ARTICLE
doi:10.1038/nature09922
Gut flora metabolism of phosphatidylcholine
promotes cardiovascular disease
Zeneng Wang1,2, Elizabeth Klipfell1,2, Brian J. Bennett3, Robert Koeth1, Bruce S. Levison1,2, Brandon DuGar1, Ariel E. Feldstein1,2,
Earl B. Britt1,2, Xiaoming Fu1,2, Yoon-Mi Chung1,2, Yuping Wu4, Phil Schauer5, Jonathan D. Smith1,6, Hooman Allayee7,
W. H. Wilson Tang1,2,6, Joseph A. DiDonato1,2, Aldons J. Lusis3 & Stanley L. Hazen1,2,6
Metabolomics studies hold promise for the discovery of pathways linked to disease processes. Cardiovascular disease
(CVD) represents the leading cause of death and morbidity worldwide. Here we used a metabolomics approach to
generate unbiased small-molecule metabolic profiles in plasma that predict risk for CVD. Three metabolites of the
dietary lipid phosphatidylcholine—choline, trimethylamine N-oxide (TMAO) and betaine—were identified and then
shown to predict risk for CVD in an independent large clinical cohort. Dietary supplementation of mice with choline,
TMAO or betaine promoted upregulation of multiple macrophage scavenger receptors linked to atherosclerosis, and
supplementation with choline or TMAO promoted atherosclerosis. Studies using germ-free mice confirmed a critical role
for dietary choline and gut flora in TMAO production, augmented macrophage cholesterol accumulation and foam cell
formation. Suppression of intestinal microflora in atherosclerosis-prone mice inhibited dietary-choline-enhanced
atherosclerosis. Genetic variations controlling expression of flavin monooxygenases, an enzymatic source of TMAO,
segregated with atherosclerosis in hyperlipidaemic mice. Discovery of a relationship between gut-flora-dependent
metabolism of dietary phosphatidylcholine and CVD pathogenesis provides opportunities for the development of new
diagnostic tests and therapeutic approaches for atherosclerotic heart disease.
The pathogenesis of CVD includes genetic and environmental factors.
A known environmental risk factor for the development of CVD is a
diet rich in lipids. A relationship between blood cholesterol and
triglyceride levels and cardiovascular risk is well established.
However, less is known about the role of the third major category
of lipids, phospholipids, in atherosclerotic heart disease pathogenesis.
Another potential yet controversial environmental factor in the
development or progression of atherosclerotic heart disease is inflammation due to infectious agents. Some studies have reported associations between coronary disease and pathogens such as cytomegalovirus
(CMV), Helicobactor pylori, and Chlamydia pneumoniae1–4. However,
prospective randomized trials with antibiotics in humans have thus far
failed to demonstrate cardiovascular benefit5–7 and studies with germfree hyperlipidaemic mice confirm that infectious agents are not
necessary for murine atherosclerotic plaque development8. Although
a definite cause-and-effect relationship between a bacterial or viral
pathogen and atherosclerosis in humans has not yet been established,
the prospect of a role for microbes in atherosclerosis susceptibility
remains enticing.
The intestinal microbiota (‘gut flora’), comprised of trillions of typically non-pathogenic commensal organisms, serve as a filter for our
greatest environmental exposure—what we eat. Gut flora have an essential role, aiding in the digestion and absorption of many nutrients9.
Animal studies have recently shown that intestinal microbial communities can influence the efficiency of harvesting energy from diet,
and consequently influence susceptibility to obesity10. Metabolomics
studies of inbred mouse strains have also recently shown that gut microbiota may have an active role in the development of complex metabolic
abnormalities, such as susceptibility to insulin resistance and nonalcoholic fatty liver disease11. A link between gut-flora-dependent
phospholipid metabolism and atherosclerosis risk through generation
of pro-atherosclerotic metabolites has not yet been reported.
Metabolomics studies identify markers of CVD
In initial studies we sought to discover unbiased small-molecule metabolic profiles in plasma that predict increased risk for CVD. An initial
‘Learning Cohort’ was used comprising plasma from stable patients
undergoing elective cardiac evaluation who subsequently experienced
a heart attack (myocardial infarction), stroke or death over the ensuing
three-year period versus age- and gender-matched subjects who did
not. Liquid chromatography with on-line mass spectrometry (LC/
MS) analysis of plasma was performed to define analytes associated
with cardiac risk as described in Methods. Of an initial 2,0001 analytes
monitored, 40 met all acceptability criteria within the Learning Cohort.
Subsequent studies within an independent ‘Validation Cohort’ led to
identification of 18 analytes that met acceptability criteria in both
Learning and Validation Cohorts (Fig. 1a, b, Supplementary Fig. 1a
and Supplementary Table 1).
The structural identity of the 18 small molecules in plasma, the
levels of which track with cardiac risks, was not known, as the compounds were screened on the basis of retention time and mass-tocharge ratio (m/z) when analysed by LC/MS. Among the 18 analytes,
those with m/z 76, 104 and 118 demonstrated significant (P , 0.001)
correlations among one another, suggesting a potential relationship
via a common biochemical pathway (Supplementary Fig. 1b). We
therefore initially sought to structurally define these three analytes.
Phosphatidylcholine metabolites are linked to CVD
The candidate compound in plasma with an m/z of 76 associated with
CVD risks was isolated and unambiguously identified as TMAO using
1
Department of Cell Biology, Cleveland Clinic, Cleveland, Ohio 44195, USA. 2Center for Cardiovascular Diagnostics and Prevention, Cleveland Clinic, Cleveland, Ohio 44195, USA. 3Department of Medicine/
Division of Cardiology, BH-307 Center for the Health Sciences, University of California, Los Angeles, California 90095, USA. 4Department of Mathematics, Cleveland State University, Cleveland, Ohio 44115,
USA. 5Bariatric and Metabolic Institute, Cleveland Clinic, Cleveland, Ohio 44195, USA. 6Department of Cardiovascular Medicine, Cleveland Clinic, Cleveland, Ohio 44195, USA. 7Department of Preventive
Medicine and Institute for Genetic Medicine, Keck School of Medicine, University of Southern California, Los Angeles, California 90089, USA.
7 A P R I L 2 0 1 1 | VO L 4 7 2 | N AT U R E | 5 7
©2011 Macmillan Publishers Limited. All rights reserved
RESEARCH ARTICLE
P for trend
<0.05
43 analytes
29 analytes
25 analytes
40 analytes
24 analytes
18 analytes
(3) Structural identiication of analytes
(4) Conirm clinical prognostic utility in independent prospective cohort (N = 1,876)
OR (95% CI)
30
132.3
34.9
34.9
b
Analytes
presumably
related via
common
pathway
20
10
1
0
50
100
150
200
m/z
Figure 1 | Strategy for metabolomics studies to identify plasma analytes
associated with cardiovascular risk. a, Overall schematic to identify plasma
analytes associated with cardiac risk over the ensuing 3-year period. CVA,
cerebrovascular accident; HPLC, high-performance liquid chromatography;
MI, myocardial infarction. b, Odds ratio (OR) and 95% confidence intervals
(CI) of incident (3-year) risk for MI, CVA or death of the 18 plasma analytes
that met all selection criteria in both Learning and Validation Cohorts; odds
ratio and 95% confidence intervals shown are for the highest versus lowest
quartile for each analyte. Filled circles represent the analytes (m/z 5 76, 104,
118) focused on in this study. m/z, mass to charge ratio.
Gut flora is needed to form TMAO from dietary PC
Intestinal microflora have a role in TMAO formation from dietary
free choline13. We therefore proposed that commensal organisms (gut
flora) might also have an obligate role in TMAO formation from
dietary PC. To test this, deuterated PC was synthesized whereby the
choline-methyl groups were deuterium labelled (that is, d9-PC) and
Betaine
a
CH
3
Choline
CH
2
CH
3
[O
2
m/z = 104
d4-choline:
(CH3)3+NCD2CD2OH
Or
al
Gu
ora
d9-DPPC gavage
0.1
olin
e
TM
0
0
2
3
CH
3
FMO
H
[O]
CH
CH
0
4
Time (h)
1.6
3
OH
3
25
0.8
0.02
Choline
TMAO
0
2
Time (h)
0
4
e
in
ol
Ch
TMAO
0
0
2
4
Time (h)
Acquistion of
microflora
0.04
ne
tai
Be
Post-antibiotics
Betaine
0
CH
50
m/z = 76
d4, m/z = 76
d9, m/z = 85
m/z = 60
d4, m/z = 60
d9, m/z = 69
Concentration (μM)
Concentration (μM)
Concentration (μM)
Ch
AO
ly
CH
3
Betaine
1.0
TMAO
TMA
CH
3
0.2
2.0
d9-DPPC i.p.
m/z = 118
d4, m/z = 120
d9, m/z = 127
Suppresion of
microflora
Pre-antibiotics
2
3
on
t fl
CH
CH
]
d9-choline:
(CD3)3+NCH2CH2OH
c
3
N+
N+
3
CH
CH
CH
HO
CH
.
i.p
or
al
Or
3
N+
multinuclear nuclear magnetic resonance (NMR), multi-stage mass
spectrometry (MSn), liquid chromatography with tandem mass spectrometry (LC/MS/MS) and gas chromatography with tandem mass
spectrometry (GC/MS/MS) after multiple derivatization strategies
(see Methods, Supplementary Figs 2a–d, and Supplementary Table
2). TMAO, an oxidation product of trimethylamine (TMA), is a relatively common metabolite of choline in animals12,13. Foods rich in the
lipid phosphatidylcholine (PC, also called lecithin), which predominantly include eggs, milk, liver, red meat, poultry, shell fish and fish,
are believed to be the major dietary sources for choline, and hence
TMAO production14. Briefly, initial catabolism of choline and other
trimethylamine-containing species (for example, betaine) by intestinal microbes forms the gas TMA13, which is efficiently absorbed and
rapidly metabolized by at least one member of the hepatic flavin
monooxygenase (FMO) family of enzymes, FMO3, to form
TMAO15,16. Identification of the plasma analyte associated with
CVD risk with an m/z of 76 as TMAO therefore indicated that the
plasma analyte with an m/z of 104 might be choline. Further, these
results also indicated that the plasma analyte with an m/z of 118
associated with CVD might be related to PC (choline) metabolism.
To test the hypothesis that the plasma analytes with m/z 76
(TMAO), 104 and 118 might all be derived from the major dietary
lipid PC, mice were fed egg-yolk PC (through oral gavage) and plasma
levels of analytes over time were monitored. In both male and female
mice, analytes with the same m/z (76, 104 and 118) and the same
retention times as the corresponding analytes of interest observed in
human plasma all increased after oral PC feeding (Supplementary Fig.
3a, b), strongly indicating that the m/z 104 analyte was choline, and the
analyte at m/z 118 was derived from PC. Confirmation that the plasma
analyte (m/z 104) associated with CVD risk was choline was achieved by
b
C
2.0
Conventionalized
0.2
TMAO
Choline
1.0
0.1
Betaine
0
0
0
2
Concentration (μM)
P for trend
<0.05
Concentration (nM)
58 analytes
HPLC-MS
Adjusted
–Log(P)>1.3
Concentration (μM)
Concentration (μM)
HPLC-MS
Adjusted
–Log(P)>1.3
MSn, LC/MS/MS and GC/MS/MS after multiple derivitization strategies
(Supplementary Fig. 4a–d and Supplementary Table 3).
We next studied the plasma analyte with m/z 118. We proposed
that the analyte was either betaine or one of several potential methylated metabolites of choline (see Supplementary Fig. 5a for structures
and strategy for discrimination among these isomers). To distinguish
between these species, and explore a role for intestinal generation of
the various metabolites, different isotopically labelled choline precursors were administered to mice either through an oral (gavage) or a
parenteral (intraperitoneal, i.p.) route. The observed m/z of new isotopically labelled analytes at the appropriate retention times identified
in plasma after these isotope tracer studies are summarized in Fig. 2a.
Oral administration of non-labelled choline resulted in time-dependent
increases in plasma levels of analytes with m/z 76, 104 and 118, consistent with TMAO, choline and either betaine or a methylated choline
species (Supplementary Fig. 6a). Use of selectively deuterated choline
species at either the trimethylamine moiety (d9 isotopomer) or the
ethyl moiety (d4 isotopomer) unambiguously confirmed the m/z 118
analyte as betaine (Fig. 2a and Supplementary Fig. 6b). Further confirmation was acquired by observing the same retention time in LC/MS
and an identical collision-induced dissociation (CID) mass spectrum
(Supplementary Fig. 5b). Moreover, supplementation of PC or choline
isotopomers via gavage or i.p. injection showed an absolute requirement for the oral route in TMAO production, whereas betaine production from PC or choline was formed via both oral and i.p. routes
(Fig. 2 and Supplementary Fig. 7a).
N+
Validation Cohort
25 cases (3 yr MI, CVA, death)
(2)
versus
25 age/gender-matched controls
HO
Learning Cohort
50 cases (3 yr MI, CVA, death)
(1)
versus
50 age/gender-matched controls
O
a
4
Time (h)
Figure 2 | Identification of metabolites of dietary PC and an obligatory role
for gut flora in generation of plasma analytes associated with CVD risks.
a, Summary schematic indicating structure of metabolites and routes (oral or
i.p.) of formation observed in choline challenge studies in mice using the
indicated isotope-labelled choline. The m/z in plasma observed for the
isotopomers of the choline metabolites are shown. b, Plasma levels of d9
metabolites after i.p. challenge with d9(trimethyl)dipalmitoylphosphatidylcholine (d9-DPPC). c, d9-TMAO production after
oral d9-DPPC administration in mice, following suppression of gut flora with
antibiotics (3 weeks), and then following placement (4 weeks) into
conventional cages with non-sterile mice (‘conventionalized’). Data are
presented as mean 6 standard error (s.e.) from four independent replicates.
5 8 | N AT U R E | VO L 4 7 2 | 7 A P R I L 2 0 1 1
©2011 Macmillan Publishers Limited. All rights reserved
ARTICLE RESEARCH
Dietary PC metabolites predict CVD risk
We next sought to independently confirm the prognostic value of
monitoring fasting plasma levels of TMAO, choline and betaine in a
large independent clinical cohort distinct from subjects examined in
both the Learning and Validation Cohorts. Stable subjects (N 5 1,876)
undergoing elective cardiac evaluations were enrolled. Clinical, demographic and laboratory characteristics of the cohort are provided in
Supplementary Table 4a. Fasting plasma levels of TMAO, choline and
betaine were quantified by LC/MS/MS using methods specific for each
analyte (Supplementary Fig. 8). Increasing levels of choline, TMAO
and betaine were all observed to show dose-dependent associations
with the presence of CVD (Fig. 3a–c) and multiple individual CVD
phenotypes including peripheral artery disease (PAD), coronary artery
disease (CAD), and history of myocardial infarction (see Supplementary Table 5a–d for multilogistic regression models, and Supplementary Table 5e for Somers’ Dxy correlation). The associations between
increased risk of all CVD phenotypes monitored and elevated systemic
levels of the three PC metabolites held true after adjustments for traditional cardiac risk factors and medication usage (Fig. 3a–c and
Supplementary Table 5a–e).
Dietary choline or TMAO promotes atherosclerosis
We next investigated whether the strong associations noted between
plasma levels of the dietary PC metabolites and CVD risk reflected some
hidden underlying pro-atherosclerotic mechanism. Atherosclerosisprone mice (C57BL/6J Apoe2/2) at time of weaning were placed on
either normal chow diet (contains 0.08–0.09% total choline, wt/wt) or
normal chow diet supplemented with intermediate (0.5%) or high
amounts of additional choline (1.0%) or TMAO (0.12%). At 20 weeks
of age increased total aortic root atherosclerotic plaque area was noted
in both male and female mice on diets supplemented with either choline
or TMAO (Fig. 3d and Supplementary Fig. 9a). Analysis of plasma
levels of choline and TMAO in each of the dietary arms showed
nominal changes in plasma levels of choline, but significant increases
of TMAO in mice receiving either choline or TMAO supplementation
(Supplementary Fig. 10). Parallel examination of plasma cholesterol,
triglycerides, lipoproteins, glucose levels and hepatic triglyceride content
in the mice failed to show significant increases that could account for
the enhanced atherosclerosis (Supplementary Table 6 and Supplementary Fig. 11). Interestingly, all dietary groups of mice revealed a significant positive correlation between plasma levels of TMAO and
atherosclerotic plaque size (Fig. 3e and Supplementary Fig. 9b). Of
note, plasma TMAO levels observed within the female mice (which
CVD
P < 0.001
16
b
c
8
2
1
8
Odds ratio
Odds ratio
4
4
2
0.5
0.4
0
15
30
0.3
Lesion (mm2)
P = 0.01
120,000
80,000
40,000
f
R = 0.42, P < 0.001
0.2
0.1
0
50
100
Human
P < 0.0001
7
0
)
)
10
22
=
=
(n
(n
14
0
175
350
TMAO (μM)
None Single Double Triple
Coronary vessel
disease
0.
12
%
TM
ho
C
AO
e
lin
e
lin
1%
C
ho
C
ho
w
(n
(n
=
=
10
31
)
)
0
5%
0
Betaine (μM)
Mouse
e
P = 0.009
0.
5 10 20 40 80
TMAO (μM)
Mouse P = 0.01
160,000
2
0.5
1 2
0
Choline (μM)
d
4
1
1
0.5
0.4
CVD
P < 0.001
12
16
8
Odds ratio
CVD
P < 0.001
32
TMAO (μM)
a
Lesion (μm2)
used as isotope tracer for feeding studies. When mice were fed
through oral gavage with d9-PC, the time-dependent appearance of
the anticipated d9 isotopomer of TMAO was observed in plasma
(Fig. 2c). Interestingly, pre-treatment of mice with a three-week
course of broad-spectrum antibiotics to suppress intestinal flora completely suppressed the appearance of d9-TMAO in plasma after oral
d9-PC administration (Fig. 2c). A similar pattern was observed after
oral administration of d9-choline to mice, with d9-TMAO produced
in untreated mice, but not in the same mice after a 3-week course of
broad-spectrum antibiotics (Supplementary Fig. 7b), or in germ-free
mice born sterilely by Caesarean section (Supplementary Fig. 7c). In a
final series of studies, mice with suppressed intestinal microflora after
antibiotics were placed in conventional cages with normal (nongerm-free) mice to permit intestinal colonization with microbes.
After four weeks, repeat oral d9-PC challenge of the now ‘conventionalized’ mice resulted in readily detectable plasma levels of d9-TMAO
(Fig. 2c). Similar results were observed after conventionalization of
germ-free mice and oral d9-choline (Supplementary Fig. 7c).
Collectively, these results show an obligate role for intestinal microbiota in the generation of TMAO from the dietary lipid PC. They also
reveal the following metabolic pathway for dietary PC producing
TMAO: PC R choline R TMA R TMAO.
