https://orcid.org/0000-0003-2450-760X
Figures
Abstract
Younger age at menarche (AAM) is associated with higher body mass index (BMI) for young women. Considering that continuous trends in decreasing AAM and increasing BMI are found in many countries, we attempted to assess whether the observed negative association between AAM and young adult BMI is causal. We included 4,093 women from the Korean Genome and Epidemiology Study (KoGES) and Healthy twin Study (HTS) with relevant epidemiologic data and genome-wide marker information. To mitigate the remarkable differences in AAM across generations, we converted the AAM to a generation-standardized AAM (gsAAM). To test causality, we applied the Mendelian randomization (MR) approach, using a genetic risk score (GRS) based on 14 AAM-associated single nucleotide polymorphisms (SNPs). We constructed MR models adjusting for education level and validated the results using the inverse-variance weighted (IVW), weighted median (WM), MR-pleiotropy residual sum and outliers test (MR-PRESSO), and MR-Egger regression methods. We found a null association using observed AAM and BMI level (conventional regression; -0.05 [95% CIs -0.10–0.00] per 1-year higher AAM). This null association was replicated when gsAAM was applied instead of AAM. Using the two-stage least squares (2SLS) approach employing a univariate GRS, the association was also negated for both AAM and gsAAM, regardless of model specifications. All the MR diagnostics suggested statistically insignificant associations, but weakly negative trends, without evidence of confounding from pleiotropy. We did not observe a causal association between AAM and young adult BMI whether we considered the birth cohort effect or not. Our study alone does not exclude the possibility of existing a weak negative association, considering the modest power of our study design.
Citation: Kim H, Choe S-A, Lee SJ, Sung J (2021) Causal relationship between the timing of menarche and young adult body mass index with consideration to a trend of consistently decreasing age at menarche. PLoS ONE 16(2): e0247757. https://doi.org/10.1371/journal.pone.0247757
Editor: Frank T. Spradley, University of Mississippi Medical Center, UNITED STATES
Received: May 8, 2020; Accepted: February 12, 2021; Published: February 26, 2021
Copyright: © 2021 Kim et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Data cannot be shared publicly because of data distribution policy of the National Biobank of Korea (http://koreabiobank.re.kr/). Data are available to researchers who meet the criteria for access. This study used the epidemiological and genomic data sets called "Combined baseline data of population-based studies (Ansan and Ansung study, HEXA study, CAVAS)" and "KoGES_twin and family study" (KoGES). The variables used in this study are from the "Core questionnaire."
Funding: This work was supported by the National Research Foundation of Korea grant (NRF-2016R1D1A1B03933410, 2018R1D1A1B07048821, 2018R1A6A3A01012922 and 2020R1C1C1013980) funded by the Korean Government. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
There is good evidence to consider menarche as important for understanding young adulthood obesity [1, 2]. Menarche is a landmark in the female reproductive timespan. It is often preceded by a weight growth spurt during adolescence. A decline in the age at menarche (AAM) has been globally observed for the past 50 years [3–5] coinciding with the obesity epidemic. Girls that experienced early menarche have been observed more prone to overweight/obesity [6]. Previous studies have shown that later spontaneous recovery of obesity established in early life is not common [1, 7]. It is worth noting that, even in the context of dramatic increase in childhood obesity since the 1990s [8], most of obese individuals reported their onset of obesity as age of 17‒18 [1]. Such evidence elicits that young adulthood (18–25 years old) obesity, rather than childhood obesity, substantially influences obesity in later life. It also aligns with the hypothesis that menarche as the marked event of female puberty which occurs at an average age of 12 years, may cause obesity in young adulthood for females.
Mendelian randomization (MR) is an established epidemiological method to infer a causal relationship between risk factors and health outcomes [9], by utilizing a genetic instrumental variable (IV) that is robustly associated with the risk factor. Owing to the random independent assortment of alleles during meiosis, genetic variants with biological effects on risk factors are free from possible confounding factors. Moreover because the transmitted germline genome cannot be affected by health outcomes, it should not be biased by reverse causality [9].
Given that the tendency toward a decrease in AAM is still in progress, identifying the causal relationship between AAM and young adult obesity is crucial for obesity control [10, 11]. This study aimed to assess the causal relationship between AAM and young adult body mass index (BMI). In evaluating the causal association between AAM and young adult obesity, birth cohort effects matter not only for the AAM, but for social determinants such as the length of education. Thus, using these covariates without considerations of birth cohort effects may result in biased estimations. For example, the average AAM decreased from 16.9 years for women born between 1920–1925 and to 13.8 years for those born between 1980–1985 [3]. The proportion of women with higher than upper-secondary education (corresponding to high school in most countries) increased from < 5% for women born in 1926, to 95% for those born in 1970 in South Korea [12]. Given the inverse association between education level and the risk of young adulthood obesity [13, 14], this transitional change in education level may confound the association between AAM and young adult BMI. To address the problem of complex confounding structures, we attempted to introduce a generation-standardized estimation for both AAM and the length of education, using a transgenerational approach. In this study, we estimated the unconfounded causal association of AAM with young adult BMI through the MR approach, using generation-standardized measures.
