Gene-specific DNA methylation profiles and LINE-1 hypomethylation are associated with myocardial infarction risk
Guarrera et al. Clinical Epigenetics
Gene-specific DNA methylation profiles and LINE-1 hypomethylation are associated with myocardial infarction risk
Simonetta Guarrera 0 1 2
Giovanni Fiorito 0 1 2
N. Charlotte Onland-Moret 3
Alessia Russo 1 2
Claudia Agnoli 7
Alessandra Allione 1 2
Cornelia Di Gaetano 1 2
Amalia Mattiello 6
Fulvio Ricceri 5
Paolo Chiodini 4
Silvia Polidoro 2
Graziella Frasca 9
Monique W. M. Verschuren 3 8
Jolanda M. A. Boer 8
Yvonne T. van der Schouw 3
Rosario Tumino 9
Paolo Vineis 2
Vittorio Krogh 7
Salvatore Panico 6
Carlotta Sacerdote 5
Giuseppe Matullo 1 2
0 Equal contributors
1 Medical Sciences Department, University of Turin , Turin , Italy
2 Human Genetics Foundation (HuGeF) , Via Nizza 52, Turin I-10126Torino , Italy
3 Julius Center for Health Sciences and Primary Care, UMC Utrecht , Utrecht , The Netherlands
4 Department of Public Health, Second University of Naples , Naples , Italy
5 Cancer Epidemiology, CPO Piemonte , Turin , Italy
6 Department of Clinical and Experimental Medicine, Federico II University , Naples , Italy
7 Epidemiology and Prevention Unit, Fondazione IRCCS Istituto Nazionale Tumori , Milan , Italy
8 Centre for Nutrition, Prevention and Health Services, National Institute for Public Health and the Environment
9 Cancer Registry and Histopathology Unit, “Civile-M.P. Arezzo” Hospital , ASP 7, Ragusa , Italy
Background: DNA methylation profiles are responsive to environmental stimuli and metabolic shifts. This makes DNA methylation a potential biomarker of environmental-related and lifestyle-driven diseases of adulthood. Therefore, we investigated if white blood cells' (WBCs) DNA methylation profiles are associated with myocardial infarction (MI) occurrence. Whole-genome DNA methylation was investigated by microarray analysis in 292 MI cases and 292 matched controls from the large prospective Italian European Prospective Investigation into Cancer and Nutrition (EPIC) cohort (EPICOR study). Significant signals (false discovery rate (FDR) adjusted P < 0.05) were replicated by mass spectrometry in 317 MI cases and 262 controls from the Dutch EPIC cohort (EPIC-NL). Long interspersed nuclear element-1 (LINE-1) methylation profiles were also evaluated in both groups. Results: A differentially methylated region (DMR) within the zinc finger and BTB domain-containing protein 12 (ZBTB12) gene body and LINE-1 hypomethylation were identified in EPICOR MI cases and replicated in the EPIC-NL sample (ZBTB12-DMR meta-analysis: effect size ± se = −0.016 ± 0.003, 95 % CI = −0.021;−0.011, P = 7.54 × 10−10; LINE-1 methylation meta-analysis: effect size ± se = −0.161 ± 0.040, 95 % CI = −0.239;−0.082, P = 6.01 × 10−5). Moreover, cases with shorter time to disease had more pronounced ZBTB12-DMR hypomethylation (meta-analysis, men: effect size ± se = −0.0059 ± 0.0017, PTREND = 5.0 × 10−4; women: effect size ± se = −0.0053 ± 0.0019, PTREND = 6.5 × 10−3) and LINE-1 hypomethylation (meta-analysis, men: effect size ± se = −0.0010 ± 0.0003, PTREND = 1.6 × 10−3; women: effect size ± se = −0.0008 ± 0.0004, PTREND = 0.026) than MI cases with longer time to disease. In the EPIC-NL replication panel, DNA methylation profiles improved case-control discrimination and reclassification when compared with traditional MI risk factors only (net reclassification improvement (95 % CI) between 0.23 (0.02-0.43), P = 0.034, and 0.89 (0.64-1.14), P < 1 × 10−5). (Continued on next page)
(Continued from previous page)
Conclusions: Our data suggest that specific methylation profiles can be detected in WBCs, in a preclinical
condition, several years before the occurrence of MI, providing an independent signature of cardiovascular
risk. We showed that prediction accuracy can be improved when DNA methylation is taken into account
together with traditional MI risk factors, although further confirmation on a larger sample is warranted. Our
findings support the potential use of DNA methylation patterns in peripheral blood white cells as promising
early biomarkers of MI.
Cardiovascular diseases (CVDs) are a leading cause of
mortality, morbidity, and hospitalization in the adult
population in Western countries and a major challenge
for developing countries that follow a
Westernizedlifestyle. Great attention has been given so far to
lifestylerelated CVD risk factors, such as unhealthy diet, smoking
habits, and lack of physical activity, whose deleterious
effects may be prevented through major lifestyle changes
or medical treatments. Apart from monogenic disorders
associated with cardiovascular risk (e.g., hypertrophic
cardiomyopathy, familial hypercholesterolemia), there is a
strong evidence that a family history of CVD and stroke
enhances individual CVD risks in relatives as compared
with a general population that points out the importance
of genetic factors in the etiology of CVDs.
Recent genome-wide association studies (GWASs)
reported several potential genetic risk factors for CVDs or
intermediate disease phenotypes such as type 2 diabetes,
obesity and overweight [
], hypertension [
], and altered
lipid profiles [
], underlying the importance of the
genetic component. However, the contribution of common
genetic variants to non-monogenic CVDs is likely to act
in combination with environmental factors or via epistatic
(gene-gene or gene-environment) interactions. As
geneenvironment interactions are thought to be mediated by
epigenetic modifications of the genome, epigenetic
regulation can be rewarded as the boundary between the
inherited genomic asset and the environment, potentially
playing a major role in disease onset and severity [
epigenetic changes are in fact dynamic, can be modified both
during the early in utero development stages and across
lifetime by environmental factors as well as diseases, and
may be reversible reflecting environmental changes [
DNA methylation at CpG dinucleotides is an epigenetic
mechanism mainly involved in gene expression regulation.
DNA methylation patterns across the genome are not
uniform: genetic regions spanning gene locations have
variable DNA methylation profiles which are linked to
regulatory functions (e.g., gene promoter
methylation/demethylation regulates gene expression) and structural
functions in shaping local chromatin structures [
instead, intergenic regions are usually heavily methylated,
since about 45 % of the mammalian genome consists of
transposable and viral elements that are silenced by
]. Methylation levels of the repetitive long
interspersed nuclear element-1 (LINE-1) are generally
considered as a proxy for global DNA methylation, as
LINE-1 elements are widely distributed in the genome
and usually heavily methylated in the majority of normal
tissues. LINE-1 hypomethylation has previously been
associated with ischemic heart disease and stroke [
with altered levels of LDL and HDL [
Altered DNA methylation profiles have been linked to
oxidative stress [
], atherosclerosis [
], ageing [
and a variety of human diseases ranging from
neurological and autoimmune disorders to cancer [
addition to individual constitutive DNA methylation
profiles that could per se be associated with
cardiovascular outcomes , subtle and progressive DNA
methylation alterations mediated by lifestyle and environmental
exposures may in fact lead to dysregulation of several
metabolic pathways during lifetime and ultimately to
cardiovascular damage and disease [
]. However, the
few reports linking cardiovascular outcomes to DNA
methylation measured in blood cells or vascular tissue
] did not provide conclusive evidences of DNA
methylation involvement in CVD.
