,833	2,821		70,091	0	0	GI disorder treatment
Lansoprazole	711	4,497	1,426		71,486	2.34e−306	2.49e−303	GI disorder treatment
Ranitidine	351	4,857	942		71,970	3.67e−118	3.91e−115	GI disorder treatment
Paracetamol	1,963	3,245	17,305		55,607	1.74e−103	1.86e−100	painkiller
Esomeprazole	107	5,101	88		72,824	1.62e−72	1.73e−69	GI disorder treatment
Tramadol	217	4,991	665		72,247	1.10e−64	1.17e−61	opioid
Co-codamol	348	4,860	1,608		71,304	4.27e−64	4.55e−61	opioid
Amitriptyline	326	4,882	1477		71,435	1.03e−61	1.10e−58	antidepressants

Med, medication used; No, no medication used; BF, Bonferroni corrected p-value.

Discussion

Evidence for comorbid relationships

Endometriosis and IBS, two of the leading causes of chronic pelvic pain, are often misdiagnosed in clinics due to non-specific symptoms.⁵⁷ Several epidemiological studies indicate women with endometriosis have an increased risk of a diagnosis of IBS,^58,59 and two studies reported increased prevalence of endometriosis among IBS patients.^21,60 Our analysis using diagnostic reports in UKB supports the previous epidemiological association between endometriosis and IBS. In addition, we identified comorbid associations between endometriosis and other GI disorders including GORD, PUD and GPM, rarely investigated in previous epidemiological studies. Endometriosis showed a stronger relationship with IBS and GORD when compared with PUD. Therefore, in addition to the clinically reported shared symptomology, this study provides further evidence of a complex phenotypic association between endometriosis and GI disorders, including GPM, a more powerful and representative disease phenotype.

Evidence of shared genetic etiology and potential drug candidates

A previous study investigating the shared genetic basis between endometriosis and depression implicated gastric mucosa abnormalities in this casual pathway.⁴¹ Adewuyi et al. also reported a strong genetic correlation between endometriosis and GORD using two early published GWAS summary datasets in the UKB.⁴¹ Our study, which included one-third more individuals for the UKB, replicated the significant genetic correlation between endometriosis and GORD and extended this result, identifying significant genetic correlation between endometriosis and three other GI disease phenotypes: IBS, PUD, and GPM, but no evidence of genetic correlations with IBD. This is consistent with the known genetic similarities among the three GI disorders (IBS, GORD, and PUD), and differences between IBD and the other three GI disorders.³⁸ The combined GPM phenotype increased power for the GWAS, resulting in a stronger genetic correlation (smaller standard error) between endometriosis and the more powerful GPM phenotype.

We next conducted MR analysis to assess the causal relationship between endometriosis and GI disorders. Given that there is a greater number of genome-wide significant SNP instruments (n_SNP = 19) for GPM compared with GORD (n_SNP = 6) and PUD (n_SNP = 8), we focused on using GPM GWAS summary statistics for the MR analysis. We identified a significant causal association between GPM and endometriosis with evidence of bidirectional relationship between IBS and endometriosis. The directions and effect size estimates remained similar for other MR approaches tested, although not all results were statistically significant. A strength of our study was that we conducted conditional analyses using both mtCOJO and MVMR analyses, which take into account the correlation between traits given the previously reported complexity among GI disorders.³⁸ The MR results were consistent after conditioning on GPM, indicating that the bidirectional causal relationship between endometriosis and IBS at the genetic level were not driven by their relationship with GPM. Estimates for the effect of GPM risk SNPs on endometriosis were similar after adjusting for genetic effects of IBS using the two different significance thresholds applied for selecting IBS SNPs from the mtCOJO analysis. Application of a relaxed threshold can improve the statistical power for MR analysis,⁶¹ but can increase the likelihood of false positivity of SNP exposure associations and introduce possible pleiotropic effects.⁶² Effect size estimates were generally consistent for different MR approaches when using different GWAS thresholds. Most of MR results were not significant after using Bonferroni test to correct for multiple tests. Shared genetic risk factors between endometriosis and IBS and/or GPM, may partly explain the significant comorbid relationship between these diseases. However, further MR analysis should be conducted when more powerful GWAS datasets become available.

