Associations between polygenic risk scores for age at menarche and menopause, reproductive timing, and serum hormone levels in multiple race/ethnic groups

Abstract

Objective:

We assessed associations of genetic loci that contribute to age at menarche and menopause with sentinel menopausal traits in multiple race/ethnic groups.

Methods:

Genetic data from the Study of Women’s Health Across the Nation (SWAN) include 738 White, 366 Black, 139 Chinese, and 145 Japanese women aged 42 to 52 at baseline. We constructed standardized polygenic risk scores (PRSs) using single nucleotide polymorphisms identified from large-scale GWAS meta-analyses of ages at menopause and menarche, evaluating associations with each trait within each race/ethnic group.

Results:

Menopause PRS was significantly associated with age at menopause in White women after Bonferroni correction (p<0.004) and nominally associated in Chinese and Japanese women (p<0.05) (7.4–8.5 month delay for one standard deviation (SD) increase in PRS). Menarche PRS was significantly associated with age at menarche in White (p<0.004) and nominally associated in Black and Japanese women (p<0.05) (2.6–4.8 month delay for one SD increase). In White women, menarche and menopause PRSs were significantly associated (p<0.004) with shorter and longer (5.9 months and 9.6 months for one SD increase) reproductive lifespans, respectively, and menopause PRS with a longer menopausal transition (7.1 months for one SD increase). We observed a significant positive association (p<0.004) between menopause PRS and E2 level two years prior to menopause and a nominal association (p<0.05) two years after menopause in Japanese women.

Conclusions:

In addition to menopausal timing, PRSs associated with onset and ending of reproductive life were associated with reproductive lifespan, length of the menopausal transition, and E2 levels in different race/ethnic groups.

INTRODUCTION

Given the increase in women’s life expectancy, currently 81.1 years in the United States,¹ and a median age of menopause of approximately 51.4 years,² over one-third of women’s lives are spent in the post-menopause. Long viewed as a marker of biologic aging,³ age at menopause is strongly associated with women’s risk of mortality and chronic disease.^4–6 For example, later menopause is associated with increased risk of breast, ovarian, and endometrial cancer^7–10 while earlier menopause is associated with increased risk of cardiovascular disease, stroke, osteoporosis, and fracture.^6,11–15 Factors that influence the timing of menopause, including environmental factors such as smoking² and perifluoroalkyal and polyfluoroalkyal substances¹⁶ (PFAS), reproductive factors such as parity², and genetic loci¹⁷, are thus particularly salient to understanding and predicting women’s risk of mortality and chronic disease.

Additionally, both age at menarche and length of the reproductive lifespan, i.e. the time from menarche to menopause, have been associated with mortality and women’s risk of chronic disease. Earlier menarche and a longer reproductive lifespan are associated with increased cancer risk,¹⁸ while a shorter reproductive lifespan is associated with higher risk of cardiovascular disease.¹⁹ Age at menarche and menopause do not appear to be correlated but women with early menarche had a longer reproductive lifespan in a study conducted in 336,788 Norwegian women.²⁰ The relationship remains to be determined in other race/ethnic groups.

Although age at menopause is defined by the final menstrual period (FMP), ovarian aging is a complex phenomenon that reflects declining ovarian reserve.^21,22 Several cohort studies of midlife women have described key characteristics of reproductive aging including the duration of the menopausal transition and the timing and longitudinal trajectories of change in follicle stimulating hormone (FSH) and estradiol (E2) from pre-menopause through post-menopause.^23–29 Like timing of menopause and duration of the reproductive lifespan, duration of the menopausal transition varies across women.^30–31 Similarly, patterns of change in FSH and E2, as well as the peak and nadir of these hormone levels vary across women.^23–28

Several candidate gene studies and small genome-wide association studies (GWAS) have been conducted on age at natural menopause,^33,34 age at menarche, and serum E2 and FSH levels.^35–39 Recently, large-scale GWAS meta-analyses have been conducted for age at natural menopause and age at menarche.^17,40 No large scale GWAS have been published, however, for E2 and FSH levels observed during the menopausal transition in women, as few studies have measures of hormones, DNA, and menopausal timing available. However, large scale GWAS are beginning to emerge for other reproductive hormone levels, as well as for E2 levels in men.¹

The largest GWAS meta-analysis of age at natural menopause, conducted in nearly 70,000 women of European ancestry, identified 54 independent single nucleotide polymorphisms (SNPs) in 44 genomic loci that were genome-wide significantly associated (p<5×10⁻⁸) with age at natural menopause.¹⁷ The genome-wide significant SNPs explained approximately 6% of the variance in age at natural menopause, and variance explained increased to 21% using independent SNPs with a nominal p-value cutoff (p<0.05). The identified loci harbored genes involved in DNA damage response as well as primary ovarian insufficiency and delayed puberty. Cross-trait analysis of GWAS results from this study showed evidence of modest pleiotropy (shared genetic influences) for age at natural menopause and anthropometric traits including adult body mass index (BMI) and waist circumference in women.

The largest GWAS meta-analysis of age at menarche, conducted in over 370,000 women of European ancestry identified 389 independent genomic loci as genome-wide significantly associated with age at menarche,⁴⁰ showing that the genetic architecture of age at menarche is highly polygenic. The 389 genome-wide significant SNPs explained over 7% of the trait variance in independent samples. The set of identified genes are most highly expressed in central nervous system tissues including the hypothalamus and pituitary gland, and several of the identified genes are disrupted in rare disorders of puberty. Approximately 10% of the identified genomic loci overlapped with those reported for adult BMI.