Apart from few reports of single CpG associations with a
disease or a phenotype, it is usually the cumulative
methylation profile of neighboring CpG sites to be more likely
associated to a potential functional effect of the methylation
status, and the search for differentially methylated regions
(DMRs) able to differentiate groups of subjects with
different phenotypes or outcomes of interest is a common
approach. Along this line, we conducted an epigenome-wide
association study (EWAS) to identify DMRs and LINE-1
methylation profiles associated to myocardial infarction
(MI) risk in the cardiovascular section (EPICOR) of the
Italian cohort of the European Prospective Investigation
into Cancer and Nutrition (EPIC) study and replicated
statistically significant findings in an independent
casecontrol study nested in the Dutch EPIC cohort (EPIC-NL)
with comparable biological samples and information.
Furthermore, we tested whether MI risk prediction
accuracy can be improved when DNA methylation profiles,
measured at baseline in a preclinical condition, are taken
into account together with traditional MI risk factors.
Descriptive statistics of the sample are reported in Table 1;
details on EPIC cohorts are provided in Additional file 1.
Statistically significant differences between cases and
controls were found in smoking habits, body mass index
(BMI) and/or waist-to-hip ratio (WHR), serum lipid
profile, and blood pressure, in both the discovery (EPICOR)
and the replication (EPIC-NL) studies (Table 1).
After raw methylation data quality controls (QCs), and
removal of cross-hybridizing and single-nucleotide
polymorphism-containing probes, 425,498 CpGs were
included into the following analyses.
Case-control differential methylation
In the EPICOR sample, 25,376 regions with correlated
methylation levels were identified with the A-clustering
] and subsequently tested for differential
methylation between cases and controls (see the “Methods”
section): the top-ranking 6 DMRs are reported in Table S1
(Additional file 2). However, only the first region reached
statistical significance (false discovery rate (FDR) Q < 0.05),
i.e., a 15-CpGs cluster within the gene body (exon 1) of the
zinc finger and BTB domain-containing protein 12 gene
(ZBTB12, gene ID: 221527) that was hypomethylated in
cases as compared to controls (effect size ± se = −0.019 ±
0.004, 95 % CI −0.03;−0.01, P = 1.94 × 10−7, Q = 0.005). To
check for sex-specific effects of the ZBTB12-DMR, we
stratified EPICOR subjects by sex and found the 15-CpGs
cluster still significantly hypomethylated in male cases
(effect size ± se = −0.023 ± 0.005, 95 % CI −0.03;−0.01, P =
1.06 × 10−6), but not in females (effect size ± se = −0.006 ±
0.006, 95 % CI −0.02;0.005, P = 0.29). Details on single
CpGs are reported in Table S2A (Additional file 2).
The genomic inflation factor for the overall EPICOR
sample was lambda = 1.023 (men, lambda = 1.043;
women, lambda = 1.017; Q-Q plots in Additional file 3:
LINE-1 differential methylation was also tested in the
EPICOR overall sample by logistic regression analysis: MI
cases had statistically significant LINE-1 hypomethylation
as compared to controls (effect size ± se = −0.511 ± 0.147,
95 % CI −0.80;−0.22, P = 5.00 × 10−4). At a sex-stratified
analysis, LINE-1 hypomethylation was still statistically
significant in men (effect size ± se = −0.520 ± 0.179,
95 % CI −0.87;−0.17, P = 0.004), but not in women (effect
size ± se = −0.496 ± 0.319, 95 % CI −1.12;−0.13, P = 0.12).
Additionally, for ZBTB12-DMR, we found a significant
sex-methylation interaction (P = 0.01), while for LINE-1
we found no evidence of interaction.
Results were replicated on the EPIC-NL panel, where
the methylation profile of the same ZBTB12-DMR
identified in the discovery phase proved consistent with that of
the EPICOR discovery sample, with a cluster of 22
contiguous CpGs significantly hypomethylated in Dutch MI
cases as compared to controls (effect size ± se = −0.013 ±
0.004, 95 % CI −0.02;−0.005 P = 5.82 × 10−4). Details on
ZBTB12-DMR single CpGs for the EPIC-NL study are
reported in Table S2B (Additional file 2).
At a sex-stratified analysis, ZBTB12-DMR was
hypomethylated in both EPIC-NL men (effect size ± se = −0.014
± 0.007, 95 % CI −0.03;−0.001, P = 0.034) and women
(effect size ± se = −0.012 ± 0.004, 95 % CI −0.02;−0.004,
P = 0.006), with effect sizes more comparable between men
and women than in the EPICOR sample.
In the EPIC-NL panel, LINE-1 mean methylation
levels were lower than those of EPICOR, with an average
methylation of about 0.8 in EPICOR subjects (men:
mean ± sd = 0.844 ± 0.007; women: mean ± sd = 0.843 ±
0.007) and about 0.6 in EPIC-NL subjects (men: mean ±
sd = 0.624 ± 0.029; women: mean ± sd = 0.613 ± 0.023).
As seen in the EPICOR panel, we found LINE-1
hypomethylation also in Dutch cases as compared to controls,
although with a milder effect (effect size ± se = −0.132 ±
0.042, 95 % CI −0.21;−0.05, P = 0.001). In EPIC-NL men,
the sex-stratified LINE-1 analysis showed an effect size
similar to that found in EPICOR (effect size ± se = −0.40 ±
0.085, 95 % CI −0.57;−0.23, P = 2.22 × 10−6), while in
EPIC-NL women the effect was much lower and
statistically non-significant (effect size ± se = −0.016 ± 0.046,
95 % CI −0.11;0.07, P = 0.73).
In the Dutch panel, we found no evidence of
sexmethylation interaction for ZBTB12-DMR, while we
found a statistically significant interaction for
LINE1 (P = 0.0003).
The observation of sex-methylation interactions in both
the discovery and replica panels, and further
considerations addressed in the “Discussion” section, suggested to
consider men and women separately in all the subsequent
To achieve an overall estimate of the effects of
ZBTB12-DMR and LINE-1 methylation across the two
subject panels, we performed a meta-analysis of the
EPICOR and EPIC-NL studies.
The estimated ZBTB12-DMR effects were effect size ±
se = −0.016 ± 0.003 in the overall sample (P = 7.54 × 10−10,
95 % CI = −0.021;−0.011, Cochran’s Q = 0.005, d.f. = 1,
PHET = 0.83), effect size ± se = −0.020 ± 0.004 in men (P =
1.82 × 10−7, 95 % CI = −0.027;−0.012, Cochran’s Q = 0.007,
d.f. = 1, PHET = 0.79), and effect size ± se = −0.010 ± 0.003
in women (P = 0.005, 95 % CI = −0.017;−0.003, Cochran’s
Q = 0.004, d.f. = 1, PHET = 0.84).
The estimated LINE-1 effects were effect size ± se =
−0.161 ± 0.040 in the overall sample (P = 6.01 × 10−5,
95 % CI = −0.239;−0.082, Cochran’s Q = 0.85, d.f. = 1,
PHET = 0.35), effect size ± se = −0.422 ± 0.076 in men
(P = 3.42 × 10−8, 95 % CI = −0.572;−0.272, Cochran’s Q =
0.06, d.f. = 1, PHET = 0.81), and effect size ± se = −0.025 ±
0.046 in women (P = 0.576, 95 % CI = −0.115;0.064,
Cochran’s Q = 0.70, d.f. = 1, PHET = 0.40).
