{"paper_id":"52eeb841-aa55-4607-8478-07825bc8d9fb","body_text":"McGrath et al. BMC Medicine          (2023) 21:482  \nhttps://doi.org/10.1186/s12916-023-03184-z\nRESEARCH ARTICLE Open Access\n© The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which \npermits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the \noriginal author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or \nother third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line \nto the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory \nregulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this \nlicence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecom-\nmons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.\nBMC Medicine\nPolygenic risk score phenome-wide \nassociation study reveals an association \nbetween endometriosis and testosterone\nIsabelle M. McGrath1*  , International Endometriosis Genetics Consortium, Grant W. Montgomery1   and \nSally Mortlock1   \nAbstract \nBackground Endometriosis affects 1 in 9 women, yet it is poorly understood with long diagnostic delays, invasive \ndiagnoses, and poor treatment outcomes. Characterised by the presence of endometrial-like tissue outside of the \nuterus, its main symptoms are pain and infertility. Endometriosis often co-occurs with other conditions, which may \nprovide insights into the origins of endometriosis.\nMethods Here a polygenic risk score phenome-wide association study of endometriosis was conducted \nin the UK Biobank to investigate the pleiotropic effects of a genetic liability to endometriosis. The relationship \nbetween the polygenic risk score for endometriosis and health conditions, blood and urine biomarkers and repro-\nductive factors were investigated separately in females, males and females without an endometriosis diagnosis. The \nrelationship between endometriosis and the blood and urine biomarkers was further investigated using genetic cor-\nrelation and Mendelian randomisation approaches to identify causal relationships.\nResults Multiple health conditions, blood and urine biomarkers and reproductive factors were associated \nwith genetic liability to endometriosis in each group, indicating many endometriosis comorbidities are not depend-\nent on the physical manifestation of endometriosis. Differences in the associated traits between males and females \nhighlighted the importance of sex-specific pathways in the overlap of endometriosis with many other traits. Notably, \nan association of genetic liability to endometriosis with lower testosterone levels was identified. Follow-up analysis \nutilising Mendelian randomisation approaches suggested lower testosterone may be causal for both endometriosis \nand clear cell ovarian cancer.\nConclusions This study highlights the diversity of the pleiotropic effects of genetic risk to endometriosis irrespective \nof a diagnosis of endometriosis. A key finding was the identification of a causal effect of the genetic liability to lower \ntestosterone on endometriosis using Mendelian randomisation.\nKeywords Endometriosis, Testosterone, PheWAS, Mendelian randomisation, Genetic, Ovarian cancer\nBackground\nEndometriosis is a poorly understood common disease \ncharacterised by the growth of endometrial-like tissue \noutside of the uterus. The diagnostic delay for endome -\ntriosis is 7–11  years, which can be attributed to lack of \ndisease awareness, variability in disease presentation, \nsymptoms that overlap other conditions and the invasive \n*Correspondence:\nIsabelle M. McGrath\nisabelle.mcgrath@uq.edu.au\n1 The Institute for Molecular Bioscience, The University of Queensland, \nBrisbane, QLD 4072, Australia\n\nPage 2 of 14McGrath et al. BMC Medicine          (2023) 21:482 \nnature of the diagnostic technique: laparoscopic surgery. \nThere is a genetic component to endometriosis, with her-\nitability estimates of 47–51% [1, 2], and the most recent \ngenome-wide association study revealed 42 loci associ -\nated with the disease, which explain up to 5.01% of dis -\nease variance [3].\nWhilst the flagship symptoms of endometriosis are \npelvic pain and infertility, it is now appreciated endome -\ntriosis patients will often exhibit disturbances to mul -\ntiple bodily systems beyond the reproductive system. \nRecently, a wealth of epidemiological data has identified \nmany conditions that are comorbid with endometrio -\nsis, multiple of which have evidence of a shared genetic \narchitecture [3–6]. An understanding of the overlapping \ntraits with endometriosis is critical for comprehensive \nhealth management of the patient, for developing predic-\ntive tools, and for elucidating the underlying biology of \nendometriosis.\nThis study expands on previous approaches to charac -\nterising the overlap of endometriosis with other traits. \nA phenome-wide association study (PheWAS) is a tech -\nnique whereby the association of an array of traits with \na particular genetic variant is determined. This approach \ncan be extended to study the association of multiple traits \nwith the genetic liability to a trait of interest through use \nof the polygenic risk score (PRS) for the trait of interest, a \nPRS-PheWAS. Compared to analysis with disease status, \nuse of the PRS for cross-trait analysis does not require a \ncohort with the presence of a trait of interest accurately \nascertained. This is advantageous as many endometriosis \ncases may be undiagnosed and thus be present in control \ncohorts given the heterogeneity in symptom severity and \nthe invasive and lengthy diagnosis process for endome -\ntriosis. Further, as a PRS-PheWAS does not use disease \nstatus, it specifically looks for pleiotropic effects of the \ndisease-associated genetic variants, meaning the effects \nof a genetic liability to endometriosis can be studied in \nindividuals without the disease.\nMethods\nA flow diagram summarising the workflow and methods \nis employed in Fig. 1.\nPhenotype data from UK Biobank\nThe UK Biobank (UKB) is a large population data -\nbase containing comprehensive health records and \ngenetic data of approximately 500,000 individuals. \nThe matched phenotype and genotype information \nallows use of the UKB for a PheWAS. Three groups of \nphenotype data were utilised: ICD10 diagnostic data, \nblood and urine biomarker data and female-specific \nfactors. ICD10 diagnostic codes were mapped to phe -\ncodes and grouped into categories using the map from \nhttps:// phewa  scata log. org/ files/ Pheco de_ map_ v1_2_ \nicd10_ beta. csv. zip and https:// phewa scata log. org/ files/ \npheco de_ defin ition s1.2. csv. zip. Where an ICD10 code \nmapped to multiple phecodes, one combination was \nselected. Phecode categories included infectious dis -\nease, neoplasms, endocrine/metabolic, haematopoietic, \nmental disorders, neurological, sense organs, circula -\ntory system, respiratory, digestive, genitourinary, der -\nmatologic, musculoskeletal, congenital abnormalities, \nsymptoms, injuries and poisonings and other. Numeri -\ncal data from the blood biochemistry and urine assays \nwere collated for each participant. In total, there were \n34 blood/urine variables available. Blood/urine bio -\nmarker data were log transformed prior to PheWAS \nanalysis. The female-specific factors age at menarche, \nage at menopause, length of menstrual cycle, num -\nber of live births, birth weight of first child and age \nat first birth were analysed in the PheWAS. For some \nUKB participants, there were multiple datapoints for \nthese female-specific factors due to follow-up visits. \nFor number of live births, the last recorded value was \nconsidered to account for any live births proceeding the \nenrolment visit. For all other female-specific factors, \nthe first recorded value was considered to minimise \nthe recall period. Quality control was conducted to \nremove extreme values that may reflect pathologies or \nvalues with less than 10 participants: age at menarche \nbetween 8 and 20 and age at first live birth between 14 \nand 43 years were considered, menopause before 40 or \nafter 63 years was excluded, menstrual cycle length was \nrestricted to 22–36 days and birth weight of first child \nbetween 3 and 12 pounds was considered. Four or more \nlive births were collapsed into a single category.