Genome-wide association study identifies a locus at 7p15.2 associated with endometriosis.

Endometriosis is a common gynecological disease associated with pelvic pain and subfertility. We conducted a genome-wide association study (GWAS) in 3,194 individuals with surgically confirmed endometriosis (cases) and 7,060 controls from Australia and the UK. Polygenic predictive modeling showed significantly increased genetic loading among 1,364 cases with moderate to severe endometriosis. The strongest association signal was on 7p15.2 (rs12700667) for 'all' endometriosis (P = 2.6 × 10⁻⁷, odds ratio (OR) = 1.22, 95% CI 1.13-1.32) and for moderate to severe disease (P = 1.5 × 10⁻⁹, OR = 1.38, 95% CI 1.24-1.53). We replicated rs12700667 in an independent cohort from the United States of 2,392 self-reported, surgically confirmed endometriosis cases and 2,271 controls (P = 1.2 × 10⁻³, OR = 1.17, 95% CI 1.06-1.28), resulting in a genome-wide significant P value of 1.4 × 10⁻⁹ (OR = 1.20, 95% CI 1.13-1.27) for 'all' endometriosis in our combined datasets of 5,586 cases and 9,331 controls. rs12700667 is located in an intergenic region upstream of the plausible candidate genes NFE2L3 and HOXA10.

Endometriosis [MIM 131200] is a disease affecting 6-10% of women of reproductive age1 with significant annual health costs2 and health burden for individuals3,4. Common symptoms include chronic pelvic pain, severe dysmenorrhea (painful periods), and sub-fertility. The causes of endometriosis remain uncertain despite over 50 years of hypothesis-driven research. Disease severity is classified using the revised American Fertility Society (rAFS) system5, assigning patients to one of four stages (I–IV, minimal-severe) based on lesion size and associated pelvic adhesions. However, it remains unclear whether the disease progresses through these stages, and it has been suggested that small lesions (stages I-II) represent an epiphenomenon rather than a disease entity6. Endometriosis risk is influenced by genetic factors7-14, with heritability estimated around 51%9.

We genotyped 3,194 unrelated cases with surgically confirmed endometriosis, recruited by the International Endogene Consortium, IEC (QIMR, Australia dataset = 2,270; Oxford, UK dataset = 924)15, using the Illumina Human670Quad Beadarray (Online Methods). Disease stage was assessed from surgical records using the rAFS classification system5,15 and grouped into two phenotypes: stage A (stage I-II or some ovarian disease plus a few adhesions; N = 1,686, 52.7%), and stage B (stage III–IV disease; N = 1,364, 42.7%), or unknown (N = 144, 4.6%) (Supplementary Table 1). Illumina Human610Quad control genotypes for QIMR cases were available for 1,870 individuals in an adolescent twin study16,17. For Oxford cases, Illumina Human1M-Duo genotypes for 5,190 UK population controls were obtained from the Wellcome Trust Case Control Consortium (WTCCC2). Although endometriosis affects women, Australian and UK control sets included men to maximise power of association detection on autosomal chromosomes (Online Methods). No significant autosomal allele frequency differences were detected between male and female control samples (Supplementary Fig. 1), indicating that association signals would not be influenced by differing female/male ratio in the cases and controls.

Studies to date have established that endometriosis is heritable, but have not addressed genetic burden for different disease stages. We used the GWA data to assess genetic loading in cases in two complementary ways. Using a novel method18 we estimated the proportion of variation in case-control status that can be explained by considering all SNPs simultaneously through inference of distant relatedness from marker data and comparing it to case-control status (Online Methods). The proportion of variation in case-control status explained by the GWA data was highly significant for both “all” and stage B endometriosis (Table 1; Supplementary Table 2). The estimate for stage B (0.34, SE: 0.04) was significantly higher than for stage A (0.15, SE: 0.04 (Table 1)).

Table 1

Estimates of proportion of variation due to common genetic variants for “all” endometriosis and stage A or B disease using genome-wide SNP data from cases and controls

Phenotypes	Cases	Controls	Proportion ofvariation (SE)	P valuea
All endometriosis	3154	6981	0.27 (0.04)	4.4 × 10−16
Stage B	1347	6981	0.34 (0.04)	4.4 × 10−16
Stage A	1666	6981	0.15 (0.04)	2.6 × 10−4

Proportion of variation and associated P values for the likelihood ratio test were estimated using a linear mixed model incorporating 203,826 SNPs from the GWA panel after additional QC. Case and control numbers are slightly lower than for the GWA analyses due to the stricter QC measures (Online Methods). Stage A and stage B estimates of the variance explained are significantly different from each other (P = 1.8 × 10−3, using a two sample t-test which is conservative since the control samples are the same). Results were verified by prediction of individual genetic risk using QIMR and Oxford as alternate “discovery” and “target” datasets (Supplementary Table 2).

