Eur J Epidemiol

DOI 10.1007/s10654-017-0272-9

RESPIRATORY EPIDEMIOLOGY

Age at menarche and lung function: a Mendelian randomization

study
Dipender Gill1,2 • Nuala A. Sheehan3 • Matthias Wielscher4 • Nick Shrine3 •
Andre F. S. Amaral5,6 • John R. Thompson3 • Raquel Granell7 • Bénédicte Leynaert8,9 •

Francisco Gómez Real10,11 • Ian P. Hall12 • Martin D. Tobin3,13 • Juha Auvinen14 •

Susan M. Ring7 • Marjo-Riitta Jarvelin4,6,15,16,17 • Louise V. Wain3,13 •
John Henderson7 • Deborah Jarvis5,6 • Cosetta Minelli5

Received: 11 March 2017 / Accepted: 7 June 2017

Ó The Author(s) 2017. This article is an open access publication

Abstract A trend towards earlier menarche in women has by classical confounding. We estimated the effects of age
been associated with childhood factors (e.g. obesity) and at menarche on forced vital capacity (FVC), a proxy for
hypothesised environmental exposures (e.g. endocrine restrictive lung impairment, and ratio of forced expiratory
disruptors present in household products). Observational volume in one second to FVC (FEV1/FVC), a measure of
evidence has shown detrimental effects of early menarche airway obstruction, in both adulthood and adolescence. We
on various health outcomes including adult lung function, derived SNP-age at menarche association estimates for 122
but these might represent spurious associations due to variants from a published genome-wide meta-analysis
confounding. To address this we used Mendelian ran- (N = 182,416), with SNP-lung function estimates obtained
domization where genetic variants are used as proxies for by meta-analysing three studies of adult women
age at menarche, since genetic associations are not affected (N = 46,944) and two of adolescent girls (N = 3025). We
investigated the impact of departures from the assumption
of no pleiotropy through sensitivity analyses. In adult
Electronic supplementary material The online version of this women, in line with previous evidence, we found an effect
article (doi:10.1007/s10654-017-0272-9) contains supplementary
material, which is available to authorized users.
9
& Cosetta Minelli UMR 1152, Univ Paris Diderot - Paris 7, Paris, France
cosetta.minelli1@imperial.ac.uk 10
Department of Gynecology and Obstetrics, Haukeland
1
University Hospital, Bergen, Norway
Department of Clinical Pharmacology and Therapeutics, 11
Imperial College London, Hammersmith Hospital, London, Department of Clinical Science, University of Bergen,
UK Bergen, Norway
12
2
St. Mary’s Hospital, Imperial College Healthcare NHS Trust, Division of Respiratory Medicine, Queen’s Medical Centre,
London, UK University of Nottingham, Nottingham, UK
13
3
Department of Health Sciences, University of Leicester, National Institute for Health Research, Leicester Respiratory
Leicester, UK Biomedical Research Unit, Glenfield Hospital, Leicester, UK
14
4
Department of Epidemiology and Biostatistics, School of Institute of Health Sciences, University of Oulu, Oulu,
Public Health, Imperial College London, London, UK Finland
15
5
Population Health and Occupational Disease, NHLI, Imperial Biocenter Oulu, University of Oulu, Oulu, Finland
College London, Emmanuel Kaye Building, 1B Manresa 16
Center for Life Course Epidemiology, Faculty of Medicine,
Road, SW3 6LR London, UK University of Oulu, Oulu, Finland
6
MRC-PHE Centre for Environment and Health, London, UK 17
Unit of Primary Care, Oulu University Hospital, Oulu,
7
School of Social and Community Medicine, University of Finland
Bristol, Bristol, UK
8
UMR 1152, Pathophysiology and Epidemiology of
Respiratory Diseases, Epidemiology Team, Inserm, Paris,
France

123
 D. Gill et al.

on restrictive lung impairment with a 24.8 mL increase in worldwide. Restriction, defined as low total lung capacity
FVC per year increase in age at menarche (95% CI (measured by plethysmography) but commonly approxi-
1.8–47.9; p = 0.035); evidence was stronger after exclud- mated by low FVC in population-based studies, is a pre-
ing potential pleiotropic variants (43.6 mL; 17.2–69.9; dictor of all-cause mortality even in the absence of chronic
p = 0.001). In adolescent girls we found an opposite effect respiratory conditions. There has been increasing interest in
(-56.5 mL; -108.3 to -4.7; p = 0.033), suggesting that understanding gender-related risk factors for lung function
the detrimental effect in adulthood may be preceded by a impairment, particularly in relation to hormonal influences
short-term post-pubertal benefit. Our secondary analyses and sexual development. Substantial evidence shows that
showing results in the same direction in men and boys, in lung function is influenced by sex hormones in women [14]
whom age at menarche SNPs have also shown association for whom low lung function has been associated with
with sexual development, suggest a role for pubertal timing irregular menstruation, menopausal transition, and both
in general rather than menarche specifically. We found no natural and surgical menopause, with report of improve-
effect on airway obstruction (FEV1/FVC). ment in lung function in postmenopausal women receiving
hormone replacement therapy [12, 15].
Keywords Mendelian randomization
Menarche
An observational study of 2873 women aged
Puberty
Lung function
FVC
FEV1/FVC 27–57 years investigated the association of early menarche
with adult lung function and found a lower FVC and FEV1,
but not FEV1/FVC, in women with early menarche [16].
Introduction This work took account of important potential confounders,
including age, height, body mass index (BMI), smoking,
The timing of sexual development in women has shown a education and birth order, as well as secular trends (age at
secular trend with a shift towards an earlier age over the menarche has changed over the years and lung function has
years [1], and this has been related to childhood life-style also changed due to changes in height). However, the effect
and social factors, including diet and obesity, psychologi- of residual confounding by unmeasured, or poorly recalled,
cal stress and deprivation, as well as environmental expo- early life and childhood factors cannot be ruled out. For
sures, including endocrine disruptors found in many example, age at menarche is influenced by childhood
household products [2]. Menarche, defined as the date of nutritional status [17], which in turn might influence lung
the first day of the first menstrual bleeding, is preceded by a function in adult life; another example is birth weight,
complex hormonal cascade and signals the initiation of the which has been associated with both age at menarche [18]
menstrual cycle in adolescent girls. Earlier age at menarche and lung function [19]. Such limitation is typical of
has been described as a risk factor for a number of adverse observational studies, where confounding can make it hard
health outcomes, including obesity [3], type 2 diabetes [4], to distinguish between causal effects and spurious
cardio-metabolic traits [5], cardiovascular morbidity and associations.
mortality [6], as well as breast [7] and ovarian [8] cancers. Mendelian randomization (MR) can help assess the
Understanding the effects of the age at menarche offers causality of an observed association by using genetic
insight into the pathophysiology of related diseases. In variants as proxies, or ‘‘instrumental variables’’, for the
particular this can highlight the potential effects of early exposure of interest [20, 21]. Genetic associations are not
and late exposure to sex hormones on health outcomes in typically affected by confounding or reverse causation
women, as well as help explain gender differences in the because genes are randomly allocated at the time of con-
risks of common diseases [9–11]. ception. Provided the underlying assumptions are satisfied
Timing of menarche has been considered an important [22], the demonstration that genetic variants known to
factor in relation to respiratory health [12]. Lung function modify age at menarche also modify lung function pro-
is an important predictor of both respiratory disease and vides indirect evidence of a causal effect of age at
overall health. After cessation of lung growth by the early menarche. The MR technique has been rapidly growing in
twenties, there is a plateau in lung function followed by popularity because of these advantages over the classical
gradual age-related decline. There are two common pat- epidemiological approach, with MR studies having previ-
terns of lung function impairment, obstruction and ously investigated age at menarche as a risk factor for
restriction, and these have a different impact on morbidity depression [23], and lung function as an outcome in rela-
and mortality [13]. Obstruction, measured as a low ratio of tion to C-reactive protein [24].
the forced expiratory volume in one second (FEV1) to the In this study we used MR to investigate the lifetime
forced vital capacity (FVC), represents an objective marker effect of age at menarche on lung function in adolescent
of chronic obstructive pulmonary disease, a major and girls and adult women, using 122 single nucleotide poly-
growing cause of morbidity, disability and death morphisms (SNPs) associated with age at menarche. We

