DOI: 10.1161/CIRCGENETICS.114.000804
DNA Methylation of Lipid-Related Genes Affects Blood Lipid Levels
Running title: Pfeiffer et al.; genome-wide DNA methylation and blood lipid levels
Liliane Pfeifferm, MSc1,2; Simone Wahl, MSc1-3; Luke C. Pilling, MSc4; Eva Reischl, PhD1,2;
Johanna K. Sandling, PhD5,6; Sonja Kunze, PhD1,2; Lesca M. Holdt, MD, PhD7; Anja
Kretschmer, PhD1,2,8; Katharina Schramm, PhD9,10; Jerzy Adamski, PhD11; Norman Klopp,
PhD12; Thomas Illig, PhD12; Åsa K. Hedman, PhD13,14; Michael Roden, MD, PhD15-17; Dena G.
Hernandez, MSc18; Andrew B. Singleton, PhD18; Wolfgang E. Thasler, MD19; Harald Grallert,
PhD1-3; Christian Gieger, PhD20; Christian Herder, PhD15,16; Daniel Teupser, MD7; Christa
Meisinger, MD2; Timothy D. Spector, MD, FRCP21; Florian Kronenberg, MD22; Holger
Prokisch, PhD9,10; David Melzer, MBBCh, PhD4; Annette Peters PhD
hD
D1,2,23; Panos
Paano
n s Deloukas,
Delo
De
louuk
lo
5,24,25
26
11,2
1,
,2
PhD
; Luigi Ferrucci, MD, PhD ; Melanie Waldenberger,
b rger
be
er,, Ph
er
PhD
D
1
Rsrch Unit of Molecular
Molecu
cula
ulaar Epidemiology,
Ep
pid
i em
mio
ology, 2Inst of Epidemiology II,
II 9Inst of Human
Human Genetics,
Genetics 11Genome
me Analysis Ctr,
Ctr Inst of Experim
Experimental
Genetics, 20Inst of
o Genetic
Gen
enet
etiic Epidemiology,
Epi
pide
pi
d miol
olog
ol
ogy, Helmholtz Zentrum München, Germa
og
German
man Rsrch Ctr for Environmental
Environm
En
nm
mental Health; 3German Ctr for Diabetes
4
5
N uheerberg, Ger
Neu
rmanyy; Epidem
Rsrch (DZD), Neuherberg,
Germany;
Epidemiology
emio
em
iolo
io
l gy & Public
P blic Health Group,
Pu
Groupp, Univ
Univ off Exeter
Exe
x ter Medical
M dical School,
Me
Scc
Exeter;
Exet
Ex
eter
et
e ; Well
Wellcome
lllco
c mee Trus
Trust
s Sanger
Inst, Wellcome T
Trust
Hinxton,
UK;
russt Genome Campus,
Camp
puss, Hi
Hinx
xto
ton,
n U
K; 6Pr
Present:
resen
e t: D
Dept
ept of
of Medical
Meddical Sc
Sciences,
cieencees, Mo
Molecular
oleccular Medici
Me
Medicine
e ine & Sc
Science
cie
iencee fo
for
or Li
Life
ife Lab
Laboratory,
b
14
Present: Dep
D
Dept
pt ooff Medical Sciences,
Scien
ncees, Molecular
Molleculaar E
Epidemiology
pidemioloogyy & Science
Sciiencee fo
for Life
fee L
Laboratory,
a oratooryy, Upp
ab
Uppsala
psaala Un
Univ,
niv, Upp
Uppsala,
psaala, Sweden;
Sw
wedeen;; 7In
Inst
n of
e ic
edic
icin
ne,
e Univ Hospital
Hoosp
s itaal Munich
Munich
h & Ludwig
Ludw
Ludw
Lu
dwig Maximilians
Max
a im
milianss Univ
Univ Munich,
Munich
Mu
c , Munich;
ch
Muunich; 8D
Laboratory Medicine,
Dept
ept off D
Dermatology,
erm
mattology,, V
Venereology
eneereeologgy & A
Allergy,
Christian Albrechts
Univ
Kiel,
b ht
brecht
htss Un
niv
v Ki
Kiel
e , Ki
el
Kiel;; 10In
Inst of
of Human
Hum
uman
um
n Genetics,
Genetic
iccs, Technical
Tec
echnnic
ical
al Univ
Uni
nivv Munich,
ni
Muni
Mu
nich
ni
c , Mu
ch
M
Munich;
n ch
ni
ch;; 12Ha
Hannover
H
nnovver U
nn
Unified
niifi
fieed B
Biobank,
ioba
bank
ba
nkk, Ha
Han
Hannover
n
Medical School, Hannover, Germany; 13We
Wellcome
W
llcome Trust Ctr for Human Genetics, Univ off Oxford,
Oxford, Oxford,, UK; 15German Ctr for Diabetes
Diabee Rsrch
(DZD); 16Inst forr Clinical Diabetology
Diabetology,
y, German
Germ
Ge
rm
man
a Diabetes
Dia
iabe
bete
tees Ctr,
Ctr, Leibniz
Lei
eibn
bniz
bn
iz Ctr
Ctr for
for Diabetes
Diab
Di
abet
ab
e es
et
e Rsrch
Rsrcch at Heinrich
Heiinr
nric
i h Heine
ic
H in
He
ne Un
Univ; 17Dept of Endocrinology
Endocrii
&
18
8
19
Diabetology, Univ
Düsseldorf,
U
Hospital, Düssel
ldo
dorf
r , Germany;
rf
Germ
Ge
rm
man
any
y; La
Laboratory
Labo
bora
bo
raato
tory
ry of
of Neurogenetics,
N urrog
Ne
ogen
en
net
etic
icss, Nat
ic
Naat Inst
Inst on Aging,
Agi
ging
n , NIH,
ng
NIH
H, Bethesda,
Beth
Be
t esda, MD; Biobankk under
th
Administrationn of
Transplantation,
Munich,
of HTCR,
HTCR
HT
CR,, Dept
CR
Dep
De
pt of General,
Gen
ener
eral
er
al,, Visceral,
al
Visc
Vi
scer
sc
eral
er
al,, Tr
al
Tran
ansp
an
spla
sp
lant
la
ntat
nt
atio
at
ion,
io
n, Vascular
Vas
ascu
cula
cu
larr & Thoracic
la
Thor
Th
orac
or
acic
ac
ic Surgery,
Sur
urge
gery
ge
ry,, Hospital
ry
Hosp
Ho
spit
sp
ital
it
al of
of the
the Un
Univ
iv of Munich,
Mun
unic
ich,
ic
h, M
2
Germany; 21Dept
pt of Twin Rsrch
Rsrcch & Genetic
Geene
Gene
neti
ticc Epidemiology,
ti
Ep
E
pid
dem
emio
iology
iolo
lo
ogy
gy, King’s
K ng
Ki
ng’ss College
Col
olle
lege
le
gee London,
Lon
ondo
don,, London,
do
Lon
ondo
don,
do
n, UK;
UK; 22
Division
D
Di
Divi
iviisi
sion
onn of
of Genetic
Gene
Ge
neeti
ticc Epidemiology,
E idemiology, Dept of
Ep
23
Medical Genetics,
Molecular
Clinical
Pharmacology,
Medical
Univ
Innsbruck,
Austria;
cs,, Mo
cs
Mole
lecu
le
cula
larr & Cl
la
Clin
inic
in
ical
ic
al P
harm
ha
rmac
acol
olog
ol
ogy,
gy
y, M
edic
ed
ical
ic
al U
nivv of IInnsbruck,
ni
nnsb
nn
sbru
sb
ruck
ck,, In
ck
Inns
nsbr
bruc
br
uck,
k, A
ustr
us
tria
ia;; Ge
ia
German
Germ
rman
an Rsrch
Rsrc
rchh Ctr
Ctr for
for Cardiovascular
Card
Ca
rdio
rd
io
o
Disease (DZHK), Partner-site Munich, Germany; 24William Harvey Rsrch Inst, Barts & The London School of Medicine & Dentistry, Queen
Mary Univ of London, London, UK; 25Princess Al-Jawhara Al-Brahim Ctr of Excellence in Rsrch of Hereditary Disorders (PACER-HD), King
Abdulaziz Univ, Jeddah, Saudi Arabia; 26Clinical Rsrch Branch, Nat Inst on Aging, Baltimore, MD
Correspondence:
Melanie Waldenberger PhD
Research Unit of Molecular Epidemiology & Institute of Epidemiology II
Helmholtz Zentrum München, Ingolstaedter Landstraße 1
D-85764 Neuherberg, Germany
Tel: +49-89-3187-1270
Fax: +49-89-3187-4567
E-mail: waldenberger@helmholtz-muenchen.de
Journal Subject Codes: [135] Risk factors, [142] Gene expression, [90] Lipid and lipoprotein
metabolism, [8] Epidemiology
1
DOI: 10.1161/CIRCGENETICS.114.000804
Abstract:
Background - Epigenetic mechanisms might be involved in the regulation of inter-individual
lipid level variability and thus may contribute to the cardiovascular risk profile. The aim of this
study was to investigate the association between genome-wide DNA methylation and blood lipid
levels HDL-C, LDL-C, triglycerides (TG) and total cholesterol (TC). Observed DNA
methylation changes were further analyzed to also examine their relationship with previous
hospitalized myocardial infarction.
Methods and Results - Genome-wide DNA methylation patterns were determined in whole
blood samples of 1776 subjects of the KORA F4 cohort using the Infinium
HumanMethylation450 BeadChip (Illumina). Ten novel lipid-related Cp
CpG
G sites
site
si
tess (CpGs)
te
(CpG
(C
pGss)
pG
s)
annotated to various genes including ABCG1, MIR33B/SREBF1 and TNIP11 we
were
re iidentified.
dent
de
ntif
nt
ifie
if
i d CpG
ie
cg06500161,
with
1, located
loca
lo
catte
ca
ted in AB
ted
ABCG1, was associated in opp
oopposite
pposite directionss wi
w
th both HDL-C
ȕ coefficient=-0.049,
p=8.26E-DQG7*OHYHOVȕ=0.070,
Eight
associations
ent=
en
=-0.049, p=
=8.