DNA methylation and MI risk
The MI risk associated to ZBTB12-DMR and LINE-1
hypomethylation was estimated in the EPIC-NL replica
panel: recursively partitioned mixture model (RPMM)
classes and LINE-1 class (as defined in the “Methods”
section) were tested for association with MI under
different models, from unadjusted to fully adjusted.
When comparing the ZBTB12-DMR lowest
methylation class (RPMM3) with the highest methylation class
(RPMM0), we found MI risk to be significantly
associated with hypomethylation in the EPIC-NL women (fully
adjusted, OR = 2.75, 95 % CI 1.39–5.45, P = 0.004), while
in EPIC-NL men the association was statistically
nonsignificant (fully adjusted, OR = 2.60, 95 % CI 0.79–8.56,
P = 0.116), although direction and effect size were similar.
We also found a higher MI risk associated with LINE-1
lower methylation class in EPIC-NL men (fully adjusted,
OR = 1.95, 95 % CI 1.02–3.71, P = 0.043, ref. group above
the median). No difference was found in EPIC-NL women
(fully adjusted, OR = 1.05, 95 % CI 0.65–1.67, P = 0.850)
(Additional file 2: Table S3A).
The same analysis was performed on the EPICOR
discovery sample: even though in this case the ORs cannot
be considered as indicative of a true estimate of risk
being EPICOR subjects the discovery panel, the analysis
was nevertheless done to assess whether the progressive
inclusion in the model of additional variables, namely
traditional risk factors (TRFs), could modify the estimate
of risk or, on the contrary, if DNA methylation may
independently contribute to MI risk. No significant
evidence of inflation/deflation of the DNA
methylationrelated MI risk estimate was found nor for the EPIC-NL
panel nor for the EPICOR panel when progressively
adding TRFs as covariates in the model (Additional file 2:
Tables S3A and B).
Discrimination, reclassification, and calibration on EPIC-NL samples
We assumed two models, including the following: (1)
TRFs only and (2) TRFs plus the ZBTB12-based RPMM
classes and LINE-1 methylation class. According to the
net reclassification improvement (NRI) and integrated
discrimination improvement (IDI) indices (Table 2), a
statistically significant improvement in prediction
performance was achieved when adding the DNA
methylation profiles to the set of baseline predictors (i.e., TRFs),
both for EPIC-NL male and female groups. Furthermore,
we found an improvement in discrimination (Table 2,
DeLong’s test) comparing the area under the receiver
operating curves (AUC) of the two models (Table 2 and
Fig. 1), although it was not statistically significant.
The calibration plots confirmed the goodness of fit of
both the TRFs only and TRFs + Methylation models
(Fig. 2, Hosmer-Lemeshow test), with a better
performance of the second one.
DNA methylation and time to disease
The trend test on EPICOR and EPIC-NL subjects,
stratified by study and by sex, highlighted a more pronounced
ZBTB12-DMR hypomethylation in cases with shorter
time to disease (EPICOR and EPIC-NL meta-analysis,
men: PTREND = 0.0005, women PTREND = 0.0065; Table 3).
Similarly, LINE-1 was hypomethylated in cases with
shorter time to disease (meta-analysis, men: PTREND =
0.0016, women PTREND = 0.026; Table 3).
At a post hoc power analysis, our study was well
powered (86 and 82 % for male and female groups,
respectively) to identify DMRs with effect sizes equal to half of
the standard deviation, considering alpha equal to the
FDR threshold of significance (Q = 0.05).
In this study, we investigated whether white blood cells’
(WBCs) DNA methylation profiles may be associated
with MI risk. We examined clusters of adjacent CpG
sites with correlated methylation levels under the
assumption that they could be more reliable indicators of
the underlying biological function than the single CpG
methylation measurement. As we found evidences of
sex-methylation interactions in both the analyzed panels,
in our study the analyses were stratified by sex, in order
to account for sex-related differences in DNA
methylation profiles of genomic regions, of which “genomic
imprinting” is a well-known example, and to account for
sex-specific cardiovascular risks. For coronary heart
disease, sex differences in incidence, disease manifestations,
and mortality are well recognized [
], and men and
women seem not to share the same cardiovascular risk
]. Moreover, patterns of sex-specific
methylation have been reported in literature, and there
is a general consensus on the occurrence of sex-biased
autosomal DNA methylation in specific genes and
regions, although with contrasting results [
Sexassociated differential DNA methylation in autosomal
loci has been reported in genes associated to
traits/diseases with different incidence rates according to sex
, as well as in hormone-related genes, suggesting a
differential regulation, potentially exerted via
]. Differential DNA methylation may account for
the differences in metabolic profiles of men and women,
possibly leading to the different incidence, prevalence,
symptoms, ages at onset, and severity of CVDs reported
In the EPICOR discovery panel, we identified a
15CpGs cluster within the ZBTB12 gene that was
significantly differentially methylated in Italian MI cases and
controls and that was also significantly hypomethylated
in MI cases in the independent Dutch panel. Moreover,
ZBTB12-DMR showed a trend towards more
pronounced hypomethylation in subjects with a short time
to disease (TTD) both in the Italian and in the Dutch
ZBTB12-DMR spans a ~250-bp region in ZBTB12
exon 1: although the role of gene-body methylation in
transcriptional regulation is not fully understood, yet
there are evidences of a role of the first exon’s DNA
methylation in transcriptional silencing and, putatively,
in alternative splicing [
]. All of our samples belong to
the EPIC cohort, for which no biospecimen suitable for
transcriptome analyses is available to address the
relationship between ZBTB12 methylation and gene
expression levels. To cope with this issue, we explored ZBTB12
DNA methylation/gene expression relationship in
cryopreserved peripheral blood mononuclear cells from ~80
healthy young subjects belonging to another ongoing
study, for which we already measured methylation and
gene expression levels: in our data, ZBTB12 messenger
aHealthy controls (TTD class 0) were used as the reference group. Cases were divided in tertiles (TTD classes 1 to 3)
bMinimum and maximum TTD (i.e., time lapse in years from enrollment to occurrence of MI) for each class
cd.f. = 1, P = not significant
RNA (mRNA) abundance was below the background
level (as assessed by Illumina HumanHT12 gene
expression BeadChip), while ZBTB12 methylation levels were
comparable to that of EPICOR and EPIC-NL controls
(data not shown). No relationship was found also with
the gene expression levels of the nearby genes (data not
shown). Data mining in freely available resources (e.g.,
BioGPS, AceView, ProteinAtlas, Genome Atlas) confirmed
the generalized low ZBTB12 mRNA level in tissues and
cell types, although ZBTB12 protein is detectable in many
tissues, including cardiovascular tissues. Although no clear
function is described for ZBTB12, this protein is probably
involved in transcriptional regulation, like other members
of the ZBTB family of methyl-CpG binding proteins
(MBPs). This is also supported by its mainly nuclear
localization. MBPs bind to methylated DNA and recruit
chromatin remodeling co-repressor complexes, resulting
in compaction of chromatin into its transcriptionally
inactive state [
]. Specifically, members of the ZBTB family
function as mediators of epigenetically controlled gene
silencing by recognizing symmetrically methylated CpG
sites and sequence-specific non-methylated sites [
According to the Human Protein Reference Database
], ZBTB12 (HPRD ID: 15691) directly interacts with
Harvey rat sarcoma viral oncogene homolog (HRAS)
and Ras-associated protein 1 (RAP1) GTPase-activating
protein 1 (RAP1GAP). RAP1GAP downregulates the
