2

0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
nature medicine advance online publication 1
The high level of meat consumption in the developed world is linked to
CVD risk, presumably owing to the large content of saturated fats and
cholesterol in meat
1,2
. However, a recent meta-analysis of prospective
cohort studies showed no association between dietary saturated fat intake
and CVD, prompting the suggestion that other environmental exposures
linked to increased meat consumption are responsible
3
. In fact, the sus-
picion that the cholesterol and saturated fat content of red meat may
not be sufficiently high enough to account for the observed association
between CVD and meat consumption has stimulated investigation of
alternative disease-promoting exposures that accompany dietary meat
ingestion, such as high salt content or heterocyclic compounds gener-
ated during cooking
4,5
. To our knowledge, no studies have yet explored
the participation of commensal intestinal microbiota in modifying the
diet-host interaction with reference to red meat consumption.
The microbiota of humans has been linked to intestinal health,
immune function, bioactivation of nutrients and vitamins, and, more
recently, complex disease phenotypes such as obesity and insulin resist-
ance
68
. We recently reported a pathway in both humans and mice link-
ing microbiota metabolism of dietary choline and phosphatidylcholine
to CVD pathogenesis
9
. Choline, a trimethylamine-containing com-
pound and part of the head group of phosphatidylcholine, is metabo-
lized by gut microbiota to produce an intermediate compound known
as TMA (Fig. 1a). TMA is rapidly further oxidized by hepatic flavin
monooxygenases to form TMAO, which is proatherogenic and associ-
ated with cardiovascular risks. These findings raise the possibility that
other dietary nutrients possessing a trimethylamine structure may
also generate TMAO from gut microbiota and promote accelerated
atherosclerosis. TMAO has been proposed to induce upregulation of
macrophage scavenger receptors and thereby potentially contribute
to enhanced forward cholesterol transport.
10
. Whether TMAO is
linked to the development of accelerated atherosclerosis through addi-
tional mechanisms, and which specific microbial species contribute
to TMAO formation, have not been fully clarified.
l-carnitine is an abundant nutrient in red meat and contains
a trimethylamine structure similar to that of choline (Fig. 1a).
Although dietary ingestion is a major source of l-carnitine in omni-
vores, it is also endogenously produced in mammals from lysine
and serves an essential function in transporting fatty acids into the
1
Department of Cellular & Molecular Medicine, Cleveland Clinic, Cleveland, Ohio, USA.
2
Center for Cardiovascular Diagnostics & Prevention, Cleveland Clinic,
Cleveland, Ohio, USA.
3
Department of Medicine, Division of Cardiology, David Geffen School of Medicine, University of CaliforniaLos Angeles, Los Angeles,
California, USA.
4
Department of Mathematics, Cleveland State University, Cleveland, Ohio, USA.
5
Department of Cardiovascular Medicine, Cleveland Clinic,
Cleveland, Ohio, USA.
6
Department of Microbiology, Center for Clinical Epidemiology and Biostatistics, Perelman School of Medicine at the University of Pennsylvania,
Philadelphia, Pennsylvania, USA.
7
Division of Gastroenterology, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, Pennsylvania, USA.
8
Department of Medicine, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, Pennsylvania, USA.
9
Department of Pathology,
Section on Lipid Sciences, Wake Forest School of Medicine, Winston-Salem, North Carolina, USA.
10
Childrens Hospital Oakland Research Institute, Oakland,
California, USA. Correspondence should be addressed to S.L.H. (hazens@ccf.org).
Received 7 December 2012; accepted 27 February 2013; published online 7 April 2013; doi:10.1038/nm.3145
Intestinal microbiota metabolism of l-carnitine,
a nutrient in red meat, promotes atherosclerosis
Robert A Koeth
1,2
, Zeneng Wang
1,2
, Bruce S Levison
1,2
, Jennifer A Buffa
1,2
, Elin Org
3
, Brendan T Sheehy
1
,
Earl B Britt
1,2
, Xiaoming Fu
1,2
, Yuping Wu
4
, Lin Li
1,2
, Jonathan D Smith
1,2,5
, Joseph A DiDonato
1,2
, Jun Chen
6
,
Hongzhe Li
6
, Gary D Wu
7
, James D Lewis
6,8
, Manya Warrier
9
, J Mark Brown
9
, Ronald M Krauss
10
,
W H Wilson Tang
1,2,5
, Frederic D Bushman
5
, Aldons J Lusis
3
& Stanley L Hazen
1,2,5
Intestinal microbiota metabolism of choline and phosphatidylcholine produces trimethylamine (TMA), which is further
metabolized to a proatherogenic species, trimethylamine-N-oxide (TMAO). We demonstrate here that metabolism by intestinal
microbiota of dietary L-carnitine, a trimethylamine abundant in red meat, also produces TMAO and accelerates atherosclerosis
in mice. Omnivorous human subjects produced more TMAO than did vegans or vegetarians following ingestion of L-carnitine
through a microbiota-dependent mechanism. The presence of specific bacterial taxa in human feces was associated with
both plasma TMAO concentration and dietary status. Plasma L-carnitine levels in subjects undergoing cardiac evaluation
(n = 2,595) predicted increased risks for both prevalent cardiovascular disease (CVD) and incident major adverse cardiac events
(myocardial infarction, stroke or death), but only among subjects with concurrently high TMAO levels. Chronic dietary L-carnitine
supplementation in mice altered cecal microbial composition, markedly enhanced synthesis of TMA and TMAO, and increased
atherosclerosis, but this did not occur if intestinal microbiota was concurrently suppressed. In mice with an intact intestinal
microbiota, dietary supplementation with TMAO or either carnitine or choline reduced in vivo reverse cholesterol transport.
Intestinal microbiota may thus contribute to the well-established link between high levels of red meat consumption and CVD risk.

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
2 advance online publication nature medicine
mitochondrial compartment
10,11
. l-Carnitine ingestion and sup-
plementation in industrialized societies have markedly increased
12
.
Whether there are potential health risks associated with the rapidly
growing practice of consuming l-carnitine supplements has not
been evaluated.
Herein we examine the gut microbiotadependent metabolism of
l-carnitine to produce TMAO in both rodents and humans (omnivores
and vegans or vegetarians). Using isotope tracer studies in humans,
clinical studies to examine the effects on cardiovascular disease risk,
and animal models including germ-free mice, we demonstrate a role
for gut microbiota metabolism of l-carnitine in atherosclerosis patho-
genesis. We show that TMAO, and its dietary precursors choline and
carnitine, suppress reverse cholesterol transport (RCT) through gut
microbiotadependent mechanisms in vivo. Finally, we define micro-
bial taxa in feces of humans whose proportions are associated with
both dietary carnitine ingestion and plasma TMAO concentrations.
We also show microbial compositional changes in mice associated
with chronic carnitine ingestion and a consequent marked enhance-
ment in TMAO synthetic capacity in vivo.
RESULTS
Metabolomic studies link L-carnitine with CVD
Given the similarity in structure between l-carnitine and choline
(Fig. 1a), we hypothesized that dietary l-carnitine in humans, like
choline and phosphatidylcholine, might be metabolized to pro-
duce TMA and TMAO in a gut microbiotadependent fashion
and be associated with atherosclerosis risk. To test this hypothesis,
we initially examined data from our recently published unbiased
small-molecule metabolomics analyses of plasma analytes and
CVD risks
9
.
An analyte with identical molecular weight and retention time to
l-carnitine was not in the top tier of analytes that met the stringent
P value cutoff for association with CVD. However, a hypothesis-driven
examination of the data using less stringent criteria (no adjustment for
multiple testing) revealed an analyte with the appropriate molecular
weight and retention time for l-carnitine that was associated with car-
diovascular event risk (P = 0.04) (Supplementary Table 1). In further
studies we were able to confirm the identity of the plasma analyte as
l-carnitine and develop a quantitative stable-isotope-dilution liquid
chromatography tandem mass spectrometry (LC-MS/MS) method for
measuring endogenous l-carnitine concentrations in all subsequent
investigations (Supplementary Figs. 13).
Human gut microbiota is required to form TMAO from L-carnitine
The participation of gut microbiota in TMAO production from
dietary l-carnitine in humans has not previously been shown. In
initial subjects (omnivores), we developed an l-carnitine challenge
test in which the subjects were fed a large amount of l-carnitine
(an 8-ounce sirloin steak, corresponding to an estimated 180 mg
of l-carnitine)
1315
, together with a capsule containing 250 mg
of a heavy isotopelabeled l-carnitine (synthetic d3-(methyl)-l-
carnitine). At visit 1 post-prandial increases in plasma d3-TMAO
and d3- l-carnitine concentrations were readily detected, and 24-h
urine collections also revealed the presence of d3-TMAO (Fig. 1be
and Supplementary Figs. 4 and 5). Figure 1 and Supplementary
Figure 4 show tracings from a representative omnivorous subject, of
five studied with sequential serial blood draws after carnitine chal-
lenge. In most subjects examined, despite clear increases in plasma
d3-carnitine and d3-TMAO concentrations over time (Fig. 1e), post-
prandial changes in endogenous (unlabeled) carnitine and TMAO
concentrations were modest (Supplementary Fig. 5), consistent with
total body pools of carnitine and TMAO that are relatively very large
in relation to the amounts of carnitine ingested and TMAO produced
from the carnitine challenge.
To examine the potential contribution of gut microbiota to
TMAO formation from dietary l-carnitine, we placed the five
volunteers studied above on oral broad-spectrum antibiotics
to suppress intestinal microbiota for a week and then performed
a second l-carnitine challenge (visit 2). We noted near complete
suppression of detectable endogenous TMAO in both plasma
and urine after a week-long treatment with the antibiotics (visit 2)
a
Atherosclerosis
Carnitine
Gut flora
Choline
TMAO TMA
FMO
b
Visit 1
Steak
+
d3-carnitine
Visit 2
Steak
+
d3-carnitine
Visit 3
Steak
+
d3-carnitine
e
P
l
a
s
m
a

(

M
)
d3-TMAO
d3-carnitine
2.50
1.25
0
0 12 24
Time (h)
d3-TMAO
d3-carnitine
2.50
1.25
0
0 12 24
Time (h)
d3-TMAO
d3-carnitine
12.00
2.50
1.25
0
0 12 24
Time (h)
c
I
n
t
e
n
s
i
t
y

(
%
)
100
50
0
0 5 10
TMAO
m/z =
76 58
100
50
0
0 5 10
TMAO
m/z =
76 58
100
50
0
0 5 10
TMAO
m/z =
76 58
d
I
n
t
e
n
s
i
t
y

(
%
)
100
50
0
0 5 10
d3-TMAO
m/z =
79 61
Time (min)
100
50
0
0 5 10
d3-TMAO
m/z =
79 61
Time (min)
100
50
0
0 5 10
d3-TMAO
m/z =
79 61
Time (min)
Gut flora
suppression of gut flora
Reacquisition
Figure 1 TMAO production from l-carnitine is a microbiota-dependent
process in humans. (a) Structure of carnitine and scheme of carnitine
and choline metabolism to TMAO. l-Carnitine and choline are both
dietary trimethylamines that can be metabolized by microbiota to TMA.
TMA is then further oxidized to TMAO by flavin monooxygenases (FMOs).
(b) Scheme of the human l-carnitine challenge test. After a 12-h
overnight fast, subjects received a capsule of d3-(methyl)-carnitine
(250 mg) alone, or in some cases (as in data for the subject shown)
also an 8-ounce steak (estimated 180 mg l-carnitine), whereupon serial
plasma and 24-h urine samples were obtained for TMA and TMAO
analyses (visit 1). After a weeklong regimen of oral broad-spectrum
antibiotics to suppress the intestinal microbiota, the challenge was
repeated (visit 2), and then again a final third time after a 3-week period
to permit repopulation of intestinal microbiota (visit 3). (c,d) LC-MS/MS
chromatograms of plasma TMAO (c) and d3-TMAO (d) in an omnivorous
subject using specific precursor product ion transitions indicated at
t = 8 h for each visit. (e) Stable-isotope-dilution LC-MS/MS time course
measurements of d3-labeled TMAO and carnitine in plasma collected from
sequential venous blood draws at the indicated time points. Data shown in
ce are from a representative female omnivorous subject who underwent
carnitine challenge. Data are organized vertically to correspond with the
visit schedule indicated in b.