Figure 3 | Plasma levels of choline, TMAO and betaine are associated with
atherosclerosis risks in humans and promote atherosclerosis in mice.
a–c, Spline models of the logistic regression analyses reflecting risk of CVD
(with 95% CI) according to plasma levels of choline, TMAO and betaine in the
entire cohort (n 5 1,876 subjects). d, Comparison in aortic lesion area among
20-week-old female C57BL/6J Apoe2/2 mice fed with chow diet supplemented
with the indicated amounts (wt/wt) of choline or TMAO from time of weaning
(4 weeks). e, Relationship between plasma TMAO levels and aortic lesion area.
f, Relationship between fasting plasma levels of TMAO versus CAD burden
among subjects (N 5 1,020). Boxes represent 25th, 50th and 75th percentile,
and whiskers 5th and 95th percentile plasma levels. Single, double and triple
coronary vessel disease refers to number of major coronary vessels
demonstrating $50% stenosis on diagnostic coronary angiography.
get enhanced atherosclerosis relative to their male counterparts), even
on normal chow diet, were significantly higher than those observed
among male mice (Supplementary Fig. 10). No significant gender differences in plasma levels of TMAO were observed in humans
(P 5 0.47); however, a clear dose–response relationship was observed
between TMAO levels and clinical atherosclerotic plaque burden in
subjects undergoing coronary angiography (Fig. 3f).
Hepatic FMOs, TMAO and atherosclerosis
Hepatic FMO3 is a known enzymatic source for TMAO in humans,
based on the recent recognition of the aetiology of an uncommon
genetic disorder called trimethylaminuria (also known as fish malodour syndrome)15,17. Subjects with this metabolic condition have
impaired capacity to convert TMA, which smells like rotting fish, into
TMAO, an odourless stable oxidation product17. We therefore sought
to identify possible sources of genetic regulation and the role of Fmo3
in atherosclerosis using integrative genetics in mice18. Expression
levels of Fmo3 were determined by microarray analysis in the livers
of mice from an F2 intercross between atherosclerosis-prone C57BL/
6J Apoe2/2 mice and atherosclerosis-resistant C3H/HeJ Apoe2/2
mice and compared with quantitative measures of atherosclerosis.
The expression level of Fmo3 showed marked differences between
genders (females .1,000 fold higher than in males). Significant
positive correlations were readily found between hepatic Fmo3 expression and atherosclerotic lesions (Fig. 4a, Supplementary Fig. 12, top
row, and Supplementary Fig. 13). Interestingly, a highly significant
negative correlation with plasma high-density lipoprotein (HDL) cholesterol levels was noted in both male and female mice (Fig. 4b and
Supplementary Fig. 12, middle row). Further, plasma levels of the PC
metabolite TMAO showed a significant positive correlation with hepatic
Fmo3 expression level in mice (Fig. 4c and Supplementary Fig. 12,
bottom row).
7 A P R I L 2 0 1 1 | VO L 4 7 2 | N AT U R E | 5 9
©2011 Macmillan Publishers Limited. All rights reserved
RESEARCH ARTICLE
Female
6,000
0
–2 –0.5 1 2.5
Fmo3 expression
(Mean Log ratio)
–2 –1
0
1
Fmo3 expression
(Mean Log ratio)
e Mouse
P = 0.0006
Aortic lesion (μm2)
500,000
s
/H
3,000
3H
er
C
1,000
2,000
Liver FMO3 mRNA
eJ
0
et
H
0
ou
0
yg
R = 0.49
P = 0.07
250,000
oz
7
P = 0.0014
J
Human
14
Plasma TMAO (μM)
10
0
–2 –1
0
1
Fmo3 expression
(Mean Log ratio)
d
20
/6
0
R = 0.80
P < 0.001
BL
0.4
c 30 Female
R = –0.44
P < 1 × 10–6
57
12,000
C
b
R = 0.29
P = 0.002
HDL cholesterol
(mg ml–1)
Lesion (mm2)
0.8
TMAO (μM)
Female
a
Apoe–/–
f Mouse peritoneal macrophages
g
Mouse peritoneal macrophages
P < 0.001
P < 0.001
P = 0.001
P < 0.001
)
10
(n
e
in
ta
Be
0%
1.
0.
=
10
=
(n
TM
%
C
0%
AO
e
lin
w
ho
ho
C
)
)
10
)
(n
=
(n
(n
e
in
ta
Be
=
10
10
)
)
0
=
10
=
(n
AO
TM
0%
1.
%
12
0.
0%
C
C
ho
ho
lin
w
e
(n
(n
=
=
10
10
)
)
0
0.2
12
0.1
1.
Surface CD36 protein
0.2
1.
0.4
Surface SR-A1 protein
P = 0.008
0.3
P = 0.04
significantly correlated with aortic lesion formation, HDL cholesterol
concentrations and plasma TMAO levels (Supplementary Figs 14–16),
suggesting that several members of the FMO family of enzymes may
participate in atherosclerosis and the PC R TMAO metabolic pathway.
To explore the relationship between hepatic FMOs and plasma TMAO
levels in humans, paired samples of liver and plasma from subjects
undergoing elective liver biopsy were examined. Among all of the
human FMO genes monitored, only a trend towards positive association
was noted between hepatic expression of FMO3 and plasma TMAO
levels (Fig. 4d and Supplementary Fig. 17).
Next, we focused on the genetic regulation of hepatic Fmo3 expression (and other Fmo genes) using expression quantitative trait locus
(eQTL) analyses in the F2 mouse intercross. The eQTL plot for Fmo3
messenger RNA levels is shown in Supplementary Fig. 18, and demonstrates a strongly suggestive cis locus (lod score 5 5.9) on mouse chromosome 1 at 151 Mb. Fmo3 (and several other Fmo genes) is located at
164.8 Mb in a region identified as non-identical by descent between
C3H/HeJ and C57BL/6 (http://mouse.cs.ucla.edu/perlegen/). This
region is just distal to the 95% confidence interval of a previously
reported murine atherosclerosis susceptibility locus20. Examining the
effect of the closest single-nucleotide polymorphism (SNP) to Fmo3
(rs3689151) as a function of alleles inherited from either parental strain
indicated a strong effect on atherosclerosis in both genders of the F2
mice (Kruskal–Wallis test, P , 1.0 3 1026). Bonferroni corrected pairwise comparisons indicated a dose-dependent significant increase in
atherosclerosis in F2 mice heterozygous or homozygous for the
C57BL/6J allele (Fig. 4e). Although the resolution on average for an
F2 intercross of this size is in excess of 20 Mb and thus does not provide
‘gene-level’ resolution, these data show that the locus encompassing the Fmo gene cluster on chromosome 1 is associated with atherosclerotic lesion size. Collectively, these results indicate that: (1) hepatic
expression levels of multiple Fmo genes are linked to plasma TMAO
levels in mice; (2) hepatic expression levels of multiple Fmo genes are
associated with both the extent of aortic atherosclerosis and HDL
cholesterol levels in mice; (3) hepatic expression levels of FMO3 indicate an association with plasma TMAO levels in humans; and (4) a
genetic locus containing the Fmo gene cluster on chromosome 1 in
mice has a strong effect on atherosclerosis.
Figure 4 | Hepatic Fmo genes are linked to atherosclerosis and dietary PC
metabolites enhance macrophage scavenger receptor expression.
a–c, Correlation between hepatic Fmo3 expression and aortic lesion, plasma
HDL cholesterol and TMAO in female mice from the F2 intercross between
atherosclerosis-prone C57BL/6J Apoe2/2 and atherosclerosis-resistant C3H/
HeJ Apoe2/2 mice. d, Correlation between human hepatic FMO3 expression
and plasma TMAO. e, Effect of Fmo3 genotype (SNP rs3689151) on aortic sinus
atherosclerosis in male mice from the C57BL6/J Apoe2/2 and C3H/HeJ
Apoe2/2 F2 intercross. f, g, Quantification of scavenger receptor CD36 and SRA1 surface protein levels in macrophages harvested from C57BL/6J mice (13
week) after three weeks of standard chow versus chow supplemented with the
indicated amounts (wt/wt) of choline, TMAO or betaine. Data are presented as
mean 6 s.e. from the indicated numbers of mice in each group.
FMO3 is one member of a family of FMO enzymes, the majority of
which are physically located as a cluster of genes on chromosome 1 in
both humans and mice. The various FMOs share sequence homology
and overlapping substrate specificities. Further, although rare mutations
in or near the FMO3 gene have been identified in individuals with
trimethylaminuria19, the impact of these mutations on other FMO genes
remains unknown. Examination of the hepatic expression levels of the
various FMO genes revealed that many are highly correlated with each
other in both mice and humans (Supplementary Table 7). Examination
of hepatic expression levels of additional Fmo genes in mice from the
atherosclerosis F2 intercross revealed that multiple Fmo genes are
Diet and gut flora alter macrophage phenotype
To explore potential mechanisms through which dietary choline and its
metabolites might exert their pro-atherosclerotic effects, C57BL/6J
Apoe2/2 mice at time of weaning were placed on a normal chow diet
supplemented with either choline, TMAO or betaine (for .3 weeks).
Both mRNA levels (Supplementary Fig. 19) and surface protein levels
(Fig. 4f, g and Supplementary Fig. 20) of two macrophage scavenger
receptors implicated in atherosclerosis, CD36 and SR-A1, were then
determined in peritoneal macrophages. Relative to normal chow diet,
mice supplemented with either choline, TMAO or betaine all showed
enhanced macrophage levels of CD36 and SR-A1. We next examined
the impact of dietary choline and gut flora on endogenous formation of
cholesterol-laden macrophage foam cells, one of the earliest cellular
hallmarks of the atherosclerotic process. Hyperlipidaemic C57BL/6J
Apoe2/2 mice were fed diets with defined levels of choline as follows:
(1) ‘control’ (0.07–0.08%, wt/wt), which is similar to the choline content
of normal chow (0.08–0.09%); versus (2) high ‘choline’, corresponding
to a .10-fold higher level of choline (1.0%, wt/wt) than normal chow.
Concomitantly, half of the mice were administered broad-spectrum
antibiotics for 3 weeks to suppress intestinal microflora, which was
confirmed by the reduction of plasma TMAO levels by .100-fold
(plasma TMAO concentrations in groups receiving antibiotics were
,100 nM). Whereas mice on the control diet showed modest evidence
of endogenous macrophage foam cell formation, as indicated by Oilred-O staining of peritoneal macrophages, mice on the 1% choline
supplemented diet showed markedly enhanced lipid-laden macrophage development (Fig. 5a). In contrast, suppression of intestinal flora
6 0 | N AT U R E | VO L 4 7 2 | 7 A P R I L 2 0 1 1
©2011 Macmillan Publishers Limited. All rights reserved
ARTICLE RESEARCH
significantly inhibited dietary-choline-induced macrophage foam cell
formation (Fig. 5a, b). These results were confirmed by microscopic
quantification of endogenous foam cell levels (Fig. 5b) and analytical
quantification of the cholesterol content of recovered macrophages
(Fig. 5c). Histopathology and biochemical studies of livers recovered
from these mice showed no evidence of steatosis (Supplementary Fig.
21). Parallel analyses of plasma PC metabolites also demonstrated no
significant changes in choline or betaine levels between the different
dietary groups, and significant increases of plasma TMAO levels only in
mice on the high-choline diet in the absence of antibiotics (males,
–Abx
6
P = 0.02
2
500 μm
500 μm
0
–Abx
P = 0.01 P = 0.04
8
4 P = 0.09
0
Control
Choline
+Abx
P = 0.37
15
P = 0.003
P = 0.02
10
5
P = 0.47
0
Control
Choline
(n 12)
=
14
)
(n
=
11
(n
)
=
10
)
=
(n 13)
=
13
)
Choline
(n
=
(n
=
24
)
21
)
Control
12
Choline
=
0
M+F
(n
30,000
h
P = 0.56
Aorta CD36 (×104 μm2)
P = 0.33
+Abx
6)
8)
Control
(n
–Abx
=
(n 13)
=
17
)
60,000
(n
M+F
(n
90,000
Aortic lesion (μm2)
g
P = 0.002 P = 0.04
(n
=
(n 8)
=
7)
P = 0.53
Female
–Abx
+Abx
Aorta macrophage
(F4/80 positive, ×104 μm2)
f
=
P = 0.06
30,000
500 μm
(n
(n
P < 0.0001 P = 0.01
=
(n 20)
=
18
)
500 μm
P = 0.13
–Abx
+Abx
60,000
=
(n 13)
=
11
)
Choline/Abx
=
=
6)
=
e Male
Choline
Aortic lesion (μm2)
Female mice
Control
Control/Abx
Choline
5)
Control
50 μm
(n
d
(n
P < 0.05
(n
Total cholesterol
(μg μg–1 DNA)
4
0
50 μm
6)
5)
6)
=
=
–Abx
+Abx P = 0.04 P < 0.001
c
Apoe–/– Abx
Choline diet
Abx
Control diet
(n
(n
50 μm
Apoe–/–
Choline
8)
Control
=
0
50 μm
+Abx
P = 0.004 P < 0.001
=
Foam cells (%)
12
(n
b
Apoe–/–
Choline diet
Apoe–/–
Control diet
(n
a
Figure 5 | Obligatory role of gut flora in dietary choline enhanced
atherosclerosis. a, Choline supplementation promotes macrophage foam cell
formation in a gut-flora-dependent fashion. C57BL/6J Apoe2/2 mice at time of
weaning (4 weeks) were provided drinking water with or without broadspectrum antibiotics (Abx), and placed on chemically defined diets similar in
composition to normal chow (control diet, 0.08 6 0.01% total choline, wt/wt)
or normal chow with high choline (choline diet, 1.00% 6 0.01% total choline,
wt/wt). Resident peritoneal macrophages were recovered at 20 weeks of age.
Typical images of Oil-red-O/haematoxylin-stained macrophages in each diet
group are shown. b, Foam cell quantification from peritoneal macrophages
recovered from mice in studies described in panel a. c, Macrophage cellular
cholesterol content. d, Representative Oil-red-O/haematoxylin-stained aortic
root sections from female C57BL/6J Apoe2/2 mice fed control and highcholine diets in the presence or absence of antibiotics. e, f, Aortic lesion area in
20 week old C57BL/6J Apoe2/2 mice off or on antibiotics and fed with control
or choline diet. g, Aortic macrophage quantification with anti-F4/80 antibody
staining. h, Quantification of the scavenger receptor CD36 in aorta within the
indicated groups. Error bars represent s.e.m. from the indicated numbers of
mice.
control versus choline diet, 2.5 6 0.1 mM versus 28.3 6 2.4 mM,
P , 0.001; for females, control versus choline diet, 4.0 6 0.5 mM versus
158.6 6 32.9 mM, P , 0.001).
Gut flora promote diet-induced atherosclerosis
In additional studies we sought to test whether gut flora is involved in
dietary choline-induced atherosclerosis. At the time of weaning (4
weeks old), atherosclerosis-prone C57BL/6J Apoe2/2 mice were
placed on either a control diet (0.08 6 0.01%, wt/wt, choline) or a diet
supplemented with 1% choline (wt/wt, choline diet). Half of the mice
were also treated with broad-spectrum antibiotics to suppress intestinal
microflora. Serial plasma measurements confirmed suppression of
TMAO levels to virtually non-detectable levels (,100 nM) throughout
the duration of the study. At 20 weeks of age, mice were killed and aortic
root lesion development was quantified. In the absence of antibiotics
(that is, with preserved intestinal microflora), choline supplementation
augmented atherosclerosis in both male and female mice nearly threefold (Figs 5d–f). In contrast, suppression of intestinal flora completely
inhibited dietary choline-mediated enhancement in atherosclerosis
(Figs 5d–f). Aortic macrophage content and scavenger receptor CD36
immunoreactive surface area within aortic lesions were markedly
increased in mice on the high-choline diet, but not when intestinal
microflora was suppressed with antibiotic treatment (Fig. 5g, h and
Supplementary Figs 22, 23). Both histopathological and biochemical
examination of liver sections from mice showed no evidence of steatosis
or altered neutral lipid (triglyceride or cholesterol/cholesterol ester)
levels on either diet in the absence or presence of antibiotics (Supplementary Fig. 21 and Supplementary Table 8). Finally, the structural
specificity of PC metabolites in promoting a pro-atherogenic macrophage phenotype was examined. Mice fed diets supplemented with
trimethylamine species (choline or TMAO) showed increased peritoneal
macrophage cholesterol content and raised plasma levels of TMAO. In
contrast, dietary supplementation with the choline analogue 3,3dimethyl-1-butanol (DMB), where the quaternary amine nitrogen of
choline is replaced with a carbon, resulted in no TMAO increase and
no increased cholesterol in macrophages (Supplementary Fig. 24).
Discussion
Using a targeted metabolomics approach aimed at identifying plasma
metabolites the levels of which predict risk of CVD in subjects, we have
identified a novel pathway linking dietary lipid intake, intestinal microflora and atherosclerosis (Fig. 6). The pathway identified (dietary PC/
choline R gut-flora-formed TMA R hepatic-FMO-formed TMAO)
represents a unique additional nutritional contribution to the pathogenesis of CVD that involves PC and choline metabolism, an obligate
role for the intestinal microbial community, and regulation of surface
expression levels of macrophage scavenger receptors known to participate in the atherosclerotic process. The pro-atherogenic gutflora-generated metabolite TMAO is formed in a two-step process
initiated by gut-flora-dependent cleavage of a trimethylamine species
(for example, PC, choline, betaine) generating the precursor TMA, and
subsequent oxidation by FMO3 and possibly other FMOs (Fig. 6). PC
is by far the most abundant dietary source of choline in most humans.
The present results indicate that both environmental exposure (dietary
lipid from predominantly animal products) and microbial flora participate in TMAO formation and producing a pro-atherogenic macrophage phenotype. Although the present genetic studies also indicate a
role for hepatic expression levels of one or more Fmo genes in both
enhanced atherosclerotic plaque and decreased HDL levels in mice, the
participation of FMO genes in human atherosclerosis and HDL cholesterol levels remains to be established. Strong associations between
systemic TMAO levels and both angiographic measures of coronary
artery atherosclerotic burden and cardiac risks were observed among
subjects; however, no correlation was observed between plasma
TMAO levels and HDL cholesterol levels in subjects. It remains to
be determined whether genetic impairment in FMO3 alone or in
7 A P R I L 2 0 1 1 | VO L 4 7 2 | N AT U R E | 6 1
©2011 Macmillan Publishers Limited. All rights reserved
RESEARCH ARTICLE
Ph
os
Heart attack
ne
choli
dyl
ati
ph
Stroke
TMA
Death
Gut flora
Atherosclerosis
FMOs
TMAO
Choline
Fatty acid
Fatty acid
O
O P O
O-
Choline
CH3
+
HO CH2 CH2 N CH3
CH3
Phosphatidylcholine
(dietary)
Gut
flora
Choline
CH3
+
H N CH3
CH3
Hepatic
FMOs
Trimethyl amine
(TMA)
CH3
+
HO N CH3
CH3
Trimethylamine N-oxide
(TMAO)
Figure 6 | Gut-flora-dependent metabolism of dietary PC and atherosclerosis. Schematic summary illustrating newly discovered pathway for gut-floramediated generation of pro-atherosclerotic metabolite from dietary PC.
combination with other FMO genes is protective for CVD. No phenotype other than the objectionable odour accompanying this disorder is
known. In fact, individuals with trimethylaminuria often become
vegans, as reducing the ingestion of dietary animal products rich in
lipids decreases TMA production and the associated noxious odour.
Little is also known about the biological functions of TMAO in humans.
TMAO apparently serves as an osmolite in the freeze-avoidance response of some species21. In vitro it can function as a small-molecule
chaperone, affecting the folding and functioning of some proteins22,23.