123
Age at menarche and lung function: a Mendelian randomization study

considered the effects in both adulthood and adolescence, based sampling frames in 14 centres, and who had lung
since we hypothesised that they could differ due to a dif- function measured in ECRHS II. Genetic associations
ferent role of menarche in lung growth, maximal lung analyses for the 122 SNPs with FVC and FEV1/FVC were
function attained, and lung function decline. adjusted for age, age2, height, centre and ancestry principal
components. NFBC1966 is a birth-cohort study performed
in the Finnish provinces of Oulu and Lapland. Pregnant
Methods women with expected date of delivery in 1966 were
recruited and their offspring followed up [28]. Among the
SNP-age at menarche association estimates offspring, we include 2680 women with spirometry data at
age 31. Analyses were adjusted for height and ancestry
We derived SNP-age at menarche association estimates principal components. UK Biobank is a prospective study
from a published genome-wide association (GWA) meta- across 22 assessment centres, aimed at identifying causes
analysis of 57 studies on 182,416 women of European of chronic disease in middle and old age [29]. We include
descent, which identified 122 independent SNPs at 106 43,195 women of European ancestry aged 40–69 recruited
genomic loci (p value \ 5 9 10-8) [25]. Overall, the 122 in 2006–2010, who had GWA and lung function data.
SNPs explained 2.7% of the variability of age at menarche Analyses were adjusted for age, age2, height, and ancestry
in the population. Age at menarche was based on self- principal components, as well as smoking pack-years
reporting and analysed as a continuous variable, with because part of the sample was ascertained by smoking
study-specific analyses adjusted for birth year, to account status [30].
for the secular trends in menarche timing, and genomic We used two studies to estimate the association of the
control, to account for population stratification [25]. We 122 SNPs with lung function in adolescent girls: Avon
assessed instrument strength for the 122 SNPs, a function Longitudinal Study of Parents and Children (ALSPAC)
of magnitude and precision of their genetic effect, using the [31] and Northern Finland Birth Cohort of 1986 (NFBC
F statistic [26]. 1986) [32] (Table 1). ALSPAC is a birth-cohort study that
initially enrolled 14,541 pregnant women in Bristol, United
SNP-lung function association estimates Kingdom, in 1990–1992 [31]. We include 1234 of their
daughters, aged around 16, with GWA and spirometry data.
Three studies were used to estimate the association of the Analyses were adjusted for height; ancestry principal
122 SNPs with lung function in adult women: European components were not included because there was no evi-
Community Respiratory Health Survey (ECRHS) [27], dence of population stratification in the study. NFBC 1986
Northern Finland Birth Cohort of 1966 (NFBC 1966) [28], is a Finnish birth-cohort study that followed up 9432 live
and UK Biobank [29] (Table 1). ECRHS is a European births to mothers in Oulu and Lapland. A total of 6642
prospective cohort study designed to identify risk factors adolescents aged 16 participated in the clinical examination
for respiratory health [27]. The study started in 1992 in 2001–2002, and here we include 1791 with available
(ECRHS I), with follow-up performed twice (ECRHS II GWA and lung function data [32]. Analyses were adjusted
and III) over the following 20 years. Here we include 1069 for height and ancestry principal components. For both
women aged 27–57 recruited at random from population- ALSPAC and NFBC 1986, robust standard errors were

Table 1 Characteristics of the study populations included for the SNP-lung function associations
Study Study design Sample Age at spirometry Age at menarche FVC (ml) FEV1 (ml) FEV1/
size (N) (years) (years) FVC (%)

