�� 2012 Nature America, Inc. All rights reserved. Nature GeNetics��� ADVANCE ONLINE PUBLICATION A rt i c l e s To���newly���identify���loci���for���age���at���natural���menopause,���we���carried���out���a���meta-analysis���of���22���genome-wide���association���studies��� (GWAS)���in���38,968���women���of���European���descent,���with���replication���in���up���to���4,435���women.���In���addition���to���four���known���loci,���we��� identified���3���loci���newly���associated���with���age���at���natural���menopause���(at���P��� ���5��������0���8).���Candidate���genes���located���at���these���newly��� associated���loci���include���genes���implicated���in���DNA���repair���(EXO1,���HELQ,���UIMC1,���FAM175A,���FANCI,���TLK1,���POLG���and���PRIM1)������ and���immune���function���(IL11,���NLRP11���and���PRRC2A���(also���known���as���BAT2)).���Gene-set���enrichment���pathway���analyses���using���the������ full���GWAS���data���set���identified���exoDNase,���NF-kB���signaling���and���mitochondrial���dysfunction���as���biological���processes���related���to��� timing���of���menopause. RESULTS In our discovery stage of 38,968 women with natural menopause aged 40���60 (Supplementary Tables 1 and 2), we identified 20 regions with SNPs meeting the genome-wide significance criterion P 5 �� 10���8 (Fig. 1). Four of these loci confirmed earlier reports of associations on chromosomes 5, 6, 19 and 20 (refs. 14,15 regions 5b, 6a, 19a and 20, respectively, in Table 1) and 16 loci were previously unidentified. We did not confirm one reported association on chromosome 13 (13q34, rs7333181, P = 0.12). The overall genomic inflation factor was 1.03 (Fig. 1, inset SNP with lowest P value from each region, Table 1). There was no between-study effect heterogeneity across discovery studies (P 0.05/20 = 0.0025) for the 20 SNP associations presented. Within the Framingham Heart Study group, we tested for differences in effect size for the 20 SNPs in retrospectively and prospectively col- lected menopause age, and found no significant differences (data not shown). The effect sizes ranged from 0.17 years (8.7 weeks) to nearly 1 year (50.5 weeks) per each copy of the minor allele. We computed the effect sizes for dichotomized age at natural menopause in women from the Women���s Genome Health Study (WGHS). For early meno- pause, we compared women with age at menopause 45 (N = 745) to those with age at menopause 45. For late menopause, we compared women with age at menopause 54 (N = 1,632) to those with age at menopause 54. The estimated odds ratios for early menopause for the menopause-decreasing allele ranged from 1.01 to 2.03. The estimated odds ratios for late menopause for the menopause-decreasing allele ranged from 0.52 to 0.96 (Supplementary Table 3). The top SNPs in regions 2c, 5a and 19b were 400 kb but 1 Mb from the top SNP in another region on the same chromosome. The top SNP in each of these primary regions had low linkage disequilibrium (LD r2 0.5) with the top SNP in the nearby region. To determine whether these associations were independent, we carried out a conditional association analysis in the discovery study samples with the most significant SNP from each of the primary 17 regions included as covariates in the analysis. For regions 5a and 19b (rs890835 and rs12461110, respectively), the effect estimates in the conditional analysis were unchanged compared Meta-analyses identify 13 loci associated with age at menopause and highlight DNA repair and immune pathways Menopause is the cessation of reproductive function of the human ovaries. This life stage is associated with one of the major hormonal changes of women, characterized by a decline in secretion of estro- gen, progesterone and, to a lesser degree, testosterone. It influences a woman���s well-being and is associated with several major age-related diseases including cardiovascular disease, breast cancer, osteoarthri- tis and osteoporosis1. Ovarian aging is reflected by the continuous decline of the primordial follicle pool, which is established during fetal life, subsequently leading to endocrine changes owing to loss of the negative feedback from ovarian hormones on the hypothalamic- pituitary axis. In addition