8 266EE-
DQ
DQG
G7*
G
7*
* OHYHO
H Vȕ
ȕ=0.0070
70, pp=1.21E-27).
=1.
1 21
21E
E-227).
). E
Ei
ighht as
asso
soociiat
ationn
were confirmed
InCHIANTI
(N=472).
Associations
rmed by replication
rm
repl
pliccattionn in
pl
in KORA
K RA F3
KO
F3 (N=499)
(N=4
(N
=499) and
an
nd InCH
C IA
IANT
TI (N
N=4
=472).
). A
sssocia
iatioo
between TG
SREBF1
ABCG1
were
found
tissue
G levels and SR
REB
EBF1 aand
nd ABCG
G1 w
e e al
er
aalso
lso
so fou
o nd
n in adipose
adip
ad
posse tiss
ssue
ss
u of the MuTHER
MuTH
H
cohort (N=634).
Expression
analysis
revealed
association
ABCG1
methylation
6 ). Expr
634)
p ession analy
lysis
i reveal
led
d an asso
ciiation between
between AB
A
CG1 methyl
CG
y ation and
lipid levels that
mediated
ABCG1
expression.
methylation
ABCG1
tha
hatt might
migh
mi
ghtt be
gh
b partly
par
artl
tlly me
medi
diat
di
atted bby
y AB
A
CG
G1 ex
expr
p essi
pres
sion
si
on.. DNA
DN
NA me
meth
thyl
th
y attio
ylat
yl
i n of AB
ABC
C
might also play a role in previous hospitalized myocardial infarction (odds ratio 1.15,
95%CI=1.06-1.25).
Conclusions - Epigenetic modifications of the newly identified loci might regulate disturbed
blood lipid levels and thus contribute to the development of complex lipid-related diseases.
Key words: epidemiology, epigenetics, gene expression, genetic epidemiology, lipids and
lipoproteins, Illumina 450K, ABCG1, expression, EWAS, myocardial infarction
2
DOI: 10.1161/CIRCGENETICS.114.000804
Introduction
Coronary artery disease (CAD) is a major cause of death in industrialized countries1. Blood lipid
levels, including high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein
cholesterol (LDL-C), triglycerides (TG) and total cholesterol (TC) levels, are considered
heritable, modifiable risk factors for this disease2.
Lipid levels can be influenced by drug therapy or lifestyle factors such as diet, physical
activity, alcohol consumption and smoking3. Several studies have also revealed a genetic impact
on disturbed blood lipid levels. Genome-wide association studies identified
tified a total of 157
577 genetic
g
loci associated with lipid levels, explaining up to 12% of trait variancee4. Beyo
Beyond
y ndd this
thi
his there
th
her
eree is
h ep
hat
epig
ig
gen
enet
etti mechanisms
etic
mechanisms are also invol
lved in inter-individ
id
dual lipid level variab
b
evidence that
epigenetic
involved
inter-individual
variability
and thus may
ay contribute
ay
contribute too thee cardiovascular
cardi
d ovasscullar ri
di
risk
iskk pro
profile.
rofilee. One
O e epigenome-wide
On
epigeno
noome
m -wide
de aanalysis
nallysiis inn
na
t fami
th
mili
mi
ili
lial
al hyp
yper
erch
ch
hol
oles
le tero
role
ro
lemia id
le
iden
enti
en
tifi
ti
fied
fi
ed
d TN
T
NT11 DNA
NT
DNA me
DN
meth
thyllat
th
atio
ion le
io
llevels
vels
ve
l tto
ls
o be
b
patients with
familial
hypercholesterolemia
identified
TNNT1
methylation
associated with
w HDL-C levels
l 5. Another
A othe
An
h r epigenome-wide
ep
pig
i enome-wiide
d analysis
analy
lysis
i in a non-population-based
non-p
poppulation-basee
cohort observed
er ed
d an association
iatii bet
bbetween
et een CP
CPT1A
CPT1
T1A
AD
DNA
NA meth
methylation
ethh lation
latii lle
levels
el
els
l andd very
er llo
low ddens
density
lipoprotein cholesterol (VLDL-C) as well as TG levels6.
The aim of this study was to systematically investigate the association between main
blood lipid levels (HDL-C, LDL-C, TG, TC) and genome-wide DNA methylation in whole
blood of a large population-based cohort as well as in adipose tissue and skin samples. The
identified associations were further explored through expression and functional studies and by
investigation of genetic confounding. Finally, the relationship between observed DNA
methylation changes and previous hospitalized myocardial infarction (MI) was explored.
Methods
The KORA study (Cooperative health research in the Region of Augsburg) consists of
3
DOI: 10.1161/CIRCGENETICS.114.000804
independent population-based samples from the general population living in the region of
Augsburg, Southern Germany. The study has been conducted according to the principles
expressed in the Declaration of Helsinki. Written informed consent has been given by each
participant. The study was reviewed and approved by the local ethical committee (Bayerische
Landesärztekammer). For the analysis, whole blood samples of the KORA F4 study were used
(N=1776). The replication was done in whole blood samples of KORA F3 (N=499) and
InCHIANTI (N=472), as well as in human adipose (N=634) and skin (N=395) samples of the
Multiple Tissue Human Expression Resource (MuTHER) study. In thee discovery
the
y and in
n th
replication cohorts genome-wide DNA methylation patterns were analyzed
Infinium
lyzedd uusing
siing tthe
he Inf
nfin
nf
ini
in
HumanMethylation450
Array (Illumina).
t yla
thy
lati
tion
ti
on45
on
4500 BeadChip
45
Be
) In
In KORA F4 andd InCHIANTI
InCHIANTI the analysis
anall
was performed
whole
blood
fasting
F33 non-fasting
med using whol
ole blo
ol
oodd DNA
D A of fa
DN
fasti
ingg participants;
parrticcip
ipan
antss; in
an
n KORA
KOR
ORA
OR
AF
noon-ffasstin
ng
participantss were
included.
KORA
morning
10.30
re aalso
l o in
ls
ncllud
uded
ded
d. In
I K
ORA bl
OR
bblood
oodd wa
oo
wass drawn
draw
dr
wn in the
he m
orni
ning
i (8
(8 – 10
0.30
30 am)
am and
stored at -80°C
analysis.
Beta-mixture
normalization
0°C until analys
y i s. B
eta-miixture qu
qquantile
antiile normali
izatiion (BMIQ)
(BM
MIQ
IQ))7 was applied
app
pplied too the
pp
DNA methylation
llation
atiio ddata
ata using
sing
in th
the
h R package
ck
k
wateRmelon,
ateRmelon
teR
R lo version
ersion
sii 11.0.3
0 38. Suppl.
S ppll Table
Tabl
ble 1 provides
pro
a summary of normalized beta values of the identified lipid related CpGs in KORA F4. KORA
F4/F3 samples were processed on 20/7 96-well plates in 9/4 batches; plate and batch effects were
investigated using principle component analysis and eigenR2 analysis9. The plate variable
explained 4.8% (F4), 6.3% (F3) and 8.1% (InCHIANTI) of variance in the DNA methylation
data. Consequently, plate was included as a random effect in the analyses.
Lipid levels were determined in fasting fresh blood samples at most six hours after
collection, except for KORA F3 which also includes non-fasting samples. In KORA F3 and F4
TC was measured using the cholesterol-esterase method (CHOL Flex, Dade-Behring, Germany).
HDL-C and TG levels were determined using the TGL Flex and AHDL Flex methods (Dade-
4
DOI: 10.1161/CIRCGENETICS.114.000804
Behring, Germany), respectively, and LDL-C was measured by a direct method (ALDL, DadeBehring, Germany). In KORA F4/F3 the intra-assay coefficient of variation (CV) for repeated
measurements was 1.85%/1.61% (TC), 2.75%/2.65% (TG), 3.25%/2.89% (HDL-C) and
2.7%/3.02% (LDL-C). In InCHIANTI TC was determined by the cholesterol-esterase method,
HDL-C was measured with the Liquid Homogeneous HDL-C assay (Alifax S.p.A., Padova,
Italy), and TG through an enzymatic colorimetric test using lipoprotein lipase, glycerokinase,
glycerol phosphate oxidase and peroxidase. All three lipids were determined using the analyzer
Modular P800 Hitachi (Roche Diagnostics, Mannheim, Germany). Thee intra-assay CV was
was
a 0.8%
(TC), 1.5% (TG) and 0.8% (HDL-C). The level of LDL-C was calculated
Friedewald
ated uusing
siing th
the F
rie
iede
dew
de
formula (LDL-C=TC-[HDL-C+(TG/5)]).
D -C=
DLC TC
T -[H
[ DL
D -C+(TG/5)]).
Detailed
methylation
analysis
ailed
ai
iled informat
information
tioon aabout
boutt tthe
bo
he ccohorts,
ohor
orts, th
tthe
he pprocess
rooces
ess of DNA
N m
met
ethy
hyyllation
ationn ana
nallysi
na
sis
is as well
as data preprocessing,
processing,
proce
esssin
ingg, quality
qua
uali
lity
li
ty assessment
ass
ssessm
ment and
an
nd further
furt
rthe
the
h r me
meth
methods
th
hod
odss such
suuch
h as
as genotyping
genoty
ge
typi
ty
piingg and
andd ggene
ene
en
expression are pprovided
the
supplemental
rovided in th
he sup
pplemental
pll
l material.
material
i l.
