Genetic Risk Score Mendelian Randomization Shows that Obesity Measured as Body Mass Index, but not Waist:Hip Ratio, Is Causal for Endometrial Cancer

Endometrial cancer (cancer of the lining of the uterine corpus) is the fourth most diagnosed cancer in European and North American women (1). Endometrial tumors are typically classified into two etiologic types (2): hormonally driven type I, usually low-grade endometrioid histology with “good” prognosis (∼80% of cases), and type II, nonendometrioid, largely serous, or clear cell histologies with poorer prognosis. Overall, the strongest known risk factor is obesity (3), with every 5 kg/m² increase in body mass index (BMI) increasing endometrial cancer risk by up to 60% (4). Women with a BMI ≥30 kg/m² have an approximately 3-fold overall increased endometrial cancer risk compared with nonobese women (BMI < 25), increasing to an 8-fold risk in women with BMI ≥40 (5). Obesity is most commonly associated with endometrioid endometrial cancer and may also modestly increase the risk of nonendometrioid tumors (3, 6). Body fat distribution, measured as waist–hip ratio (WHR) or waist circumference (WC), may influence endometrial cancer risk, but the evidence is weaker (4, 7). In addition, whether the WHR/WC associations are independent of BMI remains to be clarified.

Association studies assessing cancer risk with variants proven to be associated with obesity may inform our understanding of the biological relationship between obesity and cancer risk, and also identify variants/genetic loci that play a direct role in the etiology of obesity-associated cancers. Genome-wide association studies (GWAS) have now identified 97 loci associated with BMI and another 49 loci independently associated with WHR adjusted for BMI (8–11). Of these, a SNP in the FTO gene, in high linkage disequilibrium with obesity SNP rs1558902, is associated with a significantly increased risk of breast cancer (12), whereas combinations of BMI-associated variants summarized by a genetic risk score (GRS) have been associated with prostate and colorectal cancers (13, 14). A recent study of 3,376 European ancestry endometrial cancer cases and 3,867 controls found an association between a 97-SNP BMI GRS and endometrial cancer, which disappeared after adjusting for BMI (15). However, a 26-SNP BMI GRS was found to be significantly associated with endometrial cancer in Chinese cases and controls independently of measured BMI (16). The relationship between WHR-associated SNPs and endometrial cancer is as yet unknown for any population.

We have investigated whether SNPs known to influence BMI (N = 77) or WHR adjusted for BMI (N = 47) in Europeans are also associated with the risk of endometrial cancer using a large sample of 6,609 endometrial cancer cases and 37,926 controls. We present the results of our association analyses for each SNP individually, and combined as a weighted genetic risk score (wGRS; ref. 17) for each adiposity measure. Furthermore, we investigated possible pleiotropy of BMI risk SNPs using a Mendelian randomization approach with a test for heterogeneity among the causal estimates from the different SNPs.

We analyzed four datasets from separate studies contributing to the Endometrial Cancer Association Consortium, as detailed previously (18, 19) and as summarized in Supplementary Table S1. The first three comprised GWAS datasets genotyped using Illumina genotyping arrays, from Australia (“ANECS/QIMR/HCS”: 606 cases, 3,083 controls), and the United Kingdom (“SEARCH/WTCCC”: 681 cases, 5,190 controls; refs. 18, 20; “NSECG/CORGI”: 919 cases, 894 controls; refs. 19, 21). The fourth dataset (“iCOGS”) was genotyped using the “iCOGs” custom Illumina Infinium iSelect genotyping array comprising 211,155 SNPs chosen for follow-up and fine mapping of hormonal cancer GWAS hits and included 4,402 cases recruited from 11 separate studies from 7 countries and 28,758 controls from the same countries.

