Genome-wide association and epidemiological analyses reveal common genetic origins between uterine leiomyomata and endometriosis

UL GWAS meta-analysis . Our discovery meta-analysis of GWAS on UL includes four population-based cohorts and one direct-to- consumer cohort of white European ancestry: Women ’s Genome Health Study (WGHS), Northern Finnish Birth Cohort (NFBC), QIMR Berghofer Medical Research Institute (QIMR), UK Biobank (UKBB), and 23andMe (Supplementary Methods, Supplementary Table 1). Imputation of genotypes was carried out using 1000 Genomes Project Phase 3 and Haplotype Reference Consortium (HRC) reference panels. UL phenotype in each cohort was ana- lyzed in a logistic regression or linear mixed model assuming additive genetic effects with multivariate adjustment for age, BMI, and/or correction for population structure. After quality control metrics were applied, including exclusion of non-informative (MAF < 0.01) and poorly imputed (r2 < 0.4) SNPs, we performed a fixed-effects, inverse-variance-weighted (IVW) meta-analysis across 35,474 cases with a clinical or self-reported history of UL and 267,505 unaffected female controls. Altogether 8,662,096 biallelic SNPs were analyzed and adjustments for genomic in fla- tion performed (Supplementary Fig. 1, Supplementary Table 2). Through linkage disequilibrium score (LDSC) regression analysis, an estimated 89.5% of the genomic in flation factor ( λ GC)o f1 . 1 2 was attributable to polygenic heritability (intercept = 1.02, s.e. = 0.0081). Overall, individual SNP-based heritability ( h2) was esti- mated to be 0.0281 (s.e. = 0.0029) on the liability scale. Risk loci associated with UL . We observe genome-wide sig- nificant associations ( P< 5×1 0 −8) at 2505 SNPs across 29 independent loci (Table 1, Supplementary Fig. 2, Supplementary Table 3). The Manhattan plot is shown in Fig. 1. We identify eight novel loci associated with UL (2p23.2, 4q22.3, 6p21.31, 7q31.2, 10p11.22, 11p14.1, 12q15, and 12q24.31), which include the fol- lowing candidate genes of interest: HMGA1, BABAM2, and WNT2. HMGA1 is a member of the high mobility group proteins and is involved in regulation of gene transcription 17. Somatic rearrangements of HMGA1 at 6p21 have been recurrently documented in UL, albeit at a much lower frequency than those of HMGA2—another member of the high mobility group protein family18–20. BABAM2 at 2p23.2 encodes a death receptor- associating intracellular protein that promotes tumor growth by suppressing apoptosis 21. Associations at 7q31.2 containing WNT2, a member of the Wnt gene family, provide support for the previously suggested role of Wnt signaling in UL 22,23. Among 29 independent loci are 21 loci previously reported to be signi ficantly associated with UL 11–16. A number of identi fied loci harbor genes previously implicated in cell growth and cancer risk in different tissue types, including cervical cancer 24, epithelial ovarian cancer 25,26, breast cancer 27,28, glioma 29,30, bladder cancer31, and pancreatic cancer 32–34. Speci fically, seven indepen- dent loci contain well-characterized oncogenes and tumor suppressor genes from the Cancer Gene Census list in COSMIC35: PDGFRA, TERT, ESR1 , WT1, ATM, FOXO1, and TP53. Using approximate conditional analysis, we identify multiple distinct association signals for UL at 10 loci (at locus-wide significance, P< 1×1 0 −5, Bonferroni correction) (Supplementary Table 4). Fine-mapping was conducted on all 43 distinct association signals arising from the 29 detected UL loci, revealing three association signals with a single variant in the 99% credible set (Fig. 2, Supplementary Table 5). The missense variant at 20p12.3 (rs16991615; E341K) maps to MCM8, a gene that encodes a protein involved in DNA double-strand break repair 36. MCM8 has also been implicated in length of reproductive lifespan, menopause, and premature ovarian failure 37,38. Another variant (rs78378222) resides in the 3 ’UTR of TP53 at 17p13.1, and has been shown to disturb 3 ’-end processing of TP53 mRNA39. This variant has been associated with both malignant and benign tumor types 39–41. UL GWAS limited by HMB . HMB, one of the major symptoms of UL, is estimated to affect up to 30% of reproductive-aged women, having a considerable impact on a woman ’s quality of life. Thus, variants speci fically associated with this symptom are of particular interest for drug target development. We performed a GWAS on UL limited by HMB using a linear mixed model across 3409 cases and 199,171 unaffected female controls from the UKBB (Supplementary Methods, Supplementary Fig. 3). We observe genome-wide signi ficant associations ( P< 5×1 0 −8)a t three of the 29 independent UL loci: 5p15.33 (rs72709458, OR [95% CI] = 0.86 [0.81 –0.91], P = 3.50 × 10−8), 5q35.2 (rs2456181, OR [95% CI] = 0.87 [0.83 –0.91], P = 4.20 × 10−10), and 11q22.3 (rs1800057, OR [95% CI] = 0.66 [0.58 –0.76], P = 2.80 x 10 −9) (Fig. 3, Supplementary Fig. 4, Supplementary Table 6). The lead SNP at 11q22.3, a missense variant in ATM, has been associated with increased risk of various cancers, such as breast cancer 42,43, while the lead SNP at 5p15.33, an intronic TERT variant, has previously been implicated in gliomas 44. The lead SNP rs2456181 at 5q35.2 resides near FGFR4, a gene encoding a cell-surface receptor for fibroblast growth factors involved in regulation of several pathways, including cell pro- liferation, differentiation, and migration. HMB GWAS . A GWAS based solely on HMB across 9813 cases and 210,946 female controls reveals one genome-wide signi ficant association at 11p14.1, one of the eight novel loci associated with ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 2 NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications UL (Supplementary Figs. 5 and 6). The lead SNPs for UL and HMB at 11p14.1 are in high LD, and the direction of the effect is the same (Supplementary Fig. 7). This locus has previously been associated with endometriosis, age at menarche, and follicle-sti- mulating/luteinizing hormone levels 45–47. According to GTEx (v7), the lead SNP for HMB (rs11031005) is a potential expression quantitative trait locus (eQTL) for ARL14EP in several tissue types, such as testis and thyroid. Mendelian randomization (MR) was used to assess the causality of genetic association between UL (exposure) and HMB (outcome). Interestingly, MR reveals that Table 1 Overview of lead SNPs with signi ficant associations at 29 independent loci in UL GWAS meta-analysis Locus Lead SNP RA OA RAF EUR PMeta OR (95% CI) Gene(s) of interest a 1p36.12b,c rs7412010 C G 0.15 2.4 × 10 −29 1.13 (1.11 –1.16) WNT4, CDC42 2p23.2 rs55819434 A G 0.91 5.6 × 10 −09 0.92 (0.90-0.95) BABAM2 2p25.1b,c rs35417544 T C 0.69 2.3 × 10 −19 1.09 (1.07 –1.10) GREB1 3q26.2c rs35446936 A G 0.24 1.0 × 10 −08 0.95 (0.93-0.96) TERC 4q12c rs62323682 T C 0.94 4.9 × 10 −18 0.87 (0.84-0.90) LNX1, PDGFRA 4q13.3c rs12640488 A G 0.52 4.0 × 10 −14 0.94 (0.92-0.96) SULT1B1 4q22.3 rs4699299 T C 0.69 4.7 × 10 −08 0.95 (0.94-0.97) PDLIM5 5p15.33c rs72709458 T C 0.23 4.7 × 10 −21 1.10 (1.08 –1.13) TERT 5q35.2c rs2456181 C G 0.49 1.1 × 10 -11 0.94 (0.93-0.96) ZNF346, UIMC1 6p21.31 rs116251328 A T 0.02 3.0 × 10 −08 1.15 (1.09 –1.21) GRM4, HMGA1 6q25.2b,c rs58415480 C G 0.84 1.9 × 10 −54 0.84 (0.82-0.86) SYNE1, ESR1 7q31.2 rs2270206 A C 0.16 4.6 × 10 −08 1.06 (1.04 –1.09) WNT2 9p24.3c rs10976689 A G 0.60 2.4 × 10 -13 0.94 (0.93-0.96) ANKRD15 10q24.3c rs9419958 T C 0.13 1.1 × 10 −16 1.10 (1.08 –1.13) OBFC1, SLK 10p11.22 rs10508765 A G 0.80 1.5 × 10 −10 1.07 (1.05 –1.09) ZEB1, ARHGAP12 11p15.5c rs547025 T C 0.92 1.5 × 10 −14 1.13 (1.09 –1.16) RIC8A, BET1L 11p14.1b rs11031006 A G 0.14 5.7 × 10 −15 0.91 (0.89-0.93) FSHB 11p13c rs61889186 C G 0.86 1.4 × 10 -25 