LETTER doi:10.

1038/nature12710

A diurnal serum lipid integrates hepatic lipogenesis

and peripheral fatty acid use
Sihao Liu1*, Jonathan D. Brown2*, Kristopher J. Stanya1, Edwin Homan3, Mathias Leidl3, Karen Inouye1, Prerna Bhargava1,
Matthew R. Gangl1, Lingling Dai1,4, Ben Hatano1{, Gökhan S. Hotamisligil1, Alan Saghatelian3, Jorge Plutzky2 & Chih-Hao Lee1

Food intake increases the activity of hepatic de novo lipogenesis, clearance of circulating [3H]oleic acid (Fig. 1f). These results dem-
which mediates the conversion of glucose to fats for storage or use. onstrate that hepatic de novo lipogenesis is linked to muscle FA use.
In mice, this program follows a circadian rhythm that peaks with Ppard expression oscillated diurnally, peaking at night, coincident
nocturnal feeding1,2 and is repressed by Rev-erba/b and an HDAC3- with messenger RNA (mRNA) levels of the molecular clock Bmal1
containing complex3–5 during the day. The transcriptional activa- (also known as Arntl) in the liver and in dexamethasone-synchronized
tors controlling rhythmic lipid synthesis in the dark cycle remain primary hepatocytes (Extended Data Fig. 2a, b). In liver-conditional
poorly defined. Disturbances in hepatic lipogenesis are also assoc- Ppard knockout (LPPARDKO) mice, induction of hepatic Acc1 during
iated with systemic metabolic phenotypes6–8, suggesting that lipo- the dark cycle was abolished; diurnal expression of Acc2, fatty acid
genesis in the liver communicates with peripheral tissues to control synthase (Fasn) and stearoyl-CoA desaturase 1 (Scd1) was also altered
energy substrate homeostasis. Here we identify a PPARd-dependent (Fig. 2a), indicating that PPARd regulates rhythmic lipogenic gene
de novo lipogenic pathway in the liver that modulates fat use by expression in the liver. Daytime-restricted feeding reversed expression
muscle via a circulating lipid. The nuclear receptor PPARd controls patterns of all major molecular clocks (Extended Data Fig. 2c)12. Peak
diurnal expression of lipogenic genes in the dark/feeding cycle. mRNA levels of Ppard and lipogenic genes also shifted to the light cycle
Liver-specific PPARd activation increases, whereas hepatocyte- in control but not in LPPARDKO mice (Fig. 2b). The expression of
Ppard deletion reduces, muscle fatty acid uptake. Unbiased meta- diglycerol acyltransferase (Dgat1, triglyceride synthesis), choline kinase
bolite profiling identifies phosphatidylcholine 18:0/18:1 (PC(18:0/ a (Chka, phosphocholine synthesis) and core circadian clock genes
18:1) as a serum lipid regulated by diurnal hepatic PPARd activity. were unchanged in LPPARDKO mice (Extended Data Fig. 2a, c).
PC(18:0/18:1) reduces postprandial lipid levels and increases fatty Body weight, feeding activity and insulin sensitivity were similar between
acid use through muscle PPARa. High-fat feeding diminishes rhy- genotypes (Extended Data Fig. 2d, e and Extended Data Table 2).
thmic production of PC(18:0/18:1), whereas PC(18:0/18:1) admin- LPPARDKO reduced muscle FA uptake in the dark cycle in vivo
istration in db/db mice (also known as Lepr2/2) improves meta- (Fig. 2c), mirroring results from LACC1KD mice and demonstrating
bolic homeostasis. These findings reveal an integrated regulatory a functional consequence of this hepatic transcriptional circuitry in
circuit coupling lipid synthesis in the liver to energy use in muscle by muscle physiology.
coordinating the activity of two closely related nuclear receptors. These Products of de novo lipogenesis can exert signalling effects, for
data implicate alterations in diurnal hepatic PPARd–PC(18:0/18:1) example, palmitoleate as a lipokine and 1-palmitoyl-2-oleoyl-sn-glycero-
signalling in metabolic disorders, including obesity. 3-phosphocholine as an endogenous ligand of the nuclear receptor
PPARd promotes fatty acid (FA) synthesis in the liver9. Surprisingly, PPARa in hepatocytes13,14. In humans and mice, serum lipid composi-
hepatic PPARd overexpression (adenoviral-mediated, adPPARd) reduced tion closely resembles that of the liver15 (Extended Data Fig. 2f), sug-
circulating triglyceride (TG) and free fatty acid (FFA) levels (Fig. 1a). gesting that changes in hepatic de novo lipogenesis may have systemic
FA uptake and b-oxidation were increased in isolated soleus muscle, metabolic effects. Indeed, serum or serum-derived lipid extracts — but
compared to control mice (adenoviral-mediated green fluorescent protein not delipidated serum — collected in the dark cycle from wild-type
expression, adGFP) (Fig. 1b), indicating that a PPARd-dependent signal mice increased FA uptake in C2C12 myotubes (versus LPPARDKO,
couples liver lipid metabolism to muscle FA oxidation. To identify candi- Fig. 2d, e). Solid phase extraction of plasma lipids (Extended Data
date molecules, we performed untargeted liquid chromatography-mass Fig. 2g) identified that the phospholipid fraction stimulated FA uptake
spectrometry (LC-MS)-based metabolite profiling of hepatic lipids10,11. in myotubes (Fig. 2f).
Metabolite set enrichment analyses ranked acetyl-CoA carboxylase To identify phospholipids mediating functional interactions between
(Acaca, also known as Acc1, a rate-limiting enzyme in de novo lipogen- PPARd, hepatic lipid synthesis and muscle FA use, we profiled serum
esis) as a top altered pathway in the adPPARd/adGFP comparison lipid metabolites of samples from wild-type and LPPARDKO mice
(Extended Data Fig. 1a and Extended Data Table 1), consistent with collected at six ZT points. 735 unique ion features were detected in positive
a positive correlation of ACC1 (also known as ACACA) and PPARD and negative ionization modes (Extended Data Fig. 2f). Metabolite
expression in human livers (Extended Data Fig. 1b). Transient liver- hierarchical clustering revealed the main differences between wild-type
specific Acc1 knockdown (LACC1KD) reduced hepatic TG content and LPPARDKO serum occurred during the dark cycle (Fig. 3a, b),
and increased serum TG and FFA levels (Fig. 1c). FA uptake was decreased when PPARd-controlled lipogenesis is most active. Daytime feeding
in isolated soleus muscle from LACC1KD mice (Fig. 1d). In vivo FA uptake led to a more pronounced discordance in serum lipidomes between
assays revealed that muscle FA uptake was decreased in LACC1KD these two genotypes, indicating that LPPARDKO mice were unable to
mice in the dark/feeding cycle, when the lipogenic program is active adjust their lipogenic gene expression program (Extended Data Fig. 3a, b).
(zeitgeber time (ZT) ZT 18 or 12 a.m.; ZT 0: lights on at 6 a.m.; ZT 12: Principal component analysis (PCA) of features in positive ionization mode,
lights off at 6 p.m.) (Fig. 1e). This defect was accompanied by slower which detects phospholipids as well as mono-, di- and triacylglycerols,
1
Department of Genetics and Complex Diseases, Division of Biological Sciences, Harvard School of Public Health, 665 Huntington Avenue, Boston, Massachusetts 02115, USA. 2Cardiovascular Division,
Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, 77 Avenue Louis Pasteur, Boston, Massachusetts 02115, USA. 3Department of Chemistry, Harvard University, 12 Oxford
Street, Cambridge, Massachusetts 02138, USA. 4Good Clinical Practice Office of XiangYa Hospital and Institute of Clinical Pharmacology, Central South University, Changsha, Hunan 410000, China.
{Present address: Clinical Science Department, Research & Development, AstraZeneca K.K., 1-1- 88 Ohyodo-Naka, Kita-Ku, Osaka 531-0076, Japan.
*These authors contributed equally to this work.