Methods
Study population and variables
The Korean Genome and Epidemiology study (KoGES, http://www.nih.go.kr/NIH/eng/main.jsp) is a community-based prospective cohort study started in 2001, and comprised of 3 sub-cohorts (rural, urban, Ansan-Ansung (KARE)) differing in residential areas of the participants [15]. The Healthy Twin Study (HTS) is a nationwide twin-family cohort study which started in 2005 [16]. Both studies recruited participants of Korean ancestry based on a shared protocol to allow pooled analysis. Detailed study protocols and information have been previously described. From both studies, we only included female participants with epidemiological and genotype information available.
We defined AAM as age in years at the onset of menstruation. Year of birth, age at menarche, and the highest educational attainment, were obtained from self-reported questionnaires. The highest level of educational attainment was categorized into 5 levels: “under elementary school”, “elementary school”, “middle school”, “high school”, and “university/college or higher”. Height measured at enrollment and self-reported body weight at 18–20 years were used to calculate the young adulthood BMI (kg/m2). We restricted our analysis to a subset of participants with information on young adulthood BMI. The differences in study variables between the overall population (N = 10,000) and the subset with young adulthood BMI available (N = 4,903) were minimal (S1 Table). The protocol of this study is approved by the Institutional Review Board of Gangnam CHA Hospital (IRB NO: GCI-17-37).
Defining birth cohorts
We constructed birth cohorts to reflect changes in the distribution of AAM, which has a similar AAM distribution within the year of birth. For example, we started from the birth year of 1927, where the AAM under 14 years was 5.08%, and expanded the birth cohort to include the next recent birth years to as far as the cumulative proportion of AAM under 14 years of the birth cohorts was maintained between 5 and 6%. The end point of the birth cohort was selected when EM at the K-birth year showed more than a 2% difference from that of the (K+1) birth year. These steps were repeated until the last birth year resulted in four birth cohorts (1927–1945, 1946–1969, 1970–1978, and 1979–2003) (S2 Table).
Calculating generation-standardized age at menarche (gsAAM) and education
To control for the effect of different birth cohorts, we developed a generation-standardized measure of AAM (gsAAM) and education level. We calculated the z-score of AAM taking individual AAM minus average AAM of the birth year, divided by the standard deviation, where both the averages and standard deviations of AAM were derived from a large population data from the Korea National Health and Nutrition Examination Survey (KNHANES, https://knhanes.cdc.go.kr/knhanes), 2001–2017.
We created an indicator for the relative level of education (“highly educated”) as a proxy for socioeconomic status, to allow us to consider the generation gap in educational attainment (S3 Table). The relative definition of those highly educated within each cohort was varied across birth cohorts: elementary school or higher for 1927–1945, high school or higher for 1946–1969, university/college or higher for both 1970–1978 and 1979–2003 (S1 Fig).
Selecting genetic markers representing younger AAM
The KoGES and HTS participants were genotyped using Illumina Omni1 (KoGES, rural), Affymatrix 6.0 (KoGES, urban; HTS) and 5.0 (KoGES, KARE) genotyping arrays. The single nucleotide polymorphisms (SNPs) were filtered by following genotype quality control (QC) criteria: (1) genotyping call rate >0.95 (2) minor allele frequency (MAF) >0.01, (3) P value in Hardy–Weinberg equilibrium (HWE) testing >10e-6 and individual QC criteria: sample call rate >0.9. Those after QC were imputed using the Korean data from Korean Reference Genome (KRG) which initiated by Center of Genome Science (CGS) of Korea National Instituted of Health (KNIH) in 2012 [17] and East Asian (EAS) data from the 1,000 Genomes Project Phase 3 (NCBI build 37) as the reference. We used IMPUTE 2.0 software to impute variants that were not directly genotyped [18]. Only SNPs with INFO score larger than 0.6 were included.
Among 367 SNPs reported to be associated with AAM from a previous study [19], 297 were available in our genotype data, and 15 SNPs were replicated at a significance level of 0.05 with directional consistency of association. One SNP, rs7132908, located at FAIM2, was excluded because it showed a pleiotropic effect with adulthood BMI [20]. Finally, 14 SNPs were selected to build the GRS representing earlier menarche: rs157877 (RXRG), rs643428 (SSBP3), rs142058842 (NR4A2), rs4588499 (GABRG1), rs3113862 (SMARCAD1), rs1428120 (GALNT10), rs13233916 (TTC26), rs7115444 (C11orf67), rs4945266 (GAB2), rs4402316 (DLG2), rs7114175 (C11orf63), rs3764002 (WSCD2), rs10143972 (UNC79), rs12915845 (DET1) (S4 Table). Childhood BMI, a potential confounder of the effect of AAM on young adulthood BMI, did not show any association at statistical significance (P < 1.0 x 10−5) with the selected 14 SNPs according to the GWAS catalogue [21].
Statistical analysis
Validating MR assumptions.
We tested whether each SNP included in the GRS and GRS as a whole satisfied the assumptions of the MR method [22]. The underlying assumptions of valid IVs are: 1) the IV has a significant association with the exposure; 2) the IV is not related to any other confounders; and 3) the IV is only related to the outcome through the exposure. F-statistics and R2 values were used to ensure that the genetic marker was strongly associated with the risk exposure. Generally, an F-statistic larger than 10 is quoted to reduce the weak instrument bias [23, 24]. The association between risk exposure and GRS quartiles was also examined for convenience of interpretation. To test the second IV assumption of independent association between the genetic IV and young adulthood BMI, the frequencies of potential confounders per risk allele-increase were examined.
Conventional regressions and genetic IV-based analyses.