0.89 (0.87-0.91) WT1 11p13c rs2785202 C G 0.55 6.9 × 10 −14 1.06 (1.05 –1.08) PDHX, CD44 11q22.3c rs149934734 T C 0.03 1.1 × 10 −27 1.33 (1.26 –1.40) C11orf65, KDELC2 12q13.11c rs2131371 A C 0.28 1.6 × 10 −18 0.93 (0.91-0.94) SLC38A2 12q15 rs11178393 T C 0.89 3.3 × 10 −08 1.08 (1.05 –1.10) PTPRR 12q24.31 rs28583837 A G 0.22 2.3 × 10 −08 0.94 (0.92-0.96) PITPNM2 13q14.11c rs117245733 A G 0.02 5.7 × 10 −14 1.31 (1.21 –1.39) FOXO1 17p13.1c rs78378222 T G 0.99 7.1 × 10 −31 0.65 (0.60-0.70) SHBG, TP53 20p12.3c rs16991615 A G 0.07 8.8 × 10 −10 1.11 (1.07 –1.14) MCM8, TRMT6 22q13.1c rs4821939 A T 0.20 7.8 × 10 −16 1.08 (1.06 –1.10) TNRC6B Xp26.2c rs12392108 A T 0.31 5.9 × 10 −46 1.13 (1.11 –1.15) RAP2C Xq13.1c rs4360450 A G 0.37 2.1 × 10 −18 1.08 (1.06 –1.10) MED12 SNP single-nucleotide polymorphism, RA risk allele, OA other allele, RAFEUR average risk allele frequency in European samples, OR odds ratio a≤300 kb distant from association signal bLoci previously associated with endometriosis cLoci previously associated with UL 50 40 30 –log10 (p) 20 10 0 1 23 4 5 6 7 8 9 Chromosome 10 11 12 13 14 15 16 17 18 20 22 23 Fig. 1 Manhattan plot for UL GWAS meta-analysis across all cohorts. Meta-analysis of GWAS including 302,979 women of white European ancestry across all cohorts identi fied 29 independent loci associated with UL. Red and blue horizontal lines indicate genome-wide signi ficant ( P< 5×1 0 −8) and suggestive ( P< 1×1 0 −5) thresholds, respectively NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 ARTICLE NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications 3 genetic predisposition to UL is causally linked to an increased risk of HMB, with the β estimate of 0.26 being signi ficant in the IVW model (P = 1.2 × 10−12) in the absence of heterogeneity ( P = 0.13) (Supplementary Table 7). The MR Egger regression shows no significant directional pleiotropy (intercept = 0.01, P = 0.36) supporting a causal relationship. Overlap of UL and endometriosis . Interestingly, signi ficant association signals are observed at several loci previously asso- ciated with endometriosis: 1p36.12 (rs7412010, OR [95% CI] = 1.13 [1.11 –1.16], P = 2.43 × 10−29), 2p25.1 (rs35417544, OR [95% CI] = 1.09 [1.07 –1.10], P = 2.32 × 10−19), 6q25.2 (rs58415480, OR [95% CI] = 1.19 [1.17 –1.22], P = 1.86 × 10−54), and 11p14.1 (rs11031006, OR [95% CI] = 1.10 [1.07 –1.12], P = 5.65 × 10−15)45,48–50. LD is strong between UL and previously reported endometriosis lead SNPs45 at all except one locus, 2p25.1 (Supplementary Table 8). In addition, the direction of effect is the same between the lead SNPs at 1p36.12. Using LDSC regres- sion, we observe a moderate genetic correlation between UL and endometriosis in women with European ancestry ( rg = 0.39, s.e = 0.05, P = 9.77 × 10−13). Endometriosis has an earlier age-of- onset than UL, with a mean age of 25 –29 years and 35 years, respectively. MR suggests that genetic predisposition to endome- triosis (exposure) is causally linked to an increased risk of UL (outcome); the β of 0.36 is signi ficant (P = 3.7 × 10−3) in the IVW model (heterogeneity P = 9.5 × 10−68) (Supplementary Table 7). Leave-one-out sensitivity analysis reveals that no single SNP alone drives the signi ficant relationship between endometriosis and UL, but instead the relationship is accounted for by contributions from multiple variants across the genome (Supplementary Fig. 8). Given the high degree of heterogeneity, the effect sizes were estimated in a minimal set of SNPs that when used as a genetic instrument eliminate the heterogeneity (Supplementary Fig. 9). The effect size estimate ( β = 0.12) from the minimal set of variants remains significant ( P = 4.3 × 10−3) in the IVW model in the absence of heterogeneity ( P = 0.23). We also applied the MR pleiotropy residual sum and outlier (MR-PRESSO) global and distortion tests to adjust for variants causing signi ficant bias in the estimates through horizontal pleiotropy. Outlier-adjusted estimates still provide significant evidence for a causal estimate of endometriosis on UL ( β = 0.29, P = 0.002) (Supplementary Table 7). 15abc 10 –log10 (p-value) –log10 (p-value) –log10 (p-value) 5 0 40 LHFP 0.8 r 2 0.6 0.4 0.2 0.8 r 2 0.6 0.4 0.2 0.8 r 2 0.6 0.4 0.2 MIR4305 COG6 rs117245733 rs78378222 rs16991615 LINC00548 LINC00332 LINC00598 MIR320D1 ASGR1 ACADVL DLG4 DVL2 PHF23 GABARAP ELP5 CLDN7 CTDNEP1 MIR324 EIF5A ACAP1 KCTD11 TMEM95 TNK1 PLSCR3 TMEM256 TNFSF13 TNFSF13TNFSF12 POLR2A FXR2 SHBG DNAH2 SLC35G6 TP53EIF4A1ZBTB4 TNFSF12NLGN2 SENP3 SENP3-EIF4A1 SNORA48 SNORD10 SCARNA21RPL29P2 EFNB3 WRAP53 KDM6B TMEM85 NAA38 CHD3 GUCY2D HES7 ALOX15B CYB5D1 LOC284023 KCNAB3 CNTROB TRAPPC1 VAMP2 LOC643406 LINC00654 LOC101929207 LOC101929225GPCPD1 CHGB CRLS1 MCM8 CASC20FERMT1TRMT6C20orf196MIR6883 PER1 ALOXE3 ALOX12B MIR4314 SPEM1 C17orf74 TMEM102 FOXO1 TPTE2P5 MRPS31 SLC25A15 MIR621 SUGT1P3 40.5 41 Position on chr13 (Mb) Position on chr17 (Mb) Position on chr20 (Mb) 41.5 0 20 40 60 80 100 0 20 40 60 80 100 Recombination rate (cM/Mb) Recombination rate (cM/Mb) Recombination rate (cM/Mb) 0 20 40 60 80 100 7.2 7.4 7.6 7.8 8 5.6 5.8 6 6.2 6.4 1 gene omitted 13 gene omitted1 gene omitted 0 0 2 4 6 8 10 5 10 15 20 25 30 Fig. 2 Fine-mapping reveals three association signals with a single driver in 99% credible set. Association with UL is expressed as −log10(P value) for the three signals on chromosomes: ( a) 13q14.11, ( b) 17p13.1, and ( c) 20p12.3. The labeled SNP represents the most signi ficant SNP for each locus. SNP association P-value is shown on the y axis, while SNP position (with gene annotation) appears on the x axis. Each SNP is colored according to the strength of LD with the lead SNP. Regional association plots were produced in LocusZoom 25 20 15 –log10 (p) 10 5 0 1 23 4 5 6 7 8 9 Chromosome 10 11 12 13 14 15 16 17 19 21 23 Fig. 3 Manhattan plot for GWAS on UL limited by heavy menstrual bleeding. GWAS across 202,580 women of white European ancestry identi fied three independent loci associated with UL limited by heavy menstrual bleeding. Red and blue horizontal lines indicate genome-wide signi ficant (P< 5×1 0 −8) and suggestive ( P< 1×1 0 −5) thresholds, respectively ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 4 NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications Endometriosis, de fined by ectopic growth of endometrial-like tissue outside the uterus, is a common in flammatory hormone- dependent disease that affects reproductive-aged women 51. Although functional studies of relevant tissue are needed to confirm consequences of the variants in regulation of gene expression, each of the four observed overlapping genomic loci contains a gene(s) known to be involved in progesterone or estrogen signaling. WNT4 at 1p36.12 encodes a secreted signaling factor that promotes female sex development, and regulates both postnatal uterine development and progesterone signaling during decidualization 52,53. Recently, SNPs at 1p36.12 associated with a greater endometriosis risk have been suggested to act through CDC42, a gene that encodes a small GTPase of the Rho family 54. GREB1 at 2p25.1 is an early response gene in the estrogen receptor (ER)-regulated pathway, and promotes growth of breast and pancreatic cancer cells 55,56. ESR1 at 6q25.2 encodes the alpha subunit of the ligand-activated nuclear ER that regulates cell proliferation in the uterus 57. FSHB at 11p14.1 encodes the biologically active subunit of follicle-stimulating hormone, which regulates maturation of ovarian follicles and release of ova during menstruation58,59. Epidemiological meta-analysis . Given shared risk loci and genetic correlation of UL and endometriosis, we conducted an epidemiological meta-analysis including 402,868 women from three