5 5 0 | N AT U R E | VO L 5 0 2 | 2 4 O C T O B E R 2 0 1 3
©2013 Macmillan Publishers Limited. All rights reserved
 LETTER RESEARCH

a adGFP
b adGFP These 14 lipid species were also the main drivers of the sample
adPPARδ

β-Oxidation (c.p.m. × 105)

150 adPPARδ 1.5 2.5 4 * clustering in PCA analyses (Extended Data Fig. 3d). We focused on
Serum TG (mg dl–1)

Ex vivo FA uptake
m/z 5 788.6, putatively identified as PC(36:1), as its levels were

Serum FFA (mM)

2.0

(fold change)
3
100 1.0
* 1.5 decreased in both LPPARDKO and LACC1KD (versus control) serum
* 1.0
2 but increased in liver tissue from PPARd-overexpressing mice (Fig. 3d),
50 0.5 correlating with the FA uptake data observed in each model. The
0.5 1
extracted ion chromatogram (EIC) showed this phospholipid displayed
0 0.0 0.0 0
c C LACC1KD d
diurnal rhythmicity peaking at night (or during the day in daytime
Acc Scramble Scramble Scramble
restricted feeding) in wild-type serum, but not in LPPARDKO serum
Actin LACC1KD LACC1KD
50 5 2.5 * 1.5
LACC1KD (Extended Data Fig. 3e, f). This phospholipid was also reduced in
(mg g–1 dry weight)

Serum FFA (mM)

LACC1KD serum and increased in adPPARd liver lysates (Extended

Ex vivo FA uptake
40 4 * 2.0
(×102 mg dl–1)

(fold change)
Hepatic TG

Serum TG

30 3 1.5 1.0 Data Fig. 3e). Co-elution experiments with authentic PC(18:0/18:1)
*
20 * 2 1.0
(also known as 1-stearoyl-2-oleoyl-sn-glycero-3-phosphocholine, SOPC)
10 1 0.5
0.5 and tandem mass spectrometry scanning16 confirmed this ion as
PC(18:0/18:1), whereas PC(18:1/18:0) or others such as PC(16:1/20:0)
0 0 0.0 0.0
were not observed (Extended Data Fig. 3g and data not shown). The
e f P = 0.06
20 concentrations of PC(18:0/18:1) in wild-type serum ranged from
[3H]FA clearance

Scramble
Scramble
LACC1KD ,50 mM at ZT 8 (day) to ,115 mM ZT 20 (night) using deuterated
(AUC–1)

10 LACC1KD
* *
3 * 3 *
2 d83-PC(18:0/18:0) as an internal standard. The night-time increase in
In vivo FA uptake

0 PC(18:0/18:1) levels was diminished in LPPARDKO mice (Fig. 3e).

(fold change)

ZT 6 ZT 18
c.p.m. (×103)

2 2 Scramble(ZT 6) PPARd synthetic ligand treatment (GW501516, 4 days) increased

*

Scramble(ZT 18)
1
LACC1KD(ZT 6) serum PC(18:0/18:1) levels in wild-type mice but not in LPPARDKO
1 1 LACC1KD(ZT 18) mice (Fig. 3f). These data identified PC(18:0/18:1) as a serum lipid
0 0
regulated by hepatic PPARd diurnally in three mouse models.
ZT 6 ZT 18 ZT 6 ZT 18 0
0 4 8 Intraperitoneal injection of escalating concentrations of PC(18:0/18:1)
Soleus Gastrocnemius Time after infusion (min)
reduced serum TG and FFA levels (Extended Data Fig. 3h), with a trend
Figure 1 | Hepatic PPARd and Acc1 are linked to muscle FA use. a, Serum of increased muscle FA uptake. Tail-vein injection of PC(18:0/18:1)
TG and FFA levels in adGFP or adPPARd mice (n 5 5). b, Ex vivo FA uptake (5 mg kg21 body weight) also reduced serum TG (Fig. 3g). Notably,
and oxidation in isolated soleus muscle. c, Hepatic TG and serum TG and FFA PC(16:0/18:1) and PC(18:1/18:1) had no effect. In myotubes, only
levels in LACC1KD or control (Scramble) mice (n 5 5). Inset, immunoblotting PC(18:0/18:1) increased FA uptake (Fig. 3h). Catheter-based, continu-
of liver ACC protein. d, Ex vivo FA uptake in isolated soleus muscle. e, In vivo
ous infusion of PC(18:0/18:1) (25 mg min21 kg21 for 200 min) through
FA uptake in soleus and gastrocnemius muscle. f, Serum [3H]oleic acid
disappearance. Inset, [3H]FA clearance. AUC, area under the curve of the jugular vein also lowered circulating TG and FFA levels (Fig. 3i). As
disappearance. *P , 0.05 (t-test), data shown as mean 6 s.e.m. such, PC(18:0/18:1) links the hepatic PPARd-controlled lipogenic pro-
gram to serum lipid concentrations and muscle fat use.
demonstrated co-clustering of LPPARDKO and LACC1KD serum Mechanistically, several FA use genes in the muscle, namely Cd36,
samples from the dark cycle (Extended Data Fig. 3c). Comparison of serum Fabp3, Fabp4, Fatp1 (also known as Slc27a1), Fatp4 (also known as
and liver metabolomes from three relevant models, LPPARDKO, Slc27a4), Ppara, Cidea and Acadm, were induced in adPPARd and/
LACC1KD and adPPARd, in positive ionization mode (Supplemen- or PC(18:0/18:1)-treated mice, but repressed in LPPARDKO and
tary Data) yielded 14 features altered in all three models (Fig. 3c, d). LACC1KD animals (Fig. 4a). Cd36 and Fabp3 are known mediators

a WT c WT
Figure 2 | Hepatic PPARd controls
LPPARDKO
2.5 LPPARDKO 2.5
liver lipogenic gene expression and
1.5 Acc1*† 1.5 Acc2† * * muscle FA uptake. a, Hepatic
In vivo FA uptake

2.0 * 2.0 *
expression

(fold change)

lipogenic gene expression in wild-

Relative

1.0 1.0
1.5 1.5 type (WT) and LPPARDKO mice
0.5 0.5 1.0 1.0 (n 5 4 per time point). White and
0.0 0.0 0.5 0.5 black bars on the x-axis represent
light and dark cycles, respectively.
2.5 Fasn† 2.0 Scd1† 0.0 0.0
ZT 6 ZT 18 ZT 6 ZT 18 b, Liver gene expression under
expression

2.0 1.5
Relative

Soleus Gastrocnemius daytime feeding (n 5 3). Grey bars,

1.5
1.0
1.0 d e time when food was available.
0.5 0.5 WT WT
LPPARDKO LPPARDKO *P , 0.05 (ANOVA), WT versus
0.0 0.0
LPPARDKO; {P , 0.05, comparing
In vitro FA uptake

In vitro FA uptake

0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28 2.0 * 1.5
*
(fold change)

(fold change)

