Neonatal DNA methylation profile in human twins is
specified by a complex interplay between intrauterine
environmental and genetic factors, subject
to tissue-specific influence
Lavinia Gordon,1Jihoon E. Joo,2,3Joseph E. Powell,4,5Miina Ollikainen,6
Boris Novakovic,2,3Xin Li,7Roberta Andronikos,3,7Mark N. Cruickshank,7
Karen N. Conneely,8Alicia K. Smith,9Reid S. Alisch,10Ruth Morley,7
Peter M. Visscher,4,5,11Jeffrey M. Craig,3,7,12,13and Richard Saffery2,3,12
1Bioinformatics Unit, Murdoch Childrens Research Institute (MCRI), Parkville, Victoria 3052, Australia;2Cancer and Developmental
Epigenetics Group, MCRI, Parkville, Victoria 3052, Australia;3Department of Paediatrics, University of Melbourne, Victoria 3052,
Australia;4University of Queensland Diamantina Institute, University of Queensland, Princess Alexandra Hospital, Brisbane,
Queensland 4102, Australia;5Queensland Institute of Medical Research, Brisbane, Queensland 4006, Australia;6Hjelt Institute,
Department of Public Health, FI-00014 University of Helsinki, Helsinki, Finland;7Early Life Epigenetics Group, MCRI, Parkville, Victoria
3052, Australia;8Department of Human Genetics, Emory University School of Medicine, Atlanta, Georgia 30322, USA;9Department
of Psychiatry & Behavioral Sciences, Emory University School of Medicine, Atlanta, Georgia 30322, USA;10Department of Psychiatry,
University of Wisconsin School of Medicine, Madison, Wisconsin 53719, USA;11The Queensland Brain Institute, The University
of Queensland, Brisbane, Queensland 4072, Australia
Comparison between groups of monozygotic (MZ) and dizygotic (DZ) twins enables an estimation of the relative contri-
bution of genetic and shared and nonshared environmental factors to phenotypic variability. Using DNA methylation
profiling of ~20,000 CpG sites as a phenotype, we have examined discordance levels in three neonatal tissues from 22 MZ
and 12 DZ twin pairs. MZ twins exhibit a wide range of within-pair differences at birth, but show discordance levels generally
distance from islands. Variance component decomposition analysis of DNA methylation in MZ and DZ pairs revealed a low
mean heritability across all tissues, although a wide range of heritabilities was detected for specific genomic CpG sites. The
largest component of variationwas attributed to the combined effects of nonshared intrauterine environment and stochastic
factors. Regression analysis of methylation on birth weight revealed a general association between methylation of genes
involved in metabolism and biosynthesis, providing further support for epigenetic change in the previously described link
between low birth weight and increasing risk for cardiovascular, metabolic, and other complex diseases. Finally, comparison
of our data with that of several older twins revealed little evidence for genome-wide epigenetic drift with increasing age. This
is the first study to analyze DNA methylation on a genome scale in twins at birth, further highlighting the importance of the
intrauterine environment on shaping the neonatal epigenome.
[Supplemental material is available for this article.]
Epigenetics has been defined as ‘‘the structural adaptation of
activity states’’ (Bird 2007). This is exemplified by the epigenetic
mark of DNA methylation, which influences a gene’s transcrip-
tional potential and plays a role in differentiation (Reik 2007;
Brunner et al. 2009; Huang and Fan 2010) and aging (Rakyan
et al. 2010; Teschendorff et al. 2010). Disruption of epigenetic
profile is a ubiquitous feature of cancers and is likely to play a role
in the etiology of other complex diseases (van Vliet et al. 2007;
Foley et al. 2009).
The DNA methylation profile is heritable through mitosis, but
the fidelity of this transmission is imperfect (Bennett-Baker et al.
2003) and may contribute to differences in gene expression and
phenotype observed between genetically identical individuals,
whether isogenic strains of mice (Gartner and Baunack 1981;
Pritchard et al. 2006) or human MZ twins (Fraga et al. 2005; Martin
Animal studies have demonstrated that the environment can
shape the epigenome, particularly during the intrauterine period,
when it demonstrates the greatest plasticity (Gluckman et al. 2007,
2010; Ozanne and Constancia 2007). The importance of this period
for human health is well documented, and mounting evidence im-
plicates the intrauterine environment in the fetal ‘‘programming’’ of
diseases of later life (Gluckman et al. 2007). Despite this, it remains
12These authors contributed equally to this work.
Article published online before print. Article, supplemental material, and pub-
lication date are at http://www.genome.org/cgi/doi/10.1101/gr.136598.111.
22:1395–1406 ? 2012, Published by Cold Spring Harbor Laboratory Press; ISSN 1088-9051/12; www.genome.org
which this interaction is sensitive to genetic influences.
Twin studies, which have traditionally enabled estimation of
genetic and environmental components to phenotypic variance,
have been used to estimate the effect of these factors on DNA
methylation, at both a gene-specific level (Heijmans et al. 2007;
Kaminsky et al. 2009; Javierre et al. 2010; Rakyan et al. 2011a,b).
Such studies are improving our understanding of the processes
involved in the regulation of epigenetic variation and are disen-
tangling the relative contributions of epigenetics, environment,
and genetic variation, together with stochastic factors, in complex
traits (Bell and Saffery 2012). This information is critical to
understanding processes of development and evolution (Feinberg
and Irizarry 2011) and for future potential epigenetic-based in-
terventions in complex disease.
To investigate the components of epigenetic variation at birth,
we have established a longitudinal cohort of 250 twin pairs with
et al. 2012) and have shown, in two tissues from 14 twin pairs at
scale (Gordon et al. 2011), most likely in response to epigenetic
variability. Furthermore, we subsequently provided direct evidence
thatDNAmethylationcanvaryconsiderably withinasingle locusin
multiple tissues from MZ twin pairs collected at birth (Ollikainen
differences in methylation within MZ and DZ twin pairs in adults
(Kaminsky et al. 2009). However, no study has yet focused on
genome-scale methylation differences within twins at birth.
In this study, we used genome-scale DNA methylation pro-
filing to measure the level of epigenetic variation present in three
estimated the within-pair variation of methylation profile gener-
ally and in the context of specific genomic features, and estimated
to DNA methylation profile. We further defined gene pathways
and networks subject to epigenetic change in association with
birth weight with the aim of investigating a possible epigenetic
link with risk for metabolic and cardiovascular disease.
Genome-scale analysis of DNA methylation of three tissues
from newborn twins
We used the Illumina Infinium HumanMethylation27 BeadChip
array (HM27) platform to profile DNA methylation in cord blood
mononuclear cells (CBMCs; 18 MZ, nine DZ pairs), human um-
bilical vascular endothelial cells (HUVECs, 14 MZ, 10 DZ pairs),
and placenta (eight MZ and seven DZ twin pairs) (Table 1). The
HM27 platform interrogates 27,578 CpG dinucleotides primarily
associated with 14,475 transcription start sites and has a high
technical reproducibility (Weisenberger et al. 2008; Bibikova et al.
2009; Rajendram et al. 2011). It contains separate probes to detect
methylated and unmethylated sequences, and data from both
probes are used to calculate a b-value between 0 and 1 (equivalent
to 0%–100% methylation) (Bibikova et al. 2009). Initial quality
control led to the removal of two twin pairs from HUVECs, one
twin pair from CBMCs, and one twin pair from placenta (Table 1).
We chose a highly conservative P-value probe cutoff of 0.001,
lower than has previously been reported, to minimize the level of
variability attributable to technical factors. After removing probes
on the sex chromosomes and probesthat hada detection P-value >
0.001, 19,350 probes for HUVECs, 19,204 for CBMCs, and 26,353
for placenta remained for subsequent analysis.
Twin pairs generally show a similar DNA methylation profile
In order to visualize the overall relationship between DNA meth-
ylation profiles of individuals, both within and between pairs,
probes that passed quality-control measures for each tissue sepa-
rately (Supplemental Figs. S1–S3). Interestingly, twins within the
same pair did not always cluster together, and the proportion of
within-pair clustering varied considerably in a tissue-dependent
manner. For example, only 29% of pairs clustered in HUVECS,
whereas intrapair similarity predominated in CBMCS and pla-
centa, with 58% and 71% of individual pairs clustering together
respectively (Supplemental Table S1). In all three tissues, a greater
proportion of MZ pairs clustered together relative to DZ pairs
facie evidence for both tissue-specific and genetic factors in de-
termining neonatal epigenetic profile.