Identification of shared genomic loci from cross-trait meta-analysis and colocalization approaches provides clues as to the possible biological mechanisms and specific pathways driving the causal relationships between the different GI disorders and endometriosis. For example, MYSM1 is a region not previously implicated in IBS or endometriosis and SMR analysis in this study identified gene-level association between MYSM1 and both endometriosis and IBS. MYSM1 plays an important role in immune function and previous studies found downregulated gene expression of MYSM1 in endometriosis compared with controls.⁶³ Another locus, FN1 on chromosome 2 shared by endometriosis and IBS, is involved in various cellular processes including cell proliferation, motility, invasion, and migration,⁶⁴ processes reported to be associated with molecular mechanisms leading to endometriosis.⁶⁵ FN1 is a genetic risk factor for endometriosis^66,67 but not previously reported for IBS. The lead SNP in the FN1 region lies within a predicted enhancer or promoter in digestive tissues and there is significant downregulation of expression of FN1 in IBS patients compared with controls, which may be responsible for increased mucosal permeability and visceral hypersensitivity of IBS through its mediation in barrier dysfunction.^68,69

Three genomic loci shared by endometriosis and GPM were identified using both cross-trait meta-analysis and GWAS-PW, including ETAA1 on chromosome 2, HOXC gene cluster, and RERG on chromosome 12. ETAA1 is an activator of ATR kinase that plays a key role in protecting the genome against both intrinsic replication problems and substantial extrinsic DNA damage.⁷⁰ The ETAA1 locus has been previously reported to be associated with risk of both traits at the genome-wide significance level.^38,66 There is evidence linking HOXC and RERG to endometriosis and GI traits. For example, altered expression of the HOXC cluster has been found in ectopic and eutopic tissues from endometriosis patients compared with controls,^71,72,73 and in GI disorders such as colorectal cancer and gastric cancer.^74,75,76 Functional annotation of independent SNPs in the HOXC region also suggest that these variants impact regulatory elements in both digestive and reproductive tissues and regulate expression of several HOXC genes (HOXC4, HOXC5, HOXC6, HOXC8, HOXC9, and HOXC-AS1). RERG and the nearby gene PTPRO, were reported as risk factors in GWAS studies of the separate traits, but p values for the lead SNP were close to genome-wide significance (p_{Endometriosis} = 9.24e−06, p_GPM = 8.80e−07), suggesting that the increased power from combining the traits was able to identify an additional risk locus for both diseases. Endometriosis is a well-known estrogen-dependent disease, while GORD and PUD have also been associated with estrogen metabolism.^77,78 Studies have shown that estradiol, via estrogen receptor beta (ERβ) and prostaglandin E, activates mRNA expression and phosphorylation of RERG, which further induces the proliferation of primary endometriotic cells.⁷⁹ In addition, ERα may upregulate the estrogen-sensitive PTPRO expression, resulting in an inhibition of cell proliferation and facilitation of apoptosis.⁸⁰

In addition to the five highlighted regions shared by endometriosis and IBS/GPM, another four regions (SEMA3F, RAB5B, CCKBR, and PDE4B) also implicate potential pathways that are associated with the two traits. These four risk loci have been implicated in previous GORD or PUD GWAS studies.^38,81 The lead SNPs in these four regions have not been formally linked to endometriosis in previous GWAS studies, but inspection of the results shows they are all nominally significant in previous GWAS studies.³⁹ They may represent target genes or pathways involved in endometriosis progression.

Evidence for the potential of drug repositioning in clinics

To verify whether the nearest genes identified have implications in clinical settings, we searched the online drug target databases and identified CCKBR and PDE4B with their encoded proteins as drug targets. While the former is currently used for treatment of PUD and GORD, the latter has been clinically trialled for both IBS and endometriosis. CCKBR encodes a G-protein-coupled receptor for gastrin and cholecystokinin. Even though CCKBR has not been targeted for treatment of endometriosis, a recent study has demonstrated that reduction of gastrin is associated with inactivation of CCKBR/ERK/P65 signaling in estrogen receptor-positive breast cancer cells, and lower expression of gastrin and CCKBR was correlated to worse prognosis in breast cancer.⁸² Therefore, the involvement of gastrin and CCKBR in estrogen metabolism suggests that this gene may be a potential drug target for endometriosis. PDE4B is mainly present in immune and epithelial cells and has a role modulating inflammation and epithelial integrity.⁸³ Considering current evidence for Pentoxifylline, which was used to treat endometriosis by targeting PDE4B encoded protein,⁵³ this study provides some evidence for the further investigation of PDE4B for the purpose of treating both endometriosis and GI disorders.