In addition, TMAO and TMA accumulate in plasma of subjects on
maintenance haemodialysis24, raising the possibility that TMAO may
contribute to the well-established enhanced CVD risk noted among
subjects with end-stage renal disease.
Choline is an essential nutrient that is usually grouped within the
vitamin B complex. Choline and its metabolite betaine are methyl
donors, along with folate, and are metabolically linked to transmethylation pathways including synthesis of the CVD risk factor homocysteine.
Deficiency in both choline and betaine have been suggested to produce
epigenetic changes in genes linked to atherosclerosis25,26, and acute
choline and methionine deficiency in rodent models causes lipid accumulation in liver (steatohepatitis), heart and arterial tissues27. Alternatively, some studies have reported an association between increased
whole blood levels of total choline and cardiovascular disease28,29. Few
clinical studies have examined the relationship between choline intake
and CVD30, probably because accurate measures of the choline content
of most foods has only recently become available14 (http://www.nal.
usda.gov/fnic/foodcomp/Data/Choline/Choln02.pdf). The association
between dietary choline (and alternative trimethyl-amine-containing
species) and atherosclerosis will be complex because, as the present
studies show, it will be influenced by inter-individual differences in
the composition of the intestinal microflora.
The human intestinal microbial community is an enormous and
diverse ecosystem with known functions in nutrition, gut epithelial
cell health, and innate immunity31. Intestinal flora recently also has
been implicated in the development of some metabolic phenotypes
such as obesity and insulin resistance, as well as alterations in immune
responses11,32–34. To our knowledge, the present studies are the first to
identify a direct link between intestinal microflora, dietary PC and
CVD risk. These results indicate that an appropriately designed probiotic intervention may serve as a therapeutic strategy for CVD.
Interestingly, production of TMAO can be altered by probiotic
administration35. Thus, in addition to the current clinical recommendation for a general reduction in dietary lipids, manipulation of
commensal microbial composition may be a novel therapeutic
approach for the prevention and treatment of atherosclerotic heart
disease and its complications. The present studies also suggest a further novel treatment for atherosclerosis—blocking the presumed
pathogenic biochemical pathway at the level of the gut flora through
use of a non-systemically absorbed inhibitor.
METHODS SUMMARY
Plasma samples and associated clinical study data were identified in patients
referred for cardiac evaluation at a tertiary care centre. All subjects gave written
informed consent and the Institutional Review Board of the Cleveland Clinic
approved all study protocols. Unbiased metabolic profiling was performed using
LC/MS. Target analyte structural identification was achieved using a combination
of LC/MS/MS, LC/MSn, multinuclear NMR, GC/MS and choline isotope tracer
feeding studies in mice as outlined in Methods. Statistical analyses were performed using R (version 2.10.1)36. Intestinal microflora were suppressed by supplementation of drinking water with a cocktail of broad-spectrum antibiotics37.
Germ-free mice were purchased from Taconic SWGF. QTL analyses to identify
atherosclerosis-related genes were performed on F2 mice generated by crossing
atherosclerosis-prone C57BL/6J Apoe2/2 mice and atherosclerosis-resistant
C3H/HeJ Apoe2/2 mice38. mRNA expression was assayed by microarray analysis
and real-time PCR. Aortic root lesion area in mice was quantified by microscopy
after staining39. Mouse peritoneal macrophages were collected by lavage for foam
cell quantification and cholesterol accumulation assay. Surface protein levels of
scavenger receptors CD36 and SR-A1 were determined by flow cytometry.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 29 July 2009; accepted 9 February 2011.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
Epstein, S. E. et al. The role of infection in restenosis and atherosclerosis: focus on
cytomegalovirus. Lancet 348 (suppl. 1), S13–S17 (1996).
Patel, P. et al. Association of Helicobacter pylori and Chlamydia pneumoniae
infections with coronary heart disease and cardiovascular risk factors. Br. Med. J.
311, 711–714 (1995).
Danesh, J., Collins, R. & Peto, R. Chronic infections and coronary heart disease: is
there a link? Lancet 350, 430–436 (1997).
Saikku, P. et al. Serological evidence of an association of a novel Chlamydia, TWAR,
with chronic coronary heart disease and acute myocardial infarction. Lancet 332,
983–986 (1988).
O’Connor, C. M. et al. Azithromycin for the secondary prevention of coronary heart
disease events—the WIZARD study: a randomized controlled trial. J. Am. Med.
Assoc. 290, 1459–1466 (2003).
Cannon, C. P. et al. Antibiotic treatment of Chlamydia pneumoniae after acute
coronary syndrome. N. Engl. J. Med. 352, 1646–1654 (2005).
Andraws, R., Berger, J. S. & Brown, D. L. Effects of antibiotic therapy on outcomes of
patients with coronary artery disease: a meta-analysis of randomized controlled
trials. J. Am. Med. Assoc. 293, 2641–2647 (2005).
Wright, S. D. et al. Infectious agents are not necessary for murine atherogenesis. J.
Exp. Med. 191, 1437–1442 (2000).
Backhed, F., Ley, R. E., Sonnenburg, J. L., Peterson, D. A. & Gordon, J. I. Hostbacterial mutualism in the human intestine. Science 307, 1915–1920 (2005).
Turnbaugh, P. J. et al. An obesity-associated gut microbiome with increased
capacity for energy harvest. Nature 444, 1027–1031 (2006).
Dumas, M. E. et al. Metabolic profiling reveals a contribution of gut microbiota to
fatty liver phenotype in insulin-resistant mice. Proc. Natl Acad. Sci. USA 103,
12511–12516 (2006).
Cashman, J. R. et al. Biochemical and clinical aspects of the human flavincontaining monooxygenase form 3 (FMO3) related to trimethylaminuria. Curr.
Drug Metab. 4, 151–170 (2003).
Al-Waiz, M., Mikov, M., Mitchell, S. C. & Smith, R. L. The exogenous origin of
trimethylamine in the mouse. Metabolism 41, 135–136 (1992).
Zeisel, S. H., Mar, M. H., Howe, J. C. & Holden, J. M. Concentrations of cholinecontaining compounds and betaine in common foods. J. Nutr. 133, 1302–1307
(2003).
6 2 | N AT U R E | VO L 4 7 2 | 7 A P R I L 2 0 1 1
©2011 Macmillan Publishers Limited. All rights reserved
ARTICLE RESEARCH
15. Lang, D. H. et al. Isoform specificity of trimethylamine N-oxygenation by human
flavin-containing monooxygenase (FMO) and P450 enzymes: selective catalysis
by fmo3. Biochem. Pharmacol. 56, 1005–1012 (1998).
16. Zhang, A. Q., Mitchell, S. C. & Smith, R. L. Dietary precursors of trimethylamine in
man: a pilot study. Food Chem. Toxicol. 37, 515–520 (1999).
17. Mitchell, S. C. & Smith, R. L. Trimethylaminuria: the fish malodor syndrome. Drug
Metab. Dispos. 29, 517–521 (2001).
18. Schadt, E. E. et al. An integrative genomics approach to infer causal associations
between gene expression and disease. Nature Genet. 37, 710–717 (2005).
19. Dolphin, C. T., Janmohamed, A., Smith, R. L., Shephard, E. A. & Phillips, I. R.
Missense mutation in flavin-containing mono-oxygenase 3 gene, FMO3, underlies
fish-odour syndrome. Nature Genet. 17, 491–494 (1997).
20. Wang, S. S. et al. Identification of pathways for atherosclerosis in mice: integration
of quantitative trait locus analysis and global gene expression data. Circ. Res. 101,
e11–e30 (2007).
21. Treberg, J. R., Wilson, C. E., Richards, R. C., Ewart, K. V. & Driedzic, W. R. The freezeavoidance response of smelt Osmerus mordax: initiation and subsequent
suppression of glycerol, trimethylamine oxide and urea accumulation. J. Exp. Biol.
205, 1419–1427 (2002).
22. Devlin, G. L., Parfrey, H., Tew, D. J., Lomas, D. A. & Bottomley, S. P. Prevention of
polymerization of M and Z a1-Antitrypsin (a1-AT) with trimethylamine N-oxide.
Implications for the treatment of a1-AT deficiency. Am. J. Respir. Cell Mol. Biol. 24,
727–732 (2001).
23. Song, J. L. & Chuang, D. T. Natural osmolyte trimethylamine N-oxide corrects
assembly defects of mutant branched-chain a-ketoacid decarboxylase in maple
syrup urine disease. J. Biol. Chem. 276, 40241–40246 (2001).
24. Bain, M. A., Faull, R., Fornasini, G., Milne, R. W. & Evans, A. M. Accumulation of
trimethylamine and trimethylamine-N-oxide in end-stage renal disease patients
undergoing haemodialysis. Nephrol. Dial. Transplant. 21, 1300–1304 (2006).
25. Dong, C., Yoon, W. & Goldschmidt-Clermont, P. J. DNA methylation and
atherosclerosis. J. Nutr. 132, 2406S–2409S (2002).
26. Zaina, S., Lindholm, M. W. & Lund, G. Nutrition and aberrant DNA methylation
patterns in atherosclerosis: more than just hyperhomocysteinemia? J. Nutr. 135,
5–8 (2005).
27. Salmon, W. D. & Newberne, P. M. Cardiovascular disease in choline-deficient rats.
Effects of choline deficiency, nature and level of dietary lipids and proteins, and
duration of feeding on plasma and liver lipid values and cardiovascular lesions.
Arch. Pathol. 73, 190–209 (1962).
28. Danne, O., Lueders, C., Storm, C., Frei, U. & Mockel, M. Whole blood choline and
plasma choline in acute coronary syndromes: prognostic and pathophysiological
implications. Clin. Chim. Acta 383, 103–109 (2007).
29. LeLeiko, R. M. et al. Usefulness of elevations in serum choline and free F2isoprostane to predict 30-day cardiovascular outcomes in patients with acute
coronary syndrome. Am. J. Cardiol. 104, 638–643 (2009).
30. Bidulescu, A., Chambless, L. E., Siega-Riz, A. M., Zeisel, S. H. & Heiss, G. Usual
choline and betaine dietary intake and incident coronary heart disease: the
Atherosclerosis Risk in Communities (ARIC) study. BMC Cardiovasc. Disord. 7, 20
(2007).
31. Eckburg, P. B. et al. Diversity of the human intestinal microbial flora. Science 308,
1635–1638 (2005).
32. Ley, R. E., Turnbaugh, P. J., Klein, S. & Gordon, J. I. Microbial ecology: human gut
microbes associated with obesity. Nature 444, 1022–1023 (2006).
33. Li, M. et al. Symbiotic gut microbes modulate human metabolic phenotypes. Proc.
Natl Acad. Sci. USA 105, 2117–2122 (2008).
34. Reigstad, C. S., Lunden, G. O., Felin, J. & Backhed, F. Regulation of serum amyloid
A3 (SAA3) in mouse colonic epithelium and adipose tissue by the intestinal
microbiota. PLoS ONE 4, e5842 (2009).
35. Martin, F. P. et al. Probiotic modulation of symbiotic gut microbial–host metabolic
interactions in a humanized microbiome mouse model. Mol. Syst. Biol. 4, 157
(2008).
36. Rizzo, M. L. Statistical Computing with R (Chapman & Hall/CRC, 2008).
37. Rakoff-Nahoum, S., Paglino, J., Eslami-Varzaneh, F., Edberg, S. & Medzhitov, R.
Recognition of commensal microflora by toll-like receptors is required for
intestinal homeostasis. Cell 118, 229–241 (2004).
38. Wang, S. et al. Genetic and genomic analysis of a fat mass trait with complex
inheritance reveals marked sex specificity. PLoS Genet. 2, e15 (2006).
39. Baglione, J. & Smith, J. D. Quantitative assay for mouse atherosclerosis in the aortic
root. Methods Mol. Med. 129, 83–95 (2006).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We wish to thank E. Sehayek for discussions, L. W. Castellani for
help with lipoprotein profile analysis, and F. McNally, M. Berk and M. Pepoy for technical
assistance. This research was supported by National Institutes of Health grants R01
HL103866, P01 HL098055, P01HL087018-020001, P01 HL28481 and P01
HL30568. B.J.B. was supported by NIH training grant T32-DK07789. The clinical study
GeneBank was supported in part by P01 HL076491-055328, R01 HL103931 and the
Cleveland Clinic Foundation General Clinical Research Center of the Cleveland Clinic/
Case Western Reserve University CTSA (1UL1RR024989). Some of the laboratory
studies (haemaglobin A1C, fasting glucose) in GeneBank were supported by R01
DK080732 and Abbott Diagnostics provided supplies for performance of some of the
fasting lipid profile, glucose, creatinine, troponin I and hsCRP measured in GeneBank.
Author Contributions Z.W. performed metabolomics analyses, and biochemical,
cellular, animal model and mass spectrometry studies. He assisted with statistical
analyses, and assisted in both drafting and critical review of the manuscript. E.K., B.D.
and J.D.S. assisted with performance of animal models and their analyses. B.S.L.
synthesized d9-DPPC and assisted in metabolomics/mass spectrometry analyses.
B.J.B., H.A. and A.J.L. performed the mouse eQTL experiments and analyses, and
assisted in both drafting and critical review of the manuscript. A.J.L. provided some
funding for the study. R.K., E.B.B., X.F. and Y.-M.C. performed mass spectrometry
analyses of clinical samples. Y.W. performed statistical analysis. A.E.F. and P.S. helped
with collection of human liver biopsy material and interpretation of biochemical and
pathological examination of animal liver for steatosis. W.H.W.T. assisted in GeneBank
study design and enrolment, as well as analyses of clinical studies and critical review of
the manuscript. J.A.D. assisted in clinical laboratory testing for human clinical studies,
animal model experimental design, and critical review of the manuscript. S.L.H.
conceived of the idea, designed experiments, assisted in data analyses, the drafting and
critical review of the manuscript, and provided funding for the study.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of this article at
www.nature.com/nature. Correspondence and requests for materials should be
addressed to S.L.H. (hazens@ccf.org).
7 A P R I L 2 0 1 1 | VO L 4 7 2 | N AT U R E | 6 3
©2011 Macmillan Publishers Limited. All rights reserved
RESEARCH ARTICLE
METHODS
40
General procedures. Lipids were extracted by chloroform:methanol (2:1, v/v) .
Cholesterol was quantified by GC/MS41. Triglyceride was quantified by GPO reagent
set (Pointe Scientific)42. Cell DNA content was quantified by PicoGreen43. RNA was
isolated by TRIZOL reagent (Invitrogen) and RNeasy Mini Kit (Qiagen). All
reagents were purchased from Sigma unless otherwise specified.
Research subjects. Plasma samples and associated clinical data were collected as
part of two studies involving stable non-symptomatic subjects undergoing elective cardiac evaluations at a tertiary care centre. The first study, GeneBank, is a
large well-characterized tissue repository with longitudinal data from subjects
undergoing elective diagnostic left heart catheterization or elective coronary
computed tomography angiography44. The second study, BioBank, includes subjects undergoing cardiac risk factor evaluation/modification in a preventive cardiology clinic45. CAD included adjudicated diagnoses of stable or unstable angina,
myocardial infarction or angiographic evidence of $50% stenosis of one or more
epicardial vessels. PAD was defined as any evidence of extra-coronary atherosclerosis. Atherosclerotic CVD was defined as the presence of either CAD or
PAD. All subjects gave written informed consent and the Institutional Review
Board of the Cleveland Clinic approved all study protocols.
Discovery metabolomics analyses began with an unbiased search for plasma
(fasting, EDTA purple top tube) analytes linked to CVD risk using a case–control
design (Learning Cohort, N 5 50 cases and 50 controls). Cases were randomly
selected from GeneBank subjects who experienced a myocardial infarction, stroke
or death over the ensuing 3-year period. An age- and gender-matched control
group was randomly selected from GeneBank subjects that did not experience a
CVD event. An independent non-overlapping Validation Cohort (N 5 25 cases
and 25 controls) was also from GeneBank. A third large (N 5 1,876) independent
study comprised of non-overlapping subjects then evaluated clinical associations
of identified analytes. Approximately half (N 5 1,020) of the subjects enrolled
were from GeneBank and the remaining (N 5 856) were from BioBank. Similar
patient characteristics within each cohort and the combined cohort are observed,
as shown in Supplementary Table 4a, b. The association of each PC metabolite
and various cardiovascular phenotypes within each cohort (GeneBank and
BioBank) are also similar (Supplementary Tables 4c–e). All subjects in the large
independent clinical study had similar inclusion and exclusion criterion, negative
cardiac enzymes (troponin I , 0.03 ng ml21) and no recent history of myocardial
infarction or coronary artery bypass graft. Estimate of glomerular filtration rate was
calculated using the MDRD formula46. Fasting blood glucose, C reactive protein,
troponin I and lipid profiles were measured on the Abbott ARCHITECT platform
(Abbott Diagnostics).
Metabolomics analyses. Plasma proteins were precipitated with four volumes of
ice-cold methanol and small-molecule analytes within supernatants were analysed after injection onto a phenyl column (4.6 3 250 mm, 5 mm Rexchrom
Phenyl; Regis) at a flow rate of 0.8 ml min21 using a Cohesive HPLC interfaced
with a PE Sciex API 365 triple quadrupole mass spectrometer (Applied
Biosystems) with Ionics HSID1, EP101, XT1 redesigned source and collision
cell as upgrades in positive MS1 mode. LC gradient (LC1) starting from 10 mM
ammonium formate over 0.5 min, then to 5 mM ammonium formate, 25% methanol and 0.1% formic acid over 3 min, held for 8 min, followed by 100% methanol
and water washing for 3 min at a flow rate of 0.8 ml min21 was used to resolve
analytes. Spectra were continuously acquired after the initial 4 min. Peaks within
reconstructed ion chromatograms at 1 AMU increments were integrated and both
retention times and m/z of analytes were used for statistical analyses.
Selection criteria for determining analytes of interest were based on the composite of MACE as the clinical phenotype, defined as incident myocardial infarction, stroke or death at 3 years, and included: (1) demonstration of a statistically
significant difference between cases versus controls using a Bonferroni adjusted
two sided t-test (P , 0.05); (2) evidence of a significant (P , 0.05) dose–response
relationship between analyte level and clinical phenotype using Cochran–
Armitage trend test; and (3) a minimal signal-to-noise ratio of 5:1 for a given
analyte.
Chemical characterization of unknown metabolites. To chemically define the
structures of the plasma analytes selected for further investigation (that is, analytes with m/z 76, 104 and 118 in positive MS1 mode), multiple approaches were
used. Analytes of interest were isolated by HPLC, vacuum dried, re-dissolved in
water and injected onto the same phenyl column with a distinct HPLC gradient
(LC2, flow rate: 0.8 ml min21) starting from 0.2% formic acid over 2 min, then
linearly to 18% acetonitrile containing 0.2% formic acid over 18 min and further
to 100% acetonitrile containing 0.2% formic acid over 3 min. The targeted analytes were identified by their m/z and the appropriate fractions recovered. After
removal of solvent, dry analytes were used for structural identification.
Samples analysed by GC/MS were derivatized using Sylon HTP kit (HMDS 1
TMCS 1 Pyridine (3: 1: 9), Supelco). Derivatization of TMAO and the plasma
analyte at m/z 76 also included initial reduction by titanium (III) chloride47 and
further reaction with 2,2,2-trichloroethylchloroformate48. Analyses were performed on the Agilent Technolgies 6890/5973 GC/MS in positive ion chemical
ionization mode. The GC/MS analyses used a J&W Scientific DB-1 column (30 m,
0.25-mm inner diameter, 0.25-mm film thickness) for separations.