Identification of factors contributing to DNA
To quantify within-pair relationships both for measures of discor-
dance and similarity, we calculated Euclidean distance (ED) and
ED and Pearson’s correlation coefficient were plotted using logit-
transformed b-values (i.e., M-values) with mean-subtracted trans-
formed values used for the latter (Fig. 1). In all instances, MZ pairs
showed a greater median within-pair similarity and lower median
within-pair discordance than DZ pairs, confirming a role for un-
derlying genetic factors in contributing to neonatal epigenetic pro-
file. These data also show that a proportion of unrelated individuals
than some co-twins, highlighting the likely role of stochastic/non-
shared environmental factors in determining the epigenetic profile.
Surprisingly, whenthe effects of chorionicity were tested (withinMZ
pairs only), dichorionic (DC) pairs were found to be generally more
similar (less discordant) epigenetically than monochorionic (MC)
pairs in both CBMCs (eight MC, nine DC) and HUVECs (eight MC,
five DC) (Fig. 1). Insufficient numbers precluded a similar examina-
tion in relation to the placental methylation profile. In addition,
there were no sex-specific significant differences in discordance or
similarity for all tissues (data not shown).
Estimation of variance components of DNA
Using methylation within our neonatal twin tissues as a variable
phenotype, we estimated the narrow sense heritability (h2, the
proportion of phenotypic variance due to additive genetic factors)
and common (intrauterine) environmental variance (c2) for all
HM27 probe-associated CpGs within our data sets using the least
against the expected null distribution with the variance equal to the
Mean heritability across all probes was 0.12 (60.0017, p = 0.0024)
for CBMCs, 0.07 (60.0014, p = 0.009) for HUVECs, and 0.05
(60.0016, p = 0.017) for placenta with calculated P-values from
Gordon et al.
empirical null distribution of?h2¼ 0 determined by a permutation
analysis, suggesting a pattern of greater estimated heritability than
that expected by chance. Because the distribution of c2did not differ
from a random distribution (median = 0), we conclude that shared
environment does not contribute significantly to methylation vari-
ation within our twins in utero and that the remaining variance in
methylation is due to the sum of unique intrauterine environment
and stochastic factors encountered by the tissues in each twin. An
investigation of probes with the highest (top 5%) heritability esti-
mates (Supplemental Table S3) revealed 961 probes in CBMCs with
h2values ranging from 0.48 to 0.94; 968 probes in HUVECs with
h2= 0.37–0.95, and 1304 probes in placenta with h2= 0.49–0.97. In
addition, only 3%–10% of highly heritable probes were shared be-
tween two tissues (28 between placenta and HUVECs, 109 between
HUVECs, and CBMCs and 58 between CBMCs and placenta). Only
three probes were highly heritable in all three tissues (cg15052901/
SLC24A4, h2= 0.49–0.62; cg14217157/WHSC2, h2= 0.52–0.63;
cg01593886/COL1A1, h2= 0.51–0.72). Ontology analysis of genes
linked to highly heritable probes revealed enrichment of genes in-
metabolism, and biosynthesis in HUVECs and signaling in placenta
(Supplemental Table S4).
Within-pair discordance varies with genomic location
Because the annotation of the HM27 array focuses on CpG is-
lands and gene promoters (Bibikova et al. 2009), we examined
how within-pair methylation discordance, measured using me-
dian within-pair standard deviation, varies in relation to these
landmarks. Previous studies in unrelated individuals (Bock et al.
suggested that such regions are more refractory to methylation
variation than non-island CpGs. To test this in our data, we
compared median within-pair methylation discordance for CpG
islands, CpG island ‘‘shores’’ (sequences up to 2 kb from CpG
islands), CpG island ‘‘shelves’’ (sequences within 2 kb and 4 kb of
CpG islands), and CpGs >4 kb from CpG islands (Fig. 3; Sandoval
et al. 2011). In agreement with previous studies (Doi et al. 2009;
Irizarry et al. 2009), we found that mean absolute methylation
levels in our samples increased as a function of distance from
CpG islands (data not shown). We also found that median
within-pair methylation discordance increased with increasing
distance from CpG islands (up to 4 kb) in all tissues for both MZ
and DZ pairs, with no evidence for a further increase at distances
Twin pair characteristics
ID no. Zygositya
Twin 1 sexTwin 2 sex
Twin 1 birth
Twin 2 birth
P, H, C
P, H, C
P, H, C
P, H, C
P, H, C
P, H, C
P, H, C
a(MZ) Monozygotic; (DZ) dizygotic.
b(MC) Monochorionic; (DC) dichorionic.
c[(Weight of heaviest twin) ? (weight of lightest twin)]/(weight of heaviest twin) 3 100.
d(H) HUVECs; (C) CBMCs; (P) placenta.
eExpression data previously generated (Gordon et al. 2011).
fHUVECs from pairs 1065 and 2027, CBMCs from pair 2064, and placenta from pair 2027 were dropped from the analysis because one twin failed quality
control during array data analysis.
Methylation discordance in twins at birth
Relationship between within-pair methylation discordance
and age: no evidence for genome-scale epigenetic drift
We have presented evidence that the nonshared intrauterine envi-
ronment contributes to methylation discordance at birth for both
MZ and DZ twins. However, little is known about how nonshared
environment can influence epigenetic discordance postnatally. We
and others have shown using array analysis that within-twin pair
discordance in the gene expression profile increases as a function of
age (Fraga et al. 2005; Gordon et al. 2011). Additionally, limited
evidence from low-resolution analysis also suggests that within-pair
DNA methylation discordance increases with age (a phenomenon
termed ‘‘epigenetic drift’’) (Fraga et al. 2005). However, a recent
evidence for such drift (Bocklandt et al. 2011). We plotted within-
pair methylation discordance using Euclidean distance for blood-
derived data from twins from birth to 73 yr ofage from ourdata and
94 MZ pairs and 17 DZ pairs from Infinium HM27 data sets from
a previously published data set (Rakyan et al. 2010), with additional
sets of unpublished data (Fig. 4; Table 2). We found no evidence for
epigenetic drift throughout the life course in either MZ or DZ pairs.
Identification of highly discordant gene classes
Despite the overall low median discordance in methylation appar-
ent in all twin pairs (Supplemental Fig. S4), almost all pairs had
several HM27 probes with a high level of methylation discordance
in excess of ;20% (Db > 0.2). To investigate the potential biological
relevance of such probes, we ranked each gene-associated CpG by
median within-pair standard deviation, a measure that summarizes
the typical discordance between co-twins for that gene, separately
for MZ and DZ pairs, for all three tissues (Supplemental Table S5).
Using a combination of Gene Ontology (Supplemental Table S6)
and pathway analysis (Supplemental Table S7), we found that genes
associated with development and morphogenesis were over-repre-
sented in MZ and DZ pairs from all three tissues, closely followed by
genes involved in response to environment and the cell cycle/cell
division. In contrast to our previous analysis of gene expression
(Gordon et al. 2011), we found no evidence that imprinted differ-
entially methylated regions (DMRs) (Choufani et al. 2011) and
housekeeping genes are significantly variably methylated within
MZ twin pairs (p > 0.14 and p > 0.37 for all tissues, respectively).
Identification of discordantly methylated genes associated
with birth weight
We are particularly interested in identifying epigenetic variation
in genes potentially associated with birth weight because of the
heritability (h2) and common (intrauterine) environmental variance (c2). Distributions of h2were also compared with the random distribution (dotted lines).
Analysis of variance components of DNA methylation in CBMCs, HUVECs, and placenta. Data from all probes were used to plot histograms of
correlation, zygosity, and chorionicity. Box-and-whisker plots of within-
pair methylation discordance (Euclidean distance) and Pearson’s corre-
lation coefficient of mean-corrected values in HUVECs, CBMCs, and pla-
centa from MZ and DZ twins, same-sex unrelated (UR) individuals, and
MC and DC MZ twins. Numbers of pairs within each category are shown
above each graph.