activity of RAP1, a small GTPase involved in several
aspects of cell adhesion, including angiogenesis [
HRAS, a member of the RAS oncogene family, is a key
transducer in several growth-signaling events that may
trigger cardiovascular complications such as
angiogenesis and vascular permeability [
] and may be involved
in inflammatory proliferative arterial diseases, including
atherosclerosis and restenosis after angioplasty [
RAS-MEK-ERK cascade has been described as implicated
in cardiac hypertrophy and heart failure, and ERK signal
transduction pathways were associated with cardiac
In addition to gene-/region-specific DNA methylation,
we investigated the cumulative DNA methylation profile
of LINE-1 repetitive sequences and found LINE-1
hypomethylation in MI cases, statistically significant in men
in both panels, but not in women. LINE-1
hypomethylation was associated to cardiovascular-related traits in
previous studies [
], and it is associated to MI and
shorter TTD in the present study. DNA
hypomethylation is regarded as a cause of genomic instability, and as
a matter of fact, LINE-1 hypomethylation was found in
several conditions, including cancer , autoimmune
], and CVDs [
]. Specifically, global
hypomethylation of genomic DNA and gene-specific methylation
profiles have been associated to conditions already known
to predispose to CVDs, such as cellular ageing [
atherosclerotic plaques [
], menopausal state, and
]. On the other hand, LINE-1 hypomethylation
could simply be a marker of increased WBC proliferation
due to inflammatory or immunological responses which
are known to be active during cardiovascular pathogenic
]. In vitro experiments on mouse embryonic
stem cells showed that folate deficiency affected the
homeostasis of folate-mediated one-carbon metabolism,
leading to reduced LINE-1 methylation [
]. In a targeted
analysis, we recently demonstrated on a subset of the
EPICOR cohort (206 MI cases and 206 matched controls) an
inverse relationship between B vitamin intake and DNA
methylation of genes belonging to one-carbon metabolism
and homocysteine pathways [
]. These previous
observations, together with our current finding of LINE-1
hypomethylation in cases compared to healthy controls,
suggest a link between DNA methylation patterns and
CVD risk conferred by low folate and B vitamin intake
that is worthy of further investigation.
Overall, this study analyzed 609 cases and 554 controls
and was sufficiently powered to detect effects of the
magnitude we found. The discovery and the replica panels
share homogeneous features: both belong to the European
EPIC cohort, subjects were all enrolled in the 1990s, and
biosamples were collected and stored at enrollment
according to shared standard protocols [
]. Nevertheless, a
limitation of the study is that while EPICOR cases and
controls were matched by age, sex, center, and season of
recruitment, this could not be achieved for the EPIC-NL
sample, since a DNA sample suitable for methylation
analysis was not available for all the subjects enrolled in
Another limitation is that the assessment of the
methylation levels was done with different methods for
the two panels. However, our approach that considered
the regional methylation profile as a whole instead of
single CpGs may contribute to overcome the bias due to
measure errors at the single CpG level, as highlighted
by the correlation between the methylation measures
of 16 control samples assayed with both BeadChip
and MassARRAY assays (Additional file 1). Moreover,
although the CpG positions assayed with the two methods
are not exactly the same due to technical constrains
(Additional file 1 and Additional file 3: Figure S4),
still the analysis of methylation data collected with
each one of the two different techniques highlighted a
cluster of CpGs with correlated methylation levels within
exon 1 of ZBTB12, hypomethylated in MI cases vs
controls. This complies with our study design assumption that
the methylation status of multiple CpGs with correlated
methylation could better describe the cumulative
methylation status of the underlying region and that this could be
potentially related to the underlying biological function, if
any. The same goes for LINE-1 methylation, which is
defined as the cumulative DNA methylation status of the
several CpGs located in LINE-1 sequences across the
genome. Also in this case, different portions of LINE-1
sequence were investigated with the two techniques, i.e.,
CpGs scattered across the whole LINE-1 sequence were
analyzed on the BeadChip, whereas CpGs within base
pairs 335–767 of the LINE-1 promoter (GenBank
accession number X58075.1) were analyzed by MassARRAY
according to Wang et al. [
] (Additional file 3: Figure S4).
Despite slight differences between EPICOR and
EPICNL panels in LINE-1 average methylation levels,
arguably due to the use of different methods and different
assayed CpGs, our results highlighted the same effect
trend in both the EPIC sub-cohorts.
When included in the same multivariate models, the
estimated risks associated to ZBTB12-DMR and LINE-1
methylation profiles were not attenuated by the
adjustment for known risk factors (Additional file 2: Table
S3A and B), suggesting that they independently
contribute to MI risk estimate.
Moreover, we observed that discrimination between
MI cases and controls and prediction accuracy both
improved when DNA methylation was taken into account
together with TRFs, suggesting the DNA methylation
could be an independent predictor of MI risk, although
further confirmation on a larger sample is warranted.
Our results highlight the possibility to identify
MIrelated methylation marks on DNA from blood samples
drawn in a preclinical condition, for some subjects many
years before the MI. Unfortunately, due to the initial
EPIC study design that envisaged only one blood
sampling at enrollment time, it was not possible to monitor
individual DNA methylation level changes at different
time points. Further replication in additional cohorts with
prospective design and biospecimens sampled at multiple
points along time is warranted to elucidate DNA
methylation changes across time, from a “healthy” status to MI.
This will allow a better estimation of the ZBTB12-DMR
and LINE1 demethylation rates associated with increased
MI risk, in the view of a personalized risk assessment that
will take into account TRFs and MI risk biomarkers, such
as DNA methylation profiles.
To the best of our knowledge, this is the first paper
reporting an association between MI risk and DNA
methylation profiles identified from epigenome-wide
data in prospectively collected subjects with well-recorded
clinical endpoints and replicated in an independent
sample form the same large European prospective
Taken together, the reported results suggest the
possible role of DNA methylation patterns in peripheral
blood white cells as promising early MI biomarkers to
be potentially used, together with TRFs, for individual
MI risk assessment.
For the discovery phase, 292 MI cases and 292 matched
healthy controls were recruited among those enrolled in
the EPICOR study [
], a case-cohort study nested
within the EPIC-Italy prospective cohort (~50,000
]. All EPICOR cases developed MI after
recruitment (average time to diagnosis 6.90 years). Cases
were identified at cohort follow-up from hospital
discharge databases and were then matched with healthy
controls from the same cohort without evidence of MI
at follow-up. Matching parameters were age at
recruitment (±1.5 years), sex, center, and season of recruitment.
Results from the discovery phase were replicated in an
independent sample of 317 Dutch subjects from the
prospective EPIC-NL cohort [
] who developed MI during
follow-up (average time to diagnosis 5.64 years) and 262
unmatched healthy controls from the same cohort. Details
on anthropometrics, lifestyle, biochemical measurements,
and MI definition are provided in Additional file 1.