Quantification of TMAO, choline and betaine. Stable isotope dilution LC/MS/
MS was used for quantification of TMAO, choline and betaine. TMAO, choline
and betaine were monitored in positive MRM MS mode using characteristic
precursor–product ion transitions: m/z 76 R 58, m/z 104 R 60 and m/z
118 R 59, respectively. The internal standards TMAO-trimethyl-d9 (d9TMAO) and choline-trimethyl-d9 (d9-choline), were added to plasma samples
before protein precipitation, and were similarly monitored in MRM mode at m/z
85 R 68 and m/z 113 R 69, respectively. Various concentrations of TMAO, choline and betaine standards and a fixed amount of internal standards were spiked
into control plasma to prepare the calibration curves for quantification of plasma
analytes. TMA was similarly quantified from acidified plasma by LC/MS/MS
using MRM mode.
Aortic root lesion quantification. Apolipoprotein E knockout mice (C57BL/6J
Apoe2/2) were weaned at 4 weeks of age and fed with either standard chow
control diet (Teklad 2018) or a custom diet comprised of normal chow supplemented with 0.5% choline (Teklad TD.07863), 1.0% choline (Teklad TD.07864)
or 0.12% TMAO (Teklad TD.07865) for 16 weeks. Mice were anaesthetized with
ketamine/xylazine before cardiac puncture to collect blood. Hearts were fixed and
stored in 4% paraformaldehyde before frozen OCT sectioning and staining with
Oil red O and haematoxylin. Aortic root lesion area was quantified as the mean
value of six sections39. Aortic sections were immunostained with rat anti-mouse
F4/80 antibody (ab6640, Abcam) followed by goat anti-rat IgG-FITC antibody
(sc-2011, Santa Cruz) and FITC-conjugated CD36 monoclonal antibody
(Cayman Chemical) for F4/80 and CD36, respectively. Sections were mounted
in Vectashield DAPI (H-1200, Vectashield) to take pictures under a Leica DMR
microscope (W. Nuhsbaum) equipped with a Q Imaging Retiga EX camera. We
used Image-Pro Plus Version 7.0 (MediaCybernetics) to integrate the positive
staining area of F4/80 and CD36 in aorta.
Flow cytometry assays on scavenger receptors. Cell surface expression of
scavenger receptors SR-A1 and CD36 were quantified on peritoneal macrophages
from female mice by flow cytometry after immunostaining with fluorochromeconjugated antibodies. Fluorescence intensity was quantified on a FACSCalibur
flow cytometry instrument with FlowJo software (BD Biosciences). More than
10,000 total events were acquired to obtain adequate macrophages numbers.
The following antibodies were used to stain macrophages: CD36 monoclonal
antibody FITC (Cayman Chemical), anti-mouse SR-AI/MSRA1 (R&D Systems),
goat anti-rat IgG-FITC (Santa Cruz Biotechnology), Alexa Fluor 647 anti-mouse
F4/80 (eBioscience), Alexa Fluor 647 anti-mouse CD11b (eBioscience) and the
isotype controls, Alexa Fluor 647 rat IgG2b (eBioscience), Alexa Fluor 647 rat
IgG2a (eBioscience), normal mouse IgA-FITC (Santa Cruz). Cells were incubated
with antibodies for 30 min at 4uC and washed with 0.1% BSA in PBS. Cells with
double-positive staining for F4/80 and CD11b were gated as macrophage49–51 for
the quantification of fluorescence intensity for CD36 and SR-A1 (Supplementary
Fig. 20), with results normalized to F4/80.
eQTL studies. C57BL/6J Apoe2/2 (B6 Apoe2/2) mice were purchased from the
Jackson Laboratory and C3H/HeJ Apoe2/2 (C3H Apoe2/2) mice were bred by
backcrossing B6 Apoe2/2 to C3H/HeJ for 10 generations. F2 mice were generated
by crossing B6 Apoe2/2 with C3H Apoe2/2 and subsequently intercrossing the
F1 mice. Mice were fed Purina Chow containing 4% fat until 8 weeks of age, and
then transferred to a Western diet (Teklad 88137) containing 42% fat and 0.15%
cholesterol for 16 weeks until euthanasia at 24 weeks of age. Mouse atherosclerotic
lesion area was quantified using standard methods39. eQTL analyses were performed as previously described38. Each individual sample was hybridized against
the pool of F2 samples. Significantly differentially expressed genes were determined
as previously described52. Expression data in the form of mean log ratios (mlratios)
were treated as a quantitative trait in eQTL analysis using Rqtl package for the R
language and environment for statistical computing (http://cran.r-project.org/).
Germ-free mice and conventionalization studies. An antibiotic cocktail
(0.5 g l21 vancomycin, 1 g l21 neomycin sulphate, 1 g l21 metronidazole, 1 g l21
ampicillin) previously shown to be sufficient to deplete all detectable commensal
bacteria37 was administered in drinking water ad libitum. In additional studies,
8-week-old female Swiss Webster germ-free mice (Taconic SWGF) underwent an
oral (gavage) choline challenge (see later) immediately after their removal from
their germ-free microisolator shipper. After the choline or PC challenge, the
germ-free mice were placed in conventional cages with non-sterile C57BL/6J
female mice to facilitate transfer of commensal organisms. Four weeks later,
the conventionalized mice underwent a second choline or PC challenge.
©2011 Macmillan Publishers Limited. All rights reserved
ARTICLE RESEARCH
In vivo macrophage studies. C57BL/6J mice or B6 Apoe 2/2 mice were fed with
either standard chow control diet (Teklad 2018) or a custom diet supplemented
with 1.0% betaine (Teklad TD.08112), 1.0% choline (Teklad TD.07864), 0.12%
TMAO (Teklad TD.07865) or 1.0% dimethylbutanol (DMB) supplemented in
drinking water for at least 3 weeks. Elicited mouse peritoneal macrophages
(MPMs) were harvested by peritoneal lavage with ice-cold PBS 3 days after i.p.
injection of 1 ml 4% thioglycollate. Some studies with mice were performed using a
custom diet with low but sufficient choline content (0.07% total; Teklad TD.09040)
versus high-choline diet (1.0% total; Teklad TD.09041) in the presence or absence
of antibiotics. Choline content of all diets was confirmed by LC/MS/MS.
Foam cell staining. Foam cells were identified by microscopy cultured peritoneal
macrophages on glass coverslips after 6 h in RPMI 1640 medium supplemented
with 3% lipoprotein-deficient serum. Cells were fixed with paraformaldehyde and
stained with Oil red O/haematoxylin53. Cells containing .10 lipid droplets were
scored as foam cells50. At least 10 fields and 500 cells per condition were counted.
Real-time PCR. Real-time PCR of Cd36, Sr-a1 and flavin monooxygenases
(mouse Fmos) was performed using Brilliant II SYBR Green qRT–PCR kit
(Strategene). The forward and reverse primers Cd36, Gapdh, Sr-a1, mouse Fmos
and F4/80 were synthesized by IDT based on sequences reported54–58. RT–PCR of
human FMOs was similarly performed using primers specific for the sequence of
each of the indicated human FMOs.
d9-DPPC synthesis and vesicle preparation. d9-DPPC was synthesized by reacting 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine (Genzyme Pharmaceuticals)
withper-deuteromethyliodide(CD3I,CambridgeIsotopeLaboratories)59,60.Theproduct
waspurifiedbypreparativesilicagelTLCandconfirmedbybothMSandNMR.Eggyolk
lecithin (Avanti Polar Lipids) and d9-DPPC liposomes used for gavage feeding and i.p.
injection of mice were prepared by the method of extrusion through polycarbonate
filters61.
Metabolic challenges in mice. C57BL/6J mice were administered (gavage)
unlabelled or the indicated stable-isotope-labelled choline or PC (egg yolk lecithin
or d9-DPPC) using a 1.5-inch 20-gauge intubation needle. Choline challenge:
gavage consisted of 150 ml of 150 mM d9-choline. PC challenge: gavage or i.p.
injection of 300 ml 5 mg ml21 of unlabelled PC or labelled d9-DPPC. Mice were
fasted 12 h before PC challenge. Plasma (50 ml) was collected via the saphenous
vein from mice at baseline and after gavage or i.p. injection time points.
Statistical analysis. Student’s t-test and Wilcoxon rank sum test were employed to
compare group means62,63. Pearson correlation, Spearman rank correlation and
Somers’ Dxy correlation were used to investigate the correlation between two
variables64,65. Comparison of categorical measures between independent groups
was done using x2 tests66. Odds ratios and 95% confidence intervals for cardiovascular phenotypes (history of myocardial infarction, CAD, PAD, CVD and
CAD1PAD) were calculated with R, version 2.10.1 (http://www.r-project.org),
using logistic regression67 with case status as the dependent variable and plasma
analyte as independent variable. Trend tests in frequencies across quartiles were
done using Cochran–Armitage trend tests68. Levels of analytes were adjusted for
traditional CAD risk factors in a multivariate logistic regression model including
individual traditional cardiac risk factors (age, gender, diabetes, smoking, hypertension, lipids, CRP and estimated creatinine clearance) and medication usage
(statin or other lipid-lowering agents, antihypertensive agents including angiotensin-converting-enzyme inhibitor, angiotensin-receptor blocking agent, diuretic,
calcium-channel blocker or beta blocker, and aspirin or other platelet inhibitors).
40. Folch, J., Lees, M. & Sloane Stanley, G. H. A simple method for the isolation and
purification of total lipides from animal tissues. J. Biol. Chem. 226, 497–509
(1957).
41. Robinet, P., Wang, Z., Hazen, S. L. & Smith, J. D. A simple and sensitive enzymatic
method for cholesterol quantification in macrophages and foam cells. J. Lipid Res.
51, 3364–3369 (2010).
42. Millward, C. A. et al. Genetic factors for resistance to diet-induced obesity and
associated metabolic traits on mouse chromosome 17. Mamm. Genome 20,
71–82 (2009).
43. Ahn, S. J., Costa, J. & Emanuel, J. R. PicoGreen quantitation of DNA: effective
evaluation of samples pre- or post-PCR. Nucleic Acids Res. 24, 2623–2625 (1996).
44. Wang, Z. et al. Protein carbamylation links inflammation, smoking, uremia and
atherogenesis. Nature Med. 13, 1176–1184 (2007).
45. Nicholls, S. J. et al. Lipoprotein (a) levels and long-term cardiovascular risk in the
contemporary era of statin therapy. J. Lipid Res. 51, 3055–3061 (2010).
46. Stoves, J., Lindley, E. J., Barnfield, M. C., Burniston, M. T. & Newstead, C. G. MDRD
equation estimates of glomerular filtration rate in potential living kidney donors
and renal transplant recipients with impaired graft function. Nephrol. Dial.
Transplant. 17, 2036–2037 (2002).
47. Barham, A. H. et al. Appropriateness of cholesterol management in primary care by
sex and level of cardiovascular risk. Prev. Cardiol. 12, 95–101 (2009).
48. daCosta, K. A., Vrbanac, J. J. & Zeisel, S. H. The measurement of dimethylamine,
trimethylamine, and trimethylamine N-oxide using capillary gas chromatographymass spectrometry. Anal. Biochem. 187, 234–239 (1990).
49. Schledzewski, K. et al. Lymphatic endothelium-specific hyaluronan receptor LYVE1 is expressed by stabilin-11, F4/801, CD11b1 macrophages in malignant
tumours and wound healing tissue in vivo and in bone marrow cultures in vitro:
implications for the assessment of lymphangiogenesis. J. Pathol. 209, 67–77
(2006).
50. Cailhier, J. F. et al. Conditional macrophage ablation demonstrates that resident
macrophages initiate acute peritoneal inflammation. J. Immunol. 174, 2336–2342
(2005).
51. Kunjathoor, V. V. et al. Scavenger receptors class A-I/II and CD36 are the principal
receptors responsible for the uptake of modified low density lipoprotein leading to
lipid loading in macrophages. J. Biol. Chem. 277, 49982–49988 (2002).
52. Yang, X. et al. Tissue-specific expression and regulation of sexually dimorphic
genes in mice. Genome Res. 16, 995–1004 (2006).
53. Zhou, J., Lhotak, S., Hilditch, B. A. & Austin, R. C. Activation of the unfolded protein
response occurs at all stages of atherosclerotic lesion development in
apolipoprotein E-deficient mice. Circulation 111, 1814–1821 (2005).
54. Miles, E. A., Wallace, F. A. & Calder, P. C. Dietary fish oil reduces intercellular
adhesion molecule 1 and scavenger receptor expression on murine macrophages.
Atherosclerosis 152, 43–50 (2000).
55. Westendorf, T., Graessler, J. & Kopprasch, S. Hypochlorite-oxidized low-density
lipoprotein upregulates CD36 and PPARc mRNA expression and modulates SR-BI
gene expression in murine macrophages. Mol. Cell. Biochem. 277, 143–152
(2005).
56. Rasooly, R., Kelley, D. S., Greg, J. & Mackey, B. E. Dietary trans 10, cis 12-conjugated
linoleic acid reduces the expression of fatty acid oxidation and drug detoxification
enzymes in mouse liver. Br. J. Nutr. 97, 58–66 (2007).
57. Zhang, J. & Cashman, J. R. Quantitative analysis of FMO gene mRNA levels in
human tissues. Drug Metab. Dispos. 34, 19–26 (2006).
58. de Vries, T. J., Schoenmaker, T., Hooibrink, B., Leenen, P. J. & Everts, V. Myeloid
blasts are the mouse bone marrow cells prone to differentiate into osteoclasts. J.
Leukoc. Biol. 85, 919–927 (2009).
59. Chen, F. C. M. & Benoiton, L. N. A new method of quatenizing amines and its use in
amino acid and peptide chemistry. Can. J. Chem. 54, 3310–3311 (1976).
60. Morano, C., Zhang, X. & Fricker, L. D. Multiple isotopic labels for quantitative mass
spectrometry. Anal. Chem. 80, 9298–9309 (2008).
61. Greenberg, M. E. et al. The lipid whisker model of the structure of oxidized cell
membranes. J. Biol. Chem. 283, 2385–2396 (2008).
62. Gauvreau, K. & Pagano, M. Student’s t-test. Nutrition 9, 386 (1993).
63. Wijnand, H. P. & van de Velde, R. Mann–Whitney/Wilcoxon’s nonparametric
cumulative probability distribution. Comput. Methods Programs Biomed. 63,
21–28 (2000).
64. Gaddis, M. L. & Gaddis, G. M. Introduction to biostatistics: part 6, correlation and
regression. Ann. Emerg. Med. 19, 1462–1468 (1990).
65. Deichmann, M. et al. S100-b, melanoma-inhibiting activity, and lactate
dehydrogenase discriminate progressive from nonprogressive American Joint
Committee on Cancer stage IV melanoma. J. Clin. Oncol. 17, 1891–1896 (1999).
66. Goodall, C. M., Stephens, O. B. & Moore, C. M. Comparative sensitivity of survivaladjusted chi-square and normal statistics for the mutagenesis fluctuation assay. J.
Appl. Toxicol. 6, 95–100 (1986).
67. Traissac, P., Martin-Prevel, Y., Delpeuch, F. & Maire, B. Logistic regression vs other
generalized linear models to estimate prevalence rate ratios. Rev. Epidemiol. Sante
Publique 47, 593–604 (1999).
68. Gautam, S. Test for linear trend in 2 3 K ordered tables with open-ended
categories. Biometrics 53, 1163–1169 (1997).
©2011 Macmillan Publishers Limited. All rights reserved
SUPPLEMENTARY INFORMATION
doi:10.1038/nature09922
Supplementary Table 1. Characteristics of the 18 analytes monitored by LC-MS
with unique m/z and specific retention time in positive ion mode which meet
acceptability criteria in both the Learning Cohort andValidation Cohort.
Analyte
(m/z)
76
87
101
104
105
115
116
118
130
133
138
139
147
147
176
178
191
196
Retention
time
(min)
7.5
8.6
4.3
9.3
9.3
5.8
4.8
6.2
4.3
4.4
6.5
6.5
4.3
5.6
5.8
5.8
5.5
6.8
Control
1.6E+08
6.2E+07
1.6E+07
5.6E+08
1.4E+08
4.4E+07
3.3E+07
9.7E+07
8.4E+07
5.5E+06
4.3E+07
6.4E+06
2.3E+08
1.4E+07
1.4E+07
4.3E+06
4.7E+06
7.7E+06
Peak area
Cases
Adjusted
-Log(P)
2.4E+08
1.85
7.8E+07
1.64
1.9E+07
3.59
6.6E+08
3.00
1.9E+08
3.71
6.0E+07
2.52
4.5E+07
4.52
4.1E+08
3.90
9.5E+07
3.28
6.9E+06
2.40
8.3E+07
3.13
1.5E+07
3.29
2.6E+08
3.91
1.8E+07
3.86
2.0E+07
2.52
1.5E+07
1.54
7.1E+06
4.30
1.3E+08
1.74
Odds ratio
(95% CI)
P for trend
Hazard ratio
(95% CI)
7.7 (2.2-27.4)
8.1 (2.2-29.5)
9.5 (2.6-34.9)
8.1 (2.2-29.5)
8.1 (2.2-29.5)
4.5 (1.3-15.3)
9.5 (2.6-34.9)
2.4 (1.1-5.3)
7.7 (2.2-27.4)
8.1 (2.2-29.5)
5.1 (1.5-17.4)
8.1 (2.2-29.5)
8.1 (2.2-29.5)
5.1 (1.5-17.4)
27.9 (5.9-132.3)
8.0 (2.2-29.2)
8.0 (2.2-29.3)
5.3 (1.5-18.2)
<0.01
<0.01
<0.01
<0.01
<0.01
0.01
<0.01
0.01
<0.01
<0.01
<0.01
<0.01
<0.01
<0.01
<0.01
<0.01
<0.01
0.01
8.4 (2.5-27.8)**
6.8 (2.2-20.5)**
7.4 (2.2-24.7)**
18.0 (4.9-66.5)**
20.8 (5.6-77.1)**
5.1 (1.6-16.1)**
13.0 (3.8-44.6)**
3.9 (1.3-12.0)**
8.8 (2.6-29.6)**
5.2 (1.8-15.2)**
7.5 (2.2-25.6)**
4.2 (1.7-10.5)**
9.7 (2.9-32.7)**
6.4 (2.0-20.6)**
19.3 (5.7-65.2)**
4.7 (1.6-14.1)**
1.3 (0.4-3.8)
4.8 (1.5-15.0)*
Plasma samples for the metabolomics discovery studies were all from the clinical study
Gene Bank, a large and well-characterized tissue repository with longitudinal data from
sequential consenting subjects undergoing elective diagnostic left heart catheterization.