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
nature medicine advance online publication 3
(Fig. 1be and Supplementary Fig. 5). Moreover, we observed vir-
tually no detectable formation of either native or d3-labeled TMAO
in all post-prandial plasma samples or 24-h urine samples examined
after carnitine challenge, consistent with an obligatory role for gut
microbiota in TMAO formation from l-carnitine (Fig. 1be and
Supplementary Fig. 4). In contrast, we detected both d3- l-carnitine
and unlabeled l-carnitine after the l-carnitine challenge, and there
was little change in the overall time course before (visit 1) versus
after (visit 2) antibiotic treatment (Fig. 1e and Supplementary
Fig. 5). We rechallenged the same subjects several weeks after
discontinuation of antibiotics (visit 3). Baseline and post-l-carnitine
challenge plasma and urine samples again showed TMAO and
d3-TMAO formation, consistent with intestinal recolonization
(Fig. 1be and Supplementary Figs. 4 and 5). Collectively, these
data show that TMAO production from dietary l-carnitine in
humans is dependent on intestinal microbiota.
Vegans and vegetarians produce less TMAO from L-carnitine
The capacity to produce TMAO (native and d3-labeled) after l-carnitine
ingestion was variable among individuals. A post hoc nutritional
survey that the volunteers completed suggested that antecedent
dietary habits (red meat consumption) may influence the capacity
to generate TMAO from l-carnitine (data not shown). To test this
prospectively, we examined TMAO and d3-TMAO production after
the same l-carnitine challenge, first in a long-term (>5 years) vegan
who consented to the carnitine challenge (including both steak and
d3-(methyl)-carnitine consumption) (Fig. 2a). Also shown for
comparison are data from a single representative omnivore with
self-reported frequent (near daily) dietary consumption of red meat
(beef, venison, lamb, mutton, duck or pork). Post-prandially, the
omnivore showed increases in TMAO and d3-TMAO concentra-
tions in both sequential plasma measurements (Fig. 2a) and in a
24-h urine collection sample (Fig. 2b). In contrast, the vegan showed
nominal plasma and urine TMAO levels at baseline, and virtually no
capacity to generate TMAO or d3-TMAO in plasma after the carnitine
challenge (Fig. 2a,b). The vegan subject also had lower fasting
plasma levels of l-carnitine compared to the omnivorous subject
(Supplementary Fig. 6).
To confirm and extend these findings, we examined additional
vegans and vegetarians (n = 23) and omnivorous subjects (n = 51).
Figure 2 The formation of TMAO from
ingested l-carnitine is negligible in vegans,
and fecal microbiota composition associates
with plasma TMAO concentrations. (a,b) Data
from a male vegan subject in the carnitine
challenge consisting of co-administration
of 250 mg d3-(methyl)-carnitine and an
8-ounce sirloin steak and, for comparison,
a representative female omnivore who
frequently consumes red meat. Plasma
TMAO and d3-TMAO were quantified after
l-carnitine challenge (a) and in a 24-h urine
collection (b). Urine TMAO and d3-TMAO
reported as ratio with urinary creatinine
(Cr) to adjust for urinary dilution. Data are
expressed as means s.e.m. (c) Baseline
fasting plasma concentrations of TMAO and
d3-TMAO from male and female vegans and
vegetarians (n = 26) and omnivores (n = 51).
Boxes represent the 25th, 50th, and 75th
percentiles and whiskers represent the 10th
and 90th percentiles. (d) Plasma d3-TMAO
concentrations in male and female vegans
and vegetarians (n = 5) and omnivores
(n = 5) participating in a d3-(methyl)-
carnitine (250 mg) challenge without
concomitant steak consumption. The P value
shown is for the comparison of the area under
the curve (AUC) of groups using the Wilcoxon
nonparametric test. Data points represent
mean s.e.m. of n = 5 per group. (e) Baseline
TMAO plasma concentrations associate with
enterotype 2 (Prevotella) in male and female
subjects with a characterized gut microbiome
enterotype. Boxes represent the 25th,
50th (middle lines) and 75th percentiles,
and whiskers represent the 10th and 90th
percentiles. (f) Plasma TMAO concentrations
(plotted on x axes) and the proportion of
taxonomic operational units (OTUs, plotted
on y axes), determined as described in
Supplementary Methods. Subjects were
grouped by dietary status as either vegan
or vegetarian (n = 23) or omnivore (n = 30). P value shown is for comparisons between dietary groups using a robust Hotelling T
2
test. Data are
expressed as means s.e.m. for both TMAO concentration (x axis) and the proportion of OTUs (y axis).
a
6
4
0
0 12 24
P
l
a
s
m
a

(

M
)
TMAO
Omnivore
Vegan
Time (h)
0.250
0.125
0
0 12 24
P
l
a
s
m
a

(

M
)
d3-TMAO
Omnivore
Vegan
Time (h)
b
50
25
0
Vegan Omnivore
Urine TMAO
T
M
A
O
/
C
r
(
m
m
o
l
/
m
o
l
)
2
1
0
Vegan Omnivore
Urine d3-TMAO
d
3
-
T
M
A
O
/
C
r
(
m
m
o
l
/
m
o
l
)
c
8
4
0
P
l
a
s
m
a

T
M
A
O

(

M
)
Omnivore
(n = 51)
Vegan/
vegetarian
(n = 26)
P < 0.05
d
30
15
0
0 12 24
P
l
a
s
m
a

d
3
-
T
M
A
O

(

M
)
P < 0.05
O
m
nivore
(n = 5)
Vegan/vegetarian
(n = 5)
Time (h)
P
r
o
p
o
r
t
i
o
n

O
T
U
s

(

1
0

4
)
Clostridiales
incertae sedis XII
4
2
0
1.8 2.7 3.6
TMAO (M)
P = 0.13
Fusibacterium
4
2
0
1.8 2.7 3.6
TMAO (M)
P = 0.13
Lachnospira
50
25
0
1.8 2.7 3.6
TMAO (M)
P < 0.05
Sporobacter
24
12
0
1.8 2.7 3.6
TMAO (M)
P = 0.10
f
40
20
0
P
r
o
p
o
r
t
i
o
n

O
T
U
s

(

1
0

4
)
Omnivore
(n = 30)
Peptostreptococcaceae
incertae sedis
Vegan/
vegetarian
(n = 23)
1.8 2.7 3.6
TMAO (M)
P < 0.05
Peptostreptococcaceae
40
20
0
1.8 2.7 3.6
TMAO (M)
P < 0.05
Clostridium
30
15
0
1.8 2.7 3.6
TMAO (M)
P < 0.05
30
15
0
1.8 2.7 3.6
TMAO (M)
P < 0.05
Clostridiaceae
e
8
4
0
P
l
a
s
m
a

T
M
A
O

(

M
) P < 0.05
Enterotype 1
Bacteroides
(n = 49)
Enterotype 2
Prevotella
(n = 4)