A recent GWAS of ages at menarche and menopause in about 67,000 Japanese women found two new loci for the former and eight new loci for latter.⁴² Most of the previously identified European menarche (81%) and menopause (88%) loci showed consistent direction of effects in Japanese women. These results indicate a shared genetic basis of reproductive aging across populations of different race/ethnicities.

Despite the complex interplay among sentinel traits of reproductive aging, no studies to date have considered how already identified genetic loci associated with age at menopause and age at menarche may be related to additional biomarkers of the reproductive lifespan and ovarian aging. In order to better understand the relationship between genetic loci that contribute to reproductive timing and additional sentinel characteristics of the menopausal transition, we constructed standardized polygenic risk scores (PRS) using the genome-wide suggestive loci (p<1×10⁻⁴) from the large-scale GWAS meta-analyses of age at natural menopause and of age at menarche described above.^17,40 We then evaluated their association with a variety of traits related to the reproductive lifespan and reproductive aging, including age at natural menopause, age at menarche, length of the reproductive lifespan, length of menopausal transition, and levels of FSH and E2 two years prior to and two years after menopause in an independent multi-racial/ethnic cohort, the Study of Women’s health Across the Nation.

MATERIALS AND METHODS

The Study of Women’s Health Across the Nation (SWAN).

SWAN is a richly phenotyped, multi-racial/ethnic cohort study that has prospectively observed the natural history of reproductive aging including age at menopause and longitudinal measures of reproductive hormones. The SWAN cohort enrolled 3,302 women (935 Black, 250 Chinese, 286 Hispanic, 281 Japanese, and 1,550 White) in 1996/1997. Details of the study have been described elsewhere.⁴³ In brief, a cross-sectional screening survey identified women eligible for the cohort study at seven clinical sites. In addition to residence in the geographic area of a site, eligibility criteria included being age 42–52 years old, self-identification as White (at all sites) or as Black (at the southeastern Michigan, Boston, Chicago, or Pittsburgh sites), Chinese (at the Northern California site), Japanese (at the Southern California site) or Hispanic (at the New Jersey site). Women had to have an intact uterus, not be pregnant or lactating, have had at least one menstrual period in the past 3 months and had not used reproductive hormones in the past 3 months. The study protocol was approved by the Institutional Review Boards at each study site. Since the baseline visit, women have been followed approximately annually for 15 follow-up visits, with the study remaining in contact with 75% of surviving participants.

Participants provided written, informed consent at each visit which included interviewer-administered and self-administered questionnaires on a broad range of topics, including menstrual characteristics and current hormone therapy use. A fasting blood draw was obtained at each clinic visit in the early follicular phase and assayed for E2 and FSH. Socio-demographic characteristics including education, smoking status, and history of contraceptive use were assessed at the baseline visit. Physical assessments at baseline included measurement of height and weight. BMI was calculated as weight in kilograms divided by height in meters squared. Women also maintained a monthly menstrual calendar where they recorded prospectively the days that they bled until they reached their FMP or had a hysterectomy.

Age at menopause, menarche, reproductive lifespan and length of the menopausal transition.

Menopausal status was based on self-report of bleeding patterns obtained at each study visit including the date of their last menstrual period. Women also maintained a monthly menstrual calendar. Age at menopause was defined as age at the FMP, with the FMP defined retrospectively after observing 12 months of amenorrhea. Surgical menopause was defined by report of either hysterectomy or bilateral oophorectomy. Age at FMP could not be determined in women who began hormone therapy and had no subsequent untreated menstrual bleeds. Women who had surgical menopause or whose FMP could not be determined because of hormone use or missing data were excluded from this analysis. Recalled age at menarche was ascertained at the baseline visit. Reproductive lifespan was calculated by subtracting age at menarche from age at menopause. Using data from the menstrual calendar, age at onset of the early menopausal transition was defined based on the STRAW+10 criteria⁴⁴ as the age at which a persistent difference of at least 7 days in the length of consecutive menstrual cycles was observed with persistence defined as recurrence within 10 cycles of the first variable length cycle. Length of the menopausal transition was calculated as the time from onset of the early menopausal transition to the FMP.⁴⁵

Hormone assays.

Women were scheduled for venipuncture before 10:00 am between days two and five of their menstrual cycle. If a follicular phase sample could not be obtained, a random fasting sample was obtained. Blood was refrigerated 1–2 h after phlebotomy, and after centrifugation, the serum was aliquoted, frozen, and batched for shipment to the SWAN central laboratory. FSH assays were conducted in singlicate and E2 assays in duplicate using an ACS-180 automated analyzer (Bayer Diagnostics Corp., Norwood, MA). E2 concentrations were measured with a modified, off-line ACS-180 (E2–6) immunoassay. Inter- and intra-assay coefficients of variation averaged 10.6 and 6.4%, respectively, over the assay range, and the lower limit of detection was 1 pg/ml. Serum FSH concentrations were measured with a two-site chemiluminometric immunoassay. Inter- and intra-assay coefficients of variation were 12.0 and 6.0%, respectively, and the lower limit of detection was 1.1 IU/liter. If the blood draw was prior to menopause, we also recorded whether the measurement was taken during cycle day two and five or outside of this window.

The SWAN Genomic Substudy.