\nDevelopment of endometriosis PRS weightings\nSummary statistics from seven European cohorts \nincluded in the Sapkota et al. 2017 meta-analysis (14,926 \ncases; 189,715 controls) [7] of endometriosis were meta-\nanalysed alongside endometriosis GWAS summary sta -\ntistics obtained from FinnGen Release 8 (13,456 cases, \n100,663 controls). The meta-analysis was conducted in \nMETAL using the classical approach with genomic con -\ntrol for each cohort. Although a more recent, better pow-\nered GWAS has been published [3], the current dataset \nwas selected to avoid sample overlap between the data \nused to generate the SNP weightings for the PRS and the \ntesting cohorts for the PRS. A Bayesian method, SBayesR \n[8], as implemented in GCTB 2.02, was used for adjusting \nthe GWAS summary statistics effect sizes. SBayesR was \nperformed with default settings, in addition to the exclu -\nsion of the MHC region, and imputation of the sample \nsize.\n\nPage 3 of 14\nMcGrath et al. BMC Medicine          (2023) 21:482 \n \nCalculation of PRS\nFirstly, two subgroups of the UKB genotype data were \ncurated: unrelated European males (n = 159,855) and \nunrelated European females (n = 188,221). Specifically, \nfemale sex was initially determined using the genotype \ndata provided by UKB and confirmed by only including \nparticipants with Sex = Female in the phenotype data. \nMales had Sex = Male and Sex Genetic = Male in the \nphenotype data. Ancestry was determined using genetic \ninformation [9].\nPRS for three cohorts: females, males and a sensitiv -\nity cohort comprising of females without an endome -\ntriosis diagnosis (n = 182,789) (henceforth referred to as \nthe sensitivity cohort) were calculated using plink1.9’s \nscore function [10] on the SBayesR weightings. The \nparticipants considered as endometriosis cases were \ndetermined by entries in 132,122–0.0 and 132,123–0.0 \n(Date and Source of N80 first report) and was further \nrefined by excluding individuals with the ICD10 diagnos -\ntic code N80.0: endometriosis of the uterus, and no other \nendometriosis diagnosis (N80.1-N80.9). N80.0 refers to \nadenomyosis, which is currently recognised as a distinct \ndisease to endometriosis.\nRunning PheWAS\nIn a PheWAS, the association of a given genotype with \nmultiple phenotypes is tested. This differs from a typi -\ncal GWAS, where for a given phenotype, the association \nwith multiple genotypes is determined. Here, the associa-\ntion of endometriosis PRS with multiple phenotypes was \ntested, i.e. a PRS-PheWAS, essentially testing whether \ngenetic liability to endometriosis affects the probability of \nFig. 1 Flow diagram of methodology. PRS: polygenic risk score; PCs: principal components; GWAS: genome-wide association study; QC: quality \ncontrol; MR: Mendelian randomisation. Created with Biorender.com\n\nPage 4 of 14McGrath et al. BMC Medicine          (2023) 21:482 \nbeing diagnosed with other traits. PRS was converted to \nz-score for PRS-PheWAS.\nThe PRS-PheWAS was conducted using R’s glm \nfunction for logistic regression for the phecodes. The \nfirst 10 genetic principal components (PCs) and age \n(calculated using year of birth) were used as covariates. \nPCs were calculated on genotype data filtered to minor \nallele frequency > 5%, genotype missingness < 5%, SNPs \npassing Hardy–Weinberg exact test with a p  value \nthreshold of 1 ×  10−6 and pruned for linkage disequi -\nlibrium (window size 50  kb, step size 5 variants,  r2 \nthreshold 0.2) for males and females, in addition to \nthe female sensitivity cohort set. PCA was performed \nusing plink2 PCA function with the approx flag [11]. \nPhecodes assigned to at least 100 participants were \nincluded in the PheWAS.\nLinear regression using the lm function in R was used for \nthe blood/urine data. Before running the PheWAS, certain \nbiomarkers affected by statins were adjusted in individu -\nals taking statins at enrolment. Adjustment for statin usage \nwas possible for blood/urine biomarker data as statin usage \nat the time of biological sample collection is available in the \nUK Biobank. The codes for the medications recognised as \nstatins were 1,141,146,234, 1,141,192,414, 1,140,910,632, \n1,140,888,594, 1,140,864,592, 1,141,146,138, 1,140,861,970, \n1,140,888,648, 1,141,192,410, 1,141,188,146, 1,140,861,958, \n1,140,881,748 and 1,141,200,040. The adjustment method \nwas based on a previously published GWAS of blood and \nurine biomarkers in the UKB [12]. Specifically, individu -\nals who were taking statins at the first repeat visit, and not \nthe enrolment visit, were identified. For each biomarker in \nthese individuals, the ratio of on-statin biomarker value/\npre-statin biomarker value was determined. The mean \nratio across all individuals of each sex for each biomarker \nwas determined. This was the sex-specific statin correction \nfactor. Then to determine which biomarkers are affected by \nstatin usage, participants taking statins at the first repeat \nvisit only were utilised for a linear regression modelling \nthe effect of each biomarker on the log ratio of pre-statin \nbiomarker value to on-statin biomarker value. In this lin -\near regression, covariates included Townsend Deprivation \nindex, the first 10 PCs for all individuals of the correspond-\ning cohort, age at enrolment and age difference between \nenrolment and first visit. For age, only the month and \nyear were utilised. Traits that were significant (P < 0.05/34 \nas 34 blood/urine traits) in this linear regression analysis \nfor males and females separately were flagged for adjust -\nment in individuals taking statins at enrolment (Additional \nfile 1: Table  S1). Specifically, for individuals taking statins \nat enrolment, their value for biomarkers affected by statins \nwas divided by the biomarker-specific statin correction \nfactor. In females, apolipoprotein B, C-reactive protein, \ncholesterol, LDL direct and triglycerides were adjusted by \nthe statin correction factor. In males, apolipoprotein B, \nC-reactive protein, cholesterol, direct bilirubin, LDL direct, \nmicroalbumin in urine, sodium in urine, testosterone and \ntriglycerides were adjusted by the statin correction factor. \nIn the PheWAS, fasting time, age at biomarker measure -\nment (utilising month and year information) and the first \n10 PCs were used as covariates.\nFor the female-specific factors, visual assessment for \nnormality of a QQ plot on inverse-rank normal trans -\nformed data determined whether each variable should \nbe treated as a continuous variable or an ordinal cat -\negorical variable. Age of menarche and age at first birth \nwere treated as linear variables so tested with linear \nregression using R’s lm function, whilst age at meno -\npause, length of menstrual cycle, number of live births \nand birth weight of first child were treated as ordinal \ncategorical variables and tested with ordinal logis -\ntic regression using the polr function from the MASS \npackage in R. Linear variables were inverse-rank nor -\nmal transformed. As ordinal logistic regression does \nnot output a P  value, P  values were calculated by com -\nparing the t -values to a standard normal distribution. \nThe first 10 genetic PCs were included as covariates for \nall female-specific factors. Age at menarche was also a \ncovariate for age at first live birth.