Evidence for associations with medication usage

Linking the diagnosis of endometriosis with pharmaceutical records revealed overlap in medication use between endometriosis and GI disorders, providing insights for disease etiology and management in clinics. The identification of a higher use of medications for IBS, GORD, and PUD in women diagnosed with endometriosis as well as the higher use of hormone therapies in women diagnosed with IBS, GORD, and PUD, but not for IBD, supports the coexistence and potential underlying pathophysiology and genetic correlations between the diseases. While not unexpected, the frequent use of NSAIDs during endometriosis treatment is a concern because NSAIDs are a well-known risk factor for PUD by destroying the mucus layer in the digestive tract.²⁶ However, the genetic correlation between the diseases suggests the relationship is not driven by a consequence of medical therapies alone. In addition, we also identified that the frequency of other treatments for GI disorders such as omeprazole and laxatives was also significantly higher in women with endometriosis. Visceral sensitivity and chronic low-grade inflammatory state have been key characteristics in both IBS and endometriosis⁸⁴ and therapies targeted at relieving pain in IBS can also relieve pain during menstruation⁸⁵ suggesting further potential for drug repositioning. A New Zealand-based study reported that women with IBS and concurrent endometriosis had a significantly higher response rate to a low FODMAP diet than those IBS patients with no known endometriosis.⁶⁰

Clinical implications

Several clinical implications can be drawn from this study. First, regarding diagnosis for both endometriosis and GI disorders, shared etiology suggests that joint or alternative diagnoses should be considered for patients presenting with symptoms related to either disease. Second, evidence from medication use suggests that caution should be taken around contraindications for some drugs such as NSAIDs, often used by endometriosis patients, as a well-known risk factor for PUD. Therefore, it may be worthwhile for clinicians to consider potential contraindications when prescribing NSAIDs for female patients presenting with symptoms shared between the diseases such as abnormal pain, bloating, or constipation. Third, support for PDE4B as a shared drug target for the treatment of both IBS and endometriosis suggests that comorbidity of endometriosis and IBS should be considered in the design and recruitment for clinical trials.

Limitations of the study

A strength of this study is the integration of large-scale population data, genetic risk factors, and medication use data for both GI disorders and endometriosis, to examine the association between these two disorders. Integrating these datasets provides strong evidence for comorbidity between these diseases. Despite multiple lines of evidence supporting comorbidity and genetic overlap, we acknowledge serval limitations. Firstly, endometriosis is a highly heterogeneous condition with variation in lesion location and grade.⁸⁶ Similarly, the four GI phenotypes derived from UKB may also introduce further heterogeneity.³⁸ It is not clear if certain subtypes of endometriosis share more genetic risk factors with digestive disorders. Therefore, the associations identified in this study should be validated in larger endometriosis datasets with more detailed phenotype information when these become available.

Secondly, there are some limitations related to use of the UKB dataset. As mentioned previously, some endometriosis treatments, such as frequent use of NSAIDs, may act as a confounder inducing GI symptoms and contributing to GI disorders.^{6,26,27,28,29} In addition, NSAIDs are over-the-counter (OTC) medications that are easily accessed and complete records of OTC NSAID usage, or when these medications were taken, are not available. It is therefore not possible for us to test whether the comorbidity between endometriosis and GI traits were impacted by the use of NSAIDs. Wu et al.³⁸ also describe that the accuracy of diagnosis for GI disorders (IBS, GORD, PUD, GPM, and IBD) in the UKB may be affected by the inclusion of self-reported illness, the existence of co-reporting of two GI diagnoses, and misdiagnosis. However, the correlation analysis between self-report and hospital admission data among GI disorders, and the sensitivity analysis from excluding individuals with more than one GI disorder diagnosis suggest that self-reported cases and co-reporting diagnosis did not impact the conclusions.³⁸ Overlapping symptoms between endometriosis and GI disorders might lead to misdiagnosis and bias in the associations identified, especially for IBS where diagnosis is made from symptoms-based criteria. The accuracy of endometriosis diagnosis may also be affected by the age of the UKB cohort. The mean age of endometriosis diagnosis in the UKB cohort was 39.2 years, which differs from recent estimates of 26–28 years in other European cohorts.^87,88 UKB participants were recruited from 2006 at an age of 40–69 years, indicating that their diagnoses occurred during a period of less disease awareness possibly resulting in diagnostic delays and lower diagnosis rates. This is also reflected by the lower frequency of endometriosis diagnosis in the UKB compared with current incidence estimates, suggesting that diagnoses may have been missed. Compared with the other four phenotypes, there are fewer IBD, PUD, and endometriosis cases available in UKB, which may limit the power for both genetic and epidemiological analyses. Results for the IBD GWAS were consistent with previously published GWAS³⁸ suggesting that the relevant results in this study are robust.

Finally, differences in the incidence of these diseases between men and women may suggest possible differences in genetic effects between gender. Despite endometriosis being very rarely found in men, we find consistent evidence of a genetic correlation between endometriosis and IBS in men and women, supporting the genetic relationship between the traits and indicating that the conclusions in this study are robust. When separate male and female GWAS summary statistics are available for the other GI traits, extra analysis should be conducted to validate the results in this study.