During follow-up visits 5 and 6, whole blood was collected to generate genetic materials including extracted DNA and immortalized B-lymphocyte cells. Collection occurred during a hiatus at the New Jersey site, thus no genetic materials are available for the SWAN Hispanic women. Of the original SWAN cohort, 60% of the Chinese and Japanese, 52% of the Whites, and 44.1% of the African-Americans participated. At the six participating sites, 1,757 (88.3%) of the 1,988 women who attended these clinic visits consented to provide genetic materials. Immortalized cell lines were developed successfully for 1,588 (90.4%) with 1,536 (96.7%) processed into distributable diluted, extracted DNA. A separate informed consent was obtained for the DNA collection.

Samples were genotyped on the Illumina (San Diego, CA) Multi-Ethnic Global Array (MEGA A1) at the Center for Inherited Disease Research (CIDR) at Johns Hopkins University. Genotype calling was performed using GenomeStudio version 2011.1, Genotyping Module version 1.9.4, GenTrain Version 1.0. Of the 1,536 women with available DNA, 1,464 were successfully genotyped at the genotyping center and available for quality control. Briefly, quality control procedures included sex check, relatedness check, identity check, and population structure analysis. Samples with missing call rate >5% or unresolved identity issues were removed, leaving a total of 1,462 participants. SNPs with missing call rate ≥2%, Hardy Weinberg equilibrium p-value <10⁻⁴, minor allele frequency=0, >1 Mendelian error, or >0 discordant calls in duplicate samples were excluded. Sex chromosome anomalies that may have caused genotyping errors were identified in two samples, and the X chromosomes for these two samples were removed. Imputation to the 1000G Phase 3 v5 reference panel was performed using the Michigan Imputation Server.⁴⁶ SWAN genetic data are available from dbGaP (accession number = phs001470.v1.p1).

The King-robust⁴⁷ method was used to select a maximum set of unrelated participants by removing one from each pair of women who showed cryptic relatedness at 3^rd degree or closer (11 participants were removed). Principal component analysis was used to further identify four unrelated homogenous analysis samples by excluding outliers who deviated more than 2 standard deviation (SD) from the mean on eigenvector 1 or 2 for Whites and Blacks, and who deviated more than 3 SD from the mean for the Chinese and Japanese. The differing exclusion criteria across groups were determined upon visual inspection of the PCA plots. Analysis samples included White (N=740), Black (N=368), Chinese (N=139) and Japanese (N=146). Genetic principal components (PCs) were calculated within each ethnicity. Women who were not included in any model due to missing data were excluded, leaving a total sample size of 738 White women, 366 Black women, 139 Chinese women, and 145 Japanese women.

Polygenic Risk Score (PRS).

We obtained the effect sizes (beta coefficients) of each SNP estimated from the most up to date GWAS meta-analysis conducted on age at menopause¹⁷ and age at menarche.³⁹ Polygenic risk scores for menopause and menarche were constructed as PRSj = ∑ β_ix_ij/∑ β_i where β_i is the effect size associated with the risk allele (increases age at menopause or age at menarche) for SNP i, and x_ij is the dosage of the risk allele for SNP i and individual j. Thus, a higher PRS would be associated with later onset of menopause or a later onset of menarche. Only SNPs with high imputation quality (R²>0.8) were included. We constructed two versions of the PRS for each trait. The first version of the PRS (“all SNPs”) included all independent SNPs with p-value<10⁻⁴ in the GWAS meta-analysis. In order to select independent SNPs, we performed clumping with an independent sample, the 1000 Genomes Reference Panel (phase 3 v5a), in PLINK 2.0^48,49 using 0.5 as the linkage disequilibrium (LD) r² threshold and a distance of 250kb from the index SNP. Since different race/ethnicities have different LD structures, we used appropriate reference panels to select the independent SNPs for each race/ethnicity (503 European (EUR) for White, 661 African (AFR) for Black, 504 East Asian (EAS) for Chinese and Japanese women). After clumping, the total number of SNPs included in the “all SNPs” PRS for age at menopause was 601, 1431, 886 and 886 for White, Black, Chinese and Japanese, respectively. The total number of SNPs included in the “all SNPs” PRS for age at menarche was 5,381, 14,333, 10,477 and 10,477 for White, Black, Chinese and Japanese, respectively. Four outliers (> 4 SD from the mean) were identified in the “all SNPs” menarche PRS score in White women and were removed from analysis.

The second version of the PRS (“top SNPs”) included only the top independent genome-wide significant SNPs reported by the authors of the GWAS meta-analysis, which corresponds to 54 SNPs (51 SNPs available in SWAN) for age at menopause and 389 SNPs (339 SNPs available in SWAN) for age at menarche. Since the age at menarche GWAS included chromosome X, the PRS for age at menarche could not be calculated for the two participants with X chromosome anomalies. Prior to the analyses, each PRS was standardized (mean=0, sd=1) within each race/ethnic analysis sample and the standardized score was used in the analyses. Primary analyses are focused on the “all SNPs” PRSs, but we also examined the “top SNPs” PRSs as a secondary analysis.

Analysis.

