DNA methylation age of blood predicts all-cause mortality in later life
Marioni et al. Genome Biology
DNA methylation age of blood predicts all-cause mortality in later life
Riccardo E Marioni 0 1
Sonia Shah 0
Allan F McRae 0
Brian H Chen
Sarah E Harris 1
Anjali K Henders
Simon R Cox 1
Nicholas G Martin
Grant W Montgomery
Andrew P Feinberg
M Daniele Fallin
Michael L Multhaup
Andrew E Jaffe
Allan C Just
Kathryn L Lunetta
Joanne M Murabito
John M Starr 1
Andrea A Baccarelli
Peter M Visscher 0 1
Naomi R Wray 0
Ian J Deary 1
0 Queensland Brain Institute, The University of Queensland , Brisbane 4072, QLD , Australia
1 Centre for Cognitive Ageing and Cognitive Epidemiology, University of Edinburgh , 7 George Square, Edinburgh EH8 9JZ , UK
Background: DNA methylation levels change with age. Recent studies have identified biomarkers of chronological age based on DNA methylation levels. It is not yet known whether DNA methylation age captures aspects of biological age. Results: Here we test whether differences between people's chronological ages and estimated ages, DNA methylation age, predict all-cause mortality in later life. The difference between DNA methylation age and chronological age (age) was calculated in four longitudinal cohorts of older people. Meta-analysis of proportional hazards models from the four cohorts was used to determine the association between age and mortality. A 5-year higher age is associated with a 21% higher mortality risk, adjusting for age and sex. After further adjustments for childhood IQ, education, social class, hypertension, diabetes, cardiovascular disease, and APOE e4 status, there is a 16% increased mortality risk for those with a 5-year higher age. A pedigree-based heritability analysis of age was conducted in a separate cohort. The heritability of age was 0.43. Conclusions: DNA methylation-derived measures of accelerated aging are heritable traits that predict mortality independently of health status, lifestyle factors, and known genetic factors.
DNA sequence variants and epigenetic marks that are
associated with changes in gene expression contribute
to interindividual variation in complex phenotypes.
Epigenetic mechanisms such as DNA methylation,
characterized by the addition of a methyl group to a cytosine
nucleotide primarily at cytosine-phosphate-guanine (CpG)
sites, play essential roles during development, acting
through the regulation of gene expression . Unlike
genomic variants, such as single nucleotide
polymorphisms (SNPs), levels of DNA methylation vary across
the life course [2-6]. DNA methylation levels are
influenced by lifestyle and environmental factors , as well
as by genetic variation [8,9].
Age-related changes in DNA methylation are also well
documented, and two recent studies used methylation
measures from multiple CpG sites across the genome to
predict chronological age in humans [10,11]. Hannum
et al. created an age predictor based on a single cohort
in which DNA methylation was measured in whole
blood . Horvath developed an age predictor using
DNA methylation data from multiple studies (including
the Hannum dataset) and multiple tissues . In both
studies, the difference between methylation-predicted
age and chronological age (that is, age) was put forth as
an index of disproportionate biological aging and was
hypothesized to be associated with risk for age-related
diseases and mortality [10,11]. Weidner et al. 
proposed an age predictor based on three CpGs taken
from a methylation array with fewer total CpG sites than
the Hannum and Horvath models (27 k probes versus
450 k probes). To date, however, no study has tested
whether DNA methylation-based age or other
genomewide DNA methylation biomarkers are significant
predictors of all-cause mortality.
Here, we tested the association of two DNA
methylation measures of age (using the Hannum and Horvath
predictors) with all-cause mortality in four cohorts: the
Lothian Birth Cohorts of 1921, and 1936 [13-15], the
Framingham Heart Study [16,17], and the Normative
Aging Study [18,19]. In addition, we estimated the
heritability of age using the Brisbane Systems Genetics Study
The association between age (DNA
methylation-predicted age minus chronological age) and mortality was
examined in four cohorts: Lothian Birth Cohort 1921
(LBC1921) (N = 446, ndeaths = 292), Lothian Birth Cohort
1936 (LBC1936) (N = 920, ndeaths = 106), the Framingham
Heart Study (FHS) (N = 2,635, ndeaths = 238), and the
Normative Aging Study (NAS) (N = 657, ndeaths = 226).
The mean ages of the cohorts were 79.1 (SD 0.6), 69.5
(SD 0.8), 66.3 (SD 8.9), and 72.9 (SD 6.9) years,
respectively. The Hannum predicted values were higher than
the participants chronological ages by a mean of 2 to
6 years (SDs approximately 5 years) across the four
cohorts. The Horvath predicted values were lower than
the chronological ages in LBC1921 and LBC1936
participants by 4 to 5 years (SD approximately 6 years) but
very similar to chronological age in the FHS (0.60 years;
SD 5.2) and the NAS (0.6 years; SD 5.8). A third
predictor, based on the Weidner predictor was also
examined, although it had a low correlation with
chronological age (LBC1921: Pearson R = 0.02; LBC1936: Pearson
R = 0.03; FHS: Pearson R = 0.25; NAS: Pearson R = 0.43)
and very large absolute median differences (LBC1921:
29.9 years, LBC1936: 19.8 years, FHS: 12.6 years, NAS:
18.4 years) so was not examined further. A full
description of the cohorts is provided in Table 1 and
Additional file 1. Combining information from these
studies, the correlation between chronological age and
predicted age was 0.83 for the Hannum measure and
0.75 for the Horvath measure (Figure 1). The
correlation between the Hannum and Horvath predictors
The corresponding increase in mortality risk for the
Horvath age was 11% (95% CI (1.05, 1.18), P = 0.0003).
Kaplan-Meier survival curves for the Horvath and
Hannum age (split into highest versus lowest quartile,
for descriptive purposes only) in these models are
presented in Figure 3 for the LBC1921 sample, which was
the study with the greatest number of deaths. The plot
illustrates the higher mortality rate for those with
Association between methylation age indicators and
white blood cell counts
It is well known that blood cell types have distinct
methylation profiles [21,22]. A sensitivity analysis adjusting for
white blood cell counts (basophils, eosinophils,
monocytes, lymphocytes, and neutrophils) resulted in mostly
minor differences to the results (Additional file 2).
