Independent Replication and Meta-Analysis for Endometriosis Risk Loci

Study Participants The cases ( n = 1,077) and controls ( n = 900) who were recruited in this study at the Leuven University Hospital, Belgium during 1993–2012 had undergone laparoscopy for sub-fertility with or without pain. Presence of endometrio- sis in cases was confirmed laparoscopically and histologi- cally based on electronic medical file records. The disease severity in women with endometriosis was prospectively graded according to the rAFS classification system. En- dometriosis cases had either minimal (stage I, n = 380), mild (stage II, n = 229), moderate (stage III, n = 174), severe (stage IV ,n = 284), or unknown ( n = 10) disease. Absence of endometriosis in controls was confirmed la- paroscopically. Both cases and controls were Caucasian in origin. All the study participants provided written informed consent, and the study was approved by the Commission of Medical Ethics of the Leuven University Hospital, Belgium and QIMR Berghofer Human Ethics Research Committee, Australia. DNA Extraction and Genotyping DNA was purified from EDTA-stabilized whole blood col- lected for routine molecular diagnostic tests at the Cen- tre for Human Genetics of University Hospitals, Leuven, Belgium. Following the manufacturer’s protocol, DNA was purified using Chemagic DNA blood special kit (Chema- gen MSM I, PerkinElmer Chemagen T echnologies GmbH, Baesweiler, Germany) based on the specific binding of DNA to paramagnetic beads, and Auto Pure LS Puregene chem- istry (Qiagen, V enlo, The Netherlands) based on salting- out extraction and a manual salting-out procedure (home- b r e w ) .T h ec h o i c eo ft h ee x t r a c t i o nm e t h o dw a sb a s e do n the available amount of blood and on the type of required molecular diagnostic test. DNA concentration was mea- sured using Victor (PerkinElmer, Massachussetts, USA). Whole genome genotyping of the DNA samples was performed using the Illumina HumanCoreExome 12v1.1 array at the Molecular Epidemiology Laboratory, QIMR Berghofer Medical Research Institute, Brisbane, Australia, following the manufacturer’s standard protocol. For a qual- ity control (QC) check, DNA concentrations of majority of the samples were re-measured on the BioTech Powerwave at the QIMR Molecular Epidemiology Laboratory before genotyping. The Illumina HumanCoreExome genotyping arrays are the newer generation of Illumina GWA arrays, which comprised /223c250,000 common tag SNPs (‘core’) and /223c250,000 predominantly rare coding variants (‘exome’). The exome variants in the Illumina HumanCoreExome were carefully selected based on exome sequencing data in /223c12,000 individuals. Genotype Calling and Quality Control Genotype data were called using a custom cluster file gen- erated using /223c2,000 good quality ( <1% missing rate) TWIN RESEARCH AND HUMAN GENETICS 519 https://doi.org/10.1017/thg.2015.61 Published online by Cambridge University Press Yadav Sapkota et al. samples and the GenCall algorithm within Illumina Genome Studio. Data were then further processed by zCall (Goldstein et al.,2012) ,ar a r ev a r i a n tc a l l e r ,i na na t t e m p tt o recall missing genotypes. Following manufacturer’s guide- lines and the protocols developed for the Exome chip data, quality control measures were applied to the Belgian GWA data. Briefly, samples with >1% missing rates, outlying heterozygosity, non-European ancestries (based on 1,000 Genomes Project’s European populations), cryptic related- ness (pi-hat > 0.2), and gender discordances were excluded. Similarly, markers with poor separation of three genotype clusters, excess heterozygosity, outlying mean theta and in- tensity values for heterozygote genotypes, >1% missing rates, the Hardy–Weinberg Equilibrium (HWE) p < 10 −6 in controls, and minor allele frequency (MAF) < .05% in either cases or controls were dropped. Association Analysis Since rs1096523 in the CDKN2BAS gene at 9p21.3 is monomorphic in populations of European ancestry, we considered the remaining nine SNP loci (rs7521902, rs13394619, rs4141819, rs6542095, rs1519761, rs7739264, rs12700667, rs1537377, and rs10859871) for further anal- ysis. T o see whether there were any novel