to follicle loss, oocyte quality diminishes with increasing age, which is believed to be due to greater meiotic nondisjunction2. Oocyte quality may be controlled at the time germ cells are formed during fetal life, but it may also reflect accumulated damage during reproductive life and/or age-related changes in granu- losa cell���oocyte communication3. Although both oocyte quantity and quality decline with increasing age, it is unclear whether they are controlled by the same mechanisms and whether they decline in parallel. The average age at natural menopause in women of Northern European descent is 50���51 years (range 40���60 years)4. Heritability estimates from twin and family studies for age at natural menopause range from 44% to 65% (refs. 5���8). Thus far most genetic associa- tion studies regarding age at menopause have focused on candidate genes9 from the estrogen pathway10,11 or vascular components12,13. Recently, two GWAS have newly identified five loci associated with age at natural menopause on chromosomes 5, 6, 13, 19 and 20 (refs. 14,15). These loci, however, explained 1.5% of the pheno- typic variation of age at natural menopause, suggesting that addi- tional loci of small effect will probably be discovered in larger samples. Therefore, we conducted a two-stage GWAS of women of European ancestry, combining the women from the two previous GWAS14,15 with new participants for a total of 38,968 women from 22 studies in the discovery stage, and 14,435 women from 21 studies in the replication stage. A full list of authors and affiliations appears at the end of the paper. Received 21 July 2011 accepted 2 December 2011 published online 22 January 2012 doi:10.1038/ng.1051

�� 2012 Nature America, Inc. All rights reserved. 2��� ADVANCE ONLINE PUBLICATION Nature GeNetics A rt i c l e s with in the discovery analysis (differences of 0.3% and 4%, respectively), and the P values were genome-wide significant. However, for region 2c, the effect size was ~12.5% lower in the conditional analysis than in the initial analysis, and the SNP P value was no longer genome-wide significant (P = 9.8 �� 10���7 Table 1), suggesting that the association with rs7606918 is not independent of the rs1018348 region 2b association. We attempted replica- tion only for the 19 SNPs that represented independent regions that reached genome- wide significance (P 5 �� 10���8), thus we did not pursue replication of rs7606918. Replication We used 21 studies contributing 14,435 women for replication of the 19 SNPs that defined the independent genome-wide signi- ficant regions from stage 1. We defined age at natural menopause using the same criteria as in the discovery studies (Supplementary Table 1). Of these studies, 17 (n = 6,639) were included in in silico replication (Supplementary Table 2) an additional 4 studies (n = 7,796) con- tributed de novo genotypes for the 19 SNPs (Supplementary Table 2 effect sizes and P values for replication and combined meta-analysis of discovery and replication samples, Table 1). There was no evidence for effect heterogeneity among the replication studies (Table 1). We also tested for heterogeneity between the in silico and de novo geno- typed studies, and found no evidence for heterogeneity of effect (data not shown), suggesting that for the significant SNPs, the genotype imputation methods did not lead to significantly different effect size estimates than would have been obtained from direct genotyping. Of the 19 SNPs, 17 were genome-wide significant and had lower P values in combined meta-analysis of the discovery and replication samples. Regions 5a and 13a showed no evidence of association in the replication samples (P 0.50) and were not genome-wide significant in combined discovery and replication meta-analysis. Four of the 17 replicated regions have been reported previously thus our analysis identified 13 regions newly associated with age at natural menopause on the basis of genome-wide significant discovery with replication. In the combined discovery and replication meta-analyses, the effect esti- mates ranged from 8.2 to 49.3 weeks per minor allele. The estimated proportion of variance explained by the 17 replicated SNPs in