In addition to the five white blood cell types, we also
examined the association between estimated naive T cell
abundance and age. However, to prevent spurious
correlations between age, which by definition correlates
negatively with age, and cell counts, we used age
acceleration defined as the residuals from a regression
of predicted age on chronological age. There were
stronger associations with these measures and the
Hannum predictor (naive CD4+ T cells, average
correlation r = 0.35, naive CD8+ T cells, average r = 0.34)
compared to the Horvath predictor (r = 0.20, and 0.20,
respectively). After adjustment for naive T cells, both
predictors were still significantly associated with mortality.
Naive T cell count was also associated with mortality in
addition to the Hannum predictor (Additional file 3).
Chronological age had a significant negative relationship
with the abundance of naive T cells (on average r = 0.12
for CD8+ and r = 0.10 for CD4+ T cells, Additional
The moderately strong correlation between naive T
cell abundance and the Hannum predictor suggests that
the latter keeps track of the age-related decline of
certain T cell populations. This makes sense in light of the
facts that: (1) the Hannum predictor was constructed on
the basis of DNA methylation data from whole blood;
and (2) that naive T cells diminish with age due to age
related thymic involution. In contrast to the Hannum
predictor, the Horvath predictor exhibits a weaker
relationship with naive T cell abundance, which probably
reflects the fact that it was constructed on a range of
different tissues and cell types.
Methylation age acceleration predicts mortality
In the meta-analyzed results across the four cohorts, a
5-year higher Hannum age was associated with a 21%
(95% CI (1.14, 1.29), P <0.0001) greater mortality risk
after adjustment for chronological age and sex (Figure 2).
Adjusting for demographic variables and risk factors
Sensitivity analyses were performed to control for
potentially confounding variables: smoking, education,
childhood IQ (LBC1921 and LBC1936 only), social
class (LBC1921 and LBC1936 only), APOE (LBC1921,
Table 1 Summary details of the four analysis cohorts
Hannum methylation age (years)
Hannum median error (years)
Horvath methylation age (years)
Horvath median error (years)
FHS: Framingham Heart Study, LBC: Lothian Birth Cohort, NAS: Normative Aging Study.
LBC1936, and NAS only), cardiovascular disease, high
blood pressure, and diabetes. When entered together
in a fully adjusted model (Figure 2) the meta-analyzed
hazards ratio (HR) per 5-year increment was 1.16 (95%
CI (1.08, 1.25), P = 6.0x10-9) for the fully adjusted
Hannum age and 1.09 (95% CI (1.02, 1.15), P = 0.0069)
for the fully adjusted Horvath age. In the LBC
datasets, which were the only datasets to contain
information on all of the covariates, inclusion of the covariates
one at a time made very little difference to basic, age-,
and sex-adjusted results (Additional file 4).
Separate age-adjusted Cox models for men and women
are presented in Additional file 5; there was no notable
difference in the relation of age to survival by sex.
Sensitivity analyses that excluded deaths within the first 2 years
of follow-up made negligible differences to the effect size
and significance of the age associations (Additional
We tested the associations between age and several
key covariates (Additional file 7). With the exception of
sex, where women had significantly lower age estimates
than men, there were no consistent associations with the
covariates. There was some evidence for an association
between the Hannum (but not Horvath) age and
childhood IQ and social class, although these covariates were
assessed in only the two LBC cohorts.
Heritability of methylation age
Using data from the BSGS cohort, the estimated
heritability for the Horvath and Hannum age was 0.43 (SE
0.07, P = 91013) and 0.42 (SE 0.07, P = 41010),
respectively, indicating that approximately 40% of
interindividual differences in age are due to genetic factors.
The contribution to heritability broken down across
relationship classes is given in Figure 4, and is
consistent with an additive genetic model of inheritance.
The difference between DNA methylation predicted age
and chronological age (age) using two sets of epigenetic
markers [10,11] is a heritable trait that is associated with
an increased risk of mortality in four independent
cohorts of older individuals. This association is
independent of life-course predictors of aging and death such as
possession of the e4 allele of APOE, education,
childhood IQ, social class, diabetes, high blood pressure, and
cardiovascular disease. Moreover, there was no clear
association between these factors and age.
A strength of our study is that we evaluate two distinct
epigenetic biomarkers of aging (age) one from DNA
methylation in whole blood, and one based on results
across multiple tissues. The associations between age
and mortality were stronger for the blood-based
predictor but the two measures became comparable after
adjusting for naive CD8 T cell abundances (Additional
file 3). It is well known that immunosenescence is
accompanied by diminishing naive T cells due to thymic
involution and that the aging immune system is a
predictor of human longevity [23,24]. While it is possible
that the association between accelerated epigenetic age
and all-cause mortality is mediated by changes in blood
cell composition, we think that this is unlikely for the
following reasons. First, the Horvath measure only
exhibited a weak relationship (average r = 0.20) with naive
T cell abundance which reflects its definition and
applicability to most tissues and cell types. Second, both
Figure 1 Plot of predicted methylation age against
chronological age and plot of Hannum versus Horvath
predicted methylation age. *To prevent the potential identification
of individual participants, only FHS data points with chronological
ages between 45 and 85, and NAS data points between ages 56
and 100 are displayed. r = Pearson correlation coefficient. FHS:
Framingham Heart Study, LBC: Lothian Birth Cohort, NAS: Normative
measures remained significant predictors of mortality in
multivariate regression models that adjusted for naive T
Given the similarity of the findings for the two
epigenetic biomarkers of aging, it is a credible hypothesis that
age is an epigenetic biomarker of the pace of biological
aging throughout life.
Nearly all biological and lifestyle factors that we
studied did not materially influence either the Hannum or
Horvath age across the four cohorts. Specifically,
vascular insults, as measured by diabetes and hypertension,
which are linked to cognitive decline, dementia, and
death [25-27], were not associated with age. There were
sex differences, with men having higher age than
women, which is consistent with previous findings .