For conventional epidemiological analyses, linear regressions were performed to examine the relationship between AAM and young adulthood BMI observationally. We constructed 4 different regression models with varied groupings of birth cohort and education as follows:
Model 1: BMIyoung adult = AAMyear + Year of birth + Educational level + ℇ
Model 2: BMIyoung adult = AAMyear + Birth cohort (defined by AAM distribution) + Educational level (generation-adjusted) + ℇ
Model 3: BMIyoung adult = gsAAM+ Educational level + ℇ
Model 4: BMIyoung adult = gsAAM + Educational level (generation-adjusted) + ℇ
Birth cohort variables were treated as dummy variables in the regression models. This analysis was performed using R (The R Foundation, version 3.3.1). We investigated the causal effect of AAM on young adulthood BMI using the two-stage least squares (2SLS) regression, under an additive model assumption through the MR approach [25]. GRS was categorized into quartiles in the first regression of 2SLS for convenience of interpretation. We applied the same four models as in the conventional analyses. The standard error of the 2SLS regressions was obtained by bootstrapping (1000 replications). All MR analyses were performed using STATA version 14 (StataCorp LLC, College Station, Texas).
Implementing the MR analysis without considering pleiotropy effects could lead to bias. To confirm that the IV shows a consistent association with the outcome without the condition of exposure (horizontal pleiotropy), the adjusted MR method using summary data estimation: weighted median(WM) and MR-Egger method using the R package ‘Mendelian Randomization’ [26] and MR-PRSSO using the R package ‘MR-PRSSO’ were employed [27]. The inverse variance weighted (IVW) method calculates the weighted mean using the inverse variance of each SNP to minimize the variance effect [28]. IVW theoretically provides asymptotically equal results from the 2SLS regression when each SNP is fully uncorrelated. The results from IVW are used as a reference compared to other adjusted MR methods. The WM method is a robust estimator that provides consistent estimates even if up to 50% of IVs are invalid [29]. The MR-Egger method is based on Egger’s regression, which is used for adjusting publication bias in meta-analyses [30]. The difference in the intercept term from the origin gives evidence of the average pleiotropic effect in IV used in MR. The coefficient from the MR-Egger regression was used to explore whether the causal effect remains after adjusting for the pleiotropic effect. Scatter plots of summary data estimation were used to graphically interpret the validity of the slope estimates and the intercept of the MR-Egger regression. Cochran’s Q statistics indicating heterogeneity of SNPs [31] were estimated.
To further investigate the influence of AAM on BMI-related traits, we performed additional analyses for non-alcoholic fatty liver disease (NAFLD) and Homeostatic Model Assessment for Insulin Resistance (HOMA-IR). Using the MR-Base (http://www.mrbase.org), a public database and analytic platform of multi-instrument MR [32], the causal association of AAM with NAFLD and HOMA-IR were investigated. Dataset for exposure AAM was selected from GWAS-catalogue [21, 33] and those for NAFLD and HOMA-IR were from prior genome-wide association studies [34–44]. Linkage disequilibrium (LD) clumping method was applied in the process of selecting SNPs to be included. As in our primary analysis, IVW, WM and MR-Egger method was used in two-sample MR analysis.
Results
In the three Korean cohort studies (KHANES, KoGES, HTS), the proportion of those who reported AAM < 12 was 5–8% in the 1970s group and was 40% in the 1990s group. Mean AAM decreased from 16.6 in women born in 1929–1945 to 12.4 in those born in 1979–1994 (Table 1). We tested the association of educational attainment with young adulthood BMI and AAM (S5 Table and S3 Fig). Overall, AAM decreased with higher educational attainment except the youngest generations (born in 1979―2003). Young adult BMI was lower with higher education attainment groups. Young adulthood BMI was generally higher in the older generations (P for trend < 0.01) (S4 Fig).
Assessment of the MR assumptions
We tested GRS to ensure that a higher GRS represents a younger AAM and thus functions as an IV (Table 2). Both AAM in years and gsAAM changed by -0.08 and -0.04 as GRS increased by one unit (both P <0.01). The F-statistics of GRS on AAM was 57.52, indicating that this can be strong IV for younger AAM. The associations between GRS and potential confounders were close to null (S6 Table).
Association of AAM and young adulthood BMI
In conventional analyses, higher AAM in year was associated with lower young adulthood BMI (coefficient: -0.05, 95% confidence intervals [CI]: -0.10–0.00; P = 0.04; Table 3). This association was not consistently observed in other models, including those that applied the gsAAM, indicating that the results of conventional analysis could be confounded by how adjustments were selected. In the 2SLS analysis, the associations of both AAM and gsAAM with young adulthood BMI were not demonstrated. However, the results showed consistent trend of negative estimates, in all types of model specifications (Fig 1).
Observational analysis, Obs; Mendelian randomization analysis, MR.
Using the adjusted MR methods (WM, MR-Egger, and MR-PRSSO), we observed larger negative estimates notwithstanding without statistical significances than that from observational studies (Table 4). The intercept estimated from the MR-Egger regression model for AAM (intercept = -0.00, 95% CI: -0.18, 0.18) and gsAAM (intercept = 0.00, 95% CI: -0.18, 0.18) showed no difference from the origin, suggesting that the average pleiotropic effect of SNPs used did not influence the results of the MR analysis. The scatter plot of genetic association with risk factors and outcomes indicated no bias from pleiotropy (S5 Fig). As the Cochran’s Q statistics estimated with 14 SNPs used for GRS indicated heterogeneity between SNPs, we performed the same regression after excluding rs1428120, which showed heterogeneity with other SNPs (S7 Table). The adjusted MR result without rs1428120 was consistent with the result showing heterogeneity.