\nPhenotypes were declared significant if they passed \na stringent Bonferroni threshold (P < 0.05/n traits), \nwhereby each set of traits (phecodes, blood/urine bio -\nmarker and female-specific factors) were considered \nseparately. Analyses were conducted for the male cohort, \nfemale cohort and the sensitivity cohort.\nGenetic investigation of blood and urine biomarkers\nThe genetic relationship of biomarkers apolipoprotein \nA, apolipoprotein B, alanine aminotransferase, testos -\nterone, bioavailable testosterone, triglycerides, HDL \ncholesterol, LDL cholesterol, albumin, calcium, sex hor -\nmone-binding globulin (SHBG) and urate with endome -\ntriosis was assessed. These traits were selected due to \ntheir significance or nominal significance in the PheWAS \nanalysis and availability of appropriate GWAS summary \nstatistics. SHBG was also included due to its previously \nreported strong correlation with testosterone. Bioavail -\nable testosterone was also included, as most testosterone \nis bound to SHBG, and thus inactive. Summary statistics \nwere downloaded from the GWAS Catalog (Table 1) [13]. \nGWAS summary statistics underwent quality control: \nmissing SNP sample size was replaced with the published \ntotal sample size, rsIDs were updated to the rsID of the \nendometriosis summary statistics to ensure maximum \nSNP overlap, in the case of duplicated rsIDs the variant \nwith the lowest P value was retained, and traits with -\nout SNP allele frequency had the allele frequencies of \n\nPage 5 of 14\nMcGrath et al. BMC Medicine          (2023) 21:482 \n \nunrelated European females in the UKB appended. Two \nendometriosis datasets were utilised. In the first, utilised \nfor genetic correlation, three endometriosis GWAS data -\nsets were meta-analysed to maximise power: Rahmioglu \n2023 (21,779 European ancestry cases, 449,087 Euro -\npean ancestry controls, 1713 Japanese ancestry cases, \n1581 Japanese ancestry controls) [3], FinnGen release \n8 (13,456 cases, 100,663 controls), and 23andMe, Inc. \n(4970 cases, 34,561 controls). SNP rsIDs were harmo -\nnised, SNPs with missing P or beta values were removed, \nSNPs with P < 0 or P > 1 were removed and in the case \nof duplicate SNPs, the SNP with the lowest p value was \nretained. The meta-analysis was conducted in METAL \nusing the classical approach with genomic control for \neach cohort. The meta-analysis result underwent a sec -\nondary round of genomic control. As the allele frequency \nwas not present in all three cohorts, the allele frequency \nfrom the largest cohort [3] was subbed in, restricting to \nSNPs present in this GWAS. The sample size was approx-\nimated by the sum of the cohorts the SNP was present in. \nThe second endometriosis dataset, detailed earlier in the \nmethods for the PheWAS, was the meta-analysis of the \nEuropean component of the 2017 endometriosis GWAS \n[7] with FinnGen release 8 (13,456 cases, 100,663 con -\ntrols). Although less powered, this second dataset was \nnecessary to avoid sample overlap with the biomarker \nGWAS, as this is known to bias estimates for Mende -\nlian randomisation. These GWAS summary statistics are \nreferred to as dataset 2. Follow-up analysis of testoster -\none [14] and ovarian cancer utilised ovarian cancer sum -\nmary statistics published in Phelan, Kuchenbaecker [15].\nGenetic correlation assesses the average genome-\nwide correlation in variant effects between traits. The \ngenetic correlation between endometriosis and the \nblood/urine traits was estimated using LDSC (v1.0.1) \nand precomputed LD scores from the 1000 Genomes \nTable 1 GWAS Summary Statistics of Blood/Urine Biomarkers utilised for cross-trait analysis with endometriosis\nMR Mendelian randomisation\nTrait Female-\nspecific\nAncestry and sample size Publication (GWAS Catalog Accession Number)\nApolipoprotein A No 311,601 European ancestry individuals, 5550 \nAfrican ancestry individuals, 6682 South Asian \nancestry individuals\nSinnott-Armstrong, Tanigawa [12]\nGCST90019495\nApolipoprotein B No 340,860 European ancestry individuals, 5962 \nAfrican ancestry individuals, 7275 South Asian \nancestry individuals\nSinnott-Armstrong, Tanigawa [12]\nGCST90019496\nAlanine aminotransferase No 342,387 European ancestry individuals, 6017 \nAfrican ancestry individuals, 7325 South Asian \nancestry individuals\nSinnott-Armstrong, Tanigawa [12]\nGCST90019492\nTestosterone Yes 230,454 European ancestry women Ruth, Day [14]\nGCST90012112\nBioavailable Testosterone Yes 188,507 European ancestry women Ruth, Day [14]\nGCST90012102\nTriglycerides (for Genetic Correlation) No 1,320,016 European ancestry individuals Graham, Clarke [16]\nGCST90239664\nTriglycerides (for MR) No 115,082 European ancestry individuals Richardson, Sanderson [17]\nGCST90092992\nUrate No 342,087 European ancestry individuals, 6011 \nAfrican ancestry individuals, 7328 South Asian \nancestry individuals\nSinnott-Armstrong, Tanigawa [12]\nGCST90019524\nHDL Cholesterol No 313,372 European ancestry individuals, 5573 \nAfrican ancestry individuals, 6689 South Asian \nancestry individuals\nSinnott-Armstrong, Tanigawa [12] GCST90019510\nLDL Cholesterol No 341,875 European ancestry individuals, 6003 \nAfrican ancestry individuals, 7319 South Asian \nancestry individuals\nSinnott-Armstrong, Tanigawa [12]\nGCST90019512\nSex hormone-binding globulin Yes 189,473 European ancestry women Ruth, Day [14]\nGCST90012107\nAlbumin No 313,032 European ancestry individuals, 5573 \nAfrican ancestry individuals, 6687 South Asian \nancestry individuals\nSinnott-Armstrong, Tanigawa [12]\nGCST90019493\nCalcium No 313,387 European ancestry individuals, 5576 \nAfrican ancestry individuals, 6696 South Asian \nancestry individuals\nSinnott-Armstrong, Tanigawa [12]\nGCST90019500\n\nPage 6 of 14McGrath et al. BMC Medicine          (2023) 21:482 \nEuropean reference set. Using the munge_sumstats.py \nscript, the alleles in the GWA summary statistics for \neach trait were crosschecked against HapMap3 SNPs \nused to estimate the LD scores using the merge-alleles \nfunction and chunk size 1,000,000. Genetic correla -\ntion was performed using the endometriosis dataset 1 \nGWAS summary statistics.\nMendelian randomisation analysis was performed \nto assess whether any of these traits could be linked to \nendometriosis via causality. Mendelian randomisation \nis a statistical technique that assesses causality through \nthe use of genetic variants as instrumental variables (IVs) \n[18]. Initially, three models were applied: inverse-vari -\nance weighted (IVW), MR-Egger (MRE) and weighted \nmedian (WM). The IVW method assumes all vari -\nants satisfy the assumptions of MR. The MRE and WM \nmethods are important sensitivity tests. MRE allows for \nan overall directional pleiotropic effect, which provides \nvalid estimates if the pleiotropic effect on the outcome \nis independent of their effects on the exposure [19]. The \nWM method provides valid estimates if at least 50% of \nthe variants are valid instrumental variables [20]. Het -\nerogeneity and pleiotropy statistics were also calculated. \nCausality was assessed with endometriosis as both the \nexposure and outcome variable. SNPs were filtered using \nthe MR Steiger directionality test to ensure they were \nmore strongly associated with the exposure than the \noutcome. FDR-adjusted P values were calculated within \neach directional analysis (i.e. endometriosis as exposure, \nendometriosis as outcome), and within each MR method. \nSignificant results (adjusted P < 0.05) were further investi-\ngated using additional MR methods GSMR [21] and MR-\nPRESSO. GSMR and MR-PRESSO detect and remove \noutlier SNPs, so the result is unlikely to be biased by \npleiotropy. As many blood biomarkers show strong phe -\nnotypic correlation and have many shared genetic risk \nloci, multivariable MR was conducted for testosterone \nwith SHBG, apolipoprotein A with HDL-C and triglyc -\nerides with LDL-C and apoliproprotein B. Multivari -\nable MR determines the causal effect of an exposure on \nthe outcome, conditional on the other exposures in the \nmodel. This is an appropriate sensitivity test to run when \npotential confounders are known. The TwoSampleMR \npackage in R was utilised for both univariate and mul -\ntivariable MR, whilst GSMR was performed as imple -\nmented in GCTA v1.94.1. The LD reference dataset for \nGSMR was the QIMRHCS cohort [7]. Independent SNPs \nwith genome-wide significance (P < 5 ×  10−8) were utilised \nas IVs. As many of the blood/urine biomarker summary \nstatistics were derived from UK Biobank data, endome -\ntriosis dataset 2 summary statistics were utilised. All IVs \nwere sufficiently powered: the F-statistics  (beta2/se2) for \neach IV for every exposure were > 10.