Childhood IQ and social class were modestly associated
with age in the two Lothian Birth Cohorts; an increased
IQ in childhood and a less deprived social class were
associated with lower age in later life, although these were
driven almost entirely by LBC1936. These correlated
variables have been repeatedly associated with a range of
health inequalities, including mortality , and it is
possible that age might offer insight into the
mechanisms by which they are linked to health outcomes.
There is continued interest in identifying new risk
factors, environmental, genetic, and epigenetic that can
improve our ability to predict disease and mortality.
Epidemiological studies have identified numerous
measures from across the human life-course that are
associated with an increased risk of mortality. These include
health factors such as cardiovascular disease, diabetes,
and hypertension , genetic factors such as presence
of the APOE e4 allele , lifestyle variables such as
smoking  and education , behavioral traits such
as cognitive ability [31,32], the personality trait of
conscientiousness , and candidate biomarkers of age
such as telomere length [34,35]. Here, we report on an
epigenetic biomarker that is predictive of human
mortality, after accounting for known risk factors. We found
that two heritable DNA methylation-based measures of
the difference between epigenetic age and chronological
age are significant predictors of mortality in our
metaanalysis of four independent cohorts of older people.
Figure 2 Meta-analysis results of age versus mortality. The basic adjusted models controlled for chronological age, sex (NAS had only male
participants), and laboratory batch (FHS only). The fully adjusted models controlled for chronological age, sex, smoking, education, childhood IQ
(LBC1921 and LBC1936 only), social class (LBC1921 and LBC1936 only), APOE (LBC1921, LBC1936, and NAS only), cardiovascular disease, high
blood pressure, and diabetes. CI: confidence interval, FHS: Framingham Heart Study, HR: hazard ratio, LBC: Lothian Birth Cohort, NAS: Normative
Aging Study, W: fixed effect weight.
Individual genetic or environmental exposures that drive
the associations are not yet known, but they appear not
to be clearly linked to classic life-course risk factors. The
difference between DNA methylation age and
chronological age predicts mortality risk over and above a
combination of smoking, education, childhood IQ, social
class, APOE genotype, cardiovascular disease, high blood
pressure, and diabetes. It may therefore be possible to
think of DNA methylation predicted age as an
epigenetic clock  that measures biological age and runs
alongside, but not always in parallel with chronological
age, and may inform life expectancy predictions. Our
Figure 4 Heritability of methylation age. (A) Intra-class correlation
of Hannum and Horvath age across relationship class. (B) Heritability
of Hannum and Horvath age in BSGS data. Both plots show estimates
with standard errors. *Pseudo-independent pairs. r represents the
degree of relatedness.
results imply that epigenetic marks, such as gene
methylation, are like other complex traits: influenced by both
genetic and environmental factors and associated with
major health-related outcomes.
Materials and methods
The Lothian Birth Cohort 1921
Data were from the Lothian Birth Cohort 1921
(LBC1921), which is the basis of a longitudinal study of
aging [13,15]. Participants were born in 1921 and most
completed a cognitive ability test at about the age of
11 years in the Scottish Mental Survey 1932 (SMS1932)
. The SMS1932 was administered nationwide to
almost all 1921-born children who attended school in
Scotland on 1 June 1932. The cognitive test was the
Moray House Test No. 12, which provides a measure of
general cognitive ability and has a scoring range between
0 and 76. The LBC1921 study attempted to follow up
individuals who might have completed the SMS1932 and
resided at about the age of 79 years in the Lothian
region (Edinburgh and its surrounding areas) of Scotland;
550 people (n = 234, 43% men) were successfully traced
and participated in the study from the age of 79 years.
To date, there have been four additional follow-up waves
at average ages of 83, 87, 90, and 92 years. The cohort
has been deeply phenotyped during the later-life waves,
including blood biomarkers, cognitive testing, and
psycho-social, lifestyle, and health measures .
Genome wide single nucleotide polymorphisms and exome
chip data are also available. DNA methylation measured
in subjects at an average age of 79 (n = 514) was used for
analyses in this report.
Lothian Birth Cohort 1936
The methylation mortality survival analysis was
investigated in a second study, the Lothian Birth Cohort 1936
(LBC1936) [13,14]. All participants were born in 1936.
Most had taken part in the Scottish Mental Survey 1947
at a mean age of 11 years as part of national testing of
almost all children born in 1936 who attended Scottish
schools on 4 June 1947 . The cognitive test
administered was the same Moray House Test No. 12 used in
the SMS1932. A total of 1,091 participants (n = 548, 50%
men) who were living in the Lothian area of Scotland
were re-contacted in later life. Extensive phenotyping
has also been carried out in this study, with data
collection waves at three time points . Genome-wide
single nucleotide polymorphisms and exome chip data are
also available. DNA methylation was measured in 1,004
subjects at Wave 1 (mean age, 70 years). To date, there
have been two additional follow-up waves at average
ages of 73 and 76 years.
The Framingham Heart Study
Framingham Heart Study (FHS) is a community-based
longitudinal study of participants living in and near
Framingham, MA, at the start of the study in 1948
. The Offspring cohort comprised the children and
spouses of the original FHS participants, as described
previously . Briefly, enrollment for the Offspring
cohort began in 1971 (n = 5,124), and in-person
evaluations occurred approximately every 4 to 8 years
thereafter. The current analysis was limited to participants
from the Offspring cohort who survived until the
eighth examination cycle (2005 to 2008) and consented
to genetics research. DNA methylation data of
peripheral blood samples collected at the eighth examination
cycle were available in 2,741 participants.
The Normative Aging Study
The US Department of Veterans Affairs (VA) Normative
Aging Study (NAS) is an ongoing longitudinal cohort
established in 1963, which included men who were aged
21 to 80 years and free of known chronic medical
conditions at entry [18,19]. Participants were subsequently
invited to medical examinations every 3 to 5 years. At
each visit, participants provided information on medical
history, lifestyle, and demographic factors, and
underwent a physical examination and laboratory tests. DNA
samples were collected from 1999 to 2007 from the 675
active participants and used for DNA methylation
analysis. We excluded 18 participants who were not of
European descent or had missing information on race,
leaving a total of 657 individuals.