Zeitgeber time Zeitgeber time * circadian patterns. c, In vivo muscle

b 1.5
1.0
WT
1.0 FA uptake (n 5 3). d–f, In vitro FA
LPPARDKO
0.5
0.5 uptake in C2C12 myotubes treated
6.0 Ppard*† 2.0 Acc2*
0.0 0.0 with serum (2%) pooled from light or
expression

1.5 Light Dark Total Delipidated dark cycle samples, serum total lipids
Relative

4.0
1.0 lipids serum
or delipidated serum (dark cycle
2.0 0.5 f samples) or serum lipid fractions
0.0 WT
0.0
LPPARDKO
(n 5 3). *P , 0.05 (t-test), data
In vitro FA uptake

4.0 Acc1* 2.0 Scd1* 1.5 shown as mean 6 s.e.m.

(fold change)

*
expression

3.0 1.5 1.0

Relative

2.0 1.0
0.5
1.0 0.5
0.0 0.0 0.0
DAG/ FFA Total
0 4 8 12 16 20 24 28 0 4 8 12 16 20 24 28
MAG PL
Zeitgeber time Zeitgeber time

2 4 O C T O B E R 2 0 1 3 | VO L 5 0 2 | N AT U R E | 5 5 1
©2013 Macmillan Publishers Limited. All rights reserved
 RESEARCH LETTER

a b d adPPARδ Figure 3 | PC(18:0/18:1) links

LPPARDKO hepatic PPARd to serum lipid levels

ZT 4
ZT 8
ZT 12
ZT 16
ZT 20
ZT 24
WT LACC1KD
LPPARDKO Z-score Z-score Z-score and muscle FA uptake. a, Serum
Controls
–10 0 10 –10 0 10 –10 0 10
Putative identity m/z lipid heat map (n 5 3 per time point).
5 MAG(18:0) 359.32
MAG(18:0) 376.34 White, light (starting at ZT 4) and
4 LysoPC(18:3) 518.32
LysoPC(22:0) 580.44 black, dark cycles. b, Dendrogram
Height
PC(34:1) 760.59
3
PC(36:3) 784.59 from hierarchical clustering.
2 PC(36:1) 788.62 c, Cross-comparison of changed
PC(38:8)/PC(36:5) 802.55
1 PC(40:6)/PC(38:3) 834.61 lipids. d, Z-score plots of 14
PC(40:5)/PC(38:2) 836.61
SM(d18:1/24:0) 837.69 commonly changed features. MAG,
0
TG(56:7)
TG(56:6)
922.79
924.81
monoacylglycerol; LysoPC,
Light Dark TG(58:6) 950.83 lysophosphatidylcholine; SM,
c e f sphingomyelin. e, Serum PC(36:1)
92
WT Vehicle quantification in WT (n 5 5) and
LPPARDKO GW1516
High

*
LPPARDKO (n 5 4) mice. f, Serum
14 150 * 80 *
PC (36:1) (μM)

PC (36:1) (μM)
38 14 PC(36:1) concentrations in WT/
60 LPPARDKO 6 GW501516 (n 5 5).
100
42 348
40 g, Serum TG changes (tail-vein
50
95 20 injection) with phospholipids in
Low

WT LPPARDKO LPPARDKO 0 0 wild-type mice (n 5 6). h, FA uptake

LACC1KD ZT 8 ZT 20 WT LPPARDKO
adPPARδ in C2C12 myotubes treated with
* phospholipids (50 mM, n 5 3).
g Serum TG h i 200 ns Vehicle
* PC(18:0/18:1) i, Serum TG and FFA levels
Serum TG

(fold change) 150

(mg dl–1)

after PC(18:0/18:1) infusion (n 5 6,

0.0

0.5

1.0

1.5

In vitro FA uptake

2.0 100
*
(fold change)

1.5 50 wild-type C57BL/6J mice). *P , 0.05

Vehicle
1.0 0 (t-test), data presented as
Baseline infusion mean 6 s.e.m.
PC(18:0/18:1)
*

0.5 ns
0 2.0 * *
PC(16:0/18:1)
Vehicle
PC(18:0/18:1)
PE(18:0/18:1)
PA(18:0/18:1)
PC(18:1/18:0)
PC(16:0/16:0)
PC(16:0/18:1)
PC(18:1/18:1)
PC(18:0/18:2)

Serum FFA

1.5
(mM)

PC(18:1/18:1) 1.0
0.5
Baseline 0.0
Post-injection Baseline infusion

of muscle FA uptake17,18. Cd36 expression at mRNA and protein levels Extended Data Fig. 4a). This diurnal pattern was disrupted in muscle
also oscillated in wild-type muscle peaking in the dark cycle, and of LPPARDKO mice. Furthermore, whereas PPARd agonist GW501516
shifted to the light cycle by daytime restricted feeding (Fig. 4b and increased muscle expression of Cd36 and Fabp3 (Fig. 4c), enhanced

a b c Figure 4 | PC(18:0/18:1) regulates

Cd36 2.5 2.5 Veh icle
WT LPPARDKO * muscle FA use through PPARa.
(fold change)

Cidea 2.0 2.0 GW 1516

Fabp4 Zeitgeber 4 8 12 16 20 24 4 8 12 16 20 24
a, Muscle gene expression. b, Muscle
Cd36

Fatp4 time CD36 1.5 1.5

Dgat1 Loading 1.0 1.0 CD36 protein (top) and gene
Ppara
0.5 0.5 (bottom, n 5 4 per time point)
Acox1 Cd36†
Fatp1 2.0 WT 0.0 0.0
Fabp3 WT LPPARDKO expression. c, Cd36 and Fabp3
Log2(fold change)

expression

1.5 LPPARDKO
Relative

Acadm
Lpl
1.0
0.7 2.5 2.5 expression in wild-type and
*
(fold change)

1.0
LPPARDKO muscle with or without
adPPARδ
PC(18:0/18:1)
LACC1KD
LPPARDKO

0.3 2.0 2.0

Fabp3

0.0 0.5
–0.3 1.5 1.5 GW501516 (n 5 5). d, Top, serum
–0.5 0.0 1.0 1.0
–1.0 0 4 8 12 16 20 24 28 TG levels in wild-type and
0.5 0.5
Zeitgeber time PPARaKO mice after vehicle or
0.0 0.0
d Vehicle e Vehicle f
WT LPPARDKO PC(18:0/18:1) infusion (n 5 6, wild-
PC(18:0/18:1) PC(18:0/18:1) * type from Fig. 3i). Bottom, in vivo
200 150 2.5 * 2.5 2.0 * soleus muscle FA uptake. e, Muscle
Vehicle
(fold change)

(fold change)

2.0 2.0
Serum TG

* PC(18:0/18:1)
FA uptake

Cd36 and Fabp3 expression in wild-

(mg dl–1)

100 1.5
Cd36

1.5 1.5
100 *
1.0 1.0 1.0 type and PPARaKO mice. f, FA
50
0.5 0.5 0.5 uptake in C2C12 myotubes (n 5 3).
0 0 0.0 0.0 0.0 Top, Ppara knockdown or control;
WT PPARαKO WT PPARαKO shControl shPpara bottom, wild-type Ppara or AF2
In vivo FA uptake

2.0 2.0 2.0 2.0 2.0 *

* mutant (AF2m). g–i, Fasting serum
(fold change)

(fold change)
(fold change)