Conclusions

This study analyzes data from several large-scale datasets to assess the relationship between endometriosis and GI disorders (IBS, GORD, PUD, GPM, and IBD). We provide multiple levels of evidence supporting shared etiological factors between endometriosis and digestive disorders and highlight target genes and pathways contributing to the shared etiology. The results suggest potential targets for treatment, considerations for disease management, and caution around contraindications for some drugs. The clinical implications could facilitate better clinical outcomes for women with both endometriosis and GI diseases.

Consortia

International Endometriosis Genetics Consortium

Yadav Sapkota, Valgerdur Steinthorsdottir, Andrew P. Morris, Amelie Fassbender, Nilufer Rahmioglu, Immaculata De Vivo, Julie E. Buring, Futao Zhang, Todd L. Edwards, Sarah Jones, Dorien O, Daniëlle Peterse, Kathryn M. Rexrode, Paul M. Ridker, Andrew J. Schork, Stuart MacGregor, Nicholas G. Martin, Christian M. Becker, Sosuke Adachi, Kosuke Yoshihara, Takayuki Enomoto, Atsushi Takahashi,¹⁸ Yoichiro Kamatani, Koichi Matsuda, Michiaki Kubo, Gudmar Thorleifsson, Reynir T. Geirsson, Unnur Thorsteinsdottir, Leanne M. Wallace, iPSYCH-SSI-Broad Groupw, Jian Yang, Digna R. Velez Edwards, Mette Nyegaard, Siew-Kee Low, Krina T. Zondervan, Stacey A. Missmer, Thomas D’Hooghe, Grant W. Montgomery, Daniel I. Chasman, Kari Stefansson, Joyce Y. Tung, and Dale R. Nyholt.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Observational data	UK Biobank	https://www.ukbiobank.ac.uk/enable-your-research/apply-for-access
Gastrointestinal traits GWAS summary statistics	Wu et al.38	https://cnsgenomics.com/content/data
Endometriosis GWAS summary statistic	International Endometriosis Genetics Consortium Sapkota et al.39	N/A (Not publicly available due to 23andMe data restrictions, but can contact author for request process)
GTEx eQTL summary statistics_V8	GTEx consortium47	https://gtexportal.org/home/datasets/GTEx_Analysis _v8_eQTL.tar
Endometrial eQTL summary statistics	Mortlock et al.48	http://reproductivegenomics.com.au/shiny/eeqtl2/
Epigenome data	Boix et al.45	http://compbio.mit.edu/epimap/
Drug targets database	Ochoa et al.50	https://platform.opentargets.org/
Medication usage data	ALSWH cohorts42,43 UK Biobank	Table S11 in this study https://www.ukbiobank.ac.uk/enable-your-research/apply-for-access
Software and algorithms
LDSC	Bulik-Sullivan et al.89	https://github.com/bulik/ldsc
GSMR	Zhu et al.44	https://yanglab.westlake.edu.cn/software/gsmr/
mtCOJO	Zhu et al.44	https://yanglab.westlake.edu.cn/software/gcta/#Overview
MetABF	Trochet et al.90	https://github.com/trochet/metabf
IVW-MR	Hemani et al.91	https://mrcieu.github.io/TwoSampleMR/
MR-Egger	Hemani et al.91	https://mrcieu.github.io/TwoSampleMR/
Weighted median	Hemani et al.91	https://mrcieu.github.io/TwoSampleMR/
Simple median	Hemani et al.91	https://mrcieu.github.io/TwoSampleMR/
MR-PRESSO	Hemani et al.91	https://mrcieu.github.io/TwoSampleMR/
Multivariable MR	Hemani et al.91	https://mrcieu.github.io/TwoSampleMR/
RE2C	Lee et al.92	https://github.com/cuelee/RE2C
GWAS-PW	Pickrell et al.93	https://github.com/joepickrell/gwas-pw
GWAS Catalog	Watanabe et al.49	https://fuma.ctglab.nl/
SMR	Zhu et al.46	https://yanglab.westlake.edu.cn/software/smr/#DataResource
EpiMap	Boix et al.45	http://compbio.mit.edu/epimap/
PhenoScanner V2	Kamat et al.94	http://www.phenoscanner.medschl.cam.ac.uk/

Resource availability

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Dr Sally Mortlock (s.mortlock@imb.uq.edu.au).

Materials availability

This study did not generate new unique reagents.