BMI information was available for subsets of cases and controls from the ANECS, SEARCH, and iCOGS datasets (Table 1 and Supplementary Table S2). Analyses that did not include BMI as a covariate included 6,609 cases and 37,296 controls; analyses including BMI as a covariate included 4,088 cases and 15,986 controls. The association between BMI and endometrial cancer risk was assessed by meta-analysis of the ANECS, SEARCH, and iCOGS datasets. There was modest evidence for heterogeneity (P_trend all cases I² = 73.4, P = 0.02), driven by a lower estimate for the SEARCH dataset, with little difference between a fixed effects and random effects model (presented in Table 1).

WHR information was available only for a subset of WTCCC controls (the 1958 Birth Cohort, N = 1,259); the association between WHR wGRS and WHR was confirmed in this subset of individuals. Analyses assessing the association between WHR wGRS and endometrial cancer risk included all cases and controls.

Our analyses included 77 SNPs recently validated as associated with BMI at a genome-wide level of significance (P < 5.0 × 10⁻⁸) in a large-scale meta-analysis, including 339,224 individuals of European ancestry from 125 separate studies conducted by the Genetic Investigation of Anthropomorphic Traits (GIANT) consortium (8, 9). Only SNPs significant in the primary analysis were included (i.e., we did not include SNPs significant only in secondary or conditional analyses, or in the analysis including other ancestries). Using the same criteria, we included 47 SNPs associated with WHR after adjustment for BMI in a GIANT meta-analysis, including 210,088 individuals from 101 studies (10, 11); 34 of these WHR SNPs had also reached genome-wide significance in analyses including only women (11). The BMI and WHR SNPs were nonoverlapping. SNPs that were not directly genotyped on either the Illumina or iCOGS platforms were imputed to the 1000 Genomes dataset v3 (April 2012 release) using IMPUTE v2 (22) as described in ref. 19. All target SNPs had imputation information scores >0.85 across datasets and minor allele frequencies >0.05.

The four datasets were analyzed separately using unconditional logistic regression with a per-allele (1 degree of freedom) model using SNPTEST v2 (23), adjusting for principal components of the genomic kinship matrix as described previously (18, 19). The GWAS datasets were each analyzed as a single stratum, the iCOGS dataset was adjusted for eight strata (six defined by country, while the large UK dataset was divided into “SEARCH” and “NSECG”). Given no indication for heterogeneity between studies, betas and their standard errors for each dataset were combined using standard fixed effects meta-analyses across studies in METAL (24). All statistical tests were two-sided. P < 4.0 × 10⁻⁴ (where P = 0.05/124) was considered significant.

We next tested for associations between endometrial cancer and the wGRS for BMI and WHR. For each individual in the study, the number of trait-increasing alleles at each SNP (between 0 and 2) was weighted by the reported effect size in the GIANT consortium meta-analysis (per-allele regression coefficient) on the relevant phenotype and then summed across SNPs (Supplementary Text; refs. 8–11). As most WHR-associated SNPs showed a significant difference in effect between the sexes, we calculated the WHRwGRS using the effect size as reported for women (11). The weighted contributions from all SNPs were summed to give a BMIwGRS and two different WHRwGRSs for each individual (a 34-SNP WHRwGRS including only SNPs reaching genome-wide significance in women, and a 47 SNP WHRwGRS including all WHR-associated SNPs for which we had data).

Associations between the BMIwGRS and BMI and the WHRwGRS and WHR were determined by linear regression and associations between the BMIwGRS, WHRwGRS, and case–control status by logistic regression. These analyses were performed separately for each study, and the results were combined using random effects meta-analysis. Associations between each wGRS and endometrial cancer were performed per GRS unit (continuous) and after stratifying into quartiles based on the distribution in controls. All wGRS analyses were performed using the R software package (http://www.r-project.org/) with two-sided P < 0.05 considered significant.