FA uptake

1.5 * 1.5 1.5 1.5 1.5 lipid concentrations (g), glucose

Fabp3

1.0 1.0 1.0 1.0 1.0 tolerance test (GTT) and insulin
0.5 0.5 0.5 0.5 0.5 tolerance test (ITT) (h), and muscle
0.0 0.0 0.0 0.0 0.0 lipid content (i) in db/db mice treated
WT PPARαKO WT PPARαKO WT AF2m
with vehicle (n 5 4) or PC(18:0/18:1)
g Vehicle h Vehicle
i Vehicle (n 5 5). *P , 0.05 (t-test); {P , 0.05
PC(18:0/18:1) PC(18:0/18:1) PC(18:0/18:1)
100 (ANOVA); data presented as
Serum TG

30 P = 0.11
(mg dl–1)

*
Muscle FFA Muscle TG

800 GTT 400 ITT mean 6 s.e.m.

(mg g–1)

20
Blood glucose

50
(mg dl–1)

† 10
0 †
Normalized

400 200 0
1.0
2.0 *
Serum FFA

0.6 15 P = 0.06
(μM g–1)
(mM)

0 40 80 120
1.0 0 0 10
0 40 80 0 40 80 120
5
0.0 Time after injection Time after injection 0
db/db (minutes) (minutes) db/db

5 5 2 | N AT U R E | VO L 5 0 2 | 2 4 O C T O B E R 2 0 1 3
©2013 Macmillan Publishers Limited. All rights reserved
 LETTER RESEARCH

muscle FA uptake and lowered serum TG levels in wild-type mice [3H]palmitic acid. Supernatants were collected and the 3H radioactivity in the
(Extended Data Fig. 4b), all these ligand effects were lost in LPPARDKO aqueous phase was quantified as described27. In vivo FA uptake: we adapted an
animals. These results indicate that hepatic PPARd may alter expres- established protocol28. Briefly, 10 mCi [3H]oleic acid in 3.5% BSA was infused
sion of muscle genes and FA use through PC(18:0/18:1). Indeed, through the portal vein (or in 5% intralipid through jugular vein in PC(18:0/
18:1) infusion experiments). Blood samples were collected at 1, 2, 5, 7 and 10 min.
PC(18:0/18:1) treatment induced Cd36/Fabp3 expression in myotubes Soleus and gastrocnemius muscles were isolated at 10 min. FA uptake was calcu-
whereas Cd36 knockdown abrogated the effect of PC(18:0/18:1) on lated as described29.
muscle cell FA uptake (Extended Data Fig. 4c, d). PPARa controls Data analysis. Sample handling (Extended Data Fig. 5, 6) and statistical tests are
FA metabolism in muscle19 and can be activated by certain PCs14. In described in Methods. Data are presented as mean 6 s.e.m. Information for
reporter assays, PC(18:0/18:1) moderately activated PPARa (Extended mouse cohorts and primer sequences are described in Extended Data Tables 3,
Data Fig. 4e). However, the effects of PC(18:0/18:1) infusion on redu- 4. For in vitro assays, the mean and s.e.m. were determined from 3–4 biological
cing serum TG levels and increasing muscle FA uptake and Cd36/ replicates for one representative experiment. Experiments were repeated at least
Fabp3 expression were abolished in Ppara knockout (PPARaKO) mice three times. Significance was set at P , 0.05.
(Fig. 4d, e). In myotubes, increased FA uptake by PC(18:0/18:1) was Online Content Any additional Methods, Extended Data display items and Source
diminished by Ppara knockdown or by a Ppara mutant lacking the Data are available in the online version of the paper; references unique to these
carboxy-terminus activation function domain (AF2), indicating that sections appear only in the online paper.
PC(18:0/18:1) or its metabolites may modulate PPARa transcriptional
Received 5 July 2012; accepted 19 September 2013.
activity in vivo (Fig. 4f). These findings demonstrate that a hepatic
PPARd–PC(18:0/18:1)–muscle PPARa signalling cascade coordinates 1. Asher, G. & Schibler, U. Crosstalk between components of circadian and metabolic
fat synthesis and use. cycles in mammals. Cell Metab. 13, 125–137 (2011).
Obesity alters circadian rhythms in multiple tissues, resulting in 2. Bass, J. & Takahashi, J. S. Circadian integration of metabolism and energetics.
Science 330, 1349–1354 (2010).
abnormal metabolism20. Diet-induced obesity altered the rhythmic 3. Feng, D. et al. A circadian rhythm orchestrated by histone deacetylase 3 controls
pattern of serum PC(18:0/18:1) (Extended Data Fig. 4f, g). In db/db hepatic lipid metabolism. Science 331, 1315–1319 (2011).
mice (a genetic model of obesity), tail-vein injection of PC(18:0/18:1) 4. Solt, L. A. et al. Regulation of circadian behaviour and metabolism by synthetic
REV-ERB agonists. Nature 485, 62–68 (2012).
(5 mg kg21 day21 for 6 days) reduced fasting TG and FFA levels
5. Cho, H. et al. Regulation of circadian behaviour and metabolism by REV-ERB-a and
(Fig. 4g). Non-fasting blood glucose levels trended lower in PC(18:0/ REV-ERB-b. Nature 485, 123–127 (2012).
18:1)-treated animals (Extended Data Fig. 4h). PC(18:0/18:1) reduced 6. Miyazaki, M. et al. Hepatic stearoyl-CoA desaturase-1 deficiency protects mice
fasting glucose and improved glucose tolerance test results (Fig. 4h and from carbohydrate-induced adiposity and hepatic steatosis. Cell Metab. 6,
484–496 (2007).
Extended Data Table 2). Glucose concentrations throughout insulin 7. Matsuzaka, T. et al. Crucial role of a long-chain fatty acid elongase,
tolerance test were lower with PC(18:0/18:1) treatment (Fig. 4h), although Elovl6, in obesity-induced insulin resistance. Nature Med. 13, 1193–1202
the per cent change did not differ. Fasting insulin levels were similar (2007).
(Extended Data Table 2). Muscle lipid contents in the PC(18:0/18:1)- 8. Choi, C. S. et al. Continuous fat oxidation in acetyl-CoA carboxylase 2 knockout