population-based cohorts: Nurses ’ Health Study II (NHSII), Women’s Health Study (WHS), and UKBB (Supplementary Methods, Supplementary Table 9), to assess the likelihood of UL diagnosis among women who had or had not been diagnosed with endometriosis. Women with endometriosis had a sig- nificantly higher likelihood of UL diagnosis (multivariable- adjusted summary relative risk (RR) [95% CI] = 2.17 [1.48–3.19]) (Fig. 4). All cohort-speci fic analyses demonstrated at least a doubling of risk, suggesting a robust association (Table 2). However, biologically and statistically signi ficant heterogeneity was observed in the pooling of effect size estimates in the meta- analysis ( P <1×1 0 −4) (Fig. 4). Therefore, absolute effect esti- mates need to be interpreted in the context of source populations. Heterogeneity could re flect various different population sampling and data collection factors among the three cohorts. First, the validity of self-reported diagnosis of endometriosis and to a lesser extent UL are known to be <75% in general population cohorts, such as UKBB, compared to more highly validated self- assessment in the medical professional NHSII and WHS cohorts7. Second, endometriosis clinical de finitions prior to the 1990s were more restrictive —often limited to the presence of endometrioma and/or “powderburn” superficial peritoneal lesions among adult women 60. Subsequently, de finitions have expanded to recognize a wide range of super ficial peritoneal phenotypic presentations, as well as incidence among adolescents and young women 61. It may be impactful therefore that the WHS participants were ≥ 45 years of age in 1992, while NHSII parti- cipants were ≥ 25 years of age in 1989, and UKBB participants were aged 40 to 69 in 2006. Thus, disease de finitions varied during the peak calendar years of incidence among the cohorts, and in addition, while the NHSII were queried about endome- triosis prospectively during their reproductive years, the WHS and UKBB cohorts were cross-sectionally asked to recall their gynecologic health experience decades earlier. It is also important to note that while WHS and NHSII participants were asked specifically about endometriosis diagnosis via questionnaire, the UKBB data collection included qualitative interviews during which endometriosis would be documented only when the par- ticipant herself raised it as a health issue. Those with mild symptoms or those past their reproductive years and thus past the moderate to severe life-impacting symptoms of the disease may have been less likely to offer endometriosis among the list of their health issues. This is supported by the low prevalence of endo- metriosis reported within the UKBB compared to WHS, NHSII, and other population-based estimates 62. However, the UKBB participants (due to the qualitative interview structure and recall bias) and the WHS participants (due to recall bias) could have been more likely to choose to report endometriosis if they also suffered from UL together, resulting in diagnostic bias and con- sequently an in flation of effect estimates. Indeed, the population heterogeneity and differing potential for diagnostic bias by cohort fits with the observed differences among effect estimate magni- tudes with the RRs and CI widths ordered from NHSII (RR = 1.56) to WHS (RR = 1.96) to UKBB (RR = 3.50) (Fig. 4). Bioinformatic analyses of UL risk SNPs and loci . To estimate the genetic correlation between UL and various reproductive traits, as well as cardiometabolic traits/diseases, we performed LD Hub analysis for a total of 21 traits/diseases (Supplementary NHSII WHS UKBB Overall 0.4 0.6 0.8 1 1.5 2 Study RR (95% Cl) 1.56 (1.45, 1.68) 1.96 (1.70, 2.25) 3.50 (2.79, 4.40) 2.17 (1.48, 3.19) Fig. 4 Epidemiologic meta-analysis demonstrates endometriosis is associated with UL. Random-effects, inverse-variance-weighted meta-analysis was performed across the effect sizes and standard errors in 402,868 women from three cohorts (NHSII, WHS, and UKBB). Squares represent point estimates from individual studies, whiskers correspond to the 95% CIs, and the diamond represents results from the meta-analysis. There was evidence of signi ficant heterogeneity based on Cochran ’s Q statistic ( P< 1×1 0 −4) NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 ARTICLE NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications 5 Data 1). We observe signi ficant correlations between increased risk of UL and earlier age of menarche ( rg = −0.16, P = 3.7 × 10−6), earlier age of first birth ( rg = −0.14, P = 1.0 × 10−3), increased levels of triglycerides ( rg = 0.13, P = 1.9 × 10−3), and increased BMI ( rg = 0.11, P = 2.0 × 10−3), as previously suggested by epidemiological studies 63,64, illustrating that common genetic factors can predispose women to both risk factors related to, for example, adverse metabolic and cardiovascular disease risk and UL. Gene-set and tissue enrichment analyses across 8971 SNPs with suggestive ( P< 1×1 0 −5) or signi ficant ( P< 5×1 0 −8)U L associations using DEPICT 65 reveal enrichments (false discovery rate (FDR) < 0.05) in gene sets, such as steroid hormone receptor (GO:0035258; P = 1.03 × 10−5), hormone receptor binding (GO:0051427; P = 9.07 × 10−5), and nuclear hormone receptor binding (GO:0035257; P = 1.53 × 10−4) (Supplementary Data 2 and 3). The results are concordant with the hormone-driven nature of UL. We did not observe any cell/tissue types sig- nificantly enriched for the expression of the genes in the asso- ciated loci (Supplementary Fig. 10). To identify potential causal genes at UL risk loci, we used a summary-data based MR (SMR) method, including both eQTL and mQTL data from peripheral blood66,67. We identify 18 potential causal genes showing no significant heterogeneity in SMR ( PHEIDI >5×1 0 −3), including WNT4 (rs55938609, PSMR = 6.92 × 10−15), GREB1 (rs35417544, PSMR = 3.93 × 10−19), WT1 (rs12280757, PSMR = 1.87 × 10−18), and FOXO1 (rs3924478, PSMR = 5.76 × 10−10) (Supplementary Data 4 and 5). FOXO1 expression in UL . To explore potential functional sig- nificance, we examined expression of the FOXO1 protein, a transcription factor that plays an important role in cell pro- liferation, apoptosis, DNA repair, and stress response 68. Inter- estingly, inactivation of FOXO1 promotes cell proliferation and tumorigenesis in several hormone-regulated malignancies, such as prostate, breast, cervical, and endometrial cancers 69–72. Con- versely, we observe a signi ficant increase in nuclear FOXO1 protein expression in UL compared to myometrial samples using immunohistochemistry on tissue microarrays (Supplementary Fig. 11). Patient-matched tumor-normal pairs show 1.69-fold higher (P = 0.01; paired t-test) nuclear FOXO1 expression in UL, while the expression is as much as 2.32-fold greater ( P = 1.52 × 10−9; Welch ’s t-test) when all 335 UL are considered (Supple- mentary Fig. 12). These results are consistent with a previous study73, which showed phosphorylated (p) FOXO1 (pSer 256)t o be predominantly present in the nucleus in UL, but sequestered in the cytoplasm of myometrium. The concomitant increase of p-FOXO1 and reduced expression of its interaction partner 14-3- 3γ in UL has been suggested to lead to impaired nuclear/cyto- plasmic shuttling of p-FOXO1, which promotes cell survival 73–75. We performed strati fication of samples by genotype, revealing a statistically signi ficant increase in FOXO1 levels of UL harboring the risk allele for rs6563799 (allelic dosage, P = 0.047; homo- zygosity for risk allele, P = 0.035) (Supplementary Figs. 11 and 13). An increase in FOXO1 levels of UL with the rs7986407 risk allele is also observed; however, the change is not statistically significant (Supplementary Figs. 11 and 13).