Adult women
ECRHS II Multicentre cohort 1069 42.9 (7.1) 12.9 (1.5) 3677 (617) 2927 (531) 80 (7.2)
NFBC 1966 Birth cohort 2680 31 (0) 12.9 (1.3) 4014 (552) 3407 (461) 85 (6.3)
UK Biobank Multicentre cross-sectional 43,195 56.6 (7.7) 12.9 (1.6) 3094 (675) 2251 (574) 75 (6.8)
Adolescent girls
ALSPAC Birth cohort 1234 15.5 (0.3) 12.5 (1.3) 3294 (599) 3005 (545) 92 (6.8)
NFBC 1986 Birth cohort 1791 16 (0) 12.5 (1.1) 3758 (497) 3353 (446) 89 (7.0)
Values reported are mean (standard deviation)

123
 D. Gill et al.

used for the analyses of FEV1/FVC to account for deviation associated with height. We then additionally excluded
from normality of the regression residuals. SNPs previously associated with obesity and related traits
Spirometry methods for all studies are reported in (weight, BMI, waist circumference), fasting insulin and
Supplementary Table 1. For adults and adolescents, esti- type 2 diabetes, and birth weight, since these might also
mates of the association with FVC and FEV1/FVC for each influence lung function. These exclusions were used in
SNP were pooled across studies using fixed-effect inverse- sensitivity analyses if there was statistical evidence of
variance weighted meta-analysis. pleiotropy.
If statistically significant heterogeneity remained after
Mendelian randomization estimates the exclusion of these SNPs, we performed two further
sensitivity analyses: a meta-analysis of MR estimates using
We used a two-sample MR approach for summary data a random-effects model instead of the fixed-effect model
with multiple instruments, where the estimate of the causal used in the main analysis [37], and MR-Egger regression
effect is obtained as the inverse-variance weighted com- [38]. The random-effects model allows for random pleio-
bination of individual MR estimates across instruments, tropic effects across SNPs, while MR-Egger regression
using fixed-effect meta-analysis [33]. Individual MR esti- provides unbiased results if pleiotropic effects are not
mates for the 122 SNPs were derived using the Wald random (i.e. do not cancel). Both analyses assume that the
estimator (ratio of SNP-lung function estimate over SNP- magnitude of the pleiotropic effects is independent of the
age at menarche estimate), with standard error derived magnitude of the corresponding SNP-age at menarche
using the delta method [34]. effects. As these approaches are less powerful than the
fixed-effect meta-analysis, particularly the MR-Egger
Investigation of pleiotropy regression [37], they were only used as further sensitivity
analyses when between-instrument heterogeneity was still
A fundamental assumption of MR is the absence of present after excluding possible pleiotropic SNPs.
pleiotropy, i.e. the genetic instruments modify lung func- In order to understand whether the observed effects of
tion only through age at menarche and no other indepen- age at menarche on FVC were due to factors specific to
dent pathways [22]. We tested for statistical evidence of menarche rather than puberty in general, we tested the
pleiotropy by using between-instrument heterogeneity as a association of our 122 SNPs with lung function in men.
proxy; in the absence of pleiotropy, all variants are valid Many of the age at menarche SNPs that we used as
instruments and their MR estimates will vary only by instruments have also been shown to regulate male
chance (no heterogeneity) [35]. We defined evidence of pubertal timing as measured by Tanner stage [25].
pleiotropy as an I2 [ 25%, where I2 describes the per- Finding evidence of an association in men would indi-
centage of total variation in MR estimates due to hetero- cate that the underlying mechanism is related to general
geneity rather than chance [35], or a statistically significant timing of puberty as opposed to a female-specific effect.
heterogeneity Cochran Q test (p \ 0.05). If pleiotropy was SNP–FVC associations in adolescent boys (N = 3421)
detected, we performed a series of sensitivity analyses to and adult men (N = 40,687) were estimated from the
address it. same studies used for women (Supplementary Table 6).
Using the PhenoScanner, a curated database of publicly For each SNP, association estimates were pooled across
available GWA findings created to inform MR studies studies using fixed-effect inverse-variance weighted
(available at www.phenoscanner.medschl.cam.ac.uk/phe meta-analysis. The individual SNP–FVC estimates were
noscanner) [36], we first checked for previous associations then meta-analysed (fixed-effect model) to provide an
of the 122 SNPs (and highly correlated SNPs; linkage overall effect of the 122 SNPs on FVC, which is
disequilibrium r2 [ 0.8) with any phenotype other than age equivalent to performing an unweighted allele score
at menarche, limiting our search to associations that had analysis with all SNPs.
been identified at a significance level of 5 9 10-8 (Sup- All analyses were performed using Stata 14 (StataCorp
plementary Table 2). For the exclusion of possible pleio- LP)
tropic SNPs in our sensitivity analyses, we then only
considered effects on phenotypes which can be related to
lung function. We adjusted for height all SNP-lung func- Results
tion analyses and therefore height could not directly induce
pleiotropy. However, height SNPs could be associated with Imputed genotype data for the 122 SNPs were available for
other markers of somatic growth, potentially exerting all studies, with the exception of one SNP (rs10423674) in
pleiotropic effects on lung function other than through NFBC 1986. The quality of imputation was very good for
height. To test this, we excluded SNPs previously all SNPs across all studies (imputation INFO or R2