Statistical analysis
l i
Discovery step
In the KORA F4 cohort, 26 of 1802 individuals were excluded from the analysis due to missing
information in covariates or due to non-fasting status at the time point of blood collection,
resulting in a final sample size of 1776 F4 subjects. Associations between DNA methylation beta
values and lipid levels were analyzed using linear mixed effects models implemented with the
nlme package in R with lipid levels as response. To normalize lipid levels, square root (TC and
LDL-C) and logarithmic (HDL-C and TG) transformations were applied, followed by
standardization to a mean of zero and a standard deviation of one. The following potential
confounders were included as covariates: age, sex, body mass index, smoking, alcohol
5
DOI: 10.1161/CIRCGENETICS.114.000804
consumption, intake of lipid-lowering drugs, physical activity, history of myocardial infarction,
current hypertension, HbA1c levels, C-reactive protein levels and white blood cell count.
Experimental plate was included as a random effect. To correct for multiple comparisons, a
genome-wide significance level of 1.1E-07 was used, determined according to the Bonferroni
procedure. Since whole blood DNA samples were used, cell heterogeneity had to be considered
as a confounder. As no measured cell count information was available for any cohort, samplespecific estimates of the proportion of the major white blood cell types were obtained using a
statistical method described by Houseman et al.10. The significant associations
ociations of the firs
first
rsst m
model
were recalculated, additionally adjusting for the estimated white bloodd celll pr
pproportions
oporti
tions (C
ti
(CD
(CD8 Tcells, CD4 T-cells,
T ce
T-ce
cell
llss, nnatural
ll
attural
atu
al killer cells, B-lymphocy
B-lymphocytes,
yte
tess, monocytes andd granulocytes). To get
gee a
measure of th
the
he variance in thee lipi
llipid
i id llevels
eveels explained
explaain
ned bby
y me
methylation
eth
thyylattio
on le
levels,
evels
ls,, R2 statistics
ls
sttatiisticss were
we
calculated accor
aaccording
rdi
ding
ing to Ed
Edwa
Edwards
ward
rds
ds et aal.
l.111, using
usin
i g the
in
th R package
pack
pa
ckaag
ck
age ppbkrtest,
age
bkr
krte
kr
test
te
stt, version
versio
ionn 0.3io
0.3-7.
3 7.
Replication
n step
p
Identified loci
the
ocii were
ere replicated
li tedd using
sing
in th
h same statistical
tatisti
i icall model
oddell in
i KORA
KORA F3
F3 (N=499)
(N 499)
499
99)) as well
e as
in InCHIANTI (N=472). In KORA F3 an adjustment for C-reactive protein was not possible
since this variable was not available for this cohort. A fixed-effects meta-analysis of KORA F3
and InCHIANTI results was conducted with the R package metafor, version 1.9-2. Results were
corrected according to the Bonferroni procedure (level of significance=4.5E-03).
For the MuTHER cohort the Infinium HumanMethylation450 BeadChip Array signal
intensities were quantile normalized and methylation beta values were calculated using R 2.12 as
previously described12. For cg06500161 no DNA methylation data were available as it did not
pass the quality control filters. Data for N=634 adipose and N=395 skin samples were available
for the final analysis. A linear mixed effects model was fitted for blood lipid values using the
6
DOI: 10.1161/CIRCGENETICS.114.000804
lme4 package in R. The model was adjusted for age, BMI, smoking, statins, technical covariates
(fixed effects) and family relationship and zygosity (random effects). A likelihood ratio test was
used to assess significance, and the p-value was calculated from the Chi2 distribution with 1df
using -2log (likelihood ratio) as the test statistic. Results were corrected according to the
Bonferroni procedure (level of significance=7.14E-03).
SNP analysis
Investigation of genetic confounding was carried out to identify whether the observed
associations between lipid and methylation levels in KORA F4 were due to singl
single
nucleotide
g e nucleo
eoti
eo
td
polymorphisms (SNPs) being associated with both lipid levels and DNA
methylation.
NA met
thyla
l ti
tion. 15
1577 llipidassociated SNPs
Genetics
S Ps identified
ide
dent
nttif
i ieed by the Global Lipids Gen
enet
en
e ics Consortium
m were
were included in the
analysis4. SNP
SN
NP rs94114899 was
wass excluded
excllud
u ed
d because
becaausee genotype
gen
notyppe ddata
ata w
were
ere no
not
ot av
available
vailaabl
ble for
fo
or th
the
he KORA
K
Genot
ottyp
ypee da
ata
t off 156
156
56 lipid-associated
lip
ipid
ip
id
d-associ
ciat
ci
ated
at
ted
e SN
SNP
Ps as well
Ps
wel
ell
el
ll as DNA
DNA
A met
ethy
et
hyllatiion data
hy
datta w
F4 dataset. Genotype
data
SNPs
methylation
were
available for
o 1710 KORA F4
or
F pparticipants.
artiicipa
ip nts. A pre-selection
p e-sele
pr
l ctiion was done
done to reveal the lipi
lipid-associated
p d-assoo
SNPs whichh were
ere at th
the
he same ti
time
i nominall
nominally
mii ll associated
ciat
i ed
d ((p<0
(p<0.05)
p<00 005)
5) with
iith
th
h di
differentiall
diff
differentially
ff
tiiall
ll meth
methylated
ethh
lipid-related CpG sites (CpGs; Suppl. Table 2). Next, models for each significant CpG–lipid pair
were recalculated with additional adjustment for the respective pre-selected SNPs to see if the
association was based on genetic confounding. Discovery, replication step and SNP analysis
were analyzed using the statistical package R, version 2.15.3.
Gene expression analysis
For the gene expression analysis, 724 KORA F4 subjects were included, as for these participants
both DNA methylation data and expression data were available. We tried to disentangle the
relationships between methylation at the CpGs, expression of the corresponding annotated gene
and lipid levels in an ad hoc approach based on a sequence of regression models with and
7
DOI: 10.1161/CIRCGENETICS.114.000804
without adjusting for the third of the three components. For each significant lipid-methylation
pair, the association between lipid level and DNA methylation was recalculated for KORA F4
(N=724). Afterwards we repeated the analysis, adjusting for the expression levels of the
annotated gene (except for cg07504977 which has no annotation to a gene according to the
UCSC genome Browser) (Suppl. Table 3). A p-value for the association was determined through
a likelihood ratio test. Similarly, the association between DNA methylation and transcript levels,
and between lipid levels and transcript levels, were determined. All models were also adjusted
for age, sex, body mass index, smoking, alcohol consumption, intake of lipid lowering ddrugs,
rug
ru
physical activity, history of myocardial infarction, current hypertension,
HbA1c
n, H
bA11c llevels
bA
evels
l an
andd Creactive protein
well
o in llevels,
otein
evel
ev
e s, ass w
el
ell as for white blood cell
ceell
l count and estimated
estim
mated white blood cell
proportions.
technical
s Models
s.
Models including
includ
udingg expression
ud
expr
preession
pr
on data
data were
were aadditionally
we
dddit
itio
iona
io
n llyy adjusted
na
adju
ustted
d for
for the
he tec
echhnic
ec
i al
variables RNA
Integrity
Number
sample
storage
RNA In
Inte
nte
teggrity Nu
N
mb
ber ((RIN),
RIN)
RI
N), samp
N)
mple
mp
le sto
ora
rage
ge ttime
ime
im
me and
andd RN
RNA
A aamplification
mpl
plif
pl
ifiicattio
if
i n ba
bbatch
tch13.
tch
The level off sig
significance
gnificance was set to 88.3E-04.
.33EE 04
04.
Association off DNA
DNA methylation
th
h l tii with
ith
ith
h prevalent
l t myocardial
di l infarction
i f tii
To assess the association of the observed lipid-related CpGs with previous hospitalized MI in
KORA F4, generalized linear mixed effects models were fitted with adaptive Gauss-Hermite
quadrature using the R package lme4, version 1.0-4. Three models were analyzed. The first
model was adjusted for age, sex and estimated white blood cell proportions. In the second model
we additionally included body mass index, smoking, alcohol consumption, physical activity,
current hypertension, HbA1c levels, C-reactive protein levels and white blood cell count as
covariates, and in the third model the lipid variables (HDL-C, LDL-C, TG, TC) were also
included. The Bonferroni correction was used with a significance level of 6.3E-03. The same
analyses were done for KORA F3 and InCHIANTI. This statistical analysis and the gene
8
DOI: 10.1161/CIRCGENETICS.114.000804
expression analysis were performed using the statistical package R, version 3.0.2.
Results
Associations between genome-wide DNA methylation and blood lipid levels
Characteristics of the discovery cohort (KORA F4) as well as the replication cohorts (KORA F3,
InCHIANTI, MuTHER cohort) are shown in Table 1.
In KORA F4, DNA methylation levels at one, 68, 17 and 80 CpGs were associated with
HDL-C, TG, LDL-C and TC levels, respectively. When white blood cell pproportions
were
p
included as covariates, the number of significant associations (p<1.1E-07)
indicating
-07) ddecreased,
ecre
ec
reas
re
ased
as
ed,, in
ed
indi
dicc
di
the presencee of bl
with
bloo
blood
ood cell
oo
c lll cconfounding.
ce
onfounding. The association
associatio
ion of methylation
onn level at one CpG w
HDL-C and
significant,
dL
LDL-C
DL-C remained
rem
maine
nedd si
sign
gnif
nificant,
ific
if
ican
ic
ant,
an
t as
as well
weelll as the
th
he association
assoociat
as
atio
on of
o 10 CpGs
CpGs with
wit
i h TG levels.
le
There were noo lo
with
TC.
llonger
nger
ng
er aany
ny as
aassociations
soci
so
c at
ci
a io
onss w
ithh TC
C. P-values
P-va
v lu
ues
e rranged
ange
an
geed fr
from
o 11.21E-27
om
. 1E
.2
1E-2
-2
27 to
o 99.66E-08
.66E
.6
6 -0
6E
0 with
percentage of explained li
lipid
level
variance
ipi
pidd le
leve
veel va
vari
rian
ri
a ce rranging
an
angi
an
g ng
gi
n ffrom
rom
ro
m 1.