In our meta-analysis of GWAS on UL, we identify 29 genomic loci to be signi ficantly associated with UL in women of white European ancestry, including eight novel and 21 previously reported loci. Candidate genes in the identi fied loci implicate pathways of estrogen and progesterone signaling ( ESR1, FSHB, GREB1, WNT2 , and WNT4), as well as cell growth ( FOXO1, PDGFRA, TERT, TERC, and TP53) in predisposing women to UL. We do not con firm five of 26 previously identi fied loci reported to be signi ficantly associated with UL 12,14–16. Two of these loci, 3p24.1 and 16q12.1, are nominally signi ficant ( P< 1×1 0 −5)i n our GWAS meta-analysis, but the remaining three loci (3q29, 17q25.3 and a distinct region at 22q13.1) do not reach nominal significance. Ancestral differences may explain the absence of the association originally identi fied in African American women in the genomic region at 22q13.1, while variation in phenotypic definitions12 may underlie the two other loci. Discovery of eight novel loci signi ficantly associated with UL reveals several candidate genes of particular interest: BABAM2, FSHB, HMGA1 , and WNT2. Because UL are benign tumors that rarely, if ever, develop into malignancy, the association between UL and multiple loci harboring well-known oncogenes and tumor suppressor genes is also worthy of note. Fine-mapping of the TP53 locus identi fies rs78378222 to be the most probable causal variant, which has been shown to disrupt the polyadenylation sequence in the 3 ’UTR of TP53 and result in reduced expression of mRNA 39. We also observe nuclear FOXO1 levels to be sig- nificantly elevated in UL when compared to myometrium. FOXO1 is a downstream target of the Akt signaling pathway that responds to hormone signaling through the progesterone receptor in UL and activates proliferative responses 76. HMB is one of the major debilitating symptoms of UL and can have a substantial impact on a woman ’s quality of life. Here, we report GWAS on both UL limited by HMB and solely on HMB, revealing potential targets for pharmacologic intervention: Table 2 Multivariable-adjusted effect estimates of the association between endometriosis and UL among women in NHSII, WHS, and UKBB cohorts Cohort n UL cases Age-adjusted Multivariable-adjusted Nurses’ Health Study II a 102,545 10,714 1.61 (1.50 –1.73) 1.57 (1.45 –1.68) Women’s Health Study b 26,868 1,262 2.04 (1.78 –2.34) 1.97 (1.71 –2.26) UK Biobank c 273,455 19,789 3.11 (2.86 –3.37) 3.50 (2.77 –4.38) aHazard ratios (95% con fidence intervals [CIs]) from Cox regression models. Multivariable model was adjusted for age (continuous), age at menarche (15), infertility (yes, no), ancestry (White, Black, Hispanic, Asian, other), parity (nulliparous, 1, 2, 3, 4 +), age at first birth (29), time since last birth (<1, 1 –3, 4 –5, 6 –7, 8 –9, 10 –12, 13 –15, ≥16), age first oral contraceptive use (13 –16, 17–20, 21–24, ≥25), BMI (<20, 20 –21.9, 22–23.9, 24–24.9, 25–26.9, 27–29.9, ≥30), menstrual cycle length (50 days), smoking (never, past, current), recent gynecologic/breast exam (no recent exam, recent exam), use of anti-hypertensive medications/diastolic blood pressure (no meds <65, no meds 65–74, no meds 75 –84, no meds 85 –89, no meds ≥90, meds <65, meds 65 –74, meds 75 –84, meds 85-89, meds ≥90), and physical activity (MET hours/week: <3, 3 –<9, 9 –<18, 18 –<27, 27 –<42, ≥42). bOdds ratios (95% CI) from logistic regression models. Multivariable model was adjusted for age at baseline (continuous), age at menarche (15), ancestry (White, Black, Hispanic, Asian, other), parity (nulliparous, 1, 2, 3, ≥4), age at first birth (29), oral contraceptive use (ever, never), BMI (<20, 20 –21.9, 22–23.9, 24–24.9, 25–26.9, 27–29.9, ≥30), smoking (never, past, current), use of anti-hypertensive medications/diastolic blood pressure (no meds <65, no meds 65 –74, no meds 75-84, no meds 85 –89, no meds ≥90, meds <65, meds 65 –74, meds 75–84, meds 85–89, meds ≥ 90), physical activity (never/rarely, <1 time/week, 1 –3 times.week, ≥4 times/week), alcohol consumption (never/rarely, 1 –3 drinks/month, 1 –6 drinks/week, ≥1 drinks/day). cOdds ratios (95% CI) from logistic regression models. Multivariable model was adjusted for age (<50, 50 –55, 56–60, ≥60), age at menarche (15), ancestry (White, Black, Asian, other), parity (nulliparous, 1, 2, 3, ≥4), age at first birth (29), oral contraceptive use (ever, never), BMI (<20, 20 –21.9, 22 –23.9, 24 –24.9, 25 –26.9, 27 –29.9, ≥30), smoking (never, past, current), physical activity (never/rarely, 1 –3 drinks/week, ≥4 drinks/week), alcohol consumption (never/rarely,1 –3/month, 1 –4/week, ≥1/day), and menopausal status (premenopausal, postmenopausal). ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 6 NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications ARL14EP, ATM, TERT, and FGFR4. In addition, MR analyses suggest that genetic predisposition to UL is causally linked to an increased risk of HMB. These results form a solid basis for further work to elucidate the mechanisms underlying UL-related HMB and towards tailored treatments of UL and HMB. Biological overlap between UL and endometriosis, two highly common gynecologic diseases has long been suspected due to similarities in molecular mechanisms and progenitor cells. Our UL GWAS meta-analysis indicates that genes previously asso- ciated with endometriosis and involved in hormone-signaling pathways are also associated with UL ( WNT4/CDC42, GREB1, ESR1, and FSHB). Overlap observed in the genetic etiology of endometriosis and UL led us to epidemiologically quantify the co- occurrence of these two diseases across three independent cohorts. The epidemiological meta-analysis indicates that women with a history of endometriosis are at elevated risk for reporting UL. Results from our MR analyses suggest that genetic predis- position to endometriosis is causally linked to increased risk of UL. Alternatively, given the discordance in the direction of allelic effects for the UL and endometriosis loci, our MR results may indicate a signi ficant overlap in the underlying biology of the two diseases. Additional work is needed to better quantify the con- tribution of genetic effects to the directional relationship between endometriosis and UL. Results of which will enable us to quantify what portion of the MR results re flect the fundamental patho- biological overlap in these two diseases of the uterus. Further characterization of the mutual pathogenic mechanisms of UL and endometriosis has the capacity to discover not only a deeper understanding of the underlying biology, but also treatments for two diseases that cause signi ficant morbidity in roughly one-third of the world ’s population.

Subjects. For UL GWAS meta-analysis, four population-based cohorts (WGHS, NFBC, QIMR and UKBB) and one direct-to-consumer cohort (23andMe) from the FibroGENE consortium were included (Supplementary Table 1), resulting in 35,474 UL cases and 267,505 female controls of white European ancestry. Sample sizes were maximized using a basic, harmonizing phenotype de finition to separate cases and controls solely based on either self-report or clinically documented UL history. Our large-scale epidemiologic analysis was comprised of three population- based cohorts (NHSII, WHS, and UKBB), totaling 402,869 women. HMB GWAS included the UKBB cohort, consisting of 220,759 women. Detailed descriptions of cohorts and sample selections are available in Supplementary Methods. All parti- cipants provided informed consent in accordance with the processes approved by the relevant jurisdiction for human subject research for each cohort: the Partners HealthCare System Human Research Committee (WHS/WGHS), the Ethical Committee of the Northern Ostrobothnia Hospital District (NFBC), the Human Research Ethics Committee at the QIMR Berghofer Medical Research Institute and the Australian Twin Registry (QIMR), the North West Multi-centre Research Ethics Committee (UKBB), Ethical and Independent Review Services (an external institutional review board; 23andMe), and the Institutional Review Boards at Harvard T.H. Chan School of Public Health and Brigham and Women ’s Hospital (Partners Human Research Committee) (NHSII). Genotyping. Several different Illumina-based genotyping platforms (Illumina Inc., San Diego, CA, USA) were used: HumanHap300 Duo ‘+’ chips or the combination of the Human-Hap300 Duo and iSelect chips (WGHS), In finium 370cnvDuo array (NFBC), 317 K, 370 K, or 610 K SNP platforms (QIMR). Genotyping of partici- pants in the UKBB was performed either on the Affymetrix UK BiLEVE or Affymetrix UK Biobank Axiom ® array with over 95% similarity. Genotyping of participants in the 23andMe cohort was performed on various versions of Illumina-based BeadChips. Quality control and imputation . Each cohort conducted quality control measures and imputation for their data. For WGHS, NFBC, QIMR, and 23andMe, all cases and controls with a genotyping call rate <0.98 were excluded from the study. Imputation was performed on both autosomal and sex chromosomes using the