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
4 advance online publication nature medicine
Fasting baseline TMAO levels were signifi-
cantly lower among vegan and vegetarian
subjects compared to omnivores (Fig. 2c). In
a subset of these individuals, we performed
an oral d3-(methyl)-carnitine challenge (but
with no steak) and confirmed that long-term
(all >1 year) vegans and vegetarians have a
markedly reduced capacity to synthesize TMAO from oral carnitine
(Fig. 2c,d). Vegans and vegetarians challenged with d3-(methyl)-
carnitine also had significantly higher post-challenge plasma con-
centrations of d3-(methyl)-carnitine compared to omnivorous
subjects (Supplementary Fig. 7), perhaps due to decreased intestinal
microbial metabolism of carnitine before absorption.
TMAO levels are associated with human gut microbial taxa
Dietary habits (for example, vegan or vegetarian versus omnivore
or carnivore) are associated with significant alterations in intestinal
microbiota composition
1618
. To determine microbiota composition,
we sequenced the gene encoding bacterial 16S rRNA in fecal samples
from a subset of the vegans and vegetarians (n = 23) and omnivores
(n = 30) studied above. In parallel, we quantified plasma TMAO,
carnitine and choline concentrations by stable-isotope-dilution
LC-MS/MS. Global analysis of taxa proportions (Supplementary
Methods) revealed significant associations with plasma TMAO
concentrations (P = 0.03), but not with plasma carnitine (P = 0.77)
or choline (P = 0.74) concentrations.
After false discovery rate (FDR) adjustment for multiple compari-
sons, several bacterial taxa remained significantly (FDR-adjusted
P < 0.10) associated with plasma TMAO concentration
(Supplementary Fig. 8). When we classified subjects into previously
reported enterotypes
19
on the basis of fecal microbial composition,
individuals with an enterotype characterized by enriched proportions
of the genus Prevotella (n = 4) had higher (P < 0.05) plasma TMAO
concentrations than did subjects with an enterotype notable for
enrichment in the Bacteroides (n = 49) genus (Fig. 2e). Examination of
the proportion of specific bacterial genera and subject plasma TMAO
concentrations revealed several taxa (at the genus level) that simulta-
neously were significantly associated with both vegan or vegetarian
versus omnivore status and plasma TMAO concentration (Fig. 2f).
These results indicate that preceding dietary habits may modulate
both intestinal microbiota composition and ability to synthesize TMA
and TMAO from dietary l-carnitine.
TMAO production from dietary L-carnitine is inducible
We next investigated the ability of chronic dietary l-carnitine intake
to induce gut microbiotadependent production of TMA and TMAO
in mice. Initial LC-MS/MS studies in germ-free mice showed no
detectable plasma d3-(methyl)-TMA or d3-(methyl)-TMAO after
oral (gastric gavage) d3-(methyl)-carnitine challenge. However, after
a several-week period in conventional cages to allow for microbial
colonization (conventionalization), previously germ-free mice
acquired the capacity to produce both d3-(methyl)-TMA and
d3-(methyl)-TMAO following oral d3-(methyl)-carnitine challenge
c
1
Taxonomy TMAO
*
*
*
*
**
**
**
**
*
**
**
*
*
*
Actinobacteria
Bacteroidetes
TMA
Bacteroidetes Bacteroidia
Bacteroidetes Bacteroidia Bacteroidales
Bacteroidetes Bacteroidia Bacteroidales Bacteroidaceae
Actinobacteria Actinobacteri
Actinobacteria Actinobacteria Bifidobacteriales
Actinobacteria Actinobacteria Bifidobacteriales Bifidobacteriaceae
Actinobacteria Actinobacteria Bifidobacteriales Bifidobacteriaceae Bifidobacterium
Bacteroidetes Bacteroidia Bacteroidales Bacteroidaceae Bacteroides
Bacteroidetes Bacteroidia Bacteroidales Unclassified
Bacteroidetes Bacteroidia Bacteroidales Unclassified Unclassified
Bacteroidetes Bacteroidia Bacteroidales Porphyromonadaceae
Bacteroidetes Bacteroidia Bacteroidales Porphyromonadaceae Barnesiella
Bacteroidetes Bacteroidia Bacteroidales Porphyromonadaceae Odoribacter
Bacteroidetes Bacteroidia Bacteroidales Porphyromonadaceae Unclassified
Bacteroidetes Bacteroidia Bacteroidales Porphyromonadaceae Parabacteroides
Bacteroidetes Bacteroidia Bacteroidales Prevotellaceae
Bacteroidetes Bacteroidia Bacteroidales Prevotellaceae Unclassified
Bacteroidetes Bacteroidia Bacteroidales Prevotellaceae Prevotella
Bacteroidetes Bacteroidia Bacteroidales Rikenellaceae
Bacteroidetes Bacteroidia Bacteroidales Rikenellaceae Alistipes
Deferribacteres
Deferribacteres Deferribacteres
Deferribacteres Deferribacteres Deferribacterales
Deferribacteres Deferribacteres Deferribacterales Deferribacteraceae
Deferribacteres Deferribacteres Deferribacterales Deferribacteraceae Mucispirillum
Firmicutes
Firmicutes Bacilli
Firmicutes Bacilli Lactobacillales
Firmicutes Bacilli Lactobacillales Lactobacillaceae
Firmicutes Bacilli Lactobacillales Lactobacillaceae Lactobacillus
Firmicutes Clostridia
Firmicutes Clostridia Clostridiales
Firmicutes Clostridia Clostridiales Lachnospiraceae
Firmicutes Clostridia Clostridiales Lachnospiraceae Dorea
Firmicutes Clostridia Clostridiales Lachnospiraceae Unclassified
Firmicutes Clostridia Clostridiales Unclassified
Firmicutes Clostridia Clostridiales Unclassified Unclassified
Firmicutes Clostridia Clostridiales Ruminococcaceae
Firmicutes Clostridia Clostridiales Ruminococcaceae Butyricicoccus
Firmicutes Clostridia Clostridiales Ruminococcaceae Oscillibacter
Firmicutes Clostridia Clostridiales Ruminococcaceae Unclassified
Firmicutes Clostridia Clostridiales Ruminococcaceae Ruminococcus
Firmicutes Erysipelotrichi
Firmicutes Erysipelotrichi Erysipelotrichales
Firmicutes Erysipelotrichi Erysipelotrichales Erysipelotrichaceae
Firmicutes Erysipelotrichi Erysipelotrichales Erysipelotrichaceae Allobaculum
Firmicutes Erysipelotrichi Erysipelotrichales Erysipelotrichaceae Unclassified
Firmicutes Erysipelotrichi Erysipelotrichales Erysipelotrichaceae Turicibacter
Proteobacteria Alphaproteobacteria
Proteobacteria
Proteobacteria Betaproteobacteria
Proteobacteria Betaproteobacteria Burkholderiales
Proteobacteria Betaproteobacteria Burkholderiales Alcaligenaceae
Proteobacteria Betaproteobacteria Burkholderiales Alcaligenaceae Parasutterella
Proteobacteria Deltaproteobacteria
Proteobacteria Deltaproteobacteria Desulfovibrionales
Proteobacteria Deltaproteobacteria Desulfovibrionales Desulfovibrionaceae
Proteobacteria Deltaproteobacteria Desulfovibrionales Desulfovibrionaceae Desulfovibrio
Proteobacteria Deltaproteobacteria Desulfovibrionales Unclassified
Proteobacteria Deltaproteobacteria Desulfovibrionales Unclassified Unclassified
Proteobacteria Epsilonproteobacteria
Proteobacteria Epsilonproteobacteria Campylobacterales
Proteobacteria Epsilonproteobacteria Campylobacterales Helicobacteraceae
Proteobacteria Epsilonproteobacteria Campylobacterales Helicobacteraceae Helicobacter
Tenericutes
Tenericutes Mollicutes
Tenericutes Mollicutes Anaeroplasmatales
Tenericutes Mollicutes Anaeroplasmatales Anaeroplasmataceae
Tenericutes Mollicutes Anaeroplasmatales Anaeroplasmataceae Anaeroplasma
Verrucomicrobia
Verrucomicrobia Verrucomicrobiae
Verrucomicrobia Verrucomicrobiae Verrucomicrobiales
Verrucomicrobia Verrucomicrobiae Verrucomicrobiales Verrucomicrobiaceae
Verrucomicrobia Verrucomicrobiae Verrucomicrobiales Verrucomicrobiaceae Akkermansia
*
FDR-adjusted P value 0.1
**
FDR-adjusted P value 0.1
b
a
Mice
30
0
d3-TMA
P
l
a
s
m
a

(

M
)
Carnitine
diet
Chow
Time (h)
15
0
6 12
500
0
d3-TMAO
Carnitine
diet
Chow
Time (h)
250
0
6 12
90
d3-carnitine
Carnitine diet
Chow
Time (h)
0 6 12
45
0
0
5.0
Carnitine
(n = 11)
Chow
(n = 10)
P < 0.01
TMAO (M)
P
r
o
p
o
r
t
i
o
n

O
T
U
s

(

1
0

2
)
Anaeroplasma
2.5
0
65 130
50
P < 0.01
TMAO (M)
Porphyromonadaceae
P
r
o
p
o
r
t
i
o
n

O
T
U
s

(

1
0

2
)
0 65 130
25
0
0
3
P < 0.05
TMA (M)
Prevotella
P
r
o
p
o
r
t
i
o
n

O
T
U
s

(

1
0

2
)
2
1
0
35 70
0
3
P < 0.05
TMA (M)
Prevotellaceae,
Unclassified
P
r
o
p
o
r
t
i
o
n

O
T
U
s

(

1
0

2
)
2
1
0
35 70
0.5
0
0.5
1
Figure 3 The metabolism of carnitine to
TMAO is an inducible trait and associates
with microbiota composition. (a) d3-carnitine
challenge of mice on either an l-carnitine
supplemented diet (1.3%) for 10 weeks and
compared to age-matched normal chowfed
controls. Plasma d3-TMA and d3-TMAO
were measured at the indicated times after
d3-(methyl)-carnitine administration by oral
gavage using stable-isotope-dilution LC-
MS/MS. Data points represent mean s.e.m.
of n = 4 per group. (b) Correlation heat map
demonstrating the association between the
indicated microbiota taxonomic genera and
TMA and TMAO concentrations (all reported
as mean s.e.m. in M) of mice grouped by
dietary status (chow, n = 10 (TMA, 1.3 0.4;
TMAO, 17 1.9); and l-carnitine, n = 11 (TMA,
50 16; TMAO, 114 16). Red denotes a
positive association, blue a negative association,
and white no association. A single asterisk
indicates a significant FDR-adjusted association
of P 0.1, and a double asterisk indicates a
significant FDR-adjusted association of
P 0.01. (c) Plasma TMAO and TMA
concentrations determined by stable-isotope-
dilution LC-MS/MS (plotted on x axes) and the
proportion OTUs (plotted on y axes). Statistical
and laboratory analyses were performed as
described in Supplementary Methods. Data are
expressed as means s.e.m. for both TMAO or
TMA concentrations (x axis) and the proportion
of OTUs (y axis).

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
nature medicine advance online publication S
(Supplementary Fig. 9). Parallel studies with non-germ-free (con-
ventional) Apoe
/
mice (lacking apolipoprotein E; on a C57BL/6J
background) that had been placed on a cocktail of oral, relatively
nonabsorbable broad-spectrum antibiotics previously shown to
suppress intestinal microbiota
9,20
showed similar results (complete
suppression of both TMA and TMAO formation; Supplementary
Fig. 10). Collectively, these studies confirm in mice an obligatory
role for gut microbiota in TMA and TMAO production from
dietary l-carnitine.
To examine whether dietary l-carnitine can induce TMA and
TMAO production from intestinal microbiota, we compared the
pre- and post-prandial plasma profiles of Apoe
/
mice on normal
chow diet versus a normal chow diet supplemented with l-carnitine
for 15 weeks. The production of both d3-(methyl)TMA and
d3-(methyl)TMAO after gastric gavage of d3-(methyl)-carnitine
was induced by approximately tenfold in mice on the l-carnitine
supplemented diet (Fig. 3a). Furthermore, plasma post-prandial
d3-(methyl)-carnitine levels in mice in the l-carnitinesupplemented
diet arm were substantially lower than those observed in mice on
the l-carnitinefree diet (normal chow), consistent with enhanced
microbiota-dependent catabolism before absorption in the
l-carnitinesupplemented mice.
Plasma TMA and TMAO associate with mouse gut microbial taxa
The marked effects of an acute l-carnitine challenge (d3-(methyl)-
carnitine by gavage) on TMA and TMAO production suggested that
chronic l-carnitine supplementation may significantly alter intesti-
nal microbial composition, with an enrichment for taxa better suited
for TMA production from l-carnitine. To test this hypothesis, we
first identified the cecum as the segment of the entire intestinal tract
of mice showing the highest synthetic capacity to form TMA from
carnitine (data not shown). We then sequenced 16S rRNA gene
amplicons from ceca of mice on either normal chow (n = 10) or
l-carnitine-supplemented (n = 11) diets and in parallel quantified
plasma concentrations of TMA and TMAO (Fig. 3b). Global analyses
of the presence of the microbiota taxa revealed that, in general, taxa that
were at a relatively high proportion coincident with high TMA plasma
concentrations also tended to be a relatively high proportion coinci-
dent with high TMAO plasma concentrations. Several bacterial taxa
remained significantly associated with plasma TMA and/or TMAO
levels after FDR adjustment for multiple comparisons (Fig. 3b).
Further analyses revealed several bacterial taxa whose proportion was
significantly associated (some positively, others inversely) with dietary
l-carnitine and with plasma TMA or TMAO concentrations (P < 0.05)
(Fig. 3c and Supplementary Fig. 11). Notably, a direct comparison of
taxa associated with plasma TMAO concentrations in humans versus in
mice failed to identify common taxa. These results are consistent with
prior reports that microbes identified from the distal gut of the mouse
represent genera that are typically not detected in humans
16,21
.
High plasma L-carnitine concentration is associated with CVD
We next investigated the relationship of fasting plasma concentra-
tions of l-carnitine with CVD risk in an large, independent cohort
of stable subjects (n = 2,595) undergoing elective cardiac evaluation.
Patient demographics, laboratory values and clinical characteris-
tics are provided in Supplementary Table 2. We observed signifi-
cant dose-dependent associations between carnitine concentration
and risks of prevalent coronary artery disease (CAD) (P < 0.05),
peripheral artery disease (PAD) (P < 0.05) and overall CVD (P <
0.05) (Fig. 4ac). Moreover, these associations remained significant
following adjustments for traditional CVD risk factors (P < 0.05)
(Fig. 4ac). Plasma concentrations of carnitine were high in sub-
jects with angiographic evidence of CAD (50% stenosis), regardless
of the extent (for example, single- versus multivessel) of CAD, as
PAD
b
0
Odds ratio
1 2 3 4
CVD
c
Odds ratio
0 1 2 3 4
a
Carnitine CAD
(M)
Q1 <31.6
Q2
Q3
Q4
31.737.8
37.945.1
>45.1
0 1 2 3 4
Odds ratio
C
a
r
n
i
t
i
n
e

(

m
)
d
50
P < 0.001
N
o
n
e
40
30
20
S
in
g
le
D
o
u
b
le
T
r
ip
le
Coronary vessel disease
Carnitine MACE (3-year)
Q1 <31.6
e
Q2
Q3
Q4
31.737.8
37.945.1
>45.1
Hazard ratio
0.5
(M)
1.0 0.5 2.0
f
Time (years)
1
100
E
v
e
n
t
-
f
r
e
e
s
u
r
v
i
v
a
l