11.6%
6% ttoo 6.5%
6 5%
6.
% (Table
(Ta
Tabble 2). CpG
cg06500161,
ABCG1,
associated
directions
with
HDL-C
(ȕ=-0.049,
11,, located
loca
lo
cate
tedd in
i AB
A
CG11, was
CG
was as
asso
soci
ciat
iated
ed
d in
in oopposite
ppos
pp
osit
itte di
dire
rect
ctio
ions
ns w
ithh HD
it
HDL
L-C
C ((ȕ=-0.0
ȕ=-0.
0.00
0.
p=8.26E-17) and TG levels (ȕ=0.070, p=1.21E-27). TG levels were associated with nine
additional CpGs located in genes including ABCG1, MIR33B, SREBF1 and CPT1A. LDL-C
showed a positive association with methylation status of one CpG located in TNIP1 (ȕ=0.040,
p=4.27E-09).
The lipid-related CpGs were carried forward to replication in a meta-analysis of the
KORA F3 and InCHIANTI cohorts. Nine of the twelve associations were confirmed (p-values
from 9.00E-11 to 3.78E-03; Table 2).
Tissue expression of candidate genes and replication in an adipose tissue cohort
To address cell- and tissue-specificity of ABCG1, CPT1A and SREBF1 expression, we quantified
their expression in human blood cell types (peripheral blood mononuclear cells (PBMC), CD14-,
9
DOI: 10.1161/CIRCGENETICS.114.000804
CD19-, CD3-, CD4-, CD8-positive cells, and regulatory T-cells) and human tissues (brain, heart,
lung, kidney, small intestine, adipose tissue, skeletal muscle) (Suppl. Figure 1). All genes were
expressed not only in blood cells but also adipose tissue. Five of the replicated associations were
also significant in adipose tissue of the MuTHER cohort (Table 3). Here, the CpG cg20544516
(MIR33B/SREBF1) showed the strongest association with TG levels (ȕ=0.012, p=1.20E-10),
followed by CpGs located in ABCG1 (cg27243685, cg07397296; ȕ=0.013, p=5.86E-08 and
ȕ=0.008, 6.59E-07, respectively) and SREBF1 (cg11024682, ȕ=0.007, p=6.72E-04). The
association between LDL-C and cg22178392 (TNIP1) was also found to be significant in adipose
a
tissue (ȕ=0.002, p=6.02E-03). In skin tissue no associations could be determined
determiinedd except
pt ffor
or TG
levels and cg110
(SREBF1)
ccg11024682
1024
10
2468
24
6822 (SR
68
SREBF1) and cg005749588 (CPT1A)
SR
(CP
C T1A) (ȕ=0.006,
(ȕ=0.006
066, p=4.07E-04 and ȕ=ȕ=
=
0.005, p=2.81E-03,
81E
E-03, respectively).
respeecttiveely
y). These
Thesee results
ressults are
are iinn line
lin
ne with
with the
the strong
strrongg expression
expres
e sion
es
on off
MIR33B/SREBF1
R BF
REBF
F1 in
in adipose
adip
i os
ip
osee tissue
tiss
ti
ssue observed
observed
bss
d in our
ourr tissue
tiss
issuee panel
pane
pa
nell (Suppl.
ne
(Sup
(S
uppll. Figure
uppl
F gu
Fi
gure
re 1).
Investigation
confounding
on of ggenetic
enetic conf
foundi
ding
i g
Known lipid-related
nominally
the
d related
latedd SN
SNPs
SNP
Ps which
hich
hich
h were
ere nominall
mii ll associated
ciat
i ed
d with
iith
thh DN
DNA
A methylation
meth
th
h llation
atiio at th
he
identified lipid-related CpGs and thus acting as potential confounders, are shown in Suppl. Table
2. P-values ranged between 4.99E-02 and 3.42E-05, except for one SNP (rs964184), which was
highly significantly associated with DNA methylation of cg12556569 (APOA5, p=3.75E-289).
After recalculation of the models for each CpG-lipid pair with additional adjustment for the
respective pre-selected SNP, only the association between methylation of APOA5 and TG was
considerably genetically confounded (Suppl. Table 4). Further analyses showed that rs964184,
from the pre-selected SNPs for APOA5, caused the genetic confounding (data not shown).
Gene expression analysis
Gene expression analysis revealed a negative association between methylation at ABCG1 and
10
DOI: 10.1161/CIRCGENETICS.114.000804
mRNA levels of the six ABCG1 transcripts (cg06500161: p=5.42E-14, cg27243685: p=1.86E07). Here, transcript ILMN_2329927 showed the highest association (cg06500161: ȕ=-0.151,
p=5.22E-15, cg27243685: ȕ=-0.185, p=4.34E-06) of the six ABCG1 transcripts. Adjusted for
lipids the association became less significant. Detailed results of the analyses are shown in
Suppl. Table 5.
The association between cg06500161 and HDL-C (p=4.10E-07) weakened (p=1.30E- 02)
after adjusting for ABCG1 transcripts. Similar results were observed for the association between
cg06500161 and TG levels, and between cg27243685 and TG levels.
ABCG1 transcript levels showed a strong positive association with HDL-C
HDL
HD
L-C
C (p=7.76E-13)
(p=7
(p
=7
7.776E
and a negative
(p=1.25E-33).
t e aassociation
tive
ssoc
ss
ocia
oc
iaati
t on
n with TG levels (p=1.25E
E-3
-33). Significancee was
was reduced when
adjusting for
ABCG1
DNA
methylation
also
Figure
Suppl.
or A
or
BCG1 DN
NA m
etthylat
yllat
ationn (see
(seee als
so Fig
gure 1 an
andd Su
S
upppl. Figure
Fig
gur
ure 2).
2)).
Functionall analysis
analy
ly
ysi
siss of cg06500161
cg0
g0065
6500
0001611 (ABCG1)
(AB
(A
ABCG1
G1)) aand
nd
d cg20544516
cg20
2054
20
5445
54
4516
45
1 (MIR33B/SREBF1)
16
(MI
M R3
R33B
B/S
/SRE
R BF
BF1)
1)
To assess the
he biological
biologgical relevance
rellevance off the
th
he DNA
DNA methylation
meth
hyllatiion status off CpGs
C Gs
Cp
G found
found to be associated
asso
with lipid levels,
cg06500161
e el
els
l electrophoretic
ell tr ho tii mobility
mobilit
bili
bi
lit shift
hifft assays
hi
assa s (EMSAs)
(EMS
(E
MSA
As)) were
ere carried
iedd ooutt ffor
o cg065
0655
06
(ABCG1). This CpG was chosen as it showed the strongest association with both HDL-C and
TG. cg20544516 was also included in the analysis because of its functionally interesting location
in SREBF1 in a region coding for a microRNA (MIR33b).
The EMSA for cg06500161 identified a higher binding affinity of a protein complex for
the unmethylated status of cg06500161 compared to the methylated status. For cg20544516 a
strong protein binding affinity was detected in the methylated status, which was not detectable in
the unmethylated status (Suppl. Figure 3).
DNA methylation and prevalent myocardial infarction
The CpGs associated with lipid levels were tested for an association with previous hospitalized
11
DOI: 10.1161/CIRCGENETICS.114.000804
MI in the discovery cohort KORA F4 (N=1776 with N=60 cases). Three models were analyzed
and CpG cg06500161, located in the ABCG1 gene, showed an association with MI independent
of lipid levels in all three models (e.g., model 3: ȕ=0.141, p=1.30E-03) (Suppl. Table 6). The
results could not be replicated in KORA F3 and InCHIANTI, possibly due to the low number of
MI cases (N=8 in KORA F3, N=36 in InCHIANTI).
Discussion
DNA methylation of genes involved in lipid metabolism is associated with HDL-C,, TG and
LDL-C levels
Our results indica
indicated
cate
ate
ted th
thatt D
DNA
NA methylation of cg0650
cg06500161
500161 in ABCG11 was associated in
opposite directions
HDL-C
TG
levels.
recttions with
re
hH
DL-C
DL
-C
C aand
nd T
G lev
eveels. Integrating
ev
Integrratin
ing gene
in
genee eexpression
ge
xppre
r ssio
on data
daata
ta revealed
rev
evea
eale
ea
ledd an
le
a
association be
between
cg06500161
methylation
lipid
which
might
mediated
etw
twee
eenn cg
ee
cg06
0650
06
5001
50
0161
01
6 me
61
meth
th
hyl
y atio
io
on an
andd li
lipi
p d levels
pi
l ve
le
vells wh
w
icch mi
migh
g t be
gh
b partly
part
pa
rtly
rt
ly m
e ia
ed
ia by
ABCG1 expression.
methylation
compared
to
p
pression.
DNA
A meth
m
me
eth
thyl
ylat
yl
atio
at
ionn at
io
a tthis
h s Cp
hi
CpG
G was
was al
aalso
soo eelevated
leeva
vate
teed in
n ccases
asses of MIs compa
a
healthy individuals.
iividuals.
vidu
vi
dual
als.
s.
One challenge of genome-wide DNA methylation analyses in blood samples is the
difference in methylation patterns between different blood cell types10, 14. In our blood cell
expression panel of ABCG1, CPT1A and SREBF1 varying expression patterns were also
detectable (Suppl. Figure 1) which underlines the issue of cell heterogeneity. After adjustment
for estimated blood cell proportions using the method proposed by Houseman et al.10, the
number of significant CpGs decreased from 166 to 12. Therefore, in all further analyses cell
proportions were included as covariates to correct for cell heterogeneity.