123
Age at menarche and lung function: a Mendelian randomization study

parameters C 0.8), except for one SNP in two studies obesity-related traits (Supplementary Table 2). The first
(rs17233066, R2 of 0.4 in ECRHS II and ALSPAC). The sensitivity analysis excluding 14 SNPs associated with
SNPs identified were strong instruments for age of height (Model 1 in Table 3) showed a larger and more
menarche. F statistics ranged from 21 to 441 across vari- highly statistically significant MR estimate (43.6 mL;
ants (Supplementary Table 3), well over the threshold of 17.2–69.9; p = 0.001). The second sensitivity analysis,
F [ 10 usually recommended as a test for weak instru- where we additionally excluded 13 SNPs associated with
ments in MR analyses [39]. other traits potentially related to lung function, showed
Individual estimates of the per-allele effects on age at very similar results (Model 2 in Table 3). Since some
menarche and lung function (FVC and FEV1/FVC) for residual between-instrument heterogeneity remained in
each SNP are provided in Supplementary Tables 3 and 4, both sensitivity analyses (Table 3), we performed a ran-
respectively. MR estimates for the causal effect of age at dom-effects meta-analysis of the MR estimates as well as
menarche on lung function obtained separately from each MR-Egger regression. For the random-effects meta-analy-
SNP are presented in Supplementary Table 5, while the sis, results were similar to those in Table 3, with an MR
combined MR estimates across the 122 SNPs are reported estimate of 40.7 mL (8.4–72.9; p = 0.013) for Model 1,
in Table 2. and 40.3 mL (5.7–74.8; p = 0.022) for Model 2. MR-Eg-
The MR estimate for age at menarche and FVC in adult ger regression, which suffers from low statistical power
women showed a statistically significant increase of [37], showed results in the same direction but with much
24.8 mL per year increase in age at menarche (95% con- larger confidence intervals and loss of statistical signifi-
fidence interval 1.8–47.9; p = 0.035), while we found no cance (Model 1: 95.7 mL, -37.0 to 228.4, p = 0.156;
effect for FEV1/FVC (Table 2). In the MR analysis for Model 2: 84.8 mL, -52.2 to 221.8, p = 0.222).
FVC, a between-instrument I2 of 45% (95% CI 31–55%; In adolescents, the MR analysis for FVC showed a
p \ 0.001) suggested the presence of pleiotropy, and we statistically significant decrease of 56.5 mL per year
repeated the analysis after excluding SNPs with potentially increase in age at menarche (95% CI -108.3 to -4.7;
pleiotropic effects. Out of the 122 SNPs, 34 had been p = 0.033), with no evidence of pleiotropy across the 122
previously associated with phenotypes other than age at SNPs (Table 2). As with adults, there appeared to be no
menarche, the large majority of which were height and causal effect of age at menarche on FEV1/FVC.

Table 2 MR estimates for the

Population Sample size MR estimate Between-instrument heterogeneity
causal effect of age at menarche
on lung function in adults and Beta (95% CI) p value I2 (95% CI) Het. p value
adolescents, obtained by fixed-
effect meta-analysis of SNP- FVC
specific MR estimates across the Adult women 46,944 24.8 (1.8 to 47.9) 0.035 45 (31 to 55) \0.001
122 SNPs
Adolescent girls 3025 -56.5 (-108.3 to -4.7) 0.033 2 (0 to 21) 0.418
FEV1/FVC
Adult women 46,944 0.0 (-0.7 to 0.7) 0.968 0 (0 to 19) 0.618
Adolescent girls 3025 0.6 (-0.2 to 1.4) 0.157 3 (0 to 23) 0.386
Beta, estimate of effect of 1 year increase in age at menarche on FVC (mL) and FEV1/FVC (%); I2 (%),
between-instrument heterogeneity; Het. p value, Q test p value
Bold values indicate statistically significant p values

Table 3 Sensitivity analyses for the MR of age at menarche and FVC in adult women
Model Number MR estimate Between-instrument
of SNPs heterogeneity
Beta (95% CI) p value I2 (95% Het. p
CI) value

Model 1 Excluding SNPs previously associated with height 108 43.6 (17.2–69.9) 0.001 31 (12–46) 0.002
Model 2 Excluding SNPs previously associated with height, obesity, weight, 95 42.9 (14.7–71.2) 0.003 31 (10–46) 0.003
BMI, waist circumference, fasting insulin, type 2 diabetes, or birth weight
Reported are MR estimates after excluding SNPs with possible pleiotropic effects (Suppl. Table 2). Beta, estimate of effect of 1 year increase in
age at menarche on FVC (mL); I2 (%), between-instrument heterogeneity; Het. p value, Q test p value
Bold values indicate statistically significant p values

123
 D. Gill et al.

The results of the secondary analyses in men were very adulthood. Our secondary analysis suggests that the same
consistent with those in women, with a statistically sig- happens in males, where lung development has also been
nificant positive association of the 122 SNPs with FVC in shown to plateau at puberty [40]. The beneficial effect of
adult men (p = 0.013) and a statistically significant nega- earlier menarche on FVC in adolescent girls may be
tive association in adolescent boys (p = 0.007). explained by the prominent truncal (as opposed to limb)
growth and increased thoracic muscle strength which occur
in puberty and which contribute to higher lung volumes
Discussion [40]. It could also be related to the direct effect of early
exposure to sex hormones, for example oestrogens, in
Our study shows a causal effect of age at menarche on lung adolescent girls, as these have been shown to affect lung
function using Mendelian randomization, a technique function in humans [12] and animal models [42]. Our
which draws on the biological principle that genes are secondary analysis suggesting a similar effect in boys
randomly allocated at conception to provide evidence not support the hypothesis of a mechanism related to factors
affected by classical confounding. We found an effect of associated with early pubertal timing in general rather than
age at menarche on restrictive lung impairment (FVC), specifically through female sex hormones. However, it is
with no evidence of an effect on airway obstruction (FEV1/ also possible that the mechanisms differ in men and
FVC). In particular, we find that early menarche increases women, for example through sex hormones in girls and
FVC in adolescence but decreases it in adulthood. The thoracic growth and muscle strength in boys (the latter
findings for adult women confirm previous observational being more pronounced in boys [40]). The complex hor-
evidence suggesting a decrease in FVC of 123 mL (95% CI monal and physiological shifts that occur in women during
27–220; p = 0.01) associated with early menarche (me- menarche [12] make it difficult to pin-point the precise
narche B10 years vs. menarche at 13), but no association mechanisms underlying our findings, and further research
with FEV1/FVC (p = 0.77) [16]. is needed to explore them.
The finding of a beneficial effect of earlier age at Our study suggests that if these same adolescents were
menarche on FVC in adolescence, as opposed to the assessed in early adult life (once they have reached maxi-
detrimental effect in adulthood, is interesting and has a mal lung function), those with early menarche would have
plausible explanation, illustrated graphically in Fig. 1. comparatively lower FVC. Large-scale studies of lung
Lung development tends to plateau following menarche function with longitudinal data across the lifespan will
[40], and therefore earlier initiation of menstruation may allow to test this hypothesis. To date few child cohorts
lead to premature completion of lung development and have lung function assessments in adolescence and at ages
lower maximally attained lung function. Given the relative associated with lung function plateau, but ongoing con-
stability of lung function over time (a phenomenon known sortium-based initiatives, such as STELAR [43] and
as ‘‘tracking’’) [41], this would translate to lower FVC in MEDALL [44], will be able to provide the relevant
information.
Our findings offer insight into plausible pathophysio-
logical mechanisms underlying the effects of early
menarche on lung function impairment. Previous work has
shown an association of early menarche with greater risk
and severity of asthma [16, 45, 46], while we did not find
FVC
Later menarche an effect of age at menarche on airway obstruction in either
(mL) adults or adolescents. This might be explained by an
Earlier menarche
association of early menarche with bronchial hyperactivity
through immunological and inflammatory effects [12, 47],
which would manifest as short-term reversible airway
obstruction not captured by the FEV1/FVC ratio from
single assessments in population-based studies. Our study
also highlights the importance of evaluating different lung
Age (years)
function parameters. Many epidemiological studies on lung
Fig. 1 Graphical representation of a possible explanation for the function have focused on low FEV1, which may arise from
discrepancy in FVC findings for adult women and adolescent girls. either obstructive or restrictive lung impariment. We found
Earlier menarche may have current benefits to the lung function in
that early menarche only affects FVC, a proxy for total
adolescents, but may also lead to premature completion of lung
development with attainment of a lower maximal lung function in lung capacity and characteristic of restrictive lung impar-
adult life iment, which is a predictor of morbidity (including