(
%
)
P < 0.001
Carnitine TMAO
Unadjusted
HR (95%)
Adjusted
HR (95%)
90
80
2 3
High
Low
Low
High High
Low
Low
High
0.9 (0.61.4)
1.0 (reference)
1.6 (1.22.0)
2.5 (1.83.4)
0.8 (0.51.3)
1.0 (reference)
1.3 (1.021.7)
2.1 (1.52.8)
Figure 4 Relationship between plasma carnitine concentration and
CVD risks. (ac) Forrest plots of the odds ratio of CAD (a), PAD (b) and
CVD (c) and quartiles of carnitine before (closed circles) and after
(open circles) logistic regression adjustments with traditional
cardiovascular risk factors, including age, sex, history of diabetes
mellitus, smoking, systolic blood pressure, LDL cholesterol and HDL
cholesterol. Bars represent 95% confidence intervals. (d) Relationship
of fasting plasma carnitine concentrations and angiographic evidence
of CAD. Boxes represent the 25th, 50th and 75th percentiles of
plasma carnitine concentration, and whiskers represent the 10th and
90th percentiles. The Kruskal-Wallis test was used to assess the degree of CAD (none, single-, double- or triple-vessel disease) association with
plasma carnitine concentrations. (e) Forrest plot of the hazard ratio of MACE and quartiles of carnitine unadjusted (closed circles) and after adjusting
for traditional cardiovascular risk factors (open circles), or traditional cardiac risk factors plus creatinine clearance, history of myocardial infarction,
history of CAD, burden of CAD (one-, two- or three-vessel disease), left ventricular ejection fraction, baseline medications (angiotensin-converting
enzyme (ACE) inhibitors, statins, beta blockers and aspirin) and TMAO levels (open squares). Bars represent 95% confidence intervals. (f) Kaplan-
Meier plot and hazard ratios with 95% confidence intervals for unadjusted model, or following adjustments for traditional risk factors as in e. Median
plasma concentration of carnitine (46.8 M) and TMAO (4.6 M) within the cohort were used to stratify subjects as having high (median) or low
(<median) values.

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
6 advance online publication nature medicine
revealed by diagnostic cardiac catheterization
(Kruskal-Wallis P < 0.001) (Fig. 4d).
We also examined the relationship
between fasting plasma concentrations
of carnitine and incident (3-year) risk
for major adverse cardiac events (MACE:
composite of death, myocardial infarction,
stroke and revascularization). Elevated
carnitine (4th quartile) concentration was
an independent predictor of MACE, even
after adjustments for traditional CVD risk
factors (Fig. 4e). After further adjustment
for both plasma TMAO concentration
and a larger number of comorbidities that
might be known at time of presentation (for
example, extent of CAD, ejection fraction,
medications and estimated renal function), the significant relation-
ship between carnitine and MACE risk was completely abolished
(Fig. 4e). Notably, we observed a significant association between
carnitine concentration and incident cardiovascular event risks
in Cox regression models after multivariate adjustment, but only
among those subjects with concurrent high plasma TMAO concen-
trations (P < 0.001) (Fig. 4f). Thus, although plasma concentra-
tions of carnitine seem to be associated with both prevalent and
incident cardiovascular risks, these results suggest that TMAO,
rather than carnitine, is the primary driver of the association
of carnitine with cardiovascular risks.
Dietary L-carnitine promotes microbiota-dependent atherosclerosis
We next investigated whether dietary l-carnitine has an impact on
the extent of atherosclerosis in the presence or absence of TMAO
formation. We fed Apoe
/
mice from the time of weaning a normal
chow diet versus the same diet supplemented with l-carnitine. Aortic
root atherosclerotic plaque quantification revealed approximately
a doubling of disease burden in l-carnitine supplemented mice
compared to normal chowfed mice (Fig. 5a,b). Parallel studies
in mice placed on an oral antibiotic cocktail to suppress intestinal
microbiota showed marked reductions in plasma TMA and TMAO
concentrations (Fig. 5c) and complete inhibition of the dietary
l-carnitinedependent increase in atherosclerosis (Fig. 5b). Of note,
the increase in atherosclerotic plaque burden with dietary l-carnitine
occurred in the absence of proatherogenic changes in plasma lipid,
lipoprotein, glucose or insulin levels; moreover, both biochemical
and histological analyses of livers from any group of the mice failed
to show evidence of steatosis (Supplementary Tables 3 and 4 and
Supplementary Fig. 12).
Plasma concentrations of carnitine were significantly higher
in l-carnitinefed mice compared to normal chowfed controls
(P < 0.05) (Fig. 5c). Plasma carnitine concentrations were even
higher in mice supplemented with l-carnitine in the antibiotic
arm of the study (Fig. 5c), presumably as a result of the reduced
capacity of microbiota to catabolize l-carnitine. However, as the
a
Chow
250 m 250 m
A
o
r
t
i
c

l
e
s
i
o
n

(

m
2
)
T
M
A
O

(

M
)
T
M
A

(

M
)
C
a
r
n
i
t
i
n
e

(

M
)
250 m 250 m
Chow + ABS
Carnitine
1.8-fold
5.0 10
5
2.5 10
5
Chow
(n = 9) (n = 11) (n = 9) (n = 10)
0
200 120 200
2 2 1.5
1.5
1.5
28% 31%
35%
P < 0.01
P < 0.01 P < 0.01 P < 0.01
P < 0.01
P < 0.01
P < 0.05
P < 0.05
P < 0.05
P < 0.05 P < 0.05 P < 0.01 P < 0.05 P = 0.77 P = 0.97
P < 0.05
Cyp27a1 Cyp7a1
Oatp1 Oatp4 Bsep Mrp2 Ephx1 Ntcp
P = 0.39
P = 0.34 P < 0.05
P = 0.22
P < 0.01
P < 0.05
P < 0.01
P < 0.01
P < 0.01
P = 0.31
P =0.84
P < 0.01
P = 0.89
Carnitine + ABS
c
b
d e
f
Carnitine Chow
+ ABS
Carnitine
+ ABS
Chow
(n = 9) (n = 11) (n = 9) (n = 11)
Carnitine Chow
+ ABS
Carnitine
+ ABS
Chow
(n = 9) (n = 11) (n = 9) (n = 11)
Carnitine Chow
+ ABS
Carnitine
+ ABS
Chow
Chow
Stool Stool Liver
Liver
+ ABS stool
R
C
T

(
%
)
R
e
l
a
t
i
v
e

e
x
p
r
e
s
s
i
o
n
R
e
l
a
t
i
v
e

e
x
p
r
e
s
s
i
o
n
(n = 9)
(n = 43) (n = 30) (n = 12) (n = 16) (n = 27) (n = 21) (n = 21) (n = 29)
(n = 11) (n = 9) (n = 11)
Carnitine
Carnitine Choline Chow Chow TMAO
(n = 16) (n = 16)
Chow TMAO
(n = 15) (n = 16)
Chow TMAO
(n = 16) (n = 16)
Chow TMAO
(n = 16) (n = 16)
Chow TMAO
(n = 16) (n = 16)
Chow TMAO
(n = 16) (n = 16)
Chow TMAO
(n = 16) (n = 16)
Chow TMAO
(n = 16) (n = 16)
Chow TMAO
Carnitine Choline
Chow
+ ABS
Carnitine
+ ABS
100
0
60
0
100
0
1
0
1
0
1.0
0.5
0
1.0
0.5
0
1.0
0.5
0
Figure 5 Dietary l-carnitine accelerates
atherosclerosis and inhibits reverse cholesterol
transport in a microbiota dependent fashion.
(a) Representative oil red Ostained aortic roots
(counterstained with hematoxylin) of 19-week-old
Apoe
/
female mice on the indicated diets
in the presence versus absence of antibiotics
(ABS) as described in the Online Methods.
(b) Quantification of mouse aortic root plaque
lesion area. Apoe
/
female mice at 19 weeks
of age were started on the indicated diets at the
time of weaning (4 weeks of age) before killing,
and lesion area was quantified as described in
the Online Methods. (c) Carnitine, TMA and
TMAO concentrations as determined using
stable-isotope-dilution LC-MS/MS analysis of
plasma recovered from mice at the time of
killing. (d) RCT (72-h stool collection) in adult
female (>8 weeks of age) Apoe
/
mice on normal
chow versus diet supplemented with either
l-carnitine or choline, as well as after suppression
of microbiota using cocktail of antibiotics
(+ ABS). Also shown are RCT (72-h stool
collection) results in adult female (>8 weeks of
age) Apoe
/
mice on normal chow versus diet
supplemented with TMAO. (e,f) Relative mRNA
levels (to Actb) of mouse liver candidate genes
involved in bile acid synthesis or transport.
Data are expressed as means s.e.m.

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
nature medicine advance online publication 7
l-carnitinesupplemented mice that received
antibiotics did not show enhanced athero-
sclerosis, these results are consistent with the
notion that it is a downstream microbiota-
dependent metabolite, not l-carnitine itself,
that promotes atherosclerosis.
TMAO inhibits RCT
To identify additional mechanisms by which
TMAO might promote atherosclerosis, we
first noted that TMAO and its trimethylamine
nutrient precursors are all cationic quater-
nary amines that could potentially compete
with arginine, thereby limiting its bioavail-
ability and reducing nitric oxide synthesis.
However, a direct test of this hypothesis with
competition studies using [
14
C]arginine and
TMAO in bovine aortic endothelial cells
demonstrated no decrease in [
14
C]arginine
transport (Supplementary Fig. 13).
In recent studies we showed that TMAO can promote macrophage
cholesterol accumulation in a microbiota-dependent manner by
increasing cell surface expression of two proatherogenic scavenger
receptors, CD36 and scavenger receptor A (SRA)
9,22,23
. We envi-
sioned three non-exclusive mechanisms through which cholesterol
can accumulate within cells of the artery wall: enhancing the rate
of influx (as noted above), enhancing synthesis or diminishing the
rate of efflux. To test whether TMAO might alter the canonical regu-
lation of cholesterol biosynthesis genes
24
, we loaded macrophages
with cholesterol in the presence or absence of physiologically rel-
evant TMAO concentrations. However, TMAO failed to alter mRNA
levels of the low-density lipoprotein (LDL) receptor or cholesterol
synthesis genes (Supplementary Fig. 14). Parallel studies examin-
ing macrophage inflammatory gene expression
25
and desmosterol
levels in the culture medium also failed to show any effect of TMAO
(Supplementary Figs. 14 and 15).
We next examined whether TMAO might inhibit cholesterol
removal from peripheral macrophages by testing whether dietary
sources of TMAO (choline or l-carnitine) inhibit RCT in vivo
using an adaptation of an established model system
26
. Mice on
either choline (1.3% choline chloride by mass)- or l-carnitine
supplemented diets showed significantly less (~30%, P < 0.05) RCT
compared to normal chowfed controls (Fig. 5d). Furthermore,
suppression of intestinal microbiota (and plasma TMAO concen-
trations) with oral broad-spectrum antibiotics completely blocked
the diet-dependent (for both choline and l-carnitine) suppression
of RCT (Fig. 5d), suggesting that a microbiota-generated product
(for example, TMAO) inhibits RCT (Supplementary Fig. 16). To
further test this hypothesis, we placed mice on a TMAO-containing
diet. They showed a 35% decrease in RCT relative to mice on a nor-
mal chow diet (Fig. 5d, P < 0.05). Further examination of plasma,
liver and bile showed significantly less [
14
C]cholesterol recovered
from plasma of TMAO-fed compared to chow-fed mice (16% lower,
P < 0.05) but no changes in counts recovered from liver or bile
(Supplementary Fig. 17).
TMAO alters sterol metabolism in vivo
To better understand the molecular mechanisms through which
TMAO suppresses RCT, we examined candidate genes and biological
processes in compartments (macrophages, plasma, liver and intes-
tine) known to participate in cholesterol and sterol metabolism and
RCT. We exposed peritoneal macrophages recovered from wild-type
C57BL/6J mice to TMAO in vitro and quantified mRNA levels of the
cholesterol transporters Abca1, Srb1 and Abcg1. TMAO treatment
led to modest but statistically significant increases in expression of
Abca1 and Abcg1 (P < 0.05; Supplementary Fig. 18). Parallel stud-
ies showed corresponding modest TMAO-dependent increases in
Abca1-dependent cholesterol efflux to apoA1 as cholesterol acceptor in
RAW 264.7 macrophages (P < 0.01; Supplementary Fig. 19). Collectively,
these results suggest that the observed global reduction in RCT
in vivo induced by TMAO is unlikely to be accounted for by changes
in the expression of these transporters. Parallel examination of plasma
recovered from mice in the RCT experiments showed no differences
in total cholesterol and high-density lipoprotein cholesterol concen-
trations (Supplementary Table 5).
P < 0.01
26%
Atherosclerosis
Forward cholesterol transport
Reverse cholesterol transport
Heart attack
TMAO
FMOs
TMA
Bile acid
pool size
TMA
Carnitine
Choline
Cholesterol
absorption
Net reverse
cholesterol
transport
NPC1L1
ASBT
ABCG5/8
Enterocyte
Cholesterol
absorption
ABCA1
OST- OST-
OATPs
EPHX1 NTCP
SRB1
ABCA1
ABCG1
BSEP
MRP2
CYP27A1
Bile acid
pool
CYP7A1
Cholesterol
ABCG5/8
Hepatocyte
SRA
SRB1
Macrophage
Foam cell
CD36
( ABCG1)
( ABCA1)
Stroke
Death
Revascularization
Chow
(n = 11)
TMAO
(n = 10)
b
c
C
h
o
l
e
s
t
e
r
o
l
a
b
s
o
r
p
t
i
o
n