The initial Learning Cohort was comprised of 50 cases (randomly selected from amongst
GeneBank subjects who suffered from non-fatal MI, stroke or death within 3 years after
enrollment) and 50 age-gender matched controls that did not experience non-fatal MI,
stroke or death in the 3 years following enrollment. The peaks in LC chromatograms
were integrated and compared between controls and cases using a Bonferoni adjusted
two sided t-test and adjusted p values (for multiple sampling) were calculated. Adjusted logP values >1.3 we reconsidered significant. Odds ratios (OR) and 95% confidence
intervals (CI) comparing top quartile versus lowest quartile to predict risk for non-fatal MI,
stroke or death in 3 years were calculated using logistic regression and odds ratios with
95% CI that failed to cross unity were consider significant. Trend tests in frequencies
across quartiles were done using Cochran-Armitage trend tests, and p<0.05 was
consider significant. Only analytes with signal to noise ratio of 5 or greater were included
in analyses. A second Validation Cohort (n=25 cases and 25 controls) from GeneBank
that is non-overlapping, and used similar selection criterion for detecting analytes
associated with CVD risks was also examined. A total of 18 analytes met all selection
criteria within both the initial Learning Cohort, and the subsequent ValidationCohort.
Hazard ratios and 95% CI for analytes as continuous variables were also calculated
using Cox regression. Data in the table are shown for cases vs. controls from the
combined Learning and Validation cohorts, including the m/z, retention time, peak areas,
OR(95% CI), and Hazard ratio (95% CI). *p<0.05; **p<0.01.
WWW.NATURE.COM/ NATURE | 1
Supplementary Table 2.
GC-MS analysis of candidate compounds with m/z=76 in positive ion mode.
System 1, trimethylchlorosilane (TMCS); System 2, thrichloroethyl chloroformate (TCECF)
Component
System1
Derivative RT Product
(min) ions
System2
Derivative
RT
(min)
Product
ions
Component
Purified from
Plasma
ND
N, N-dimthyl
trichloroethylcarbmate
4.3
219, 184,
131, 117
95, 88
72, 61, 44
TMAO
ND
N, N-dimthyl
trichloroethylcarbmate
4.3
219, 184,
131, 117
95, 88
72, 61, 44
1-Amino-2propanol
TMS-
4.4
147, 130,
116, 102
86, 73,
59, 44
ND
2-Amino-1propanol
DiTMS-
3.8
220, 204,
188, 147
132, 116
93, 73,
59, 44
ND
3-Amino-1propanol
DiTMS-
3.7
220, 204,
188, 147
132, 116
93, 73,
59, 44
ND
Methylaminoethanol
TMS-
5.2
147, 131,
117, 100,
73, 59, 44
ND
Glycolamide
DiTMS-
5.3
220, 204,
,
188, 132,
73, 59,
44
ND
Hydroxyguanidine
TMS-
4.3
147, 131,
117, 100,
87, 73,
59, 45
ND
Glycine
TriTMS-
5.3
294, 278,
220, 188,
147, 131,
116, 73,
59, 45
ND
4.3
147, 131,
117, 100,
73, 59, 45
ND
N-IsopropylTMShydroxylamine
WWW.NATURE.COM/ NATURE | 2
Supplementary Table 3.
GC-MS analysis of candidate compounds with m/z=104 in positive ion mode.
Candidate compounds were analyzed following derivitization with trimethylsilyl using
Sylon HTP kit.
Compound
Derivative
RT
Product ions
(min)
Component from plasma
Choline
2-Amino-3-Methyl-1butanol
3-Amino-isobutyric Acid
DiTMS
2.6
248, 218, 188, 160, 144, 73
DiTMS
2.9
2-Aminobutyric acid
DiTMS
3.0
TetraTMS
3.2
DiTMS
2.7
248, 218, 201, 174, 147, 130,
102, 73, 45
248, 220, 174, 147, 131, 100,
73, 45
391, 345, 329, 242, 215, 189,
172, 147, 131, 100, 73, 45
248, 218, 190, 174, 147, 131,
100, 73, 45
Diethylenetriamine
TetraTMS
3.4
392, 354, 282, 258, 231, 208,
189, 171, 143, 126, 102, 73, 56
Ethyl-N-hydroxylacetimidate
benzonitrile
TMS
2.6
177, 160, 144, 125, 117, 100,
73, 45
DiTMS
2.7
2-Isopropylaminoethanol
TMS
5.6
247, 223, 188, 171, 147, 130,
79, 73, 42
175, 103, 72, 44
1-dimethylaminopropanol
TMS
2.6
Biuret
N,N-Dimethylglycine
2-propylaminoethanol
_
176, 160, 144, 126, 117, 100,
73, 45
WWW.NATURE.COM/ NATURE | 3
Supplementary Table 4a .
Subject characteristics for entire cohort examined
Characteristic
Whole cohort (n=1876)
Age
64.2 ± 10.3
Women (%)
51.2
Diabetes (%)
31.2
Hypertension (%)
70.1
History of smoking (%)
59.4
Current smoking (%)
10
LDL
98 (81- 120)
HDL
35 (29- 43)
Triglycerides
115 (83- 164)
CRP
2.4 (1.0- 5.9)
Framingham Risk Score
10.0 (7.0- 12.0)
MDRD (GFR)
73.0 (60.4- 86.0)
ACE (%)
47.6
STATIN (%)
52.8
ASPIRIN (%)
69.0
TMANO
3.9 (2.6- 6.3)
Choline
10.7 (8.7- 13.4)
Betaine
41.2 (33.1- 51.0)
CVD (%)
65.3
PAD (%)
23.1
CAD (%)
42.2
History of MI (%)
27.0
Values expressed in mean ± standard deviation or median (interquartile range).
WWW.NATURE.COM/ NATURE | 4
Supplementary Table 4b. Comparison of clinical characteristics
(Genebank vs. BioBank subjects)
Characteristic
GeneBank
BioBank
P value
(n=1,020)
(n=856)
63.9 ± 9.3
64.5±11.5
0.221
Women (%)
50.7
51.7
0.681
Diabetes (%)
33.9
28.0
0.07
Hypertension (%)
68.5
72.0
0.116
History of smoking (%)
55.0
64.6
<0.001
LDL
98 (81- 121)
99 (80- 120)
0.996
HDL
35 (28- 43)
36 (30- 43)
0.06
Triglycerides
117 (83- 166)
112 (82- 163)
0.597
CRP
2.5 (1.0- 5.8)
2.3 (1.0- 5.9)
0.743
Framingham Risk Score
10.0 (7.0- 12.0)
10.0 (7.0- 12.0)
0.690
MDRD (GFR)
72.6 (60.6- 85.5)
73.8 (60.2- 86.5)
0.358
ACE inhibitors (%)
48.2
46.9
0.583
Statins (%)
51.1
54.8
0.121
Aspirin (%)
67.7
70.5
0.204
CVD (%)
65.4
65.2
0.941
PAD (%)
24.5
21.5
0.140
CAD (%)
41.0
43.7
0.249
Age
Plasma from subjects examined in the third large independent cohort
for verification of the association between plasma levels of choline,
TMAO and betaine vs. various cardiovascular phenotypes were
enrolled in two studies, GeneBank and BioBank. GeneBank and
BioBank have similar inclusion criterion. Each is comprised of stable
consenting subjects undergoing elective cardiac evaluation for either
cardiovascular risk factor evaluation/modification (BioBank) and/or
diagnostic coronary angiography (GeneBank). The clinical characteristics of subjects analyzed from GeneBank and BioBank are listed above,
and reveal similar overall clinical characteristics for cardiac risk factors
and prevalent cardiovascular disease.
Values expressed in mean ± standard deviation or median (interquartile range).
WWW.NATURE.COM/ NATURE | 5
Supplementary Table 4c. Comparison of odds ratio and 95% confidence intervals (CIs)
for association between plasma choline levels and the indicated cardiovascular
phenotypes within Genebank and BioBank subjects
GeneBank (n=1,020)
Characteristic
BioBank (n=856)
Unadjusted
Adjusted
Unadjusted
Adjusted
1.36 (0.95-1.95)
(p=0.09)
1.76 (1.23-2.54)
(p=0.002)
2.98 (2.03-4.38)
(p<0.001)
1.22 (0.82-1.81)
(p=0.33)
1.36 (0.91-2.04)
(p=0.13)
1.91 (1.24-2.96)
(p=0.004)
1.79 (1.21-2.64)
(p=0.004)
1.80 (1.21-2.66)
(p=0.003)
3.23 (2.13-4.91)
(p<0.001)
1.67 (1.10-2.54)
(p=0.02)
1.56 (1.03-2.38)
(p=0.04)
2.70 (1.71-4.27)
(p<0.001)
1.43 (0.88-2.32)
(p=0.15)
2.51 (1.57-4.01)
(p<0.001)
4.92 (3.06-7.91)
(p<0.001)
1.38 (0.79-2.42)
(p=0.25)
1.88 (1.08-3.27)
(p=0.03)
2.87 (1.61-5.12)
(p<0.001)
1.44 (0.81-2.53)
(p=0.21)
1.99 (1.14-3.47)
(p=0.02)
4.31 (2.48-7.48)
(p<0.001)
1.20 (0.65-2.20)
(p=0.56)
1.60 (0.88-2.91)
(p=0.12)
2.91 (1.57-5.38)
(p=0.001)
1.45 (0.86-2.46)
(p=0.17)
2.80 (1.70-4.62)
(p<0.001)
5.69 (3.44-9.39)
(p<0.001)
1.43 (0.76-2.71)
(p=0.27)
2.00 (1.06-3.76)
(p=0.03)
3.07 (1.60-5.89)
(p=0.001)
1.30 (0.72-2.37)
(p=0.38)
1.97 (1.10-3.50)
(p=0.02)
4.01 (2.27-7.08)
(p<0.001)
1.04 (0.55-1.99)
(p=0.89)
1.45 (0.78-2.72)
(p=0.24)
2.47 (1.30-4.68)
(p=0.006)
1.36 (0.96-1.94)
(p=0.09)
1.74 (1.22-2.50)
(p=0.003)
2.92 (1.99-4.28)
(p<0.001)
1.24 (0.84-1.83)
(p=0.27)
1.36 (0.92-2.02)
(p=0.12)
1.93 (1.26-2.97)
(p=0.003)
1.86 (1.26-2.76)
(p=0.002)
1.86 (1.26-2.76)
(p=0.002)
3.39 (2.23-5.15)
(p<0.001)
1.76 (1.16-2.67)
(p=0.008)
1.64 (1.08-2.49)
(p=0.02)
2.94 (1.86-4.62)
(p<0.001)
1.57 (0.98-2.51)
(p=0.06)
1.68 (1.05-2.69)
(p=0.03)
2.64 (1.67-4.17)
(p<0.001)
1.29 (0.78-2.13)
(p=0.31)
1.16 (0.70-1.92)
(p=0.57)
1.52 (0.92-2.5)
(p=0.10)
1.60 (1.05-2.43)
(p=0.03)
1.32 (0.86-2.02)
(p=0.20)
2.13 (1.40-3.25)
(p<0.001)
1.60 (1.02-2.49)
(p=0.04)
1.23 (0.78-1.93)
(p=0.37)
2.02 (1.27-3.22)
(p=0.003)
CAD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
PAD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
CAD + PAD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
CVD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
MI
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
Plasma from subjects examined in the third large independent cohort for verification of
the association between plasma levels of choline, TMAO and betaine vs. various cardiovascular phenotypes were enrolled in two studies, GeneBank and BioBank. GeneBank
and BioBank have similar inclusion criterion. Each is comprised of stable consenting
subjects undergoing elective cardiac evaluation for either cardiovascular risk factor
evaluation/modification (BioBank) and/or diagnostic coronary angiography (GeneBank).
Shown are the unadjusted and adjusted odds ratio and 95%CI for plasma choline levels
within each sub-cohort and the indicated cardiovascular phenotypes. Multivariate logistic
regression analyses were adjusted for traditional cardiovascular risk factors, CRP, renal
function and medication use, as indicated in Supplemental Methods. Significant Odds
ratios (95% CIs) are shown in bold with p values in red.
WWW.NATURE.COM/ NATURE | 6
Supplementary Table 4d. Comparison of odds ratio and 95% confidence intervals (CIs)
for association between plasma TMAO levels and the indicated cardiovascular phenotypes
within Genebank and BioBank subjects
GeneBank (n=1,020)
Characteristic
Unadjusted
Adjusted
BioBank (n=856)
Unadjusted
Adjusted
CAD
Quartile 2 vs 1
1.33 (0.93-1.90)
(p=0.12)
1.56 (1.18-2.07)
(p=0.002)
1.86 (1.26-2.75)
(p=0.002)
1.85 (1.22-2.8)
(p=0.004)
Quartile 3 vs 1
1.59 (1.11-2.29)
(p=0.01)
3.53 (2.36-5.26)
(p<0.001)
1.44 (1.08-1.91)
(p=0.01)
2.59 (1.88-3.55)
(P<0.001)
1.87 (1.27-2.77)
(p=0.002)
3.74 (2.44-5.72)
(p<0.001)
1.71 (1.12-2.61)
(p=0.01)
3.08 (1.94-4.88)
(p<0.001)
1.15 (0.70-189)
(p=0.58)
2.11 (1.32-3.36)
Quartile 3 vs 1
(p=0.002)
6.01 (3.72-9.72)
Quartile 4 vs 1
(p<0.001)
CAD + PAD
1.49 (0.99-2.24)
(p=0.05)
1.66(1.11-2.49)
(p=0.01)
3.43(2.26-5.21)
(p<0.001)
2.17 (1.23-3.84)
(p=0.008)
1.97 (1.11-3.48)
(p=0.02)
4.96 (2.82-8.73)
(p<0.001)
2.07( 1.12-3.82)
(p=0.02)
1.66 (0.89-3.09)
(p=0.11)
3.75 (2.00-7.03)
(p<0.001)
Quartile 2 vs 1
1.36 (0.79-2.34)
(p=0.26)
2.58 (1.55-4.28)
(p<0.001)
7.47 (4.45-12.52)
(p<0.001)
1.86 (1.18-2.93)
(p=0.008)
1.85 (1.17-2.91)
(p=0.009)
4.03 (2.54-6.40)
(p<0.001)
2.51 (1.37-4.59)
(p=0.003)
2.12 (1.15-3.92)
(p=0.02)
5.60 (3.08-10.2)
(p<0.001)
2.44 (1.26-4.71)
(p=0.008)
1.79 (0.91-3.52)
(p=0.09)
4.00 (2.04-7.82)
(p<0.001)
1.26 (0.89-1.80)
(p=0.50)
1.53 (1.07-2.18)
(p=0.02)
3.41 (2.29-5.06)
(P<0.001)
1.50 (1.14-1.98)
(p=0.004)
1.39 (1.05-1.83)
(p=0.02)
2.54 (1.86-3.47)
(p<0.001)
1.80 (1.22-2.66)
(p=0.003)
1.87 (1.26-2.75)
(p=0.002)
3.70 (2.42-5.64)
(p<0.001)
1.78 (1.18-2.69)
(p=0.006)
1.70 (1.12-2.58)
(p=0.01)
3.08 (1.96-4.86)
(p<0.001)
0.98 (0.60-1.59)
(p=0.93)
1.86 (1.19-2.90)
(p=0.006)
2.23 (1.43-3.46)
(p<0.001)
0.95 (0.57-1.59)
(p=0.86)
1.53 (0.95-2.46)
(p=0.08)
1.47 (0.90-2.40)
(p=0.12)
1.48 (0.97-2.27)
(p=0.07)
1.37 (0.89-2.1)
(p=0.15)
2.47 (1.62-3.75)
(p<0.001)
1.51 (0.96-2.36)
(p=0.07)
1.30 (0.82-2.05)
(p=0.26)
2.11 (1.33-3.34)
(p=0.001)
Quartile 4 vs 1
PAD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
CVD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
MI
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
Plasma from subjects examined in the third large independent cohort for verification of
the association between plasma levels of choline, TMAO and betaine vs. various cardiovascular phenotypes were enrolled in two studies, GeneBank and BioBank. GeneBank
and BioBank have similar inclusion criterion. Each is comprised of stable consenting
subjects undergoing elective cardiac evaluation for either cardiovascular risk factor
evaluation/modification (BioBank) and/or diagnostic coronary angiography (GeneBank).
Shown are the unadjusted and adjusted odds ratio and 95%CI for plasma TMAO levels
within each sub-cohort and the indicated cardiovascular phenotypes. Multivariate logistic
regression analyses were adjusted for traditional cardiovascular risk factors, CRP, renal
function and medication use, as indicated in Supplemental Methods. Significant Odds
ratios (95% CIs) are shown in bold with p values in red.
WWW.NATURE.COM/ NATURE | 7
Supplementary Table 4e. Comparison of odds ratio and 95% confidence intervals(CIs)
for association between plasma betaine levels and the indicated cardiovascular
phenotypes within Genebank and BioBank subjects.
Characteristic
GeneBank (n=1,020)
BioBank (n=856)
Unadjusted
Adjusted
Unadjusted
Adjusted
1.48 (1.03-2.14)
(p=0.04)
1.41 (0.98-2.03)
(p=0.07)
1.63 (1.13-2.35)
(p=0.009)
1.62 (1.20-2.17)
(p=0.001)
1.36 (1.01-1.82)
(p=0.04)
2.30 (1.69-3.14)
(p<0.001)
1.88 (1.26-2.79)
(p=0.002)
1.29 (0.87-1.9)
(p=0.20)
3.23 (2.11-4.94)
(p<0.001)
1.88 (1.22-2.90)
(p=0.004)
1.30 (0.84-1.99)
(p=0.24)
3.66 (2.28-5.86)
(p<0.001)
1.50 (0.95-2.38)
(p=0.09)
1.78 (1.14-2.79)
(p=0.01)
1.87 (1.18-2.94)
(p=0.007)
1.40 (0.93-2.09)
(p=0.10)
1.57 (1.04-2.36)
(p=0.03)
1.87 (1.24-2.83)
(p=0.003)
1.48 (0.86-2.55)
(p=0.16)
1.65 (0.96-2.83)
(p=0.07)
2.40 (1.41-4.08)
(p=0.001)
1.50 (0.82-2.74)
(p=0.19)
1.67 (0.91-3.07)
(p=0.10)
2.75 (1.49-5.08)
(p=0.001)
1.92 (1.16-3.17)
(p=0.01)
2.23 (1.37-3.64)
(p=0.001)
2.53 (1.55-4.14)
(p<0.001)
1.71 (1.10-2.67)
(p=0.02)
1.76 (1.12-2.76)
(p=0.01)
2.14 (1.37-3.36)
(p=0.001)
1.34 (0.76-2.35)
(p=0.32)
1.64 (0.94-2.86)
(p=0.08)
2.27 (1.31-3.93)
(p=0.003)
1.36 (0.72-2.58)
(p=0.34)
1.64 (0.86-3.13)
(p=0.13)
2.51 (1.31-4.79)
(p=0.005)
1.38 (0.96-1.98)
(p=0.08)
1.34 (0.93-1.92)
(p=0.11)
1.49 (1.04-2.14)
(p=0.03)
1.57 (1.18-2.10)
(p=0.001)
1.38 (1.03-1.84)
(p=0.03)
2.21 (1.63-3.00)
(p<0.001)
1.84 (1.24-2.72)
(p=0.003)
1.25 (0.85-1.83)
(p=0.25)
3.35 (2.18-5.13)
(p<0.001)
1.83 (1.20-2.80)
(p=0.005)
1.28 (0.84-1.95)
(p=0.25)
3.84 (2.40-6.15)
(p<0.001)
1.42 (0.90-2.23)
(p=0.13)
1.29 (0.81-2.04)
(p=0.29)
1.72 (1.09-2.69)
(p=0.02)
1.02 (0.62-1.67)
(p=0.95)
0.86 (0.52-1.43)
(p=0.56)
1.10 (0.66-1.82)
(p=0.71)
1.30 (0.86-1.96)
(p=0.22)
1.22(0.80-1.86)
(p=0.35)
1.95 (1.29-2.95)
(p=0.002)
1.33 (0.85-2.07)
(p=0.22)
1.22 (0.77-1.92)
(p=0.40)
2.09 (1.33-3.28)
(p=0.001)
CAD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
PAD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
CAD + PAD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
CVD
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
MI
Quartile 2 vs 1
Quartile 3 vs 1
Quartile 4 vs 1
Plasma from subjects examined in the third large independent cohort for verification of
the association between plasma levels of choline, TMAO and betaine vs. various cardiovascular phenotypes were enrolled in two studies, GeneBank and BioBank. GeneBank
and BioBank have similar inclusion criterion. Each is comprised of stable consenting
subjects undergoing elective cardiac evaluation for either cardiovascular risk factor
evaluation/modification (BioBank) and/or diagnostic coronary angiography (GeneBank).