Our study complies with the Declaration of Helsinki
principles and conforms to ethical requirements. All
volunteers signed an informed consent form at enrollment in
the respective studies. The EPIC study protocol was
approved by Ethics Committees of the International Agency
for Research on Cancer (Lyon, France), as well as by local
Ethical Committees of the participant centers. The
EPICOR study was approved by the Ethical Committee of the
Human Genetics Foundation (Turin, Italy). For the Dutch
EPIC samples, approval was obtained by the Institutional
Review Board of the University Medical Center Utrecht
(Utrecht, the Netherlands) and the Medical Ethical
Committee of TNO Nutrition and Food Research (Zeist, the
DNA methylation measurement
DNA methylation was measured in DNA from WBCs
collected at subject enrollment into EPIC and stored in liquid
]. The Infinium HumanMethylation450
BeadChip (Illumina Inc., San Diego, CA, USA) and the
MALDI-TOF mass spectrometry methylation assay
(Sequenom Inc., San. Diego, CA, USA) were used for the
discovery phase and the replication phase, respectively.
Laboratory methods for DNA methylation level
measurements are detailed in Additional file 1.
Whole-genome methylation data quality control and normalization procedures
DNA methylation levels were measured as Beta-values,
ranging from 0 to 1. We excluded the following from the
analyses: (i) single Beta-values with detection P value ≥
0.01; (ii) CpG loci with detection P value ≥ 0.01 in more
than 20 % of the assayed samples; (iii) probes containing
SNPs with MAF ≥ 0.05 in the CEPH (Utah residents with
ancestry from Northern and Western Europe, CEU)
population; and (iv) samples with a global call rate ≤95 %.
From the 435,457 CpGs that passed QCs (~95 % of
BeadChip content), we further removed 9959 CpGs whose
methylation signal was detected by cross-hybridizing and
SNPs-containing probes [
A total of 292 matched case-control pairs and 425,498
CpG sites were used in the following analyses.
Background normalization was performed on raw
methylation data according to Marabita et al. [
Statistical analyses were conducted using the open
source R v3.0.1 package [
Analyses were performed stratifying by sex, in order to
account for the occurrence of sex-specific DNA
methylation and for the different cardiovascular risk profiles
between men and women (see the “Discussion” section).
Descriptive statistics of sample characteristics,
anthropometrics, lipid profiles, hypertension, and lifestyle habits
(smoke, alcohol consumption) was performed.
Case-control DMR analysis
We analyzed the EPICOR methylation data (discovery
phase, 425,498 CpGs) with the A-clustering algorithm
] to identify clusters of two or more neighboring CpGs
with correlated methylation levels.
The association between each one of the identified
methylation clusters and case-control status was tested
by generalized estimating equations (GEE) [
identify DMRs between MI cases and controls. We adjusted
the analyses for matching variables (age at recruitment,
center, season of recruitment, sex in the overall
analyses), estimated WBC composition (for the EPICOR
panel only), and the major cardiovascular risk factors
], i.e., smoking status, BMI, blood pressure, and
physical activity (for the EPICOR panel only). EPICOR
sample analyses were additionally adjusted for “control
probes” principal components, while EPIC-NL analyses
did not require batch correction (see Additional file 1:
Removal of technical biases).
As fasting glucose measurement was missing for >20 %
of the EPICOR and EPIC-NL samples, glucose level was
excluded from the adjustment covariates. Lipid levels were
missing for 48 EPICOR subjects: lipid levels were omitted
as covariates too, after verification that inclusion or
exclusion of this parameter did not substantially affect the
results (Additional file 1).
Due to the small number of subjects with incident
diabetes identified at follow-up (n = 9), diabetes was not
included in the covariate list.
DMRs with FDR Q value < 0.05 were considered
statistically significant and investigated in the EPIC-NL
sample with the same statistical approach. The Q statistic
] was used to assess heterogeneity between the two
sample panels: provided no heterogeneity was found, an
inverse variance-weighted fixed effect meta-analysis was
additionally carried out to achieve an overall estimate of
the two studies.
Case-control LINE-1 methylation analyses
To analyze LINE-1 methylation levels from BeadChip
data, we first identified all the BeadChip’s CpGs lying in
LINE-1 sequences according to the UCSC Genome
Browser database. The cumulative DNA methylation level
of LINE-1 sequences was computed, for each subject, as
the average methylation level across the 12,762 CpGs, out
of the >450K assayed on the BeadChip, that were
annotated in LINE-1 sequences. Case-control differences were
assayed by logistic regression, with methylation levels as a
continuous variable and the same adjustment used for the
case-control DMR discovery and replication analyses. For
replication purposes, the same analysis was performed on
the EPIC-NL samples using LINE-1 methylation data
from MassARRAY analysis (Additional file 1). A LINE-1
methylation meta-analysis of the two studies was also
done as described above.
DNA methylation and MI risk
EPICOR and EPIC-NL subjects, stratified by sex and by
study, were clustered with a RPMM algorithm [
four classes according to their ZBTB12-DMR methylation
profile, irrespective of case-control status. Each subject
was also allocated to a LINE-1 methylation class (above/
below the median). The association between MI and DNA
methylation (as RPMM class or LINE-1 methylation
profile) was evaluated on the EPIC-NL panel by logistic
regression analysis, stratifying by sex.
Moreover, to test the dependence/independence of the
DNA methylation effects from the TRFs, we compared
the ORs associated to each RPMM class and to LINE-1
methylation status under three logistic regression models,
progressively including additional covariates at each step.
To this purpose, the same analysis was done on the
EPICOR discovery panel as well, under the caveat that the
estimated ORs in this case should not be considered as a
risk estimate, being assessed in the discovery panel and, as
such, putatively inflated. Briefly, model 1 included the
matching variables only, model 2 included the whole set
of covariates used for the case-control DMR discovery
and replication analyses, and model 3 was fully adjusted
with the comprehensive set of variables as available in the
two studies. Further methodological details are provided
in Additional file 1.
Discrimination, reclassification, and calibration
We tested for the improvement in the performance of MI
risk prediction when including DMR and LINE-1 profiles
identified in the EPICOR dataset (discovery phase) by
running discrimination and reclassification analyses on the
independent EPIC-NL dataset. Two models were
compared: the first one included only TRFs that were
significantly associated to MI in our study or reported in the
literature to be associated to MI (Fig. 1, legend); the
second one comprised TRFs as model 1 plus
ZBTB12RPMM classes and LINE-1 methylation class.
For discrimination, we compared the AUC of the two
models by the DeLong test [
]. For reclassification, we
computed the NRI and IDI indices [
]. The goodness of
fit was evaluated by the Hosmer-Lemeshow (HL) test [
in order to assess the proper calibration of the model.
DNA methylation and TTD
Being EPICOR and EPIC-NL prospective cohorts with
incident MI cases identified during cohort follow-up, we
investigated the relationship between methylation and
TTD, i.e., the time lapse between blood collection and
the MI event. EPICOR and EPIC-NL cases, stratified by
study and by sex, were divided in tertiles according to
TTD. Control groups were used as reference. The
occurrence of a linear trend between DNA methylation levels
and TTD, as the ordinal categorical variable, was tested
by GEE (details in Additional file 1).