Finally, we used Mendelian randomization, with BMIwGRS as the instrumental variable, to assess the causality of BMI for endometrial cancer. We genetically predicted the effect of a 5 kg/m² increase in BMI on endometrial cancer risk by scaling the natural logarithm of the OR of endometrial cancer per unit increase in the BMIwGRS on BMI. Using the Mendelian randomization approach, if BMI is causal for endometrial cancer, then the observed BMI OR for endometrial cancer should be consistent with that predicted using the scaled BMIwGRS. A larger observed than predicted OR would suggest that at least part of the observed BMI–endometrial cancer association is attributable to bias or confounding, inflating the observed estimates. Conversely, a larger predicted than observed OR might indicate pleiotropy or bias or confounding that has reduced the observed estimate toward the null. To formally test the Mendelian randomization assumption of no pleiotropy, we used the Mendelian randomization adaptation of the Egger test, a method originally developed for assessing small-study bias in meta-analysis (25). In this setting, each point on the funnel plot represents the causal estimate derived from one BMI SNP, and we are testing whether the causal estimates from weaker SNPs (those less strongly associated with BMI) are skewed toward either high or low values, compared with stronger variants. We used Cochran's Q test as a further test for heterogeneity in the causal estimates of the individual SNPs (where the analysis is over the 77 SNPs rather than over multiple studies, as would be more usual in a meta-analysis context) and used the result of this test to guide whether the best estimate of the causal effect of BMI on endometrial cancer is the combined estimate from the fixed effects or from the random effects inverse-variance weighted meta-analysis of the per-SNP causal estimates.

There was evidence of association between endometrial cancer and one BMI-associated SNP at P < 4.0 × 10⁻⁴, SNP rs2075650 located within TOMM40 on chromosome 19 [per-allele OR = 1.13; 95% confidence interval (CI), 1.05–1.21; P = 2.4 × 10⁻⁴; Supplementary Table S3). The signal was similar in the subset of samples with BMI information (OR = 1.18; 95% CI, 1.10–1.26; P = 2.0 × 10⁻⁴) and remained significant after including BMI as a covariate (OR = 1.16; 95% CI, 1.07–1.24; P = 3.7 × 10⁻⁴). For the individual SNPs, there was a very modest positive correlation between the published effect on BMI and the estimated effect on endometrial cancer risk (Pearson R = 0.26, P = 0.02), which was attenuated when only samples with BMI information were included (Pearson R = 0.19, P = 0.09) and disappeared completely after conditioning on BMI (Pearson R = 0.004, P = 0.96; Supplementary Fig. S1).

There was also evidence for association with one WHR SNP, rs10842707 at the ITPR2-SSPN locus on chromosome 12 (OR = 1.09; 95% CI, 1.05–1.13; P = 3.7 × 10⁻⁴; Supplementary Table S4), although this signal fell below our study-wide significance threshold after adjusting for BMI (OR = 1.08; 95% CI, 1.01–1.14; P = 1.1 × 10⁻²; unadjusted OR for the subset with BMI information was 1.07, 95% CI, 1.01–1.13; P = 2.4 × 10⁻²). There was no obvious correlation between published effect sizes for WHR SNPs and endometrial cancer risk (Pearson R = −0.19, P = 0.09; Supplementary Fig. S2).

As expected, self-reported BMI was highly significantly associated with endometrial cancer risk overall, with an OR = 1.55 (95% CI, 1.44–1.68, per 5 kg/m²; P = 1.8 × 10⁻²⁶) for every 5 kg/m² increase in BMI (Table 1): ORs were somewhat greater for endometrioid (OR = 1.56; 95% CI, 1.42–1.72) than nonendometrioid/mixed (OR = 1.50; 95% CI, 1.43–1.57) histologies. The association between BMI and the BMIwGRS was significant in both cases and controls (Supplementary Fig. S3); overall, each weighted allele (i.e., each unit increase in the BMIwGRS) was associated with a 4.83 kg/m² increase in BMI (95% CI, 4.33–5.32; P = 1.2 × 10⁻⁸¹), indicating the suitability of this score as an instrumental variable for BMI in our dataset (F statistic on a pooled analysis adjusting for study 587.7).