Shown are the unadjusted and adjusted odds ratio and 95%CI for plasma betaine levels
within each sub-cohort and the indicated cardiovascular phenotypes. Multivariate logistic
regression analyses were adjusted for traditional cardiovascular risk factors, CRP, renal
function and medication use, as indicated in Supplemental Methods. Significant Odds
ratios (95% CIs) are shown in bold with p values in red.
WWW.NATURE.COM/ NATURE | 8
Supplementary Table 5a. Odds ratio (95% CI) of cardiovascular phenotypes
based on quartiles of plasma levels of choline. (n=1876)
Risk
MI
Model
1
unadjusted 1
adjusted
1
unadjusted 1
PAD
adjusted
1
unadjusted 1
CAD
adjusted
1
unadjusted 1
PAD+
CAD
adjusted
1
unadjusted 1
CVD
adjusted
1
2
1.30 (0.96-1.76)
(p=0.09)
1.26 (0.31-1.71)
(p=0.75)
1.82 (1.25-2.66)
(p=0.002)
1.51 (0.98-2.34)
(p=0.06)
1.58(1.21-2.06)
(p=0.001)
1.48(1.09-2.02)
(p=0.01)
1.89 (1.27-2.84)
(p=0.002)
1.47 (0.90-2.40)
(p=0.12)
1.68 (1.29-2.18)
(p<0.001)
1.56 (1.16-2.10)
(p=0.009)
Quartiles
3
1.21 (0.89-1.64)
(p=0.22)
1.15 (0.85-1.57)
(p=0.37)
2.23 (1.54-3.24)
(p<0.001)
1.52 (0.99-2.35)
(p=0.06)
1.66 (1.27-2.17)
(p<0.001)
1.30 (0.95-1.77)
(p=0.10)
2.31 (1.55-3.44)
(p<0.001)
1.49 (0.92-2.41)
(p=0.10)
1.73 (1.33-2.26)
(p<0.001)
1.35(1.00-1.83)
(p=0.05)
4
1.50 (1.11-2.03)
(p=0.008)
1.39 (1.03-1.89)
(p=0.03)
5.39 (3.70-7.84)
(p<0.001)
3.46 (2.21-5.41)
(p<0.001)
2.83 (2.14-3.76)
(p=0.001)
1.92 (1.38-2.69)
(p<0.001)
5.44 (3.67-8.07)
(p<0.001)
3.02 (1.84-4.95)
(p<0.001)
3.02 (2.28-3.99)
(p<0.001)
2.18 (1.57-3.02)
(p<0.001)
P for
trend
0.02
<0.001
<0.001
<0.001
<0.001
Adjusted for Framingham risk factors (age, gender, SBP, DBP, HDL, LDL, smoking,
diabetes) and medication: statin, angiotensin converting enzyme (ACE), aspirin (ASA),
β-blocker, and renal function (estimated creatinine clearance). Significant Odds ratios
(95% CIs) are shown in bold with p values in red.
WWW.NATURE.COM/ NATURE | 9
Supplementary Table 5b. Odds ratio (95% CI) of cardiovascular phenotypes
based on quartiles of plasma levels of TMAO. (n=1876)
Risk
MI
PAD
CAD
PAD+
CAD
CVD
Model
unadjusted
1
1
adjusted
1
unadjusted
1
adjusted
1
unadjusted
1
adjusted
1
unadjusted
1
adjusted
1
unadjusted
1
adjusted
1
2
1.39 (1.01-1.91)
(p=0.04)
1.37 (1.00-1.89)
(p=0.05)
1.66 (1.13-2.43)
(p=0.009)
1.58(1.02-2.44)
(p=0.04)
1.57 (1.20-2.04)
(p=0.001)
1.66 (1.22-2.26)
(p=0.001)
2.05 (1.35-3.11)
(p=0.001)
2.07 (1.25-3.43)
(p=0.005)
1.52 (1.17-1.97)
(p=0.002)
1.56 (1.16-2.11)
(p=0.003)
Quartiles
3
1.54 (1.12-2.11)
(p=0.008)
1.48 (1.08-2.04)
(p=0.01)
2.26 (1.55-3.28)
(p<0.001)
1.78 (1.15-2.77)
(p=0.01)
1.70 (1.30-2.22)
(p<0.001)
1.59 (1.16-2.18)
(p=0.004)
2.62 (1.74-3.96)
(p<0.001)
2.10 (1.26-3.5)
(p=0.004)
1.63 (1.25-2.11)
(p<0.001)
1.5 (1.11-2.04)
(p=0.008)
4
2.23 (1.64-3.03)
(p<0.001)
2.08 (1.52-2.84)
(p<0.001)
5.83 (4.00-8.49)
(p<0.001)
4.32 (2.73-6.84)
(p<0.001)
3.55 (2.65-4.74)
(p<0.001)
3.25 (2.29-4.63)
(p<0.001)
7.23 (4.81-10.9)
(p<0.001)
5.71 (3.36-9.7)
(p<0.001)
3.46 (2.59-4.62)
(p<0.001)
3.06 (2.17-4.31)
(p<0.001)
P for
trend
<0.001
<0.001
<0.001
<0.001
<0.001
Adjusted for Framingham risk factors (age, gender, SBP, DBP, HDL, LDL, smoking,
diabetes) and medication: statin, angiotensin converting enzyme (ACE), aspirin (ASA),
β-blocker, and renal function (estimated creatinine clearance). Significant Odds ratios
(95% CIs) are shown in bold with p values in red.
WWW.NATURE.COM/ NATURE | 10
Supplementary Table 5c. Odds ratio (95% CI) of cardiovascular phenotypes
based on quartiles of plasma levels of betaine. (n=1876)
Risk
MI
Model
1
unadjusted 1
adjusted
1
unadjusted 1
PAD
adjusted
1
unadjusted 1
CAD
adjusted
1
unadjusted 1
PAD+
CAD
adjusted
1
unadjusted 1
CVD
adjusted
1
2
1.49 (1.09-2.03)
(p=0.01)
1.52 (1.11-2.08)
(p=0.009)
1.48 (1.03-2.11)
(p=0.03)
1.15 (0.75-1.78)
(p=0.52)
1.56 (1.19-2.04)
(p=0.001)
1.51(1.1-2.08)
(p=0.01)
1.73 (1.18-2.53)
(p=0.005)
1.43 (0.88-2.34)
(p=0.15)
1.56 (1.20-2.04)
(p=0.001)
1.45(1.07-1.98)
(p=0.02)
Quartiles
3
1.45 (1.06-1.99)
(p=0.02)
1.50 (1.09-2.06)
(p=0.01)
1.74 (1.22-2.47)
(p=0.002)
1.38 (0.89-2.15)
(p=0.15)
1.39 (1.06-1.81)
(p=0.02)
1.35 (0.98-1.87)
(p=0.07)
1.95 (1.34-2.85)
(p=0.001)
1.60 (0.97-2.65)
(p=0.07)
1.40 (1.07-1.82)
(p=0.01)
1.37(1.00-1.87)
(p=0.05)
4
1.88 (1.38-2.56)
(p<0.001)
1.96 (1.44-2.68)
(p<0.001)
2.18 (1.53-3.11)
(p<0.001)
1.65(1.06-2.56)
(p=0.03)
2.26 (1.71-2.99)
(p<0.001)
2.16 (1.54-3.03)
(p<0.001)
2.52 (1.73-3.66)
(p<0.001)
2.00 (1.21-3.31)
(p=0.007)
2.20 (1.67-2.9)
(p<0.001)
2.00 (1.44-2.78)
(p<0.001)
P for
trend
<0.001
<0.001
<0.001
<0.001
<0.001
Adjusted for Framingham risk factors (age, gender, SBP, DBP, HDL, LDL, smoking,
diabetes) and medication: statin, angiotensin converting enzyme (ACE), aspirin (ASA),
β-blocker, and renal function (estimated creatinine clearance). Significant Odds ratios
(95% CIs) are shown in bold with p values in red.
WWW.NATURE.COM/ NATURE | 11
Supplementary Table 5d. Odds ratio (95% CI) of cardiovascular phenotypes
based on plasma levels of choline, TMAO and betaine (as continuous
variables, n=1876)
Risk
Model
unadjusted
MI
adjusted
unadjusted
PAD
adjusted
unadjusted
CAD
adjusted
unadjusted
PAD+
CAD
adjusted
unadjusted
CVD
adjusted
choline
1.23(1.10-1.37)
(p<0.001)
1.20 (1.07-1.33)
(p=0.001)
2.01(1.74-2.32)
(p<0.001)
1.68(1.43-1.97)
(p<0.001)
1.62(1.45-1.80)
(p<0.001)
1.43(1.28-1.61)
(p<0.001)
2.05(1.77-2.39)
(p<0.001)
1.64(1.38-1.94)
(p<0.001)
1.62(1.46-1.80)
(p<0.001)
1.45(1.3-1.63)
(p<0.001)
TMAO
1.37(1.23-1.52)
(p<0.001)
1.34(1.20-1.49)
(p<0.001)
2.11(1.82-2.45)
(p<0.001)
1.71(1.46-2)
(p<0.001)
1.73(1.55-1.94)
(p<0.001)
1.53(1.36-1.73)
(p<0.001)
2.24(1.91-2.62)
(p<0.001)
1.78(1.5-2.12)
(p<0.001)
1.72(1.53-1.92)
(p<0.001)
1.52(1.35-1.72)
(p<0.001)
betaine
1.36(1.22-1.53)
(p<0.001)
1.38(1.23-1.55)
(p<0.001)
1.43(1.26-1.63)
(p<0.001)
1.32(1.13-1.54)
(p<0.001)
1.43(1.29-1.58)
(p<0.001)
1.41(1.25-1.57)
(p<0.001)
1.53(1.33-1.77)
(p<0.001)
1.41(1.19-1.67)
(p<0.001)
1.4(1.27-1.54)
(p<0.001)
1.38(1.23-1.54)
(p<0.001)
Adjusted for Framingham risk factors (age, gender, SBP, DBP, HDL, LDL, smoking,
diabetes) and medication: statin, angiotensin converting enzyme (ACE), aspirin (ASA),
β-blocker, and renal function (estimated creatinine clearance).
WWW.NATURE.COM/ NATURE | 12
T able 5e. Somers' Dxy rank correlation between plasma analytes and cardiac
phenotypes (n=1876).
Cardiac Phenotype
Analyte
Dxy
p-value
MI
MI
MI
Choline
TMAO
Betaine
0.084
0.184
0.139
0.006
<0.001
<0.001
CAD
CAD
CAD
Choline
TMAO
Betaine
0.228
0.263
0.172
<0.001
<0.001
<0.001
PAD
PAD
PAD
Choline
TMAO
Betaine
0.342
0.364
0.177
<0.001
<0.001
<0.001
CAD+PAD
CAD+PAD
CAD+PAD
Choline
TMAO
Betaine
0.349
0.393
0.205
<0.001
<0.001
<0.001
CVD
CVD
CVD
Choline
TMAO
Betaine
0.231
0.259
0.163
<0.001
<0.001
<0.001
Somers’ Dxy rank correlation was used to examine the association between the
continuous variables (plasma levels of analytes) including choline, TMAO
(trimethylamine N-oxide) and betaine, and the binary variables (cardiac phenotypes)
including MI (myocardial infarction), CAD (coronary artery disease), PAD (peripheral
artery disease), CVD (cardiovascular disease) and the combined presence of CAD
and PAD (CAD+PAD).
WWW.NATURE.COM/ NATURE | 13
Supplementary Table 6.
Plasma and liver lipids and glucose in 20 week C57BL/6J.Apoe-/- mice on either normal
chow or normal chow supplemented with the indicated concentrations of choline or
TMAO.
male mice
Lipids
Triglyceride
(mg/dL)
Cholesterol
(mg/dL)
Plasma
Liver
Chow
(n=9)
114±16
426±28
LDLc (mg /dL)
324±36
HDLc (mg /dL)
61.3±7.3
Glucose (mg/dL)
238±28
Triglyceride (mg/g
protein)
Cholesterol (mg/g
protein)
34.9±4.4
10.1±0.8
0.5% choline
(n=10)
150±22
(p=0.20)
387±25
(p=0.32)
291±26
(p=0.50)
63.4±5.8
(p=0.83)
290±18
(p=0.15)
42.9±8.2
(p=0.47)
9.6±0.6
(p=0.64)
1.0% choline
(n=10)
99±13
(p=0.46)
346±10
(p=0.02)
308±29
(p=0.74)
54.8±6.8
(p=0.55)
191±33
(p=0.29)
48.0±5.3
(p=0.07)
10.4±0.4
(p=0.72)
0.12% TMAO
(n=13)
104±8
(p=0.58)
352±12
(p=0.04)
285±13
(p=0.38)
49.6±4.0
(p=0.23)
248±20
(p=0.78)
47.2±8.5
(p=0.22)
12.0±1.2
(p=0.18)
1.0% choline
(n=11)
97±7
(p=0.57)
384±12
(p=0.46)
234±78
(p=0.70)
46.7±6.5
(p=0.91)
238±10
(p=0.93)
85.5±14.9
(p=0.08)
13.7±1.1
(p=0.40)
0.12% TMAO
(n=10)
104±10
(p=0.99)
347±15
(p=0.53)
271±24
(p=0.84)
31.6±8.5
(p=0.28)
238±21
(p=0.96)
103.7±26.1
(p=0.10)
14.3±1.1
(p=0.21)
female mice
Lipids
Triglyceride (mg/dL)
Chow
(n=11)
103±9
Cholesterol (mg/dL)
364±23
LDLc (mg/dL)
271±24
Plasma
HDLc (mg/dL)
Liver
48.2±9.6
Glucose (mg/dL)
239±16
Triglyceride (mg/g
protein)
Cholesterol (mg/g
protein)
53.8±8.1
12.6±0.8
0.5% choline
(n=10)
101±12
(p=0.88)
356±25
(p=0.81)
251±12
(p=0.50)
34.2±7.5
(p=0.31)
230±12
(p=0.63)
85.4±11.8
(p=0.06)
14.4±1.1
(p=0.20)
Data are presented as mean±SE from the indicated numbers of mice in each
group. P values were given by student’s t test with 2 tails versus normal chow
feeding group in the same gender.
WWW.NATURE.COM/ NATURE | 14
Supplementary Table 7.
Correlation of hepatic gene exression levels for seven (mouse, n=285) or six
(human, n=465) members of the Fmo Gene Family.
Mouse
Fmo1
Fmo2
Fmo3
Fmo4
Fmo5
Fmo6
Fmo9
Fmo1
1
0.76
(<0.001)
0.75
(<0.001)
0.78
(<0.001)
-0.47
(<0.001)
-0.35
(<0.001)
0.001
(0.98)
Fmo2
Fmo3
Fmo4
Fmo5
Fmo6
0.85
(<0.001)
0.73
(<0.001)
-0.64
(<0.001)
-0.36
(<0.001)
-0.06
(0.31)
1
0.78
(<0.001)
-0.71
(<0.001)
-0.26
(<0.001)
-0.022
(0.72)
1
-0.57
(<0.001)
-0.35
(<0.001)
-0.069
(0.25)
1
0.11
(0.06)
0.069
(0.24)
1
Fmo1
1
0.03
(0.55)
0.37
(<0.001)
0.26
(<0.001)
0.25
(<0.001)
0.03
(0.59)
Fmo2
Fmo3
Fmo4
Fmo5
Fmo6
Fmo9
1
0.092
(0.12)
1
Human
Fmo1
Fmo2
Fmo3
Fmo4
Fmo5
Fmo6
1
-0.15
(0.001)
-0.26
(<0.001)
-0.10
(0.03)
-0.19
(<0.001)
1
0.52
(<0.001)
0.82
(<0.001)
-0.12
(0.009)
1
0.46
(<0.001)
0.25
(<0.001)
1
-0.19
(<0.001)
1
Data are shown as correlation coefficient values (r) between hepatic mRNA
levels of the indicated FMO enzymes. Human hepatic Fmos expression
data were taken from a publicly available dataset of Caucasian liver samples, as
previously reported by Schadt et al. (PLoS Biol, 2008. 6:e107). Prior to analysis,
expression data for each gene was averaged across two or more probes present
on the custom Agilent microarray.
WWW.NATURE.COM/ NATURE | 15
Supplementary Table 8.
Liver lipids in 20 week C57BL/6J.Apoe-/- mice on either control diet (choline%, 0.08±
0.01) or choline supplemented diet (choline%, 1.0%) off and on antibiotics (Abx).
Male mice
Control
(n=11)
Choline
(n=5)
36.1±8.2
1.76±0.06
diet
lipids
Triglyceride
(mg/g liver)
Cholesterol
(mg/g liver)
P (choline
vs control)
Control/Abx Choline/Abx P (choline
(n=10)
(n=5)
/Abx vs
control/Abx)
58.9±4.8
0.03
46.8±8.9
56.8±8.2
0.42
1.62±0.06
0.15
2.09±0.11
2.24±0.12
0.35
P (choline
vs control)
Control/Abx Choline/Abx
(n=10)
(n=5)
Female mice
diet
lipids
Triglyceride
(mg/g liver)
Cholesterol
(mg/g liver)
Control
(n=9)
Choline
(n=5)
P (choline
/Abx vs
control/Abx)
121.6±31.7 121.3±22.0
0.99
71.5±16.6
79.4±11.5
0.70
3.07±0.27
0.94
2.49±0.14
2.21±0.21
0.31
3.04±0.20
Data are presented as mean±SE from the indicated numbers of mice in each
group. Liver tissues, harvested from mice, were snap-frozen and stored at
-80oC until use. Cholesterol was extracted with chloroform and methanol as
described in Methods and quantified by stable isotope dilution GC/MS using
cholesterol-2,2,3,4,4,6-d6 as internal standard. Triglyceride was measured by
GPO reagent set (Pointe Scientific) after saponification to release glycerol followed by glycerol kinase and glycerol phosphate oxidase catlayzed reaction to
produce H2O2, which can react with 3-hydroxy-2,4,6-tribomobenzoic acid
(TBHB) catalyzed by peroxidase to yield a red colored quinoneimine dye. The
optical density at 540 nm indicates the concentration of triglyceride.