(
%
)
70
60
50
40
G
u
t flora
a
100
P < 0.01
26%
Chow
(n = 11)
TMAO
(n = 10)
*
*
B
i
l
e

a
c
i
d

p
o
o
l

s
i
z
e
(

m
o
l

p
e
r

1
0
0

g
b
o
d
y

w
e
i
g
h
t
)
50
0
Taurodeoxycholate
Tauroursodeoxycholate
Tauro--muricholate
Taurocholate
Figure 6 Effect of TMAO on cholesterol and
sterol metabolism. (a,b) Measurement of total
bile acid pool size and composition (a) and
cholesterol absorption (b) in adult female
(>8 weeks of age) Apoe
/
mice on normal chow
diet versus diet supplemented with TMAO for
4 weeks. Data are expressed as means s.e.m.
(c) Summary scheme outlining the proposed
pathway by which microbiota participate in
atherosclerosis. The microbiota metabolizes
dietary l-carnitine and choline to form TMA
and TMAO. TMAO affects cholesterol and sterol
metabolism in macrophages, liver and intestine.

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
8 advance online publication nature medicine
In parallel studies, we examined the mRNA levels of known cho-
lesterol transporters (Srb1, Abca1, Abcg1, Abcg5, Abcg8 and Shp)
in mouse liver but found only a modest difference for Srb1 expres-
sion (Supplementary Fig. 20). Western blot analysis of liver from
TMAO-supplemented mice, however, showed no change in the
abundance of Srb1 protein compared to chow (control) mouse livers
(Supplementary Fig. 21). In contrast, mRNA levels in the liver of
the key bile acid synthetic enzymes Cyp7a1 and Cyp27a1 were sig-
nificantly lower in mice supplemented with dietary TMAO, with no
change in expression of the upstream regulator Shp (P < 0.05 for each;
Fig. 5e and Supplementary Fig. 20). Multiple bile acid transporters
in the liver (Oatp1, Oatp4, Mrp2, and Ntcp) also showed significant
dietary TMAOinduced decreases in expression (P < 0.05 each);
however, Bsep and Ephx1 did not (Fig. 5f). In contrast to the liver,
TMAO-induced changes in bile acid transporter gene expression were
not observed in the gut (Supplementary Fig. 22). Taken together,
these data show that the gut microbiotadependent metabolite TMAO
affects a major pathway for cholesterol elimination from the body, the
bile acid synthetic pathway, at multiple levels.
Consistent with the effects of TMAO on bile acid transporter
gene expression, mice supplemented with TMAO had a significantly
smaller total bile acid pool size compared to normal chowfed mice
(P < 0.01) (Fig. 6a). Dietary supplementation with TMAO also mark-
edly lowered mRNA expression of both types of intestinal cholesterol
transporters: Npc1L1, which transports cholesterol into enterocyte
from the gut lumen
27
, and Abcg5-Abcg8, which transport cholesterol
out of enterocytes into the gut lumen
27
(Supplementary Fig. 23).
Previous studies using either Cyp7a1- or Cyp27a1-null mice demon-
strated a reduction in cholesterol absorption
28,29
. In separate studies,
dietary TMAO supplementation compared to normal chow similarly
promoted a decrease (26% reduced compared to normal chowfed
mice, P < 0.01) in total cholesterol absorption (Fig. 6b).
DISCUSSION
The dietary nutrient l-carnitine has been studied for over a cen-
tury
30
. Although eukaryotes can endogenously produce l-carnitine,
only prokaryotic organisms can catabolize it
11
. A role for intestinal
microbiota in TMAO production from dietary l-carnitine was first
suggested by studies in rats
31
. Although TMAO production from
alternative dietary trimethylamines has been suggested in humans,
a role for the microbiota in the production of TMAO from dietary
l-carnitine in humans has not previously been demonstrated
3133
.
The present studies reveal an obligatory role of gut microbiota in the
production of TMAO from ingested l-carnitine in humans. They
also suggest a new nutritional pathway in CVD pathogenesis that
involves dietary l-carnitine, the intestinal microbial community and
production of the proatherosclerotic metabolite TMAO. Finally, these
studies show that TMAO modulates cholesterol and sterol metabolism
at multiple anatomic sites and processes in vivo, with a net effect of
increasing atherosclerosis.
Our results also suggest a previously unknown mechanism for
the observed relationship between dietary red meat ingestion and
accelerated atherosclerosis. Consuming foods rich in l-carnitine
(predominantly red meat) can increase fasting human l-carnitine
concentrations in the plasma
34
. Meats and full-fat dairy products
are abundant components of the Western diet and are commonly
implicated in CVD. Together, l-carnitine and choline-containing
lipids can constitute up to 2% of a Western diet
14,15,35
. Numerous
studies have suggested a decrease in atherosclerotic disease risk in
vegan and vegetarian individuals compared to omnivores; reduced
levels of dietary cholesterol and saturated fat have been suggested as
the mechanism explaining this decreased risk
36,37
. Notably, a recent
4.8-year randomized dietary study showed a 30% reduction in car-
diovascular events in subjects consuming a Mediterranean diet (with
specific avoidance of red meat) compared to subjects consuming a
control diet
38
. The present studies suggest that the reduced inges-
tion of l-carnitine and total choline by vegans and vegetarians,
with attendant reductions in TMAO levels, may contribute to their
observed cardiovascular health benefits. Conversely, an increased
capacity for microbiota-dependent production of TMAO from
l-carnitine may contribute to atherosclerosis, particularly in omni-
vores who consume high amounts of l-carnitine.
One proatherosclerotic mechanism observed for TMAO in the cur-
rent studies is suppression of RCT (Fig. 6c). Dietary l-carnitine and
choline each suppressed RCT (P < 0.05), but only in mice with intact
intestinal microbiota and increased TMA and TMAO concentrations.
Suppression of the intestinal microbiota completely eliminated choline-
and l-carnitine-dependent suppression of RCT. Moreover, direct
dietary supplementation with TMAO promoted a similar suppression
of RCT. These results are consistent with a gut microbiotadependent
mechanism whereby generation of TMAO impairs RCT, potentially
contributing to the observed proatherosclerotic phenotype of these
interventions. Another mechanism by which TMAO may promote
atherosclerosis is through increasing macrophage SRA and CD36
surface expression and foam cell formation
9
(Fig. 6c). Within macro-
phages, TMAO does not seem to alter known cholesterol biosynthetic
and uptake pathways
24,39
or the more recently described regulatory
role of desmosterol in integrating macrophage lipid metabolism and
inflammatory gene responses
25
. In the liver, TMAO decreased the bile
acid pool size and lowered the expression of key bile acid synthesis
and transport proteins (Fig. 6c). However, it is unclear whether these
changes contribute to the impairment of RCT. Of note, TMAO lowered
expression of Cyp7a1, the major bile acid synthetic enzyme and rate-
limiting step in the catabolism of cholesterol. The effect of TMAO is
thus consistent with reports of human Cyp7a1 gene variants that are
associated with reduced expression of Cyp7a1, leading to decreased bile
acid synthesis, decreased bile acid secretion and enhanced atheroscle-
rosis
4042
. Furthermore, upregulation (as opposed to downregulation)
of Cyp7a1 has been reported to lead to expansion of the bile acid pool,
increased RCT and reduced atherosclerotic plaque area in susceptible
mice
4345
. Within the intestine, we found that TMAO concentration
was also associated with changes in cholesterol metabolism. However,
the reduction in cholesterol absorption observed, although consistent
with the reduction in intestinal Npc1L1 expression
46
(as well as hepatic
Cyp7a1 and Cyp27a1 expression
28,29
), cannot explain the suppression
of RCT observed after dietary supplementation with TMAO.
Thus, the molecular mechanisms through which gut microbiota
formation of TMAO leads to inhibition of RCT are not entirely clear.
It is also not known whether TMAO interacts directly with a specific
receptor or whether it acts to alter signaling pathways indirectly by
altering protein conformation (that is, via allosteric effects). Whereas
TMA has been reported to influence signal transduction by direct
interaction with a family of G proteincoupled receptors
47,48
, TMAO,
a small quaternary amine with aliphatic character, is reportedly capa-
ble of directly inducing conformational changes in proteins, stabilizing
protein folding and acting as a small-molecule protein chaperone
49,50
.
It is thus conceivable that TMAO may alter many signaling pathways
without directly acting at a TMAO receptor.
A noteworthy finding is the magnitude with which long-term dietary
habits affect TMAO synthetic capacity in both humans (vegans and