association sig- nals for endometriosis in the Belgian cohort, GWA analy- sis of the observed genotypes was performed using –assoc (for ‘core’ SNPs) and –fisher (for ‘exome’ variants) com- mands in Plink for data, including all endometriosis cases (‘All’) and controls. Considering the relatively greater ge- netic loading of moderate-to-severe (rAFS stage III/IV or ‘Grade_B’) endometriosis compared with mild or minimal (rAFS stage I/II or ‘Grade_A’) disease (Nyholt et al., 2012; Painter et al., 2011; Sapkota et al., 2015a), additional anal- ysis for ‘Grade_B’ endometriosis cases versus controls was also performed. Strengths of association of SNPs with en- dometriosis are reported in terms of odds ratio (ORs) and confidence intervals (CIs). Imputation and Meta-Analysis Of the nine SNPs, only six (rs7521902, rs13394619, rs4141819, rs6542095, rs7739264, and rs12700667) are assayed on Illumina HumanCoreExome 12v1.1 geno- typing platform. Therefore, we imputed genotypes in chromosomes containing the nine SNP loci in the Belgian GWA data, using a reference panel of 1,000 Genomes Project (March 2012 release). Imputation was carried out using SHAPEIT (Delaneau et al., 2012) and minimac programs (Li et al., 2009, 2010) and following the two-step approach outlined in the online Minimac: 1,000 Genomes Imputation Cookbook (http://genome.sph.umich.edu/ wiki/Minimac:_1000_Genomes_Imputation_Cookbook). Quality of the imputed genotypes was assessed byr 2 metric, which estimates the squared correlation between true and imputed genotypes. Poorly imputed SNPs indicated by r2 < .3 were excluded from downstream analyses. Association TABLE 1 Summary of the Datasets Used in the Current Study Endometriosis GWA study cases (‘Grade_B’) Controls Ethnicity Belgian 998 (423) 783 Caucasian QIMRHCS 2,262 (905) 2,924 European OX 919 (452) 5,151 European BBJ 1,423 1,318 Japanese Albertsen et al. (2013) 1,514 12,660 Caucasian — discovery Albertsen et al. (2013) 505 1,811 Caucasian — replication Adachi et al. (2010) 696 825 Japanese Note: QIMRHCS = Queensland Institute of Medical Research and Hunter Community Study; OX = Oxford (UK); BBJ = BioBank of Japan. analyses of imputed genotype dosage scores of the nine SNP loci were conducted using Plink for ‘All’ and ‘Grade_B’ endometriosis cases separately. After combining results of imputed dosage scores for the nine SNP loci from Belgian data with the published results of Nyholt et al. (2012), Adachi et al. (2010), Albertsen et al. (2013), and Sapkota et al. ( 2015b), we performed meta- analysis for ‘All’ endometriosis cases and controls. A brief summary of the datasets used in this study is provided in Table 1.R e s u l t sf r o mN y h o l te ta l .(2012) included summary statistics of rs7521902, rs13394619, rs4141819, rs7739264, rs12700667, rs1537377, and rs10859871 obtained from the European (QIMRHCS+OX) and Japanese (BBJ) GWA data (Table 1). Similarly, we included results of rs13394619 and rs6542095 from Adachi et al. ( 2010), obtained from the combined analysis of Affymetrix 500K and 6.0 arrays in 696 cases and 825 controls of Japanese ancestry. Furthermore,

from Albertsen et al. ( 2013) included summary statistics of rs1519761 in their discovery and replication stages. Finally, we obtained summary results of rs6542095 in the QIMRHCS, OX, and BBJ imputed data from Sapkota et al. (2015b). Initial meta-analysis was conducted using a fixed-effect (inverse variance-weighted) model implemented in the GWAMA program (Magi & Morris, 2010). Heterogeneity of allelic associations was examined using the Cochran’s Q statistic p het < .1 (Cochran, 1954), as well as the I2 index (Ioannidis et al., 2007), which indicates the proportion of variance attributable to between-study heterogeneity. Meta- analysis of SNPs associated in fixed-effect model with an evidence of heterogeneity ( p < .1) was carried out using the Han–Eskin random-effects model (RE2; Han & Eskin, 2011) implemented in the METASOFT program. In con- trast with the conventional random-effects model, the RE2 model increases power under heterogeneity. Furthermore, additional meta-analysis for the nine SNP loci was also performed by restricting to ‘Grade_B’ endometriosis cases (wherever available) versus controls.