WWW.NATURE.COM/ NATURE | 16
Supplementary Fig. 1
- LOG(P)
a
6
5
4
3
2
1
0
Analytes
presumably
related via
common
pathway
50
100
150
200
m/z
r=0.63
m/z=76
p<0.001
m/z=104
25
0.
1
r=
00
0.
p<
r=
0.
48
p<
0.
00
1
b
m/z=118
Supplementary Figure 1. Identification of plasma analytes associated with risk
for near-term heart attack, stroke and death. a, Analytes identified in m/z 50-200
range with statistically significant difference in levels among subjects who
experienced a major adverse cardiovascular event (MACE = myocardial
infarction (MI), stroke or death) over the ensuing 3 yr period (cases) vs. those
who did not experience MACE (controls) within the Learning and Validation
Cohorts. Significant differences in analyte levels between cases vs. controls were
determined following adjustments for multiple sampling, with –log(p) values
plotted on Y axis versus mass to charge ratio (m/z) of analyte on X axis. Filled
circles represent the analytes (m/z 76, m/z 104, m/z 118), which are focused on
in this study. b, A significant correlation was noted among three analytes (m/z 76,
m/z 104, m/z 118) whose levels tracked with MACE risk in both Learning and
Validation Cohorts, suggesting their potential involvement in a common
biochemical pathway.
WWW.NATURE.COM/ NATURE | 17
Supplementary Fig. 2
a
MH+
Candidate plasma analyte m/z=76
Trimethylamino (CH3)3N+OH
N-oxide
(TMAO)
CH3CH(+NH3)CH2OH
3-Amino1-propanol
+NH CH CH CH OH
3
2
2
2
Methylaminoethanol
CH3+NH2CH2CH2OH
Glycolamide
HOCH2CO+NH3
0
100
76
HONHC(+NH3)=NH
Glycine
+NH CH COOH
3
2
1.60
.0
.0
0
100
76
76
76
76
Hydroxyguanidine
MH+=76
8.22
Relative Intensity (%)
2-Amino1-propanol
LC2
MH+=76
0
100
76
CH3CH(OH)CH2+NH3
1-Amino2-propanol
LC1
100
.0
0
100
.
0
100
0
100
0
100
76
0
100
76
0
100
N-Isopropyl- (CH3)2CH+NH3OH
hydroxylamine
76
0
2
4
6
8 10 12
2
Retention time (min)
100
Plasma
58
59
Deriv-TMAO
(standard)
100
100
Trimethylamine N-oxide
(TMAO)
59
58
76
0
Relative Intensity (%)
Relative Intensity (%)
76
0
- Cl-
c
44
0
100
40 60
m/z
80
44
N 72
O
88
M+=219
131
117
184
d
Cl
Cl
40
72
88
95
M+
219
80
M+
131
117
120
184
6
8 10 12
Derivatized-TMAO
(standard)
100
Cl
Deriv-plasma m/z=76 analyte
44
0
72
88
95
95
184
131
117
O
Relative Intensity (%)
Analyte m/z=76
b
4
Retention time (min)
0
Derivatized-plasma
analyte m/z=76
100
219
0
160
m/z
200
240
3.5 4.0 4.5 5.0
Time (min)
WWW.NATURE.COM/ NATURE | 18
Supplementary Figure 2. Identification of the plasma analyte at m/z 76
associated with cardiac risk to be trimethylamine N-oxide (TMAO). a,
Positive ion chromatograms and the 9 potential structures of candidate
compounds with m/z 76 in positive ion mode are shown, with separations
performed on a phenyl column under two distinct chromatographic conditions
(LC1 and LC2, which employed methanol/ammonium formate and
acetonitrile/formate gradients, respectively) developed to help distinguish these
candidate species. Note that most of these analytes share the identical elemental
composition and that the unidentified plasma analyte with m/z 76 shows identical
retention time only to authentic TMAO standard. b, Demonstration of identical
collision induced dissociation (CID) mass spectra of TMAO standard and the
isolated plasma analyte with m/z 76. c, Demonstration of identical full scan mass
spectra of the trichloroethyl formate derivatives of TiCl3 reduced TMAO standard
and the isolated plasma analyte with m/z 76 associated with CVD risks. d,
Demonstration of identical retention times of gas chromatography mass
spectrometry analyses of trichloroethyl formate derivatives of TiCl3 reduced
authentic TMAO standard and the plasma analyte with m/z 76 in positive ion
electron impact mode.
WWW.NATURE.COM/ NATURE | 19
Supplementary Figure 3
a
Relative
Intensity (%)
100
Relative
Intensity (%)
human
male mouse
100 m/z=76
m/z=76
m/z=104
m/z=118
0
Relative
Intensity (%)
2 6 10
Time (min)
8 12 16
Time (min)
2 6 10
Time (min)
m/z=104
0
0.5
1.0
2.5
3.0
3.5
4.0
4.5
m/z=118
5.0
8.5
female mouse
100 m/z=76
7.5
8.0
8.5
7.0
7.5
8.0
8.5
2.5
3.5
4.0
4.5
5.0
5.5
6.0
6.5
2 6 10
Time (min)
8 12 16
Time (min)
2 6 10
Time (min)
7.0
m/z=118
m/z=104
0
2 6 10
Time (min)
8 12 16
Time (min)
2 6 10
Time (min)
1
0
0 1 2 3 4
Time (hr)
Intensity (x105)
Intensity (x105)
Male mouse, 1.5 mg egg yolk PC gavage
m/z=76
m/z=104
2
4
2
0
0 1 2 3 4
Time (hr)
Intensity (x105)
b
Intensity (x105)
Intensity (x105)
Intensity (x105)
Female mouse, 1.5 mg egg yolk PC gavage
m/z=76
m/z=104
9
8
6
4
3
0
0
0 1 2 3 4
0 1 2 3 4
Time (hr)
Time (hr)
16
m/z=118
8
0
0 1 2 3 4
Time (hr)
m/z=118
4
2
0
0 1 2 3 4
Time (hr)
WWW.NATURE.COM/ NATURE | 20
Supplementary Figure 3. Demonstration of analytes at m/z 76, 104 and 118
in plasma from mice and humans with identical retention times, and their
origins from dietary phosphatidylcholine. a, Typical LC chromatograms
(phenyl column) for plasma from human and both male and female C57BL/6J
mice, extracted at m/z 76, 104 and 118 in positive-ion MS1 mode. b, The
increased integrated peak area (intensity) in extracted ion LC chromatograms at
m/z 76, 104 and 118 from mice plasma after 1.5 mg bolus supplementation
(gavage) of phosphatidylcholine (egg yolk PC). Mice were fasted overnight prior
to supplementation. Data are presented as mean ± SE from 3 independent
replicates.
WWW.NATURE.COM/ NATURE | 21
Supplementary Figure 4
a
MH+
MS1, m/z=104
100
Plasma analyte m/z=104
OH
N+
choline
2-amino-3-methyl1-butanol
0
100
104
OH
0
100
104
NH2
0
100
O
H2N
104
OH
0
100
O
NH2
O
104
NH2
N
H
Relative Intensity (%)
O
H2N
biuret
104
OH
2-aminobutyric
acid
O
N,N-dimethyl
glycine
N
104
OH
CN
104
benzonitrile
diethylenetriamine
H2N
ethyl-N-hydroxyl
acetimidate
2-isopropylaminoethanol
HO
45
N
100
40
60
c
60
80
100
choline
5x
120
104
45
60
0
40
60
80
m/z
100
6 .3 8
7 .2 2
7 .9 4
8 .8 4
9 .6 8
1 0 .9 2
1 1 .1 7
1 2 .9 1
1 3 .2 3
1 4 .1 0
1 5 .2 9 1 6 .0 8
1 7 .0 7
1 .6 4
1 .7 3
1 .9 1
2 .7 1
3 .5 7
4 .3 9
6 .1 6
9 .0 8
6 .8 8
0 .1 8
7 .6 0
1 3 .6 5
1 2 .4 6
1 4 .3 2
1 6 .2 6
1 7 .1 4
0
100
0 .2 1
2 .0 3
3 .1 6
3 .5 7
5 .6 6 5 .8 2
8 .0 9
9 .6 3
1 0 .4 6
1 1 .5 4
1 2 .3 6
1 2 .7 2
1 4 .2 7
1 5 .6 8
1 4 .8 9
1 5 .2 1
1 7 .2 2
6 .5 2
0
100
0 .3 4
7 .8 1
1 .2 6
2 .2 1
3 .0 8 3 .7 0
4 .4 4
8 .5 4
9 .3 8
1 0 .7 4
1 1 .0 4
1 3 .4 7
1 3 .8 0
1 6 .8 3 1 7 .0 4
0
100
104
0
4 8 12 16 20
Retention time (min)
4 4 .9
0
4 .4 4 5 .0 8
5 .6 4
0 .8 6
104
OH
104
5 9 .9
3 .7 9
0 .0
0 .4 5
104
120
Relative Intensity (%)
Relative Intensity (%)
5x
2 .8 8
0
100
OH
Analyte m/z=104
Plasma
0
100
104
N
H
N
H
0
100
0
100
O
1-dimethyl-amino2-propanol
100
104
N-OH
2-propylaminoethanol
b
NH2
N
H
0
100
100
d
1-dimethylamino
-2-propanol
0
40
100
60
80
100
120
benzonitrile
0
40
60
80
m/z
100
120
Relative Intensity (%)
3-aminoisobutyric
acid
Plasma
MRM, m/z=104
60
100 Plasma
0
100
choline
0
4
8
12
16 20
Retention time (min)
WWW.NATURE.COM/ NATURE | 22
Supplementary Figure 4. Identification of analyte at m/z 104 to be choline. a,
Candidate compounds with m/z 104 and extracted ion chromatograms in
positive-ion MS1 mode of the component purified from plasma with m/z 104
associated with CVD risks, choline and other standards by reverse phase HPLC
coupled to tandem mass spectrometry. Sample (20 µl) was injected onto a
phenyl column (4.6 × 250 mm, 5 µm Rexchrom Phenyl) (Regis) at a flow rate of
0.8 ml/min. The separation was performed using a gradient starting from 10 mM
ammonium formate over 0.5 min, then to 5 mM ammonium formate, 25 %
methanol and 0.1 % formic acid over 3 min, held for 8 min, followed by 100%
methanol and water washing for 3 min each. b, Demonstration of identical
collision induced dissociation (CID) positive ion mass spectra of isolated human
plasma analyte with m/z 104 correlated with CVD risks and choline standard. c,
Demonstration of CID mass spectrum of 1-dimethylamino-2-propanol and
benzonitrile, a candidate compound with similar retention time to choline, but
distinct mass spectra from the human plasma analyte with m/z 104 correlated
with CVD risks. d, LC/MS/MS analysis (reverse phase column) in positive-ion
multiple reaction monitoring mode with parent to daughter transition, 104→60, of
both the human plasma analyte with m/z 104 correlated with CVD and choline
standard.
WWW.NATURE.COM/ NATURE | 23
Supplementary Figure 5
CH
O
3
al
Or
or
IP
3
CH
HO
CH
3
3
CD
CD
2
CH
3
CH
3
CD
CH
3
CD2
3
2-methyl-choline-d3
m/z=121
l- se
hy ra
et fe
M ans
tr
d4-choline
m/z=108
1-Methyl-choline-d3
m/z=121
CH
3
CH
3
N+
CH
2
Methyltransferase
CH
3
HO
N+
2
CD
N+
CD
CH
3 CH
3
N+
3
betaine-d2
m/z=120
3
3
CH
2
CH
[O]
CH
C
CD
HO
N+
CH
HO
a
CD
CD
2
2
CH
CH3
O
3
choline methyl ether-d4
m/z=122
b
Plasma analyte at m/z=118 100
CID MS spectrum for plasma analyte
at m/z=118
Relative Intensity (%)
Relative Intensity (%)
100
0
118
58 59
0
4
6
8
10
30
60
90
Time (min)
100
120
m/z
100
CID MS spectrum for betaine
at m/z=118
118
Relative Intensity (%)
Relative Intensity (%)
betaine at m/z=118
0
CH
CH
3
3
N+ CH
.N+
CH
CH
3
HO
CH
C
CH2
3
N+ CH
3
CH
3
3
2
m/z=58
O
CH
3
m/z=118
m/z=59
58 59
0
4
6
8
Time (min)
10
30
60
90
120
m/z
WWW.NATURE.COM/ NATURE | 24
Supplementary Figure 5. Choline isotope labeling strategy for
identification of human plasma analyte with m/z 118 correlated
with CVD risks as betaine. PC and choline feeding studies confirmed
that the plasma analyte with m/z 118 was a metabolite of both PC and
choline. To discriminate amongst multiple potential methylated choline
metabolites, synthetic cholines with regiospecific isotope labeling were
used since metabolites produced from the distinct choline isotopomers
would generate isotopomeric forms with distinct m/z as illustrated. a,
Scheme shown is for methylation of choline [1,1, 2, 2-d4 isotopomer].
Plasma was collected from mice 4 h following gavage and the m/z for
each potential candidate molecule were scanned by mass spectrometry
in positive MS1 mode. b, The LC chromatogram of plasma analytes
extracted at m/z 118 and its CID mass spectrum show identical pattern
to those of betaine.
WWW.NATURE.COM/ NATURE | 25
Supplementary Fig. 6
m/z
104 (choline)
5
Fold Increase
b
Choline gavage
4
)
MAO
T
(
6
7
3
2
1
0
118
(betaine, methylcholine,
or choline methyl ether)
0
2
Time (hour)
4
Relative Ion counts
a
d4-Choline gavage
1.0
d2-betaine
0.5
d3-methylcholine
0
d4-choline methyl ether
0
2
4
Time (hour)
Supplementary Figure 6. Metabolites of choline monitored and
determination of the metabolite with m/z=118. a, Demonstration that oral
(gavage) choline supplementation of mice results in time-dependent increase in
plasma analytes with m/z 76, m/z 104 and m/z 118 possessing identical retention
times to the analytes with m/z 76, m/z 104 and m/z 118 identified in human
plasma associated with CVD risk. b, Identification of betaine as the plasma
analyte with m/z 118 through use of isotope tracer studies following (4h) oral
administration of choline-1,1,2,2-d4. Data are presented as mean ± SE from 4 (2
males and 2 females) independent replicates.
WWW.NATURE.COM/ NATURE | 26
Supplementary Figure 7
ine
a
t
be
0.5
TMAO
TMA
choline
TMAO
TMA
0.0
0
1
2
3
Time (h)
4
Supplementary Figure 7a. Gut flora is
required for the appearance of TMA and
TMAO in plasma from oral route. Plasma
levels of isotope labeled choline metabolites
following oral (gavage) vs. intraperitoneal (i.p.)
challenge with d9(trimethyl)-choline. Note that
the appearance of detectable trimethylamine
(TMA) and TMAO has an obligatory
requirement for oral route, whereas choline
and betaine do not. For the mice receiving oral
d9(trimethyl)-choline, concentrations are
normalized to the isotope labeled choline at 1
hour; for the i.p. mice, concentrations of
analytes are normalized to the isotope labeled
betaine at 1 hour. Data presented are mean ±
SE from 2 independent experiments with at
least 3 mice in each experiment .
Post-antibiotics
25
0
0
0
1
2
Time (h)
Choline
Betaine
9
TMAO
0
0
1
2
Time (h)
TMAO
0
1
2
Time (h)
4
Conventionalized
18
Conc (µM)
Conc (µM)
18
9
0
4
Germ - free
c
Choline
Betaine
10
TM
AO
9
18
Conc (µM)
50
Choline
Betaine
TM
AO
Conc (µM)
18
Conc (µM)
Pre-antibiotics
b
Betaine
9
5
Choline
0
4
Conc (µM)
1.0
oral d9-choline (gavage)
i.p. d9-choline
chol
ine
e
in
ta
be
Relative Conc
a
0
0
1
2
Time (h)
4
WWW.NATURE.COM/ NATURE | 27
Supplementary Figure 7b,c. Gut flora plays an obligatory role in generation
of TMAO. b, Demonstration of obligatory role for gut flora in TMAO, but not
choline or betaine appearance in plasma of mice following choline oral challenge.
c, Independent confirmation of requirement for gut flora in TMAO production from
oral choline using germ-free mice. Note that germ-free mice fail to produce
TMAO following choline challenge. However, following housing of germ-free
mice within conventional cages with non-sterile mice for several weeks (i.e. –
“conventionalization”), there is recovery of the capacity to formTMAO from oral
choline challenge. Data presented are mean ± SE from 5 independent replicates.
WWW.NATURE.COM/ NATURE | 28
Supplementary Figure 8
2.5
2.0
choline-d9
Peak area of choline /
Development of stable isotope dilution HPLC with on-line tandem mass
spectrometry assay to quantify plasma levels of choline, TMAO and betaine
1.5
1.0
y = 0.0188x + 0.3032
R2 = 0.9995
0.5
0.0
100
2.0
1.5
1.0
y = 0.0219x + 0.0203
R2 = 0.9998
0.5
0.0
0
50
TMAO (µM)
100
1.6
TMAO-d9
Peak area of betaine /
50
choline (µM)
2.5
TMAO-d9
Peak area of TMANO /
0
1.2
0.8
y = 0.0111x + 0.3718
R2 = 0.9969
0.4
0
0
50
betaine (µM)
100
Supplementary Figure 8. Standard curves for LC/ESI/MS/MS analysis of choline,
TMAO and betaine within plasma matrix. Varying levels of each analyte were spiked
into 20 µl control plasma followed by precipitation with 80 µl methanol containing 0.8
nmol choline-d9 and 0.8 nmol TMAO-d9. 20 µl supernatant was loaded to LC/MS.
Analyses were performed using electrospray ionization in positive-ion mode with multiple
reaction monitoring of parent and characteristic daughter ions and retention times
specific for components monitored. The transitions monitored were mass-to-charge ratio
(m/z): m/z 104.1 → 60.1 for choline; 76.0 → 58.0 for TMAO; m/z 118.1 →59.1 for
betaine; m/z 113.1 → 69.1 for choline-d9; and m/z 85.0 → 66.0 for TMAO-d9. Standard
curves of choline, TMAO and betaine were generated by plotting peak area ratio versus
concentration spiked into plasma for each analyte.
WWW.NATURE.COM/ NATURE | 29
Supplementary Figure 9
mouse
Lesion (µm2)
50000
p=0.050
p=0.045
p=0.20
mouse
b
100000
R=0.68
p<0.001
Lesion (µm2)
a
50000
25000
0
Chow 0.5% 1% 0.12%
choline choline TMAO
(n=8) (n=10) (n=10) (n=13)
0
0
20
40
TMAO (µM)
Supplementary Figure 9. Dietary supplementation of mice with choline,
TMAO or betaine promotes atherosclerosis and plasma TMAO is positively
correlated with aortic lesion area in male mice. a, Comparison in aortic lesion
area among 20 week old male C57BL/6J.Apoe- /- mice fed with chow diet
supplemented with the indicated amounts (wt/wt) of choline or TMAO from time
of weaning (4 weeks). b, Relationship between plasma TMAO levels and aortic
lesion area.