We observed a null association between AAM and NAFLD/HOMA-IR which is consistent with the MR result for adult BMI (S8 Table).
Discussion
Our study suggests that younger AAM may not be causally associated with BMI in young adulthood. The weak but consistently negative association between the AAM and young adulthood BMI in various MR sensitivity analyses, however, also might suggest a negative association, given the relatively low study power. To reflect the evident secular changes in AAM and educational attainment, we redefined birth cohorts, classifying the participants so that the AAM distributions were similar across birth cohorts. The inverse relationship (earlier menarche associated with higher BMI in young adulthood) reported in observational studies may have been confounded by the secular change of AAM. This notion is supported by the result of 2SLS analysis. Interpretations of the MR analysis requires caution because all the MR models and their adjusted diagnostic models suggest weak but constantly negative associations between AAM and young adulthood BMI. Considering the limited power of the study, we do not conclude that a negative association exist between AAM and young adulthood BMI. Overall, our study suggests a null association, but ours does not exclude the possibility that AAM is one of the determinants of young adulthood obesity, but at least it is not likely that AAM be a major determinant even if so. While the relationships between early puberty and childhood and/or post-pubertal obesity have been investigated, these have not been thoroughly explicated to explain how birth cohort effects may influence the association. To the best of our knowledge, this study is the first to consider generational differences in AAM in exploring the causal association between AAM and young adulthood BMI.
The NAFLD and HOMA-IR are strongly associated with BMI [45, 46]. Also, several observational studies reported that earlier menarche is associated with a higher risk of NAFLD and insulin resistance, independent of young-adult BMI [47, 48]. We tried to explore the causal association of AAM and NAFLD/HOMA-IR in public database. The null association between AAM and NAFLD/HOMA-IR was observed which indicates that AAM does not affect the incident of NAFLD/high HOMA-IR even through obesity.
According to Gill et al. [49], earlier AAM was causally associated with higher adult BMI, even pleiotropic SNPs that were also associated with childhood BMI were excluded. Yet, in another MR study, the inverse association of AAM with BMI at age 18 years was negated when adjusted for childhood BMI [50]. There is another evidence that childhood BMI roles as a confounder of effects of AAM on adulthood BMI [51, 52]. In the 1950s cohort study, the inverse association of AAM with adulthood obesity was not explained by the confounding effect of early childhood BMI [53] and was not replicated in the Australian cohort [54]. The effect of childhood BMI on AAM is somewhat evident but it is still unclear that earlier AAM is one of the cause of young adulthood obesity [55, 56]. The inconsistency between studies may be attributed to the different population distributions in adulthood BMI and AAM.
Given that the trend of AAM in Korea over generations are more dramatically decreased compared to that in western countries and similar patterns are observed worldwide in generational prevalence of childhood obesity [57–59], generational trend of AAM in Korea is not thought to be only caused by increasing childhood obesity. The gap between the trend of AAM and childhood obesity make it necessary to consider the generation effect in the study of causal association between AAM and young adulthood BMI.
The impact of demographic changes over time at a population level has not been explored in previous studies [10, 60]. In the setting of rapid socioeconomic changes, such as in Korea, the historical association between young adulthood obesity and early menarche would not have been evident. We postulated that old Korean women who survived the Korean War had experienced a delayed onset of menarche due to the stressful circumstances, despite most having sufficient body weight to initiate the menstrual cycle [61]. In contrast, in the younger generations, social pressure to maintain a low or normal body weight may have contributed to an overall lower average young adulthood BMI, compared to older women [62, 63]. Individual socioeconomic factors may also confound the association between the timing of menarche. Moreover, SES factors have changed across generations, which further necessitates deciphering the generation effect.
There are a few limitations to this study. First, there may be possibility of recall biases. Those who were obese at the time of the survey might have overestimated their weight in childhood, assuming that they maintained a similar body shape in the past. However, a number of previous studies revealed that self-reported past body weights can provide reliable estimates independent of current weight status [20, 64]. Second, a potential pleiotropic effect may have existed between AAM and young adulthood BMI via residual associations from SNPs not included in the study. As genetic architecture of childhood adiposity and AAM is highly overlapped [19], using subset of AAM SNPs which is only associated with AAM not with childhood adiposity as IV does not assure that the confounding effects of childhood obesity is fully removed. Although we examined most of the established genes and the presence of pleiotropy of IV, there could be a possibility of remaining effect by other genes that were not included in the study. However, we believe this effect would be minimal given the high genetic homogeneity of the Korean population [65]. Third, although our MR results showed consistent negative trend, the association was null and with wide CIs. Both limitations in the size and the strength of the association have resulted in the modest study power, so that further studies are required to confirm the association. Lastly, childhood BMI was not available and could not be included in our analysis. Although we believe the potential confounding role of childhood adiposity was minimized with the use of IV, future studies may benefit from including the information of childhood adiposity.
In conclusion, our findings alone do not exclude the possibility that AAM is one of the determinants of young adulthood obesity, but at least it is not likely that AAM be a major determinant even if so.