Additional file 1: Supplementary methods. A document with
supplementary materials, including the following: (1) subjects: cohort
details; lifestyle, anthropometrics, and biochemical measurements; and
outcome definition; (2) laboratory methods: EPICOR sample preparation;
discovery phase: Illumina Human450K Methylation Assay; replication
phase on EPIC-NL sample: Sequenom MassARRAY; and (3) supplementary
statistical methods: case-control differential methylation; removal of technical
biases; DNA methylation and MI risk; DNA methylation and time to disease
(TTD); supplementary references. (DOCX 73 kb)
Additional file 2: Supplemental Tables S1, S2, S3, and S4. Table S1.
top 6 genic DMRs in EPICOR MI overall cases vs controls. Table S2A.
details of the ZBTB12-DMR CpGs in EPICOR subjects. Table S2B. details
of the ZBTB12-DMR CpGs in EPIC-NL subjects. Table S3A. EPIC-NL MI risk,
adjusted models. Table S3B. EPICOR, adjusted models. Table S4A.
EPICOR case-control differential methylation analysis: comparison of
models with and without lipids adjustment. Table S4B. EPIC-NL case-control
differential methylation analysis: comparison of models with and without
lipids, batch, and WBCs adjustments. (DOCX 55 kb)
Additional file 3: Supplemental Figures S1, S2, S3, and S4.
Figure S1. quantile-quantile plot, EPICOR overall subjects. Figure S2.
quantile-quantile plot, EPICOR men. Figure S3. quantile-quantile plot,
EPICOR women. Figure S4. locations of ZBTB12 and LINE-1 CpG sites
investigated by Sequenom MassARRAY. CpGs (in red) investigated within
ZBTB12-DMR, LINE-1, and flanking primers (upper case: complementary to
DNA; lower case: T7-promoter sequence and 10mer tag). CpG sites that
could not be tested individually due to MassARRAY technology constrains,
but had to be tested jointly with neighboring CpGs as a single unit, are
underlined: the methylation level is the cumulative value of all the sites
within the CpG unit. (ZIP 91 kb)
AUC: area under the receiver operating curve; Beta-value: estimate of
methylation level at each CpG; BMI: body mass index; CVD: cardiovascular
disease; DMR: differentially methylated region; EPIC: European Prospective
Investigation into Cancer and Nutrition; EWAS: epigenome-wide association
study; FDR: false discovery rate; GEE: generalized estimating equations; IDI
index: integrated discrimination improvement index; LINE-1: long
interspersed nuclear element-1; MBPs: methyl-CpG binding proteins;
MI: myocardial infarction; NRI index: net reclassification improvement index;
QC: quality control; RPMM: recursively partitioned mixture model;
TRFs: traditional risk factors; TTD: time to disease; WBCs: white blood cells;
WHR: waist-to-hip ratio.
The authors declare that they have no competing interests.
GM, SPa, VK, CS, SG, and GFi conceived the study. SPa, AM, PC, VK, CA, RT,
GFr, PV, FR, CS, and GM for EPIC-Italy (EPICOR) and MWMV, JMAB, NCOM,
and YTvdS for EPIC-NL enrolled the subjects; managed personal information
databases (data from questionnaires, clinical data, cohort follow-up data) as
responsible of the respective cohorts; and coordinated biospecimens storage,
retrieval, and shipment to the analytical laboratories at HuGeF. GFi, PC, and FR
contributed to sample selection and EPIC-Italy (EPICOR) database management.
SG, AA, AR, and SPo carried out all the laboratory analyses from DNA extraction
(EPICOR) to DNA methylation analyses (EPICOR and EPIC-NL). LI carried out all
the biochemical measurements on the EPICOR samples. GFi carried out
all the statistical analyses. SG, AA, AR, CDG, GM substantially contributed
to the interpretation of results. SG wrote the manuscript. GM, GFi, CDG,
MWMV, JMAB, NCOM, YTvdS, PV, CS, LI, VK, and SPa critically revised the
manuscript content and provided important intellectual content. All authors
read and approved the final manuscript.
This work was supported by the Compagnia di San Paolo for the EPIC-Italy
and EPICOR projects to SPa, VK, RT, PV, LI, CS, and GM; the Human Genetics
Foundation-Torino (HuGeF) to GM and PV; and the MIUR ex60% grant to
GM. EPIC-Italy is further supported by a grant from the “Associazione Italiana
per la Ricerca sul Cancro” (AIRC, Milan) to SPa.
The EPIC-NL study was funded by “Europe against Cancer” Programme of
the European Commission (SANCO), Dutch Ministry of Public Health, Welfare
and Sports (VWS), Netherlands Cancer Registry (NKR), LK Research Funds,
Dutch Prevention Funds, Dutch Cancer Society, ZonMW—the Netherlands
Organisation for Health Research and Development, and World Cancer
Research Fund (WCRF) (the Netherlands).
The EPIC study is also supported by the European Union.
The authors wish to thank all who participated in, or collaborated with, EPIC;
the Italian AVIS blood donor organization and the Sicilian Government; and
Statistics Netherlands and Pharmo for follow-up data on cardiovascular disease
and causes of death for EPIC-NL.
Bilthoven, The Netherlands. 10Department of Epidemiology and Prevention,
IRCCS Istituto Neurologico Mediterraneo Neuromed, Pozzilli, IS, Italy.
11Epidemiology and Public Health, Imperial College London, London, UK.
1. Karaderi T , Drong AW , Lindgren CM . Insights into the genetic susceptibility to type 2 diabetes from genome-wide association studies of obesity-related traits . Curr Diab Rep . 2015 ; 15 ( 10 ): 83 . doi: 10 .1007/s11892-015-0648-8.
2. Ehret GB , Munroe PB , Rice KM , Bochud M , Johnson AD , Chasman DI , et al. Genetic variants in novel pathways influence blood pressure and cardiovascular disease risk . Nature . 2011 ; 478 ( 7367 ): 103 - 9 . doi: 10 .1038/nature10405.
3. Lange LA , Willer CJ , Rich SS . Recent developments in genome and exome-wide analyses of plasma lipids . Curr Opin Lipidol . 2015 ; 26 ( 2 ): 96 - 102 . doi: 10 .1097/MOL.0000000000000159.
4. Bell CG , Beck S. The epigenomic interface between genome and environment in common complex diseases . Brief Funct Genomics . 2010 ; 9 ( 5 -6): 477 - 85 . doi: 10 .1093/bfgp/elq026.
5. Kanherkar RR , Bhatia-Dey N , Csoka AB . Epigenetics across the human lifespan . Front Cell Dev Biol . 2014 ; 2 : 49 . doi: 10 .3389/fcell. 2014 . 00049 .
6. Mathers JC , Strathdee G , Relton CL . Induction of epigenetic alterations by dietary and other environmental factors . Adv Genet . 2010 ; 71 : 3 - 39 . doi: 10 .1016/B978-0 -12-380864-6 . 00001 - 8 .
7. Reddington JP , Pennings S , Meehan RR . Non-canonical functions of the DNA methylome in gene regulation . Biochem J . 2013 ; 451 ( 1 ): 13 - 23 . doi: 10 .1042/BJ20121585.
8. Klose RJ , Bird AP . Genomic DNA methylation: the mark and its mediators . Trends Biochem Sci . 2006 ; 31 ( 2 ): 89 - 97 . doi: 10 .1016/j.tibs. 2005 . 12 .008.
9. Schulz WA , Steinhoff C , Florl AR . Methylation of endogenous human retroelements in health and disease . Curr Top Microbiol Immunol . 2006 ; 310 : 211 - 50 .
10. Baccarelli A , Wright R , Bollati V , Litonjua A , Zanobetti A , Tarantini L , et al. Ischemic heart disease and stroke in relation to blood DNA methylation . Epidemiology . 2010 ; 21 ( 6 ): 819 - 28 . doi: 10 .1097/Ede.0b013e3181f20457.