Brisbane Systems Genetics Study
The Brisbane Systems Genetic Study (BSGS)  is a
cohort comprising adolescent monozygotic (MZ) and
dizygotic (DZ) twins, their siblings, and their parents.
They were originally recruited into an ongoing study of
the genetic and environmental factors influencing
cognition and pigmented nevi. DNA methylation was
measured on 614 individuals from 117 families of European
descent. Families consist of adolescent monozygotic
(MZ; n = 67 pairs) and dizygotic (DZ; n = 111 pairs) twins,
their siblings (n = 119), and their parents (n = 139).
Children have a mean age of 14 years (age range, 923 years)
and parents 47 years (age range, 3375 years).
Following informed consent, venesected whole blood
was collected for DNA extraction in both LBC1921 and
LBC1936. Ethics permission for the LBC1921 was
obtained from the Lothian Research Ethics Committee
(Wave 1: LREC/1998/4/183). Ethics permission for the
LBC1936 was obtained from the Multi-Centre Research
Ethics Committee for Scotland (Wave 1: MREC/01/0/
56), the Lothian Research Ethics Committee (Wave 1:
LREC/2003/2/29). Written informed consent was
obtained from all subjects.
All participants provided written informed consent at
the time of each examination visit. The study protocol
was approved by the Institutional Review Board at
Boston University Medical Center (Boston, MA, USA).
The NAS study was approved by the Institutional Review
Boards (IRBs) of the participating institutions. Participants
have provided written informed consent at each visit.
The BSGS study was approved by the Queensland Institute
for Medical Research Human Research Ethics Committee.
All participants gave informed written consent.
DNA methylation measurement
In all cohorts, bisulphite converted DNA samples
were hybridised to the 12 sample Illumina
HumanMethylation450BeadChips  using the Infinium HD
Methylation protocol and Tecan robotics (Illumina, San
Diego, CA, USA).
LBC1921 and LBC1936 DNA methylation
DNA was extracted from 514 whole blood samples in
LBC1921 and from 1,004 samples in LBC1936. Samples
were extracted at MRC Technology, Western General
Hospital, Edinburgh (LBC1921) and the Wellcome Trust
Clinical Research Facility (WTCRF), Western General
Hospital, Edinburgh (LBC1936), using standard methods.
Methylation typing of 485,512 probes was performed at
the WTCRF. Raw intensity data were
backgroundcorrected and methylation beta-values generated using the
R minfi package . Quality control analysis was
performed to remove probes with a low (<95%) detection rate
at P <0.01. Manual inspection of the array control probe
signals was used to identify and remove low quality
samples (for example, samples with inadequate hybridization,
bisulfite conversion, nucleotide extension, or staining
signal). The Illumina-recommended threshold was used to
eliminate samples with a low call rate (samples with
<450,000 probes detected at P <0.01). Since the LBC
samples had previously been genotyped using the Illumina
610-Quadv1 genotyping platform, genotypes derived from
the 65 SNP control probes on the methylation array using
the wateRmelon package  were compared to those
obtained from the genotyping array to ensure sample
integrity. Samples with a low match of genotypes with SNP
control probes, which could indicate sample
contamination or mix-up, were excluded (n = 9). Moreover, eight
subjects whose predicted sex, based on XY probes, did
not match reported sex were also excluded.
FHS DNA methylation
Peripheral blood samples were collected at the eighth
examination samples (2005 to 2008). Genomic DNA was
extracted from buffy coat using the Gentra Puregene
DNA extraction kit (Qiagen) and bisulfite converted
using EZ DNA Methylation kit (Zymo Research
Corporation). DNA methylation quantification was conducted
in two laboratory batches. Methylation beta values were
generated using the Bioconductor minfi package with
background correction. Sample exclusion criteria
included poor SNP matching of control positions,
missing rate >1%, outliers from multi-dimensional scaling
(MDS), and sex mismatch. Probes were excluded if
missing rate >20%. In total, 2,635 samples and 443,304
CpG probes remained for analysis.
NAS DNA methylation
DNA was extracted from buffy coat using the QIAamp
DNA Blood Kit (QIAGEN, Valencia, CA, USA). A total
of 500 ng of DNA was used to perform bisulfite conversion
using the EZ-96 DNA Methylation Kit (Zymo Research,
Orange, CA, USA). To limit chip and plate effects, a
twostage age-stratified algorithm was used to randomize
samples and ensure similar age distributions across chips and
plates; we randomized 12 samples - which were sampled
across all the age quartiles - to each chip, then chips were
randomized to plates (each housing eight chips). Quality
control analysis was performed to remove samples where
>1% of probes had a detection P value >0.05. The
remaining samples were preprocessed using the
Illuminatype background correction without normalization as
reimplemented in the Bioconductor minfi package, which
was used to generate methylation beta values . All
485,512 CpG and CpH probes were in the working set.
BSGS DNA methylation
DNA was extracted from peripheral blood lymphocytes
by the salt precipitation method  from samples that
were time matched to sample collection of PAXgene
tubes for gene expression studies in the Brisbane
Systems Genetics Study . Bisulphite converted DNA
samples were hybridized to the 12 sample Illumina
HumanMethylation450 BeadChips using the Infinium
HD Methylation protocol and Tecan robotics (Illumina,
San Diego, CA, USA). Samples were randomly placed
with respect to the chip they were measured on and to
the position on that chip in order to avoid any
confounding with family. Box-plots of the red and green intensity
levels and their ratio were used to ensure that no chip
position was under- or over-exposed, with any outlying
samples repeated. Similarly, the proportion of probes with
detection P value less than 0.01 was examined to confirm
strong binding of the sample to the array. Raw intensity
values were background corrected using the Genome
Studio software performing normalization to internal controls
and background subtraction.
LBC mortality ascertainment
For both LBC1921 and LBC1936, mortality status was
obtained via data linkage from the National Health Service
Central Register, provided by the General Register Office
for Scotland (now National Records of Scotland).
Participant deaths and cause of death are routinely flagged to the
research team on approximately a 12-weekly basis.