Relationship between within-pair methylation discordance/
Gordon et al.
1398 Genome Research
numerous previous studies linking low birth weight with later de-
velopment of cardiovascular, metabolic, and other complex diseases
et al. 2007). We have previously identified genes whose expression
levels correlate with birthweight ina subset of tissues from newborn
twins and shown that these genes are enriched for functions and
pathways associated with metabolism, growth, and cardiovascular
disease (Gordon et al. 2011). In the present study, we performed
a similar regression analysis of methylation M-values on birth
weight, using a statistical model that estimates association based on
within-pair variation of both DNA methylation and birth weight.
MZ HUVECs) was significantly associated with birth weight after
adjustment for multiple testing (adjusted P-value < 0.1) (Supple-
mental Table S8) and included genes with links to metabolism,
growth, and cardiovascular disease. For example, APOLD1 (apoli-
poprotein L domaincontaining1), identified inHUVECsand EDG1
(sphingosine-1-phosphate receptor 1), identified in CBMCs, both
regulate vascular function in mice and humans (Liu et al. 2000;
Regard et al. 2004; Simonsen et al. 2010; Gordon et al. 2011). Even
though no genes in placenta reached significance after multiple
testing, polymorphisms in the genes ranked 1 and 2, HLA-B (ma-
jor histocompatibility complex class 1B) and SCD (stearoyl-CoA
desaturase), have been associated with low birth weight and meta-
and Ntambi 2008; Capittini et al. 2009; Shin et al. 2010).
To investigate pathways and processes that may be subject to
epigenetic variation in association with birth weight, genes from
the above regression analysis were ranked by adjusted P-values for
birth weight and analyzed using Gene Ontology (Supplemental
Table S9) and pathway analysis (Supplemental Table S10). An
over-representation of genes associated with metabolism and bio-
synthesis was found for all three tissues for both MZ and DZ pairs,
genes associated with cardiovascular function/disease were over-
pairs, and genes associated with growth and proliferation were
over-represented in only CBMCs for both MZ and DZ pairs. To test
for an association of specific classes of genes previously linked to
birth weight and metabolic function, we performed gene set test
analysis on a list of 167 imprinted DMRs (Choufani et al. 2011),
247 cardiovascular-associated genes (http://www.ucl.ac.uk/silva/
cordance is plotted as box-and-whisker plots against probes depending on location in relation to CpG islands. Data are plotted for CpG islands, CpG
island shores (0–2 kb from islands), CpG island shelves (2–4 kb from islands), and ‘‘open sea’’ probes (>4 kb from CpG islands).
Relationship between methylation discordance to location within CpG islands, shores, and shelves. Median within-pair methylation dis-
with age in blood-derived tissues. Euclidean distance is plotted, as in
Methods, for twins from birth to 73 yr of age from our data and 94 MZ
pairs and 17 DZ pairs from previously published and unpublished Infinium
HM27 data sets (see text and Table 2).
Relationship between within-pair methylation discordance
Methylation discordance in twins at birth
cardiovasculargeneontology) (Lovering et al. 2008), and 41 genes
previouslylinked with obesityby genome-wide associationstudies
associated genes with birth weight was found in any tissue from MZ
or DZ twins (p > 0.15 and p > 0.2, respectively), whereas some
evidence for a moderate relationship between obesity-associated
genes and birth weight was found in HUVECs from DZ (p = 0.01)
and MZ pairs (p = 0.11), but not in other tissues (p > 0.56).
To further validate the observed associations, we performed
locus-specific DNA methylation analysis on three genes using the
Sequenom EpiTYPER platform, which we have previously found
to be accurate and reproducible (Novakovic et al. 2010, 2011;
Ollikainen et al. 2010). Methylation assays were designed to in-
terrogate specific CpG sites (plus adjacent sites) measured by the
HM27 platform, with measurements performed using the same
sample set. Absolute DNA methylation was highly correlated
across the two platforms (Spearman’s correlation coefficient
between within-pair methylation discordance and birth weight
discordance (Supplemental Figs. S6–S8). This analysis also high-
lighted that the observed association of birth weight with meth-
ylation level is regional rather than localized to a specific CpG site.
Despite the growing awareness of the importance of inter-
individual epigenetic variation in modulating the risk associated
such variation arises in humans. In this study, we capitalized on
widespread, genome-scale epigenetic discordance between genet-
ically identical humans at birth. These findings highlight the im-
portance of the intrauterine environment in determining the
neonatal DNA methylation profile and reveal the presence of tissue-
specific variability in response to such factors.
Nonshared environmental/stochastic factors predominate
in determining neonatal DNA methylation
Unsupervised clustering and pair-specific analysis of discordance
and correlation revealed compelling evidence for a limited genetic
contribution to the neonatal methylation variability, supported by
analysis of variance components of DNA methylation, which pro-
duced mean levels of heritability of 0.05–0.12 depending on tissue
(Fig. 2). However, the HM27 platform we used for this analysis
contains probes associated with gene promoters and CpG islands,
and previous studies have demonstrated that genomic regions
showing a high heritability of DNA methylation are under-repre-
sented in CpG islands (Gertz et al. 2011). As such, our heritability
of the genomeas a whole. Additionally, it is important to remember
that epigenetic heritability estimates will not only be population-
specific, but also cell-, tissue-, time-, and locus-specific. These will
also be largely dependent on the sensitivity, resolution, and cover-
age of the specific epigenetic assay used for measurement.
The suggestion that the DNA methylation profile is only
minimally influenced by genetic variation agrees with a previous
array-based, genome-scale study of DNA methylation in buccal cell
of DNA methylation at 1760 CpG sites in CD4+lymphocytes from
adult twins (Gervin et al. 2011). One study that measured DNA
methylationat;1500 CpG sitesinwholebloodfrom 43 MZ and43
DZ twin pairs found that 23% of all CpG sites displayed significant
heritability of methylation level (Boks et al. 2009). Although low
power makes comparison difficult, the range of heritability within
the top 5% of probes in the Boks and colleagues study (0.62–0.94) is
similar to that found for the top 5% of probes for CBMCs in our
study (0.48–0.94). Studies of allele-specific methylation (ASM) have
also found evidence for a high heritability of DNA methylation at
a subsetof genomic loci,with proportions varying with the method
of analysis and the tissue examined (Kerkel et al. 2008; Boks et al.
2009; Zhang etal.2009, 2010;Meaburn etal.2010; Schalkwyketal.
2010; Shoemaker et al. 2010; Gertz et al. 2011).
Surprisingly, we found little evidence for an effect of common
environment on the overall DNA methylation profile at birth using
variance component analysis. This does not rule out the possibility
that a minority of genes are influenced by a common environment.
Indeed, previous genome-scale studies of DNA methylation have
found a range of probes significantly associated with maternal en-
vironmental, from 0.6% (Breton et al. 2009) and 1.1% (Fryer et al.
2011) to 23% (Katari et al. 2009). Because the accuracy of the esti-
mation of effect size depends on study power, further investigation
in larger numbers of twins is needed (Visscher 2004).
Given the small genetic effect and lack of evidence for wide-
spread common environmental effects observed in our study, the
largest residual variance component contributing to overall DNA
methylation profile represents cumulative nonshared (individual
intrauterine) environment and stochastic factors. There is prior
evidence that nonshared environmental factors can influence
Details of the Infinium HumanMethylation27 data sets used for Figure 4
TissueAge range (yr) Number of twin pairsCohort detailsFigure 4 letter
Whole cord blood0 3 DZ
CBMCs0 Phenotypically normalb
Whole peripheral blood
Discordant for mild psoriasisc
Whole peripheral blood
Whole peripheral blood
Autism cohort, normal pairse
Source of data:aK.N.C. and A.K.S.;bthis study;cRobert Lyle, Kristina Gervin, and Jennifer Harris, Oslo;dJordana Bell and Pei-Chien Tsai, UK;eJonathan Mill
and Chloe Wong, UK;fR.A.
Gordon et al.