\nResults\nPRS-PheWAS results\nThe PheWAS of phecodes tested association of multiple \nphenotypes with endometriosis PRS. Phecodes aggre -\ngate multiple ICD10 codes corresponding to a similar \nphenotype. There were 17, 11 and 2 significant phe -\ncodes in the female, sensitivity and male cohort Phe -\nWASs, respectively (Additional file  1: Tables S2-S4). \nAll significant phecodes were positively associated with \nendometriosis PRS. In the female phecode PheWAS, the \ntop associated phecode was 615: Endometriosis (Fig.  2). \nOther highly associated traits included excessive/fre -\nquent menstruation, uterine leiomyoma, ovarian cyst, \nand pelvic peritoneal adhesions. All phecodes signifi -\ncant in the female analysis replicated at least nominal \nsignificance in the female sensitivity cohort, except for \nchronic inflammatory pelvic disease (P  = 0.064). All 11 \nphecodes significant in the sensitivity cohort were also \nsignificant in the whole female cohort. The phecode \nfor endometriosis was also significant in the sensitiv -\nity cohort. As the phecode definition of endometriosis \nincludes the ICD10 code N80.0: endometriosis of the \nuterus, which here was excluded from the endometrio -\nsis definition, the endometriosis phecode in the female \nsensitivity cohort is representative of adenomyosis. In \nthe male analysis, the two traits significantly associ -\nated with endometriosis PRS were abdominal pain and \nhyperplasia of prostate.\nIn the female-specific factor analysis, younger age at \nmenopause, menarche and first live birth, and shorter \nlength of menstrual cycle were significantly associ -\nated with endometriosis PRS (Table  2, Additional file  1: \nTable  S5). This was replicated in the sensitivity cohort, \nalthough age at menopause was nominally significant \n(P = 0.047) (Additional file 1: Table S6).\nMultiple blood and urine biomarkers were associ -\nated with endometriosis PRS (Tables  3, Additional file  1: \nTables S7-S9). All thirteen biomarkers significant in the \nfemale analysis replicated at the Bonferroni-corrected \nthreshold or nominal significance in the female sensitiv -\nity cohort. In the male analysis five biomarkers were sig -\nnificant, four were also significant in the female analysis \n(triglycerides, HDL cholesterol, calcium, apolipoprotein \nA), whilst alkaline phosphatase had nominal significance \nin the female analysis.\nThe relationship of endometriosis with a subset of the \nblood/urine biomarkers was further explored. Genetic \ncorrelation analysis revealed apolipoprotein A, HDL \ncholesterol, triglycerides, testosterone, SHBG and ala -\nnine aminotransferase were all significantly correlated \nwith endometriosis and passed the stringent Bonferroni-\ncorrected P value threshold (Fig.  3, Additional file  1: \nTable S10).\n\nPage 7 of 14\nMcGrath et al. BMC Medicine          (2023) 21:482 \n \nA negative causal effect of testosterone on endome -\ntriosis was supported by the IVW method (b =  − 0.20, \nP = 2.12 ×  10−7, Padjusted = 2.54 ×  10−6) Fig.  4, Additional \nfile  1: Table  S11). Although heterogeneity was present, \nthis relationship was not driven by any individual SNP , \nas indicated by leave-one-out analysis (Additional file  1: \nTable S12A). Whilst MRE and WM were not significant, \nthe direction of effect was concordant with the IVW \nresult, and the unadjusted MRE result was significant \n(b =  − 0.14, P = 0.049, Padjusted = 0.26). Complementary \ntests MR-PRESSO and GSMR supported the result: the \nFig. 2 Female PRS-PheWAS for endometriosis in the UK Biobank. In total, 841 phecodes were tested for their association with endometriosis \nPRS. Traits are grouped and colour-coded into categories. The P value threshold for significance (dotted line) is 5.95 ×  10−5  (Bonferroni-corrected \nthreshold). The P values were generated from logistic regression, with the first ten genetic principal components and age as covariates. Traits \nwith significant P values are annotated\nTable 2 Female-specific factors associated with endometriosis \nPRS. Females in the UK Biobank were utilised. Estimate refers \nto the regression coefficient from linear regression (age at \nmenarche and age at first live birth), or from ordinal logistic \nregression (age at menopause and length of menstrual cycle)\nSE Standard error, P P value\nBiomarker Estimate SE P\nAge at menopause  − 0.016 0.005 2.64 ×  10−3\nLength of menstrual cycle  − 0.062 0.011 3.38 ×  10−8\nAge at menarche  − 0.017 0.002 4.72 ×  10−14\nAge at first live birth  − 0.021 0.002 1.08 ×  10−13\n\nPage 8 of 14McGrath et al. BMC Medicine          (2023) 21:482 \nMR-PRESSO method was significant, even after adjust -\nment for outliers (b =  − 0.18, P = 2.14 ×  10−6), and GSMR \nwas also significant (b =  − 0.14, P = 8.41 ×  10−7). As the \nmajority of testosterone is bound to SHBG and inactive, \nwe also considered the effect of bioavailable testoster -\none on endometriosis. The IVW method for bioavailable \ntestosterone was also significant with a negative causal \neffect (b =  − 0.16, P = 8.04 ×  10−3) (Fig. 4). Although there \nwas significant heterogeneity in the IVW model, the \nresult was robust to leave-one-out analysis (Additional \nfile  1: Table  S12B). MR-PRESSO and GSMR supported \na negative causal effect of bioavailable testosterone on \nendometriosis (MR-PRESSO outlier adjusted: b =  − 0.18, \nP = 1.53 ×  10−3, GSMR: b =  − 0.20, P = 7.75 ×  10−7). In a \nmultivariable model including SHBG, the effect of overall \ntestosterone on endometriosis was retained (b =  − 0.21, \nP = 5.03 ×  10−5).