The BMIwGRS was significantly associated with endometrial cancer risk in the entire dataset, with a per weighted allele OR = 2.11 (95% CI, 1.94–2.28; P = 3.4 × 10⁻¹⁷; Table 2). Scaling according to the magnitude of the effect of the score on BMI (β = 4.83 kg/m²), we find the endometrial cancer OR per 5 kg/m² of genetically predicted BMI to be 2.06 (95% CI, 1.89–2.21; Fig. 1). This effect is apparently driven by an association with endometrioid disease (scaled OR = 2.21; 95% CI, 2.03–2.38; P = 6.6 × 10⁻¹²). The overall association was similar for the subset with BMI information (scaled OR = 2.18; 95% CI, 1.96–2.41; P = 4.2 × 10⁻¹²) and attenuated but remained significant after including BMI as a covariate in the model (scaled OR = 1.22; 95% CI, 1.12–1.34; P = 5.3 × 10⁻⁴).

According to the P test, there was no significant evidence of directional pleiotropy (P = 0.53), despite some possible asymmetry in the funnel plot (Supplementary Fig. S4). However, Cochran's Q test did show some significant heterogeneity in the causal estimates from the individual SNPs (P = 1.5 × 10⁻⁴); hence, the causal effect of BMI on endometrial cancer would be more appropriately estimated from the inverse-variance weighted random effects meta-analysis of the 77 BMI SNPs (P = 1.8 × 10⁻⁹). Unfortunately, this effect estimate cannot be interpreted as the SNP–BMI regression coefficients presented by the GIANT consortium are for an inverse-normalized transformation of BMI, from which effects on the kg/m² scale cannot be derived. Nevertheless, we note that the causal lnOR estimate from the random effects analysis of individual SNPs is approximately 10% higher than that from the equivalent fixed effects analysis (Supplementary Fig. S4); thus, we infer that the true causal effect of BMI on endometrial cancer is slightly larger than our best estimate under the assumption of no directional pleiotropy, that is, OR > 2.06 per 5 kg/m²as predicted in our dataset. This is somewhat larger than the observed OR = 1.55 (95% CI, 1.44–1.68) per 5 kg/m² of reported BMI in this dataset, and also larger than previously published estimates of the effect of reported BMI on endometrial cancer in epidemiologic studies [e.g., OR = 1.54; 95% CI, 1.47–1.61, per 5 kg/m² (4); OR = 1.57; 95% CI, 1.54–1.61, per 5 kg/m² for “type I” largely endometrioid endometrial cancer (3)].

Both WHRwGRS were significantly associated with WHR in the WTCCC control group (34-SNP WHRwGRS β = 0.05; 95% CI, 0.02–0.08; P = 2.2 × 10⁻³; 47-SNP WHRwGRS β = 0.05; 95% CI, 0.03–0.08; P = 1.8 × 10⁻⁴). As expected, neither WHRwGRSs were associated with BMI (P ≥ 0.80). The results for the 34-SNP WHRwGRS were very similar to those from secondary analyses using all 47 WHR SNPs, neither of which were significantly associated with endometrial cancer risk (OR = 1.02; 95% CI, 0.99–1.04; P = 0.09 and OR = 0.97; 95% CI, 0.63–1.31; P = 0.86, respectively; Table 3), or with risk stratified by histology (data not shown).

In this study, we assessed whether SNPs associated with increased BMI or WHR are also associated with increased endometrial cancer risk, either individually or in combination, and whether these genetic associations are independent of BMI. Although BMI is clearly recognized as a major risk factor for endometrial cancer, the role of WHR, independent of BMI, is less clear. Most studies including WHR have reported evidence for an association with endometrial cancer (4, 26–32) but only four presented analyses adjusting for BMI, with the results suggesting that the WHR-endometrial cancer risk association was attenuated in Caucasians (29–31), but not in Asians (32).