Following the QC steps, a total of 998 endometriosis cases and 783 disease-free controls with 316,467 markers 520 TWIN RESEARCH AND HUMAN GENETICS https://doi.org/10.1017/thg.2015.61 Published online by Cambridge University Press Replication Study for Endometriosis Risk Loci TABLE 2 Summary Results of the Nine Known SNP Loci for Endometriosis in the Current Study ‘All’ ‘Grade_B’ Chr SNP Position RA OA RAF cases RAFcontrols OR (95% CI) p RAFcases RAFcontrols OR (95% CI) p 1 rs7521902 22490724 A C 0.243 0.221 1.13 (0.97–1.32) .125 0.271 0.221 1.30 (1.08–1.58) .007 2 rs13394619 11727507 G A 0.538 0.503 1.14 (1.00–1.30) .045 0.533 0.503 1.13 (0.95–1.33) .164 2 rs4141819 67864675 C T 0.307 0.328 0.91 (0.79–1.05) .183 0.311 0.328 0.92 (0.77–1.11) .384 2 rs6542095 113529183 C T 0.324 0.295 1.14 (0.99–1.32) .066 0.346 0.295 1.26 (1.06–1.51) .010 2 rs1519761 151633204 G A 0.410 0.414 0.98 (0.86–1.13) .814 0.408 0.414 0.97 (0.82–1.16) .755 6 rs7739264 19785588 T C 0.520 0.515 1.02 (0.89–1.16) .806 0.545 0.515 1.12 (0.95–1.32) .172 7 rs12700667 25901639 A G 0.758 0.747 1.06 (0.91–1.24) .445 0.772 0.747 1.15 (0.94–1.40) .166 9 rs1537377 22169700 C T 0.391 0.413 0.91 (0.79–1.04) .178 0.415 0.413 1.01 (0.85–1.20) .935 12 rs10859871 95711876 C A 0.322 0.327 0.98 (0.84–1.13) .747 0.329 0.327 1.01 (0.84–1.21) .915 Note: Chr = chromosome; Position = chromosomal position (bp) based on Human Build 37 (GRCh37/hg19); RA = risk allele from original study; OA = other allele; OR = odds ratio; CI = confidence interval. remained in the Belgian GWA data for downstream anal- ysis. Of these, 246,071 were ‘core’ SNPs whereas 70,396 were ‘exome’ variants. The GWA analysis of observed geno- types of the 316,467 markers in the Belgian GWA study alone did not produce any genome-wide significant hits in either ‘All’ or ‘Grade_B’ analysis, with few suggestive (p .95. We also com- pared imputed genotypes (dosage scores) of six SNP loci (rs7521902, rs13394619, rs4141819, rs6542095, rs7739264, and rs12700667) for endometriosis with the observed true genotypes available in the Belgian data. Genotype concor- dances (as measured by the Pearson’s correlation coeffi- cient) between two sets of genotypes for the six SNPs were >0.99 (p < 2.2 × 10 −16). Association analysis of the dosage scores of the nine im- plicated SNP loci in the Belgian data provided further in- sights into the associations of these SNPs with endometrio- sis. Risk alleles and their frequencies of all nine SNPs were similar to the ones reported in the original studies (Table 2; Albertsen et al.,2013; Nyholt et al.,2012; Painter et al.,2011; Sapkota et al., 2015b) and their associations were stronger with ‘Grade_B’ than ‘All’ endometriosis. Furthermore, ef- fect directions of seven out of nine tested SNPs in either ‘All’ or ‘Grade_B’ endometriosis were in line with the published results. Three SNPs showed statistically significant associa- tion with endometriosis in either ‘All’ or ‘Grade_B’ disease at a nominal p < .05. SNP rs7521902 showed borderline marginal association (OR = 1.13; p = .12) with ‘All’ en- dometriosis. As expected, its association was stronger and statistically significant (OR= 1.30; p= .007) with ‘Grade_B’ cases. A statistically significant association (OR= 1.14; p = .045) for rs13394619 was also observed for ‘All’ endometrio- sis; however, the signal was slightly weaker (OR = 1.13; p = .164) in ‘Grade_B’ cases. A borderline association (OR = 1.14; p = .06) with ‘All’ endometriosis was observed for rs6542095, which was stronger and significant (OR= 1.26; p = .01) in ‘Grade_B’ cases. Meta-analysis, including imputed data from Belgian co- hort and the published results, provide insights into SNP loci associated with endometriosis. Six SNP loci showed associations with either ‘All’ or ‘Grade_B’ endometriosis at genome-wide significance level ( p < 5 × 10 −8)a n dw i t h similar directions of effect across all studies included in the analysis (Table 3). Of these, three SNPs were associated with both ‘All’ and ‘Grade_B’ endometriosis, with a genome- wide significant evidence in the fixed-effect meta-analysis. These included: SNP rs7521902 near WNT4 (‘All’ , OR= 1.17; 95% CI = 1.11–1.23; p = 3.63 × 10 −8;‘ G