We also assessed the genetic loading of the different stages using a “prediction” approach (Online Methods)19 in which we used the Oxford data as a “discovery” set to identify increasingly large SNP sets ranked on significance of association (“allele specific scores”), and used these scores to predict disease status in “target” samples from QIMR. The discovery and target sets were then reversed (Supplementary Fig. 2). Oxford “all” endometriosis predicted endometriosis in the QIMR sample, with the smallest P value (8.4 × 10−6) obtained for a score set including ~75% of SNPs (Fig. 1). This result was highly significant, although the proportion of variance explained was small (maximum Nagelkerke R2 of 0.007; 0.7% of the variance). For stage B cases the proportion of variance explained by most score sets was higher (e.g. the score set including the ~20% most associated SNPs (P = 3.5 × 10−7) explained 1.3% of the variance, consistent with a greater (polygenic) genetic loading for stage B disease.

Figure 1

Allele specific score prediction for (a) “all” endometriosis and (b) stage B endometriosis, using Oxford as the “discovery” and QIMR as the “target” dataset. Variance explained in the target dataset on the basis of allele specific scores derived in the discovery dataset for eight significance thresholds (P<0.01, P<0.05, P<0.1, P<0.2, P<0.3, P<0.4, P<0.5, P<0.75, plotted left to right in each study). The y-axis indicates Nagelkerke’s pseudo R2 representing the proportion of variance explained. The number above each bar is the P value for the target dataset analysis. This figure shows that the results were not driven by a few highly associated regions, indicating a substantial number of common variants underlying disease.

We performed two GWA analyses stratified by dataset (QIMR and Oxford) using: 1) “all” 3,194 endometriosis cases, and 2) 1,364 stage B cases, given their substantially greater genetic loading (Online Methods). For ‘all’ endometriosis, the strongest signal was observed for rs12700667 in an inter-genic region on chromosome 7p15.2 (P = 2.6 × 10−7, OR = 1.22 (1.13-1.32), Table 2). As predicted from our quantitative genetic analyses, stronger signals of association across the genome were observed for stage B disease compared to ‘all’ endometriosis (Supplementary Figure 3). The 7p15.2 signal for stage B endometriosis was considerably stronger producing P = 1.5 × 10−9, OR = 1.38 (1.24-1.53) (Table 2) for rs12700667, and P = 6.0 × 10−8, OR = 1.34 (1.21-1.49) for nearby rs7798431 (r2 = 0.87). A second strong association was found for rs1250248 (2q35) within the FN1 gene (P = 3.2 × 10−8) (Supplementary Table 3). Results for SNPs rs12700667, rs7798431 and rs1250248 remained genome-wide significant after adjustment for multiple testing in the two non-independent GWA analyses via permutation (Online Methods). Only one of the permuted GWAs produced an independent P value less than that observed for rs12700667 (P = 0.001). The SNPs rs12700667 and rs7798431 lie in a narrow region of strong LD (r2 > 0.8) that extends for approximately 48 kb. Following imputation with 1000 Genomes and HapMap data (Fig 2; Supplementary Methods), conditioning on the effect of rs12700667 in logistic regression analysis showed no other independent associations with “all” or stage B endometriosis in the region.

Table 2

GWA, replication and meta-analysis results for rs12700667

Analysis	Number ofCases/Controls	Risk allele (A)frequency incontrols	P value	OR (95% CIs)	Heterogeneitytest P value
1. GWA – All endometriosis
QIMR	2270/1870	0.73	1.5 × 10−5	1.25 (1.13-1.38)	-
Oxford	924/5190	0.74	3.9 × 10−3	1.19 (1.06-1.34)	-
Combined	3194/7060	0.74	2.6 × 10−7	1.22 (1.13-1.32)	0.56
2. GWA – Stage B
QIMR	910/1870	0.73	8.3 × 10−7	1.40 (1.22-1.60)	-
Oxford	454/5190	0.74	4.2 × 10−4	1.35 (1.14-1.60)	-
Combined	1364/7060	0.74	1.5 × 10−9	1.38 (1.24-1.53)	0.75
3. Replication NHSII –All endometriosis a	2392/2271	0.73	1.2 × 10−3	1.17 (1.06-1.28)	-
4. Meta-analysis –  All Endometriosis (1+3)	5586/9331	0.74	1.4 × 10−9	1.20 (1.13-1.27)	0.64

Stage was unknown for cases in the NHSII replication cohort, though estimated to include ~40% stage B21.

Figure 2

Evidence for association with (a) “all” endometriosis and (b) stage B endometriosis across the chromosome 7 region following imputation using HapMap 3 and 1000 Genomes project CEU and TSI reference panels. SNP rs12700667 is represented by a purple diamond. All other SNPs are colour coded according to the strength of LD (as measured by r2) with rs12700667.