\nA causal effect of testosterone on ovarian cancer risk \nhas previously been reported using MR approaches \n[14]. Likewise, a causal effect of endometriosis on ovar -\nian cancer has also been reported [22]. Therefore, we \ninvestigated the causal pathways between these three \nTable 3 Significant blood/urine biomarkers associated with \nendometriosis PRS in females in the UK Biobank. Biomarkers \nwere corrected for statin usage. Estimate refers to the regression \ncoefficient from linear regression\nSE Standard error, P P value\nBiomarker Estimate SE P\nTriglycerides 0.0140 0.0020 4.56E − 12\nCalcium 0.0013 0.0002 2.56E − 08\nAlanine aminotransferase 0.1396 0.0291 1.62E − 06\nUrate 0.7117 0.1525 3.07E − 06\nHDL cholesterol  − 0.0042 0.0009 6.37E − 06\nOestradiol 9.8021 2.3176 2.35E − 05\nGamma glutamyltransferase 0.3345 0.0794 2.53E − 05\nAlbumin 0.0265 0.0064 3.55E − 05\nApolipoprotein B 0.0022 0.0005 4.35E − 05\nTestosterone  − 0.0060 0.0016 2.61E − 04\nLDL direct 0.0065 0.0019 8.74E − 04\nApolipoprotein A  − 0.0022 0.0007 8.80E − 04\nTotal protein 0.0323 0.0100 1.23E − 03\nFig. 3 Genetic correlation of blood and urine traits with endometriosis. Genetic correlation (rg) was determined using LDSC. Traits were considered \nsignificantly genetically correlated with endometriosis if P < 4.17 ×  10−3 . Error bars represent standard errors\n\nPage 9 of 14\nMcGrath et al. BMC Medicine          (2023) 21:482 \n \ntraits. Firstly, we evaluated the causal effect of testos -\nterone on ovarian cancer histotypes, as the previously \nreported analysis was limited to overall ovarian cancer. \nA negative causal effect of bioavailable testosterone was \nidentified on one histotype: clear cell carcinoma ovarian \ncancer with the IVW method (b =  − 0.44, P = 6.61 ×  10−3, \nPadjusted = 0.046), in the absence of heterogeneity (Fig.  4, \nAdditional file 1: Table S13). The unadjusted P values and \ndirection of effect for the WM (b =  − 0.63, P = 0.0432) \nand MRE (b =  − 0.66, P = 0.0430) methods supported \nthis result for bioavailable testosterone (Fig.  4, Addi -\ntional file  1: Table S13). The IVW result was not driven \nby an individual SNP (Additional file 1: Table S12C). Both \nMR-PRESSO (b =  − 0.41, P = 0.014, no outliers detected) \nand GSMR (b =  − 0.48, P = 6.48 ×  10−4) also supported \nthe causal effect. Total testosterone also had concordant \ndirections of effect on clear cell carcinoma (Additional \nfile  1: Table  S13). When considering the effect of bio -\navailable testosterone on clear cell ovarian cancer in a \nmultivariable model with endometriosis, the causal effect \nof bioavailable testosterone was no longer significant \n(P = 0.073), whilst endometriosis retained its causative \neffect (b = 0.78, P = 1.80 ×  10−13) (Fig.  4, Additional file  1: \nTable S14).\nDiscussion\nIn this study, we use a PheWAS approach to identify con-\nditions, female-specific traits, and blood/urine biomark -\ners associated with the genetic liability to endometriosis. \nA PheWAS approach differs from epidemiological and \ngenomic studies through the integrated analysis of phe -\nnotype and genotype data. This approach enabled inves -\ntigation of the effects of genetic liability to a disease in the \nabsence of disease: utilisation of a female cohort without \nendometriosis diagnoses and a male cohort suggested \nthat association of many traits with endometriosis cannot \nsolely be attributed to the physical presence of endome -\ntriosis. Using Mendelian randomisation approaches, we \nalso identify a possible causal effect of the genetic liabil -\nity to lower testosterone on endometriosis and clear cell \novarian cancer, which, following further validation, may \nhave important clinical implications for both endometri -\nosis and ovarian cancer.\nGiven many endometriosis diagnoses are likely missed \nin the UK Biobank participants [23], the exclusion of \nindividuals with an endometriosis diagnosis from the \nsensitivity cohort was likely imperfect. In addition to the \nendometriosis sensitivity cohort, males were used as a \nhigh-confidence endometriosis-free group to investigate \nthe effects of genetic liability to endometriosis in the \nabsence of the physical presence of endometriosis. There \nwere two phecodes associated with the endometriosis \nPRS in males: abdominal pain and hyperplasia of pros -\ntate. The relationship with prostate hyperplasia is intrigu-\ning, as although the cause of benign prostate hyperplasia \nis unclear, inflammation and proliferation, also character-\nistic features of endometriosis, are key [24]. Given these \nindividuals did not have a history of endometriosis, the \nassociation of abdominal pain with the endometriosis \nPRS must be explained by factors beyond presence of \nthe lesion. A shared genetic background has previously \nbeen identified between endometriosis and multiple \npain traits [3, 5]. Multiple other traits previously identi -\nfied to have a shared genetic background with endome -\ntriosis were significant in the female analysis but not the \nmale analysis. This may be explained through sex-specific \nhormonal-related pathways being involved in the overlap \nof endometriosis with these traits. Pain is an incentive \nto seeking an endometriosis diagnosis, so in addition to \nthe endometriosis genetic risk signals capturing variants \nassociated with lesion growth, gynaecological healthcare-\nseeking factors such as sensitivity to pain signals may also \nbe captured. Therefore, the association of abdominal pain \nin males with the endometriosis PRS could be explained \nby the endometriosis risk variants being enriched for \npain sensitivity signals. However, this explanation does \nnot negate pleiotropic effects of the genetic variants on \nboth endometriosis and pain sensitivity, or pain-causing \neffects of the endometriosis lesions, given most individu -\nals report a reduction in pain in the short term follow -\ning surgical removal of the lesions [25]. Disentangling the \ngenetic effects of the risk signals for endometriosis on \nlesion characteristics and endometriosis symptoms will \nbe of interest for future studies. Further, the association \nof traits with the endometriosis PRS in males and in the \nfemale sensitivity cohort suggests epidemiological stud -\nies should look for symptoms and traits in relatives of \nFig. 4 Causal relationships between testosterone, endometriosis and clear cell carcinoma ovarian cancer determined by Mendelian randomisation. \nA causal effect of testosterone and bioavailable testosterone on endometriosis is supported by multiple methods. A causal effect of bioavailable \ntestosterone on clear cell carcinoma ovarian cancer was identified using univariate approaches, but adjustment for endometriosis using \nmultivariable MR attenuated the effect and it was non-significant. MR: Mendelian randomisation, GSMR: generalised summary Mendelian \nrandomisation, SHBG: sex hormone-binding globulin, IVW: inverse-variance weighted. MR-PRESSO: Mendelian Randomization Pleiotropy Residual \nSum and Outlier\n(See figure on next page.)\n\nPage 10 of 14McGrath et al. BMC Medicine          (2023) 21:482 \nFig. 4 (See legend on previous page.)\n\nPage 11 of 14\nMcGrath et al. BMC Medicine          (2023) 21:482 \n \nindividuals with endometriosis, which could prove valu -\nable disease-predictive factors.\nThe traits in the PRS-PheWAS analysis identified to be \nassociated with endometriosis genetic risk are directly \ndependent on the characteristics of the cohorts used to \ngenerate the GWAS summary statistics used to calculate \nthe PRSs. Endometriosis is a highly heterogenous condi -\ntion, with a large spectrum in symptom intensity. Further, \nwomen face a multitude of barriers in accessing a diag -\nnosis including the trivialisation of their symptoms, the \nnon-specificity of their symptoms, cost of seeking health-\ncare and the invasiveness of the gold-standard diagno -\nsis (laparoscopic surgery). Therefore, women with more \nsevere symptoms and fewer barriers to seeking health -\ncare are more likely to be diagnosed, and thus present \nin the case sample of a GWAS study. Likewise, there are \nlikely many undiagnosed cases in the control cohort, \nreducing the power of the GWAS. If there is variability \nin the genetic architecture of endometriosis that is cor -\nrelated with healthcare-seeking factors, the results of the \nPRS-PheWAS analysis may be restricted to a subset of \nendometriosis cases.