Supporting information
S1 Fig. Young adulthood BMI and education attainment across year of birth in the Korean Genome and Epidemiology Study (KoGES) and Healthy Twin Study (HTS), (n = 4,093 women).
https://doi.org/10.1371/journal.pone.0247757.s001
(DOCX)
S2 Fig.
Distribution of genetic risk score (AAM) for younger age at menarche (AAM).
https://doi.org/10.1371/journal.pone.0247757.s002
(DOCX)
S3 Fig. Trend of AAM/young adulthood BMI by educational attainment.
https://doi.org/10.1371/journal.pone.0247757.s003
(DOCX)
S4 Fig. Association between age and young-adulthood BMI.
https://doi.org/10.1371/journal.pone.0247757.s004
(DOCX)
S5 Fig. Scatter plot of adjusted MR estimation.
https://doi.org/10.1371/journal.pone.0247757.s005
(DOCX)
S1 Table. Characteristic comparison of full data (N = 10,000) and data with young adulthood BMI available, Korean Genome and Epidemiology Study (KoGES) and Healthy Twin Study (HTS) (N = 4,093).
https://doi.org/10.1371/journal.pone.0247757.s006
(DOCX)
S2 Table. Definition of birth cohorts based on the distribution of age at menarche (AAM) from the data of Korea National Health and Nutrition Examination Survey (KNHANES), Korean Genome and Epidemiology Study (KoGES) and Healthy Twin Study (HTS) (N = 169,571).
https://doi.org/10.1371/journal.pone.0247757.s007
(DOCX)
S3 Table. Distribution of highest educational attainment by predefined birth cohort from the data of Korean Genome and Epidemiology Study (KoGES) and Healthy Twin Study (HTS) (N = 10,000).
https://doi.org/10.1371/journal.pone.0247757.s008
(DOCX)
S4 Table. SNPs replicated with AAM in the Korean Genome and Epidemiology Study (KoGES) and Healthy Twin Study (HTS), (n = 118,569 women).
https://doi.org/10.1371/journal.pone.0247757.s009
(DOCX)
S5 Table. The association of educational attainment and AAM/young-adulthood BMI.
https://doi.org/10.1371/journal.pone.0247757.s010
(DOCX)
S6 Table. Association of potential confounders with instrumental variable (IV) in the Korean Genome and Epidemiology Study (KoGES) and Healthy Twin Study (HTS), (n = 10,000).
https://doi.org/10.1371/journal.pone.0247757.s011
(DOCX)
S7 Table. Result of two-sample summary MR method for exploring the association between age at menarche (AAM) and young adulthood body mass index without rs1428120.
https://doi.org/10.1371/journal.pone.0247757.s012
(DOCX)
S8 Table. Result of MR-base for exploring the association between age at menarche (AAM) and NAFLD/HOMA-IR.
https://doi.org/10.1371/journal.pone.0247757.s013
(DOCX)
Acknowledgments
We used bioresources from the National Biobank of Korea, the Centers for Disease Control and Prevention, Republic of Korea (Approval No. 2017–04).
References
- 1. McTigue KM, Garrett JM, Popkin BM. The natural history of the development of obesity in a cohort of young U.S. adults between 1981 and 1998. Annals of internal medicine. 2002;136(12):857–64. Epub 2002/06/19. pmid:12069559.
- 2. Braddon FE, Rodgers B, Wadsworth ME, Davies JM. Onset of obesity in a 36 year birth cohort study. British medical journal (Clinical research ed). 1986;293(6542):299–303. Epub 1986/08/02. pmid:3089493; PubMed Central PMCID: PMC1340984.
- 3. Cho GJ, Park HT, Shin JH, Hur JY, Kim YT, Kim SH, et al. Age at menarche in a Korean population: secular trends and influencing factors. Eur J Pediatr. 2010;169(1):89–94. Epub 2009/06/09. pmid:19504269.
- 4. Nichols HB, Trentham-Dietz A, Hampton JM, Titus-Ernstoff L, Egan KM, Willett WC, et al. From menarche to menopause: trends among US Women born from 1912 to 1969. Am J Epidemiol. 2006;164(10):1003–11. Epub 2006/08/25. pmid:16928728.
- 5. Bellis MA, Downing J, Ashton JR. Adults at 12? Trends in puberty and their public health consequences. J Epidemiol Community Health. 2006;60(11):910–1. Epub 2006/10/21. pmid:17053275; PubMed Central PMCID: PMC2465479.
- 6. Bralić I, Tahirović H, Matanić D, Vrdoljak O, Stojanović-Spehar S, Kovacić V, et al. Association of early menarche age and overweight/obesity. J Pediatr Endocrinol Metab. 2012;25(1–2):57–62. Epub 2012/05/11. pmid:22570951.
- 7. Guo SS, Roche AF, Chumlea WC, Gardner JD, Siervogel RM. The predictive value of childhood body mass index values for overweight at age 35 y. The American journal of clinical nutrition. 1994;59(4):810–9. Epub 1994/04/01. pmid:8147324.
- 8. Ebbeling CB, Pawlak DB, Ludwig DS. Childhood obesity: public-health crisis, common sense cure. Lancet (London, England). 2002;360(9331):473–82. Epub 2002/09/21. pmid:12241736.
- 9.
Smith GD, Ebrahim S. Mendelian Randomization: Genetic Variants as Instruments for Strengthening Causal Inference in Observational Studies. Washington DC: National Academies Press (US); 2008.