panel from the 1000 Genomes Project European dataset (1000 G EUR) Phase 3. Imputation was carried out using ShapeIt2 and IMPUTE2 softwares 77,78. SNPs with call rates of <99% or with deviation from Hardy-Weinberg equilibrium (P ≤ 1×1 0 −6) were excluded from further analyses. Population strati fication for the data was examined with principal component analysis (PCA) using EIGENSTRAT79. The four HapMap populations were used as reference groups: Europeans (CEU), Africans (YRI), Japanese (JPT), and Chinese (CHB). All observed outliers were removed from the study. UKBB data QC and imputation were performed centrally, prior to public release of the data 80. Genotype data used in the present analyses were imputed up to the Haplotype Reference Consortium (HRC) panel. We applied additional quality control filters to exclude poorly imputed SNPs ( r 2 < 0.4) and SNPs with a MAF of <1%. Association analyses. Using additive encoding of genotypes and adjusting for age, BMI, and/or the first five PCs, logistic regression analysis was performed in WGHS, NFBC, QIMR, and 23andMe cohorts and summary statistics were provided, including beta coef ficients, χ2 values, and standard errors, for genotyped and imputed SNPs. The UKBB association analyses were conducted using a linear mixed model (BOLT-LMM v.2.3.2) 81 adjusting for the two array types used, age and BMI ( fixed effects), and a random effect accounting for relatedness between women. Effect size estimates ( β and SE) from the linear mixed-model were con- verted to log-odds scale prior to meta-analysis. A fixed-effects, inverse-variance- weighted (IVW) meta-analysis on summary statistics was conducted using METAL82 across all cohorts (Supplementary Data 6). A total of 8,662,096 SNPs were available from at least two of the five cohorts. A quantile-quantile plot of the

from meta-analysis across all GWAS cohorts is shown in Supplementary Fig. 1. Details on the overall genomic in flation factor and number of analyzed SNPs for each cohort are provided in Supplementary Table 2. For GWAS meta-analysis, independence of genetic association with UL was de fined as SNPs in low LD ( r 2 0.6) with index SNPs. Any adjacent regions within 250 kb of one another were combined and classified as a single locus of association. All associated genomic regions were confirmed to have lead SNPs that were either directly genotyped or that met a rigorously high quality imputation threshold (INFO > 0.9) in at least two cohorts. Linkage disequilibrium score regression (LDSC) . Analysis of residual in flation in test statistics was conducted using univariate LDSC regression. Individual χ2 values for each SNP analyzed in the GWAS meta-analysis was regressed onto LD scores estimated from the 1000 G EUR panel. Heritability calculations can be derived from analyzing the slope and y-axis intercept of the slope of the regression line. Percent impact of confounders, such as population strati fication, on test statistic inflation are quanti fied as the LDSC ratio [((intercept –1))/((mean χ 2–1))] × 100%. Remaining effects [(1 –LDSC ratio) × 100%] represent the percentage of in flation attributed to polygenic heritability. Univariate LDSC regression was conducted using the LDSC software ( https://github.com/bulik/ldsc.git). Adjustment of herit- ability ( h 2) calculations to the liability scale were performed by accounting for the prevalence of UL in the sample (~0.132) compared to the general population (~0.300). LDSC software was also used to estimate the genetic correlation between UL and endometriosis (Endo) using endometriosis GWA meta-analysis summary data from Sapkota et al. 45 consisting of only European cohorts. The heritability and LD score intercepts for both traits were computed, in this analysis with SNPs present in both datasets for LDSC regression again using LD scores from the 1000 G EUR panel. Genetic correlation between traits was estimated as the genetic covariance among SNPs / √ h 2UL × h2Endo. Approximate conditional analysis . Approximate conditional analysis, imple- mented in GCTA 83, was conducted to dissect distinct signals of association at each locus. Of note, where lead SNPs at adjacent loci mapped within 1 Mb of each other, loci were combined as a single region for conditional analysis, to account for potential LD between SNPs in different loci. GCTA makes use of meta-analysis association summary statistics (log-OR and corresponding standard error) and a

panel of individual-level genotype data to obtain LD between all pairs of SNPs at a locus (or region) that approximates the covariance in effect estimates in a joint model. For these analyses, we made use of 5000 randomly selected white British women (of European descent) as reference. We used the -cojo-slct option to select index variants for each distinct association signal, at a locus-wide signi ficance threshold of P <1 0 −5, which is a conservative Bonferroni correction for the number of SNPs mapping to a locus. For loci with multiple distinct association signals, we obtained the conditional association summary statistics for each by conditioning on all other index SNPs at the locus (or region) using the -cojo-cond option. Fine-mapping distinct association signals . For each distinct association signal, association summary statistics (log-OR and corresponding standard error) were extracted from the meta-analysis for all SNPs at the locus (or region). For loci with a single signal of association, we made use of association summary statistics from the unconditional meta-analysis. For loci with multiple signals of association, we made use of association summary statistics from the approximate conditional analysis. For each SNP j, we calculated an approximate Bayes ’ factor in favor of NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 ARTICLE NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications 7 association84, given by Λj ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffi ffi Vj Vj þ ω s exp ωβ2 j 2Vj Vj þ ω /C16/C17 2 4 3 5; ð1Þ where βj and Vj denote the estimated log-OR and corresponding variance from the meta-analysis. The parameter ω denotes the prior variance in allelic effects, taken here to be 0.04 for a disease outcome 84. We then calculated the posterior prob- ability, πj, that the jth SNP is causal for the association signal, given by πCj ¼ ΛjP k Λk ; ð2Þ where the summation is over all retained variants in the locus (or region). The 99% credible set for each signal was then constructed by: (i) ranking all variants according to their Bayes ’ factor, Λ j; and (ii) including ranked variants until their cumulative posterior probability of causality is at least 0.99. Heavy menstrual bleeding (HMB) GWAS . The HMB GWAS was conducted using data from the UKBB cohort (Supplementary Methods). Both hospital-linked medical records and self-report were considered to identify women with a history of UL, while for HMB only hospital-linked medical records were taken into account. Controls had no previous history of either UL or HMB. Association analyses were performed using a linear mixed model (BOLT-LMM v.2.3.2) 81. Effect size estimates ( β and SE) from the linear mixed-model were converted to log- odds scale. Mendelian randomization (MR) . MR analyses were performed using the Two Sample Mendelian Randomization R package. GWAS summary statistics on HMB from the UKBB cohort were used to create outcome data for MR between UL (exposure) and HMB (outcome). To avoid overlap between samples in the exposure and outcome cohorts, we performed UL GWAS excluding all the HMB cases 85.L D pruning was performed to con firm no duplication of exposure haplotypes or SNPs. Subsequently, data were harmonized to ensure the same reference alleles were used in exposure and outcome GWAS and that the variants were present in both GWAS datasets. Thirteen independent SNPs associated with UL from our GWAS meta- analysis were available in the HMB GWAS summary data to test for a causal effect of UL on HMB. There were too few signi ficant SNPs available for HMB to test for a causal effect of HMB on UL. GWAS summary statistics on endometriosis (with laparoscopy, without laparoscopy, and all self-reported endometriosis cases) from the WHS cohort were used to create outcome data for MR between UL (exposure) and endometriosis (outcome). To avoid overlap between samples in the exposure and outcome cohorts, WGHS was excluded from the UL GWAS for MR analysis. Twenty-two independent SNPs associated with UL were available in the endometriosis GWAS summary data to test for a causal effect of UL on endometriosis. For reverse causation model, summary statistics from seven GWAS listing ‘endometriosis’ as the phenotype of interest were available from the EMBL-EBI NHGRI GWAS catalog (Study Accession: GCST000797, GCST001894, GCST001720, GCST005906, GCST000916, GCST004549, GCST004873). Due to a low number of cases/controls or insuf ficient number of SNPs after LD pruning and data harmonizing, only one of the studies (GCST004549) was included in the analysis. Sixteen independent SNPs associated with endometriosis were available in our UL GWAS summary data to test for a causal effect of endometriosis on UL. The IVW model was used to test causality between exposure and outcome. In addition, the IVW (Q) method was used to test for heterogeneity, leave-one-out sensitivity analysis to identify the effect of individual SNPs, and MR Egger for horizontal pleiotropy. Due to heterogeneity in our initial MR estimates, we have now leveraged a similar approach to the one published in Corbin et al., 2016, to identify the minimum set of variants that when used as a genetic instrument eliminate heterogeneity 86. We also conducted the MR-PRESSO test to identify and adjust for variants causing signi ficant bias through horizontal pleiotropy87. MR-PRESSO method (1) applies a global test to evaluate whether horizontal pleiotropy is present, (2) calculates the causal estimates incorporating correction for the detected horizontal pleiotropy, and (3) applies a distortion test to evaluate if the causal estimate is signi ficantly different after adjustment for outliers. We have reported the initial estimates along with the outlier-adjusted estimates as both the global and distortion tests showed signi ficant results. Co-morbidity analyses. Each cohort was analyzed individually with study-speci fic