Combined as a wGRS, the 77 BMI-associated SNPs were highly significantly associated with BMI, even though the BMIwGRS explained only approximately 1% of the variance in BMI in our sample (less than the estimated 2.7% of the variance in BMI explained by 97 BMI-associated SNPs across ancestries in the discovery dataset; ref. 9). The BMIwGRS was also significantly associated with endometrial cancer, explaining approximately 0.1% of the variance in risk and confirming the causal nature of the association between BMI and endometrial cancer. Indeed, the association between genetically predicted BMI (based on the 77 SNP BMIwGRS) and endometrial cancer risk was somewhat larger than that between observed BMI (i.e., that calculated from self-reported height and weight) and endometrial cancer risk, and we identified significant heterogeneity in the per-SNP causal estimates, both of which suggest some modest degree of directional pleiotropy. Furthermore, the overall association signal attenuated, but did not disappear, when adjusting for BMI (OR = 1.22 vs. 2.06 per 5 kg/m² genetically predicted BMI), which also suggests that these SNPs mainly, but not entirely, operate to increase endometrial cancer risk via BMI. In particular, we note that one BMI SNP, rs2075650, was found to be associated with endometrial cancer risk independent of BMI in our dataset.

Our result could also (or instead) suggest that the aspect of body composition most relevant for endometrial cancer risk is only partially captured by BMI; although BMI is widely used as a convenient proxy measure for adiposity, it is by no means a perfect measure (33). One would expect that the SNPs identified to date in GWAS of BMI, at least on aggregate, are more strongly associated with adiposity than with its proxy, BMI. Hence, the combined effect of the 77 BMI SNPs might be a better predictor of risk due to adiposity than BMI self-reported at a single time point (which could be subject to regression dilution). The effect of BMI on endometrial cancer risk has been reported to be attenuated among ever users of hormone replacement therapy (HRT), as compared with never users (34). Although we were unable to stratify our analyses according to HRT use, we are confident that the difference between the effects of observed and genetically predicted BMI on endometrial cancer risk seen in our study is not attributable to an interaction between BMI and HRT use, as both analyses were based on the same set of women, and so will necessarily have included the same proportions of current, previous, and never HRT users. However, the discrepancy between the observed and predicted effects of BMI on endometrial cancer could theoretically point to negative confounding between measured BMI and endometrial cancer, via HRT use and some other factor (e.g., socioeconomic status) associated with both higher BMI and less frequent HRT use.

The evidence for modest pleiotropy for BMI SNPs and endometrial cancer risk has been reported previously in a study of Chinese women. A study of 26 SNPs then reported as (nominally) associated with different measures of obesity in GWAS datasets (35) identified a GRS–endometrial cancer association in Chinese women (16), which attenuated but remained significant after adjusting for BMI. Direct comparison with our findings is difficult, due to differences in SNP selection and overlap and also because the relationship between BMI and percentage body fat differs among ethnic groups (36). However, the results from our European ancestry study contrast with those from another recent analysis of 3,376 European ancestry endometrial cancer cases and 3,867 controls from the E2C2 consortium (15); neither cases nor controls from the E2C2 analysis overlap with those presented here. Although the E2C2 analysis also identified a significant association between a 97-SNP GRS and endometrial cancer risk (P = 0.002), this association ablated after adjustment for BMI (P = 0.78). The differences in findings between the two European studies may possibly reflect the BMI profiles of the two studies; although the mean BMI of the controls did not differ between the two studies (P = 0.11), the mean BMI of cases in the E2C2 study was greater than that for cases in our study (P = 0.017, mean difference of 0.43 kg/m²). However, the differences are more likely to reflect the increased power of our larger study to detect modest effects. This is particularly pertinent to the single SNP findings. Although BMI SNP rs2075650 was found to be significantly associated with endometrial cancer risk independent of BMI in our dataset (per allele OR of 1.13; 95% CI, 1.05–1.21), this same SNP was not significantly associated with endometrial cancer risk in the E2C2 analysis (OR_BMI-adjusted = 1.00; 95% CI, 0.90–1.10; ref. 15), although there was some overlap between the 95% CIs. We found no evidence in support of the E2C2 tentative finding of a protective effect on endometrial cancer of the subset of five BMI risk alleles at loci known to be involved in monogenic obesity syndromes, with OR point estimates above unity observed for the four loci we investigated in our study (rs6567160, MC4R, OR = 1.02 (0.98–1.06); P = 0.3; rs11030104, BDNF, OR = 1.05 (1.01–1.09); P = 0.05; rs10182181, POMC/ADCY3, OR = 1.02 (0.98–1.06); P = 0.4; Supplementary Table S3).