the four replication studies with de novo���genotyped SNPs varied from 2.5% (Osteos) to 3.7% (EPOS and BWHHS) to 4.1% (PROSPECT-EPIC). We used the largest study contributing data to our discovery GWAS (WGHS, n = 11,379) to explore whether substantial SNP-SNP inter- actions are present among the 17 replicated SNPs. We tested all 136 pairs of SNPs and found no evidence for interaction (all P 0.01). Roles���of���genes���at���or���near���newly���identified���loci All but two of the replicated SNPs are intronic or exonic to known genes (Table 2). The top SNPs in regions 6b, 12, 19b and 20 are mis- sense polymorphisms. Three of the four have been predicted to have damaging protein function by SIFT16, and one by PolyPhen2 (ref. 17). Using dbSNP and LocusZoom18, we identified the genes underlying the newly identified top regions. We used SCAN (see URLs) to identify all genes with SNPs that are in LD (r2 0.5) with our SNPs (Table 2). We identified all SNPs with r2 ��� 0.8 with our top SNPs and used several databases to determine whether the SNPs are associated with expression (Table 2). The strongest new signal was on chromosome 4 (region 4, rs4693089 P = 2.4 �� 10���19). The SNP is located in an intron of HELQ, which encodes the protein HEL308, a DNA-dependent ATPase and DNA helicase19. The second strongest new signal was on chromosome 12 (region 12, rs2277339 P = 2.5 �� 10���19). This SNP is a nonsynonymous variant in exon 1 of PRIM1. The top SNP was significantly associated with expression of PRIM1 in visual cortex, cerebellum and prefrontal cortex (Table 2). Several other previously unidentified signals are located in introns of genes for which mouse models exist. These were region 8 in ASH2L (rs2517388 P = 9.3 �� 10���15), region 15 in POLG (rs2307449 P = 3.6 �� 10���13) and region 1b in EXO1 (rs1635501 P = 8.5 �� 10���10). ASH2L encodes a trithorax group protein, and is involved in X chromosome inactivation in women20. POLG encodes the catalytic subunit of mito- chondrial DNA polymerase, the enzyme responsible for replication and repair21 of mitochondrial DNA. EXO1 is a member of the RAD2 nuclease family of proteins, which is involved in DNA replication, repair and recombination, and the top hit is in LD (r2 = 0.83) with a functional polymorphism in EXO1 that affects a transcription factor���binding site in the promoter. Region 11 (rs12294104 P = 1.5 �� 10���11) is near and in LD (r2 = 0.92) with SNPs in FSHB. Transcription of FSHB limits the rate of production of the heterodimeric follicle- stimulating hormone (FSH), a key pituitary gland���expressed hor- mone that stimulates maturation of follicles. Region 19a (rs11668344 P = 1.5 �� 10���59) is in tight LD with SNPs in IL11 this cytokine stimulates the T cell���dependent development of immunoglobulin- producing B cells. The top SNPs in two other previously unknown regions are non- synonymous coding variants. Region 6b, rs1046089 (P = 1.6 �� 10���16), is in exon 22 of PRRC2A and was associated with expression of several transcripts in the human leukocyte antigen (HLA) region in several tissues (Table 2). Region 19b, rs12461110 (P = 8.7 �� 10���10) is in exon 5 of NLRP11. PRRC2A encodes HLA-B associated transcript 2 and has several microsatellite repeats. NLRP11 encodes the nucleotide- binding domain and leucine-rich repeat���containing (NLR) family pyrin domain���containing 11 protein, which is implicated in the acti- vation of proinflammatory caspases22. Of the remaining five new regions, the top SNPs for regions 1a, 2a, 2b and 13b are located in introns. These were rs4246511 in RHBDL2 (0.24 years per minor allele, P = 9.1 �� 10���17), which is thought to func- tion as an intramembrane serine protease rs2303369 in FNDC4, which encodes fibronectin type III domain���containing 4 (P = 2.3 �� 10���12) 50 50 40 30 20 10 0 0 1 2 3 4 5 6 ���log 10 P ���log 10 P (obs.) ���log10 P (exp.) 40 30 20 10 0 1 2 3 4 5 6 7 8 9 Chromosome 10 11 12 13 14 15 16 17 18 19 20 2122 Figure 1 Discovery GWAS results. Manhattan plot of discovery meta-analysis. Inset, quantile- quantile plot of discovery primary analysis (red) and double genomic control���adjusted primary analysis (black). Obs., observed exp., expected.