- 10. Anderson SE, Must A. Interpreting the continued decline in the average age at menarche: results from two nationally representative surveys of U.S. girls studied 10 years apart. The Journal of pediatrics. 2005;147(6):753–60. Epub 2005/12/17. pmid:16356426.
- 11. Yu EJ, Choe S-A, Yun J-W, Son M. Association of early menarche with adolescent health in the setting of rapidly decreasing age at menarche. Journal of Pediatric and Adolescent Gynecology. pmid:31874313
- 12. Yoo SH. Educational differentials in cohort fertility during the fertility transition in South Korea. Demographic Research. 2014;30(53):1463–94.
- 13. Watts AW, Mason SM, Loth K, Larson N, Neumark-Sztainer D. Socioeconomic differences in overweight and weight-related behaviors across adolescence and young adulthood: 10-year longitudinal findings from Project EAT. Prev Med. 2016;87:194–9. Epub 03/10. pmid:26970036.
- 14. Gonzalez A, Boyle MH, Georgiades K, Duncan L, Atkinson LR, MacMillan HL. Childhood and family influences on body mass index in early adulthood: findings from the Ontario Child Health Study. BMC public health. 2012;12:755. Epub 2012/09/11. pmid:22958463; PubMed Central PMCID: PMC3490808.
- 15. Kim Y, Han BG. Cohort Profile: The Korean Genome and Epidemiology Study (KoGES) Consortium. International journal of epidemiology. 2017;46(2):e20. Epub 2016/04/17. pmid:27085081; PubMed Central PMCID: PMC5837648.
- 16. Sung J, Cho SI, Lee K, Ha M, Choi EY, Choi JS, et al. Healthy Twin: a twin-family study of Korea—protocols and current status. Twin research and human genetics: the official journal of the International Society for Twin Studies. 2006;9(6):844–8. Epub 2007/01/27. pmid:17254419.
- 17. Cho YS, Kim H, Kim HM, Jho S, Jun J, Lee YJ, et al. An ethnically relevant consensus Korean reference genome is a step towards personal reference genomes. Nat Commun. 2016;7:13637. Epub 2016/11/25. pmid:27882922; PubMed Central PMCID: PMC5123046.
- 18. Howie BN, Donnelly P, Marchini J. A flexible and accurate genotype imputation method for the next generation of genome-wide association studies. PLoS Genet. 2009;5(6):e1000529. Epub 2009/06/23. pmid:19543373; PubMed Central PMCID: PMC2689936.
- 19. Day FR, Thompson DJ, Helgason H, Chasman DI, Finucane H, Sulem P, et al. Genomic analyses identify hundreds of variants associated with age at menarche and support a role for puberty timing in cancer risk. Nature Genetics. 2017;49(6):834–41. pmid:28436984
- 20. Casey VA, Dwyer JT, Coleman KA, Krall EA, Gardner J, Valadian I. Accuracy of recall by middle-aged participants in a longitudinal study of their body size and indices of maturation earlier in life. Annals of human biology. 1991;18(2):155–66. Epub 1991/03/01. pmid:2024949.
- 21. Buniello A, MacArthur JAL, Cerezo M, Harris LW, Hayhurst J, Malangone C, et al. The NHGRI-EBI GWAS Catalog of published genome-wide association studies, targeted arrays and summary statistics 2019. Nucleic acids research. 2019;47(D1):D1005–d12. Epub 2018/11/18. pmid:30445434; PubMed Central PMCID: PMC6323933.
- 22. Didelez V, Sheehan N. Mendelian randomization as an instrumental variable approach to causal inference. Stat Methods Med Res. 2007;16(4):309–30. Epub 2007/08/24. pmid:17715159.
- 23. Staiger D, Stock JH. Instrumental Variables Regression with Weak Instruments. Econometrica. 1997;65(3):557–86.
- 24. Lawlor DA, Harbord RM, Sterne JA, Timpson N, Davey Smith G. Mendelian randomization: using genes as instruments for making causal inferences in epidemiology. Statistics in medicine. 2008;27(8):1133–63. Epub 2007/09/22. pmid:17886233.
- 25. Burgess S, Small DS, Thompson SG. A review of instrumental variable estimators for Mendelian randomization. Statistical Methods in Medical Research. 2015;26(5):2333–55. pmid:26282889
- 26. Yavorska OO, Burgess S. MendelianRandomization: an R package for performing Mendelian randomization analyses using summarized data. International journal of epidemiology. 2017;46(6):1734–9. pmid:28398548
- 27. Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet. 2018;50(5):693–8. Epub 2018/04/25. pmid:29686387; PubMed Central PMCID: PMC6083837.
- 28. Burgess S, Bowden J. Integrating summarized data from multiple genetic variants in Mendelian randomization: bias and coverage properties of inverse-variance weighted methods. arXiv preprint arXiv:151204486. 2015.
- 29. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genetic epidemiology. 2016;40(4):304–14. Epub 2016/04/12. pmid:27061298; PubMed Central PMCID: PMC4849733.
- 30. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. International journal of epidemiology. 2015;44(2):512–25. PMC4469799. pmid:26050253
- 31. Greco MF, Minelli C, Sheehan NA, Thompson JR. Detecting pleiotropy in Mendelian randomisation studies with summary data and a continuous outcome. Statistics in medicine. 2015;34(21):2926–40. Epub 2015/05/08. pmid:25950993.