Date and code availability

• •Data: This paper analyses existing and publicly available data which are listed in the key resource table. Endometriosis GWAS summary statistics from the International Endometriosis Genetics Consortium are not publicly available due to data restrictions for summary statistics containing data from 23andMe, Inc. GWAS summary statistics from 23andMe, Inc. were made available under a data use agreement that protects participant privacy. For more information and to apply for access to 23andMe data Please contact dataset-request@23andme. com or visit research.23andMe.com/collaborate. Further information on access to the full endometriosis summary statistics under existing restrictions can be obtained from the lead contact Sally Mortlock.
• •GTEx_Analysis _v8_eQTL.tar were download from the GTEx consortium website (https://gtexportal.org/home/datasets) for the SMR analysis in this study. The specific files (tissue_name_v8.signif_variant_gene_pairs.txt.gz) were formatted following the instructions of SMR analysis.
• •Data from the UK Biobank was made avaliable under data use agreements (No. 54861 and 12505). Please visit https://www.ukbiobank.ac.uk/enable-your-research/apply-for-access for more information and to apply to access the data. The UKB data fields used are provided in the STAR method details section.
• •Code: The paper does not generate original code. We utilized publicly available software in all the analyses. These are listed with appropriate citations in the methods. The paper does not generate original code.
• •Any additional information required to analyze the data reported in this work paper is available from the lead contact Sally Mortlock upon request.

Experimental model and study participants details

This study was based on computational analysis. Additional details are in the Methods Details section.

Method details

Data resources

The large-scale GWAS summary statistics for endometriosis and five gastrointestinal (GI) disorder phenotypes, utilized in this study, have been well described in previous studies.^38,39 Summary data for endometriosis were restricted to eight European ancestry cohorts^39,95 for the purpose of this analysis. A total number of 14,926 cases and 189,715 controls genotyped across 7,899,415 SNPs were included in the endometriosis meta-analysis. GWAS summary statistics for four gastrointestinal disorders, gastro-oesophageal reflux disease (GORD) (n_case = 39,851, n_control = 416,563), peptic ulcer disease (PUD) (n_case = 12,226, n_control = 444,188), irritable bowel syndrome (IBS) (n_case = 14994, n_control = 441420) and inflammatory bowel disease (IBD) (n_case = 6,115, n_control = 450,299), were previously generated using genetic data from individuals in the UK Biobank (UKB), using the health-related outcomes data from combing self-reported, primary care, death register and hospital reported diagnoses. As medications used for PUD also have a therapeutic effect on GORD, a fifth phenotype for gastrointestinal disorders is the combined GORD and PUD and individuals taking medications for GORD/PUD making a total of 75,192 cases and 381,222 controls in the GORD/PUD Medicated (GPM) phenotype.³⁸ Gender stratified GWAS summary statistics for IBS were also generated in this study for the purpose of exploring potential gender bias in the relationship between endometriosis and IBS.

Quantification and statistical analysis

Comorbidity analysis

As a cross-sectional analysis, the comorbid relationship between endometriosis and each GI disorder (IBS, GORD, PUD, GPM and IBD) described above were investigated among unrelated European female individuals in the UKB, with ancestry definition described previously.⁹⁶ Phenotypes were defined using self-reported, hospital admission, death register or primary care record data. Endometriosis cases were defined using date and source of endometriosis reported (UKB data fields:132122 & 132123), ICD10 diagnosis (UKB data field:41270), ICD9 diagnosis (UKB data field: 41271) and self-report (UKB data field: 20002), totalling 5,392 cases (excluding endometriosis of the uterus/adenomyosis). Gastrointestinal disorders definitions were similar as previously described by Wu et al.³⁸ however, these were also restricted to women. A total of 16,330 IBS cases were included (UKB data field: 131639) alongside 22,383 GORD cases (UKB data field: 131585). IBD and PUD were defined using a combination of disease codes, IBD cases were a combination of Crohn’s diseases (UKB data field: 131627) and ulcerative colitis (UKB data field: 131629) diagnoses totalling 2,708 cases and PUD cases were a combination of gastric ulcer cases (UKB data field: 131591), duodenal ulcer cases (UKB data field: 131593), other site peptic ulcer cases (UKB data field: 131595) and gastro-jejunal ulcer cases (UKB data field: 131597) totalling 5,208. We firstly measured whether individuals diagnosed with endometriosis were more likely to have a diagnosis of IBS, GORD, PUD, GPM and/or IBD using Fisher’s exact test, and vice vera in which whether women with each of those five GI disorder phenotypes were more likely to be diagnosed with endometriosis. Next, we conducted a competitive comorbidity analysis to test which digestive disorder is more prone to be comorbid with endometriosis among these four GI disorders (IBS, GORD, PUD and IBD). Briefly, the proportion of endometriosis cases in each of the four digestive diseases were calculated (N_Endo/N_IBS, N_Endo/N_GORD, N_Endo/N_PUD and N_Endo/N_IBD) and then compared in pairs using a two-proportion Z-test. To meet the prerequisite of this analysis that samples in each pair are independent, we removed overlapping samples among four GI disorders (IBS, GORD, PUD and IBD) when calculating the proportion. Also, due to this prerequisite, the GPM phenotype was not considered for competitive comorbidity analysis.