mice increases total energy expenditure, reduces fat mass, and improves insulin
treated group trended lower (Fig. 4i), consistent with the notion that sensitivity. Proc. Natl Acad. Sci. USA 104, 16480–16485 (2007).
PC(18:0/18:1) promotes fat use in the muscle. 9. Liu, S. et al. Role of peroxisome proliferator-activated receptor d/b in hepatic
The data presented here reveal that diurnal oscillations of hepatic metabolic regulation. J. Biol. Chem. 286, 1237–1247 (2011).
10. Homan, E. A., Kim, Y. G., Cardia, J. P. & Saghatelian, A. Monoalkylglycerol ether
de novo lipogenesis-derived lipid metabolites coordinate metabolic func- lipids promote adipogenesis. J. Am. Chem. Soc. 133, 5178–5181 (2011).
tions between liver and muscle (Extended Data Fig. 4i). The finding 11. Brown, J. D., Oligino, E., Rader, D. J., Saghatelian, A. & Plutzky, J. VLDL hydrolysis by
also adds PC(18:0/18:1) to an emerging network of signalling mole- hepatic lipase regulates PPARd transcriptional responses. PLoS ONE 6, e21209
(2011).
cules mediating inter-organ communications13,21–24. The approxi- 12. Damiola, F. et al. Restricted feeding uncouples circadian oscillators in peripheral
mately twofold change in PC(18:0/18:1) concentrations is similar to tissues from the central pacemaker in the suprachiasmatic nucleus. Genes Dev.
other lipid mediators, including two gut-derived lipids that regulate 14, 2950–2961 (2000).
satiety: N-acylphosphatidylethanolamine and oleylethanolamide21,22, 13. Cao, H. et al. Identification of a lipokine, a lipid hormone linking adipose tissue to
systemic metabolism. Cell 134, 933–944 (2008).
suggesting that physiological fluctuations in levels of lipid mediators 14. Chakravarthy, M. V. et al. Identification of a physiologically relevant endogenous
are sufficient to stimulate specific biological functions. Specificity is ligand for PPARa in liver. Cell 138, 476–488 (2009).
further supported by data showing that systemic treatment with 15. Kotronen, A. et al. Comparison of lipid and fatty acid composition of the liver,
subcutaneous and intra-abdominal adipose tissue, and serum. Obesity 18,
PC(16:0/18:1), a hepatic PPARa ligand14, did not lower serum TG or 937–944 (2010).
stimulate FA uptake (Fig. 3g, h), nor did it activate PPARa in muscle 16. Hsu, F. F., Bohrer, A. & Turk, J. Formation of lithiated adducts of
cells (Extended Data Fig. 4j, k). An association between serum PC(36:1) glycerophosphocholine lipids facilitates their identification by electrospray
levels and diabetes mellitus in humans has recently been reported25. ionization tandem mass spectrometry. J. Am. Soc. Mass Spectrom. 9, 516–526
(1998).
Herein, diet-induced obesity dysregulated temporal PC(18:0/18:1) 17. Glatz, J. F., Luiken, J. J. & Bonen, A. Membrane fatty acid transporters as regulators
production, whereas PC(18:0/18:1) treatment improved lipid and glu- of lipid metabolism: implications for metabolic disease. Physiol. Rev. 90, 367–417
cose metabolism in db/db mice. Although reduced ectopic fat accu- (2010).
18. Shearer, J. et al. Heart-type fatty acid-binding protein reciprocally regulates
mulation in PC(18:0/18:1)-treated muscle would be predicted to improve glucose and fatty acid utilization during exercise. Am. J. Physiol. Endocrinol. Metab.
metabolic homeostasis26, future studies are required to determine how 288, E292–E297 (2005).
PC(18:0/18:1) lowers fasting glucose, how twofold fluctuations in 19. Finck, B. N. et al. A potential link between muscle peroxisome proliferator-
serum PC(18:0/18:1) levels transduce physiological effects and how activated receptor-a signaling and obesity-related diabetes. Cell Metab. 1,
133–144 (2005).
PC(18:0/18:1) achieves specificity towards muscle PPARa. Nevertheless, 20. Kohsaka, A. et al. High-fat diet disrupts behavioral and molecular circadian
mechanisms that restore the rhythmic activity of the PPARd–PC(18:0/ rhythms in mice. Cell Metab. 6, 414–421 (2007).
18:1) axis may provide new therapeutic opportunities to treat meta- 21. Gillum, M. P. et al. N-acylphosphatidylethanolamine, a gut-derived circulating
factor induced by fat ingestion, inhibits food intake. Cell 135, 813–824
bolic diseases. (2008).
22. Fu, J. et al. Oleylethanolamide regulates feeding and body weight through
METHODS SUMMARY activation of the nuclear receptor PPAR-a. Nature 425, 90–93 (2003).
Functional assays. In vitro FA uptake: C2C12 myotubes were pre-treated with 23. Lee, J. M. et al. A nuclear-receptor-dependent phosphatidylcholine pathway with
lipids complexed in 0.2 % FA-free BSA overnight. Cells were washed before sub- antidiabetic effects. Nature 474, 506–510 (2011).
24. Fu, S. et al. Aberrant lipid metabolism disrupts calcium homeostasis causing liver
jecting to a 5 min FA loading with 1 mCi ml21 [3H]oleic acid in Krebs–Ringer
endoplasmic reticulum stress in obesity. Nature 473, 528–531 (2011).
HEPES (KRH) buffer, 1% BSA and 100 mM oleic acid. Intracellular 3H radioacti- 25. Floegel, A. et al. Identification of serum metabolites associated with risk
vity was determined and normalized to protein concentration. Ex vivo FA oxida- of type 2 diabetes using a targeted metabolomic approach. Diabetes 62, 639–648
tion: freshly isolated soleus muscles were incubated at 37 uC for 30 min with (2013).