123
Age at menarche and lung function: a Mendelian randomization study

cardiovascular morbidity) and mortality even in the pleiotropy and made the result stronger in our secondary
absence of chronic respiratory conditions [13]. analysis in adult women supports our hypothesis that these
All 122 SNPs used in our MR study were ‘‘strong’’ SNPs could be associated with other markers of somatic
instruments, with strength reflecting not only the magni- growth, exerting pleiotropic effects on lung function other
tude of the genetic effects on age at menarche but also the than through height. Robustness of our finding for FVC in
precision of their estimates. This is important since the use women was also confirmed by a further analysis to account
of ‘‘weak’’ instruments can bias the MR estimate [39], with for residual pleiotropic effects using a random-effects
such bias resulting in an attenuation of the causal effect in meta-analysis, while MR-Egger regression resulted in a
the context of a two-sample MR analysis like ours [48]. To loss of statistical significance likely explained by its low
improve precision, we used the results from the GWA statistical power [37]. Interestingly, there was no statistical
discovery rather than replication analysis from Perry et al. evidence of pleiotropy for FVC in adolescents, as shown by
[25], as the former was over 20 times larger (182,416 vs. an I2 of 2% for the between-instrument heterogeneity (95%
8689). These estimates might have been affected by the CI 0–21%; p = 0.42). A possible explanation for this is
upward bias typical of the discovery stage (‘‘winner’s that some of our genetic instruments may have effects on
curse’’) [49], but this is likely to be very limited in our secondary phenotypes related to lung function that only
study given the strong p values. Moreover, any resulting become apparent during adult life.
overestimation of the SNP-age at menarche association A potential weakness of our study may arise from the
would have pulled the MR estimate towards the null, presence of gene-environment interactions [51], since our
leading to underestimation of the true causal effect rather MR analyses assume no interactions for the SNP-age at
than to a false positive result. menarche and SNP-lung function relationships. For
Like any other instrumental variable approach, MR example, the data used in our study for the analysis of
tends to suffer from limited statistical power when the adults and adolescents cover different ‘‘cohorts’’ of
effect of the instruments on the exposure is relatively women, potentially exposed to different environmental
small, as typically happens with common genetic variants exposures. ‘‘Cohort effects’’ are known to affect both age at
[50]. Despite this, we were able to identify statistically menarche and lung function. The MR approach is not
significant effects of age at menarche on FVC, although the susceptible to confounding from environmental exposures,
confidence intervals of our MR estimates are large, par- including those inducing cohort effects, but they might bias
ticularly for adolescents where the sample size for the the results if they interacted with the genetic variants used
genetic associations with lung function is much smaller. as instrumental variables [52]. Although the practical rel-
MR is not affected by classical confounding encoun- evance of gene-environment interactions for our MR
tered in observational studies, and yet there is a form of analyses of age at menarche and lung function is not clear,
confounding specific to MR, pleiotropy, whereby the it remains a theoretical possibility.
genetic instrument modifies lung function through sec- Finally, our MR estimate of the effect of age at
ondary phenotypes other than age at menarche [22]. menarche on FVC needs to be interpreted as a population-
Heterogeneity in the MR estimates obtained from the averaged causal effect rather than the effect for an indi-
individual instruments can be used as a proxy for pleio- vidual and are based on the assumption of a linear rela-
tropy [35]. In our study there was evidence of hetero- tionship. Parametric and non-parametric methods to
geneity in the MR estimates for the analysis of FVC in address non-linearity in MR have been proposed, including
adults; the exclusion of SNPs with possible pleiotropic stratification of the exposure to estimate localized average
effects, in particular SNPs previously associated with causal effects (LACE) [53, 54], although they typically
height, showed consistent and much stronger results than require individual-level data.
the main analysis, thus demonstrating robustness of our In conclusion, our study provides evidence of a causal
findings. Height is strongly influenced by genetic and effect of early sexual development in women on lung
environmental factors regulating growth and development, function later in life, with our secondary findings in men
and is also a strong predictor of FVC. In order to clearly suggesting a role for pubertal timing in general rather than
disentangle the effect of the age at menarche SNPs on FVC menarche specifically. This, together with evidence of
from any possible effect on height, we adjusted all our detrimental effects on other adverse health outcomes,
SNP-lung function analyses for height. Indeed, it is FVC including cardiometabolic outcomes and cancer [3–8], has
standardised for height which is clinically of interest, and public health implications given that factors predisposing
FVC is often expressed as a percentage of a normal ref- to early sexual development in women could be targeted at
erence value (percent predicted) based on the individual’s a population level to contrast the secular trend towards
height as well as sex and age. The fact that removal of earlier puberty. These include a number of established
SNPs previously associated with height reduced the childhood life-style and social factors, such as diet and