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
nature medicine advance online publication 9
vegetarians versus omnivores) and mice (normal chow versus chronic
l-carnitine supplementation). Analyses of microbial composition in
human feces and mice cecal contents revealed specific taxa that seg-
regate with both dietary status and plasma TMAO concentrations.
Recent studies have shown that changes in enterotype are associ-
ated with long-term dietary patterns
19
. We observed that plasma
TMAO concentration varied significantly (P < 0.05) according
to previously reported enterotypes. We also showed an obligatory role
for gut microbiota in TMAO formation from dietary l-carnitine in
mice and humans. The differences observed in TMAO production
after an l-carnitine challenge in omnivore versus vegan subjects is
striking, and is consistent with the observed differences in micro-
bial community composition. Recent reports have shown differences
in microbial communities among vegetarians and vegans versus
omnivores
51
. Of note, we observed an increase in baseline plasma
TMAO concentrations in what has historically been called entero-
type 2 (Prevotella), a relatively rare enterotype that in one study was
associated with low animal-fat and protein consumption
19
. In our
study, three of the four individuals classified into enterotype 2 are
self-identified omnivores, suggesting more complexity in the human
gut microbiome than anticipated. Indeed, other studies have demon-
strated variable results in associating human bacterial genera, includ-
ing Bacteroides and Prevotella, to omnivorous and vegetarian eating
habits
18,52
. This complexity is no doubt in part attributable to the fact
that there are many species within any genus, and distinct species
within the same genus may have different capacities to use l-carnitine
as a fuel and form TMA. Indeed, prior studies have suggested that mul-
tiple bacterial strains can metabolize l-carnitine in culture
53
, and spe-
cies within the genus Clostridium differ in their ability to use choline as
the sole source of carbon and nitrogen in culture
54
. Our results suggest
that multiple proatherogenic (that is, TMA- and TMAO-producing)
species probably exist. Consistent with this supposition, others have
reported that several bacterial phylotypes are associated with a history
of atherosclerosis and that human microbiota biodiversity may in part
be influenced by carnivorous eating habits
16,19,55
.
The association between plasma carnitine concentrations and
cardiovascular risks further supports the potential pathophysiologi-
cal importance of a carnitine gut microbiota TMA/TMAO
atherosclerosis pathway (Fig. 6c). The association between high
plasma carnitine concentration and CVD risk disappeared after
TMAO levels were added to the statistical model. These observations
are consistent with a proposed mechanism whereby oral l-carnitine
ingestion contributes to atherosclerotic CVD risk via the microbiota
metabolite TMAO. There are only a few reports of specific intestinal
anaerobic and aerobic bacterial species that can use l-carnitine as a
carbon nitrogen source
10,11,56
.
l-carnitine is essential for the import of activated long-chain fatty
acids from the cytoplasm into mitochondria for -oxidation, and dietary
supplementation with l-carnitine has been widely studied. Some case
reports of l-carnitine supplementation have reported beneficial effects
in individuals with inherited primary and acquired secondary carnitine
deficiency syndromes
13
. l-Carnitine supplementation studies in chronic
disease states have reported both positive and negative results
57,58
. Oral
l-carnitine supplementation in subjects on hemodialysis raises plasma
l-carnitine concentrations to normal levels but also substantially
increases TMAO levels
57
. A broader potential therapeutic scope for
l-carnitine and two related metabolites, acetyl-l-carnitine and
propionyl-l-carnitine, has also been explored for the treatment of
acute ischemic events and cardiometabolic disorders (reviewed in
ref. 58). Here too, both positive and negative results have been reported.
Potential explanations for the discrepant findings of various l-carnitine
intervention studies are differences in the duration of dosing or in the
route of administration. In many studies, l-carnitine or a related mol-
ecule is administered over a short interval or via the parenteral route,
thereby bypassing gut microbiota (and hence TMAO formation).
Discovery of a link between l-carnitine ingestion, gut microbiota
metabolism and CVD risk has broad health-related implications. Our
studies reveal a new pathway potentially linking dietary red meat inges-
tion with atherosclerosis pathogenesis. The role of gut microbiota in this
pathway suggests new potential therapeutic targets for preventing CVD.
Furthermore, our studies have public health relevance, as l-carnitine
is a common over-the-counter dietary supplement. Our results suggest
that the safety of chronic l-carnitine supplementation should be exam-
ined, as high amounts of orally ingested l-carnitine may under some
conditions foster growth of gut microbiota with an enhanced capacity
to produce TMAO and potentially advance atherosclerosis.
METHODS
Methods and any associated references are available in the online
version of the paper.
Note: Supplementary information is available in the online version of the paper.
ACKnOWLEDGMEnTS
We thank L. Kerchenski and C. Stevenson for assistance in performing the clinical
studies; A. Pratt, S. Neale, M. Pepoy and B. Sullivan for technical assistance with
human specimen processing and routine clinical diagnostic testing; E. Klipfell,
F. McNally and M. Berk for technical assistance; and the subjects who consented to
participate in these studies. Mass spectrometry instrumentation used was housed
within the Cleveland Clinic Mass Spectrometry Facility with partial support
through a Center of Innovation by AB SCIEX. Germ-free mice were obtained from
the University of North Carolina Gnotobiotic Facility, which is supported by
P30-DK034987-25-28 and P40-RR018603-06-08. This research was supported by
US National Institutes of Health grants R01 HL103866 (S.L.H.), P20 HL113452 (S.L.H.
and W.H.W.T.), PO1 HL30568 (A.J.L.), PO1 H28481 (A.J.L.), R00 HL096166 (J.M.B.),
UH3-DK083981 (J.D.L.), 1RC1DK086472 (R.M.K.) and the Leducq Foundation
(S.L.H.). The clinical study GeneBank was supported in part by P01 HL076491, P01
HL098055, R01 HL103931 and the Cleveland Clinic Foundation General Clinical
Research Center of the Cleveland Clinic/Case Western Reserve University Clinical
and Translational Science Award (1UL1RR024989). S.L.H. is also partially supported
by a gift from the Leonard Krieger Fund. Z.W. was partially supported by a Scientist
Development Grant from the American Heart Association. E.O. was supported by a
MOBILITAS Postdoctoral Research Grant (MJD252). R.A.K. was supported in part
by US National Institutes of Health grant T32 GM007250.
AUTHOR COnTRIBUTIOnS
R.A.K. participated in laboratory, mouse and human studies, assisted in statistical
analyses, helped design the experiments and drafted the manuscript. Z.W.
performed the initial metabolomics study and assisted with mouse and mass
spectrometry analyses. B.S.L. synthesized d3- and d9-carnitine for studies, assisted
with mass spectrometry analyses and helped draft the manuscript. E.B.B. and
X.F. assisted in performance of mass spectrometry analyses of the large human
clinical cohort study. Y.W. and L.L. performed the statistical analyses and critically
reviewed the manuscript. J.D.S. helped with aortic root atherosclerosis analyses and
critical review of the manuscript. J.A.D. assisted in experimental design. J.A.B. and
B.T.S. assisted in laboratory and mouse experiments. E.O. and A.J.L. performed and
helped interpret mouse cecal microbiota analyses. J.C., F.D.B., H.L., G.D.W., J.D.L.
and R.M.K. assisted in human subject microbiota analyses and helped interpret
human microbiota data. M.W. and J.M.B. assisted with measurement of bile acid
pool size and helped with critical review of the manuscript. W.H.W.T. helped with
human studies and critical review of the manuscript. S.L.H. conceived of the idea,
helped design the experiments, provided the funding for the study and helped draft
and critically revise the manuscript.
COMPETInG FInAnCIAL InTERESTS
The authors declare competing financial interests: details are available in the online
version of the paper.
Reprints and permissions information is available online at http://www.nature.com/
reprints/index.html.