In addition to the three genome-wide significant SNPs, we genotyped 70 SNPs producing nominal evidence of association with “all” (P < 1.0 × 10−4) or stage B endometriosis (P <1.0 × 10−4 in stage B and P <1.0 × 10−3 in “all” endometriosis analyses; Online Methods) in an independent IEC dataset encompassing 2,392 self-reported surgically confirmed cases from the Nurses’ Health Study II (NHSII) and 2,271 controls from GWAs of breast cancer20 and kidney function from the Nurses’ Health Study (NHS) I and II. Stage information was not available for NHSII cases, but the proportion likely to have stage B disease has been estimated at approximately 40%21, similar to that observed in the QIMR case set (Supplementary Table 1). Association with “all” endometriosis for the two SNPs on 7p15.2 was replicated in the US dataset, with P = 1.2 × 10−3, OR 1.17 (1.06-1.28) for rs12700667 and P = 1.6 × 10−3, OR = 1.17 (1.06-1.28) for rs7798431 (Supplementary Table 3). There was no evidence (nominal P ≤ 0.05) for replication of rs12540248 (FN1), or association with the remaining 70 SNPs (Supplementary Table 3).

Analysis of all 5,586 cases and 9,331 controls from the combined QIMR, Oxford and NHS cohorts further confirmed association between “all” endometriosis and 7p15.2, producing P values of 1.4 × 10−9 (OR = 1.20 (1.13-1.27)) for rs12700667 and 1.1 × 10−7 (OR = 1.18 (1.11-1.25)) for rs7798431 (Table 2). Although effect sizes from discovery datasets may be inflated22, the similarity of ORs for “all” endometriosis in our discovery (GWA) and replication datasets (Table 2), suggests this type of bias has not played a major role. Assuming the estimated OR of 1.20 and allele frequency of 0.74 for the rs12700667 A allele, a multiplicative risk model, and a population prevalence of 8%10,21,23, the estimated percentage of “all” endometriosis variance explained by rs12700667 is 0.36 (or 0.69% of the estimated 51% heritability of endometriosis9).

The associated SNPs are located in a ~924 kb inter-genic region containing at least one non-coding RNA (AK057379), predicted transcripts and regulatory elements and a miRNA (hsa-mir-148a) ~88 kb upstream of rs12700667. The closest gene NFE2L3, highly expressed in placenta, is located ~331 kb downstream of rs12700667. Two endometriosis candidate genes HOXA10 and HOXA1124,25, members of the homeobox A family of transcription factors that play a role in uterine development, lie ~1.35 Mb downstream.

Among reported candidate gene associations for endometriosis14 the only gene with a P value <10−3 for SNPs in the GWA data was PGR on chromosome 11 (Supplementary Table 3), but the result for this SNP was not significant in the replication stage. A recent GWA scan in Japanese women reported significant association of endometriosis with rs10965235 (P = 5.8 × 10−12, OR = 1.44) located on chromosome 9p21, and possible associations with rs13271465 on 8p22 and rs16826658 on 1p3626. The Japanese GWA study did not report our 7p15.2 signal among their 100 top SNPs followed up for replication, but with 1,423 cases and 1,318 controls they would have had only 13% power to detect the effect of rs12700667 with P ≤ 1.8 × 10−4 (Online Methods). We found no evidence for association with rs10965235 (which is monomorphic in individuals of European descent reflecting the different genetic (“ancestral”) backgrounds between the studies) or any other SNP in LD (r2 > 0.5 in HapMap JPT) in the QIMR/Oxford data (Supplementary Table 4). We also found no evidence of association to 8p22. We did find evidence for replication of SNP rs7521902 on 1p36 close to WNT4 for both “all” endometriosis (P = 9.0 × 10−5; OR = 1.16 (1.08-1.25)) and stage B (P = 7.5 × 10−6; OR = 1.25 (1.13-1.38)), the stronger signal in stage B providing additional empirical evidence for the benefit of examining stage B cases. Importantly, meta-analysis of the QIMR and Oxford “all” endometriosis OR with the reported Japanese OR of 1.25 (1.12-1.39) for rs7521902 produced a genome-wide significant P value of 4.2 × 10−8 (OR = 1.19 (1.12-1.27)). The frequency of the rs7521902 risk allele (A) was 0.57 and 0.51 in the Japanese GWAs cases and controls, and 0.26 and 0.24 in our combined GWAs cases and controls. WNT4 is important for development of the female reproductive tract27, ovarian follicle development and steroidogenesis28,29, and a plausible biological candidate.

We have identified a novel locus on chromosome 7p15.2 significantly associated with risk of endometriosis in women of European ancestry and confirm a previously reported suggestive association for SNPs close to the WNT4 locus. Our analyses also demonstrate a higher genetic loading for moderate-severe (stage B) endometriosis, and consistent with these results the strongest association signals were observed with stage B disease. Our predictive modelling demonstrates there are additional common variants contributing to risk for this disease and that future larger studies enriched for laparoscopically-confirmed moderate-severe cases will be better powered to identify risk loci and aberrant pathways contributing to the development of endometriosis.