FHS mortality ascertainment
Deaths that occurred prior to 1 January 2013 were
ascertained using multiple strategies, including routine
contact with participants for health history updates,
surveillance at the local hospital and in obituaries of
the local newspaper, and queries to the National Death
Index. We requested death certificates, hospital and
nursing home records prior to death, and autopsy
reports. When cause of death was undeterminable, the
next of kin were interviewed. The date and cause of
death were reviewed by an endpoint panel of three
NAS mortality ascertainment
Regular mailings to study participants have been used to
maintain vital-status information, and official death
certificates were obtained for decedents from the appropriate
state health department. Death certificates were reviewed
by a physician, and cause of death coded by an
experienced research nurse using ICD-9. Both participant deaths
and cause of death are routinely updated by the research
team and last update available was 31 December 2013.
Mortality-associated variables assessed in LBC1921 and
LBC1936 were used as covariates in the statistical
models: educational attainment, age-11 cognitive ability,
APOE e4 status (carriers versus non-carriers), smoking
status, and the presence or absence of diabetes, high
blood pressure, or cardiovascular disease. Age-11
cognitive ability (age-11 IQ) was measured in 1932 for
LBC1921 and in 1947 for LBC1936 using the Moray
House Test Number 12, described above. All other
variables were measured at the late-life baseline waves (age
79 years for LBC1921 and age 70 years for LBC1936).
APOE was genotyped from venous blood using PCR
amplification of a 227-bp fragment of the APOE gene,
which contains the two single nucleotide polymorphisms
that are used to define the e2, e3, and e4 alleles  in
LBC1921, and by TaqMan technology (Applied
Biosystems, Carlsbad, CA, USA) in LBC1936. Subjects were
then categorized by the presence or absence of the e4
allele. Social class was based on the most prestigious
occupation held by the participant prior to retirement. It was
grouped into five categories in LBC1921 and six
categories in LBC1936, where Class III was split into manual
and non-manual professions [43,44]. It was treated as a
continuous variable with lower values representing the
more prestigious classes. The other variables were
determined via self-report: number of years of education
(measured as a continuous variable), diabetes (yes/no),
high blood pressure (yes/no), cardiovascular disease (yes/
no), and categorical smoking status (current/ex-smoker,
Given the known influence of blood cell count on
methylation , we adjusted for five types of white
blood cell count (basophils, monocytes, lymphocytes,
eosinophils, and neutrophils) that were measured at on the
same blood that was analyzed for methylation. These
data were collected and processed the same day;
technical details are reported in McIllhagger et al. .
At the eighth in-person examination visit participants
completed a questionnaire that inquired about their
education, occupation, smoking status, and disease status.
Highest levels of educational attainment was assessed by
eight categories - no schooling, grades 1 to 8, grades 9
to 11, completed high school or GED, some college but
no degree, technical school certificate, associate degree,
Bachelors degree, graduate or professional degree.
Smoking status was dichotomized as current/past smokers and
those who reported to never have smoked. Diabetes was
defined as having fasting blood glucose 126 mg/dl or
current treatment for diabetes. Hypertension was defined
as having systolic blood pressure 140 mmHg, diastolic
blood pressure 90 mmHg, or current treatment for
hypertension. Cardiovascular disease was determined by a
panel of three physicians, who reviewed participants
medical records, laboratory findings, and clinic exam notes.
At each in-person examination visit, participants
completed a questionnaire that enquired about their smoking
status, education, diabetes (self-reported diagnosis and/or
use of diabetes medications), and diagnosis of coronary
heart disease (validated on medical records, ECG, and
physician exams). High blood pressure was defined as
antihypertensive medication use or SBP 140 mmHg or
DBP 90 mmHg at study visit. APOE-e4 allele status was
assessed through genotyping on a Sequenom MassArray
MALDI-TOF mass spectrometer.
Estimated naive T cell abundance
In LBC1921, LBC1936, FHS, and NAS, we considered
the abundance of defined different subtypes of T cells:
Naive T cells were defined as RA+ IL7 Receptor + cells.
Central Memory T cells = RA negative IL7 Receptor
positive Effector memory = RA negative IL7 Receptor
negative. To estimate the naive T cells in our cohort
studies, we used a prediction method that was developed
on an independent dataset. The predictor of T cell counts
(that is, naive CD4 T cell count) was found by applying a
penalized regression model (elastic net) to regress T cell
counts (dependent variable) on a subset of CpGs reported
in Supplemental Table 3 from Zhang et al. . By
applying this resulting penalized regression model to our data,
we arrived at predicted T cell counts.
LBC methylation data have been submitted to the European
Genome-phenome Archive under accession number
EGAS00001000910; phenotypic data are available at
dbGaP under the accession number phs000821.v1.p1.
The FHS and NAS data are available at dbGaP under
the accession numbers phs000724.v2.p9 phs000853.v1.
p1, respectively. BSGS methylation data are available
from the NCBI Gene Expression Omnibus under
accession number GSE56105.
Two measures of DNA methylation age (mage) were
calculated. The Horvath  mage uses 353 probes
common to the Illumina 27 K and 450 K Methylation arrays
using data from a range of tissues and cell types. The
Hannum  mage is based on 71 methylation probes
from the Illumina 450 K Methylation array derived as
the best predictors of age using data generated from
whole blood. Of the Hannum age predictor probes, 70,
71, and 71 were included in the LBC, NAS, and FHS
data, respectively. mage was calculated as the sum of the
beta values multiplied by the reported effect sizes for the
Hannum predictor. For the Horvath predictor, mage was
determined in all cohorts using the online calculator
(http://labs.genetics.ucla.edu/horvath/dnamage/). A third
predictor, based on the three probes highlighted in the
Weidner et al. paper , was also examined although,
due to its poorer predictive accuracy, it was not included
for the main analyses. To account for technical
variability in the measurement of the methylation CpGs in the
LBC studies, mage was adjusted for plate, array, position
on the chip, and hybridisation date (all treated as fixed
effect factors) using linear regression. In a sensitivity
analysis, additional adjustments were made for white
blood cell counts (the number of basophils, monocytes,
lymphocytes, eosinophils, and neutrophils per volume of
blood) or DNA methylation-estimated cell counts, as
described elsewhere . The residuals from these models
were added to the mean predicted methylation age to
give the new, adjusted measure of mage. The two
methylation age predictors contained six overlapping probes.