For age at menopause, age at menarche, reproductive lifespan and length of the menopausal transition, we constructed linear models with each variable as the outcome and the menarche and menopause PRSs as predictors. Covariates included the first five genetic PCs, and study site for the White and Black women. For age at menopause, reproductive lifespan, and length of the menopausal transition, we additionally adjusted for baseline BMI, smoking (current, former/never), education (high school degree or less, some college, college degree or more), and history of oral contraceptive use (yes, no). For age at menarche, we additionally adjusted for year of birth in order to account for differences in nutrition or environmental exposures, in accordance with the age at menarche GWAS.⁴⁰ We also performed two sensitivity analyses. Since reproductive lifespan is measured directly from ages at menopause and menarche, we adjusted for each of these ages in the reproductive lifespan model, separately. Similarly, we adjusted for age at menopause in the model for length of the menopausal transition.

SWAN has described the pattern of change in FSH and E2 as women transition through menopause and into the post menopause.²³ In this analysis, we aligned each women’s hormone measurements with the date of her FMP and selected the measurement that was closest to the date two years before and two years after the FMP.²³ These measures correspond with the hormone values shortly before the accelerated rise of FSH and the sharp decline in E2 associated with the start of the late menopausal transition and the time in post-menopause corresponding to when these hormone levels have stabilized.. We examined the association between hormone levels and the menopause PRS using a linear model with BMI, smoking, education, oral contraceptive use, study site and the first 5 genetic PCs as the covariates; and when modeling hormone values two years prior to menopause, an indicator as to whether the measurement was taken within the early follicular phase window (cycle day two-five) or not. For FSH, we performed a sensitivity analysis to conditionally adjust one SNP (rs11031005) that is near FSHB and has been repeatedly shown to be associated with FSH level as well as age at menopause.^{36,38,50, 51} Due to their skewed distribution, all hormone measurements were natural log transformed prior to analyses. All analyses were conducted separately by race/ethnicity. We tested 12 PRS-phenotype combinations in total for each race/ethnicity, thus statistical significance was defined using Bonferroni-corrected p-value<0.05/12 = 0.004. However, we note that this is a conservative approach, since many phenotypes are correlated. Thus we are also interested in nominal association results at p-value<0.05.

RESULTS

Descriptive Statistics

The population characteristics of the study sample stratified by race/ethnicity are shown in Table 1. These characteristics differed among the race/ethnic groups for all characteristics except median age at menopause, median length of menopausal transition in years, and median E2 measured two years prior to menopause (p<0.05). As expected, duration of the reproductive lifespan was positively correlated with age at menopause (r ranged from 0.80 to 0.88) and negatively correlated with age at menarche (r ranged from −0.39 to −0.60) across race/ethnic groups (see Supplemental Digital Content 1). Duration of the menopausal transition was positively correlated with both age at menopause (r ranged from 0.50 to 0.55) and reproductive lifespan (r ranged from 0.44 to 0.53). Measures of E2 and FSH obtained two years prior to menopause were negatively correlated (r ranged from −0.33 to −0.40), as were measures obtained two years post menopause (r ranged from −0.27 to −0.412).

Table 1:

Descriptive statistics of the Study of Women’s Health Across the Nation (SWAN) Genomic Study sample

	White (N=738)		Black (N=366)		Chinese (N=139)		Japanese (N=145)		P valuee
	N, Median (IQR)		N, Median (IQR)		N, Median (IQR)		N, Median (IQR)
Age at menopause (years)	453	52.3 (3.7)	249	52.2 (3.9)	107	52.1 (3.7)	105	52.5 (3.4)	0.22
Age at menarche (years)	734	12.0 (1.0)	361	12.0 (2.0)	139	13.0 (2.0)	145	12.0 (1.0)	0.0001
Reproductive lifespan (years)	449	39.9 (4.2)	244	39.7 (4.7)	107	38.7 (4.3)	105	40.2 (4.4)	0.0302
Length of menopausal transition (years)	329	5.2 (3.5)	137	5.4 (3.7)	86	4.9 (3.3)	81	5.0 (4.2)	0.675
BMI (kg/m2)	733	25.6 (8.4)	361	30.2 (9.4)	139	22.2 (4.0)	143	22.2 (4.3)	<0.0001
E2 prior to menopause (pg/ml)	399	50.7 (95.9)	203	42.6 (70.4)	96	38.2 (89.4)	92	46.0 (94.9)	0.0762
E2 after menopause (pg/ml)	395	16.5 (12.4)	211	21.7 (14.3)	98	14.2 (9.1)	86	12.5 (7.4)	0.0207
FSH prior to menopause (IU/L)	399	25.1 (28.7)	203	26.9 (31.9)	96	38.9 (37.9)	92	24.7 (30.0)	0.0115
FSH after menopause (IU/L)	395	105.7 (58.0)	211	81.6 (60.0)	99	105.8 (55.9)	86	112.0 (42.0)	<0.0001
Menopause PRS (“all SNPs”)a	738	1.02 (0.06)	366	1.03 (0.05)	139	1.20 (0.03)	145	1.21 (0.03)	N.A.
Menarche PRS (“all SNPs”)b	733	1.19 (0.01)	366	1.05 (0.01)	139	0.98 (0.01)	145	0.97 (0.01)	N.A.
Menopause PRS (“top SNPs”)c	738	0.89 (0.11)	366	0.88 (0.10)	139	0.98 (0.09)	145	1.00 (0.09)	<0.0001
Menarche PRS (“top SNPs”)d	737	0.91 (0.04)	366	0.91 (0.04)	139	0.90 (0.04)	145	0.90 (0.04)	<0.0001
	N (%)		N (%)		N (%)		N (%)
Education, N		736		360		139		142	< 0.0001
High school degree or less, N (%)	101 (13.7)		111 (30.8)		35 (25.2)		20 (13.8)
Some college, N (%)	218 (29.6)		152 (42.2)		31 (22.3)		53 (38.6)
College degree or more, N (%)	417 (56.7)		97 (26.9)		73 (52.5)		69 (47.6)
Oral contraceptive use, N	735		365		139		143		<0.0001
Yes, N (%)	582 (79.2)		292 (80.0)		91 (65.5)		67 (46.9)
No, N (%)	153 (79.2)		73 (20.0)		48 (34.5)		76 (46.9)
Current smoker, N	736		359		139		144		<0.0001
Yes, N (%)	97 (13.2)		88 (24.5)		2 (1.4)		16 (11.1)
No, N (%)	639 (86.8)		271 (75.5)		137 (98.6)		128 (88.9)