\nA key finding of this study was the suggestion of a \ncausal effect of genetically predicted lower testosterone \non endometriosis using MR. This effect was consist -\nent across multiple models, and when using multiple \nmeasures of testosterone: overall testosterone and bio -\navailable testosterone. Although not all MR models were \nsignificant, the direction of effect was consistent between \nmodels. A limitation of this approach to modelling the \nrelationship with the MR techniques utilised is that a lin -\near effect is assumed: meaning the model suggests lower \ntestosterone is causative, and higher testosterone is pre -\nventative of endometriosis. However, this may be a sim -\nplification of the relationship if testosterone only exerts \nits causative/preventative effect in one of these direc -\ntions. The testosterone SNPs used for MR were female-\nspecific which is important given there is no genetic \ncorrelation between testosterone in males and females \n[14]; however, they were generated from a GWAS of \nmostly postmenopausal adults. Given endometriosis \nonsets in the early reproductive years, assessment of the \ncausative effect using adolescent, childhood and/or pre -\nnatal testosterone-associated SNPs should be assessed if \nthe genetic control of testosterone differs at these stages.\nNevertheless, a role for alterations to the hypotha -\nlamic-pituitary–gonadal (HPG) axis, and a role for tes -\ntosterone, specifically lower prenatal testosterone in \nendometriosis, has been previously discussed [26– 28]. \nDifferences in anogenital distance, a proxy for prenatal \ntestosterone, between endometriosis cases and controls \nimplicates lower prenatal testosterone in endometriosis \ncases [29]. In females, testosterone is synthesised from \ncholesterol in the adrenal gland and ovaries, and the pri -\nmary regulator of testosterone levels in females is this \nsteroid biosynthesis pathway [30]. In the ovary, testos -\nterone is converted to oestradiol by aromatase, which \nis under the control of follicle-stimulating hormone \n(FSH). Elevated FSH has been reported in endometriosis \ncases (although some studies find no difference), along -\nside increased oestradiol only in the menstrual fluid \n(not in circulation) and increased aromatase activity in \nthe eutopic endometrium [27]. In the PheWAS, oestra -\ndiol levels were positively correlated with endometriosis \nPRS; however, this should be considered cautiously as \nmost women in the UKB are postmenopausal. In con -\ntrast, a reduced ovarian oestrogen to testosterone ratio, \nreduced FSH and reduced ovarian aromatase activity are \nobserved in polycystic ovarian syndrome (PCOS). Alter -\nnate alleles of SNPs in strong LD upstream of FSHB  are \nalternatively associated with endometriosis and PCOS: \ni.e. one haplotype confers risk to endometriosis, the \nother to PCOS [31].\nThere are a few possible mechanisms of lower tes -\ntosterone impacting endometriosis risk. In support \nof Sampson’s retrograde menstruation hypothesis for \nendometriosis, lower prenatal and postnatal testoster -\none cause earlier menarche, shorter menstrual cycles \nand thicker endometrial lining, increasing the exposure \nto menstruation [27]. Oestrogens have inflammatory \neffects, whilst androgens have anti-inflammatory effects, \nso a high oestrogen:testosterone ratio could promote \ninflammation in response to ectopic endometrial tis -\nsue [27]. The Müllerian remnants are another hypothe -\nsis for endometriosis, whereby misplaced stem cells are \nactivated by a stimulus [32, 33]. Low testosterone could \ncontribute to the activation of these stem cells by facili -\ntating a high inflammatory environment, and/or through \nfacilitating deposition of these stem cells [27]. Testoster -\none is a controller of HOXA10 expression [34], which \nis a key player in the development of the female repro -\nductive tract, and has been implicated in endometriosis \n[35]. Low testosterone could also contribute to pain in \nendometriosis, through mechanisms such as the inverse \nassociation with inflammation, and/or through links \nwith β-endorphin levels within the central nervous sys -\ntem [28]. Importantly, a role for prenatal testosterone \nand early disturbance of the HPG axis in endometriosis \nwould imply a developmental origin for endometriosis \n[28].\nPreviously a causative effect of lower testosterone on \novarian cancer was reported using MR [14]. Analysis \nof the major ovarian cancer histotypes indicated this \ncausative effect was restricted to clear cell carcinoma; \nhowever, this effect was attenuated and non-signifi -\ncant when considered in a multivariable MR model \n\nPage 12 of 14McGrath et al. BMC Medicine          (2023) 21:482 \nwith endometriosis. This suggests the effect of lower \ntestosterone on clear cell carcinoma may be partially \nmediated through endometriosis; however, confidence \nintervals were large, so validation in larger datasets is \nnecessary. Endometriosis has been previously identi -\nfied as a cause of multiple histotypes of ovarian cancer: \nmost strongly clear cell carcinoma and endometrioid \novarian cancer [22]. As lower testosterone did not show \na causative effect on endometrioid ovarian cancer, tes -\ntosterone may play a role in determining the histotype \nof ovarian cancer resulting from the endometriosis. \nInterestingly, polycystic ovarian syndrome, for which \nhigh testosterone is a characteristic feature and identi -\nfied as causative using MR [14], has been determined as \npreventative for endometrioid ovarian cancer using MR \ntechniques [36].\nFour female-specific factors were associated with the \nendometriosis PRS: earlier age at menopause, shorter \nlength of menstrual cycle, earlier age at menarche and \nearlier age at first live birth. A significant genetic corre -\nlation of endometriosis with these traits has previously \nbeen reported [3 ]. Shorter length of menstrual cycle \nand earlier age at menarche in endometriosis, also \nobserved in epidemiological data, mean a greater expo -\nsure to menstruation, in support of Sampson’s theory \nof retrograde menstruation for endometriosis [37, 38]. \nThe association of earlier age at first live birth, a proxy \nfor fertility, with endometriosis PRS likely owes to the \nprogressive nature of fertility issues. Age at first birth \nremained significant in the female sensitivity cohort. \nThis may be explained by the presence of undiagnosed \nendometriosis cases in this cohort, and/or the direct \neffects of some endometriosis risk loci on infertility. \nWell-designed epidemiological studies of the relation -\nship between age of menopause and endometriosis are \nlacking [39]. One study reported an increased risk of \nearly natural menopause (< 45  years) in endometriosis \npatients [40]. Potential mechanisms include effects of \nendometrioma (endometriosis on the ovary) on ovarian \nfunction, effects of surgical excision of endometrioma \n[41] and effects of endometriosis-related traits such as \nreduced body mass index [39]. The attenuation of the \neffect of endometriosis PRS on age at menopause when \nendometriosis cases were excluded may point to effects \nof lesion presence on age at menopause, but further \nvalidation is needed.\nThe association of blood biomarkers with the endo -\nmetriosis PRS provides promising evidence these bio -\nmarkers could be useful in predicting endometriosis. \nHowever, as the biomarker measurements from the \nUKB utilised here were measured in women older than \nthe typical age for seeking an endometriosis diagnosis, \ntheir prediction accuracy should be assessed in a \nyounger cohort, prior to surgical excision of the endo -\nmetriosis lesions. Given most women in the UKB are \npostmenopausal, this is particularly relevant for bio -\nmarkers that are strongly affected by menopause, such \nas oestrogen. The association of various lipid-related \nbiomarkers aligns with the association of endometrio -\nsis with cardiovascular traits in the phecode PheWAS \nand has been previously reported in epidemiological \ndata [ 42]. Triglycerides show significant differences \nbetween endometriotic and normal endometrium of \nendometriosis patients [43]. The absence of any causal \nrelationships between endometriosis and most blood/\nurine biomarkers may suggest pleiotropic genetic effects \nand/or non-genetic factors may be responsible for the \noverlap. One limitation of the genetic analysis is that \nfor most traits the GWAS summary statistics were not \nfemale-specific, nor specific to young reproductive aged \nwomen. Timely and female-specific genetic associations \nmay be necessary to reveal causal relationships and the \ntrue magnitude of genetic overlap with endometriosis. \nFurther, not all GWAS summary statistics were derived \nfrom a purely European ancestry sample; however, the \nnon-European component is small, and thus any bias is \nexpected to be small.