r a d e _ B ’ ,O R = 1.25; 95% CI= 1.17–1.34; p = 1.72 × 10−10), rs13394619 in GREB1 (‘All’ , OR= 1.15; 95% CI = 1.10–1.20; p = 9.13 × 10−9;‘ G r a d e _ B ’ ,O R= 1.17; 95% CI = 1.11–1.24; p = 3.02 × 10−8), and rs12700667 at 7p15.2 (‘All’ , OR= 1.19; 95% CI = 1.13–1.26; p = 7.10 × 10−10;‘ G r a d e _ B ’ ,O R= 1.29; 95% CI = 1.20–1.39; p = 1.47 × 10−11). The IL1A SNP (rs6542095) was genome-wide significantly associated with only ‘Grade_B’ (OR = 1.22; 95% CI = 1.14–1.30; p = 1.00 × 10 −9) endometriosis in fixed-effect meta-analysis, but after appropriate modeling for between-study hetero- geneity in ‘All’ endometriosis ( p het = .007) in the RE2 model, the association reached genome-wide significance (p = 3.35 × 10 −8). Statistical significance of association of rs6542095 with ‘Grade_B’ endometriosis also became stronger ( p = 4.90 × 10 −10)i nt h eR E 2m o d e la f t e ra c - counting for between-study heterogeneity ( phet = .01). A strong association between rs7739264 near ID4 and ‘All’ endometriosis (p = 1.93 × 10−7) was observed, and the sig- nal was further enriched in ‘Grade_B’ endometriosis (OR = 1.20; 95% CI = 1.12–1.27; p = 1.98 × 10 −8), achieving a genome-wide significance. Similarly, near genome-wide significant evidence for association between rs1537377 near CDKN2B-AS1 and ‘Grade_B’ endometriosis (OR = 1.19; 95% CI= 1.12–1.27; p = 9.27 × 10 −8)w a so b s e r v e d ,w h i c h was genome-wide significant (p = 4.80 × 10−8)a f t e rm o d - eling for borderline between-study heterogeneity ( phet = .1). However, the effect of rs1537377 in ‘All’ endometriosis was in opposite direction of the published results (Table 2). While the remaining three SNPs (rs4141819, rs1519761, and rs10859871) did not produce genome-wide signifi- cant evidence for association with either ‘All’ or ‘Grade_B’ TWIN RESEARCH AND HUMAN GENETICS 521 https://doi.org/10.1017/thg.2015.61 Published online by Cambridge University Press Yadav Sapkota et al. TABLE 3 Meta-Analysis for the Nine Known Endometriosis SNP Loci After Combining Summary Statistics From Current Study With the Published Results Position No. of Chr SNP (bp) RA OA Cases OR (95% CI) pp het I2 studies Direction RE2 p 1 rs7521902 22490724 A C ‘All’ 1.17 (1.11–1.23) 3.63 × 10−8 0.533 0.00 4 ++++ NC ‘Grade_B’ 1.25 (1.17–1.34) 1.72 × 10−10 0.762 0.00 4 ++++ NC 2 rs13394619 11727507 G A ‘All’ 1.15 (1.10–1.20) 9.13 × 10−9 0.183 0.36 5 +++++ NC ‘Grade_B’ 1.17 (1.11–1.24) 3.02 × 10−8 0.243 0.27 5 +++++ NC 2 rs4141819 67864675 C T ‘All’ 1.12 (1.07–1.19) 2.38 × 10−5 0.017 0.71 4 +++-1 . 8 8 × 10−5 ‘Grade_B’ 1.17 (1.09–1.25) 1.01 × 10−5 0.004 0.77 4 +++-3 . 6 3 × 10−7 2 rs6542095 113529183 C T ‘All’ 1.15 (1.09–1.21) 1.05 × 10−7 0.007 0.72 5 +++++ 3.35 × 10−8 ‘Grade_B’ 1.22 (1.14–1.30) 1.00 × 10−9 0.014 0.68 5 +++++ 4.90 × 10−10 2 rs1519761 151633204 G A ‘All’ 1.15 (1.08–1.22) 5.62 × 10−6 0.025 0.73 3 ++-1 . 0 7 × 10−5 ‘Grade_B’ 1.17 (1.09–1.24) 1.94 × 10−6 0.067 0.63 3 ++-7 . 9 9 × 10−6 6 rs7739264 19785588 T C ‘All’ 1.14 (1.08–1.20) 1.93 × 10−7 0.359 0.07 4 ++++ NC ‘Grade_B’ 1.20 (1.12–1.27) 1.98 × 10−8 0.756 0.00 4 ++++ NC 7 rs12700667 25901639 A G ‘All’ 1.19 (1.13–1.26) 7.10 × 10−10 0.405 0.00 4 ++++ NC ‘Grade_B’ 1.29 (1.20–1.39) 1.47 × 10−11 0.338 0.11 4 ++++ NC 9 rs1537377 22169700 C T ‘All’ 1.10 (1.05–1.16) 9.88 × 10−5 0.041 0.64 4 +++-1 . 0 0 × 10−4 ‘Grade_B’ 1.19 (1.12–1.27) 9.27 × 10−8 0.101 0.52 4 ++++ 4.80 × 10−8 12 rs10859871 95711876 C A ‘All’ 1.16 (1.09–1.22) 4.29 × 10−7 0.137 0.46 4 +++-N C ‘Grade_B’ 1.17 (1.10–1.25) 2.46 × 10−6 0.385 0.02 4 ++++ NC Note: Chr = chromosome; Position = chromosomal position (bp) based on Human Build 37 (GRCh37/hg19); RA = risk allele; OA = other allele; OR = odds ratio; CI = confidence interval; Phet = Cochran’s Q between-study heterogeneity test p value; I2, percentage of variance attributable to between-study heterogeneity; RE2 = Han Eskin’s random effects model meta-analysis; NC = not calculated. endometriosis in fixed-effect meta-analysis, they still showed strong associations with the disease ( p < 2.38 × 10−5). Nonetheless, the effects of rs4141819 and rs1519761 in both ‘All’ and ‘Grade_B’ endometriosis, and that of rs10859871 in ‘All’ endometriosis were in opposite