WWW.NATURE.COM/ NATURE | 30
60
Supplementary Figure 10
mouse
Male
choline (µM)
80
p=0.38
p=0.20
0
c
50
p=0.74
40
p=0.0001
p=0.02
p=0.005
10
0
chow 0.5% 1% 0.12%
choline choline TMAO
(n=9) (n=10) (n=10) (n=13)
TMAO (µM)
TMAO (µM)
20
25
d
30
p=0.74
p=0.0003
p=0.30
0
chow 0.5% 1% 0.12%
choline choline TMAO
(n=9) (n=10) (n=10) (n=13)
Male
Female
b
choline (µM)
a
chow 0.5% 1% 0.12%
choline choline TMAO
(n=11) (n=10) (n=11) (n=10)
Female p=0.0002
p<0.0001
200
p<0.0001
100
0
chow 0.5% 1% 0.12%
choline choline TMAO
(n=11) (n=10) (n=11) (n=10)
Supplementary Figure 10. Plasma levels of choline (a, b) and TMAO (c, d) in
mice after supplementation with choline and TMAO. Male and female
C57BL/6J.Apoe- /- mice at time of weaning (4 weeks) were placed on diets
consisting of normal chow or normal chow supplemented with the indicated
amount (wt/wt) of choline or TMAO. Plasma was isolated after 16 weeks on the
diets at time of aorta recovery for quantification of aortic sinus lesions. Data are
presented as mean ± SE for the indicated numbers of mice in each group.
WWW.NATURE.COM/ NATURE | 31
Supplementary Figure 11
Male, TC
Female, TC
20
25
VLDL
VLDL
chow
C50
C100
TMAO
20
TMAO
Optical Density (AU)
Optical Density (AU)
16
12
LDL
8
HDL
15
LDL
10
4
5
0
0
0
10
20 30
Fraction
40
chow
C50
C100
50
HDL
0
10
20 30
Fraction
40
50
Supplementary Figure 11. FPLC analysis of plasma lipoprotein on the chow
and 0.5% choline (C50), 1.0% choline (C100) and 0.12% TMAO
supplemented chow diets. Plasma from mice on diets consisting of normal
chow (chow) or normal chow supplemented with 0.5% (wt/wt) choline (C50),
1.0% choline (C100) or 0.12% TMAO TMAO) was collected after 16 weeks on
the diets at time of aorta recovery. Pooled aliquots from at least 6 mice in each
group were analyzed by gel filtration FPLC for separation of lipoprotein fractions.
WWW.NATURE.COM/ NATURE | 32
Supplementary Figure 12
Mouse
R = 0.35 (p<1x
400000
200000
R = 0.05 (p=0.61)
250000
-2.0
-1.0
0
1.0
0
-2.0
-1.0
Fmo3 Expression
(Mean Log Ratio)
1.0
-2.0
50
R = -0.19 (p=0.032)
100
0
0
-2.0
1.0
3
Plasma TMAO (µM)
0
0
1.0
-0.5
1
2.5
Fmo3 Expression
(Mean Log Ratio)
-1.0
0
1.0
Fmo3 Expression
(Mean Log Ratio)
Female
30
R = 0.85
p=0.004
2
1
0
-2
-2.0
Male
R = 0.80
p<0.001
10
-1.0
Fmo3 Expression
(Mean Log Ratio)
Fmo3 Expression
(Mean Log Ratio)
Male+Female
20
R = -0.44 (p<1x 10-6)
0
Plasma TMAO (µM)
-1.0
1.0
60
0
-2.0
0
Female
120
HDL Cholesterol
(mg/dL)
HDL Cholesterol
(mg/dL)
100
-1.0
Fmo3 Expression
(Mean Log Ratio)
Male
200
10-15)
150
Plasma TMAO (µM)
HDL Cholesterol
(mg/dL)
0
Fmo3 Expression
(Mean Log Ratio)
Male+Female
R = -0.47 (p<1x
R = 0.29 (p=0.002)
400000
0
0
30
Female
800000
Aortic Lesion (µm 2)
600000
Male
500000
10-7)
Aortic Lesion (µm 2)
Aortic Lesion (µm 2)
Male+Female
R = 0.80
p<0.001
20
10
0
-2
-1.5
Fmo3 Expression
(Mean Log Ratio)
-1
-2
-0.5
1
2.5
Fmo3 Expression
(Mean Log Ratio)
Supplementary Figure 12. Relation between hepatic Fmo3 expression
levels in mice from the C57BL/6J.Apoe- /- and C3H/HeJ Apoe- /- F2
intercross and various phenotypes. The correlations shown are between
hepatic expression levels in combined male and female mice (left column), male
mice (middle column), and female mice (right column) from the F2 intercross
versus aortic sinus atherosclerotic lesion area (top row), HDL cholesterol levels
(middle row) and plasma TMAO levels (bottom row).
WWW.NATURE.COM/ NATURE | 33
Supplementary Figure 13
Hepatic expression levels of Fmo3 in adult male mice is extremely low. This
may have accounted for the lack of correlation with lesion area observed in the
microarry study (Supplementary Fig 12, top panel, middle). We therefore
performed separate analyses using a more sensitive method to quantify Fmo3
mRNA (qRT-PCR), which is shown below. Note that using the more sensitive
and accurate qRT-PCR data, a significant positive correlation between hepatic
Fmo3 expression levels and quantitative measures of atherosclerotic plaque
burden in males were observed.
Male mouse hepatic Fmo3 mRNA expression level relative to Gapdh
quantified by RT-PCR and the lowest value was normalized to 1.
100000
r=0.58; p<0.001
Aortic lesion (µm2)
80000
60000
40000
20000
0
0
5
10
15
20
25
30
35
Fmo3 mRNA
Supplementary Figure 13. Correlation between hepatic Fmo3 mRNA
expression levels and aortic lesion. Male C57BL/6J.Apoe- /- mice at time of
weaning (4 weeks) were placed on diets consisting of normal chow or normal
chow supplemented with varying amounts of choline or TMAO as indicated in Fig.
3d of the manuscript. Liver was isolated after 16 weeks on the diets at time of
aorta recovery for quantification of aortic sinus lesions. Hepatic Fmo3 mRNA
expression levels were quantified by RT-PCR. Expression levels are normalized
to Gapdh mRNA, and the minimum value was set to 1.0. Aortic lesion area was
quantified after staining with Oil-red-O / haematoxylin.
WWW.NATURE.COM/ NATURE | 34
Supplementary Figure 14
mouse
r=0.10; p=0.09
r=0.19; p=0.002
600000
Aortic lesion (µm2)
Aortic lesion (µm2)
600000
300000
300000
0
0
-0.2
0
0.2
Fmo1 Expression
(Mean Log Ratio)
-0.8
-0.4
0
0.4
Fmo2 Expression
(Mean Log Ratio)
r=0.35; p<0.001
r=0.12; p=0.04
600000
Aortic lesion (µm2)
Aortic lesion (µm2)
600000
300000
0
-2
-1
0
Fmo3 Expression
(Mean Log Ratio)
300000
300000
0
0
1
-0.2
r=-0.23; p<0.001
r=-0.01; p=0.87
600000
Aortic lesion (µm2)
Aortic lesion (µm2)
600000
300000
0
300000
0
-0.4
-0.2
0
Fmo5 Expression
(Mean Log Ratio)
0.2
r=0.001; p=0.98
600000
Aortic lesion (µm2)
0
0.2
Fmo4 Expression
(Mean Log Ratio)
300000
0
-0.6
-0.2
0.2
Fmo9 Expression
(Mean Log Ratio)
0.6
-0.2
0
Fmo6 Expression
(Mean Log Ratio)
0.2
Supplementary Figure 14. Correlation
between expression levels of multiple
members of the murine family of
hepatic flavin monooxygenase (Fmos)
and aortic lesion. Hepatic expression
levels of the Fmos in t he F2 intercross
between atherosclerosis prone
C57BL/6J.Apoe- /- and atherosclerosis
resistant C3H/HeJ Apoe- /- mice were
determined by microarray analysis
(described in Methods) and aorta
lesions were quantified after staining
with Oil-red-O/haematoxylin.
WWW.NATURE.COM/ NATURE | 35
Supplementary Figure 15
mouse
200
r=-0.31; p<0.001
100
100
0
0
-0.2
-0.8
0
0.2
Fmo1 Expression
(Mean Log Ratio)
200
r=-0.48; p<0.001
HDL Cholesterol
(mg/dL)
HDL Cholesterol
(mg/dL)
200
100
0
200
-1
0
Fmo3 Expression
(Mean Log Ratio)
r=-0.36; p<0.001
100
1
-0.2
200
r=0.38; p<0.001
HDL Cholesterol
(mg/dL)
HDL Cholesterol
(mg/dL)
-0.4
0
0.4
Fmo2 Expression
(Mean Log Ratio)
0
-2
100
0
0
0.2
Fmo4 Expression
(Mean Log Ratio)
r=0.34; p=0.57
100
0
-0.4
200
HDL Cholesterol
(mg/dL)
r=-0.38; p<0.001
HDL Cholesterol
(mg/dL)
HDL Cholesterol
(mg/dL)
200
-0.2
0
Fmo5 Expression
(Mean Log Ratio)
0.2
r=0.012; p=0.84
100
0
-0.6
-0.2
0.2
Fmo9 Expression
(Mean Log Ratio)
0.6
-0.2
0
Fmo6 Expression
(Mean Log Ratio)
0.2
Supplementary Figure 15. Correlation
between expression levels of multiple
members of the murine family of
hepatic flavin monooxygenase (Fmos)
and plasma HDL cholesterol. Hepatic
expression levels of the Fmos in the F2
intercross between atherosclerosis prone
C57BL/6J.Apoe- /- and atherosclerosis
resistant C3H/HeJ Apoe- /- mice were
determined by microarray analysis
(described in Methods) and plasma HDL
cholesterol levels were measured by
Cholesterol LiquiColor® Test (Enzymatic)
(Stanbio, Boerne, Texas).
WWW.NATURE.COM/ NATURE | 36
Supplementary Figure 16
mouse
TMAO (µM)
3
0
-0.1
0
0.1
Fmo1 Expression
(Mean Log Ratio)
3
TMAO (µM)
0
0
Fmo2 Expression
(Mean Log Ratio)
6
0
Fmo4 Expression
(Mean Log Ratio)
0.2
r=-0.24;
p=0.33
3
0
-0.1
0
Fmo9 Expression
(Mean Log Ratio)
0.1
10
3
-2
-0.5
1
Fmo3 Expression
(Mean Log Ratio)
2.5
r=-0.27; p=0.29
6
0
-0.2
20
0.5
r=-0.68; p=0.002
6
3
r=0.80; p<0.001
0
-0.5
0.2
6 r=0.27; p=0.28
TMAO (µM)
30
0
-0.2
TMAO (µM)
r=0.41; p=0.05
TMAO (µM)
TMAO (µM)
6
TMAO (µM)
r=-0.028; p=0.91
6
3
0
-0.4
-0.2
0
Fmo5 Expression
(Mean Log Ratio)
0.2
-0.1
0
Fmo6 Expression
(Mean Log Ratio)
Supplementary Figure 16. Correlation between
expression levels of multiple members of the
murine family of hepatic flavin monooxygenase
(Fmos) and plasma levels of TMAO. Hepatic
expression levels of the Fmos in t he F2 intercross
between atherosclerosis prone C57BL/6J.Apoe- /and atherosclerosis resistant C3H/HeJ Apoe-/- mice
were determined by microarray analysis (described
in Methods), and plasma TMAO levels were
measured using stable isotope dilution LC/MS/MS
as described in Methods.
WWW.NATURE.COM/ NATURE | 37
0.1
Supplementary Figure 17
Human (female) hepatic Fmos mRNA and plasma levels of TMAO
relative to GAPDH
15
r=-0.11
p=0.76
10
TMAO (µM)
TMAO (µM)
15
5
0
5
0
0
4
Fmo1 mRNA
8
0
5
Fmo2 mRNA
10
15
TMAO (µM)
15
TMAO (µM)
r=-0.31
p=0.28
10
10
r=0.43
p=0.13
10
r=-0.49
p=0.07
5
0
5
0
0
1000
2000
Fmo3 mRNA
3000
0
6
Fmo4 mRNA
12
TMAO (µM)
15
10
r=0.31
p=0.31
5
0
0
200
Fmo5 mRNA
400
Supplementary Figure 17. Correlation between expression levels of
multiple members of the human family of hepatic flavin monooxygenases
(Fmos) and plasma levels of TMAO. Human liver biopsies and paired samples
of fasting blood for plasma isolation were collected. Hepatic expression levels for
the Fmos were determined by qRT-PCR in each sample and normalized to the
level of Gapdh mRNA in that sample. Plasma levels of TMAO were quantified
by stable isotope dilution LC/MS/MS.
WWW.NATURE.COM/ NATURE | 38
Supplementary Figure 18
mouse
8
LOD
6
4
2
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14 15 16
17 18
Chromosome Number
Supplementary Figure 18. Genome scan results for Fmo3 expression in
mice from the C57BL/6J.Apoe- /- and C3H/HeJ.Apoe- /- F2 intercross. LOD
score plots for Fmo3 mRNA levels across the 20 autosomal chromosomes in
mice using a sex interaction model adjusting for sex, and sex × QTL. The
horizontal lines represent the significant (Black, p < 0.05) and suggestive (Red, p
< 0.63) LOD thresholds as determined by 1000 permutations. A suggestive ciseQTL for Fmo3 expression (and other Fmo within the Fmo gene cluster) is
found on chromosome 1 over the physical location of the Fmo gene cluster.
WWW.NATURE.COM/ NATURE | 39
19 20
Supplementary Figure 19
mouse
Peritoneal macrophages
p=0.075
p=0.22
Cd36 mRNA
relative to F4/80
3
p=0.075
2
1
0
chow
(n=5)
choline TMAO
(n=5)
(n=4)
betaine
(n=5)
p=0.047
p=0.014
Sr-a1 mRNA
relative to F4/80
6
p=0.075
4
2
0
chow
(n=5)
choline TMAO
(n=5)
(n=4)
betaine
(n=5)
Supplementary Figure 19. Effect of dietary choline, TMAO and betaine on
expression of scavenger receptors by RT-PCR. Female C57BL/6J.Apoe- /mice were placed on diets consisting of normal chow or normal chow
supplemented with 1.0% (wt/wt) of choline, 0.12% TMAO or 1.0% betaine at time
of weaning (4 weeks). After 1 month, peritoneal macrophages were harvested by
lavage, RNA was extracted, and Cd36 and Sr -a1 mRNA levels quantified by
RT-PCR. Expression levels are normalized to control F4/80 mRNA for
comparison.
.
WWW.NATURE.COM/ NATURE | 40
Supplementary Figure 20
a
b
c
- +
-
IgG2b-Alexa Fluor 647
CD11b Fluor 647
IgG2a-Alexa Fluor 647
F4/80 Fluor 647
d
F4/80 positive cells
f
+
e
CD11b positive cells
F4/80 and CD11b positive cells
WWW.NATURE.COM/ NATURE | 41
Supplementary Figure 20. Expression of surface scavenger receptor by
flow cytometry. Cells sorted with the macrophage specific F4/80 and CD11b
double positive staining were used to measure the level of CD36 and SR-A1
protein on the surface of thioglycollate-elicited macrophages in mice. Anti-F4/80
antibody was conjugated with Alexa® 647, anti-CD36 antibody and anti-rat Ig
antibody for SR-A1 were conjugated with FITC. a, A typical flow cytometry
scattergram to sort cells by forward size channel (FSC) and scattering size
channel (SSC). b-c, Histograms of the cells immunostained with rat anti -mouse
F4/80 and anti-mouse CD11b and their corresponding isotope controls, rat IgG2a
and IgG2b, conjugated with Alexa Fluor® 647, respectively. A dotted black line
inside the histograms split cells into F4/80, CD11b positive (+) staining and
negative (-) staining ones. d-e, The extracted F4/80 and CD11b positive staining
cells, respectively. f, The final extracted double positive cells.
WWW.NATURE.COM/ NATURE | 42
Supplementary Figure 21
mouse liver
H/E
Oil red O
MCD
control
diet
choline
diet
100 µm
Supplementary Figure 21. Representative examples showing haematoxylin/eosin
(H/E) and oil-red-O / haematoxylin stainings of liver from mice fed with different
levels of choline diet. C57BL/6J.Apoe-/- mice were placed on the different diets at
weaning (4 weeks) and sacrificed 16 weeks later. The liver sections were stained with
haematoxylin/eosin (H/E) and oil-red-O/haematoxylin. Parallel biochemical analyses on
liver tissues for triglyceride and cholesterol content were also performed. Control diet,
(0.08±0.01)% (wt/wt) total choline; choline diet, 1.0% (wt/wt) total choline; MCD
(methonine and choline deficient diet classically used for induction of steatohepatitis),
0% (wt/wt) total choline.
WWW.NATURE.COM/ NATURE | 43
Supplementary Figure 22
mouse aortic root
Mouse aorta immunohistochemical staining for macrophages using rat anti mouse F4/80
antibody followed by goat anti rat IgG conjugated with FITC
control diet
choline diet
control diet / Abx
choline diet / Abx
200 µm
Supplementary Figure 22. Representative immunohistochemical staining of
aorta for F4/80 in C57BL/6J.Apoe- /- mouse on chow (control diet,
(0.08±0.01)% choline) or choline supplemented diet (choline diet, 1.0%
choline) off or on antibiotics (Abx). The aortic root was sectioned and
immunohistochemically stained with rat anti mouse F4/80 antibody followed by
goat anti rat IgG conjugated with FITC. Section was counterstained with DAPI.
WWW.NATURE.COM/ NATURE | 44
Supplementary Figure 23
mouse aortic root
Aortic lesion imunohistochemical staining for macrophage scavenger receptor CD36 with
anti-CD36 conjugated with FITC and counterstained with DAPI
Control
Choline
Control/Abx
Choline/Abx
200 µm
Supplementary Figure 23. Representative immunohistochemical staining of
aorta for CD36 in C57BL/6J.Apoe- /- mouse on chow (control diet,
(0.08±0.01)% choline) or choline supplemented diet (choline diet, 1.0%
choline) off or on antibiotics (Abx). The aortic root was sectioned and
immunohistochemically stained with rat anti mouse CD36 antibody conjugated
with FITC. Section was counterstained with DAPI.
WWW.NATURE.COM/ NATURE | 45
Supplementary Figure 24
p=0.94
Total Cholesterol (µg/µg DNA)
4
p=0.01
CH
p=0.05
HO CH
CH
2
CH
3
2
+ CH
N
CH
HO
3
3
CH
3
3
Trimethylamine N-oxide
(TMAO)
choline
2
3
+ CH
N
CH
HO CH
2
0
3
CH
2
C
CH
3
CH
3
control choline TMAO DMB
(n=10) (n=8) (n=7) (n=11)
3,3-dimethyl-1-butanol (DMB)
Supplementary Figure 24. Difference in cholesterol accumulated in
macrophages among structurally similar chemical compounds, choline,
TMAO and 3,3-dimethyl-1-butanol (DMB). Male C57BL/6J.Apoe- /- mice (15
week old) were placed on normal chow (control) alone or supplemented in the
presence of either choline (1.0%), TMAO (0.12%), or DMB (1.0%). Peritoneal
macrophages were recovered from the indicated number of mice at 20 weeks of
age and cellular cholesterol content was quantified by stable isotope dilution
GC/MS, and normalized to DNA content.
WWW.NATURE.COM/ NATURE | 46