models chosen and covariates coded as appropriate for each cohort ’s data structure (Supplementary Methods). The study-speci fic effect estimates were combined using meta-analysis to obtain a summary RR. Between study heterogeneity was assessed with Cochran Q statistic and the I 2 statistic88. Because heterogeneity among the studies was identi fied, we reported a random-effects IVW effect estimate based on the DerSimonian and Laird method 89. LD Hub, gene-set, cell/tissue enrichment, and SMR analyses . LD Hub ana- lysis90 was conducted using summary-level results data of UL GWAS meta-analysis to estimate the genetic correlation between UL and 21 different traits/diseases, including various reproductive traits and cardiometabolic traits/diseases that have publicly available GWAS results on the LD Hub repository. Multiple-testing cor- rection was performed (0.05/21 = 2.4 × 10−3). For gene-set and cell/tissue enrichment, summary statistics from the set of 8971 SNPs with suggestive ( P< 1× 10−5) or signi ficant associations ( P< 5×1 0 −8) were analyzed using the Data- driven Expression-Prioritized Integration for Complex Traits (DEPICT) soft- ware 65. Using the 1000 G EUR panel as a reference for LD calculations and the ‘clumping’ algorithm in PLINK 91, we identi fied 104 independent loci at the sug- gestive threshold for DEPICT analyses (Supplementary Data 2). FDR < 0.05 was considered statistically signi ficant. For SMR analysis, SNPs present in at least two studies in the summary statistics were considered. The analysis was run using eQTL data from the CAGE blood dataset 66 and mQTLs from the LBC_BSGS blood dataset67. FOXO1 immunohistochemistry and genotyping . FOXO1 immunostaining was performed on two replicate tissue microarrays (TMAs) containing 335 UL and 36 myometrium tissue samples from 200 white women of European ancestry obtained from myomectomies and hysterectomies. Tissue cores on the replicate TMAs represent different regions of the same samples, which include corresponding tumor-normal tissue pairs from 35 women. Immunohistochemistry was carried out using the BOND staining system (Leica Biosystems, Buffalo Grove, IL) with a primary antibody dilution 1:100 (clone C29H4, Cell Signaling Technology, Dan- vers, MA) and hematoxylin as the counterstain. Immunostaining was analyzed using Aperio ImageScope software (Leica Biosystems). Each core was evaluated for the ratio of stain to counterstain taking into account variable cellularity between cores. Only nuclear labeling of the protein was evaluated. The average stain-to- counterstain ratio was compared between patient-matched UL and myometrium samples using a paired t-test (two-tailed), while an unpaired t-test (Welch ’s t-test, two-tailed) was applied to compare all UL and myometrium samples. Genomic DNA from 109 UL on the TMA was available for genotyping. These UL were genotyped for two SNPs with genome-wide signi ficance at the 13q14.11 locus: rs6563799 and rs7986407. For each SNP, the average FOXO1 stain-to-counterstain ratio was compared across increasing dosage of the risk allele using a one-way analysis of variance test (two-tailed). We also performed an unpaired t-test to compare mean expression of UL homozygous for the risk variant against the other genotypes (Welch ’s t-test, two-tailed). P-values < 0.05 were considered statistically significant. URLs. For WHS see http://whs.bwh.harvard.edu/; for NFBC see http://www.oulu. fi/nfbc/; for QIMR see http://www.qimrberghofer.edu.au/; for UK Biobank see http://www.ukbiobank.ac.uk/; for 23andMe see https://research.23andme.com/; for METAL see http://csg.sph.umich.edu/abecasis/metal/; for LDSC see https://github. com/bulik/ldsc.git; for DEPICT see https://data.broadinstitute.org/mpg/depict/; for SMR see http://cnsgenomics.com/software/smr/; and for PLINK see http://pngu. mgh.harvard.edu/purcell/plink/. Reporting summary . Further information on research design is available in the Nature Research Reporting Summary linked to this article. Data availability The authors declare that the data supporting the findings of this study are available within the article and its Supplementary Information files. Summary statistics for the top 10,000 UL GWAS meta-analysis variants are provided in Supplementary Data 6. UL GWAS meta-analysis summary statistics (without 23andMe), UL GWAS limited by HMB and HMB GWAS summary statistics will be made available through the NHGRI- EBI GWAS Catalog https://www.ebi.ac.uk/gwas/downloads/summary-statistics.T o request access to 23andMe GWAS summary statistics, please visit https:// research.23andme.com/dataset-access/. Received: 28 February 2019; Accepted: 10 September 2019; Published online: 24 October 2019

1. Stewart, E. A. Clinical practice. Uterine fibroids. N. Engl. J. Med. 372, 1646–1655 (2015). 2. Cramer, S. F. & Patel, A. The frequency of uterine leiomyomas. Am. J. Clin. Pathol. 94, 435 –438 (1990). 3. Marino, J. L. et al. Uterine leiomyoma and menstrual cycle characteristics in a population-based cohort study. Hum. Reprod. 19, 2350 –2355 (2004). 4. Pavone, D., Clemenza, S., Sorbi, F., Fambrini, M. & Petraglia, F. Epidemiology and risk factors of uterine fibroids. Best Pr. Res Clin. Obstet. Gynaecol. 46, 3–11 (2018). 5. Treloar, S. A., Martin, N. G., Dennerstein, L., Raphael, B. & Heath, A. C. Pathways to hysterectomy: Insights from longitudinal twin research. Am. J. Obstet. Gynecol. 167,8 2 –88 (1992). ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 8 NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications 6. Vikhlyaeva, E. M., Khodzhaeva, Z. S. & Fantschenko, N. D. Familial predisposition to uterine leiomyomas.Int. J. Gynecol. Obstet. 51,1 2 7–131 (1995). 7. Marshall, L. M. et al. Variation in the incidence of uterine leiomyoma among premenopausal women by age and race. Obstet. Gynecol. 90, 967 –973 (1997). 8. Luoto, R. et al. Heritability and risk factors of uterine fibroids-the Finnish Twin Cohort study. Maturitas 37,1 5 –26 (2000). 9. Faerstein, E., Szklo, M. & Rosenshein, N. Risk factors for uterine leiomyoma: a practice-based case-control study. I. African-American heritage, reproductive history, body size, and smoking. Am. J. Epidemiol. 153,1 –10 (2001). 10. Van Voorhis, B. J., Romitti, P. A. & Jones, M. P. Family history as a risk factor for development of uterine leiomyomas. Results of a pilot study. J. Reprod. Med. 47, 663 –669 (2002). 11. Cha, P. C. et al. A genome-wide association study identi fies three loci associated with susceptibility to uterine fibroids. Nat. Genet. 43, 447 –450 (2011). 12. Eggert, S. L. et al. Genome-wide linkage and association analyses implicate FASN in predisposition to uterine leiomyomata. Am. J. Hum. Genet. 91, 621–628 (2012). 13. Gallagher, C. S. et al. Genome-wide association analysis identi fies 27 novel loci associated with uterine leiomyomata revealing common genetic origins with endometriosis. Preprint at https://www.biorxiv.org/content/10.1101/324905v1 (2018). 14. Rafnar, T. et al. Variants associating with uterine leiomyoma highlight genetic

shared by various cancers and hormone-related traits. Nat. Commun. 9, 3636 (2018). 15. Välimäki, N. et al. Genetic predisposition to uterine leiomyoma is determined by loci for genitourinary development and genome stability. Elife 7, e37110 (2018). 16. Hellwege, J. N. et al. A multi-stage genome-wide association study of uterine fibroids in African Americans. Hum. Genet. 136, 1363 –1373 (2017). 17. Fusco, A. & Fedele, M. Roles of HMGA proteins in cancer. Nat. Rev. Cancer 7, 899–910 (2007). 18. Schoenberg Fejzo, M. et al. Translocation breakpoints upstream of the HMGIC gene in uterine leiomyomata suggest dysregulation of this gene by a mechanism different from that in lipomas. Genes Chromosomes Cancer 17, 1–6 (1996). 19. Williams, A. J., Powell, W. L., Collins, T. & Morton, C. C. HMGI(Y) expression in human uterine leiomyomata. Involvement of another high- mobility group architectural factor in a benign neoplasm. Am. J. Pathol. 150, 911–918 (1997). 20. Sornberger, K. S. et al. Expression of HMGIY in three uterine leiomyomata with complex rearrangements of chromosome 6. Cancer Genet. Cytogenet. 114,9 –16 (1999). 21. Chan, B. C. et al. BRE enhances in vivo growth of tumor cells. Biochem Biophys. Res. Commun. 326, 268 –273 (2005). 22. Ono, M. et al. Paracrine activation of WNT/ β-catenin pathway in uterine leiomyoma stem cells promotes tumor growth. Proc. Natl Acad. Sci. USA 110, 17053–17058 (2013). 23. Mehine, M. et al. Integrated data analysis reveals uterine leiomyoma subtypes with distinct driver pathways and biomarkers. Proc. Natl Acad. Sci. USA 113, 1315–1320 (2016). 24. Shi, Y. et al. A genome-wide association study identi fies two new cervical cancer susceptibility loci at 4q12 and 17q12. Nat. Genet. 45, 918 –922 (2013). 25. Kuchenbaecker, K. B. et al. Identi fication of six new susceptibility loci for invasive epithelial ovarian cancer. Nat. Genet. 47, 164 –171 (2015). 26. Phelan, C. M. et al. Identi fication of 12 new susceptibility loci for different histotypes of epithelial ovarian cancer. Nat. Genet. 49, 680 –691 (2017). 27. Haiman, C. A. et al. A common variant at the TERT-CLPTM1L locus is associated with estrogen receptor-negative breast cancer. Nat. Genet . 43, 1210–1214 (2011). 28. Hamdi, Y. et al. Association of breast cancer risk in BRCA1 and BRCA2 mutation carriers with genetic variants showing differential allelic expression: identification of a modi fier of breast cancer risk at locus 11q22.3. Breast Cancer Res. Treat. 161, 117 –134 (2017). 29. Shete, S. et al. Genome-wide association study identi fies five susceptibility loci for glioma. Nat. Genet. 41, 899 –904 (2009). 