2 4 O C T O B E R 2 0 1 3 | VO L 5 0 2 | N AT U R E | 5 5 3
©2013 Macmillan Publishers Limited. All rights reserved
 RESEARCH LETTER

26. Samuel, V. T. & Shulman, G. I. Mechanisms for insulin resistance: common threads Health grants R01DK075046 (C.-H.L.), R01HL048743 (J.P.) and K08HL105678
and missing links. Cell 148, 852–871 (2012). (J.D.B.).
27. Reilly, S. M. et al. Nuclear receptor corepressor SMRT regulates mitochondrial
oxidative metabolism and mediates aging-related metabolic deterioration. Cell Author Contributions S.L., A.S., J.P. and C.-H.L. designed the research. S.L. performed
Metab. 12, 643–653 (2010). most of the experiments with technical assistance from K.J.S., P.B., M.R.G. and L.D.; S.L.,
28. Bartelt, A. et al. Brown adipose tissue activity controls triglyceride clearance. Nature J.D.B., E.H., M.L. and A.S. developed and performed untargeted and targeted metabolite
Med. 17, 200–205 (2011). profiling. B.H. generated adGFP and adPPARd virus. K.I. performed metabolic cage
29. Shearer, J. et al. Long chain fatty acid uptake in vivo: comparison of [125I]-BMIPP and lipid infusion experiments. G.S.H., J.P., A.S. and J.D.B. provided critical
and [3H]-bromopalmitate. Lipids 43, 703–711 (2008). intellectual inputs and manuscript editing. S.L. and C.-H.L. analysed the data and wrote
the paper.
Supplementary Information is available in the online version of the paper.
Author Information Reprints and permissions information is available at
Acknowledgements We thank C. Newgard for providing shACC1 and shScramble www.nature.com/reprints. The authors declare no competing financial interests.
adenovirus, U. Unluturk, X. Li and A. N. White for technical help and D. E. Cohen, Readers are welcome to comment on the online version of the paper.
S. Watkins and D. Jacobi for comments. This work is supported by the American Heart Correspondence and requests for materials should be addressed to C.-H.L.
Association and the American Diabetes Association (C.-H.L.) and National Institutes of ([email protected]).