123
 D. Gill et al.

obesity, psychological stress and deprivation, as well as European Commission (018996), Fondo de Investigación Sanitaria
hypothesised environmental exposures, such as endocrine (91/0016-060-05/E, 92/0319, 93/0393, 97/0035-01, 99/0034-01 and
99/0034-02), Hospital General de Albacete, Hospital General Ramón
disrupting chemicals (EDCs) found in many household Jiménez, Consejerı́a de Sanidad del Principado de Asturias, CIRIT
products [2]. Of these, childhood obesity is the most (1997SGR 00079, 1999SGR 00241), and Servicio Andaluz de Salud,
worrisome given the current childhood obesity epidemic. SEPAR, Public Health Service (R01 HL62633-01), RCESP (C03/09),
EDCs may prove to be a substantial concern due to their Red RESPIRA (C03/011), Basque Health Department, Swiss National
Science Foundation, Swiss Federal Office for Education and Science,
widespread presence and the potential persistence of their Swiss National Accident Insurance Fund (SUVA), GSF-National
effects on menarche for generations without further expo- Research Centre for Environment and Health, Deutsche
sure (transgenerational inherited effects) [55], although Forschungsgemeinschaft (DFG) (FR 1526/1-1, MA 711/4-1), Pro-
these effects remain controversial. Our study also illus- gramme Hospitalier de Recherche CliniqueDRC de Grenoble 2000
No. 2610, Ministry of Health, Direction de la Recherche Clinique,
trates the value of the MR approach, which exploits Ministere de l’Emploi et de la Solidarite, Direction Generale de la
increasingly available genetic data from large datasets, as a Sante, CHU de Grenoble, Comite des Maladies Respiratoires de
tool to investigate causal effects of childhood events on l’Isere, UCB-Pharma (France), Aventis (France), Glaxo France,
adult health, an area of epidemiological research which is Estonian Science Foundation, and AsthmaUK (formerly known as
National Asthma Campaign UK). ALSPAC study: GWAS data was
particularly problematic due to the presence of confound- generated by Sample Logistics and Genotyping Facilities at the
ing factors very difficult to measure and partly unknown. Wellcome Trust Sanger Institute and LabCorp (Laboratory Corpor-
tation of America) using support from 23andMe. The UK Medical
Acknowledgements UK Biobank: This research has been conducted Research Council and the Wellcome Trust (Grant Ref: 102215/2/13/
using the UK Biobank Resource under Application Number 648, and 2) and the University of Bristol provide core support for ALSPAC
we thank the participants, field workers, and data managers for their study.
time and cooperation. The project used the ALICE and SPECTRE
High Performance Computing Facilities at the University of Leices- Compliance with ethical standards
ter. NFBC studies: We thank the late Professor Paula Rantakallio for
the launch of the studies, Ms Outi Tornwall and Ms Minttu Jussila for Conflict of interest The authors declare that they have no conflict of
the DNA biobanking, and the late Academian of Science Leena interest.
Peltonen for her contribution. ECRHS study: We thank the partici-
pants, field workers and researchers who have participated in the Open Access This article is distributed under the terms of the
ECRHS study for their time and cooperation. ALSPAC study: We are Creative Commons Attribution 4.0 International License (http://crea
extremely grateful to all the families who took part, the midwives for tivecommons.org/licenses/by/4.0/), which permits unrestricted use,
their help in recruiting them, and the whole ALSPAC team, which distribution, and reproduction in any medium, provided you give
includes interviewers, computer and laboratory technicians, clerical appropriate credit to the original author(s) and the source, provide a
workers, research scientists, volunteers, managers, receptionists and link to the Creative Commons license, and indicate if changes were
nurses. made.

Funding This project has received funding from the European

Union’s Horizon 2020 research and innovation programme under
Grant Agreement No. 633212. IPH was supported by an MRC pro-
gramme Grant (G1000861). RG was supported by the UK Medical References
Research Council (Grant Ref: G0902125). The work of MDT, IPH
and LVW was funded by a Medical Research Council (MRC) 1. Adams Hillard PJ. Menstruation in adolescents: what’s normal,
strategic award (MC_PC_12010). MDT was supported by MRC fel- what’s not. Ann N Y Acad Sci. 2008;1135:29–35. doi:10.1196/
lowships G0501942 and G0902313. This article presents independent annals.1429.022.
research funded partially by the National Institute for Health Research 2. Fisher MM, Eugster EA. What is in our environment that effects
(NIHR). The views expressed are those of the authors and not nec- puberty? Reprod Toxicol. 2014;44:7–14. doi:10.1016/j.reprotox.
essarily those of the NHS, the NIHR or the Department of Health. 2013.03.012.
NFBC1966 study: NFBC1966 received financial support from the 3. Pierce MB, Leon DA. Age at menarche and adult BMI in the
Academy of Finland (Project Grants 104781, 120315, 129269, Aberdeen children of the 1950s cohort study. Am J Clin Nutr.
1114194, 24300796, Center of Excellence in Complex Disease 2005;82(4):733–9.
Genetics and SALVE), University Hospital Oulu, Biocenter, 4. He C, Zhang C, Hunter DJ, et al. Age at menarche and risk of
University of Oulu, Finland (75617), NHLBI Grant 5R01HL087679- type 2 diabetes: results from 2 large prospective cohort studies.
02 through the STAMPEED program (1RL1MH083268-01), NIH/ Am J Epidemiol. 2010;171(3):334–44. doi:10.1093/aje/kwp372.
NIMH (5R01MH63706:02), ENGAGE Project and Grant Agreement 5. Remsberg KE, Demerath EW, Schubert CM, Chumlea WC, Sun
HEALTH-F4-2007-201413, EU FP7 EurHEALTHAgeing 277849, SS, Siervogel RM. Early menarche and the development of car-
the Medical Research Council, UK (G0500539, G0600705, diovascular disease risk factors in adolescent girls: the fels lon-
G1002319, PrevMetSyn/SALVE) and the MRC, Centenary Early gitudinal study. J Clin Endocrinol metab. 2005;90(5):2718–24.
Career Award. The program is currently being funded by the H2020- doi:10.1210/jc.2004-1991.
633595 DynaHEALTH action and academy of Finland EGEA-pro- 6. Lakshman R, Forouhi NG, Sharp SJ, et al. Early age at menarche
ject. The DNA extractions, sample quality controls, biobank associated with cardiovascular disease and mortality. J Clin
upkeeping and aliquotting were performed in the National Public Endocrinol Metab. 2009;94(12):4953–60. doi:10.1210/jc.2009-
Health Institute, Biomedicum Helsinki, Finland and were supported 1789.
financially by the Academy of Finland and Biocentrum Helsinki. 7. Hsieh CC, Trichopoulos D, Katsouyanni K, Yuasa S. Age at
ECRHS study: This work was supported by a contract from the menarche, age at menopause, height and obesity as risk factors