We identified seven new lipid-related CpGs located in ABCG1 (HDL-C, TG),
MIR33B/SREBF1, in an intergenic region (TG) and in TNIP1 (LDL-C). In addition, we
replicated one CpG (cg00574958 in CPT1A) which was found to be associated with TG levels in
12
DOI: 10.1161/CIRCGENETICS.114.000804
CD4+ T-cells in the GOLDN study (N=991)6. Five of the associations were also found in
adipose tissue, of which the strongest associations were observed between TG levels and
MIR33B/SREBF1 as well as ABCG1 DNA methylation. Both genes are highly expressed in
adipose tissue (> 1.0E07 copies/μg RNA, Suppl. Figure 1). In skin TG levels were associated
with SREBF1 and CPT1A DNA methylation but there was no significant association with
ABCG1 methylation. These results indicate a tissue-specific association between TG levels and
MIR33B/SREBF1 and ABCG1 DNA methylation.
Additionally, we examined whether the observed associations between lipi
lipid
p d andd
methylation levels in KORA F4 were based on confounding by lipid-associated
SNPs.
Most
associiatted
dS
NPs. Mo
NP
Mos
associationss rem
remained
emai
em
aine
ai
nedd sign
ne
significant
gnificant after additional adjustment
gn
adj
djuustment for SNPs
Ps which were nominally
nominall
associated w
with
ithh DNA meth
methylation
thylaatiion at
th
a thee respective
resspecttiv
ve CpG
CpG ssite.
ite
te. Only
On y one
onne C
CpG-lipid
pG
G-lipi
pidd association
pi
assocciaatioo
as
was found to be confounded.
conf
co
n ound
nf
ndded
ed.
d Th
The
he asso
association
sooci
ciationn between
bettwee
be
tweenn DN
DNA
NA m
methylation
eth
thyl
th
ylat
yl
attion
ion off ccg12556569
g1125556
5656
6569
5669 (located
((lo
(l
o
in the promoter
m
moter
region
regi
g on of APOA5)
APOA
AP
OA5)) andd TG
OA
G llevels
evells was confounde
confounded
d d bbyy rrs964184,
s99644184,, which is kn
known
n
to primarily affect
ff t TG le
llevels
el
els
l 4. O
One
n st
study
d hhad
add pre
previously
iio
o sll id
identified
tifi
ified
d this
thi
hi SNP
SNP as an mQTL
QTL
(cytosine modification quantitative trait loci)15. Our results indicate that the nominal associations
between trait-associated SNPs and DNA methylation of lipid-related CpCs were dependent on
lipids and that the identified lipid–DNA methylation associations were not due to genetic
confounding.
Interaction of genes of lipid-associated CpGs
Interestingly, three of the genes where the lipid-related CpGs are localized - ABCG1,
MIR33B/SREBF1 and CPT1A - and their gene products interact with one another (Figure 2).
SREBF1 and SREBF2 (sterol regulatory element-binding transcription factor 1 and 2) code for
the membrane-bound transcription factors SREBP1 and SREBP2, which activate the synthesis of
13
DOI: 10.1161/CIRCGENETICS.114.000804
fatty acid and the synthesis and uptake of cholesterol16, 17. The intronic microRNAs 33a and 33b
(MIR33a/b) are located within SREBF2 and SREBF1, respectively. Coincident with transcription
of SREBF2/1, the embedded MIR33a/b is co-transcribed18. MIR33a/b act as negative regulators,
repressing a number of genes involved in fatty acid oxidation and cholesterol transport18-23 such
as carnitine palmitoyltransferase 1A (CPT1A), which is important for the transport of fatty acids
into the mitochondria for their oxidation24. Studies also identified a role for MIR33a/b in the
repression of the ABC transporters ABCA1 and ABCG120, 25. ABCG1 encodes the ABCtransporter G1, a cholesterol transporter which plays a role in cellular lipid homeostasis.. It
I has
been shown that ABCG1 functions cooperatively with ABCA126. ABCA1
CA1 transports
t ansportts
tr
ids and
and cholesterol
c ol
ch
olesste
terol to lipid-poor HDL su
ubc
b lasses such as aapoA-I,
po
poA-I,
whereas ABCG
phospholipids
subclasses
ABCG1
27,
7, 228
8
has more ma
m
mature
ature HDL pa
particles
artticlees as its
its acceptor
acc
ccepptoor27
.
h pre
he
rese
re
sent
se
n study
nt
stu
udy tthe
h m
he
ethy
hyla
hy
lattion levels
la
lev
evel
e s off tthese
el
h see genes,
he
gen
enes, MI
en
MIR3
R3
33B
3 /S
/SRE
REBF
RE
F1, AB
ABC
CG and
In the
present
methylation
MIR33B/SREBF1,
ABCG1
CPT1A, aree associated withh bl
blood
loodd TG llevels,
evells,, sug
suggesting
gge
g stiingg an epigenetic
epi
piige
g netiic mod
modulation
dulation of lipi
lipid
p d and
fatty acid metabolism
metabolism.
etab
boli
lis H
Here,
Here AB
ABCG1
ABCG
CG11 mi
might
ight
h play
pla
pll a key
ke
k role
oll since
sii
one CpG
CpG
G (cg06500161)
(cg065
06500
0016
161)
1) located
llo
in this gene is associated with both HDL-C and TG levels. The function of ABCG1 in HDL-C
metabolism has been recorded in several studies and reviews29-31; however no report yet exists
about a direct role of ABCG1 in connection with TG levels. One study showed that genetic
variants in the ABCG1 promoter were associated with ABCG1 expression which showed an
influence on the bioavailability of lipoprotein lipase (LPL). Accordingly, ABCG1 regulates the
bioavailability of macrophage-secreted LPL, thereby promoting lipid accumulation, primarily in
the form of TG, in primary human macrophages32.
TNIP1, the methylation of which was associated with LDL-C levels in this study,
encodes the TNFĮ-induced protein 3 (TNFAIP3)-interacting protein 1. This protein appears to be
14
DOI: 10.1161/CIRCGENETICS.114.000804
important in regulating multiple receptor-mediated transcriptional activity of peroxisome
proliferator activated receptors (PPAR)33 and retinoic acid receptors (RAR)34. Interestingly,
ligand activated RAR increases ABCA1 and ABCG1 expression in human macrophages
modulating ABCG1 promoter activity via LXR responsive elements-dependent mechanisms35.
Additionally, studies revealed that PPARĮ/Ȗ-activators induce ABCA1 expression in
macrophages36 and PPARȖ induce ABCG1 expression37. Therefore TNIP1 might have an indirect
impact on the expression of ABCA1/G1.
Methylation of ABCG1 is associated with ABCG1 transcripts
The identified negative association between ABCG1 methylation (cg06500161,
65001
0161
01
61, cg27243685)
61
cg27
2724
27
24436
3688 and
ABCG1 mRNA
methylation-dependent
R A levels
RNA
leeve
vels
lss is po
possibly mediated by methy
hyla
hy
l tion-dependent
ntt ttranscription
ranscription factor
f
binding, as observed
ABCG1
levels
were
obsserved in the
th
he EMSA
EM
MSA
S experiments.
exper
erim
er
imen
im
nts.. A
BCG
CG1 mRNA
mRNA le
evells we
wer
re additionally
add
ddit
dd
i io
onallly
associated with
w hH
HDL-C
D -C aand
DL
ndd T
TG
G le
llevels
vels
ls iin
n oppo
opposite
posi
po
sitte
si
te directions.
direc
irecti
tion
ti
ons.
on
n The
Thee nnegative
egat
ativ
ti e ass
aassociation
as
ssociat
attion
ion between
betw
betw
ABCG1 methylation
e ylation (cg06500161)
ethy
(cg
g06
650
5 01
0161
61)) and
61
d HD
HDL-C
DL-C
C migh
might
ight
h bbee partly
p rtly
pa
ly mediated
med
diatedd by
by the expression
exppressionn of
ABCG1. These
results
the
hese res
h
lts
lt demonstrate
de
trat the
th
h complexity
comple
le iitt off th
h relationship
latii shi
hip bbetween
bet
et een DNA
DNA methylation
meth
ethh
and gene expression.
Our findings could provide the missing link between disturbed blood lipid levels and
changed expression patterns of ABCG1. Studies have shown that in patients with type 2 diabetes
mellitus the ABCG1 expression in macrophages is reduced, leading to decreased cholesterol
efflux to HDL38. Interestingly, a recent study shows an association between the methylation
status of cg06500161 (ABCG1) and fasting insulin as well as with HOMA-IR (homeostatic
model assessment), a surrogate marker of insulin resistance39. All results indicate a key role of
DNA methylation of ABCG1 in the development of complex lipid-related diseases.
15
DOI: 10.1161/CIRCGENETICS.114.000804
ABCG1 – an epigenetic link between blood lipid levels and myocardial infarction?
DNA methylation has been linked to biological processes of cardiovascular disease such as
atherosclerosis40. An association between ABCA1 methylation and HDL-C levels as well as CAD
in patients with familial hypercholesterolemia has been reported41.
We identified a positive association between cg06500161 (ABCG1) and MI in the KORA
F4 cohort: DNA methylation levels of cg06500161 are higher in subjects with previous
hospitalized MI compared to healthy people. Since the number of subjects with self-reported
hospitalized MI was low in KORA F3 and InCHIANTI, no replicationn was achieved. The
These
hese
he
s
results need further confirmation by prospective genome-wide DNA methyl
methylation
yll tion stud
ylat
studies.
t diees.
42 43
n c va
netic
vari
rian
ri
antts in
an
in ABCG1 were shown to be associated with C
AD42,
. However,
Genetic
variants
CAD
nothing is yet
yet known
ye
known about
abooutt ann epigenetic
epig
ig
genetic
i imp
ic
impact
mpacct of ABCG1
ABCG
AB
CG11 onn the
the development
deeveelo
opmen
nt off M
MI.