We also, for the first time, used a genetic approach to assess the influence of body fat distribution on endometrial cancer risk, an epidemiologic association that is less clear than that of adiposity as measured by BMI. Combined as a wGRS, 34 SNPs reported as significantly associated with WHR in women (11) were not significantly associated with endometrial cancer in our sample. We focused on the 34-SNP WHRwGRS due to the marked sexual dimorphism among WHR-associated loci (11); however, the results did not differ when 47 SNPs were included in the WHRwGRS. Together, the 49 SNPs now reported as associated with WHR explain approximately 2.4% of the variance in WHR in women (∼1.4% in both sexes combined; ref. 11). As this is similar to the proportion of variation in BMI explained by currently known BMI-associated SNPs, it seems most likely that the lack of association between the WHRwGRS and endometrial cancer is due to a true lack of association between WHR and endometrial cancer, rather than the smaller number of SNPs included in the WHRwGRS, particularly as the WHR–endometrial cancer association seen in epidemiologic studies seems to be accounted for by BMI (4). Rather than WHR, WC may be a more relevant measure of central adiposity, with evidence that the association between WC and endometrial cancer is independent of BMI (4). However, analysis of the genetic association between WC and endometrial cancer awaits the discovery of additional WC-associated SNPs, as only six have been reported to date (four in Caucasians), none of which reached genome-wide significance in women only (11).

In summary, our combined results from wGRSs and Mendelian randomization analysis provide a further line of evidence that increasing BMI has a direct effect on endometrial cancer risk, and thus that interventions aimed at weight loss should reduce that risk (5). We also found that SNP alleles associated with increased BMI have an aggregate effect on endometrial cancer risk that is over and above that predicted by their effects on BMI. This suggests a possible degree of pleiotropy in SNP functions, indicating that these SNPs, and potentially other BMI-associated SNPs yet to be discovered, would be more useful components in an endometrial cancer risk prediction model than BMI itself. In contrast, our genetic findings indicate that WHR is not independently associated with endometrial cancer risk. These findings support the value of genetic approaches to verify causal relationships between epidemiologic risk factors and cancer risk.

P.A. Fasching reports receiving commercial research grants from Amgen and Novartis and is a consultant/advisory board member for Novartis, Pfizer, Roche, and Genomic Health. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.N. Painter, P. Hall, E.L. Goode, A.M. Dunning, D.F. Easton, A.B. Spurdle

Development of methodology: J.N. Painter, E.L. Goode, J.L. Hopper, D.F. Easton, D.J. Thompson, A.B. Spurdle