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
arti cl e s
10 advance online publication nature medicine
1. Bernstein, A.M. et al. Major dietary protein sources and risk of coronary heart
disease in women. Circulation 122, 876883 (2010).
2. Micha, R., Wallace, S.K. & Mozaffarian, D. Red and processed meat consumption
and risk of incident coronary heart disease, stroke, and diabetes mellitus:
a systematic review and meta-analysis. Circulation 121, 22712283 (2010).
3. Siri-Tarino, P.W., Sun, Q., Hu, F.B. & Krauss, R.M. Meta-analysis of prospective
cohort studies evaluating the association of saturated fat with cardiovascular
disease. Am. J. Clin. Nutr. 91, 535546 (2010).
4. Bibbins-Domingo, K. et al. Projected effect of dietary salt reductions on future
cardiovascular disease. N. Engl. J. Med. 362, 590599 (2010).
5. Hansen, E.S. International Commission for Protection Against Environmental
Mutagens and Carcinogens. ICPEMC Working Paper 7/1/2. Shared risk factors for
cancer and atherosclerosisa review of the epidemiological evidence. Mutat. Res.
239, 163179 (1990).
6. Turnbaugh, P.J. et al. A core gut microbiome in obese and lean twins. Nature 457,
480484 (2009).
7. Turnbaugh, P.J. et al. An obesity-associated gut microbiome with increased capacity
for energy harvest. Nature 444, 10271031 (2006).
8. Goodman, A.L. & Gordon, J.I. Our unindicted coconspirators: human metabolism
from a microbial perspective. Cell Metab. 12, 111116 (2010).
9. Wang, Z. et al. Gut fora metabolism of phosphatidylcholine promotes cardiovascular
disease. Nature 472, 5763 (2011).
10. Bremer, J. Carnitinemetabolism and functions. Physiol. Rev. 63, 14201480
(1983).
11. Rebouche, C.J. & Seim, H. Carnitine metabolism and its regulation in microorganisms
and mammals. Annu. Rev. Nutr. 18, 3961 (1998).
12. Brass, E.P. Carnitine and sports medicine: Use or abuse? Ann. NY Acad. Sci. 1033,
6778 (2004).
13. Stanley, C.A. Carnitine defciency disorders in children. Ann. NY Acad. Sci. 1033,
4251 (2004).
14. Demarquoy, J. et al. Radioisotopic determination of l-carnitine content in foods
commonly eaten in western countries. Food Chem. 86, 137142 (2004).
15. Rigault, C., Mazue, F., Bernard, A., Demarquoy, J. & Le Borgne, F. Changes in
l-carnitine content of fsh and meat during domestic cooking. Meat Sci. 78,
331335 (2008).
16. Ley, R.E. et al. Evolution of mammals and their gut microbes. Science 320,
16471651 (2008).
17. Muegge, B.D. et al. Diet drives convergence in gut microbiome functions across
mammalian phylogeny and within humans. Science 332, 970974 (2011).
18. Zimmer, J. et al. A vegan or vegetarian diet substantially alters the human colonic
faecal microbiota. Eur. J. Clin. Nutr. 66, 5360 (2012).
19. Wu, G.D. et al. Linking long-term dietary patterns with gut microbial enterotypes.
Science 334, 105108 (2011).
20. Rakoff-Nahoum, S., Paglino, J., Eslami-Varzaneh, F., Edberg, S. & Medzhitov, R.
Recognition of commensal microfora by Toll-like receptors is required for intestinal
homeostasis. Cell 118, 229241 (2004).
21. Ley, R.E. et al. Obesity alters gut microbial ecology. Proc. Natl. Acad. Sci. USA
102, 1107011075 (2005).
22. Febbraio, M. et al. Targeted disruption of the class B scavenger receptor CD36
protects against atherosclerotic lesion development in mice. J. Clin. Invest. 105,
10491056 (2000).
23. Suzuki, H. et al. A role for macrophage scavenger receptors in atherosclerosis and
susceptibility to infection. Nature 386, 292296 (1997).
24. Brown, M.S. & Goldstein, J.L. The SREBP pathway: regulation of cholesterol
metabolism by proteolysis of a membrane-bound transcription factor. Cell 89,
331340 (1997).
25. Spann, N.J. et al. Regulated accumulation of desmosterol integrates macrophage
lipid metabolism and infammatory responses. Cell 151, 138152 (2012).
26. Rader, D.J. Regulation of reverse cholesterol transport and clinical implications.
Am. J. Cardiol. 92, 42J49J (2003).
27. Jia, L., Betters, J.L. & Yu, L. Niemann-pick C1-like 1 (NPC1L1) protein in intestinal
and hepatic cholesterol transport. Annu. Rev. Physiol. 73, 239259 (2011).
28. Schwarz, M., Russell, D.W., Dietschy, J.M. & Turley, S.D. Marked reduction in bile
acid synthesis in cholesterol 7-hydroxylase-defcient mice does not lead to
diminished tissue cholesterol turnover or to hypercholesterolemia. J. Lipid Res. 39,
18331843 (1998).
29. Repa, J.J. et al. Disruption of the sterol 27-hydroxylase gene in mice results in
hepatomegaly and hypertriglyceridemia. Reversal by cholic acid feeding. J. Biol.
Chem. 275, 3968539692 (2000).
30. Gulewitsch, W. & Krimberg, R. Zur Kenntnis der Extrakivstoffe der Muskein, II. Mitteilung.
Uber das Carnitin. Hoppe-Seylers Z. Physiol. Chem. 45, 326330 (1905).
31. Rebouche, C.J., Mack, D.L. & Edmonson, P.F. l-Carnitine dissimilation in the
gastrointestinal tract of the rat. Biochemistry 23, 64226426 (1984).
32. Rebouche, C.J. & Chenard, C.A. Metabolic fate of dietary carnitine in human adults:
identifcation and quantifcation of urinary and fecal metabolites. J. Nutr. 121,
539546 (1991).
33. Zhang, A.Q., Mitchell, S.C. & Smith, R.L. Dietary precursors of trimethylamine in
man: a pilot study. Food Chem. Toxicol. 37, 515520 (1999).
34. Delany, J.P., Snook, J.T., Vivian, V.M. & Cashmere, K. Metabolic effects of a
carnitine-free diet fed to college students. Fed. Proc. 45, 815 (1986).
35. Zeisel, S.H., Mar, M.H., Howe, J.C. & Holden, J.M. Concentrations of choline-
containing compounds and betaine in common foods. J. Nutr. 133, 13021307
(2003).
36. Fraser, G.E. Vegetarian diets: what do we know of their effects on common chronic
diseases? Am. J. Clin. Nutr. 89, 1607S1612S (2009).
37. Key, T.J. et al. Mortality in vegetarians and nonvegetarians: detailed fndings from
a collaborative analysis of 5 prospective studies. Am. J. Clin. Nutr. 70, 516S524S
(1999).
38. Estruch, R. et al. Primary prevention of cardiovascular disease with a Mediterranean
diet. N. Engl. J. Med. published online, http://dx.doi.org/10.1056/NEJMoa1200303
(25 February 2013).
39. Brown, M.S. & Goldstein, J.L. Expression of the familial hypercholesterolemia gene
in heterozygotes: mechanism for a dominant disorder in man. Science 185, 6163
(1974).
40. Charach, G., Rabinovich, A., Argov, O., Weintraub, M. & Rabinovich, P. The role of
bile Acid excretion in atherosclerotic coronary artery disease. Int. J. Vasc. Med.
2012, 949672 (2012).
41. Charach, G. et al. Decreased fecal bile acid output in patients with coronary
atherosclerosis. J. Med. 29, 125136 (1998).
42. Lu, Y., Feskens, E.J., Boer, J.M. & Muller, M. The potential infuence of genetic
variants in genes along bile acid and bile metabolic pathway on blood cholesterol
levels in the population. Atherosclerosis 210, 1427 (2010).
43. Miyake, J.H. et al. Transgenic expression of cholesterol-7--hydroxylase prevents
atherosclerosis in C57BL/6J mice. Arterioscler. Thromb. Vasc. Biol. 22, 121126
(2002).
44. Post, S.M., de Crom, R., van Haperen, R., van Tol, A. & Princen, H.M. Increased
fecal bile acid excretion in transgenic mice with elevated expression of human
phospholipid transfer protein. Arterioscler. Thromb. Vasc. Biol. 23, 892897
(2003).
45. Zong, C. et al. Chitosan oligosaccharides promote reverse cholesterol transport and
expression of scavenger receptor BI and CYP7A1 in mice. Exp. Biol. Med. (Maywood)
237, 194200 (2012).
46. Altmann, S.W. et al. Niemann-Pick C1 like 1 protein is critical for intestinal
cholesterol absorption. Science 303, 12011204 (2004).
47. Liberles, S.D. & Buck, L.B. A second class of chemosensory receptors in the
olfactory epithelium. Nature 442, 645650 (2006).
48. Suska, A., Ibanez, A.B., Lundstrom, I. & Berghard, A. G proteincoupled
receptor mediated trimethylamine sensing. Biosens. Bioelectron. 25, 715720
(2009).
49. Bai, C., Biwersi, J., Verkman, A.S. & Matthay, M.A. A mouse model to test the
in vivo effcacy of chemical chaperones. J. Pharmacol. Toxicol. Methods 40, 3945
(1998).
50. Mello, C.C. & Barrick, D. Measuring the stability of partly folded proteins using
TMAO. Protein Sci. 12, 15221529 (2003).
51. Cordain, L. et al. Origins and evolution of the Western diet: health implications for
the 21st century. Am. J. Clin. Nutr. 81, 341354 (2005).
52. Liszt, K. et al. Characterization of bacteria, clostridia and Bacteroides in faeces of
vegetarians using qPCR and PCR-DGGE fngerprinting. Ann. Nutr. Metab. 54,
253257 (2009).
53. Elssner, T., Preusser, A., Wagner, U. & Kleber, H.P. Metabolism of l-carnitine by
Enterobacteriaceae under aerobic conditions. FEMS Microbiol. Lett. 174, 295301
(1999).
54. Mller, B., Hippe, H. & Gottschalk, G. Degradation of various amine compounds
by mesophilic clostridia. Arch. Microbiol. 145, 8590 (1986).
55. Eckburg, P.B. et al. Diversity of the human intestinal microbial fora. Science 308,
16351638 (2005).
56. Kleber, H.P. Bacterial carnitine metabolism. FEMS Microbiol. Lett. 147, 19 (1997).
57. Hedayati, S.S. Dialysis-related carnitine disorder. Semin. Dial. 19, 323328
(2006).
58. Mingorance, C., Rodriguez-Rodriguez, R., Justo, M.L., Alvarez de Sotomayor, M. &
Herrera, M.D. Critical update for the clinical use of l-carnitine analogs in
cardiometabolic disorders. Vasc. Health Risk Manag. 7, 169176 (2011).

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
nature medicine doi:10.1038/nm.3145
ONLINE METHODS
Mice and general procedures. Breeders of all conventional mice (C57BL/6J
and Apoe
/
mice on a C57BL/6J background) were obtained from Jackson
Laboratories. All animal studies were performed under approval of the Animal
Research Committee of the Cleveland Clinic. Liver cholesterol was quanti-
fied by gas chromatography MS, and liver triglyceride was measured using
glycerol phosphate oxidase reagent as described in Supplementary Methods.
Mouse plasma lipids and glucose and human fasting lipid profile, C-reactive
protein (CRP) and glucose were measured using the Abbott ARCHITECT
platform, Model ci8200 (Abbott Diagnostics). Mouse HDL was isolated using
density ultracentrifugation, and insulin levels were quantified by enzyme-
linked immunosorbent assay as described in Supplementary Methods.
Human plasma myeloperoxidase was measured using US Food and Drug
Administrationcleared CardioMPO assay (Cleveland Heart Lab).
Research subjects. All research subjects gave written informed consent. All
protocols were approved by the Cleveland Clinic Institutional Review Board.
Two cohorts of subjects were used in the present studies. The first group of
volunteers (n = 30 omnivores and n = 23 vegetarians or vegans) had extensive
dietary questioning and stool, plasma and urine collection. A subset of subjects
with stool collected also underwent oral l-carnitine challenge (n = 5 omnivores
and n = 5 vegans), consisting of d3(methyl)-carnitine (250 mg within a veggie
capsule (Wonder Laboratories)). Where indicated, additional omnivores and
one vegan also underwent l-carnitine challenge testing with combined inges-
tion of the synthetic d3-(methyl)- l-carnitine capsule (250 mg) and an 8-ounce
beef steak (consumed within 10 min). Male and female volunteers were at least
18 years of age. Volunteers participating in the l-carnitine challenge tests were
excluded if they were pregnant, had chronic illness (including a known history
of heart failure, renal failure, pulmonary disease, gastrointestinal disorders or
hematologic diseases), had an active infection, had received antibiotics within
2 months of study enrollment, had used any over-the-counter or prescriptive
probiotic or bowel cleansing preparation within the past 2 months, had ingested
yogurt within the past 7 d, or had undergone bariatric or other intestinal (for
example, gallbladder removal, bowel resection) surgery. All other research sub-
jects were derived from GeneBank, a large longitudinal tissue repository with
connecting clinical database from sequential consenting stable subjects under-
going elective cardiac evaluation. Further description of the GeneBank cohort
can be found in Supplementary Methods.
Human l-carnitine challenge test. Consenting adult men and women fasted
overnight (12 h) before performing the l-carnitine challenge test, which involved
baseline blood and spot urine collection, and then oral ingestion (t = 0 at time of
initial ingestion) of a veggie capsule (size O) (Wonder Laboratories) containing
250 mg of a stable isotopelabeled d3-l-(methyl)-carnitine (under an
Investigational New Drug exemption from the US Food and Drug Administration).
Where indicated, for a subset of subjects, the l-carnitine challenge also included a
natural source of l-carnitine (a cooked 8-ounce sirloin steak) eaten over a 10-min
period concurrent with taking the capsule containing the d3-l-(methyl)-
carnitine. After combined ingestion of the steak and d3-l-(methyl)-carnitine,
a series of sequential venous blood draws were performed at the indicated time
points, and a 24-h urine collection was performed. An ensuing 1-week treatment
period of oral antibiotics (metronidazole 500 mg and ciprofloxacin 500 mg twice
daily) was given to suppress intestinal microbiota that use carnitine to form TMA
and TMAO; the l-carnitine challenge was then repeated. After at least 3 weeks off
of all antibiotics to allow reacquisition of intestinal microbiota, a third and final
l-carnitine challenge test was performed. Dietary habits (vegan versus ominivore)
were determined using a questionnaire assessment of dietary l-carnitine intake,
similar to that conducted by the Atherosclerotic Risk in Community study
59
. d3-
l-(methyl)-carnitine was prepared by taking sodium l-norcarnitine dissolved in
methanol and reacting it with d3-methyl iodide (Cambridge Isotope) in the pres-
ence of potassium hydrogen carbonate to give d3- l-(methyl)-carnitine. Further
details regarding d3- l-(methyl)-carnitine synthesis, purification and characteri-
zation are described in Supplementary Methods.
Metabolomics study. We previously reported results from a metabolomics study
where small-molecule analytes were sought that associated with cardiovascular
risks
9
. The metabolomics study had a two-stage screening strategy. In the first
phase, unbiased metabolomics studies were performed on randomly selected
plasma samples from a learning cohort generated from Genebank subjects who
had experienced a major adverse cardiovascular event (defined as nonfatal
myocardial infarction, stroke or death) (n = 50) in the 3-year period following
enrollment versus age- and gender-matched controls (n = 50) who had not
experienced an event. A second phase (validation cohort) of unbiased metabo-
lomics analyses was then performed on a nonoverlapping second cohort of cases
(n = 25) and age- and gender-matched controls (n = 25) using identical inclu-
sion and exclusion criteria. Further details regarding the unbiased metabolomic
approach can be found in Supplementary Methods.
Identification of l-carnitine and quantification of TMAO, TMA and l-carnitine.
Matching collision-induced dissociation (CID) spectra of an unknown plasma
metabolite with identical retention time and mass-to-charge ratio (m/z) as
authentic l-carnitine (m/z = 162) were obtained as described in Supplementary
Methods. Concentrations of carnitine, TMA and TMAO isotopologues in
mouse and human plasma samples were determined by stable-isotope-dilution
LC-MS/MS in positive multiple reaction monitoring (MRM) mode using deuter-
ated internal standards on an AB Sciex API 5000 triple quadrupole mass spec-
trometer (Applied Biosystems) as described in Supplementary Methods. In
studies quantifying endogenous carnitine and ingested d3-l-(methyl)-carnitine,
d9-carnitine was used as internal standard. d9-carnitine was prepared by dissolv-
ing 3-hydroxy-4-aminobutyric acid (Chem-Impex Intl.) in methanol and exhaus-
tive reaction with d3-methyl iodide (Cambridge Isotope Labs) in the presence of
potassium hydrogen carbonate. Further details regarding synthesis, purification
and characterization of d9-carnitine can be found in Supplementary Methods.
Human microbiota analyses. Stool samples were stored at 80 C, and DNA
for the gene encoding 16S rRNA was isolated using the MoBio PowerSoil kit
according to the manufacturers instructions. DNA samples were amplified
using V1-V2 region primers targeting bacterial 16S genes and sequenced using
454/Roche Titanium technology. Sequence reads from this study are available
from the Sequence Read Archive (controlled feeding experiment: SRX037803,
SRX021237, SRX021236, SRX020772, SRX020771, SRX020588, SRX020587,
SRX020379, SRX020378 (metagenomic); cross-sectional study of diet and stool
microbiome: SRX020773, SRX020770). The overall association between TMAO
measurements and microbiome compositions was assessed using PermanovaG
60