^aMenopause PRS was constructed using all available independent SNPs that were significant at p<10⁻⁴ in GWAS meta-analysis³⁸. Independent SNPs were selected using a clumping approach (LD r² threshold = 0.5, distance from index SNP = 250kb).

^bMenarche PRS was constructed using all available independent SNPs that were significant at p<10⁻⁴ in GWAS meta-analysis³⁹. Independent SNPs were selected using a clumping approach (LD r² threshold = 0.5, distance from index SNP = 250kb).

^cMenopause PRS was constructed using the 54 independent genome-wide significant SNPs (51 SNPs available in SWAN) that were previously reported for the age at menopause³⁸.

^dMenarche PRS was constructed using the 389 independent genome-wide significant SNPs (339 SNPs available in SWAN) that were previously reported for age at menarche³⁹.

^eP-values are from one-way ANOVA across ethnic groups for continuous variables and Chi-square test for categorical variables. P-values <0.05 are in bold.

PRS, polygenic risk score; E2, estradiol; FSH, follicle stimulating hormone; N.A., not applicable.

For the “all SNPs” PRS analyses, different sets of independent SNPs were selected at the clumping step for each race/ethnic group; thus the median scores are not directly comparable across race/ethnic groups. The “top SNPs” PRSs were based on the same set of SNPs and thus are directly comparable across the four race/ethnicity participant groups. As shown in Table 1, both menopause PRS and menarche PRS differ across the four race/ethnic groups. Both the “all SNPs” and “top SNPs” menopause and menarche PRSs were positively correlated with each other in all race/ethnic groups (see Supplemental Digital Content 2). The menopause and menarche PRS were not correlated in general.

Association between age at menopause, age at menarche, reproductive lifespan, length of the menopausal transition and polygenic risk scores.

As shown in Table 2, the “all SNPs” menopause PRSs were at least nominally associated with age at menopause in three race/ethnic groups with a significant association observed in White women (beta=0.62, p=5.94e-06), and weaker, nominal associations in Japanese (beta=0.67, p=0.0048), and Chinese (beta=0.71, p=0.0053) women. An increase of one standard deviation of the PRS was associated with 0.62–0.71 year (7.4 to 8.5 month) delay in menopause. The menopause PRS was not associated with age at menarche in any of the race/ethnic groups.

Table 2:

Association between “all SNPs” polygenic risk scores (PRS) and reproductive timing, SWAN Genomic Study

Outcome	Race/Ethnicity	N	Menopause PRSe		Menarche PRSf
			Betag	P value	Betag	P value
Age at menopausea (years)	White	443	0.62	5.94E-06	−0.07	0.595
	Black	236	0.30	0.1116	0.11	0.590
	Chinese	107	0.71	0.0053	0.10	0.673
	Japanese	101	0.67	0.0048	0.56	0.025
Age at menarcheb (years)	White	729	−0.10	0.068	0.40	6.56E-14
	Black	361	0.12	0.247	0.22	0.0375
	Chinese	139	0.02	0.903	0.20	0.195
	Japanese	145	0.03	0.836	0.28	0.039
Reproductive lifespanc (years)	White	439	0.80	5.91E-08	−0.49	0.00082
	Black	231	0.26	0.281	−0.01	0.969
	Chinese	107	0.62	0.028	−0.11	0.688
	Japanese	101	0.65	0.026	0.29	0.337
Length of menopausal transitiond (years)	White	321	0.59	0.00014	0.06	0.687
	Black	133	−0.39	0.118	0.07	0.815
	Chinese	86	0.03	0.896	0.13	0.565
	Japanese	78	0.54	0.078	0.18	0.601

^aAge at menopause = menopause PRS + menarche PRS + BMI + education + smoking + oral contraceptive use + study site (not applicable for Chinese or Japanese) + 5 genetic PCs

^bAge at menarche = menopause PRS + menarche PRS + year of birth + study sites (not applicable for Chinese or Japanese) + 5 genetic PCs

^cReproductive lifespan = menopause PRS + menarche PRS + BMI + education + smoking + oral contraceptive use + study site (not applicable for Chinese or Japanese) + 5 genetic PCs

^dLength of menopausal transition = menopause PRS + menarche PRS + BMI + education + smoking + oral contraceptive use + study site (not applicable for Chinese or Japanese) + 5 genetic PCs

^eMenarche PRS was constructed using all available independent SNPs that were significant at p<10⁻⁴ in GWAS meta-analysis³⁹. Independent SNPs were selected using a clumping approach (LD r² threshold = 0.5, distance from index SNP = 250kb).