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.A. O'Mara, P.M. Webb, T. Cheng, M. McEvoy, R.J. Scott, S. Ahmed, C.S. Healey, M. Shah, M. Gorman, L. Martin, S.V. Hodgson, M.W. Beckmann, P.A. Fasching, A. Hein, M. Rübner, K. Czene, P. Hall, J. Li, T. Dörk, M. Dürst, P. Hillemanns, I.B. Runnebaum, F. Amant, J. Depreeuw, D. Lambrechts, P. Neven, J.M. Cunningham, S.C. Dowdy, E.L. Goode, B.L. Fridley, T.S. Njølstad, H.B. Salvesen, J. Trovik, H.M.J. Werner, K.A. Ashton, G. Otton, A. Proietto, M. Mints, E. Tham, Q. Wang, J.L. Hopper, J. Peto, A.J. Swerdlow, B. Burwinkel, H. Brenner, A. Meindl, H. Brauch, A. Lindblom, J. Chang-Claude, F.J. Couch, G.G. Giles, V.N. Kristensen, A. Cox, P.D.P. Pharoah, I. Tomlinson, D.F. Easton, A.B. Spurdle

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.N. Painter, T.A. O'Mara, L. Marquart, P.M. Webb, J. Attia, S.E. Medland, J. Dennis, E.G. Holliday, A. Hein, J. Li, D. Lambrechts, E.L. Goode, G. Otton, K. Michailidou, J.P. Tyrer, J.L. Hopper, A. Meindl, F.J. Couch, A.M. Dunning, D.F. Easton, A.B. Spurdle

Writing, review, and/or revision of the manuscript: J.N. Painter, T.A. O'Mara, L. Marquart, P.M. Webb, J. Attia, S.E. Medland, E.G. Holliday, M. McEvoy, R.J. Scott, M.W. Beckmann, A.B. Ekici, P.A. Fasching, K. Czene, H. Darabi, P. Hall, J. Li, T. Dörk, P. Hillemanns, F. Amant, P. Neven, J.M. Cunningham, S.C. Dowdy, S.J. Winham, T.S. Njølstad, H.B. Salvesen, J. Trovik, G. Otton, E. Tham, M.K. Bolla, J.P. Tyrer, J.L. Hopper, J. Peto, A.J. Swerdlow, B. Burwinkel, H. Brenner, H. Brauch, A. Lindblom, J. Chang-Claude, F.J. Couch, G.G. Giles, V.N. Kristensen, A. Cox, I. Tomlinson, D.F. Easton, D.J. Thompson, A.B. Spurdle

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.N. Painter, T.A. O'Mara, J. Dennis, R.J. Scott, S. Ahmed, C.S. Healey, M. Shah, A.B. Ekici, A. Hein, M. Rübner, T. Dörk, I.B. Runnebaum, D. Annibali, D. Lambrechts, B.L. Fridley, T.S. Njølstad, M.K. Bolla, Q. Wang, J. Peto, B. Burwinkel, H. Brauch, P.D.P. Pharoah, A.M. Dunning, A.B. Spurdle

Study supervision: M. Shah, T. Dörk, J.L. Hopper, H. Brauch, D.F. Easton, A.B. Spurdle

The authors thank the many individuals who participated in this study and the numerous institutions and their staff who supported recruitment, detailed in full in the Supplementary Text. We also thank Felix Day and Stephen Burgess for helpful discussions.

The iCOGS endometrial cancer analysis was supported by NHMRC project grant (ID#1031333; to A.B. Spurdle, D.F. Easton, and A.M. Dunning). A.B. Spurdle and P.M. Webb are supported by the NHMRC Fellowship scheme. IT is supported by Cancer Research UK and the Oxford Comprehensive Biomedical Research Centre. T. Cheng is supported by the Rhodes Trust and the Nuffield Department of Medicine. Funding for the iCOGS infrastructure came from the European Community's Seventh Framework Programme under grant agreement nr. 223175 (HEALTH-F2-2009-223175; COGS), Cancer Research UK(C1287/A10118, C1287/A 10710, C12292/A11174, C1281/A12014, C5047/A8384, C5047/A15007, C5047/A10692, and C8197/A16565), the NIH (CA128978) and Post-Cancer GWAS initiative(1U19 CA148537, 1U19 CA148065, and 1U19 CA148112 - the GAME-ON initiative), the Department of Defence (W81XWH-10-1-0341), the Canadian Institutes of Health Research (CIHR) for the CIHR Team in Familial Risks of Breast Cancer, Komen Foundation for the Cure, the Breast Cancer Research Foundation, and the Ovarian Cancer Research Fund.