123
Age at menarche and lung function: a Mendelian randomization study

for breast cancer: associations and interactions in an international 26. Palmer TM, Lawlor DA, Harbord RM, et al. Using multiple
case-control study. Int J Cancer. 1990;46(5):796–800. genetic variants as instrumental variables for modifiable risk
8. Gong TT, Wu QJ, Vogtmann E, Lin B, Wang YL. Age at factors. Stat Methods Med Res. 2012;21(3):223–42. doi:10.1177/
menarche and risk of ovarian cancer: a meta-analysis of epi- 0962280210394459.
demiological studies. Int J Cancer. 2013;132(12):2894–900. 27. Burney PG, Luczynska C, Chinn S, Jarvis D. The European
doi:10.1002/ijc.27952. Community Respiratory Health Survey. Eur Respir J.
9. Global Burden of Disease Study C. Global, regional, and national 1994;7(5):954–60.
incidence, prevalence, and years lived with disability for 301 28. Rantakallio P. The longitudinal study of the northern Finland birth
acute and chronic diseases and injuries in 188 countries, cohort of 1966. Paediatr Perinat Epidemiol. 1988;2(1):59–88.
1990–2013: a systematic analysis for the Global Burden of Dis- 29. Sudlow C, Gallacher J, Allen N, et al. UK biobank: an open
ease Study 2013. Lancet. 2015;386(9995):743–800. doi:10.1016/ access resource for identifying the causes of a wide range of
S0140-6736(15)60692-4. complex diseases of middle and old age. PLoS Med.
10. Mendelsohn ME, Karas RH. Molecular and cellular basis of 2015;12(3):e1001779. doi:10.1371/journal.pmed.1001779.
cardiovascular gender differences. Science. 2005;308(5728): 30. Wain LV, Shrine N, Miller S, et al. Novel insights into the
1583–7. doi:10.1126/science.1112062. genetics of smoking behaviour, lung function, and chronic
11. Carey MA, Card JW, Voltz JW, et al. It’s all about sex: gender, obstructive pulmonary disease (UK BiLEVE): a genetic associ-
lung development and lung disease. Trends Endocrinol Metab. ation study in UK Biobank. Lancet Respir Med.
2007;18(8):308–13. doi:10.1016/j.tem.2007.08.003. 2015;3(10):769–81. doi:10.1016/S2213-2600(15)00283-0.
12. Macsali F, Svanes C, Bjorge L, Omenaas ER, Gomez RF. Res- 31. Boyd A, Golding J, Macleod J, et al. Cohort Profile: the ‘children
piratory health in women: from menarche to menopause. Exp Rev of the 90s’—the index offspring of the Avon Longitudinal Study
Respir Med. 2012;6(2):187–200. doi:10.1586/ers.12.15. of Parents and Children. Int J Epidemiol. 2013;42(1):111–27.
13. Burney PG, Hooper R. Forced vital capacity, airway obstruction doi:10.1093/ije/dys064.
and survival in a general population sample from the USA. 32. Jaaskelainen A, Schwab U, Kolehmainen M, et al. Meal fre-
Thorax. 2011;66(1):49–54. doi:10.1136/thx.2010.147041. quencies modify the effect of common genetic variants on body
14. Becklake MR, Kauffmann F. Gender differences in airway mass index in adolescents of the northern Finland birth cohort
behaviour over the human life span. Thorax. 1999;54(12): 1986. PLoS ONE. 2013;8(9):e73802. doi:10.1371/journal.pone.
1119–38. 0073802.
15. Amaral AF, Strachan DP, Gomez Real F, Burney PG, Jarvis DL. 33. Burgess S, Butterworth A, Thompson SG. Mendelian random-
Lower lung function associates with cessation of menstruation: ization analysis with multiple genetic variants using summarized
UK Biobank data. Eur Respir J. 2016;48(5):1288–97. doi:10. data. Genet Epidemiol. 2013;37(7):658–65. doi:10.1002/gepi.
1183/13993003.00412-2016. 21758.
16. Macsali F, Real FG, Plana E, et al. Early age at menarche, lung 34. Thompson JR, Minelli C, Del Greco MF. Mendelian random-
function, and adult asthma. Am J Respir Crit Care Med. ization using public data from genetic consortia. Int J Biostat.
2011;183(1):8–14. doi:10.1164/rccm.200912-1886OC. 2016;. doi:10.1515/ijb-2015-0074.
17. Yermachenko A, Dvornyk V. Nongenetic determinants of age at 35. Del Greco MF, Minelli C, Sheehan NA, Thompson JR. Detecting
menarche: a systematic review. Biomed Res Int. 2014;2014:371 pleiotropy in Mendelian randomisation studies with summary
583. doi:10.1155/2014/371583. data and a continuous outcome. Stat Med. 2015;34(21):2926–40.
18. Adair LS. Size at birth predicts age at menarche. Pediatrics. doi:10.1002/sim.6522.
2001;107(4):E59. 36. Staley JR, Blackshaw J, Kamat MA, et al. PhenoScanner: a
19. Barker DJ, Godfrey KM, Fall C, Osmond C, Winter PD, Shaheen database of human genotype-phenotype associations. Bioinfor-
SO. Relation of birth weight and childhood respiratory infection matics. 2016;32(20):3207–9. doi:10.1093/bioinformatics/btw373.
to adult lung function and death from chronic obstructive airways 37. Bowden J, Del Greco MF, Minelli C, Davey Smith G, Sheehan N,
disease. BMJ. 1991;303(6804):671–5. Thompson J. A framework for the investigation of pleiotropy in
20. Davey Smith G, Ebrahim S. What can Mendelian randomisation two-sample summary data Mendelian randomization. Stat Med.
tell us about modifiable behavioural and environmental expo- 2017;36:1783–1802. doi:10.1002/sim.7221.
sures? BMJ. 2005;330(7499):1076–9. doi:10.1136/bmj.330.7499. 38. Bowden J, Davey Smith G, Burgess S. Mendelian randomization
1076. with invalid instruments: effect estimation and bias detection
21. Smith GD, Ebrahim S. Mendelian randomization: can genetic through Egger regression. Int J Epidemiol. 2015;44(2):512–25.
epidemiology contribute to understanding environmental deter- doi:10.1093/ije/dyv080.
minants of disease? Int J Epidemiol. 2003;32(1):1–22. 39. Lawlor DA, Harbord RM, Sterne JA, Timpson N, Davey SG.
22. Sheehan NA, Didelez V, Burton PR, Tobin MD. Mendelian Mendelian randomization: using genes as instruments for making
randomisation and causal inference in observational epidemiol- causal inferences in epidemiology. Stat Med.
ogy. PLoS Med. 2008;5(8):e177. doi:10.1371/journal.pmed.0050 2008;27(8):1133–63. doi:10.1002/sim.3034.
177. 40. Neve V, Girard F, Flahault A, Boule M. Lung and thorax
23. Sequeira ME, Lewis SJ, Bonilla C, Davey Smith G, Joinson C. development during adolescence: relationship with pubertal sta-
Association of timing of menarche with depressive symptoms and tus. Eur Respir J. 2002;20(5):1292–8.
depression in adolescence: Mendelian randomisation study. Br J 41. Twisk JW, Staal BJ, Brinkman MN, Kemper HC, van Mechelen
Psychiatry. 2016;. doi:10.1192/bjp.bp.115.168617. W. Tracking of lung function parameters and the longitudinal
24. Bolton CE, Schumacher W, Cockcroft JR, et al. The CRP relationship with lifestyle. Eur Respir J. 1998;12(3):627–34.
genotype, serum levels and lung function in men: the Caerphilly 42. Massaro D, Massaro GD. Estrogen regulates pulmonary alveolar
Prospective Study. Clin Sci (Lond). 2011;120(8):347–55. doi:10. formation, loss, and regeneration in mice. Am J Physiol Lung
1042/CS20100504. Cell Mol Physiol. 2004;287(6):L1154–9. doi:10.1152/ajplung.
25. Perry JR, Day F, Elks CE, et al. Parent-of-origin-specific allelic 00228.2004.
associations among 106 genomic loci for age at menarche. Nat- 43. Custovic A, Ainsworth J, Arshad H, Bishop C, Buchan I, Cullinan
ure. 2014;514(7520):92–7. doi:10.1038/nature13545. P, et al. The Study Team for Early Life Asthma Research