A methylation-based age acceleration index (age) was
calculated for all subjects, defined as the adjusted
methylation age in years minus chronological age at
sample collection in years (age = mage - chronological
Cox proportional hazards regression models were used
to test the association between the Horvath and Hannum
measures of age and mortality, adjusting for age at
sample collection, and sex. Cox models in FHS further
adjusted for laboratory batch (fixed effect) and used a robust
variance estimator to account for familial relatedness.
Hazard ratios for age were expressed per 5 years of
methylation age acceleration. Schoenfeld residuals were
examined to test the proportional hazards assumption.
Sensitivity analyses, also using Cox proportional hazards
regression, excluded deaths within the first 2 years of
follow-up to eliminate the potential influences of (fatal)
acute illness on the methylation measurements. Analyses
to account for possible confounders/mediators included
potential life-course predictors of mortality: age-11 IQ
(LBC only), education in years, social class (LBC only),
APOE e4 carrier status (LBC and NAS), smoking status,
and self-reported diabetes, high blood pressure, and
cardiovascular disease. A fully adjusted model was tested,
in which all variables were entered together.
Chronological age- and sex-adjusted linear regression models
were used to explore the relationship between age and
the additional covariates; for example, does methylation
age acceleration depend on smoking or diabetes?
The results from the individual cohorts were
metaanalyzed using the meta package in R . The
cohorts were weighted based on the standard errors of
the log hazard ratios. There was no evidence of cohort
heterogeneity in the primary Cox model analyses
according to the DerSimonian-Laird estimator of
between-study variance so fixed effects models were
All analyses were performed in the statistical
software R  with the Cox models utilizing the 'survival'
Finally, we calculated the heritability of age in the
BSGS cohort. As mage was a better predictor of
chronological age in the adult compared to adolescent samples,
the difference between methylation age and
chronological age was firstly standardized within generations
(parents and offspring). Regression models were fitted to
methylation age removing the effects of age and sex.
Additionally, the regression on the adolescent samples
included age2 to account for the non-linearity between
chronological and methylation age . The residuals
from these regressions were standardized to have a
variance of 1 before combining the generations. See
Additional file 8 for a graphical representation of the
For each probe, the Intra Class Correlation of age for
the various relative pairs was calculated using ANOVA
where MSB is the Mean Square Between pairs and MSW
is the Mean Square Within. The confidence intervals
were based on the number of pseudo-independent
relative pair for each relationship.
The heritability for each probe was estimated by
partitioning its variance into additive genetic (Va) and
environmental (Ve) component by fitting a linear mixed model of
where y is the vector of adjusted methylation age, a is
the additive genetic effects and e is the unique
environmental effects (residuals). The model was fitted using
Additional file 1: Contains a table with summary information for
the additional covariate data in the four cohorts.
Additional file 2: Contains a table with the white blood cell-adjusted
associations of Horvath and Hannum age with mortality.
Additional file 3: Presents the results from the analyses that
accounted for differences in naive T cell abundance. It contains a
table and cohort specific figures that assess the association between
naive T cell abundance and age acceleration, chronological age, and
mortality. Cox model output is also included to show the association
between methylation age acceleration and mortality after adjusting for
naive T cells.
Additional file 4: Contains a table of the associations of age (per
5 years) with mortality in LBC1921 and LBC1936 after individual
adjustment for covariates. The basic adjustment model controls for
age and sex. A separate Cox model adjusting for age, sex, and a single
covariate was analyzed along with a saturated model that included age,
sex, and all covariates together.
Additional file 5: Contains a figure with the meta-analysis results of
sex-stratified, age-adjusted models of age against mortality.
Additional file 6: Contains a figure with the meta-analysis results of
age- and sex-adjusted age against mortality with a 2-year time lag.
Additional file 7: Contains a table with the associations of age
with known mortality risk factors. Separate linear regression analyses
were performed for each covariate. All models adjusted for sex except for
NAS, which only had male participants. Analysis of FHS data was adjusted
for laboratory batch and family structure.
Additional file 8: Contains a figure illustrating the model fitting in
BSGS to create residuals from a regression of predicted age against
chronological age. A quadratic chronological age term was included in
the adolescent model along with chronological age and sex (red line). A
linear model adjusting for age and sex was included in the adult model
Contributed equally: REM, SS, AFM, BHC, EC, SH, AAB, DL, PMV, NRW, and IJD.
Supervised research: GWM, KLL, JS, SH, AAB, DL, PMV, NRW, IJD, and JMM.
Conceived and designed the experiments: GWM, JS, SH, AAB, DL, PMV, NRW,
and IJD. Performed the experiments: SEH, JG, LM, GWM, APF, MLM, MDF, and
AEJ. Performed the statistical analyses: REM, SS, AFM, BHC, EC, MLM, AEJ, and
SH. Analyzed the data: REM, SS, AFM, BHC, and EC. Contributed reagents/
materials/analysis tools: SEH, JG, AKH, PR, SRC, AP, LM, NGM, ACJ, KLL, JMM,
JMS, APF, MLM, RJ, KLL, MDF, and AEJ. Wrote the paper: REM, SS, AFM, BHC,
EC, JMM, SH, AAB, DL, PMV, NRW, and IJD. All authors read and approved the
We thank the cohort participants and team members who contributed to
these studies. This work was supported by numerous funding bodies.
Phenotype collection in the Lothian Birth Cohort 1921 was supported by the
UKs Biotechnology and Biological Sciences Research Council (BBSRC), The
Royal Society and The Chief Scientist Office of the Scottish Government.