Genetic correlation

Genetic correlation attributable to the genome-wide common SNPs between endometriosis and each of the other five gastrointestinal disorder phenotypes (IBS, IBD, GORD, PUD, GPM) was estimated using bivariate linkage disequilibrium score regression (LDSC)⁸⁹ and their respective GWAS summary statistics. GWAS summary data was formatted using the function ‘munge_sumstats.py’ outlined in the LDSC manual and the genetic correlation for each pair was estimated. The European 1000 genome reference data was adopted in the calculation of linkage disequilibrium (LD) scores. A gender-stratified analysis was conducted to further investigate whether the genetic correlation between endometriosis and IBS is gender dependent. Given that there was no sample overlap in GWAS studies of endometriosis and gastrointestinal disorders and all participants are of European ancestry, we also reduced standard error of genetic correlations by constraining the intercept, which was used to protect bias from population stratification and sample overlap in different GWAS studies.

Assessing potential causal relationships

Mendelian randomisation (MR) uses genetic variants that are robustly associated with the exposure of interest to test whether those genetic variants also increase the risk of another trait.²⁴ MR has emerged as a valuable tool to assess the causal effect of one trait on another. The genetic variants selected are robust and are not associated with other confounders and will only influence the outcome trait through the trait of interest if there is a causal association, thus less susceptible to confounding, measurement error, and reverse causation when compared with conventional observation studies.^24,97,98 In this study, the causal relationships between endometriosis and gastrointestinal disorder phenotypes (IBS, GORD, PUD, GPM) was investigated using one widely-accepted MR method called generalised summary-data-based Mendelian randomization (GSMR).⁴⁴ The combined phenotype of GPM was used in place of individual GWAS for PUD and GORD to increase study power. GSMR uses all significantly associated SNPs as SNP instruments to test for causality. To reduce the influence of horizontal pleiotropy (a single locus directly affecting multiple phenotypes), one potential confounding factor for Mendelian randomization analysis, we also applied the HEIDI-outlier analysis to detect SNPs having obvious pleotropic effect on both risk factor and diseases. A p-value of <0.05 was considered significant. In some cases there was an insufficient number of SNPs to use as instruments and so the GWAS threshold was relaxed (P < 5e-6) to allow at least ten SNPs for each phenotype, following the author’s recommendation to include at least 10 SNP instruments during GSMR analysis to achieve robust results. It has been shown that relaxing the GWAS significance threshold yields more associations, which can lead to better prediction accuracy of the exposure and improve the statistical power in MR analysis.⁶¹ The limitation of this approach is the increased likelihood of false positive SNP-exposure associations, which could increase the chance of introducing pleiotropic effects.⁶² In this study, MR results based on two GWAS significance thresholds (P < 5e-8 and P < 5e-6) were reported.

It has been recommended that multiple MR approaches should be applied to all MR analyses to demonstrate results under different assumptions.⁹¹ In addition to the GSMR method,⁴⁴ we also used other MR approaches to further infer the causal relationship between endometriosis and IBS and GPM. These include Inverse variance weighted (IVW)-MR, MR-Egger, Weighted median, Simple median and MR-PRESSO, each with different assumptions.⁹¹ IVW-MR analysis assumes that all SNP instruments are valid and combines individual Wald ratio together in an either fixed or random effect meta-analysis, while MR-Egger relaxes this assumption by allowing directional or unbalanced horizontal pleiotropy across all SNP instruments.^61,91 The robustness of IVW and MR-Egger depend on the Instrument Strength Independent of Direct Effect (InSIDE) assumption which is likely violated when a proportion of horizontal pleiotropy operates through a confounder.⁶¹ An alternative approach is to assume that only half of SNP instruments are valid and then take the median effect of available SNP instruments to estimate causal effects, including Simple median and Weighted median.⁹⁹ Another existing strategy is to reduce heterogeneity among Wald ratio for each SNP instrument by removing SNPs that contribute to the heterogeneity than expected. Such outlier removal-based MR approaches include the MR-PRESSO¹⁰⁰ and GSMR.⁴⁴ Nevertheless, there is a potential to remove the most biologically reliable SNP instruments.⁶¹ We also applied various sensitivity tests, including F statistics to demonstrate the strength of instrument-exposure association, MR-Egger intercept to indicate the presence of directional horizontal pleiotropy, and Cochran’s Q statistic to test heterogeneity.⁹¹ All these MR analyses and sensitivity tests were performed using the well-established protocol implemented in the “TwoSampleMR” R package. Causal estimates with a p value < 0.05 were considered significant. SNP instruments with F statistics greater than 10 were considered strong. A p value < 0.05 in the MR-Egger intercept pleiotropy test suggested the presence of directional pleiotropy and a Q p value <0.05 indicated significant heterogeneity among SNPs. MR analyses were also performed using a relaxed GWAS significance threshold P < 5e-6. MR Results based on two GWAS significance thresholds (P < 5e-8 and P < 5e-6) were reported, followed by the sensitivity test results.