\nConclusions\nWe have performed a comprehensive PRS-PheWAS for \nendometriosis. Associations of traits with genetic liabil -\nity to disease, rather than disease presence, has provided \ninteresting insights into the comorbidity of these traits \nwith endometriosis and has ramifications for co-treat -\nment of these diseases. Validation of a causal effect of \nlower testosterone on endometriosis using MR, and the \nfinding of an effect of lower testosterone on clear cell \ncarcinoma, prompts further investigation into the devel -\nopmental origins of endometriosis and the malignant \ntransformation of endometriosis into ovarian cancer in \nrelation to testosterone.\nAbbreviations\nFSH   Follicle-stimulating hormone\nGSMR  Generalised summary Mendelian randomisation\nHPG  Hypothalamic-pituitary–gonadal\nIVs  Instrumental variables\nIVW  Inverse-variance weighted\nMRE  MR-Egger\nMR-PRESSO  Mendelian Randomization Pleiotropy Residual Sum and Outlier\nPCOS  Polycystic ovarian syndrome\nPCs  Principal components\nPheWAS  Phenome-wide association study\nPRS  Polygenic risk score\nSHBG  Sex hormone-binding globulin\nUKB  UK Biobank\nWM  Weighted median\n\nPage 13 of 14\nMcGrath et al. BMC Medicine          (2023) 21:482 \n \nSupplementary Information\nThe online version contains supplementary material available at https:// doi. \norg/ 10. 1186/ s12916- 023- 03184-z.\nAdditional file 1: Table S1. Statin Adjustment Factors and Linear Regres-\nsion Analysis of Effect of Statin Usage on Biomarker Levels. Table S2. \nPRS-PheWAS of phecodes in females utilising endometriosis PRS. \nTable S3. PheWAS of phecodes in females without endometriosis utilising \nendometriosis PRS. Table S4. PheWAS of phecodes in males utilising \nendometriosis PRS. Table S5. PheWAS of female-specific factors in females \nutilising endometriosis PRS. Table S6. PheWAS of female-specific factors \nin females without endometriosis utilising endometriosis PRS. Table S7. \nPRS-PheWAS of blood/urine biomarkers in females utilising endometriosis \nPRS. Table S8. PRS-PheWAS of blood/urine biomarkers in females without \nendometriosis utilising endometriosis PRS. Table S9. PRS-PheWAS of \nblood/urine biomarkers in males utilising endometriosis PRS. Table S10. \nGenetic Correlation (rg) between endometriosis and blood/urine \nbiomarkers. Table S11. Mendelian randomisation of endometriosis and \nblood/urine biomarkers. Tables S12. A-C Leave one out analyses. S13. \nMendelian randomisation ovarian cancer and testosterone. S14. Multivari-\nable Mendelian randomisation.\nAcknowledgements\nSummary statistics from the endometriosis GWAS used in this study contain \ndata from 23andMe and FinnGen. We want to acknowledge the participants \nand investigators of the FinnGen study and would like to thank the research \nparticipants and employees of 23andMe, Inc. for making this work possible. \nThis research has been conducted using the UK Biobank Resource under \nApplication Number 54861. Individual-level phenotype information (including \nage and IDC10 diagnosis) and genotype data for genetic risk assessment \nwas accessed under Application Number 54861. We thank the UK Biobank \nparticipants and the UKB research teams for their generous contributions to \ngenerating an important research resource. This work uses data provided by \npatients and collected by the NHS as part of their care and support. Copyright \n© 2023, NHS England. Re-used with the permission of the UK Biobank. All \nrights reserved.\nAuthors’ contributions\nIMM, GWM and SM contributed to the study conception and design. Data \nanalysis was performed by IMM and was interpreted by IMM, GWM and SM. \nIEGC generated, and provided access to, the endometriosis GWA meta-analy-\nsis. The first draft of the manuscript was written by IMM, and IMM, GWM and \nSM commented on previous versions of the manuscript. IMM, GWM and SM \nread and approved the final manuscript.\nAuthors’ Twitter handles\nTwitter handles: @SallyMortlock (Sally Mortlock); @IsabelleMcGr (Isabelle M. \nMcGrath).\nFunding\nOpen Access funding enabled and organised by CAUL and its Member \nInstitutions. GWM was supported by NHMRC Fellowship GNT1177194. SM was \nsupported by Medical Research Future Fund Research Grant MRF1199785.\nAvailability of data and materials\nData can be accessed from UK Biobank (https:// www. ukbio bank. ac. uk/) as per \ntheir published data access procedures. Summary data for the FinnGen Endo-\nmetriosis GWAS is available from their results portal (https:// www. finng en. fi/ \nen/ access_ resul ts). Publicly available GWAS summary data is cited in Table 1 or \ndetailed in text. Endometriosis GWAS summary statistics, where not publicly \navailable, are available on request. Any additional data supporting the conclu-\nsions of this article are included within the article and its additional files.\nDeclarations\nEthics approval and consent to participate\nThis study employed data from the UK Biobank and summary statistics from \nadditional sources and were approved by the UK Biobank and the Human \nResearch Ethics Committee of The University of Queensland (Project 2020/\nHE002852).\nConsent for publication\nNot applicable.\nCompeting interests\nThe authors declare that they have no competing interests.\nReceived: 13 August 2023   Accepted: 20 November 2023\nReferences\n 1. Treloar SA, O’Connor DT, O’Connor VM, Martin NG. Genetic influ-\nences on endometriosis in an Australian twin sample. Fertil Steril. \n1999;71(4):701–10.\n 2. Saha R, Pettersson HJ, Svedberg P , Olovsson M, Bergqvist A, Marions L, \net al. Heritability of endometriosis. Fertil Steril. 2015;104(4):947–52.\n 3. Rahmioglu N, Mortlock S, Ghiasi M, Moller PL, Stefansdottir L, Galarneau \nG, et al. The genetic basis of endometriosis and comorbidity with other \npain and inflammatory conditions. Nat Genet. 2023;55(3):423–36.\n 4. McGrath IM, Montgomery GW, Mortlock S. Insights from Mendelian \nrandomization and genetic correlation analyses into the relationship \nbetween endometriosis and its comorbidities. Hum Reprod Update. \n2023;29(5):655–74.\n 5. McGrath IM, International Endometriosis Genetics Consortium, Mont-\ngomery GW, Mortlock S. Genomic characterisation of the overlap of \nendometriosis with 76 comorbidities identifies pleiotropic and causal \nmechanisms underlying disease risk. Hum Genet. 2023;142(9):1345–60.\n 6. Yang F, Wu Y, Hockey R, Doust J, Mishra GD, Montgomery GW, et al. \nEvidence of shared genetic factors in the etiology of gastrointestinal \ndisorders and endometriosis and clinical implications for disease man-\nagement. Cell Rep Med. 2023;4(11):101250.\n 7. Sapkota Y, Steinthorsdottir V, Morris AP , Fassbender A, Rahmioglu N, De \nVivo I, et al. Meta-analysis identifies five novel loci associated with endo-\nmetriosis highlighting key genes involved in hormone metabolism. Nat \nCommun. 2017;8:15539.\n 8. Lloyd-Jones LR, Zeng J, Sidorenko J, Yengo L, Moser G, Kemper KE, et al. \nImproved polygenic prediction by Bayesian multiple regression on sum-\nmary statistics. Nat Commun. 2019;10(1):5086.