1400 Genome Research
phenotype (Bergvall and Cnattingius 2008; Plomin 2011; Torche
and Echevarria 2011). Such factors in twin pregnancies include
discordant placental weight, discordant placental umbilical cord
insertion (both resulting in differential fetal blood supply), or
differential exposure to infection (Stromswold 2006; Antoniou
et al. 2011). Further evidence that factors specific to each twin can
in response to environment are discordantly methylated (this
study) or expressed (Gordon et al. 2011) within twin pairs.
Our finding that MZ twins sharing a single placenta (MC) were
more discordant for methylation profile than MZ twins with their
own placenta (DC) is counterintuitive but similar to that reported
previously (Kaminsky et al. 2009). It was suggested that the earlier
embryo splitting associated with DC pairs potentially reflects epi-
(Kaminsky et al. 2009), although there is no direct evidence sup-
porting this hypothesis. An alternative explanation could be that
MC twins are more likely to experience competition for resource
allocation, with or without associated vascular connections that in
extreme cases can lead to a large flow of blood in one direction
known as twin-to-twin transfusion syndrome (TTTS). Although
each of these scenarios is possible, it is worth noting that we have
shown previously that the effect of chorionicity on DNA methyla-
2011). Clearly, larger studies are needed to resolve this issue.
Extensive evidence exists for a role of stochastic influence in
determining DNA methylation and other epigenetic marks dur-
ing early development (for review, see Whitelaw et al. 2010). An
elegant example of this was reported by Feinberg and colleagues,
who demonstrated that the genome-wide DNA methylation
profile in the livers of newborn inbred mice raised in identical
environments was hypervariable at certain loci, termed variably
methylated regions (VMRs) (Feinberg and Irizarry 2011). VMRs
were enriched in genes involved in development and morpho-
genesisbothinmice andunrelatedhumans (Feinbergetal.2010),
and such genes are also discordantly methylated within our twin
pairs (Supplemental Tables S9, S10). Because early development is
associated with rapid cell division, this could explain our finding
that genes involved in the cell cycle are discordantly methylated
within twins, which was also found in adult twins (Kaminsky
et al. 2009). We also found that genes associated with response to
environment are hypervariable within MZ and DZ twin pairs in
multiple tissues (Supplemental Tables S9, S10), which agrees with
previous studies of expression discordance within MZ twin pairs
(Sharma et al. 2005; Choi and Kim 2007).
Evidence for a tissue-specific effect on heritability
of methylation profile
We found no compelling evidence for a common set of highly
heritable DNA methylation variants across different tissues, sup-
porting similar findings in a study of ASM in multiple human cell
lines (Shoemaker et al. 2010) and recent data using gene expression
as a phenotypic outcome (Powell et al. 2012). This contrasts with
a subset of recently described ASM events described in a single in-
dividual that appear to be shared across kidney and skeletal muscle
(Gertz et al. 2011). Such tissue-specific differences in the heritability
of DNA methylation (and expression) may arise due to the possi-
in control of the expression of the same gene in different tissues, in
association with the biological function of specific cells (Altschuler
and Wu 2010).
Genomic regions with high within-pair methylation
discordance; constraint at CpG islands?
in DNA methylation increased with increasing distance from CpG
islands (Fig. 3). This agrees with a similar finding for highly var-
iably methylated regions (VMRs) between unrelated individuals
(Zhang et al. 2010) and with data showing that interindividual
differences in methylation are lowest in unmethylated CpG is-
lands and highestin methylatedregions of the genome(Bock et al.
2008). Furthermore, our data have shown that such regions are
hypervariable irrespective of zygosity, in support of our findings of
only a minor genetic effect in determining the overall DNA
methylation profile. In combination, these data suggest that
methylation levels are more constrained in CpG-dense genomic
regions, possibly because of a stabilizing influence from neigh-
boring CpGs (Bock et al. 2008).
What function does variable methylation serve?
Because MZ twins share the same mother and genotype, factors
present with the residual variance component ‘‘nonshared envi-
ronment’’ are most likely driving variation here (see above). Al-
though the relative proportions of nonshared (intrauterine) envi-
ronment versus stochastic influences on the neonatal epigenome
remain to be demonstrated, Feinberg and colleagues have proposed
a model in which a genetically inherited propensity to stochastic
variability in DNA methylation has evolved to increase fitness in
a varying environment (Feinberg and Irizarry 2011). Furthermore,
a combination of influences have been proposed to explain the
phenomenon of metastable epialleles, which are variably expressed
due to epigenetic modifications that are established in a stochastic
manner during early development, and that are also environmen-
tally labile (for review, see Bernal and Jirtle 2010).
DNA methylation in relation to low birth weight:
A mechanistic link with complex disease in later life
We found that the genes whose methylation was tightly associated
with growth, metabolism, and cardiovascular disease (Supplemental
Tables S9, S10). A subset of these was confirmed by locus-specific
DNA methylation analysis that revealed up to 60% methylation
discordance between heaviest versus lightest twins (Supplemental
Figs. S6–S8). Taking into account our previous data from expres-
sion arrays, we speculate that DNA methylation and expression
levels of key genes associated with cardiovascular and metabolic
function can be set in utero to confer elevated risk for disease in
later life and that this setting is linked with low birth weight.
Lack of evidence for genome-wide epigenetic drift throughout
Our data support the idea that a combination of stochastic and
nonshared intrauterine environment can generate a net within-
pair difference in DNA genome-wide epigenomic profiles at birth.
But do such factors influencetheepigenomeafter birthin a similar
way? The cumulative effects of environmental and stochastic
variation on changing epigenetic profile (known as ‘‘epigenetic
drift’’) were first described by a study that examined both genome-
wide and locus-specific DNA methylation variation in a small
number of young and middle-aged MZ twins, which found
a greater within-pair discordance in the latter (Fraga et al. 2005).
Methylation discordance in twins at birth
Our cross-sectional analysis of genome-scale data from HM27 ar-
rays from blood-derived DNA from twins from birth to >70 yr of
age does not support a generalized age-related epigenetic drift at
the genome-scale level (Fig. 4) and agree with other recent data
from saliva using the same platform (Bocklandt et al. 2011). These
discrepancies are most likely due to differences in methodology,
sample size, and genomic regions the genome studied. Despite our
lack of compelling evidence for epigenetic drift over the life course,
others haveshownthat DNAmethylation can vary over time,using
cross-sectional (Boks et al. 2009; Christensen et al. 2009; Rakyan
et al. 2010) or longitudinal (Bjornsson et al. 2008; Feinberg et al.
2010) approaches. Further studies are needed that focus on both
time, ideallyaddressingthesuggestion thatepigeneticdrift could be
driven by differences in environment (Fraga et al. 2005).
Strengths and weaknesses of this study
solely during intrauterine life. This is a unique time during the life
span in which individuals share the same maternal environment,
thus minimizing differences in shared environment. This has also
enabled us to use traditional modeling of variance components to
estimate the relative influence of genetic, common, and nonshared
has measured chorionicity in twins (Kaminsky et al. 2009), this has
enabled us to investigate the effects of sharing or having separate
placentas on the methylome. Studying twins at birth has also en-
abled us to identify a possible epigenetic mechanism linking low
birth weight and risk for complex disease in later life. This has im-
plications for future minimization of the poor health outcomes as-
sociated with low birth weight, through reversing the associated
epigenetic changes. Technologically, we have used a reproducible
and accurate method of genome-scale measurement of DNA
methylation, which compares well to and is more quantifiable than
immunoprecipitation or enzyme-based methods (Bock et al. 2010).
We have also looked at multiple tissues, which has enabled us to
determine whether methylation variation occurs in a tissue-specific
manner. Twin studies sometimes attract the criticism that they may
not be applicable to the rest of the population. However, apart from
sharing a placenta, and a relatively smaller birth size and lower
gestational age, twins have similar health outcomes to singletons
(Morley et al. 2003; Morley and Dwyer 2005). Greater than 95% of
twins and all singletons have their own amniotic sac and therefore
would have similar issues of nonshared environment, albeit with
different magnitudes of variation in placenta weight, cord place-
ment, and cord thickness (Antoniou et al. 2011).
The main weakness of this and similar studies (e.g., Boks et al.