5 5 4 | N AT U R E | VO L 5 0 2 | 2 4 O C TO B E R 2 0 1 3
©2013 Macmillan Publishers Limited. All rights reserved
 LETTER RESEARCH

METHODS db/db mice. Eight-week-old male FVB/NJ-db/db mice were injected with a
9 bolus of 5mg kg21 body weight PC(18:0/18:1) or vehicle carried by 5% intralipid
Materials. Ppard or GFP adenovirus was generated as described . The shScramble
through tail vein once daily for 6 days (n 5 4–5 per treatment), followed by meta-
and shAcc1 adenovirus were provided by C. Newgard30. Small hairpin RNA
bolic studies.
sequences against Cd3631, Ppara (59-CCCTTATCTGAAGAATTCTTA-39) or
Metabolic studies. Metabolic cage studies were performed in a Comprehensive
luciferase (control) were cloned in the pSIREN-RetroQ vector. PparaAF2 mutant
Lab Animal Monitoring System (Columbus Instruments). Data were collected for
construct was generated through site-directed mutagenesis to introduce a stop
48 h starting at the beginning of the dark cycle. TG and FFA were determined by
codon in front of the AF2 domain using wild-type Ppara construct as the template.
colorimetric methods (Thermo and Wako). Hepatic TG was determined from
The oligonucleotide used for mutagenesis was: 59-GAGCATGCGCAGCTCGA
chloroform:methanol (2:1 v/v) extracts of vacuum dried liver samples. Glucose
GTAGGTCATCAAGAAGACC-39. Full-length Ppara or PparaAF2 mutant com-
and insulin tolerance test were performed on overnight fasted animals. Blood
plementary DNA was cloned in the pBabe retroviral vector.
glucose levels were determined at indicated time points after administration of
Cell lines. All cell lines were obtained from ATCC: C2C12 (CRL-1772); HEK293
1.5 mg kg21 body weight glucose or 1U kg21 body weight insulin.
(CRL-1573). C2C12 myoblasts were infected with retroviral particles and selected
Lipid extraction, fractionation and treatments. Serum lipids were diluted with
against puromycin to generate stable lines. All stable C2C12 myoblasts were able to
PBS followed by a liquid/liquid extraction with chloroform and methanol (final
differentiate into myotubes with no apparent defects. Differentiation of C2C12
concentrations of chloroform:methanol: PBS were 2:1:1 v/v). This extraction mix-
myoblasts was performed in 2% horse serum, high-glucose DMEM for 8 days.
ture phase separated to provide an aqueous layer (top) and an organic layer
FA uptake/oxidation assays. In vitro FA uptake. C2C12 myotubes were pre-
(bottom), which contains all lipids. The lipid-containing layer was concentrated
treated with lipids complexed in 0.2% BSA (FA-free) overnight. Cells were thor-
to dryness using a constant stream of nitrogen and dissolved in chloroform,
oughly washed before subjecting to a 5-min FA loading with 1 mCi ml21 [3H]oleic
followed by fractionation using a simple column purification method, as
acid in Krebs–Ringer HEPES (KRH) buffer, 1% FA-free BSA and 100 mM oleic
described32. Briefly, aminopropyl columns (Sep-Pak Vac NH2 cartridge 3 ml per
acid. Intracellular 3H radioactivity was determined and normalized to protein
500 mg 55–105 mm, Waters) were equilibrated 3 times with acetone/water (7:1).
concentration.
Lipids in chloroform were dried under nitrogen and re-dissolved in hexane/
Ex vivo FA oxidation. Freshly isolated soleus muscles were incubated at 37 uC methyl-butyl-tert-ether (MBTE)/acetic acid (100:3:0.3). Lipids were loaded onto
for 30 min with 2% FA-free BSA containing KRH buffer supplemented with the equilibrated column and were eluted sequentially with hexane, hexane/chlo-
0.2 mM palmitic acid and 4 mCi ml21 [3H] palmitic acid. Supernatants were col- roform/ethyl acetate (100:5:5), chloroform/2-propanol (2:1) (diacylglycerol/
lected and the 3H radioactivity in the aqueous phase was quantified as described27. monoacylglycerol fraction), chloroform: methanol/acetic acid (100:2:2) (FFA frac-
In vivo FA uptake. We adapted an established protocol28. Briefly, 10 mCi [3H]oleic tion), and methanol/chloroform/water (10:5:4) (phospholipids fraction, Extended
acid in 3.5% FA-free BSA was infused through portal vein (or in 5% intralipid Data Fig. 2g). Each fraction was dried under nitrogen and dissolved in chloroform.
through jugular vein in PC(18:0/18:1) infusion experiments). Blood samples were For in vitro experiments, lipids were dissolved in 0.2% FA-free BSA in DMEM with
collected at 1, 2, 5, 7 and 10 min after infusion to determine radioactivity. At 2% double stripped FBS (charcoal stripped and lipoprotein deficient) and applied
10 min, soleus and gastrocnemius muscles were isolated. FA uptake was calculated to cells overnight. Cells were washed extensively before functional assays.
as described29. Primary hepatocytes and in vitro synchronization. Primary hepatocytes were
Animals. Mice used in the current study were on the C57BL/6J background, isolated as described33. 100 nM of dexamethasone was applied for 1 h to synchron-
except for wild-type FVB/NJ and FVB/NJ-db/db mice used for PC(18:0/18:1) ize cells. After thorough washing, fresh culture media was added and cells were
tail-vein injection (see Extended Data Table 3 for detail). Liver-specific Ppard collected at the indicated time after dexamethasone removal.
knockout mice were generated by crossing albumin-cre transgenic mice to Ppard Gene expression and western blots. Gene expression was determined by SYBR
f/f mice. Ppara knockout mice (PPARaKO), FVN/NJ and FVB/NJ-db/db mice Green based quantitative PCR with reverse transcription (qRT–PCR) using 36b4
were purchased from Jackson Laboratory. Animals were on chow diet (with the as an internal standard. A relative standard curve method was used to calculate the
exception of Extended Data Fig. 4f, g) and housed in a barrier facility with 12-h relative expression of genes. For high-throughput RT-qPCR array used for muscle
light and dark cycles. ZT 0: lights on at 6 a.m.; ZT 12: lights off at 6 p.m. All animal gene expression, the DDCt method was used to measure relative expression.
studies were approved by the Harvard Medical Area Standing Committee on Hierarchical clustering and heat map were generated by Cluster and Java
Animals. TreeView. The primers used in this study were obtained from Primer Bank34
Adenovirus-mediated liver-specific overexpression or knockdown. Adenovirus and listed in Extended Data Table 4. Protein levels of CD36 were determined by
was injected through the tail vein (109 plaque-forming units per mouse). Subsequent western blotting of muscle lysates using antibody against CD36 (SC-9154, Santa
metabolic characterizations were carried out 4 days post injection. AdPPARd/adGFP Cruz). For circadian studies, a pooled sample from wild-type or LPPARDKO mice
was repeated in 3 cohorts (8–10 weeks old male, n 5 4–6) and LACC1KD was con- (n 5 4) at each time point was used. For in vivo Acc1 knockdown, the knockdown
ducted in 2 cohorts (8–10 weeks old male, n 5 5). efficiency was determined by western blotting (n 5 5). Two representative animals
Circadian gene expression. 5 cohorts were used for circadian studies (8 weeks from each group were shown (Fig. 1c).
old, 4 male and 1 female cohorts, showing similar results). For circadian studies, Liquid-chromatography mass-spectrometry (LC-MS). A 2:1:1 chloroform:
animals were killed every 4 h starting at 10 a.m. (ZT 4) for 24 h (n 5 3–4 per methanol:PBS solution was prepared for lipid extraction to isolate organic soluble
genotype per time point) with free access to food and water. For dark cycle time metabolites. Following brief vortexing, samples were centrifuged at 2,500g at 4 uC
points, animals were killed under red safety light before dissection. For daytime for 10 min. The organic layer (bottom) was transferred to a new vial and solvents
restricted feeding studies, animals were fed daily between 6 a.m. (ZT 0) and 2 p.m. were evaporated under a stream of nitrogen. Samples were then dissolved in
(ZT 8) for 7 days under 12-h light and dark cycles. On the 8th day, animals were chloroform (120 ml) to provide a mass spectrometry ready solution and stored
killed every 4 h starting at 6 a.m. (ZT 0) for 24 h (n 5 3 per genotype per time point). at 280 uC until LC-MS analysis (within 48 h of extraction). For both positive and
GW501516 treatment. Wild-type and LPPARDKO mice (n 5 4–5 per genotype negative ionization mode LC-MS runs, 20 ml of extract was injected. LC-MS ana-
per treatment) were force-fed with 2 mg kg21 body weight day21 GW501516 lysis was performed using an Agilent 6210 Accurate-Mass time-of-flight LC-MS
carried by 0.5% methylcellulose solution for 4 days. Animals were killed 4 h after system as described10,11. For LC analysis in negative mode, a Gemini (Phenomenex)
the last gavage. C18 column (5 mm, 4.6 3 50 mm) was used together with a pre-column (C18,
PC(18:0/18:1) injection studies. For the pilot experiment, 8–10-week old male 3.5 mm, 2 3 20 mm). Mobile phase A consisted of 95:5 water:methanol and mobile
C57BL/6J mice were injected intraperitoneally with 1.25–5 mg kg21 PC(18:0/ phase B was composed of 60:35:5 isopropanol:methanol:water. Both A and B were
18:1). Circulating lipids levels were determined 2 and 4 h after injection to deter- supplemented with 0.1% ammonium hydroxide solution (28% in water). The flow
mine the biological activity and dosing for PC(18:0/18:1). 5 mg kg21 in 5% intra- rate for each run was 0.5 ml min21. The gradient started at 0% B for 5 min and
lipid was later used for tail-vein injection in FVB/NJ and serum lipids were linearly increased to 100% B over 40 min, was then maintained at 100% B for 8 min
measured 4 h later. PC(18:0/18:1) showed similar lipid lowering effects when before re-equilibrating for 7 min at 0% B. For the LC analysis in positive mode, a
injection was performed during the dark (ZT 12) or light (ZT 8) cycle. FVB/NJ Luna (Phenomenex) C5 column (5 mm, 4.6 3 50 mm) was used together with a
mice were used for these studies for technical reasons (ease of tail-vein injection). pre-column (C4, 3.5 mm, 2 3 20 mm). Mobile phase A and B and the gradient
PC(18:0/18:1) infusion studies. 8–10-week old male C57BL/6J and PPARaKO were the same as for negative mode, but supplemented with 0.1% formic acid and
mice (n 5 6–7 per genotype per treatment) were catheterized through the jugular 5 mM ammonium formate. MS analysis was performed with an electrospray
vein. 5 days post-operation, animals were infused with PC(18:0/18:1) or vehicle source ionization (ESI) interface. The capillary voltage was set to 3.0 kV and the
carried by 5% intralipid at a rate of 25 mg kg21 min21 for 200 min at ZT 4 (10 a.m.). fragmentor voltage to 100 V. The drying gas temperature was 350 uC, the drying
After infusion, a bolus of 10 mCi [3H]oleic acid was infused to determine the in vivo gas flow was 10 l min21, and the nebulizer pressure was 45 p.s.i. Data were collected
FA uptake rate as described in the method section. using a mass range from 100–1,500 Da. For wild-type and LPPARDKO serum