11. Cash HL , McGarvey ST , Houseman EA , Marsit CJ , Hawley NL , LambertMesserlian GM , et al. Cardiovascular disease risk factors and DNA methylation at the LINE-1 repeat region in peripheral blood from Samoan Islanders . Epigenetics. 2011 ; 6 ( 10 ): 1257 - 64 . doi: 10 .4161/epi.6.10.17728.
12. Valinluck V , Tsai HH , Rogstad DK , Burdzy A , Bird A , Sowers LC . Oxidative damage to methyl-CpG sequences inhibits the binding of the methyl-CpG binding domain (MBD) of methyl-CpG binding protein 2 (MeCP2) . Nucleic Acids Res . 2004 ; 32 ( 14 ): 4100 - 8 . doi: 10 .1093/nar/gkh739.
13. Lund G , Zaina S. Atherosclerosis: an epigenetic balancing act that goes wrong . Curr Atheroscler Rep . 2011 ; 13 ( 3 ): 208 - 14 . doi: 10 .1007/s11883-011-0174-3.
14. 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 ( 8 ), e1000602 . doi: 10 .1371/journal.pgen. 1000602 .
15. 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 ( 4 ): 234 - 9 . doi: 10 .1016/j.mad. 2008 . 12 .003.
16. Portela A , Esteller M. Epigenetic modifications and human disease . Nat Biotechnol . 2010 ; 28 ( 10 ): 1057 - 68 . doi: 10 .1038/nbt.1685.
17. Klengel T , Pape J , Binder EB , Mehta D. The role of DNA methylation in stress-related psychiatric disorders . Neuropharmacology . 2014 ;80C: 115 - 32 . doi: 10 .1016/j.neuropharm. 2014 . 01 .013.
18. Dang MN , Buzzetti R , Pozzilli P. Epigenetics in autoimmune diseases with focus on type 1 diabetes . Diabetes Metab Res Rev . 2013 ; 29 ( 1 ): 8 - 18 . doi: 10 .1002/dmrr.2375.
19. Breton CV , Park C , Siegmund K , Gauderman WJ , Whitfield-Maxwell L , Hodis HN , et al. NOS1 methylation and carotid artery intima-media thickness in children . Circ Cardiovasc Genet . 2014 ; 7 ( 2 ): 116 - 22 . doi: 10 .1161/CIRCGENETICS.113.000320.
20. Fiorito G , Guarrera S , Valle C , Ricceri F , Russo A , Grioni S , et al. B-vitamins intake, DNA-methylation of one carbon metabolism and homocysteine pathway genes and myocardial infarction risk: the EPICOR study . Nutr Metab Cardiovasc Dis . 2013 . doi: 10 .1016/j.numecd. 2013 . 10 .026.
21. Sharma P , Garg G , Kumar A , Mohammad F , Kumar SR , Tanwar VS , et al. Genome wide DNA methylation profiling for epigenetic alteration in coronary artery disease patients . Gene . 2014 ; 541 ( 1 ): 31 - 40 . doi: 10 .1016/j. gene. 2014 . 02 .034.
22. Soriano-Tarraga C , Jimenez-Conde J , Giralt-Steinhauer E , Mola M , Ois A , Rodriguez-Campello A , et al. Global DNA methylation of ischemic stroke subtypes . PLoS One . 2014 ; 9 ( 4 ), e96543 . doi: 10 .1371/journal.pone. 0096543 .
23. Yamada Y , Nishida T , Horibe H , Oguri M , Kato K , Sawabe M . Identification of hypo- and hypermethylated genes related to atherosclerosis by a genome-wide analysis of DNA methylation . Int J Mol Med . 2014 ; 33 ( 5 ): 1355 - 63 . doi: 10 .3892/ijmm. 2014 . 1692 .
24. Sofer T , Schifano ED , Hoppin JA , Hou L , Baccarelli AA . A-clustering: a novel method for the detection of co-regulated methylation regions, and regions associated with exposure . Bioinformatics . 2013 ; 29 ( 22 ): 2884 - 91 . doi: 10 .1093/bioinformatics/btt498.
25. Lerner DJ , Kannel WB . Patterns of coronary heart disease morbidity and mortality in the sexes: a 26-year follow-up of the Framingham population . Am Heart J . 1986 ; 111 ( 2 ): 383 - 90 .
26. Anand SS , Islam S , Rosengren A , Franzosi MG , Steyn K , Yusufali AH , et al. Risk factors for myocardial infarction in women and men: insights from the INTERHEART study . Eur Heart J . 2008 ; 29 ( 7 ): 932 - 40 . doi: 10 .1093/eurheartj/ehn018.
27. Tziomalos K , Giampatzis V , Baltatzi M , Efthymiou E , Psianou K , Papastergiou N , et al. Sex-specific differences in cardiovascular risk factors and blood pressure control in hypertensive patients . J Clin Hypertens (Greenwich) . 2014 ; 16 ( 4 ): 309 - 12 . doi: 10 .1111/jch.12289.
28. Andreasson AN , Unden AL , Elofsson S , Brismar K. Leptin and adiponectin: distribution and associations with cardiovascular risk factors in men and women of the general population . Am J Hum Biol . 2012 ; 24 ( 5 ): 595 - 601 . doi: 10 .1002/ajhb.22279.
29. Caviezel S , Dratva J , Schaffner E , Schindler C , Zemp Stutz E , de Groot E , et al. Sex-specific associations of cardiovascular risk factors with carotid stiffness-results from the SAPALDIA cohort study . Atherosclerosis . 2014 ; 235 ( 2 ): 576 - 84 . doi: 10 .1016/j.atherosclerosis. 2014 . 05 .963.
30. El-Maarri O , Becker T , Junen J , Manzoor SS , Diaz-Lacava A , Schwaab R , et al. Gender specific differences in levels of DNA methylation at selected loci from human total blood: a tendency toward higher methylation levels in males . Hum Genet . 2007 ; 122 ( 5 ): 505 - 14 . doi: 10 .1007/s00439-007-0430-3.
31. Inoshita M , Numata S , Tajima A , Kinoshita M , Umehara H , Yamamori H , et al. Sex differences of leukocytes DNA methylation adjusted for estimated cellular proportions . Biol Sex Differ . 2015 ; 6 : 11 . doi: 10 .1186/s13293-015-0029-7.
32. Liu J , Morgan M , Hutchison K , Calhoun VD . A study of the influence of sex on genome wide methylation . PLoS One . 2010 ; 5 ( 4 ), e10028 . doi: 10 .1371/journal.pone. 0010028 .
33. Boks MP , Derks EM , Weisenberger DJ , Strengman E , Janson E , Sommer IE , et al. The relationship of DNA methylation with age, gender and genotype in twins and healthy controls . PLoS One . 2009 ; 4 ( 8 ), e6767 . doi: 10 .1371/journal.pone. 0006767 .
34. Brenet F , Moh M , Funk P , Feierstein E , Viale AJ , Socci ND , et al. DNA methylation of the first exon is tightly linked to transcriptional silencing . PLoS One . 2011 ; 6 ( 1 ), e14524 . doi: 10 .1371/journal.pone. 0014524 .
35. Clouaire T , Stancheva I . Methyl-CpG binding proteins: specialized transcriptional repressors or structural components of chromatin? Cell Mol Life Sci . 2008 ; 65 ( 10 ): 1509 - 22 . doi: 10 .1007/s00018-008-7324-y.