Conditional MR analysis

Considering the associations between GPM and IBS, GPM and IBS could act as a confounder in the MR framework, which can bias the result. To remove potential confounding, we used mtCOJO⁴⁴ to correct both the exposure and outcome GWAS summary statistics for the confounding effect from either GPM or IBS. We then used the adjusted GWAS summary statistics to repeat the GSMR analysis.

Multivariable MR (MVMR) is an extension of MR that uses genetic variants associated with multiple exposures to estimate direct effect of each exposure on an outcome conditional on other exposures.¹⁰¹ It is well-known that MVMR can also be used to estimate the causal effect to adjust for potential confounding or possible pleiotropy effect.¹⁰² Therefore, in addition to mtCOJO, MVMR, implemented in the “TwoSampleMR” R package was also used in this study to estimate the causal effect of GPM, and IBS, on endometriosis conditional on effects from IBS and GPM, respectively. The opposite direction was also tested setting endometriosis as the exposure. Similar to the standard univariable MR analysis, both GWAS p-value thresholds (P < 5e-8 and P < 5e-6) were used in the MVMR analysis. Under each GWAS threshold, two MVMR analyses were conducted. First, we included both GPM and IBS as exposures to examine the direct causal effect on endometriosis. Second, we include both GPM and endometriosis as exposures, to examine the direct causal effect on IBS.

Cross-trait meta-analysis of endometriosis and gastrointestinal diseases

We next adopted two complementary cross-trait meta-analysis methods, MetABF⁹⁰ and Eskin random-effects model (RE2C),⁹² to identify whether there are shared risk loci between endometriosis and the digestive disorders (IBS, GPM), as well as additional risk loci for each disease. MetABF performs the multi-trait meta-analysis based on the Bayesian framework. Effect alleles were harmonised across all three GWAS. Both fixed and independent effect models were used when performing this meta-analysis. The prior parameter accounting for effects of heterogeneity in two diseases was set as 0.1, which is typically used in complex diseases. As a result, SNPs with a logABF >4 and at least a normally significant p-value <0.05 in each individual disease GWAS analysis were defined as significant in the MetABF analysis.

To validate the MetABF results, we used a complementary cross-trait meta-analysis approach, RE2C, which dramatically increases power when statistics among different studies are correlated compared with other methods. RE2C also accounts for the heterogeneous effects within studies using a statistic model. Similar to MetABF, effect alleles were harmonised across the GWAS prior to being used as input for the RE2C analysis. As a result, a SNP meeting the p-value threshold of < 5e-8 in either fixed (Lin-Sullivan method) or random (RE2C) effects model and having at least a normally significant p-value <0.05 in the individual disease GWAS, were deemed as significant in the meta-analysis. SNPs meeting both thresholds of MetABF and RE2C were selected for further fine mapping analysis in Functional Mapping and Annotation (FUMA)⁴⁹ to identify independent risk loci using a threshold of r² < 0.6 and then the lead SNPs at a threshold of r² < 0.1. The maximum distance between LD blocks to merge into a locus was set to 250kb.

Colocalization analysis

To identify specific genomic regions that have the same causal variant for each disease we conducted a pairwise GWAS (GWAS-PW)⁹³ analysis. Again, this analysis was restricted to comparisons between endometriosis and IBS and GPM. The input to GWAS-PW is a set of estimated effect sizes and standard error for each SNP on each of the paired diseases. The whole genome was split into 1,703 LD independent blocks, and the probability is estimated for four models extended from Giambartolomei et al.,¹⁰³ that a given region (a) contains a genetic variant that impacts the first disease (PPA1); (b) contains a genetic variant that impacts the second disease (PPA2); (c) contains a genetic variant that affects both diseases; (PPA3) or (d) contains two distinct variants that influence each disease separately (PPA4). Paired summary statistics for endometriosis with IBS or GPM were analyzed. Any regions that were identified with a PPA3/PPA4 >0.5 were considered to show evidence of a shared causal variant and two distinct causal variants respectively.