Phenotype collection in the Lothian Birth Cohort 1936 was supported by
Age UK (The Disconnected Mind project). Methylation typing was supported
by the Centre for Cognitive Ageing and Cognitive Epidemiology (Pilot Fund
award), Age UK, The Wellcome Trust Institutional Strategic Support Fund, The
University of Edinburgh, and The University of Queensland. REM, SEH, SRC,
JMS, PMV, and IJD are members of the University of Edinburgh Centre for
Cognitive Ageing and Cognitive Epidemiology (CCACE). CCACE is supported
by funding from the BBSRC, the Medical Research Council (MRC), and the
University of Edinburgh as part of the cross-council Lifelong Health and
Wellbeing initiative (MR/K026992/1). Research reported in this publication
was supported by National Health and Medical Research Council (NHMRC)
project grants 613608, APP496667, APP1010374, and APP1046880. NHMRC
Fellowships to GWM, PMV, and NRW (613602) and Australia Research Council
(ARC) Future Fellowship to NRW (FT0991360). The content is solely the
responsibility of the authors and does not necessarily represent the official
views of the NHMRC or ARC. The Framingham Heart Study is funded by
National Institutes of Health contract N01-HC-25195. The laboratory work for
this investigation was funded by the Division of Intramural Research, National
Heart, Lung, and Blood Institute, National Institutes of Health. The analytical
component of this project was funded by the Division of Intramural Research,
National Heart, Lung, and Blood Institute, and the Center for Information
Technology, National Institutes of Health, Bethesda, MD, USA. This study
utilized the high-performance computational capabilities of the Biowulf
Linux cluster (http://biowulf.nih.gov) and Helix Systems (http://helix.nih.gov)
at the National Institutes of Health, Bethesda, MD, USA. JMM and KLL were
supported by R01AG029451. The present work on the US Department of
Veterans Affairs (VA) Normative Aging Study has been supported by funding
from the U.S. National Institute of Environmental Health Sciences (NIEHS)
(R01ES015172, R01ES021733). The VA Normative Aging Study is supported by
the Cooperative Studies Program/ERIC, US Department of Veterans Affairs, and
is a research component of the Massachusetts Veterans Epidemiology Research
and Information Center (MAVERIC). Additional support to the VA Normative
Aging Study was provided by the US Department of Agriculture, Agricultural
Research Service (contract 53-K06-510). The views expressed in this paper are
those of the authors and do not necessarily represent the views of the
US Department of Veterans Affairs. We thank Stuart J Ritchie for his
helpful comments and suggestions on the initial draft of the
1. Thornburg KL , Shannon J , Thuillier P , Turker MS . In utero life and epigenetic predisposition for disease . Adv Genet . 2010 ; 71 : 57 - 78 .
2. Bell JT , Tsai PC , Yang TP , Pidsley R , Nisbet J , Glass D , et al. Epigenome-wide scans identify differentially methylated regions for age and age-related phenotypes in a healthy ageing population . PLoS Genet . 2012 ; 8 : e1002629 .
3. Bollati V , Schwartz J , Wright R , Litonjua A , Tarantini L , Suh H , et al. Decline in genomic DNA methylation through aging in a cohort of elderly subjects . Mech Ageing Dev . 2009 ; 130 : 234 - 9 .
4. Christensen BC , Houseman EA , Marsit CJ , Zheng S , Wrensch MR , Wiemels JL , et al. Aging and environmental exposures alter tissue-specific DNA methylation dependent upon CpG island context . PLoS Genet . 2009 ; 5 : e1000602 .
5. Fraga MF , Esteller M. Epigenetics and aging: the targets and the marks . Trends Genet . 2007 ; 23 : 413 - 8 .
6. Rodriguez-Rodero S , Fernandez-Morera JL , Fernandez AF , Menendez-Torre E , Fraga MF. Epigenetic regulation of aging . Discov Med . 2010 ; 10 : 225 - 33 .
7. Mathers JC , Strathdee G , Relton CL . Induction of epigenetic alterations by dietary and other environmental factors . Adv Genetics . 2010 ; 71 : 3 - 39 .
8. Bell JT , Spector TD . DNA methylation studies using twins: what are they telling us? Genome Biol . 2012 ; 13 : 172 .
9. Kilpinen H , Dermitzakis ET . Genetic and epigenetic contribution to complex traits . Hum Mol Genet . 2012 ; 21 : R24 - 8 .
10. Hannum G , Guinney J , Zhao L , Zhang L , Hughes G , Sadda S , et al. Genomewide methylation profiles reveal quantitative views of human aging rates . Mol Cell . 2013 ; 49 : 359 - 67 .
11. Horvath S. DNA methylation age of human tissues and cell types . Genome Biol . 2013 ; 14 :R115.
12. Weidner CI , Lin Q , Koch CM , Eisele L , Beier F , Ziegler P , et al. Aging of blood can be tracked by DNA methylation changes at just three CpG sites . Genome Biol . 2014 ; 15 :R24.
13. Deary IJ , Gow AJ , Pattie A , Starr JM . Cohort profile: the Lothian Birth Cohorts of 1921 and 1936 . Int J Epidemiol . 2012 ; 41 : 1576 - 84 .
14. Deary IJ , Gow AJ , Taylor MD , Corley J , Brett C , Wilson V , et al. The Lothian Birth Cohort 1936 : a study to examine influences on cognitive ageing from age 11 to age 70 and beyond . BMC Geriatr . 2007 ; 7 : 28 .
15. Deary IJ , Whiteman MC , Starr JM , Whalley LJ , Fox HC . The impact of childhood intelligence on later life: following up the Scottish mental surveys of 1932 and 1947 . J Pers Soc Psychol . 2004 ; 86 : 130 - 47 .
16. Dawber TR , Meadors GF , Moore Jr FE . Epidemiological approaches to heart disease: the Framingham Study . Am J Public Health Nations Health . 1951 ; 41 : 279 - 81 .
17. Kannel WB , Feinleib M , McNamara PM , Garrison RJ , Castelli WP . An investigation of coronary heart disease in families. The Framingham offspring study . Am J Epid . 1979 ; 110 : 281 - 90 .
18. Bell B , Rose CL , Damon A. The Veterans Administration longitudinal study of healthy aging . The Gerontologist . 1966 ; 6 : 179 - 84 .
19. Spiro 3rd A, Vokonas P. The VA Normative Aging Study . In: Markides KS, editor. Encyclopedia of health and aging . Thousand Oaks, CA: Sage; 2007 . p. 421 - 2 .