30. Melin, B. S. et al. Genome-wide association study of glioma subtypes identi fi es specific differences in genetic susceptibility to glioblastoma and non- glioblastoma tumors. Nat. Genet. 49, 789 –794 (2017). 31. Figueroa, J. D. et al. Genome-wide association study identi fies multiple loci associated with bladder cancer risk. Hum. Mol. Genet . 23, 1387 –1398 (2014). 32. Petersen, G. M. et al. A genome-wide association study identi fies pancreatic cancer susceptibility loci on chromosomes 13q22.1, 1q32.1 and 5p15.33. Nat. Genet. 42, 224 –228 (2010). 33. Wolpin, B. M. et al. Genome-wide association study identi fies multiple susceptibility loci for pancreatic cancer. Nat. Genet. 46, 994 –1000 (2014). 34. Zhang, M. et al. Three new pancreatic cancer susceptibility signals identi fied on chromosomes 1q32.1, 5p15.33 and 8q24.21. Oncotarget 7, 66328–66343 (2016). 35. Forbes, S. A. et al. COSMIC: mining complete cancer genomes in the Catalogue of Somatic Mutations in Cancer. Nucleic Acids Res. 39, D945–D950 (2011). 36. Lutzmann, M. et al. MCM8- and MCM9-de ficient mice reveal gametogenesis defects and genome instability due to impaired homologous recombination. Mol. Cell 47, 523 –534 (2012). 37. He, C. et al. Genome-wide association studies identify loci associated with age at menarche and age at natural menopause. Nat. Genet. 41, 724 –728 (2009). 38. AlAsiri, S. et al. Exome sequencing reveals MCM8 mutation underlies ovarian failure and chromosomal instability. J. Clin. Invest. 125, 258 –262 (2015). 39. Stacey, S. N. et al. A germline variant in the TP53 polyadenylation signal confers cancer susceptibility. Nat. Genet. 43, 1098 –1103 (2011). 40. Enciso-Mora, V. et al. Low penetrance susceptibility to glioma is caused by the TP53 variant rs78378222. Br. J. Cancer 108, 2178 –2185 (2013). 41. Diskin, S. J. et al. Rare variants in TP53 and susceptibility to neuroblastoma. J. Natl Cancer Inst. 106, dju047 (2014). 42. Johnson, N. et al. Counting potentially functional variants in BRCA1, BRCA2 and ATM predicts breast cancer susceptibility. Hum. Mol. Genet . 16, 1051–1057 (2007). 43. Schumacher, F. R. et al. Association analyses of more than 140,000 men identify 63 new prostate cancer susceptibility loci. Nat. Genet. 50, 928 –936 (2018). 44. Kinnersley, B. et al. Genome-wide association study identi fies multiple susceptibility loci for glioma. Nat. Commun. 6, 8559 (2015). 45. Sapkota, Y. et al. Meta-analysis identi fies five novel loci associated with endometriosis highlighting key genes involved in hormone metabolism. Nat. Commun. 8, 15539 (2017). 46. Ruth, K. S. et al. Genome-wide association study with 1000 genomes imputation identi fies signals for nine sex hormone-related phenotypes. Eur. J. Hum. Genet . 24, 284 –290 (2016). 47. Pickrell, J. K. et al. Detection and interpretation of shared genetic in fluences on 42 human traits. Nat. Genet. 48, 709 –717 (2016). 48. Uno, S. et al. A genome-wide association study identi fies genetic variants in the CDKN2BAS locus associated with endometriosis in Japanese. Nat. Genet. 42, 707 –710 (2010). 49. Nyholt, D. R. et al. Genome-wide association meta-analysis identi fies new endometriosis risk loci. Nat. Genet . 44, 1355 –1359 (2012). 50. Albertsen, H. M., Chettier, R., Farrington, P. & Ward, K. Genome-wide association study link novel loci to endometriosis. PLoS One 8, e58257 (2013). 51. Bulun, S. E. Endometriosis. N. Engl. J. Med. 360, 268 –279 (2009). 52. Biason-Lauber, A., Konrad, D., Navratil, F. & Schoenle, E. J. A WNT4 mutation associated with Mullerian-duct regression and virilization in a 46,XX woman. N. Engl. J. Med. 351, 792 –798 (2004). 53. Franco, H. L. et al. WNT4 is a key regulator of normal postnatal uterine development and progesterone signaling during embryo implantation and decidualization in the mouse. FASEB J. 25, 1176 –1187 (2011). 54. Powell, J. E. et al. Endometriosis risk alleles at 1p36.12 act through inverse regulation of CDC42 and LINC00339. Hum. Mol. Genet. 25, 5046 –5058 (2016). 55. Rae, J. M. et al. GREB 1 is a critical regulator of hormone dependent breast cancer growth. Breast Cancer Res. Treat. 92, 141 –149 (2005). 56. Rae, J. M. et al. GREB1 is a novel androgen-regulated gene required for prostate cancer growth. Prostate 66, 886 –894 (2006). 57. Bondesson, M., Hao, R., Lin, C. Y., Williams, C. & Gustafsson, J. A. Estrogen receptor signaling during vertebrate development. Biochim Biophys. Acta 1849, 142 –151 (2015). 58. Layman, L. C. et al. Delayed puberty and hypogonadism caused by mutations in the follicle-stimulating hormone beta-subunit gene. N. Engl. J. Med. 337, 607–611 (1997). 59. Demeestere, I. et al. Follicle-stimulating hormone accelerates mouse oocyte development in vivo. Biol. Reprod. 87,1 –11 (2012). 60. Missmer, S. A. & Cramer, D. W. The epidemiology of endometriosis. Obstet. Gynecol. Clin. North Am. 30,1 –19 (2003). 61. Zondervan, K. T. et al. Endometriosis. Nat. Rev. Dis. Prim. 4, 9 (2018). 62. Shafrir, A. L. et al. Risk for and consequences of endometriosis: a critical epidemiologic review. Best. Pr. Res. Clin. Obstet. Gynaecol. 51,1 –15 (2018). 63. Marshall, L. M. et al. A prospective study of reproductive factors and oral contraceptive use in relation to the risk of uterine leiomyomata. Fertil. Steril. 70, 432 –439 (1998). 64. Uimari, O. et al. Uterine fibroids and cardiovascular risk. Hum. Reprod. 31, 2689–2703 (2016). 65. Pers, T. H. et al. Biological interpretation of genome-wide association studies using predicted gene functions. Nat. Commun. 19, 5890 (2015). 66. Lloyd-Jones, L. R. et al. The genetic architecture of gene expression in peripheral blood. Am. J. Hum. Genet. 100, 228 –237 (2017). 67. McRae, A. et al. Identi fication of 55,000 Replicated DNA Methylation QTL. Sci. Rep. 8, 17605 (2018). NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 ARTICLE NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications 9 68. Xing, Y. Q. et al. The regulation of FOXO1 and its role in disease progression. Life Sci. 193, 124 –131 (2018). 69. Jackson, J. G., Kreisberg, J. I., Koterba, A. P., Yee, D. & Brattain, M. G. Phosphorylation and nuclear exclusion of the forkhead transcription factor FKHR after epidermal growth factor treatment in human breast cancer cells. Oncogene 19, 4574 –4581 (2000). 7 0 . H u a n g ,H . ,M u d d i m a n ,D .C .&T i n d a l l ,D .J .A n d r o g e n sn e g a t i v e l yr e g u l a t e forkhead transcription factor FKHR (FOXO1) through a proteolytic mechanism in prostate cancer cells. J. Biol. Chem. 279, 13866 –13877 (2004). 71. Goto, T. et al. Mechanism and functional consequences of loss of FOXO1 expression in endometrioid endometrial cancer cells. Oncogene 27,9 –19 (2008). 72. Zhang, B., Gui, L. S., Zhao, X. L., Zhu, L. L. & Li, Q. W. FOXO1 is a tumor suppressor in cervical cancer. GMR 14, 6605 –6616 (2015). 73. Kovacs, K. A. et al. Involvement of FKHR (FOXO1) transcription factor in human uterus leiomyoma growth. Fertil. Steril. 94, 1491 –1495 (2010). 74. Lv, J. et al. Reduced expression of 14-3-3 gamma in uterine leiomyoma as identified by proteomics. Fertil. Steril. 90, 1892 –1898 (2008). 75. Shen, Q. et al. Overexpression of the 14-3-3gamma protein in uterine leiomyoma cells results in growth retardation and increased apoptosis. Cell Signal 45,4 3 –53 (2018). 76. Hoekstra, A. V. et al. Progestins activate the AKT pathway in leiomyoma cells and promote survival. J. Clin. Endocrinol. Metab. 94, 1768 –1774 (2009). 77. Howie, B. N., Donnelly, P. & Marchini, J. A flexible and accurate genotype imputation method for the next generation of genome-wide association studies. PLoS Genet. 5, e1000529 (2009). 78. Delaneau, O., Marchini, J. & Zagury, J. F. A linear complexity phasing method for thousands of genomes. Nat. Methods 9, 179 –181 (2011). 79. Price, A. L. et al. Principal components analysis corrects for strati fication in genome-wide association studies. Nat. Genet . 38, 904 –909 (2006). 80. Bycroft, C. et al. The UK Biobank resource with deep phenotyping and genomic data. Nature 562, 203 –209 (2018). 81. Loh, P. R. et al. Ef fi cient Bayesian mixed-model analysis increases association power in large cohorts. Nat. Genet. 47, 284 –290 (2015). 82. Willer, C. J., Li, Y. & Abecasis, G. R. METAL: fast and ef ficient meta-analysis of genomewide association scans. Bioinformatics 26, 2190–2191 (2010). 83. Yang, J. et al. Conditional and joint multiple-SNP analysis of GWAS summary statistics identi fies additional variants in fluencing complex traits. Nat. Genet. 44, 369 –375 (2012). 84. Wake field, J. A. Bayesian measure of the probability of false discovery in genetic epidemiology studies. Am. J. Hum. Genet . 81, 208 –227 (2007). 85. Burgess, S., Davies, N. M. & Thompson, S. G. Bias due to participant overlap in two-sample Mendelian randomization. Genet. Epidemiol. 40, 597 –608 (2016). 86. Corbin, L. J. et al. BMI as a modi fiable risk factor for type 2 diabetes: re fining and understanding causal estimates using Mendelian randomization. Diabetes 65, 3002 –3007 (2016). 87. Verbanck, M., Chen, C. Y., Neale, B. & Do, R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat. Genet. 50, 693 –698 (2018). Erratum in: Nat Genet 50, 1196 (2018). 88. Higgins, J. P., Thompson, S. G., Deeks, J. J. & Altman, D. G. Measuring inconsistency in meta-analyses. BMJ 327, 557 –560 (2003). 89. DerSimonian, R. & Laird, N. Meta-analysis in clinical trials. Control Clin. Trials 7, 177 –188 (1986). 90. Zheng, J. et al. LD Hub: a centralized database and web interface to perform LD score regression that maximizes the potential of summary level GWAS data for SNP heritability and genetic correlation analysis. Bioinformatics 33, 272–279 (2017). 91. Purcell, S. et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am. J. Hum. Genet . 81, 559 –575 (2007).