2009)isthe relatively smallsample size.However, ourprimaryfocus
is not on specific genes, but on groups of genes with shared ontol-
between the methylomes of twins. We have also focused on gene
promoters and CpG islands, which constitute a relatively small
proportion of the genome, and acknowledge that some of our
findings may not be reflected in the rest of the genome.
In our calculations of within-pair discordance and heritabil-
ity, it is possible that genetic differences within DZ pairs result in
differing probe-hybridization efficiencies within such pairs. How-
ever, our unpublished studies have shown that such sequence
variants do not affect the estimation of DNA methylation at the
CpG associated with each probe, essentially because any changes
to hybridization kinetics will affect the methylation-specific
probes and the non-methylation-specific probes in the same
manner. But what of the effect of SNPs at CpG sites assayed by
Infinium HM27 probes? Of a small number of probes for which
this is the case (Chen et al. 2011), only one appeared in our list of
;1000 genes with a high heritability (data not shown).
In summary, our study uses biological samples from twins at
birth and contributes to the understanding of prenatal human
development and the factors by which it is influenced. It is es-
sential to understand these factors in healthy individuals in order
to compare with states of disease and disease predisposition.
Subjects and tissues
Sample collection from twins at the time of delivery was performed
with appropriate human ethics clearances from the Royal Women’s
Hospital (06/21),Mercy Hospitalfor Women (R06/30),andMonash
Medical Center (06117C), Melbourne. The twin pairs chosen for
methylation array analysis are shown in Table 1. They shared
of 250 pairs. For CBMCs, we studied 18 MZ and nine DZ pairs; from
HUVECs, 14 MZ and 10 DZ pairs; and from placenta, eight MZ and
seven DZ pairs. HUVECs and CBMCs were examined in combina-
tion for 19 pairs, and all three tissues were profiled in seven pairs
Sample preparation and DNA extraction
positive HUVECS were isolated using collagenase and magnetic
sorting, and full-thickness placental samples were isolated as de-
scribed previously (Novakovic et al. 2010; Gordon et al. 2011).
DNA was extracted using phenol:chloroform as described pre-
viously (Novakovic et al. 2010).
Infinium methylation analysis
DNA samples were processed using the MethylEasy Xceed bisulphite
conversion kit (Human Genetic Signatures), according to the
manufacturer’s instructions. Genome-wide DNA methylation anal-
ysis was performed by the Australian Genome Research Facility
(Melbourne, Australia) or ServiceXS (Leiden, The Netherlands).
Infinium HM27 BeadChip arrays (Illumina) were hybridized and
scanned as per the manufacturer’s instructions. Raw data were
exported from BeadStudio (Illumina). All statistical analysis was
performed in R (version 2.12) (R_Development_Core_Team 2009)
using packages from the Bioconductor project (Gentleman et al.
2004) and in-house scripts. Data quality was confirmed using
arrayQualityMetrics (Kauffmann et al. 2009). Probes on the X and Y
chromosomes were removed from further analysis to eliminate
sex-specific differences in methylation. The lumi package, which is
specifically designed for Illumina data, was used to calculate the
log2ratio for methylated probe intensity to unmethylated probe in-
tensity, the M-value (Du et al. 2008, 2011). These values underwent
background correction and normalization using lumi. Possible batch
effects from samples processed at different times were compensated
for with ‘‘color adjustment’’ from lumi. Any probe within a sample
with a highly conservative detection P-value of 0.001 or greater was
excluded from further analysis. For correlation coefficients, M-values
each probe. This was then subtracted across all of the samples to
create a mean-corrected set of M-values. The correlation coefficients
for each pair for MZ, DZ, MZMC, and MZDC were then calculated
from these values. Correlation coefficients were also calculated only
Gordon et al.
within each array, for every unrelated individual (UR) comparison,
i.e., pairwise comparison of unrelated individuals. Euclidean dis-
tances were also calculated using M-values from all probes.
Differential methylation analyses were performed using the
limma package (Smyth 2005), which is designed for the analysis of
microarray data. Array quality weights were estimated to allow for
the possibility of variable DNA quality between the samples
(Ritchie et al. 2006), and these weights were incorporated in all
differential methylation analyses. To study discordance between
co-twins, a linear model was fitted to the M-values for each gene
with twin-pair as a predictive factor, equivalent to a one-way
ANOVA analysis in which variability is broken up into between-
twin and within-twin components. This analysis yielded a within-
twin standard deviation for each CpG across all twin pairs, which
was taken to summarize the level of discordance between co-twins
for that gene. To examine the relationship between birth weight
and DNA methylation, genewise linear models were fitted with
twin-pair as a factor and log-weight as a covariate. The linear
models also included a correction for possible batch effects. The
association of birth weight on each gene was assessed using
moderated t-statistics (Smyth 2004). Genes were judged to be dif-
algorithm (Hochberg and Benjamini 1990).
Limma was also used to conduct several gene set tests. These
of genes tends to be more highly ranked according to some given
criterion than the remaining genes (Michaud et al. 2008). To test
whether certain functional categories of genes tend to be more
discordant than others, genes were ranked by within-pair standard
deviation. This approachwas applied to 45 imprinted genes and to
568 housekeeping genes (Eisenberg and Levanon 2003). To test
whether certain functional groups of genes are associated with
birth weight, genes were ranked by moderated t-statistic when
a list of 247 cardiovascular-associated genes from the Cardiovas-
cular Gene Ontology Annotation Initiative (Lovering et al. 2008).
Estimating variance components of DNA methylation
Using genetic (twin) models, we are able decompose the observed
variance in methylation levels into its additive genetic and com-
mon environmental components (Neale and Cardon 1992). Ad-
ditive genetic variance (A) denotes the variance resulting from the
sum of allelic effects throughout the genome, whereas common
environmental variance (C) relates to the environmental influences
shared within twin pairs. The remaining variance (E) is nonshared
(individual) environmental effects and includes error terms. We can
estimate A, C, and E for each methylation probe based on the
intraclass correlation between MZ (rMZ) and DZ (rDZ) twin pairs using
a least-squares estimator (Visscher 2004). Variance components
A, C, and E are calculated as follows: A = 2(rMZ? rDZ); C = rMZ? A;
E = 1 ? rMZ. Heritability (h2) is calculated as h2= A/P, where P is the
sum of A, C, and E (observed phenotypic variance).
Locus-specific methylation analysis
Sequenom MassARRAY EpiTYPING was performed as previously
described (Ollikainen et al. 2010; Novakovic et al. 2011). The
primers are listed in Supplemental Table S11. In brief, amplifica-
tion was performed after bisulfite conversion of genomic DNA
with the MethylEasy Xceed bisulphite conversion kit (Human
Genetic Signatures). All PCR amplifications and downstream pro-
cessing were performed at least in duplicate, and the mean meth-
ylation level at specific CpG sites was determined. Raw data
obtained from MassArray EpiTYPING were cleaned systematically
using an R-script to remove samples that failed to generate data for
>70% of CpG sites tested. Also, technical replicates showing $5%
absolute difference from the median value of the technical repli-
cates were removed, and only samples with at least two successful
technical replicates were analyzed. Samples were compared across
each analyzable CpG site in the amplicon, as well as the mean
across the whole amplicon.
Gene Ontology and pathway analysis
Gene Ontology (GO) analyses were conducted using GOrilla soft-
ware (Eden et al. 2009), with the default options and searching all
ontologies. This program identifies enriched GO terms from
ranked gene lists using all ranked genes, an approach that is
analogous to the limma Wilcoxon gene set tests. Genes were
ranked for GOrilla analyses using the same ranking statistics as
described above for the gene sets tests. Pathway analysis was per-
formed using Ingenuity Pathways Analysis (IPA) software (In-
genuity Systems). The functional analysis identified the biological
data sets. IPA was used to identify enriched canonical pathways,
gene networks, and functional classes. Genes corresponding to all
Infinium probes that passed QC were used as a reference set.
Public microarray data analysis
Illumina HumanMethylation27 BeadChip twin data were obtained
from various sources (see Table 2). The data sets used were restricted
to blood and, wherever possible, phenotypically normal twins.
correlation coefficients and Euclidean distances were calculated
from all probes as described above (Fig. 4).