20. Powell JE , Henders AK , McRae AF , Caracella A , Smith S , Wright MJ , et al. The Brisbane Systems Genetics Study: genetical genomics meets complex trait genetics . PLoS One . 2012 ; 7 : e35430 .
21. Houseman EA , Accomando WP , Koestler DC , Christensen BC , Marsit CJ , Nelson HH , et al. DNA methylation arrays as surrogate measures of cell mixture distribution . BMC Bioinformatics . 2012 ; 13 : 86 .
22. Jaffe AE , Irizarry RA . Accounting for cellular heterogeneity is critical in epigenome-wide association studies . Genome Biol . 2014 ; 15 :R31.
23. Larbi A , Franceschi C , Mazzatti D , Solana R , Wikby A , Pawelec G. Aging of the immune system as a prognostic factor for human longevity . Physiology (Bethesda) . 2008 ; 23 : 64 - 74 .
24. Fagnoni FF , Vescovini R , Passeri G , Bologna G , Pedrazzoni M , Lavagetto G , et al. Shortage of circulating naive CD8(+) T cells provides new insights on immunodeficiency in aging . Blood . 2000 ; 95 : 2860 - 8 .
25. Biessels GJ , Strachan MW , Visseren FLJ , Kappelle LJ , Whitmer RA . Dementia and cognitive decline in type 2 diabetes and prediabetic stages: towards targeted interventions . Lancet Diabetes Endocrinol . 2014 ; 2 : 246 - 55 .
26. Qiu C , Winblad B , Fratiglioni L. The age-dependent relation of blood pressure to cognitive function and dementia . Lancet Neurol . 2005 ; 4 : 487 - 99 .
27. World Health Organization. Global Health Risks: Mortality and burden of disease attributable to selected major risks . Geneva: World Health Organization ; 2009 .
28. Deary IJ , Weiss A , Batty GD . Intelligence and personality as predictors of illness and death: How researchers in differential psychology and chronic disease epidemiology are collaborating to understand and address health inequalities . Psychol Sci Publ Interest . 2010 ; 11 : 53 - 79 .
29. Schchter F , Faure-Delanef L , Gunot F , Rouger H , Froguel P , Lesueur-Ginot L , et al. Genetic associations with human longevity at the APOE and ACE loci . Nat Genet . 1994 ; 6 : 29 - 32 .
30. Doll R , Peto R , Boreham J , Sutherland I. Mortality in relation to smoking: 50 years' observations on male British doctors . BMJ . 2004 ; 328 : 1519 .
31. Batty GD , Kivimaki M , Deary IJ . Intelligence, education, and mortality. BMJ . 2010 ; 340 :c563.
32. Calvin CM , Deary IJ , Fenton C , Roberts BA , Der G , Leckenby N , et al. Intelligence in youth and all-cause-mortality: systematic review with meta-analysis . Int J Epidemiol . 2011 ; 40 : 626 - 44 .
33. Kern ML , Friedman HS . Do conscientious individuals live longer? A quantitative review . Health Psychol . 2008 ; 27 : 505 - 12 .
34. Bendix L , Thinggaard M , Fenger M , Kolvraa S , Avlund K , Linneberg A , et al. Longitudinal changes in leukocyte telomere length and mortality in humans . J Gerontol A Biol Sci Med Sci . 2014 ; 69 : 231 - 9 .
35. Cawthon RM , Smith KR, O'Brien E , Sivatchenko A , Kerber RA . Association between telomere length in blood and mortality in people aged 60 years or older . Lancet . 2003 ; 361 : 393 - 5 .
36. Scottish Council for Research in Education. The Intelligence of Scottish Children: a National Survey of an Age-Group . London: University of London Press; 1933 .
37. Scottish Council for Research in Education. The Trend of Scottish Intelligence. London: University of London Press; 1949 .
38. Bibikova M , Barnes B , Tsan C , Ho V , Klotzle B , Le JM , et al. High density DNA methylation array with single CpG site resolution . Genomics . 2011 ; 98 : 288 - 95 .
39. Aryee MJ , Jaffe AE , Corrada-Bravo H , Ladd-Acosta C , Feinberg AP , Hansen KD , et al. Minfi: A flexible and comprehensive Bioconductor package for the analysis of Infinium DNA Methylation microarrays . Bioinformatics . 2014 ; 30 : 1363 - 9 .
40. Schalkwyk LC , Pidsley R , Wong CCY . wateRmelon: Illumina 450 methylation array normalization and metrics . R package version 1 .2.2. 2013 . Accessed September 2013 .
41. Miller SA , Dykes DD , Polesky HF . A simple salting out procedure for extracting DNA from human nucleated cells . Nucleic Acids Res . 1988 ; 16 : 1215 .
42. Wenham PR , Price WH , Blandell G . Apolipoprotein E genotyping by one-stage PCR . Lancet. 1991 ; 337 : 1158 - 9 .
43. Office of Population Censuses and Surveys . Classification of Occupations 1980 . London: HMSO ; 1980 .
44. General Register Office. Census 1951 : Classification of occupations . London: HMSO ; 1956 .
45. McIlhagger R , Gow AJ , Brett CE , Corley J , Taylor M , Deary IJ , et al. Differences in the haematological profile of healthy 70 year old men and women: normal ranges with confirmatory factor analysis . BMC Blood Disord . 2010 ; 10 :4.
46. Zhang Y , Maksimovic J , Naselli G , Qian J , Chopin M , Blewitt ME , et al. Genome-wide DNA methylation analysis identifies hypomethylated genes regulated by FOXP3 in human regulatory T cells . Blood . 2013 ; 122 : 2823 - 36 .
47. Schwarzer G . meta: Meta-Analysis with R . R package version 2.16. 2014 . Accessed January 2014 .
48. R Core Team . R: A Language and Environment for Statistical Computing . 2013 . Accessed September 2013 .
49. Therneau TM . A Package for Survival Analysis in S. R package version 2.37-7 . 2014 . Accessed January 2014 .
50. Abecasis GR , Cardon LR , Cookson WO. A general test of association for quantitative traits in nuclear families . Am J Hum Genet . 2000 ; 66 : 279 - 92 .