^fMenopause PRS was constructed using all available independent SNPs that were significant at p<10⁻⁴ in GWAS meta-analysis³⁸. Independent SNPs were selected using a clumping approach (LD r² threshold = 0.5, distance from index SNP = 250kb).

^gBeta represents the change in the outcome for the participants who had a PRS that is one standard deviation above the mean compared to the participants who had a mean PRS.

P-values <0.05 are in bold.

PRS, polygenic risk score.

The “all SNPs” menarche PRS was at least nominally associated with age at menarche in White (beta=0.40, p=6.56e-14), Black (beta=0.22, p=0.0375) and Japanese (beta=0.28, p=0.039) women. An increase of one standard deviation of PRS was associated with 0.22–0.40 year (2.6 to 4.8 month) delay in menarche. The age at menarche PRS was not significantly associated with age at menopause, except for a nominal association in Japanese women (beta=0.56, p=0.025). All observed associations in White women were the only ones to remain significant after Bonferroni correction.

When reproductive lifespan was examined as the outcome of interest, both the menarche PRS (beta=−0.49, p=0.00082) and the menopause PRS (beta=0.80, p=5.91e-08) were significantly associated with shorter and longer reproductive lifespans, respectively, in White women, even after Bonferroni correction. In addition, the menopause PRS was nominally positively associated with reproductive lifespan in Chinese (beta=0.62, p=0.028) and Japanese women (beta=0.65, p=0.026). In sensitivity analysis (see Supplemental Digital Content 3), after adjusting for age at menopause, the menopause PRS (beta=0.20, p=0.005) and the menarche PRS (beta=−0.40, p=1.44e-08) remained nominally or significantly associated with reproductive lifespan in White women. However, as expected due to the relatively high correlation between age at menopause and reproductive lifespan, the strength of association between the menopause PRS and reproductive lifespan was partially attenuated. When adjusting the reproductive lifespan analysis for age at menarche, the menopause PRS remained significantly associated in White women (beta=0.67, p=9.76e-07), but the menarche PRS association in White women was completely attenuated.

The length of menopausal transition was also significantly associated with the menopause PRS (beta=0.59, p=0.00014) in White women, but not in the other race/ethnic groups. An increase in one standard deviation of PRS was associated with 0.59 years (7.1months) longer transition. After adjusting for age at menopause, the association was reduced but still nominally associated (beta=0.28, p=0.041; see Supplemental Digital Content 3).

As a secondary analysis, we explored the same associations as above using the “top SNPs” PRSs (see Supplemental Digital Content 4). Associations observed in White women in the “all SNPs” analyses remained significant in the “top SNPs” analyses.

To assess how much these variants, taken from studies in European ancestry women, together explain variation in age of menopause and age of menarche, we also calculated the percent variance explained by the corresponding PRSs for the two traits in White women in SWAN. The “all SNPs” and “top SNPs” menopause PRS explained about 4% and 8% of total variance in age of menopause. The “all SNPs” and “top SNPs” menarche PRS explained about 7% and 5% of variance in age of menarche.

The association between the menopause polygenic risk scores and E2 and FSH levels

The associations between the menopause PRSs and E2 and FSH levels are presented in Table 3. We observed a positive association between the “all SNPs” menopause PRS and E2, measured both prior to and after menopause, only in Japanese women (E2 prior to menopause: beta=0.338, p=0.002; E2 after menopause: beta=0.119, p=0.032). We also observed a negative association between the “all SNPs” menopause PRS and FSH, measured prior to menopause, in Japanese women (beta=−0.185, p=0.04). The association between menopause PRS and E2 prior to menopause in Japanese women was significant after Bonferroni correction while the other associations were nominal. It is possible,however, that the significant association was due to chance given the small sample size in Japanese women and the lack of any association in the larger sample of White women. In secondary analyses with “top SNPs” PRS, the PRS was only nominally associated with E2 prior to menopause in Japanese women (beta=0.324, p=0.008, see Supplemental Digital Content 5).

Table 3:

Association between the “all SNPs” polygenic risk score (PRS) for age at menopause and hormone levels, SWAN Genomic Study

Outcome	Race/Ethnicity	N	Menopause PRSc
			Betad	P value
E2 prior to menopause (pg/ml)a	White	393	0.023	0.669
	Black	193	0.022	0.740
	Chinese	96	0.030	0.797
	Japanese	89	0.338	0.002
E2 after menopause (pg/ml)b	White	390	0.052	0.103
	Black	199	0.019	0.656
	Chinese	98	−0.047	0.521
	Japanese	83	0.119	0.032
FSH prior to menopause (IU/L)a	White	393	−0.031	0.432
	Black	193	−0.026	0.692
	Chinese	96	−0.067	0.482
	Japanese	89	−0.185	0.040
FSH after menopause (IU/L)b	White	390	−0.027	0.205
	Black	199	−0.061	0.187
	Chinese	99	−0.027	0.540
	Japanese	83	−0.047	0.266

^aHormone level = menopause PRS + indicator (measurement taken within cycle or not) + BMI + education + smoking + oral contraceptive use + study site (not applicable for Chinese or Japanese) + 5 genetic PCs

^bHormone level = menopause PRS + BMI + education + smoking + oral contraceptive use + study site (not applicable for Chinese or Japanese) + 5 genetic PCs

Hormone was log transferred prior to analyses via log (hormone+1)

^cMenopause PRS was constructed using all available independent SNPs that were significant at p<10⁻⁴ in GWAS meta-analysis³⁸. Independent SNPs were selected using a clumping approach (LD r² threshold = 0.5, distance from index SNP = 250kb).