I. A hhuman
cell culture study
dy showed
dy
sho
h weed a reduction
redduct
re
duc ionn of macrophage
maccro
roph
ph ge ABCG1
phag
A CG
AB
CG11 expression
expr
pres
pr
essiion when
essi
whe
henn high
he
higher
ghher T
TG
G llevels
were present
n in the culture media.
nt
medi
d a. Th
di
The auth
author
horr sug
suggests
gge
g sts th
that
hat hhypertriglyceridemia
ypertriig
yp
igly
l ceriidemia mayy incre
increase
e
the risk of CAD
favoring
cell
thus
leading
CAD
CA
D via
ia
i direct
di ct actions
ctiio on macrophages
ha
fa
fa oring
rii foam
f
ll fformation,
formation
tii
th
h s lleadi
di to
the development of atherosclerotic plaque44. Changes in ABCG1 DNA methylation might
mediate the development of atherosclerotic plaques in response to high TG levels. Thus, with
this study we found hints for a new perspective on the molecular background of CAD.
Conclusions
We found associations between DNA methylation and lipid levels for genes contributing to the
modulation of cholesterol and fatty acid metabolism. Epigenetic modification of ABCG1 and its
regulatory network could play a key role on the path from disturbed blood lipid levels to the
development of complex lipid-related diseases. These results indicate an epigenetic impact on
metabolic regulation in humans and give new insights into the complex picture of lipid-related
16
DOI: 10.1161/CIRCGENETICS.114.000804
complex diseases.
Acknowledgments: The authors thank Nadine Lindemann, Viola Maag and Franziska Scharl for
technical support and acknowledge the support of the nonprofit foundation HTCR, which holds
human tissue on trust, making it broadly available for research on an ethical and legal basis.
Funding Sources: The KORA study was initiated and financed by the Helmholtz Zentrum
München – German Research Center for Environmental Health, which is funded by the German
Federal Ministry of Education and Research (BMBF) and by the State of Bavaria. This project
has received funding from the European Union Seventh Framework Programme (FP7/20072013) under grant agreement n° 261433 (Biobank Standardisation and Harmonisation for
der ggrant
rant
ra
nt aagreement:
gree
gr
eeme
ee
ment
me
nt::
nt
Research Excellence in the European Union - BioSHaRE-EU) and under
al Min
inis
in
istr
is
tryy of H
tr
ealt
ea
lthh
lt
603288. The German Diabetes Center is funded by the German Federal
Ministry
Health
rch off tthe
he S
tate
ta
te of
of North
N
No
(Berlin, Germany) and the Ministry of Innovation, Science and Research
State
tphalia
tph
hal
a iaa (Düsseldorf,
(Düüssel
eldo
el
d rf, Germany). This stu
udy
d was supported
s
rted in
in part
rt by
b a grant
ant from the
Rhine-Westphalia
study
deral
de
eraal Minis
Ministry
sttrry of
o Education
Edu
duca
caatiionn and
andd Research
Res
esea
earchh (BMBF)
ea
(BMB
MB
MBF)
BF) tto
o th
thee Ge
G
rm
man C
en
nter
teer fo
or Di
German Federal
German
Center
for
Diabetes
DZD e.V.). Th
DZ
he MuT
uT
THE
ER Studyy was
wass funded
fundeed bby
y th
tthee WT
WT (0
0819917/Z/0
/0
07/Z)
Z) and
nd core
coo
Research (DZD
The
MuTHER
(081917/Z/07/Z)
th
he Wellcome
Welllco
We
ome Trust
Tru
ustt Centre
Cen
en
ntrre for
fo Human
Huma
Hu
m n Genetics
ma
G neetiics (09
Ge
0905
09
090532
0 32
05
32)). Twi
TwinsUK
wiins
nsUK was
was
as funded
ffunde
u de by
un
funding forr the
(090532).
me Trust; European
Euro
ope
p an
nC
ommu
om
muni
mu
nity
ni
ty
y’s S
even
ev
enth
en
th Framework
Fra
rame
mewo
me
woork Pro
Programme
r gr
ro
g am
amme
m (FP7/2007-2
the Wellcome
Community’s
Seventh
(FP7/2007-2013).
a receives su
uppo
port
r ffrom
rt
rom
ro
m th
thee Na
Nati
tion
ti
onal
on
al Institute
Ins
nsti
tittute for
ti
f r Health
fo
H al
He
alth
t R
th
esea
es
earrch (NIHR)-fun
ea
n
The study also
support
National
Research
(NIHR)-funded
cce,
e, Clinical
Clinicaal Research
Rese
Re
sear
se
a ch Facility
ar
Fac
acil
ilit
il
ityy and
it
an
nd Biomedical
Biom
Bi
omed
om
edic
ed
iccal
ical
a R
Research
esearc
es
esea
eaarch
rch Centre
Cent
Ce
Cent
ntre
re based
re
bas
ased
ed
d at
at Guy's an
n St
BioResource,
and
HS
H
S Foundation
F d tii Trust
T t iin partnership
t
hi with
iith
th
h Ki
K
i ' C
ll
L
d
SNP
SNP genotyping
t i
Thomas' NHS
King's
College
London.
was performed by The Wellcome Trust Sanger Institute and National Eye Institute via
NIH/CIDR. PD’s work forms part of the research themes contributing to the translational
research portfolio of Barts Cardiovascular Biomedical Research Unit, supported and funded by
the National Institute for Health Research.
Conflict of Interest Disclosures: None
References:
1. Andrawes WF, Bussy C, Belmin J. Prevention of cardiovascular events in elderly people.
Drugs Aging. 2005;22:859-876.
2. Elder SJ, Lichtenstein AH, Pittas AG, Roberts SB, Fuss PJ, Greenberg AS, et al. Genetic and
environmental influences on factors associated with cardiovascular disease and the metabolic
syndrome. J Lipid Res. 2009;50:1917-1926.
17
DOI: 10.1161/CIRCGENETICS.114.000804
3. American Heart Association Nutrition C, Lichtenstein AH, Appel LJ, Brands M, Carnethon
M, Daniels S, et al. Diet and lifestyle recommendations revision 2006: A scientific statement
from the american heart association nutrition committee. Circulation. 2006;114:82-96.
4. Global Lipids Genetics C, Willer CJ, Schmidt EM, Sengupta S, Peloso GM, Gustafsson S, et
al. Discovery and refinement of loci associated with lipid levels. Nat Genet. 2013;45:1274-1283.
5. Guay SP, Voisin G, Brisson D, Munger J, Lamarche B, Gaudet D, et al. Epigenome-wide
analysis in familial hypercholesterolemia identified new loci associated with high-density
lipoprotein cholesterol concentration. Epigenomics. 2012;4:623-639.
6. Irvin MR, Zhi D, Joehanes R, Mendelson M, Aslibekyan S, Claas SA, et al. Epigenome-wide
association study of fasting blood lipids in the genetics of lipid lowering drugs and diet network
study. Circulation. 2014;130:565-572.
7. Teschendorff AE, Marabita F, Lechner M, Bartlett T, Tegner J, Gomez-Cabrero
D,, ett aal.
mez
ez--Ca
Cabr
brer
br
eroo D
er
l. A
beta-mixture quantile normalization method for correcting probe design
bias
illumina
gn bi
ias iin
n il
ill
lumina
i a iinfinium
nf
450 k DNA methylation data. Bioinformatics. 2013;29:189-196.
8. Pidsley R,, CC
Mill
Schalkwyk
CC YW, Volta
Vol
olta
t M,
M, Lunnon
Lunn
Lu
nnon
nn
on K,
K, Mi
ill
l J,
J, S
chalk
kwy
wykk LC.
L . A data-driven
LC
d ta-d
da
-ddri
-dri
rive
ven approach
ve
appr
ap
proa
pr
oach to
oa
preprocessing
Genomics.
ing illumina 450k
in
450
50k methylation
50
meth
me
hyl
y atio
on array
arrray data.
data. BM
da
BMC
C Ge
G
noomics. 22013;14:293.
013
13;14::29
13
2 3.
9. Chen LS,, Storey
dissecting
Stor
orrey JD.
JD. Eigen-r2
Eig
igen
en-r2
r2 for di
issecti
ting
ti
ng vvariation
aria
i ti
ia
tioon iin
n high
hhigh-dimensional
hi
igh
gh-d
h dim
imen
ensi
sional
i al sstudies.
tudi
dies
di
es.
Bioinformatics.
2008;24:2260-2262.
a
atics.
2008;24:22
2260
22
600-2
-226
2662.
2.
10. Houseman
Accomando
man
m
an EA, A
ccom
cc
om
oman
mando
ando W
WP,
P K
P,
Koestler
oest
oe
stle
st
tle
lerr DC,
DC, Ch
DC
Chri
Christensen
rist
ri
sten
st
en
ense
nsen
sen BC
BC,
C, Ma
Mars
Marsit
rsit
rs
itt C
CJ,
J, Nelson
J,
Nel
e son HH
HH,
H, et al.
DNA methylation
measures
mixture
distribution.
Bioinformatics.
llation
atiio arrays
arra s as ssurrogate
rrogate
te meas
res off cell
ell
ll mi
i t re di
dist
distrib
rib
ib ti
tion
i BM
BMC
C Bi
i f
2012;13:86.
11. Edwards LJ, Muller KE, Wolfinger RD, Qaqish BF, Schabenberger O. An r2 statistic for
fixed effects in the linear mixed model. Stat Med. 2008;27:6137-6157.
12. Grundberg E, Meduri E, Sandling JK, Hedman AK, Keildson S, Buil A, et al. Global analysis
of DNA methylation variation in adipose tissue from twins reveals links to disease-associated
variants in distal regulatory elements. Am J Hum Genet. 2013;93:876-890.
13. Schurmann C, Heim K, Schillert A, Blankenberg S, Carstensen M, Dorr M, et al. Analyzing
illumina gene expression microarray data from different tissues: Methodological aspects of data
analysis in the metaxpress consortium. PLoS One. 2012;7:e50938.
14. Reinius LE, Acevedo N, Joerink M, Pershagen G, Dahlen SE, Greco D, et al. Differential
DNA methylation in purified human blood cells: Implications for cell lineage and studies on
disease susceptibility. PLoS One. 2012;7:e41361.