The authors thank all of the women and their families who participated in WGHS, NFBC, QIMR, UK Biobank, 23andMe, and NHSII, and acknowledge the Channing Division of Network Medicine, Department of Medicine, Brigham and Women ’s Hos- pital and Harvard Medical School. This study was supported by the U.S. National Institutes of Health (NIH)/Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) grant HD060530 to C.C.M. C.C.M. is also sup- ported by the NIHR Manchester Biomedical Research Centre. N.M. acknowledges support from the Academy of Finland (295693) and Orion Research Foundation. H.R.H. is supported by NIH K22 CA193860. T.F. is supported by the NIHR Biomedical Research Centre, Oxford. S.E.M. is supported by the National Health and Medical Research Council (NHMRC) Fellowship Scheme (1103623). We thank the Specialized Histo- pathology Core of the Dana-Farber/Harvard Cancer Center for FOXO1 immunostaining. The Dana-Farber/Harvard Cancer Center is supported in part by an NCI Cancer Center Support Grant P30 CA06516. Further acknowledgements are provided in Supplementary Note 1. Author contributions C.S.G., S.A.M., K.T.Z. and C.C.M. designed the study. O.U., C.M.B., H.M., M.-R.J., J.E.B, S.E.M., D.R.N., P.A.L., J.N.P. and the 23andMe Research team contributed to pheno- typic/clinical aspects of the cohorts. O.U., J.P.C., N.R., T.F., D.R.V.-E., T.L.E., F.D., V.K., P.M.R., S.D.G., S.E.M., G.W.M., D.R.N., D.A.H., J.Y.T., the 23andMe Research team, J.R.B.P., P.A.L., J.N.P., N.G.M., A.P.M., D.I.C. and K.T.Z. contributed to genotyping, quality control, imputation, and/or association analysis of the genotyping data. C.S.G. and N.R. performed the UL GWAS meta-analysis. N.R. conducted the HB GWAS. R.M.C., A.P.M. and D.I.C. provided statistical genetics advice. C.S.G., N.M., N.R., Z.R., S.M., G.W.M. and A.P.M. carried out or assisted with GWAS downstream analyses. C.S.G., H.R.H., O.U., N.S., N.R., K.L.T, J.E.B, S.A.M. and K.T.Z. contributed to large-scale epidemiologic analysis. N.M., C.S.G. and H.R.H. drafted the paper. G.W.M., N.G.M., A.P.M., D.I.C., S.A.M., K.T.Z and C.C.M provided critical comments on the paper, draft, and analysis. All authors read and approved the final paper. Competing interests K.T.Z and C.M.B through Oxford University have research collaborations in benign gynecology with Bayer AG, Roche Diagnostics, Volition UK, and M DNA Life Sciences. D.A.H., J.Y.T., and members of the 23andMe Research Team are employees of 23andMe, Inc., and hold stock or stock options in 23andMe. The remaining authors declare no competing interests. Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41467- 019-12536-4. Correspondence and requests for materials should be addressed to N.M. or C.C.M. Peer review information Nature Communications thanks Siddhartha Kar and Joellen Schildkraut for their contribution to the peer review of this work. Peer reviewer reports are available. Reprints and permission information is available at http://www.nature.com/reprints Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional af filiations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party

in this article are included in the article ’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. © The Author(s) 2019, corrected publication 2022 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 10 NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications C.S. Gallagher 1,30, N. Mäkinen 2,30, H.R. Harris 3,30, N. Rahmioglu 4,30, O. Uimari 5,6, J.P. Cook 7, N. Shigesi 5, T. Ferreira4,8, D.R. Velez-Edwards 9, T.L. Edwards 10, S. Mortlock 11, Z. Ruhioglu 2, F. Day 12, C.M. Becker 5, V. Karhunen 13,14,15, H. Martikainen 6, M.-R. Järvelin 13,14,15,16,17, R.M. Cantor 18, P.M. Ridker 19, K.L. Terry 20,21, J.E. Buring19, S.D. Gordon 22, S.E. Medland 23, G.W. Montgomery 11,22, D.R. Nyholt 22,24, D.A. Hinds 25, J.Y. Tung 25, the 23andMe Research Team, J.R.B. Perry 12, P.A. Lind 23, J.N. Painter 23, N.G. Martin 22, A.P. Morris 4,7, D.I. Chasman 19,31, S.A. Missmer 21,26,31, K.T. Zondervan 4,5,31 & C.C. Morton 2,27,28,29,31 1Department of Genetics, Harvard Medical School, Boston, MA 02115, USA. 2Department of Obstetrics and Gynecology, Brigham and Women ’s Hospital, Harvard Medical School, Boston, MA 02115, USA. 3Program in Epidemiology, Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, WA 98109, USA. 4Wellcome Centre for Human Genetics, University of Oxford, Oxford OX3 7BN, UK. 5Endometriosis CaRe Centre, Nuf field Department of Women ’s and Reproductive Health, University of Oxford, John Radcliffe Hospital, Oxford OX3 9DU, UK. 6Department of Obstetrics and Gynecology, Oulu University Hospital and PEDEGO Research Unit & Medical Research Center Oulu, University of Oulu and Oulu University Hospital, 90220 Oulu, Finland. 7Department of Biostatistics, University of Liverpool, Liverpool L69 3GL, UK. 8Big Data Institute, Li Ka Shing Center for Health Information and Discovery, Oxford University, Oxford OX3 7LF, UK. 9Vanderbilt Genetics Institute, Vanderbilt Epidemiology Center, Institute for Medicine and Public Health, Department of Obstetrics and Gynecology, Vanderbilt University Medica l Center, Nashville, TN 37203, USA. 10Division of Epidemiology, Department of Medicine, Institute for Medicine and Public Health, Vanderbilt Genetics Institute, Vanderbilt University Medical Center, Nashville, TN 37203, USA. 11Institute for Molecular Bioscience, University of Queensland, Brisbane, QLD 4072, Australia. 12MRC Epidemiology Unit, University of Cambridge School of Clinical Medicine, Institute of Metabolic Science, Cambridge Biomedical Campus, Cambridge CB2 0QQ, UK. 13Center for Life Course Health Research, Faculty of Medicine, University of Oulu, 90220 Oulu, Finland. 14Unit of Primary Health Care, Oulu University Hospital, 90220 Oulu, Finland. 15Department of Epidemiology and Biostatistics, MRC-PHE Centre for Environment and Health, School of Public Health, Imperial College London, London W2 1PG, UK. 16Biocenter Oulu, University of Oulu, 90220 Oulu, Finland. 17Department of Life Sciences, College of Health and Life Sciences, Brunel University London, Uxbridge, Middlesex UB8 3PH, UK. 18Department of Human Genetics, David Geffen School of Medicine, University of California at Los Angeles, Los Angeles, CA 90095, USA. 19Division of Preventative Medicine, Brigham and Women ’s Hospital, Harvard Medical School, Boston, MA, USA. 20Obstetrics and Gynecology Epidemiology Center, Brigham and Women ’s Hospital and Harvard Medical School, Boston, MA 02115, USA. 21Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, MA 02115, USA. 22Genetic Epidemiology, QIMR Berghofer Medical Research Institute, Brisbane, QLD 4006, Australia. 23Psychiatric Genetics, QIMR Berghofer Medical Research Institute, Brisbane, QLD 4006, Australia. 24Institute of Health and Biomedical Innovation and School of Biomedical Science, Queensland University of Technology, Brisbane, QLD 4059, Australia. 2523andMe, Mountain View, CA 94041, USA. 26Department of Obstetrics, Gynecology, and Reproductive Biology, College of Human Medicine, Michigan State University, Grand Rapids, MI 49503, USA. 27Department of Pathology, Brigham and Women ’s Hospital, Harvard Medical School, Boston, MA 02115, USA. 28Broad Institute of MIT and Harvard, Cambridge, MA 02142, USA. 29Manchester Centre for Audiology and Deafness, Manchester Academic Health Science Center, University of Manchester, Manchester M13 9PL, UK. 30These authors contributed equally: C.S. Gallagher, N. Mäkinen, H.R. Harris, N. Rahmioglu. 31These authors jointly supervised this work: D.I. Chasman, S.A. Missmer, K.T. Zondervan, C.C. Morton. A full list of consortium members appears at the end of the paper. the 23andMe Research Team Michelle Agee 25, Babak Alipanahi 25, Adam Auton 25, Robert K. Bell 25, Katarzyna Bryc 25, Sarah L. Elson 25, Pierre Fontanillas 25, Nicholas A. Furlotte 25, Karen E. Huber 25, Aaron Kleinman 25, Nadia K. Litterman 25, Matthew H. McIntyre 25, Joanna L. Mountain 25, Elizabeth S. Noblin 25, Carrie A.M. Northover 25, Steven J. Pitts 25, J. Fah Sathirapongsasuti 25, Olga V. Sazonova 25, Janie F. Shelton 25, Suyash Shringarpure 25, Chao Tian 25, Vladimir Vacic 25 & Catherine H. Wilson 25 NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12536-4 ARTICLE NATURE COMMUNICATIONS | (2019)10:4857 | https://doi.org/10.1038/s41467-019-12536-4 | www.nature.com/naturecommunications 11