\n 9. Yengo L, Sidorenko J, Kemper KE, Zheng Z, Wood AR, Weedon MN, et al. \nMeta-analysis of genome-wide association studies for height and body \nmass index in ∼700000 individuals of European ancestry. Hum Mol \nGenet. 2018;27(20):3641–9.\n 10. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, et al. \nPLINK: a tool set for whole-genome association and population-based \nlinkage analyses. Am J Hum Genet. 2007;8181:559–75.\n 11. Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ. Second-\ngeneration PLINK: rising to the challenge of larger and richer datasets. \nGigascience. 2015;4(1):7.\n 12. Sinnott-Armstrong N, Tanigawa Y, Amar D, Mars N, Benner C, Aguirre M, \net al. Genetics of 35 blood and urine biomarkers in the UK Biobank. Nat \nGenet. 2021;53(2):185–94.\n 13. Sollis E, Mosaku A, Abid A, Buniello A, Cerezo M, Gil L, et al. The NHGRI-EBI \nGWAS Catalog: knowledgebase and deposition resource. Nucleic Acids \nRes. 2023;51(D1):D977–85.\n 14. Ruth KS, Day FR, Tyrrell J, Thompson DJ, Wood AR, Mahajan A, et al. Using \nhuman genetics to understand the disease impacts of testosterone in \nmen and women. Nat Med. 2020;26(2):252–8.\n 15. Phelan CM, Kuchenbaecker KB, Tyrer JP , Kar SP , Lawrenson K, Winham SJ, \net al. Identification of 12 new susceptibility loci for different histotypes of \nepithelial ovarian cancer. Nat Genet. 2017;49(5):680–91.\n 16. Graham SE, Clarke SL, Wu KH, Kanoni S, Zajac GJM, Ramdas S, et al. The \npower of genetic diversity in genome-wide association studies of lipids. \nNature. 2021;600(7890):675–9.\n 17. Richardson TG, Sanderson E, Palmer TM, Ala-Korpela M, Ference BA, \nDavey Smith G, et al. Evaluating the relationship between circulating \nlipoprotein lipids and apolipoproteins with risk of coronary heart disease: \n\nPage 14 of 14McGrath et al. BMC Medicine          (2023) 21:482 \n•\n \nfast, convenient online submission\n •\n  \nthorough peer review by experienced researchers in your field\n• \n \nrapid publication on acceptance\n• \n \nsupport for research data, including large and complex data types\n•\n  \ngold Open Access which fosters wider collaboration and increased citations \n \nmaximum visibility for your research: over 100M website views per year •\n  At BMC, research is always in progress.\nLearn more biomedcentral.com/submissions\nReady to submit y our researc hReady to submit y our researc h  ?  Choose BMC and benefit fr om: ?  Choose BMC and benefit fr om: \na multivariable Mendelian randomisation analysis. PLoS Med. 2020;17(3): \ne1003062.\n 18. Smith GD, Ebrahim S. ‘Mendelian randomization’: can genetic epide-\nmiology contribute to understanding environmental determinants of \ndisease? Int J Epidemiol. 2003;32(1):1–22.\n 19. Burgess S, Butterworth A, Thompson SG. Mendelian randomization \nanalysis with multiple genetic variants using summarized data. Genet \nEpidemiol. 2013;37(7):658–65.\n 20. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent estima-\ntion in Mendelian randomization with some invalid instruments using a \nweighted median estimator. Genet Epidemiol. 2016;40(4):304–14.\n 21. Zhu Z, Zheng Z, Zhang F, Wu Y, Trzaskowski M, Maier R, et al. Causal asso-\nciations between risk factors and common diseases inferred from GWAS \nsummary data. Nat Commun. 2018;9(1):224.\n 22. Mortlock S, Corona RI, Kho PF, Pharoah P , Seo JH, Freedman ML, et al. A \nmulti-level investigation of the genetic relationship between endome-\ntriosis and ovarian cancer histotypes. Cell Rep Med. 2022;3(3): 100542.\n 23. Blass I, Sahar T, Shraibman A, Ofer D, Rappoport N, Linial M. Revisiting the \nrisk factors for endometriosis: a machine learning approach. J Pers Med. \n2022;12(7):1114.\n 24. Devlin CM, Simms MS, Maitland NJ. Benign prostatic hyperplasia – what \ndo we know? BJU Int. 2021;127(4):389–99.\n 25. Leonardi M, Gibbons T, Armour M, Wang R, Glanville E, Hodgson R, \net al. When to do surgery and when not to do surgery for endometrio-\nsis: a systematic review and meta-analysis. J Minim Invasive Gynecol. \n2020;27(2):390-407.e3.\n 26. Dinsdale N, Nepomnaschy P , Crespi B. The evolutionary biology of endo-\nmetriosis. Evol Med Public Health. 2021;9(1):174–91.\n 27. Dinsdale NL, Crespi BJ. Endometriosis and polycystic ovary syndrome are \ndiametric disorders. Evol Appl. 2021;14(7):1693–715.\n 28. Crespi BJ, Evans SF. Prenatal origins of endometriosis pathology and \npain: reviewing the evidence of a role for low testosterone. J Pain Res. \n2023;16:307–16.\n 29. Sanchez-Ferrer ML, Mendiola J, Jimenez-Velazquez R, Canovas-Lopez L, \nCorbalan-Biyang S, Hernandez-Penalver AI, et al. Investigation of anogeni-\ntal distance as a diagnostic tool in endometriosis. Reprod Biomed Online. \n2017;34(4):375–82.\n 30. Sinnott-Armstrong N, Naqvi S, Rivas M, Pritchard JK. GWAS of three \nmolecular traits highlights core genes and pathways alongside a highly \npolygenic background. Elife. 2021;10: e58615.\n 31. McGrath IM, Mortlock S, Montgomery GW. Genetic regulation of \nphysiological reproductive lifespan and female fertility. Int J Mol Sci. \n2021;22(5):2556.\n 32. Von Recklinghausen F. Adenomyomas and cystadenomas of the wall of \nthe uterus and tube: their origin as remnants of the wolffian body. Wien \nKlin Wochenschr. 1896;8:530.\n 33. Longo LD. Classic pages in obstetrics and gynecology. Aberrant portions \nof the mullerian duct found in an ovary: William Wood Russell Johns \nHopkins Hospital bulletin, vol. 10, pp. 8--10, 1899. Am J Obstet Gynecol. \n1979;134(2):225–6.\n 34. Cermik D, Selam B, Taylor HS. Regulation of HOXA-10 expression by \ntestosterone in vitro and in the endometrium of patients with polycystic \novary syndrome. J Clin Endocrinol Metab. 2003;88(1):238–43.\n 35. Zanatta A, Rocha AM, Carvalho FM, Pereira RM, Taylor HS, Motta EL, \net al. The role of the Hoxa10/HOXA10 gene in the etiology of endo-\nmetriosis and its related infertility: a review. J Assist Reprod Genet. \n2010;27(12):701–10.\n 36. Harris HR, Cushing-Haugen KL, Webb PM, Nagle CM, Jordan SJ, Australian \nOvarian Cancer Study G, et al. Association between genetically predicted \npolycystic ovary syndrome and ovarian cancer: a Mendelian randomiza-\ntion study. Int J Epidemiol. 2019;48(3):822–30.\n 37. Treloar SA, Bell TA, Nagle CM, Purdie DM, Green AC. Early menstrual char-\nacteristics associated with subsequent diagnosis of endometriosis. Am J \nObstet Gynecol. 2010;202(6):534.e1-6.\n 38. Sampson JA. Peritoneal endometriosis due to the menstrual dissemi-\nnation of endometrial tissue into the peritoneal cavity. Am J Obstet \nGynecol. 1927;14:422–69.\n 39. Alio L, Angioni S, Arena S, Bartiromo L, Bergamini V, Berlanda N, et al. \nEndometriosis: seeking optimal management in women approaching \nmenopause. Climacteric. 2019;22(4):329–38.\n 40. Thombre Kulkarni M, Shafrir A, Farland LV, Terry KL, Whitcomb BW, Elias-\nsen AH, et al. Association between laparoscopically confirmed endome-\ntriosis and risk of early natural menopause. JAMA Netw Open. 2022;5(1): \ne2144391.\n 41. Tan Z, Gong X, Li Y, Hung SW, Huang J, Wang CC, et al. Impacts of endo-\nmetrioma on ovarian aging from basic science to clinical management. \nFront Endocrinol (Lausanne). 2022;13:1073261.\n 42. Nahar K, Khanam NN, Chowdhury AA, Khan NJ, Mohamed Z. Association \nof dyslipidemia with endometriosis: a case control study. Mymensingh \nMed J. 2023;32(1):118–24.\n 43. Adamyan LV, Starodubtseva N, Borisova A, Stepanian AA, Chagovets V, \nSalimova D, et al. Direct mass spectrometry differentiation of ectopic and \neutopic endometrium in patients with endometriosis. J Minim Invasive \nGynecol. 2018;25(3):426–33.\nPublisher’s Note\nSpringer Nature remains neutral with regard to jurisdictional claims in pub-\nlished maps and institutional affiliations.","source_license":"CC0","license_restricted":false}