18
DOI: 10.1161/CIRCGENETICS.114.000804
15. Moen EL, Zhang X, Mu W, Delaney SM, Wing C, McQuade J, et al. Genome-wide variation
of cytosine modifications between european and african populations and the implications for
complex traits. Genetics. 2013;194:987-996.
16. Edwards PA, Tabor D, Kast HR, Venkateswaran A. Regulation of gene expression by srebp
and scap. Biochim Biophys Acta. 2000;1529:103-113.
17. Horton JD. Sterol regulatory element-binding proteins: Transcriptional activators of lipid
synthesis. Biochem Soc Trans. 2002;30:1091-1095.
18. Marquart TJ, Allen RM, Ory DS, Baldan A. Mir-33 links srebp-2 induction to repression of
sterol transporters. Proc Natl Acad Sci U S A. 2010;107:12228-12232.
19. Najafi-Shoushtari SH, Kristo F, Li Y, Shioda T, Cohen DE, Gerszten RE, et al. Microrna-33
and the srebp host genes cooperate to control cholesterol homeostasis. Science. 2010;328
2010;328:1566288:1
:
1569.
20. Rayner KJ, Suarez Y, Davalos A, Parathath S, Fitzgerald ML, Tamehiro
Mir-33
mehiro
o N,
N et al.
al M
ir 3
contributes to the
homeostasis.
2010;328:1570-1573.
the
h regulation
reg
egul
ulattio
ul
on of cholesterol homeost
staasis. Science. 2010
st
0;3
; 28:1570-1573.
21. Davaloss A,, Goedeke L,
Smibert
Ramirez
U,, ett aal.
Mir-33a/b
L Smi
mibert
mi
r P, Ra
rt
R
mirez CM,
mi
CM, Warrier
Waarrie
ier NP
ie
NP, Andreo
Andr
dreo
dr
oU
l. M
ir-3
-3
33aa
contribute to
regulation
metabolism
insulin
signaling.
Natl
t th
the
he regula
lation
la
onn off ffatty
attyy ac
acid m
etaabooliism
m aand
ndd in
insu
ulin si
ign
g aliing.
g. Proc
P occ N
Pr
atl Acad
at
Accad Sci U
S A. 2011;108:9232-9237.
1 92332108:9
2 9237
37.
22. Gerin I,, Clerbaux LA,, Ha
AK,
Burant
CF,
al.
Expression
H
Haumont
umont O,
O L
Lanthier
anth
hier N, Das
Das A
K, B
urant CF
F, et al
l. Exp
pression oof mir33 from an srebp2 in
intron
export
oxidation.
ntron
trron
on iinhibits
nhib
nh
ib
ibit
bit
itss cholesterol
chol
ch
olles
oles
este
tero
te
rol
ro
ol ex
exp
port aand
port
po
nd ffatty
atty
at
ty aacid
ty
cid ox
ci
cid
oxid
id
dat
atio
ion.
io
n. J Biol
n.
Bio Chem.
Bi
2010;285:33652-33661.
3652
36
52 333661
3661
36
61
23. Horie T, Ono K, Horiguchi M, Nishi H, Nakamura T, Nagao K, et al. Microrna-33 encoded
by an intron of sterol regulatory element-binding protein 2 (srebp2) regulates hdl in vivo. Proc
Natl Acad Sci U S A. 2010;107:17321-17326.
24. McGarry JD, Brown NF. The mitochondrial carnitine palmitoyltransferase system. From
concept to molecular analysis. Eur J Biochem. 1997;244:1-14.
25. Rotllan N, Fernandez-Hernando C. Microrna regulation of cholesterol metabolism.
Cholesterol. 2012;2012:847849.
26. Gelissen IC, Harris M, Rye KA, Quinn C, Brown AJ, Kockx M, et al. Abca1 and abcg1
synergize to mediate cholesterol export to apoa-i. Arterioscler Thromb Vasc Biol. 2006;26:534540.
27. Wang N, Lan D, Chen W, Matsuura F, Tall AR. Atp-binding cassette transporters g1 and g4
mediate cellular cholesterol efflux to high-density lipoproteins. Proc Natl Acad Sci U S A.
2004;101:9774-9779.
19
DOI: 10.1161/CIRCGENETICS.114.000804
28. Kennedy MA, Barrera GC, Nakamura K, Baldan A, Tarr P, Fishbein MC, et al. Abcg1 has a
critical role in mediating cholesterol efflux to hdl and preventing cellular lipid accumulation.
Cell Metab. 2005;1:121-131.
29. Tarling EJ, Edwards PA. Atp binding cassette transporter g1 (abcg1) is an intracellular sterol
transporter. Proc Natl Acad Sci U S A. 2011;108:19719-19724.
30. Tarling EJ, Edwards PA. Dancing with the sterols: Critical roles for abcg1, abca1, mirnas,
and nuclear and cell surface receptors in controlling cellular sterol homeostasis. Biochim Biophys
Acta. 2012;1821:386-395.
31. Ramasamy I. Recent advances in physiological lipoprotein metabolism. Clin Chem Lab Med.
2013:1-33.
32. Olivier M, Tanck MW, Out R, Villard EF, Lammers B, Bouchareychas
Human
ychas L, et al. Huma
man atpma
binding cassette g1 controls macrophage lipoprotein lipase bioavailability
ili
lity
ty and
and promotes
pro
romo
mottes foam
mo
foa
fo
cell formation. Arterioscler Thromb Vasc Biol. 2012;32:2223-2231.
33. Flores AM,
Aneskievich
M, Gurevich
Gur
urev
evic
ev
i h I,
I Zhang C, Ramirez VP, Devens
De
TR, Anesk
kie
i vich BJ. Tnip1 is a
corepressorr of agonist
agonist-bound
Biochem
Biophys.
t-bound
ouund ppars.
ppa
pars
r . Arch
Arrch
c B
ioch
io
chem
ch
em B
iophhys
ys.. 20
22011;516:58-66.
1 ;5
11
;516
16:5
6: 8-66
66..
66
34. Gurevich
Aneskievich
BJ.
Liganded
repressed
ch
h II,, An
A
eski
kiev
ki
vicch BJ
J. Li
iga
g nded
ed rraralpha
arallph
pha an
and rargamma
r rg
ra
gam
amma
ma interact
inter
erracct with
with
h but
butt aare
re re
repr
epr
by tnip1. Biochem
Biophys
Ress Co
Commun.
2009;389:409-414.
i he
ioche
h m Bi
B
ophy
h s Re
R
C
mm
mun. 20
0099;3
3899:4
409
0 -4
414
14.
35. Ayaori M,, Yakushiji
Uto-Kondo
Yakushijji E, Ogura
Ogu
g ra M,
M, Nakaya
Naka
k ya
y K,
K, Hisada
Hiisadda T,, Ut
U
o-Ko
K nddo H, et al. Retinoic acid
ac
receptor agonists
regulate
onists regu
guula
late
te eexpression
xpre
xp
resssio
re
ionn of aatp-binding
tp
p-b
bin
indi
ding cassette
ding
cas
asse
seett
ttee transporter
tran
tr
an porte
ansp
orteer g1 iinn macrophages.
or
macr
ma
c ophagess
Biochim Biophys
2012;1821:561-572.
h Acta.
A t 20
2012;1821:561
2012
12;182
18211:56
5611 572
572
36. Chinetti G, Lestavel S, Bocher V, Remaley AT, Neve B, Torra IP, et al. Ppar-alpha and ppargamma activators induce cholesterol removal from human macrophage foam cells through
stimulation of the abca1 pathway. Nat Med. 2001;7:53-58.
37. Li AC, Binder CJ, Gutierrez A, Brown KK, Plotkin CR, Pattison JW, et al. Differential
inhibition of macrophage foam-cell formation and atherosclerosis in mice by pparalpha,
beta/delta, and gamma. J Clin Invest. 2004;114:1564-1576.
38. Mauldin JP, Nagelin MH, Wojcik AJ, Srinivasan S, Skaflen MD, Ayers CR, et al. Reduced
expression of atp-binding cassette transporter g1 increases cholesterol accumulation in
macrophages of patients with type 2 diabetes mellitus. Circulation. 2008;117:2785-2792.
39. Hidalgo B, Irvin MR, Sha J, Zhi D, Aslibekyan S, Absher D, et al. Epigenome-wide
association study of fasting measures of glucose, insulin, and homa-ir in the genetics of lipid
lowering drugs and diet network study. Diabetes. 2014;63:801-807.
40. Wierda RJ, Geutskens SB, Jukema JW, Quax PH, van den Elsen PJ. Epigenetics in
atherosclerosis and inflammation. J Cell Mol Med. 2010;14:1225-1240.
20
DOI: 10.1161/CIRCGENETICS.114.000804
41. Guay SP, Brisson D, Munger J, Lamarche B, Gaudet D, Bouchard L. Abca1 gene promoter
DNA methylation is associated with hdl particle profile and coronary artery disease in familial
hypercholesterolemia. Epigenetics. 2012;7:464-472.
42. Xu Y, Wang W, Zhang L, Qi LP, Li LY, Chen LF, et al. A polymorphism in the abcg1
promoter is functionally associated with coronary artery disease in a chinese han population.
Atherosclerosis. 2011;219:648-654.
43. Schou J, Frikke-Schmidt R, Kardassis D, Thymiakou E, Nordestgaard BG, Jensen G, et al.
Genetic variation in abcg1 and risk of myocardial infarction and ischemic heart disease.
Arterioscler Thromb Vasc Biol. 2012;32:506-515.
44. Wong BX, Kyle RA, Croft KD, Quinn CM, Jessup W, Yeap BB. Modulation of macrophage
fatty acid content and composition by exposure to dyslipidemic serum in vitro. Lipids.
2011;46:371-380.