by combining both the weighted and unweighted UniFrac distances. Associations
between TMAO measurements and individual taxa proportions were assessed by
Spearmans rank correlation test. False discovery rate (FDR) control based on the
Benjamini-Hochberg procedure was used to account for multiple comparisons
when evaluating these associations. Each of the samples was assigned to an
enterotype category on the basis of their microbiome distances (Jensen-Shannon
distance) to the medoids of the enterotype clusters as defined in the COMBO
data
19
. Association between enterotypes and plasma TMAO concentration
was assessed by Wilcoxon rank-sum test. Students t-test was used to test the
difference in means of TMAO concentration between omnivores and vegans.
A robust Hotelling T
2
test was used to examine the association between both
the proportion of specific bacterial taxa and TMAO concentrations in groups
using R software version 2.15 (ref. 61).
Mouse microbiota analysis. Microbial community composition was assessed
by pyrosequencing 16S rRNA genes derived from ceca of mice fed a normal
chow (n = 11) or l-carnitine (n = 13) diet. DNA was isolated using the MoBio
PowerSoil DNA Isolation Kit. The V4 region of the 16S ribosomal DNA gene
was amplified using bar-coded fusion primers (F515/R806) with the 454 A
Titanium sequencing adaptor as further described in Supplementary Methods.
The relative abundances of bacteria at each taxonomic level were computed for
each mouse, a single representative sequence for each OTU was aligned using
PyNAST and a phylogenetic tree was built using FastTree as further described
in Supplementary Methods. Spearman correlations were calculated to assess
correlations between relative abundance of gut microbiota and mouse plasma
TMA and TMAO concentrations. False discovery rates (FDR) of the multi-
ple comparisons were estimated for each taxon based on the P values resulted
from correlation estimates, as further described in Supplementary Methods.

2
0
1
3

N
a
t
u
r
e

A
m
e
r
i
c
a
,

I
n
c
.


A
l
l

r
i
g
h
t
s

r
e
s
e
r
v
e
d
.
nature medicine doi:10.1038/nm.3145
A robust Hotelling T
2
test was used to examine the association between both
the proportion of specific bacterial taxa and mouse plasma TMA and TMAO
concentrations in groups using R software version 2.15 (ref. 61).
Aortic root lesion quantification. Apolipoprotein Eknockout mice on a
C57BL/6J background (Apoe
/
) were weaned at 28 d of age and placed on a
standard chow control diet (Teklad 2018). l-Carnitine was introduced into the
diet by supplementing mouse drinking water with 1.3% l-carnitine (Chem-
Impex Intl.), 1.3% l-carnitine and antibiotics, or antibiotics alone. The antibi-
otic cocktail dissolved in mouse drinking water has previously been shown to
suppress commensal gut microbiota and included 0.1% ampicillin sodium salt
(Fisher Scientific), 0.1% metronidazole, 0.05% vancomycin (Chem Impex Intl.)
and 0.1% neomycin sulfate (Gibco)
20
. Mice were anaesthetized with ketamine
and xylazine before terminal bleeding by cardiac puncture to collect blood.
Mouse hearts were fixed and stored in 10% neutral-buffered formalin before
being frozen in optimal cutting temperature medium for sectioning. Aortic root
slides were stained with oil red O and counterstained with hematoxylin. The
aortic root atherosclerotic lesion area was quantified as the mean of sequential
sections of 6 microns approximately 100 microns apart
9
.
Germ-free mice and conventionalization studies. 10-week-old female Swiss
Webster germ-free mice (SWGF) were obtained from the University of North
Carolina Gnotobiotics Core Facility. Germ-free mice underwent gastric
gavage with the indicated isotopologues of l-carnitine (see below for details of
l-carnitine challenge) immediately following removal from the germ-free
microisolator shipper. After the l-carnitine challenge, germ-free mice were
conventionalized by being housed in cages with nonsterile C57BL/6J female
mice. Approximately 4 weeks later, the l-carnitine challenge was repeated.
Quantification of natural abundance and isotope-labeled l-carnitine, TMA
and TMAO in mouse plasma was performed using stable-isotope-dilution
LC/MS/MS as described above.
Mouse l-carnitine challenge studies. C57BL/6J female or Apoe
/
female mice
were given synthetic d3-l-carnitine (150 l of a 150 mM stock) dissolved in
water via gastric gavage using a 1.5-inch 20-gauge intubation needle. Plasma
was collected from the saphenous vein at baseline and at the indicated time
points. Apoe
/
female mice were used in the study examining the inducibility of
microbiota to generate TMA and TMAO following carnitine feeding. For these
studies, mice were placed on an l-carnitinesupplemented diet (1.3% l-carnitine
in drinking water) for 10 weeks. Quantification of the abundance of native and
isotope-labeled forms of carnitine, TMA and TMAO in mouse plasma was per-
formed using stable-isotope-dilution LC-MS/MS as described above.
Mouse reverse cholesterol transport, cholesterol absorption and bile acid
pool size studies. Adult female (>8 weeks of age) Apoe
/
mice were placed on
either a chow diet or an l-carnitine, choline- or TMAO-supplemented diet for
4 weeks before performance of reverse cholesterol transport, cholesterol absorp-
tion or bile acid pool size/composition studies as described in Supplementary
Methods. In some RCT experiments, mice were treated with a cocktail of oral
antibiotics (as in atherosclerosis studies described above) for 4 weeks before
enrollment. RCT studies were performed using subcutaneous (in the back)
injection of [
14
C]cholesterol-labeled bone marrowderived macrophages, as
further detailed in Supplementary Methods. Feces were collected and analyzed
as described in Supplementary Methods. For cholesterol absorption experi-
ments, mice were fasted 4 h before gavage with olive oil supplemented with
[
14
C]cholesterol and [
3
H]-sitostanol. Feces were collected over a 24-h period
and analyzed as described in Supplementary Methods. Total bile acid pool size
and composition were determined in female Apoe
/
mice, with analysis of the
combined small intestine, gallbladder, and liver, which were extracted together
in ethanol with nor-deoxycholate (Steraloids) added as an internal standard.
The extracts were filtered (Whatman paper #2), dried and resuspended in water.
The samples were then passed through a C18 column (Sigma) and eluted with
methanol. The eluted samples were again dried down and resuspended in metha-
nol. A portion of the sample was subjected to HPLC using Waters Symmetry
C18 column (4.6 250 mm no. WAT054275, Waters Corp.) and a mobile phase
consisting of methanol:acetonitrile:water (53:23:24) with 30 mM ammonium
acetate, pH 4.91, at a flow rate of 0.7 ml min
1
. Bile acids were detected by an
evaporative light spray detector (Alltech ELSD 800, nitrogen at 3 bar, drift tube
temperature 40 C) and identified by comparing their respective retention times
to those of standards (taurocholate and tauro--muricholate from Steraloids;
taurodeoxycholate and taurochenodeoxycholate from Sigma; tauroursodeoxy-
cholate from Calbiochem). For quantification, peak areas were integrated using
Chromperfect Spirit (Justice laboratory software) and bile acid pool size was
expressed as mol per 100 g body weight after correcting for procedural losses
based on the nor-deoxycholate internal standard.
Effects of TMAO on macrophage cholesterol biosynthesis, cholesterol efflux,
inflammatory genes and desmosterol levels. The effects of cholesterol loading
on the expression of macrophage cholesterol biosynthetic and inflammatory
genes, macrophage LDL receptor gene expression and macrophage desmos-
terol abundance were analyzed as previously described
25
. Thioglycollate-
elicited mouse peritoneal macrophages (MPMs) were harvested and cultured
in RPMI 1640 supplemented with 10% FCS and penicillin plus streptomycin.
MPMs were then lipoprotein-starved further in culture for 18 h in the absence
versus presence of increasing concentrations of cholesterol, acetylated LDL or
vehicle with or without 300 M TMAO dehydrate (Sigma). Desmosterol in the
cholesterol-loading studies was quantified by stable-isotope-dilution GC/MS
analysis. Further details of these studies and cholesterol efflux studies are
described in Supplementary Methods.
RNA preparation and real-time PCR analysis. RNA was purified from tissue
(macrophage, liver or gut) using the animal tissue protocol from the Qiagen
RNeasy mini kit. Small bowel used for RNA purification was sectioned sequen-
tially in five equal segments from the duodenum to illeum before RNA prepara-
tion. Purified total RNA and random primers were used to synthesize first-strand
cDNA using the High Capacity cDNA Reverse Transcription Kit (Applied
Biosystems, Foster City, CA) reverse transcription protocol. Quantitative real-
time PCR was performed using Taqman quantitative RT-PCR probes (Applied
Biosystems, Foster City, CA) and normalized to tissue -actin by the C
T

method using StepOne Software v2.1 (Applied Biosystems, Foster City, CA).
Statistical analyses. Students t-test or a Wilcoxon nonparametric test were
used to compare group means as deemed appropriate. The analysis of variance
(ANOVA, if normally distributed) or Kruskal-Wallis test (if not normally dis-
tributed) was used for multiple group comparisons of continuous variables, and a
Chi-square test was used for categorical variables. Odds ratios for various cardiac
phenotypes (CAD, PAD and CVD) and corresponding 95% confidence intervals
were calculated using logistic regression models. Kaplan-Meier analysis with Cox
proportional hazards regression was used for time-to-event analysis to determine
hazard ratio and 95% confidence intervals for adverse cardiac events (death,
myocardial infarction, stroke and revascularization). Adjustments were made
for individual traditional cardiac risk factors (age, gender, diabetes mellitus,
systolic blood pressure, former or current cigarette smoking, LDL cholesterol,
HDL cholesterol), extent of CAD, left ventricular ejection fraction, history of
myocardial infarction, baseline medications (aspirin, statins, beta blockers and
angiotensin-converting-enzyme (ACE) inhibitors) and renal function by esti-
mated creatinine clearance. Kruskal-Wallis test was used to assess the effect of the
degree of coronary vessel disease on l-carnitine levels. A robust Hotelling T
2
test
was used to examine the difference in the proportion of specific bacterial genera
along with subject TMAO levels between the different dietary groups
61
. All data
were analyzed using R software version 2.15 and Prism (Graphpad Software).
Additional methods. Detailed methodology is described in the Supplementary
Methods.
59. The ARIC investigators. The Atherosclerosis Risk in Communities (ARIC) Study:
design and objectives. Am. J. Epidemiol. 129, 687702 (1989).
60. Chen, J. et al. Powerful statistical analysis for associating microbiomes to
enviromental covariates using generalized Unifrac distances. Bioinformatics 28,
21062113 (2012).
61. Willems, G., Pison, G., Rousseeuw, P.J. & Van Aelst, S. A robust Hotelling test.
Metrika 55, 125138 (2002).