36. Keshava Prasad TS , Goel R , Kandasamy K , Keerthikumar S , Kumar S , Mathivanan S , et al. Human Protein Reference Database-2009 update. Nucleic Acids Res . 2009 ; 37 (Database issue): D767 - 72 . doi: 10 .1093/nar/gkn892.
37. Carmona G , Gottig S , Orlandi A , Scheele J , Bauerle T , Jugold M , et al. Role of the small GTPase Rap1 for integrin activity regulation in endothelial cells and angiogenesis . Blood . 2009 ; 113 ( 2 ): 488 - 97 . doi: 10 .1182/blood-2008 -02-138438.
38. Serban D , Leng J , Cheresh D. H-ras regulates angiogenesis and vascular permeability by activation of distinct downstream effectors . Circ Res . 2008 ; 102 ( 11 ): 1350 - 8 . doi: 10 .1161/CIRCRESAHA.107.169664.
39. Ueno H , Yamamoto H , Ito S , Li JJ , Takeshita A . Adenovirus-mediated transfer of a dominant-negative H-ras suppresses neointimal formation in balloon-injured arteries in vivo . Arterioscler Thromb Vasc Biol . 1997 ; 17 ( 5 ): 898 - 904 .
40. Bueno OF , Molkentin JD . Involvement of extracellular signal-regulated kinases 1/2 in cardiac hypertrophy and cell death . Circ Res . 2002 ; 91 ( 9 ): 776 - 81 .
41. Hur K , Cejas P , Feliu J , Moreno-Rubio J , Burgos E , Boland CR , et al. Hypomethylation of long interspersed nuclear element-1 (LINE-1) leads to activation of proto-oncogenes in human colorectal cancer metastasis . Gut . 2014 ; 63 ( 4 ): 635 - 46 . doi: 10 .1136/gutjnl-2012-304219.
42. Nakkuntod J , Avihingsanon Y , Mutirangura A , Hirankarn N. Hypomethylation of LINE-1 but not Alu in lymphocyte subsets of systemic lupus erythematosus patients . Clin Chim Acta . 2011 ; 412 ( 15 -16): 1457 - 61 . doi: 10 .1016/j.cca. 2011 . 04 .002.
43. Sidler C , Woycicki R , Kovalchuk I , Kovalchuk O. WI -38 senescence is associated with global and site-specific hypomethylation . Aging (Albany NY) . 2014 ; 6 ( 7 ): 564 - 74 .
44. Turunen MP , Aavik E , Yla-Herttuala S . Epigenetics and atherosclerosis . Bba-Gen Subjects . 2009 ; 1790 ( 9 ): 886 - 91 . doi: 10 .1016/j.bbagen. 2009 . 02 .008.
45. Jintaridth P , Tungtrongchitr R , Preutthipan S , Mutirangura A . Hypomethylation of Alu elements in post-menopausal women with osteoporosis . PLoS One . 2013 ; 8 ( 8 ), e70386 . doi: 10 .1371/journal.pone. 0070386 .
46. Chang SY , Wang L , Guan YQ , Shangguan SF , Du QA , Wang Y , et al. Long interspersed nucleotide element-1 hypomethylation in folate-deficient mouse embryonic stem cells . J Cell Biochem . 2013 ; 114 ( 7 ): 1549 - 58 . doi: 10 .1002/Jcb.24496.
47. Riboli E , Kaaks R , The EPIC Project: rationale and study design . European Prospective Investigation into Cancer and Nutrition. Int J Epidemiol . 1997 ; 26 Suppl 1 : S6 - S14 .
48. Wang L , Wang F , Guan J , Le J , Wu L , Zou J , et al. Relation between hypomethylation of long interspersed nucleotide elements and risk of neural tube defects . Am J Clin Nutr . 2010 ; 91 ( 5 ): 1359 - 67 . doi: 10 .3945/ajcn. 2009 . 28858 .
49. Bendinelli B , Masala G , Saieva C , Salvini S , Calonico C , Sacerdote C , et al. Fruit, vegetables, and olive oil and risk of coronary heart disease in Italian women: the EPICOR study . Am J Clin Nutr . 2011 ; 93 ( 2 ): 275 - 83 . doi: 10 .3945/ ajcn.110.000521.
50. Palli D , Berrino F , Vineis P , Tumino R , Panico S , Masala G , et al. A molecular epidemiology project on diet and cancer: the EPIC-Italy prospective study. Design and baseline characteristics of participants . Tumori . 2003 ; 89 ( 6 ): 586 - 93 .
51. Beulens JW , Monninkhof EM , Verschuren WM , van der Schouw YT , Smit J , Ocke MC , et al. Cohort profile: the EPIC-NL study . Int J Epidemiol . 2010 ; 39 ( 5 ): 1170 - 8 . doi: 10 .1093/ije/dyp217.
52. Chen YA , Lemire M , Choufani S , Butcher DT , Grafodatskaya D , Zanke BW , et al. Discovery of cross-reactive probes and polymorphic CpGs in the Illumina Infinium HumanMethylation450 microarray . Epigenetics . 2013 ; 8 ( 2 ): 203 - 9 . doi: 10 .4161/epi.23470.
53. Marabita F , Almgren M , Lindholm ME , Ruhrmann S , Fagerstrom-Billai F , Jagodic M , et al. An evaluation of analysis pipelines for DNA methylation profiling using the Illumina HumanMethylation450 BeadChip platform . Epigenetics . 2013 ; 8 ( 3 ): 333 - 46 . doi: 10 .4161/epi.24008.
54. R-CoreTeam. R. A language and environment for statistical computing . Vienna: R Foundation for Statistical Computing; 2014 . http://www.R-project. org/.
55. Liang KY , Zeger SL . Longitudinal data-analysis using generalized linear-models . Biometrika . 1986 ; 73 ( 1 ): 13 - 22 . doi: 10 .1093/biomet/73.1.13.
56. Mendis S , Puska P , Norrving B , editors. Global atlas on cardiovascular disease prevention and control . Geneva: World Health Organization (in collaboration with the World Heart Federation and World Stroke Organization); 2011 .
57. DerSimonian R , Laird N. Meta-analysis in clinical trials . Control Clin Trials . 1986 ; 7 ( 3 ): 177 - 88 . doi:0197- 2456 ( 86 ) 90046 - 2 .
58. Houseman EA , Christensen BC , Yeh RF , Marsit CJ , Karagas MR , Wrensch M , et al. Model-based clustering of DNA methylation array data: a recursive-partitioning algorithm for high-dimensional data arising as a mixture of beta distributions . BMC Bioinformatics . 2008 ; 9 : 365 . doi: 10 .1186/ 1471 -2105-9-365.
59. Delong ER , Delong DM , Clarkepearson DI . Comparing the areas under 2 or more correlated receiver operating characteristic curves-a nonparametric approach . Biometrics . 1988 ; 44 ( 3 ): 837 - 45 . doi: 10 .2307/2531595.
60. Pencina MJ , D'Agostino Sr RB , Steyerberg EW . Extensions of net reclassification improvement calculations to measure usefulness of new biomarkers . Stat Med . 2011 ; 30 ( 1 ): 11 - 21 . doi: 10 .1002/sim.4085.
61. Hosmer DW , Lemeshow S , Sturdivant RX . Applied logistic regression, 3rd edition . Wiley Ser Probab St . 2013 : 1 - 500 . doi: 10 .1002/9781118548387.