The data from this study have been submitted to the NCBI Gene
Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/)
under accession number GSE36642.
We thank the following people for kindly sharing their un-
published data: Robert Lyle and Kristina Gervin, Department of
Medical Genetics, Oslo University Hospital and University of
Oslo, Norway; Jennifer Harris, Division of Epidemiology, Nor-
wegian Institute of Public Health, Oslo, Norway; Jordana Bell and
Pei-Chien Tsai, Department of Twin Research and Genetic Epi-
demiology, King’s College and St. Thomas’ Hospital, London,
England; and Jonathan Mill and Chloe Wong, King’s College,
London, England. We also thank John Carlin for his contribu-
tions to establishing the PETS cohort and for biostatistical sup-
port; obstetricians Mark Umstad, Royal Women’s Hospital, Mel-
bourne; Euan Wallace, Monash Medical Centre, Melbourne; and
Mark Permezel, Mercy Hospital for Women, Melbourne for their
contributions to establishing the PETS cohort and access to study
participants; Sarah Healy, Tina Vaiano, Nicole Brooks, Jennifer
Foord, Sheila Holland, Anne Krastev, Siva Illancheran, and
Joanne Mockler for recruitment and sample collection; and
Technical Officer Anna Czajko, Study Coordinator Geraldine
McIlroy,andallmothersandtwins thatparticipated inthisstudy.
Finally, we are grateful to all of the families at the participating
SFARI Simplex Collection (SSC) sites, as well as the principal
Methylation discordance in twins at birth
investigators (A. Beaudet, R. Bernier, J. Constantino, E. Cook, E.
Fombonne, D. Geschwind, D. Grice, A. Klin, D. Ledbetter, C.
Lord, C. Martin, D. Martin, R. Maxim, J. Miles, O. Ousley, B.
Peterson, J. Piggot, C. Saulnier, M. State, W. Stone, J. Sutcliffe, C.
Walsh, E. Wijsman). We also appreciate the access to phenotypic
data on SFARI Base. Approved researchers can obtain the SSC
population data set described in this study by applying at https://
base.sfari.org. This work was supported by grants from the
Australian National Health and Medical Research Council (grant
numbers 437015 and 607358 to J.C. and R.S.), the Bonnie Babes
Foundation (grant number BBF20704 to E.J.), the Sigrid Juselius
Foundation (to M.O.), the Academy of Finland (to M.O.), the
Finnish Cultural Foundation (to M.O.), the Financial Markets
Foundation for Children (grant no. 032-2007), and by the
Victorian Government’s Operational Infrastructure Support
Altschuler SJ, Wu LF. 2010. Cellular heterogeneity: Do differences make
a difference? Cell 141: 559–563.
Antoniou EE, Derom C, Thiery E, Fowler T, Southwood TR, Zeegers MP.
2011. The influence of genetic and environmental factors on the
etiology of the human umbilical cord: The East Flanders prospective
twin survey. Biol Reprod 85: 137–143.
BarkerDJ. 1990. Thefetaland infant origins ofadultdisease.BMJ301: 1111.
Bell J, Saffery R. 2012. The value of twins in epigenetic epidemiology. Int J
Epidemiol 41: 140–150.
Bennett-Baker PE, Wilkowski J, Burke DT. 2003. Age-associated activation of
epigenetically repressed genes in the mouse. Genetics 165: 2055–2062.
Bergvall N, Cnattingius S. 2008. Familial (shared environmental and
genetic) factors and the foetal origins ofcardiovascular diseases and type
2 diabetes: A review of the literature. J Intern Med 264: 205–223.
Bernal AJ, Jirtle RL. 2010. Epigenomic disruption: The effects of early
developmental exposures. Birth Defects Res A Clin Mol Teratol 88: 938–
Bibikova M, Le J, Barnes B, Saedinia-Melnyk S, Zhou L, Shen R, Gunderson
KL. 2009. Genome-wide methylation profiling using Infinium assay.
Epigenomics 1: 177–200.
Bird A. 2007. Perceptions of epigenetics. Nature 447: 396–398.
Bjornsson HT, Sigurdsson MI, Fallin MD, Irizarry RA, Aspelund T, Cui H, Yu
W, Rongione MA, Ekstrom TJ, Harris TB, et al. 2008. Intra-individual
change over time in DNA methylation with familial clustering. JAMA
Bock C, Walter J, Paulsen M, Lengauer T. 2008. Inter-individual variation of
DNA methylation and its implications for large-scale epigenome
mapping. Nucleic Acids Res 36: e55. doi: 10.1093/nar/gkn122.
Bock C, Tomazou EM, Brinkman AB, Muller F, Simmer F, Gu H, Jager N,
Gnirke A, Stunnenberg HG, Meissner A. 2010. Quantitative comparison
of genome-wide DNA methylation mapping technologies. Nat
Biotechnol 28: 1106–1114.
Bocklandt S, Lin W, Sehl ME, Sanchez FJ, Sinsheimer JS, Horvath S, Vilain E.
2011. Epigenetic predictor of age. PLoS ONE 6: e14821. doi: 10.1371/
Boks MP, Derks EM, Weisenberger DJ, Strengman E, Janson E, Sommer IE,
Kahn RS, Ophoff RA. 2009. The relationship of DNA methylation with
age, gender and genotype in twins and healthy controls. PLoS ONE 4:
e6767. doi: 10.1371/journal.pone.0006767.
Breton CV, Byun HM, Wenten M, Pan F, Yang A, Gilliland FD. 2009. Prenatal
tobacco smoke exposure affects global and gene-specific DNA
methylation. Am J Respir Crit Care Med 180: 462–467.
Brunner AL, Johnson DS, Kim SW, Valouev A, Reddy TE, Neff NF, Anton E,
Medina C, Nguyen L, Chiao E, et al. 2009. Distinct DNA methylation
patterns characterize differentiated human embryonic stem cells and
developing human fetal liver. Genome Res 19: 1044–1056.
Burdge GC, Hanson MA, Slater-Jefferies JL, Lillycrop KA. 2007. Epigenetic
regulation of transcription: A mechanism for inducing variations in
phenotype (fetal programming) by differences in nutrition during early
life? Br J Nutr 97: 1036–1046.
Capittini C, Pasi A, Bergamaschi P, Tinelli C, De Silvestri A, Mercati MP,
Badulli C, Garlaschelli F, Sbarsi I, Guarene M, et al. 2009. HLA
haplotypes and birth weight variation: Is your future going to be light or
heavy? Tissue Antigens 74: 156–163.
ChenYA,Choufani S,Ferreira JC,Grafodatskaya D,ButcherDT,WeksbergR.
2011. Sequence overlap between autosomal and sex-linked probes on
the Illumina HumanMethylation27 microarray. Genomics 97: 214–222.
Choi JK, Kim SC. 2007. Environmental effects on gene expression
phenotype have regional biases in the human genome. Genetics 175:
Choufani S, Shapiro JS, Susiarjo M, Butcher DT, Grafodatskaya D, Lou Y,
Ferreira JC, Pinto D, Scherer SW, Shaffer LG, et al. 2011. A novel
approach identifies new differentially methylated regions (DMRs)
associated with imprinted genes. Genome Res 21: 465–476.
Christensen BC, Houseman EA, Marsit CJ, Zheng S, Wrensch MR, Wiemels
JL, Nelson HH, Karagas MR, Padbury JF, Bueno R, et al. 2009. Aging and
environmental exposures alter tissue-specific DNA methylation
dependent upon CpG island context. PLoS Genet 5: e1000602. doi:
C, Rho J, Loewer S, et al. 2009. Differential methylation of tissue- and
cancer-specific CpG island shores distinguishes human induced
pluripotent stem cells, embryonic stem cells and fibroblasts. Nat Genet
Du P, Kibbe WA, Lin SM. 2008. lumi: A pipeline for processing Illumina
microarray. Bioinformatics 24: 1547–1548.
Du P, Zhang X, Huang CC, Jafari N, Kibbe WA, Hou L, Lin SM. 2011.
Comparison of Beta-value and M-value methods for quantifying
methylation levels by microarray analysis. BMC Bioinformatics 11: 587.