- 32. Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D, et al. The MR-Base platform supports systematic causal inference across the human phenome. eLife. 2018;7. Epub 2018/05/31. pmid:29846171; PubMed Central PMCID: PMC5976434.
- 33. Horikoshi M, Day FR, Akiyama M, Hirata M, Kamatani Y, Matsuda K, et al. Elucidating the genetic architecture of reproductive ageing in the Japanese population. Nature communications. 2018;9(1):1977–. pmid:29773799.
- 34. Sun BB, Maranville JC, Peters JE, Stacey D, Staley JR, Blackshaw J, et al. Genomic atlas of the human plasma proteome. Nature. 2018;558(7708):73–9. Epub 2018/06/08. pmid:29875488; PubMed Central PMCID: PMC6697541.
- 35. Astle WJ, Elding H, Jiang T, Allen D, Ruklisa D, Mann AL, et al. The Allelic Landscape of Human Blood Cell Trait Variation and Links to Common Complex Disease. Cell. 2016;167(5):1415–29.e19. Epub 2016/11/20. pmid:27863252; PubMed Central PMCID: PMC5300907.
- 36. Wu Y, Byrne EM, Zheng Z, Kemper KE, Yengo L, Mallett AJ, et al. Genome-wide association study of medication-use and associated disease in the UK Biobank. Nat Commun. 2019;10(1):1891. Epub 2019/04/25. pmid:31015401; PubMed Central PMCID: PMC6478889.
- 37. Lutz SM, Cho MH, Young K, Hersh CP, Castaldi PJ, McDonald ML, et al. A genome-wide association study identifies risk loci for spirometric measures among smokers of European and African ancestry. BMC Genet. 2015;16:138. Epub 2015/12/05. pmid:26634245; PubMed Central PMCID: PMC4668640.
- 38. Wheeler HE, Gamazon ER, Frisina RD, Perez-Cervantes C, El Charif O, Mapes B, et al. Variants in WFS1 and Other Mendelian Deafness Genes Are Associated with Cisplatin-Associated Ototoxicity. Clin Cancer Res. 2017;23(13):3325–33. Epub 2017/01/01. pmid:28039263; PubMed Central PMCID: PMC5493516.
- 39. Consortium WTCC. Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature. 2007;447(7145):661–78. Epub 2007/06/08. pmid:17554300; PubMed Central PMCID: PMC2719288.
- 40. Liu M, Jiang Y, Wedow R, Li Y, Brazel DM, Chen F, et al. Association studies of up to 1.2 million individuals yield new insights into the genetic etiology of tobacco and alcohol use. Nat Genet. 2019;51(2):237–44. Epub 2019/01/16. pmid:30643251; PubMed Central PMCID: PMC6358542.
- 41. Shahin MH, Conrado DJ, Gonzalez D, Gong Y, Lobmeyer MT, Beitelshees AL, et al. Genome-Wide Association Approach Identified Novel Genetic Predictors of Heart Rate Response to β-Blockers. J Am Heart Assoc. 2018;7(5). Epub 2018/02/27. pmid:29478026; PubMed Central PMCID: PMC5866313.
- 42. Lauc G, Huffman JE, Pučić M, Zgaga L, Adamczyk B, Mužinić A, et al. Loci associated with N-glycosylation of human immunoglobulin G show pleiotropy with autoimmune diseases and haematological cancers. PLoS Genet. 2013;9(1):e1003225. Epub 2013/02/06. pmid:23382691; PubMed Central PMCID: PMC3561084.
- 43. Mayerle J, den Hoed CM, Schurmann C, Stolk L, Homuth G, Peters MJ, et al. Identification of genetic loci associated with Helicobacter pylori serologic status. Jama. 2013;309(18):1912–20. Epub 2013/05/09. pmid:23652523.
- 44. Jensen RA, Sim X, Li X, Cotch MF, Ikram MK, Holliday EG, et al. Genome-wide association study of retinopathy in individuals without diabetes. PLoS One. 2013;8(2):e54232. Epub 2013/02/09. pmid:23393555; PubMed Central PMCID: PMC3564946.
- 45. Ahmed ML, Ong KK, Dunger DB. Childhood obesity and the timing of puberty. Trends in endocrinology and metabolism: TEM. 2009;20(5):237–42. Epub 2009/06/23. pmid:19541497.
- 46. Chen G, Liu C, Yao J, Jiang Q, Chen N, Huang H, et al. Overweight, obesity, and their associations with insulin resistance and β-cell function among Chinese: a cross-sectional study in China. Metabolism: clinical and experimental. 2010;59(12):1823–32. Epub 2010/07/27. pmid:20655552.
- 47. Mueller NT, Pereira MA, Demerath EW, Dreyfus JG, MacLehose RF, Carr JJ, et al. Earlier menarche is associated with fatty liver and abdominal ectopic fat in midlife, independent of young adult BMI: The CARDIA study. Obesity (Silver Spring). 2015;23(2):468–74. Epub 12/17. pmid:25521620.
- 48. Mueller NT, Duncan BB, Barreto SM, Chor D, Bessel M, Aquino EML, et al. Earlier age at menarche is associated with higher diabetes risk and cardiometabolic disease risk factors in Brazilian adults: Brazilian Longitudinal Study of Adult Health (ELSA-Brasil). Cardiovasc Diabetol. 2014;13:22–. pmid:24438044.