123
 D. Gill et al.

(STELAR) consortium ‘Asthma e-lab’: team science bringing data, 50. Pierce BL, Ahsan H, Vanderweele TJ. Power and instrument
methods and investigators together. Thorax 2015;70(8):799–801. strength requirements for Mendelian randomization studies using
44. Bousquet J, Anto J, Auffray C, Akdis M, Cambon-Thomsen A, multiple genetic variants. Int J Epidemiol. 2011;40(3):740–52.
Keil T, et al. MeDALL (Mechanisms of the Development of doi:10.1093/ije/dyq151.
ALLergy): an integrated approach from phenotypes to systems 51. Dudbridge F, Fletcher O. Gene-environment dependence creates
medicine. Allergy 2011;66:596–604. spurious gene-environment interaction. Am J Hum Genet.
45. Salam MT, Wenten M, Gilliland FD. Endogenous and exogenous 2014;95(3):301–7. doi:10.1016/j.ajhg.2014.07.014.
sex steroid hormones and asthma and wheeze in young women. 52. Brennan P. Commentary: Mendelian randomization and gene-
J Allergy Clin Immunol. 2006;117(5):1001–7. doi:10.1016/j.jaci. environment interaction. Int J Epidemiol. 2004;33(1):17–21.
2006.02.004. doi:10.1093/ije/dyh033.
46. Varraso R, Siroux V, Maccario J, Pin I, Kauffmann F. Asthma 53. Burgess S, Davies NM, Thompson SG. Instrumental variable anal-
severity is associated with body mass index and early menarche ysis with a nonlinear exposure-outcome relationship. Epidemiology.
in women. Am J Res Criti Care Med. 2005;171(4):334–9. doi:10. 2014;25(6):877–85. doi:10.1097/ede.0000000000000161.
1164/rccm.200405-674OC. 54. Silverwood RJ, Holmes MV, Dale CE, et al. Testing for non-
47. Haggerty CL, Ness RB, Kelsey S, Waterer GW. The impact of linear causal effects using a binary genotype in a Mendelian
estrogen and progesterone on asthma. Ann Allergy Asthma Immu- randomization study: application to alcohol and cardiovascular
nol. 2003;90(3):284–91. doi:10.1016/S1081-1206(10)61794-2. traits. Int J Epidemiol. 2014;43(6):1781–90. doi:10.1093/ije/
48. Pierce BL, Burgess S. Efficient design for Mendelian random- dyu187.
ization studies: subsample and 2-sample instrumental variable 55. Zama AM, Uzumcu M. Epigenetic effects of endocrine-disrupt-
estimators. Am J Epidemiol. 2013;178(7):1177–84. doi:10.1093/ ing chemicals on female reproduction: an ovarian perspective.
aje/kwt084. Front Neuroendocrinol. 2010;31(4):420–39. doi:10.1016/j.yfrne.
49. Ioannidis JP, Ntzani EE, Trikalinos TA, Contopoulos-Ioannidis 2010.06.003.
DG. Replication validity of genetic association studies. Nat
Genet. 2001;29(3):306–9. doi:10.1038/ng749.

123