Eden E, Navon R, Steinfeld I, Lipson D, Yakhini Z. 2009. GOrilla: A tool for
discovery and visualization of enriched GO terms in ranked gene lists.
BMC Bioinformatics 10: 48. doi: 10.1186/1471-2105-10-48.
Eisenberg E, Levanon EY. 2003. Human housekeeping genes are compact.
Trends Genet 19: 362–365.
Feinberg AP, Irizarry RA. 2011. Evolution in health and medicine Sackler
colloquium: Stochastic epigenetic variation as a driving force of
development, evolutionary adaptation, and disease. Proc Natl Acad Sci
(Suppl 1) 107: 1757–1764.
Feinberg AP, Irizarry RA, Fradin D, Aryee MJ, Murakami P, Aspelund T,
Eiriksdottir G, Harris TB,Launer L,Gudnason V,et al.2010. Personalized
epigenomic signatures that are stable over time and covary with body
mass index. Sci Transl Med 2: 49ra67. doi: 10.1126/
Foley DL, Craig JM, Morley R, Olsson CA, Dwyer T, Smith K, Saffery R. 2009.
Prospects for epigenetic epidemiology. Am J Epidemiol 169: 389–400.
Fraga MF, Ballestar E, Paz MF, Ropero S, Setien F, Ballestar ML, Heine-Suner D,
Cigudosa JC, Urioste M, Benitez J, et al. 2005. Epigenetic differences arise
during the lifetime of monozygotic twins. Proc Natl Acad Sci 102: 10604–
2011. Quantitative, high-resolution epigenetic profiling of CpG loci
identifies associations with cord blood plasma homocysteine and birth
weight in humans. Epigenetics 6: 86–94.
Gartner K, Baunack E. 1981. Is the similarity of monozygotic twins due to
genetic factors alone? Nature 292: 646–647.
Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, Ellis B,
Gautier L, Ge Y, Gentry J, et al. 2004. Bioconductor: Open software
development for computational biology and bioinformatics. Genome
Biol 5: R80.
Gertz J, Varley KE, Reddy TE, Bowling KM, Pauli F, Parker SL, Kucera KS,
Willard HF, Myers RM. 2011. Analysis of DNA methylation in a three-
generation family reveals widespread genetic influence on epigenetic
regulation. PLoS Genet 7: e1002228. doi: 10.1371/
Gervin K, Hammero M, Akselsen HE, Moe R, Nygard H, Brandt I, Gjessing
HK, Harris JR, Undlien DE, Lyle R. 2011. Extensive variation and low
heritability of DNA methylation identified in a twin study. Genome Res
Gluckman PD, Hanson MA, Beedle AS. 2007. Early life events and their
consequences for later disease: A life history and evolutionary
perspective. Am J Hum Biol 19: 1–19.
Gluckman PD, Hanson MA, Buklijas T. 2010. A conceptual framework for the
developmentaloriginsofhealthand disease.JDev OrigHealthDis1:6–18.
Gordon L, Joo JH, Andronikos R, Ollikainen M, Wallace EM, Umstad MP,
Permezel M, Oshlack A, Morley R, Carlin JB, et al. 2011. Expression
discordance of monozygotic twins at birth: Effect of intrauterine
environment and a possible mechanism for fetal programming.
Epigenetics 6: 579–592.
Heid IM, Jackson AU, Randall JC, Winkler TW, Qi L, Steinthorsdottir V,
Thorleifsson G, Zillikens MC, Speliotes EK, Magi R, et al. 2010. Meta-
analysis identifies 13 new loci associated with waist–hip ratio and
reveals sexual dimorphism in the genetic basis of fat distribution. Nat
Genet 42: 949–960.
Gordon et al.
Heijmans BT, Kremer D, Tobi EW, Boomsma DI, Slagboom PE. 2007.
Heritable rather than age-related environmental and stochastic factors
dominate variation in DNA methylation of the human IGF2/H19 locus.
Hum Mol Genet 16: 547–554.
Hochberg Y, Benjamini Y. 1990. More powerful procedures for multiple
significance testing. Stat Med 9: 811–818.
Huang K, Fan G. 2010. DNA methylation in cell differentiation and
reprogramming: An emerging systematic view. Regen Med 5: 531–544.
Irizarry RA, Ladd-Acosta C, Wen B, Wu Z, Montano C, Onyango P, Cui H,
Gabo K, Rongione M, Webster M, et al. 2009. The human colon cancer
methylome shows similar hypo- and hypermethylation at conserved
tissue-specific CpG island shores. Nat Genet 41: 178–186.
Javierre BM, Fernandez AF, Richter J, Al-Shahrour F, Martin-Subero JI,
Rodriguez-Ubreva J, Berdasco M, Fraga MF, O’Hanlon TP, Rider LG, et al.
2010. Changes in the pattern of DNA methylation associate with twin
discordance in systemic lupus erythematosus. Genome Res 20: 170–179.
Kaminsky ZA, Tang T, Wang SC, Ptak C, Oh GH, Wong AH, Feldcamp LA,
Virtanen C, Halfvarson J, Tysk C, et al. 2009. DNA methylation profiles
in monozygotic and dizygotic twins. Nat Genet 41: 240–245.
Katari S, Turan N, Bibikova M, Erinle O, Chalian R, Foster M, Gaughan JP,
Coutifaris C, Sapienza C. 2009. DNA methylation and gene expression
differences in children conceived in vitro or in vivo. Hum Mol Genet 18:
Kauffmann A, Gentleman R, Huber W. 2009. arrayQualityMetrics—a
bioconductor package for quality assessment of microarray data.
Bioinformatics 25: 415–416.
Kerkel K, Spadola A, Yuan E, Kosek J, Jiang L, Hod E, Li K, Murty VV, Schupf
N, Vilain E, et al. 2008. Genomic surveys by methylation-sensitive SNP
analysis identify sequence-dependent allele-specific DNA methylation.
Nat Genet 40: 904–908.
2008. Aberrant DNA methylation associated with bipolar disorder
VE, Chae SS, Lee MJ, et al. 2000. Edg-1, the G protein-coupled receptor
for sphingosine-1-phosphate, is essential for vascular maturation. J Clin
Invest 106: 951–961.
Lovering RC, Dimmer E, Khodiyar VK, Barrell DG, Scambler P, Hubank M,
Apweiler R, Talmud PJ. 2008. Cardiovascular GO annotation initiative
year 1 report: Why cardiovascular GO? Proteomics 8: 1950–1953.
Martin GM.2005.Epigenetic drift inaging identicaltwins. Proc Natl Acad Sci
Meaburn EL, Schalkwyk LC, Mill J. 2010. Allele-specific methylation in the
human genome: Implications for genetic studies of complex disease.
Epigenetics 5: 578–582.
Michaud J, Simpson KM, Escher R, Buchet-Poyau K, Beissbarth T,
Carmichael C, Ritchie ME, Schutz F, Cannon P, Liu M, et al. 2008.
Integrative analysis of RUNX1 downstream pathways and target genes.
BMC Genomics 9: 363. doi: 10.1186/1471-2164-9-363.
Morley R, Dwyer T. 2005. Studies of twins: What can they tell us about the
fetal origins of adult disease? Paediatr Perinat Epidemiol (Suppl 1) 19: 2–7.
Morley R, Dwyer T, Carlin JB. 2003. Studies of twins: Can they shed light on
the fetal origins of adult disease hypothesis? Twin Res 6: 520–525.
Neale MC, Cardon LR. 1992. Methodology for genetic studies of twins and
families. Kluwer Academic Publishers, Dordrecht.
Novakovic B, Wong NC, Sibson M, Ng HK, Morley R, Manuelpillai U, Down
T, Rakyan VK, Beck S, Hiendleder S, et al. 2010. DNA methylation-
mediated down-regulation of DNA methyltransferase-1 (DNMT1) is
coincident with, but not essential for, global hypomethylation in
human placenta. J Biol Chem 285: 9583–9593.
Novakovic B, Gordon L, Wong NC, Moffett A, Manuelpillai U, Craig JM,
Sharkey A, Saffery R. 2011. Wide ranging DNA methylation differences
of primary trophoblast cell populations and derived-cell lines:
Mol Hum Reprod 17: 344–353.