^dBeta represents the change in the outcome for the participants who had a PRS that is one standard deviation above the mean compared to the participants who had a mean PRS. P-values <0.05 are in bold.

PRS, polygenic risk score; E2, estradiol; FSH, follicle stimulating hormone.

We conducted sensitivity analysis by adjusting for a variant in the FSHB gene that is known to be associated with FSH levels, rs11031005 (see Supplemental Digital Content 6). As expected, the C allele of rs11031005 was associated with decreased FSH level after menopause (beta=−0.11, p=0.0052) in White women; however, it was not associated with FSH level prior to menopause. The association between the menopause PRS and FSH stayed substantively similar for all race/ethnic groups after adjusting for rs11031005.

DISCUSSION

This study enhances scientific understanding of menopause phenotypes. By investigating whether genetic loci associated with reproductive timing may be associated with additional menopausal traits, it identified a potentially shared genetic basis among multiple sentinel indicators of reproductive aging. It is among the first to evaluate polygenic risk scores for age at menarche and at menopause, based on the most recent GWAS meta-analyses for age at menopause¹⁷ and age at menarche,⁴⁰ and their association with multiple sentinel traits of reproductive aging in different race/ethnic groups. Notably, we found that both the menarche and the menopause PRSs were associated with additional traits including length of the reproductive lifespan, timing of reproductive aging, and E2 levels before menopause.

Although the menarche GWAS and the menopause GWAS included only European ancestry populations, we found that a higher PRS for menarche was nominally associated with later menarche in White, Black and Japanese women and a higher PRS for menopause was nominally associated with later menopause in White, Chinese and Japanese women, suggesting a shared genetic basis across different populations. This conclusion is supported by results of a recent GWAS of menarche and menopause in about 67,000 Japanese women which found shared biologic pathways with European ancestry GWAS.⁴² It is also consistent with findings from a trans-ethnic meta-analysis of menarche and menopause risk loci in African, Hispanic/Latina, Asian American, and American Indian/Alaska Native women.⁵² The Population Architecture using Genomics and Epidemiology (PAGE) Study found that some, but not all, of the genomic regions identified in European ancestry GWAS harbored SNPs associated with age at menopause/menarche in non-European ancestry groups as well, but that the specific SNPs most strongly associated with the traits were often different across groups.

The magnitude of the association of the menopause PRS, approximately 7–8 months for one standard deviation, is comparable to the magnitude of being a non-smoker, having higher socio-economic status and having ever used oral contraceptives.² The lack of association between menopause PRS and age of menarche or menarche PRS with age of menopause suggest that the two traits share limited genetic risk factors, consistent with prior literature.^53–55 Also a higher menarche PRS and higher menopause PRS were associated with shorter and longer reproductive lifespans, respectively, although findings were significant only in White women. Additionally, when we examined the duration of the menopausal transition, we found that a higher menopause PRS was associated with a longer menopausal transition in White women.

The associations detected in this study were most robust in White women, probably at least in part because the PRS was constructed from GWAS conducted in exclusively European ancestry participants. Many studies have shown that predictive accuracy of polygenic scores is substantially reduced when the GWAS and target samples are genetically divergent.^56–58 The PAGE trans-ethnic GWAS also found additional genomic regions associated with age at menopause or menarche that were not identified in European ancestry groups alone.⁵² Given these findings, it is unlikely that the menopause or menarche PRSs in this study have captured all of the key risk variants in Black, Chinese or Japanese women. However, the nominal associations in these non-European racial/ethnic groups suggest that the PRS probably has captured at least some of the risk variants. Future large scale GWASs that focus on non-European or multi-ethnic GWAS are needed to better understand the genetic risk profiles of reproductive aging in those populations.

This is the first study, to our knowledge, that evaluates PRS associations with hormonal parameters of reproductive aging. Menopause is characterized by a transition to high levels of FSH and low levels of E2.⁵⁹ During the menopausal transition, FSH rises and E2 levels decline. Given that hormonal changes are one of the most noticeable biological feature of menopause, it is imperative to understand whether genetic factors that influence timing of menopause also are related to levels of these hormone levels before and after menopause. Higher menopause PRSs was significantly associated with higher E2 levels prior to menopause in Japanese women. This PRS was nominally associated with higher E2 levels after menopause in these same women. To date, a handful of SNPs have been found to be significantly associated with circulating levels of E2. In particular, CYP19A1 shows the strongest evidence of association with E2 in both men and women in exclusively European ancestry and in trans-ethnic studies.^37,60,61 A recent GWAS of 2,913 individuals from the Twins UK study, which included about 11% perimenopausal and 52% postmenopausal women, reported an intronic SNP in ANO2, rs117585797, associated with E2 levels.³⁵ However, neither of the loci are included in the menopause PRS evaluated here, because none of the SNPs in those loci passed the p-value threshold (p<1e-04) in the menopause GWAS¹⁷ to be included in the PRS. Therefore, the potential association between the menopause PRS with E2 was not due to these known risk loci.