21
DOI: 10.1161/CIRCGENETICS.114.000804
Table 1: Characteristics of subjects of the discovery cohort and the replication cohorts
Characteristic
Age [years]
Sex=male
BMI* [kg/m2]
Current smoker
Physically active
Alcohol consumption [g/day]
HDL-C [mg/dl]
LDL-C [mg/dl]
Triglyceride [mg/dl]
Cholesterol [mg/dl]
C-reactive protein [mg/l]
Leukocytes count [/nl]
HbA1c‡ [%]
Self-reported history:
Hypertension||
Hospitalized myocardial infarction
Intake of lipid-lowering drugs (excl. herbal substances)
Fasting at time of blood collection#
KORA F4
(N=1776)
KORA F3
(N=499)
InCHIANTI
(N=472)
MuTHER cohort
(N=856)
60.8 (8.9)
867 (48.8%)
28.2 (4.8)
258 (14.5%)
1021 (57.5%)
15.5 (20.4)
56.5 (14.6)
140.
14
0 0 (3
0.
(35.
5 1))
140.0
(35.1)
1333.
3 1 (94.7)
( 4.7))
(9
133.1
221.
22
1.9 (39.
1.9
9 3)
221.9
(39.3)
2.5 (5
.11)
2.5
(5.1)
5.9
5.99 (1
5.
(1.6
(1.6)
.66)
5.6 (0.6)
52.9 (9.6)
259 (51.9%)
27.2 (4.5)
.9%))
249 (49.9%)
.9%
9%
%)
249 (49.9%)
16.1 (19.6)
9.6)
58.2
2 (17
(17.8)
17.8)
17
131.
13
.0 (3
(33.2)
2)
131.0
164.6 (121.9)
(1211.9)
(1
.99
164.6
2220.5
20.
0 5 (3
(38
8.2))
(38.2)
†
NA
7.33 (2
7.
(2.1
(2.1)
.1
1)
5.3 (0.5)
71.2 (16.0)
215 (45.6%)
27.0 (4.3)
2006 (43.7%)
(
206
238 (50.5%)
(
238
12.6
12
. (15.1)
56.8 (14.8)
124.
12
4 5 (32.4)
124.5
119
9.77 (57.9)
119.7
2205.2
20
05.22 (37.9)
4.0 ((8.0)
6.3 ((1.6)
4.9 (0.8)§
59.4 (9.0)
0 (0.0%)
26.6 (4.9)
84 (9.8%)
NA†
NA†
71.5 (18.2)
124.5 (37.9)
99.2 (49.6)
218.9 (38.7)
NA†
6.5 (1.8)
NA†
811 (45.7%)
60 (3.4%)
290 (16.3%)
1776 (100.0%)
211 (42.3%)
8 (1.6%)
31 (6.2%)
47 (9.4%)
124 (26.3%)
36 (7.6%)
61 (13.0%)
472 (100.0%)
172 (20.1%)
NA†
69 (8.1%)
844 (98.5%)
Continuous and categorical characteristics are given as mean (sd) or absolute numbers and relative proportions, respectively. *BMI: body mass
index †NA: variable not available ‡HbA1c: hemoglobin A1c §in InCHIANTI, HbA1c levels were calculated using the formula (46.7+glucose
level)/28.7, in KORA F3/F4 they were analyzed using the HPLC method ||>140/90 mmHg or medically controlled #overnight fast of at least 8
hours
1
DOI: 10.1161/CIRCGENETICS.114.000804
Table 2: Associations between genome-wide DNA methylation and lipid levels
#
KORA F4
KORA F3
Lipid
CpG
Chr†
Gene
ȕFRHI‡
SE§
p-value
exp var
(%)||
ȕ coef
HDL-C
cg06500161
21
ABCG1
-0.049
0.006
8.26E-17
3.9
cg06500161
21
ABCG1
0.070
0.006
1.21E-27
cg19693031
1
TXNIP
-0.030
0.003
cg11024682
17
SREBF1
0.059
cg00574958
11
CPT1A
cg27243685
21
cg07504977
TG
LDL-C
metaanalysis**
InCHIANTI
SE
p-value
ȕ coef
SE
p-value
p-value
-0.065
0.014 2.97E-06
-0.071
0.016
1.13E-05
9.00E-11#
6.5
0.072
1.89E-06
0.015 1.89E
066
0.063
0.016
1.03E-04
5.56E-10#
1.89E-17
4.1
-0.014
00.007
.00
007
00
7 5.
5.65
5.65E-02
65E
65
E-02
02
-0.023
0.011
3.54E-02
5.67E-03
0.008
5.54E-14
3.2
0.031
00.014
014 22.89E-02
89E
89
E 02
0.030
0.013
2.36E-02
1.60E-03#
-0.118
8
0.01
0.016
16
33.15E-13
3.
.15E-13
33.2
.2
-0.
-0.103
0 103
0.
0.02
0.028
28 2.42
2.42E-04
4 E-04
-0.058
0.020
4.86E-03
7.88E-06#
ABCG1
0.064
0.009
0.00
009
3.24E-13
3.24E
E-13
33.0
.0
00.050
.0
050
00.014
.0
014 5.
5.87E-04
.87E-0
04
0.054
0.022
1.63E-02
2.49E-05#
10
NA**
0.026
0
.02
0 6
0.004
0..00
0
004
4
3.93E-12
3.93E-1
12
22.7
.7
0.02
0.026
0 6
02
0.00
0.008
0008 1.
11.87E-03
87E-03
03
0.027
0.009
3.74E-03
1.91E-05#
cg20544516
17
MIR33B/
SREBF1
0.043
0.04
043
04
3
0.007
0.00
007
00
7
2.84E-09
2.84
84E
84
E-09
09
22.7
.7
7
00.032
.03
032
03
2
00.013
.01
013
01
3 11.39E-02
.39
39E
39
E-02
02
0.032
0.018
7.16E-02
2.22E-03#
cg12556569
11
APOA5
0.005
0.001
6.43E-09
1.9
0.002
0.002 2.56E-01
0.004
0.002
1.25E-02
1.20E-02
cg07397296
21
ABCG1
0.027
0.005
9.48E-08
2.1
0.034
0.010 1.03E-03
0.008
0.011
4.63E-01
3.78E-03#
cg07815238
15
NA**
0.048
0.009
9.66E-08
1.6
0.015
0.017 3.69E-01
0.003
0.014
8.41E-01
4.61E-01
cg22178392
5
TNIP1
0.040
0.007
4.27E-09
2.1
0.049
0.015 1.11E-03
0.020
0.014
1.45E-01
1.04E-03#
*meta-analysis of results of replication in KORA F3 and InCHIANTI †Chr: chromosome ‡ȕFRHIȕFRHIILFLHQW§SE: standard error ||exp var:
explained variance #CpG with association confirmed by replication meta-analysis; level of significance: 1.1E-07 (discovery cohort), 4.5E-03
(replication meta-analysis) **no gene annotation for this CpG according to the UCSC Genome Browser
2
DOI: 10.1161/CIRCGENETICS.114.000804
Table 3: Association between blood lipid levels and lipid-related CpGs in adipose and skin tissue of the MuTHER cohort
Adipose (N=634)
Skin (N=395)
Lipid
CpG
Gene
ȕFRHI*
SE†
p-value
ȕFRHI*
SE†
p-value
HDL-C
cg06500161
ABCG1
na‡
na‡
na‡
na‡
na‡
na‡
cg06500161
ABCG1
na‡
na‡
naa‡
na‡
na‡
na‡
cg11024682
SREBF1
F1
F1
0.007
0.
.00
0077
0.002
0 0002
0.
6.72E-04
6 722E6.
E-04
04
0.006
0.002
4.07E-04
cg00574958
CPT1A
A
0.0001
0.00
0001
00
01
0.001
0 0001
0.
8.16E-01
8 16
8.
16EE-01
E01
-0.005
0.002
2.81E-03
cg27243685
ABCG11
00.013
.01
0133
01
0.00
0022
00
0.002
5.86
86E86
E-08
E08
5.86E-08
0.003
0.003
3.91E-01
cg07504977
NA§
0.006
0.003
4.60E-02
0.004
0.002
1.56E-01
cg20544516
MIR33B/ SREBF1
0.012
0.002
1.20E-10
-0.001
0.002
5.04E-01
cg07397296
ABCG1
0.008
0.002
6.59E-07
0.001
0.002
6.64E-01
cg22178392
TNIP1
0.002
0.001
6.02E-03
0.0003
0.001
7.49E-01
TG
LDL-C
*
ȕ coef: ȕ coefficient †SE: standard error ‡no DNA methylation data available for this CpG site §no gene annotation for this CpG according to the UCSC
Genome Browser. Level of significance: 7.14E-03
3
DOI: 10.1161/CIRCGENETICS.114.000804
Figure Legends:
Figure 1: Results of expression analysis between cg06500161 (ABCG1) and HDL-C levels.
ABCG1 methylation (cg06500161) was negatively associated with ABCG1 mRNA levels.
ABCG1 mRNA levels were positively associated with HDL-C levels. Adjustment analysis
indicated that the association between ABCG1 methylation and HDL-C might be partly mediated
by the expression of ABCG1. Detailed association results can be found in Suppl. Table 5. N=724,
level of significance=8.3E-04
Figure 2: Interrelation
methylation
associated
n rr
nterr
rrel
elat
atio
tio
ionn off genes
genes whose DNA methy
yla
l ti
t on level is asso
oci
c ated with lipid levels.
levee
When SREBF1/2
transcriptionally
activated,
MIR33a/b
co-transcribed,
which
play
BF1
F1/2 is transcr
crriptiional
ally
l act
ly
tiv
ivaateed, MI
M
R333a/b
/b aare
re co
o-ttransscrib
bed, wh
whic
i h pla
ay a role in
repression of
o ABCG1
AB
BCG
CG11 and
an
nd CPT1A.
C T11A.
CP
4