Ollikainen M, Smith KR, Joo EJ, Ng HK, Andronikos R, Novakovic B, Abdul
Aziz NK, Carlin JB, Morley R, Saffery R, et al. 2010. DNA methylation
intrauterine components to variation in the human neonatal
epigenome. Hum Mol Genet 19: 4176–4188.
Ozanne SE, Constancia M. 2007. Mechanisms of disease: The
developmental origins of disease and the role of the epigenotype. Nat
Clin Pract Endocrinol Metab 3: 539–546.
Plomin R. 2011. Commentary: Why are children in the same family so
different? Non-shared environment three decades later. Int J Epidemiol
Powell JE, Henders AK, McRae AF, Wright MJ, Martin NG, Dermitzakis ET,
Montgomery GW, Visscher PM. 2012. Genetic control of gene
expression in whole blood and lympho blastoid cell lines is largely
independent. Genome Res 22: 456–466.
Pritchard C, Coil D, Hawley S, Hsu L, Nelson PS. 2006. The contributions
of normal variation and genetic background to mammalian gene
expression. Genome Biol 7: R26. doi: 10.1186/gb-2006-7-3-r26.
R Development Core Team. 2009. R: A language and environment for
statistical computing. R Foundation for Statistical Computing, Vienna,
Rajendram R, Ferreira JC, Grafodatskaya D, Choufani S, Chiang T, Pu S,
Butcher DT, Wodak SJ, Weksberg R. 2011. Assessment of methylation
level prediction accuracy in methyl-DNA immunoprecipitation and
sodium bisulfite based microarray platforms. Epigenetics 6: 410–415.
Rakyan VK, Down TA, Maslau S, Andrew T, Yang TP, Beyan H, Whittaker P,
McCann OT, Finer S, Valdes AM, et al. 2010. Human aging-associated
DNA hypermethylation occurs preferentially at bivalent chromatin
domains. Genome Res 20: 434–439.
Rakyan VK, Beyan H, Down TA, Hawa MI, Maslau S, Aden D, Daunay A,
Busato F, Mein CA, Manfras B, et al. 2011a. Identification of type 1
diabetes–associated DNA methylation variable positions that precede
disease diagnosis. PLoS Genet 7: e1002300. doi: 10.1371/journal.
studies for common human diseases. Nat Rev Genet 12: 529–541.
Regard JB, Scheek S, Borbiev T, Lanahan AA, Schneider A, Demetriades AM,
Hiemisch H, Barnes CA, Verin AD, Worley PF. 2004. Verge: A novel
vascular early response gene. J Neurosci 24: 4092–4103.
Reik W. 2007. Stability and flexibility of epigenetic gene regulation in
mammalian development. Nature 447: 425–432.
Ritchie ME, Diyagama D, Neilson J, van Laar R, Dobrovic A, Holloway A,
Smyth GK. 2006. Empirical array quality weights in the analysis of
microarray data.BMC Bioinformatics 7: 261. doi: 10.1186/1471-2105-7-
Saffery R, Morley R, Carlin JB, Joo JH, Ollikainen M, Novakovic B,
Andronikos R, Li X, Loke YJ, Carson N, et al. 2012. Cohort profile: The
peri/post-natal epigenetic twins study. Int J Epidemiol 41: 55–61.
Sampath H, Ntambi JM. 2008. Role of stearoyl-CoA desaturase in human
metabolic disease. Future Lipidol 3: 163.
Sandoval J, Heyn HA, Moran S, Serra-Musach J, Pujana MA, Bibikova M,
Esteller M. 2011. Validation of a DNA methylation microarray for
450,000 CpG sites in the human genome. Epigenetics 6: 692–702.
Schalkwyk LC, Meaburn EL, Smith R, Dempster EL, Jeffries AR, Davies MN,
Plomin R, Mill J. 2010. Allelic skewing of DNA methylation is
widespread across the genome. Am J Hum Genet 86: 196–212.
Sharma A, Sharma VK, Horn-Saban S, Lancet D, Ramachandran S,
Brahmachari SK. 2005. Assessing natural variations in gene expression
in humans by comparing with monozygotic twins using microarrays.
Physiol Genomics 21: 117–123.
Shin S, Yoon JH, Lee HR, Hwang SM, Roh EY. 2010. Association of HLA-A, -B
and -DRB1 genotype with birthweight and CD34+ cell content: Analysis
of Korean newborns and their cord blood. Mol Hum Reprod 16: 338–346.
ShoemakerR,Deng J,WangW,ZhangK.2010. Allele-specificmethylationis
prevalent and is contributed by CpG-SNPs in the human genome.
Genome Res 20: 883–889.
Simonsen ML, Alessio HM, White P, Newsom DL, Hagerman AE. 2010.
Acute physical activity effects on cardiac gene expression. Exp Physiol
Smyth GK. 2004. Linear models and empirical Bayes methods for assessing
differential expression in microarray experiments. Stat Appl Genet Mol
Biol 3: 1544–6115.
Smyth GK. 2005. Limma: Linear models for microarray data. In
Bioinformatics and computational biology solutions using R and Bioconducto
(ed. R Gentleman et al.), pp. 397–420. Springer, New York.
Speliotes EK, Willer CJ, Berndt SI, Monda KL, Thorleifsson G, Jackson AU,
Allen HL, Lindgren CM, Luan J, Magi R, et al. 2010. Association analyses
of 249,796 individuals reveal 18 new loci associated with body mass
index. Nat Genet 42: 937–948.
Stromswold K. 2006. Why aren’t identical twins linguistically identical?
Genetic, prenatal and postnatal factors. Cognition 101: 333–384.
Teschendorff AE, Menon U, Gentry-Maharaj A, Ramus SJ, Weisenberger DJ,
Shen H, Campan M, Noushmehr H, Bell CG, Maxwell AP, et al. 2010.
Age-dependent DNA methylation of genes that are suppressed in stem
cells is a hallmark of cancer. Genome Res 20: 440–446.
Thorleifsson G, Walters GB, Gudbjartsson DF, Steinthorsdottir V, Sulem P,
et al. 2009. Genome-wide association yields new sequence variants at
seven loci that associate with measures of obesity. Nat Genet 41: 18–24.
Torche F, EchevarriaG. 2011. The effectof birthweight on childhood cognitive
development in a middle-income country. Int J Epidemiol 40: 1008–1018.
van Vliet J, Oates NA, Whitelaw E. 2007. Epigenetic mechanisms in the
context of complex diseases. Cell Mol Life Sci 64: 1531–1538.
Visscher PM. 2004. Power of the classical twin design revisited. Twin Res 7:
Methylation discordance in twins at birth
Warensjo E, Ingelsson E, Lundmark P, Lannfelt L, Syvanen AC, Vessby B,
Riserus U. 2007. Polymorphisms in the SCD1 gene: Associations with
body fat distribution and insulin sensitivity. Obesity (Silver Spring) 15:
Weisenberger DJ, Van Den Berg D, Pan F, Berman BP, Laird PW. 2008.
Comprehensive DNA methylation analysis on the Illumina Infinium Assay
Platform. Illumina, San Diego.
Whitelaw NC, Chong S, Whitelaw E. 2010. Tuning in to noise: Epigenetics
and intangible variation. Dev Cell 19: 649–650.
AL, Jackson AU, Lamina C, et al. 2009. Six new loci associated with body
mass index highlight a neuronal influence on body weight regulation.
Nat Genet 41: 25–34.
J. 2010. A longitudinal study of epigenetic variation in twins. Epigenetics
Zhang Y, Rohde C, Reinhardt R, Voelcker-Rehage C, Jeltsch A. 2009. Non-
imprinted allele-specific DNA methylation on human autosomes.
Genome Biol 10: R138. doi: 10.1186/gb-2009-10-12-r138.
Gershon ES, Liu C. 2010. Genetic control of individual differences in
gene-specific methylation in human brain. Am J Hum Genet 86: 411–419.
Received December 16, 2011; accepted in revised form April 26, 2012.
Gordon et al.
1406 Genome Research