With respect to FSH, we did not observe evidence of association between menopause PRS and either FSH measures, except for a nominal association with FSH measure prior to menopause in Japanese women. However in recent literature, FSHB has emerged as a key locus associated with both FSH level and timing of menopause.^36,38,50,51 The Twins UK GWAS reported that the C allele of an FSHB intergenic SNP, rs11031005, was associated with lower circulating levels of FSH.³⁵ Interestingly, the same locus was among the top genome wide significant loci associated with delayed menopause from the GWAS we used to construct the menopause PRS (p-value=9.0e-14).¹⁷ In a sensitivity analysis, as expected, we found that rs11031005 in FSHB was nominally associated with FSH levels after menopause in the expected direction in White women (p=0.0052). However, further conditional analysis adjusting for this SNP did not substantially change the relationship between the menopause PRS and FSH. All of this suggests that menopause and FSH share limited genetic factors other than FSHB locus.

Given that FSH and E2 fluctuate during the menopausal transition and across the menstrual cycle, one annual hormone assessment may not measure between-woman differences precisely, which may lead to weakened associations of PRSs with hormone levels, particularly prior to menopause. This is supported by the observation that the known FSH risk variant was only associated with FSH after menopause. Prior to menopause, the changing patterns of hormones or their variance might be biologically more relevant than absolute levels.^62,63 Larger scale genetic studies with multiple hormone measurements across the menopausal transition are needed to better understand their underlying genetic architecture.

In this study, the “all SNPs” PRS was constructed with the hypothesis that it may capture relevant pathways or mechanisms involved in reproductive aging that are represented by SNPs that did not reach genome-wide significance in the original GWAS. However, the predictive performance of the “all SNPs” PRS were similar to that of the “top SNPs” PRS. The percent variance explained by “top SNPs” and “all SNPs” PRSs ranges between 4–8% for age at menopause and 5–7% for age at menarche in White women, which was comparable to the variance explained by the top SNPs in GWAS studies (6% for menopause and 7% for menarche).^17,40 This suggests that the SNPs with a p-value falling between 5.0e-08 and 1.0e-04 from the menopause or menarche GWAS may not have a strong genetic contribution to the corresponding phenotype compared to the genome-wide significant SNPs, at least in this sample.

This study has some limitations. First, Black women were more likely than women of other race/ethnicities to have had a hysterectomy at the time of enrollment making them ineligible to participate in the SWAN cohort, which may have led to differential patterns of selection bias. As noted above, Black women were also somewhat less likely to participate in the SWAN Genetic Study, but Genetic Study participants did not differ from non-participants by age at menarche or age at menopause, or by education, smoking status, BMI or contraceptive use. Also, the small sample size, particularly for the Chinese and Japanese women, limited study power. Finally, because to date SWAN is the largest cohort to have followed women through the menopausal transition and one of the only to have obtained genome-wide genotype data to date, we were unable to replicate our findings in external cohorts. The study also has several strengths including the multi-ethnic population, the prospectively collected data on traits associated with reproductive aging ensuring more precise and accurate assessment of menopause timing, and the availability of serial hormone measurements enabling assessment of hormone levels before and after the FMP.

CONCLUSION

In conclusion, our results highlight a shared genetic basis among multiple facets of reproductive aging across four race/ethnic populations, including age at menopause, length of the menopausal transition and the reproductive lifespan, and hormone levels before and after the FMP. This paper elucidates genetic loci potentially common across reproductive traits as well as genetic loci that likely differ across reproductive traits, expanding scientific understanding of the interlinkages between sentinel traits of the reproductive life span. Notably, the menarche PRS does not appear to be informative about age at menopause, consistent with the lack of correlation between ages at menarche and menopause, but both the menarche and the menopause PRSs appear to be informative about length of the reproductive life span, which itself has been associated with risk of chronic disease.^{13, 15, 18,19} Larger scale genetic studies with multiple hormone measurements across the menopausal transition are needed, as are studies that more explicitly consider the shared genetic basis of complex biological events such as reproductive aging.

Supplementary Material

Supplemental Data File (.doc, .tif, pdf, etc.)_1

Supplemental Digital Content 1: Table S1: Correlation among the outcome variables in the Study of Women’s Health Across the Nation (SWAN) Genomic Study sample DOCX

Supplemental Data File (.doc, .tif, pdf, etc.)_2

Supplemental Digital Content 2: Table S2: Correlation among the polygenic risk scores (PRS) in the Study of Women’s Health Across the Nation (SWAN) Genomic Study sample DOCX

Supplemental Data File (.doc, .tif, pdf, etc.)_3

Supplemental Digital Content 3: Table S3: Association between “all SNPs” polygenic risk scores (PRS) and reproductive timing after adjusting for age at menopause or age at menarche, SWAN Genomic Study DOCX

Supplemental Data File (.doc, .tif, pdf, etc.)_4

Supplemental Digital Content 4: Table S4: Association between “top SNPs” polygenic risk scores (PRS) and age at menopause, age at menarche, and reproductive lifespan, SWAN Genomic Study DOCX

Supplemental Data File (.doc, .tif, pdf, etc.)_5

Supplemental Digital Content 5: Table S5: Association between the “top SNPs” polygenic risk score (PRS) for age at menopause and hormone levels, SWAN Genomic Study DOCX

Supplemental Data File (.doc, .tif, pdf, etc.)_6

Supplemental Digital Content 6: Table S6: Association between the “all SNPs” polygenic risk score (PRS) for age at menopause and follicular stimulating hormone (FSH) adjusting for rs11031005, SWAN Genomic Study DOCX