- 49. Gill D, Brewer CF, Del Greco MF, Sivakumaran P, Bowden J, Sheehan NA, et al. Age at menarche and adult body mass index: a Mendelian randomization study. International journal of obesity (2005). 2018;42(9):1574–81. Epub 2018/03/20. pmid:29549348.
- 50. Bell JA, Carslake D, Wade KH, Richmond RC, Langdon RJ, Vincent EE, et al. Influence of puberty timing on adiposity and cardiometabolic traits: A Mendelian randomisation study. PLoS medicine. 2018;15(8):e1002641. Epub 2018/08/29. pmid:30153260; PubMed Central PMCID: PMC6112630 following competing interests: GDS is a member of the Editorial Board of PLOS Medicine. MVH has collaborated with Boehringer Ingelheim in research, and in accordance with the policy of The Clinical Trial Service Unit and Epidemiological Studies Unit (University of Oxford), did not accept any personal payment. There are no further interests to declare.
- 51. Kivimäki M, Lawlor DA, Smith GD, Elovainio M, Jokela M, Keltikangas-Järvinen L, et al. Association of age at menarche with cardiovascular risk factors, vascular structure, and function in adulthood: the Cardiovascular Risk in Young Finns study. The American journal of clinical nutrition. 2008;87(6):1876–82. Epub 2008/06/11. pmid:18541580.
- 52. Sandhu J, Ben-Shlomo Y, Cole TJ, Holly J, Davey Smith G. The impact of childhood body mass index on timing of puberty, adult stature and obesity: a follow-up study based on adolescent anthropometry recorded at Christ’s Hospital (1936–1964). International journal of obesity (2005). 2006;30(1):14–22. Epub 2005/12/14. pmid:16344844.
- 53. Pierce MB, Leon DA. Age at menarche and adult BMI in the Aberdeen Children of the 1950s Cohort Study. The American journal of clinical nutrition. 2005;82(4):733–9. pmid:16210700
- 54. Le-Ha C, Beilin LJ, Burrows S, Huang R-C, Hickey M, Mori TA, et al. Age at menarche and childhood body mass index as predictors of cardio-metabolic risk in young adulthood: A prospective cohort study. PLOS ONE. 2018;13(12):e0209355. pmid:30576345
- 55. Chen YC, Fan HY, Yang C, Hsieh RH, Pan WH, Lee YL. Assessing causality between childhood adiposity and early puberty: A bidirectional Mendelian randomization and longitudinal study. Metabolism: clinical and experimental. 2019;100:153961. Epub 2019/08/20. pmid:31422054.
- 56. Mumby HS, Elks CE, Li S, Sharp SJ, Khaw KT, Luben RN, et al. Mendelian Randomisation Study of Childhood BMI and Early Menarche. J Obes. 2011;2011:180729. Epub 2011/07/21. pmid:21773002; PubMed Central PMCID: PMC3136158.
- 57. Meng X, Li S, Duan W, Sun Y, Jia C. Secular Trend of Age at Menarche in Chinese Adolescents Born From 1973 to 2004. Pediatrics. 2017;140(2). Epub 2017/07/19. pmid:28716824; PubMed Central PMCID: PMC5527668 conflicts of interest to disclose.
- 58. Kim JH, Moon JS. Secular Trends in Pediatric Overweight and Obesity in Korea. Journal of obesity & metabolic syndrome. 2020;29(1):12–7. Epub 2020/03/20. pmid:32188238; PubMed Central PMCID: PMC7118001.
- 59. Skinner AC, Ravanbakht SN, Skelton JA, Perrin EM, Armstrong SC. Prevalence of Obesity and Severe Obesity in US Children, 1999–2016. Pediatrics. 2018;141(3). Epub 2018/02/28. pmid:29483202; PubMed Central PMCID: PMC6109602.
- 60. Kaplowitz PB. Link between body fat and the timing of puberty. Pediatrics. 2008;121 Suppl 3:S208–17. Epub 2008/02/15. pmid:18245513.
- 61. Choe SA, Sung J. Trends of Premature and Early Menopause: a Comparative Study of the US National Health and Nutrition Examination Survey and the Korea National Health and Nutrition Examination Survey. Journal of Korean medical science. 2020;35(14):e97. Epub 2020/04/14. pmid:32281314; PubMed Central PMCID: PMC7152531.
- 62. Hosseini M, Kelishadi R, Baikpour M, Ataei N, Qorbani M, Yousefifard M, et al. Age-Period-Cohort Analysis of Obesity and Overweight in Iranian Children and Adolescents. International journal of endocrinology and metabolism. 2017;15(4):e13561. Epub 2018/01/19. pmid:29344031; PubMed Central PMCID: PMC5750447.
- 63. Kim Y, Choi S, Chun C, Park S, Khang YH, Oh K. Data Resource Profile: The Korea Youth Risk Behavior Web-based Survey (KYRBS). International journal of epidemiology. 2016;45(4):1076–e. Epub 2016/07/07. pmid:27380796.
- 64. Must A, Willett WC, Dietz WH. Remote recall of childhood height, weight, and body build by elderly subjects. Am J Epidemiol. 1993;138(1):56–64. Epub 1993/07/01. pmid:8333427.
- 65. Jin H-J, Tyler-Smith C, Kim W. The peopling of Korea revealed by analyses of mitochondrial DNA and Y-chromosomal markers. PloS one. 2009;4(1):e4210–e. Epub 01/16. pmid:19148289.