ANECS recruitment was supported by project grants from the NHMRC (ID#339435), The Cancer Council Queensland (ID# 4196615), and Cancer Council Tasmania (ID#403031 and ID#457636). SEARCH recruitment was funded by a programme grant from Cancer Research UK (C490/A10124). Stage I and stage II case genotyping was supported by the NHMRC (ID#552402, ID#1031333). This study makes use of data generated by the Wellcome Trust Case-Control Consortium (WTCCC). A full list of the investigators who contributed to the generation of the data is available from www.wtccc.org.uk. Funding for the project was provided by the Wellcome Trust under award 076113. We acknowledge use of DNA from the British 1958 Birth Cohort collection, funded by the Medical Research Council grant G0000934 and the Wellcome Trust grant 068545/Z/02; funding for this project was provided by the Wellcome Trust under award 085475. NSECG was supported by the EU FP7 CHIBCHA grant, Wellcome Trust Centre for Human Genetics Core grant 090532/Z/09Z, and CORGI was funded by Cancer Research UK. Recruitment of the QIMR Berghofer controls was supported by the NHMRC. The University of Newcastle, the Gladys M Brawn Senior Research Fellowship scheme, The Vincent Fairfax Family Foundation, the Hunter Medical Research Institute, and the Hunter Area Pathology Service all contributed toward the costs of establishing the Hunter Community Study.

The Bavarian Endometrial Cancer Study (BECS) was partly funded by the ELAN fund of the University of Erlangen. The Leuven Endometrium Study (LES) was supported by the Verelst Foundation for Endometrial Cancer. The Mayo Endometrial Cancer Study (MECS) and Mayo controls (MAY) were supported by grants from the NCI of United States Public Health Service (R01 CA122443, P30 CA15083, P50 CA136393, and GAME-ON the NCI Cancer Post-GWAS Initiative U19 CA148112), the Fred C and Katherine B Andersen Foundation, the Mayo Foundation, and the Ovarian Cancer Research Fund with support of the Smith family, in memory of Kathryn Sladek Smith. MoMaTEC received financial support from a Helse Vest Grant, the University of Bergen, Melzer Foundation, The Norwegian Cancer Society (Harald Andersens legat), The Research Council of Norway and Haukeland University Hospital. The Newcastle Endometrial Cancer Study (NECS) acknowledges contributions from the University of Newcastle, The NBN Children's Cancer Research Group, Jennie Thomas, and the Hunter Medical Research Institute. RENDOCAS was supported through the regional agreement on medical training and clinical research (ALF) between Stockholm County Council and Karolinska Institutet (numbers: 20110222, 20110483, 20110141, and DF 07015], The Swedish Labor Market Insurance (number 100069) and The Swedish Cancer Society (number 11 0439). The Cancer Hormone Replacement Epidemiology in Sweden Study (CAHRES, formerly called The Singapore and Swedish Breast/Endometrial Cancer Study; SASBAC) was supported by funding from the Agency for Science, Technology and Research of Singapore (A*STAR), the U.S. NIH, and the Susan G. Komen Breast Cancer Foundation.

The Breast Cancer Association Consortium (BCAC) is funded by Cancer Research UK (C1287/A10118, C1287/A12014). The Ovarian Cancer Association Consortium (OCAC) is supported by a grant from the Ovarian Cancer Research Fund thanks to donations by the family and friends of Kathryn Sladek Smith (PPD/RPCI.07), and the UK National Institute for Health Research Biomedical Research Centres at the University of Cambridge. Additional funding for individual control groups is detailed in the Supplementary Information.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.