Functional annotation and gene mapping

It has been suggested that the causal gene in each locus is often the gene closest to the lead SNP¹⁰⁴ but this is not always the case as demonstrated in the recent paper by John A. Morris et al.¹⁰⁴ Therefore, to understand the potential regulatory function of identified risk loci and prioritize candidate genes for endometriosis and IBS and GPM, we conducted extra functional annotation analyses EpiMap⁴⁵ and SMR.⁴⁶ We used EpiMap⁴⁵ to perform epigenomic functional annotation for the genome-wide significant SNPs from the cross-trait GWAS meta-analysis. Using available epigenome maps we mapped the genome-wide significant SNPs to predicted enhancers in both reproductive (uterus, ovary) and gastrointestinal tissues (gastroesophageal sphincter, Peyer’s patch, esophagus, stomach, colon, intestine, rectum) and identify gene targets of those enhancers.

SMR⁴⁶ analysis is a powerful approach to identify likely causal or pleiotropic relationships between the trait-associated SNPs and gene expression by integrating GWAS summary statistics with eQTL (expression quantitative trait QTL) information where genetic variants affect gene expression. SMR can distinguish causality and pleiotropy from the linkage association which is less interesting compared with the other two biological associations. We performed SMR analysis on endometriosis, IBS and GPM respectively using the eQTL data from 12 digestive and reproductive tissues in the GTEx project⁴⁷ and endometrium.⁴⁸ The selection of 12 digestive and reproductive tissues in the GTEx project, including Colon_Sigmoid (n = 318), Colon_Transverse (n = 368), Esophagus_Gastroesophageal_Junction (n = 330), Esohagus_Muscosa (n = 497), Esophagus_Muscularis (n = 465), Small_Intestine_Terminal_Ileum (174), Stomach (n = 324), Minor_Salivary_Gland (n = 144), Pancreas (n = 305), Ovary (n = 167), Uterus (n = 129) and Vagina (n = 141), was based on their general function in either GI disorders or reproductive disorders. Numbers in each bracket indicate the number of tissue samples that have been used to generate eQTLs. Of those, salivary gland and pancreas secrete various enzymes and play multiple roles in the gastrointestinal tract,^105,106 and salivary enzymes have also been associated with some GI disorders, including PUD.¹⁰⁷ Because of the limited power of eQTL datasets (most tissues n < 500), we considered the SMR association with P_SMR < 0.05 (nominally significant) and P_HEIDI > 0.05 as significant in this study. Then, we determined if any SMR significant genes were shared between the diseases. Shared causal associations indicated that SNPs may be associated with both diseases through the regulation of expression of the same gene. As a result, we prioritized genes that were nearest to the independent lead SNPs and also had evidence from either SMR or Epimap analyses.

Phenome-wide association

In order to investigate whether the correlation between endometriosis and each of IBS and GPM can be explained by the genetic susceptibility to any other traits or diseases, we searched traits that were associated with genome-wide significant independent SNPs and SNPs in LD (r² > 0.8) from the cross-trait meta-analysis. We used information from GWAS catalog implemented in FUMA⁴⁹ and PhenoScanner.⁹⁴

Drug target analysis

Using the online Open-targets drug database,⁵⁰ we investigated if any known drug targets are common across endometriosis and gastrointestinal disorders (GORD, PUD, IBS) and if any genes functionally annotated to shared risk loci are potential drug targets for either endometriosis and/or digestive disorders.

Medication usage

To investigate the implications of medication use on the relationship between endometriosis and gastrointestinal disorders we analyzed Pharmaceutical Benefits Scheme (PBS) data by endometriosis status from the 1973-78 and 1989-95 cohorts in the Australian Longitudinal Study on Women’s Health (ALSWH).^42,43 The frequency of medications used in these women was calculated. In addition, we characterized the medication use (UKB data field: 20003) in unrelated women in the UK Biobank, including 5,392 women diagnosed with endometriosis, 16,330 women with IBS, 22,383 women with GORD, 5,208 women with PUD and 2,708 women with IBD. We randomly selected age-matched controls to avoid the potential bias caused by differences in age distribution between cases and controls. Differences in the proportion of women, with and without an above diagnosis, using reported medications was tested using fisher test. After correcting for multiple testing using Bonferroni analysis, a p-value of <0.05 was considered significant.

Published: October 26, 2023