direc- t i o n so ft h ep u b l i s h e dr e s u l t s(Table 2). SNP rs4141819 showed between-study heterogeneity ( p het < .01) in both ‘All’ and ‘Grade_B’ endometriosis, and after accounting for this heterogeneity in the RE2 model, association of rs4141819 with ‘Grade_B’ disease became stronger with near genome-wide significant evidence (p = 3.63 × 10 −7). Significant between-study heterogeneity for rs1519761 was also observed in both ‘All’ and ‘Grade_B’ endometriosis, but its association with the disease (‘All’ ,p = 5.62 × 10 −6; ‘Grade_B’ ,p = 1.94 × 10−6) slightly diluted in the RE2 model (‘All’ , p = 1.07 × 10−5;‘ G r a d e _ B ’ ,p = 7.99 × 10−6). A near genome-wide significant association between rs10859871 near VEZT and ‘All’ endometriosis (OR= 1.16; 95% CI = 1.09–1.22; p = 4.29 × 10−7) was observed, with slightly larger effect size (OR = 1.17; 95% CI = 1.10–1.25) in ‘Grade_B’ disease, although statistical significance of the signal was weaker (p = 2.46 × 10 −6).

Endometriosis is a complex disease and studies have shown that genetic risk factors substantially contribute to the risk of endometriosis. Genetic studies, especially the GWA stud- ies for endometriosis, have identified 10 SNP loci, of which nine are polymorphic in the populations of European ori- gin (Albertsen et al., 2013; Nyholt et al.,2012; Painter et al., 2011; Sapkota et al., 2015b). While much larger and well- powered GWA studies are needed to identify additional genetic risk factors involved in the risk of endometriosis, replication studies are crucial to provide credibility that the initial genotype–phenotype associations are valid. Repeated observation of such associations in independent popula- tions of similar ethnicity adds evidence that the associations are not due to chance alone. The previous three replication studies for endometriosis have investigated only a handful of the 10 implicated SNP loci to date. Here we report the most comprehensive replication study performed to date, in which we examine all nine implicated SNP risk loci for endometriosis that are polymorphic in populations of Eu- ropean ancestry, by utilizing GWA data in uniquely char- acterized 998 endometriosis cases and 783 controls from Belgium. The risk alleles and their frequencies for all the nine SNPs in the Belgian replication cohort were comparable with the original studies (Table 2; Albertsen et al.,2013;N y - holt et al., 2012; Painter et al., 2011; Sapkota et al., 2015b). Moreover, direction of effects for seven of the nine SNPs for either ‘All’ or ‘Grade_B’ endometriosis was also con- sistent with the published results. Among these, we could successfully replicate associations of three SNPs (rs7521902, rs13394619, and rs6542095) with either ‘All’ or ‘Grade_B’ endometriosis at nominalp < .05, which is more often than by chance alone (p = .008; one-sided binomial test). Signif- icant association of rs6542095 at the IL1A locus with ‘All’ (p = .066) and ‘Grade_B’ (p = .01) endometriosis is note- worthy as this is the first successful replication in an inde- pendent population, providing further supporting evidence for a potential link between inflammation and endometrio- sis pathogenesis. More importantly, all the SNPs showed larger effect sizes with ‘Grade_B’ than ‘All’ endometriosis — an observation consistent with the previous reports 522 TWIN RESEARCH AND HUMAN GENETICS https://doi.org/10.1017/thg.2015.61 Published online by Cambridge University Press Replication Study for Endometriosis Risk Loci TABLE 4 Association Statistics for the ‘Best’ SNPs at the Six Genome-Wide Significant Loci for Endometriosis Reported in Nyholt et al. (2012) ‘All’ ‘Grade_B’ Chr SNP Position RA OA RAF cases RAFcontrols OR (95% CI) p RAFcases RAFcontrols OR (95% CI) p 1 rs56318008 22470407 T C 0.178 0.151 1.23 (1.02–1.48) .031 0.190 0.151 1.33 (1.06–1.66) .014 2 rs77294520 11660955 C G 0.166 0.147 1.18 (0.97–1.44) .098 0.175 0.147 1.28 (1.00–1.63) .051 2 rs2861694 67845739 G A 0.692 0.670 1.11 (0.96–1.28) .161 0.688 0.670 1.09 (0.91–1.30) .368 6 rs6901079 19776659 C T 0.249 0.245 1.02 (0.88–1.19) .762 0.260 0.245 1.09 (0.90–1.32) .402 9 rs7041895 22162794 A C 0.599 0.583 1.07 (0.93–1.23) .327 0.576 0.583 0.97 (0.81–1.15) .716 12 rs11107968 95690444 A G 0.680 0.676 1.02 (0.89–1.18) .779 0.672 0.676 0.98 (0.82–1.17) .846 Note: Chr = chromosome; Position = chromosomal position (bp) based on Human Build 37 (GRCh37/hg19); RA = risk allele; OA = other allele; OR = odds ratio; CI = confidence interval. supporting greater genetic loading in moderate-to-severe disease (Nyholt et al., 2012; Painter et al., 2011;S a p k o t a et al., 2015a). Our meta-analysis, including results from the cur- rent replication study and the published results, pro- duced genome-wide significant evidence for six (rs7521902 near WNT4, rs13394619 in GREB1, rs6542095 in IL1A, rs7739264 near ID4, rs12700667 at 7p15.2, and rs1537377 near CDKN2B-AS1) of the nine implicated SNPs in ei- ther ‘All’ or ‘Grade_B’ endometriosis, after accounting for between-study heterogeneity using the RE2 model, wher- ever appropriate (Table 3). With the exception of rs1519761 at 2q23.3 reported by Albertsen et al. (2013), the other two SNP loci (rs4141819 at 2p14 and rs10859871 near VEZT) also showed near genome-wide significance for ‘Grade_B’ endometriosis in the RE2 model (p = 3.63 × 10 −7)a n df o r ‘All’ endometriosis in the fixed-effect model ( p = 4.29 × 10−7). The association signal for rs1519761 was the weak- est (‘All’ ,p = 5.62 × 10−6;‘ G r a d e _ B ’ ,p = 1.94 × 10−6) among the nine risk loci, and the signal was slightly diluted after accounting for observed between-study heterogene- ity in the RE2 model (‘All’ , p = 1.07 × 10 −5;‘ G r a d e _ B ’ ,p = 7.99 × 10−6). Association signal at this locus was also not replicated in a recent meta-analysis for endometriosis (Rahmioglu et al., 2014), suggesting that further investiga- tion is required to confirm a role for this locus in the risk of endometriosis. In our multi-ethnic GWA meta-analysis that strongly associated seven risk loci with endometriosis, we found stronger associations at six loci (rs56318008 at 1p36.12, rs77294520 at 2p25.1, rs2861694 at 2p14, rs6901079 at 6p22.3, rs7041895 at 9p21.3, and rs11107968 at 12q22) when we imputed genotypes in the region 2,500 kb up- stream and downstream of the most significant geno- typed SNP using the full reference panel from the 1,000 Genomes Project Interim Phase 1 Haplotypes (2010–2011 data freeze). For the risk loci at 7p15.2, the genotyped SNP rs12700667 was the best signal. For the remaining six loci with stronger association signals (‘best’ SNPs) post- imputation than the genotyped SNP , we assessed for their replication in the Belgian cohort ( Table 4). All six SNPs were accurately imputed in the current study withr 2 > .85. Association results were consistent with that of the orig- inal genotyped SNPs, as shown in Table 2, in particular for SNPs rs56318008 at 1p36 and rs77294520 at 2p25.1, which showed nominally significant associations with ‘All’ (p < .098) and ‘Grade_B’ ( p < .051), providing further supporting evidence for implication of these risk loci in endometriosis. As a first step to help identify causal variants at nine SNP loci, we interrogated the ExomeChip data for puta- tively functional coding variants within genes harboring or closest to GWA SNPs. For the GREB1 locus, we found an- other coding variant, rs10929757, showing nominally sig- nificant association ( p = .015) with endometriosis. The effect size of rs10929757 was similar (OR = 1.18) to the GWA SNP rs13394619, although they are poorly correlated (r 2 = .25). Similarly, we also observed nominally significant association (p < .018) for two coding variants (rs2383207 and rs4977574) in CDKN2B-AS1 —t h ec l o s e s tg e n et ot h e GWA SNP rs1537377 at 9p21.3. In spite of lack of corre- lation ( r 2 = .011 and .005, respectively) with rs1537377, the effect sizes for both variants were similar (ORs = 1.17 and 1.19 respectively). While these data may suggest inde- pendent association signals at GREB1 and 9p21.3 loci, the three coding variants need to be further investigated in a larger sample size for a more conclusive interpretation. We did not observe evidence of association for other coding variants at nominal p < .05, even though the effect sizes f o rs o m ew e r ec o m p a r a b l ew i t hG W AS N P sa te a c hr i s k loci (data not shown). We did not detect any rare coding variants at GWA loci despite the ExomeChip data. This may be due, in part, to reduced power in the Belgian sample to detect such rare variants. Assuming a disease prevalence of 8%, our sample size only had 45% power to detect alle- les of frequency .20 contributing to genotype relative risk of 1.15 (Purcell et al., 2003). As such, larger ExomeChip studies may be required to adequately investigate potential role of coding/rare variants in the risk of endometriosis and other complex traits. We cannot rule out the possibil- ity of other types of rare functional variants at these loci, which are not adequately captured by either ExomeChip or current imputation methods, contributing to increased risk of endometriosis. These issues may be addressed by TWIN RESEARCH AND HUMAN GENETICS 523 https://doi.org/10.1017/thg.2015.61 Published online by Cambridge University Press Yadav Sapkota et al. the future studies utilizing larger sample sizes, coupled with re-sequencing and further fine-mapping required to identify causal variants within the implicated GWA loci. Furthermore, controls used in this study were clinic-based endometriosis-free individuals who presented with symp- toms of sub-fertility. As such, they may have different allele frequencies as compared with ‘population-based’ controls used in most GWA studies, and hence may partly explain the opposite direction of effect sizes observed for some SNPs. However, this needs to be investigated further using a larger sample size with population-based controls, and therefore caution should be used interpreting these results. Overall, results from the current replication study pro- vide further supporting evidence for associations of the implicated SNP loci with endometriosis. Meta-analysis for these loci after including additional published results produced genome-wide significant evidence for six loci, with similar magnitudes and directions of effect across studies, and hence provided further evidence against any possibility of inflated genetic effects due to the ‘winner’s curse’ bias in the original study. More importantly, all the nine SNPs showed larger effect sizes with stage III/IV en- dometriosis than all cases, corroborating our previous ob- servation for greater genetic loading in moderate-to-severe endometriosis. Acknowledgments We would like to thank all the participants in the en- dometriosis studies who were included in this analysis. We also thank many hospital directors and staff, gynecologists, general practitioners, and pathology services who provided assistance with confirmation of diagnoses. Dale R. Nyholt was supported by an NHMRC Fellowship (613674) and ARC Future Fellowship (FT0991022) schemes, and Grant W. Montgomery was supported by the NHMRC Fellow- ships Scheme (339446, 619667).

A d a c h i ,S . ,T a j i m a ,A . ,Q u a n ,J . ,H a i n o ,K . ,Y o s h i h a r a ,K . , Masuzaki, H., . . . Tanaka, K. (2010). Meta-analysis of genome-wide association scans for genetic susceptibility to endometriosis in Japanese population.Journal of Human Genetics, 55, 816–821. Albertsen, H. M., Chettier, R., Farrington, P ., & Ward, K. (2013). Genome-wide association study link novel loci to endometriosis. PLoS One, 8, e58257. American Society for Reproductive Medicine. (1997). Revised American Society for Reproductive Medicine classification of endometriosis: 1996. Fertility and Sterility, 67 , 817–821. Cochran, W. G. (1954). The combination of estimates from different experiments. Biometrics, 10 , 101–129. Delaneau, O., Marchini, J., & Zagury, J. F. (2012). A linear com- plexity phasing method for thousands of genomes. Nature Methods, 9, 179–181. Fung, J. N., Holdsworth-Carson, S. J., Sapkota, Y., Zhao, Z. Z., Jones, L., Girling, J.E., . . . Montgomery, G. W. (2015). Functional evaluation of genetic variants associated with endometriosis near GREB1. Human Reproduction, 30 , 1263–1275. Gao, X., Outley, J., Botteman, M., Spalding, J., Simon, J. A., & Pashos, C. L. (2006). Economic burden of endometriosis. Fertility and Sterility, 86 , 1561–1572. Goldstein, J. I., Crenshaw, A., Carey, J., Grant, G. B., Maguire, J., Fromer, M., . . . Neale, B. M. (2012). zCall: A rare variant caller for array-based genotyping: Genetics and population analysis. Bioinformatics, 28, 2543–2545. Han, B., & Eskin, E. (2011). Random-effects model aimed at discovering associations in meta-analysis of genome-wide association studies. American Journal of Human Genetics, 88, 586–598. Hata, Y., Nakaoka, H., Y oshihara, K., Adachi, S., Haino, K., Y amaguchi, M., . . . Tanaka, K. (2013). A non-synonymous variant of IL1A is associated with endometriosis in Japanese population. Journal of Human Genetics, 58 , 517– 520. Ioannidis, J. P ., Patsopoulos, N. A., & Evangelou, E. (2007). Heterogeneity in meta-analyses of genome-wide associa- tion investigations. PLoS One, 2, e841. Li, Y., Willer, C. J., Ding, J., Scheet, P ., & Abecasis, G. R. (2010). MaCH: Using sequence and genotype data to estimate hap- lotypes and unobserved genotypes. Genetic Epidemiology, 34, 816–834. Li, Y., Willer, C., Sanna, S., & Abecasis, G. (2009). Genotype imputation. Annual Review of Genomics and Human Genet- ics, 10, 387–406. Magi, R., & Morris, A. P . (2010). GWAMA: Software for genome-wide association meta-analysis.BMC Bioinformat- ics, 11, 288. Nyholt, D. R., Low, S. K., Anderson, C. A., Painter, J. N., Uno, S., Morris, A. P ., . . . Montgomery, G. W. (2012). Genome- wide association meta-analysis identifies new endometrio- sis risk loci. Nature Genetics, 44, 1355–1359. Pagliardini, L., Gentilini, D., Sanchez, A. M., Candiani, M., Vigano, P ., & Di Blasio, A. M. (2015). Replication and meta- analysis of previous genome-wide association studies con- firm vezatin as the locus with the strongest evidence for association with endometriosis. Human Reproduction, 30 , 987–993. Pagliardini, L., Gentilini, D., Vigano, P ., Panina-Bordignon, P ., Busacca, M., Candiani, M., . . . Di Blasio, A. M. (2013). An Italian association study and meta-analysis with previous GWAS confirm WNT4, CDKN2BAS and FN1 as the first identified susceptibility loci for endometriosis. Journal of Medical Genetics, 50, 43–46. Painter, J. N., Anderson, C. A., Nyholt, D. R., Macgregor, S., Lin, J., Lee, S. H., . . . Zondervan, K. T. (2011). Genome- wide association study identifies a locus at 7p15.2 associated with endometriosis. Nature Genetics, 43, 51–54. Purcell, S., Cherny, S. S., & Sham, P . C. (2003). Genetic power calculator: Design of linkage and association genetic map- ping studies of complex traits. Bioinformatics, 19 , 149– 150. 524 TWIN RESEARCH AND HUMAN GENETICS https://doi.org/10.1017/thg.2015.61 Published online by Cambridge University Press Replication Study for Endometriosis Risk Loci Rahmioglu, N., Nyholt, D. R., Morris, A. P ., Missmer, S. A., Montgomery, G. W., & Zondervan, K. T. (2014). Genetic variants underlying risk of endometriosis: Insights from meta-analysis of eight genome-wide association and repli- cation datasets. Human Reproduction Update, 20, 702–716. Sapkota, Y., Attia, J., Gordon, S. D., Henders, A. K., Holliday, E. G., Rahmioglu, N., . . . Nyholt, D. R. (2015a). Genetic burden associated with varying degrees of disease severity in endometriosis. Molecular Human Reproduction, 21, 594– 602. S a p k o t a ,Y . ,L o w ,S .K . ,A t t i a ,J . ,G o r d o n ,S .D . ,H e n d e r s ,A .K . , Holliday, E. G., . . . Nyholt, D. R. (2015b). Association be- tween endometriosis and the interleukin 1A (IL1A) locus. Human Reproduction, 30, 239–248. Sundqvist, J., Xu, H., Vodolazkaia, A., Fassbender, A., Kyama, C., Bokor, A., . . . Falconer, H. (2013). Replica- tion of endometriosis-associated single-nucleotide poly- morphisms from genome-wide association studies in a Caucasian population. Human Reproduction, 28, 835–839. Treloar, S. A., O’Connor, D. T., O’Connor, V . M., & Martin, N. G. (1999). Genetic influences on endometriosis in an Australian twin sample. Fertility and Sterility, 71 , 701– 710. Uno, S., Zembutsu, H., Hirasawa, A., Takahashi, A., Kubo, M., Akahane, T., . . . Nakamura, Y. (2010). A genome- wide association study identifies genetic variants in the CDKN2BAS locus associated with endometriosis in Japanese. Nature Genetics, 42, 707–710.TWIN RESEARCH AND HUMAN GENETICS 525 https://doi.org/10.1017/thg.2015.61 Published online by Cambridge University Press