©2013 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

samples, all samples of each genotype from different time points were detected in a Identification of significant features. The empirical P value for each pair of
single consecutive run. To validate the results, samples from ZT 8, ZT 16 and ZT 24 comparison was calculated by randomly permuting sample labels for 1,000 times
were subject to a second run (Extended Data Fig. 5). For Scramble and LACC1KD to obtain the null distribution. The analysis was carried out in Multiple Experiment
serum or adGFP and adPPARd liver, the entire sample set was run in a single Viewer40. False discovery rate was determined by Benjamini–Hochberg method. A
session. feature is considered significant for downstream cross-comparison with unad-
Targeted analysis of phosphocholine species. Side-chain composition of phos- justed P , 0.05. Significantly changed features in wild-type and LPPARDKO mice
phocholine species. Phosphatidylcholine fatty acyl chain composition was analysed serum at night (n 5 6, pooled sample set from ZT 16 and ZT 20), Scramble and
separately based on established conditions15 on an Agilent 6410 triple quadrupole- LACC1KD mice serum (n 5 5), and adGFP and adPPARd liver lysates (n 5 4)
mass spectrometer (QQQ-MS) by direct injection of 1 ml of serum lipid extracts were compared and visualized in Venn diagram. A total of 158, 189 and 418 features
without chromatography. This approach generates lithiated phosphocholine were significantly altered in LPPARKO/wild-type (serum samples at ZT 16/ZT 20,
adducts, which unlike protonated adducts, give strong signals in tandem MS P , 0.05, corresponding to 19.6% FDR, Supplementary Data), LACC1KD/scramble
spectra, and reveals the position of various acyl chains because of a stronger signal control (serum samples at ZT 16, P , 0.05, FDR 5 17%) and adPPARd/adGFP
in the tandem MS coming from the loss of the acyl chain in the sn-1 position over (liver lysates, P , 0.05, FDR 5 11.3%) comparisons, respectively.
the sn-2 position. The QQQ-MS was operated in multiple reaction monitoring Metabolites set enrichment analysis (MSEA). Significantly altered features in
mode (MRM), targeting lithium adduct precursors and product ions. The MRM the adPPARd/adGFP liver lysate comparison were subjected to database search to
transitions and parameters for PC(18:0/18:1) or PC(18:1/18:0) are listed in assign putative identities. Among those, 26 were matched to human metabolites
Extended Data Fig. 3g. Mobile phase was comprised of 98:2 methanol:water with database (HMDB) (Extended Data Table 1). The mapped species were assigned a
1 mM LiCl to facilitate the formation of lithium adducts for analysis. Samples were HMDB ID for subsequent MSEA analysis implemented in the Metaboanalyst39.
run in positive ionization mode with fragmentor voltage of 150 V, collision energy Statistical test. Power. Owing to the multitude of measurements on each animal
of 35 V and a dwell time of 25 ms. cohort, it is not feasible to pre-determine a sample size that achieves the same
Quantification of phosphocholine species by stable isotope dilution mass spec- power of all subsequent measurements. Therefore, we determined the minimal
trometry. 200 pmol of 1,2-distearoyl(d70)-sn-glycero-3-phosphocholine-1,1,2,2- number of animals required to detect a pre-defined difference in serum TG, a key
d4-N,N,N-trimethyl-d9 (D83 PC(18:0/18:0)) was spiked into 50 ml of serum as the readout throughout the study. Our pilot studies in wild-type mice have indicated
recovery standard. Serum was extracted as above. LC-MS/MS analysis was per- that to detect an effect size of 50% reduction in serum TG with a power of 80%, 3–8
formed using an Agilent 6410 QQQ-MS in positive ionization mode equipped mice are required per group, depending on time of the day (as TG levels vary). We
with an electrospray source ionization interface and an Agilent 1200 Binary Pump. determined the actual number of animals used for each study based on the above
For LC analysis, a Gemini (Phenomenex) C18 column (50 mm 3 2.0 mm, 3-mm sample size estimation in conjunction with the feasibility of experimental approaches.
particle size with 100 Å pore) was used with a 50-mm steel mesh filter. Mobile Replication. Animal experiments were performed on multiple cohorts (Extended
phase A consisted of 95:5 water:methanol and mobile phase B consisted of 80:20 Data Table 3). In vitro experiments were performed at least 3 times.
isopropanol:methanol. Both A and B were supplemented with 0.1% formic acid. Randomization. The randomized block design was used for all animal experi-
The flow rate was 0.3 ml min21. The gradient started at 20% B and linearly ments. We identified the age, sex, body weight, cage effect and timing of experi-
increased to 100% B over 45 min, was maintained at 100% B for 10 min before ments as blocking factors. Therefore, all animal experiments were carried out on
equilibrating for 5 min at 20% B. The QQQ-MS was operated in MRM mode and age-matched animals of the same sex. Body weight was measured before assigning
PCs were targeted using the m/z [M 1 H]1 to m/z 281.2 transition for all PCs. treatment groups. Cage effect was controlled in pharmacological treatment studies
Capillary voltage was set to 3.0 kV, the fragmentor voltage to 200 V with a collision by randomly assigning animals to the placebo or treatment group from the same
energy of 35 V. The drying gas temperature was 350 uC, the drying gas flow was cage. To control for the timing of experiments, alternating genotypes were drawn
10 l min21 and the nebulizer pressure was 45 p.s.i. The integrated peak area for for each measurement. Subsequent assays (gene expression, PC(18:0/18:1) con-
each species was normalized to the peak area of the recovery standard. centration measurement, etc) were performed in a blinded fashion, that is, every
Data analysis (Extended Data Fig. 6). Data preprocessing. Raw data files were sample was assigned a number without genotype or treatment labelled and the
converted to mzXML files and subsequently aligned by XCMS35. The resulting assays were performed sequentially based on the sample number. Often, samples
aligned features derived from wild-type, LPPARDKO, Scramble and LACC1KD were intercalated from different groups.
serum were compared to identify common features using metaXCMS36 with a Sample exclusion and statistical tests. Pre-determined sample exclusion cri-
mass tolerance of 0.01 and retention time tolerance of 60 s. Identical procedures terion was established for technical failures. In addition, the 1.5 inter-quartile
were carried out to generate common features from adPPARd and adGFP liver range rule was used to exclude additional outliers. Two-tailed unpaired student’s
lysates. Subsequently, these features from serum and liver lysates samples were t-test was used to compare two groups/treatments for experiments considered
processed by an automated workflow37 to identify isotopic peaks and assign putat- normal distribution (for example, cultured cells). For time-series data, the two-
ive identity with 3 p.p.m. mass tolerance. All isotopic peaks were removed and the way ANOVA procedure was used. For metabolomics data analysis, the methods
remaining data were cutoff for features with median intensity less than 5 3 104. are detailed in metabolomics data analysis section. Equal variance among groups
The reproducibility of the untargeted metabolomics platform was evaluated from was assumed.
two independent runs of 6 samples. The Spearman’s rank correlation coefficient 30. Ronnebaum, S. M. et al. Chronic suppression of acetyl-CoA carboxylase 1 in b-cells
was calculated and the duplicate pair with lowest correlation coefficient was plot- impairs insulin secretion via inhibition of glucose rather than lipid metabolism.
ted (Extended Data Fig. 5a). J. Biol. Chem. 283, 14248–14256 (2008).
Data normalization. We adapted methods from ref. 38. Briefly, each sample was 31. Lobo, S., Wiczer, B. M. & Bernlohr, D. A. Functional analysis of long-chain acyl-CoA
synthetase 1 in 3T3-L1 adipocytes. J. Biol. Chem. 284, 18347–18356 (2009).
centred by median and scaled by its inter-quartile range (IQR). The normalized
32. Agren, J. J., Julkunen, A. & Penttila, I. Rapid separation of serum lipids for fatty acid
distributions of samples were plotted in Extended Data Fig. 5b as box-and-whisker analysis by a single aminopropyl column. J. Lipid Res. 33, 1871–1876 (1992).
plot. 33. Kang, K. et al. Adipocyte-derived Th2 cytokines and myeloid PPARd regulate
Hierarchical clustering. Both positive and negative ionization mode features macrophage polarization and insulin sensitivity. Cell Metab. 7, 485–495 (2008).
from wild-type and LPPARDKO serum around the clock were mean-centred and 34. Spandidos, A., Wang, X., Wang, H. & Seed, B. PrimerBank: a resource of human and
mouse PCR primer pairs for gene expression detection and quantification. Nucleic
scaled by standard deviation on a per-feature basis (auto-scaling). To simplify the
Acids Res. 38, D792–D799 (2010).
visualization, only the mean value of each feature from every time point was used 35. Smith, C. A., Want, E. J., O’Maille, G., Abagyan, R. & Siuzdak, G. XCMS: processing
for the construction of heat map. The resulting data sets of each genotype were mass spectrometry data for metabolite profiling using nonlinear peak alignment,
clustered using Euclidean distance as the similarity metric in Cluster 3.0. Heat matching, and identification. Anal. Chem. 78, 779–787 (2006).
maps were generated by Java TreeView. Heat map of LPPARDKO serum was 36. Tautenhahn, R. et al. metaXCMS: second-order analysis of untargeted
aligned to wild type for comparison. Dendrogram of samples was plotted based on metabolomics data. Anal. Chem. 83, 696–700 (2011).
37. Brown, M. et al. Automated workflows for accurate mass-based putative
Spearman correlation with Ward linkage. metabolite identification in LC/MS-derived metabolomic datasets. Bioinformatics
Principal component analysis. Auto-scaling was applied on a per metabolite 27, 1108–1112 (2011).
basis to each biological group (wild type versus LPPARDKO and Scramble versus 38. Sreekumar, A. et al. Metabolomic profiles delineate potential role for sarcosine in
LACC1KD). Principal component analysis was performed in Metaboanalyst39. prostate cancer progression. Nature 457, 910–914 (2009).
The three-dimensional view of the first 3 principal components was plotted. In 39. Xia, J. & Wishart, D. S. Web-based inference of biological patterns, functions and
pathways from metabolomic data using MetaboAnalyst. Nature Protocols 6,
addition, score plot of the first and third principal components, showing the 743–760 (2011).
separation between sample groups and the loading plot of these two principal 40. Saeed, A. I. et al. TM4 microarray software suite. Methods Enzymol. 411, 134–193
components were generated (Extended Data Fig. 3c, d). (2006).

©2013 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Figure 1 | Analyses of liver lipid metabolites altered by

PPARd overexpression. a, Metabolite set enrichment analysis (MSEA) of
lipids from adGFP and adPPARd liver lysates (n 5 4). Metabolites were
identified based on database search of matching mass-charge ratio and
retention time. Identified metabolites and their relative quantity were used to
calculate the enrichment and statistical significance. Top 30 perturbed enzyme
or pathways were shown. List of metabolites recognized by the Metaboanalyst
program and subsequently used for the MSEA analysis is shown in Extended
Data Table 1. b, Correlation of hepatic PPARD and ACC1 expression in human
liver. Human liver gene expression microarray data was downloaded from gene
expression omnibus (GSE9588) and analysed using GraphPad Prism.
*P , 0.05 (t-test).

©2013 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Figure 2 | Molecular clock expression, food intake and feeding (n 5 3, each time point). Red bar, time when food was available. d, Food
glucose metabolism in wild-type and LPPARDKO mice. a, Liver gene intake in wild-type and LPPARDKO mice measured by metabolic cages
expression in wild-type and LPPARDKO mice (n 5 4, each time point). White (n 5 8). e, glucose tolerance test and insulin tolerance test in wild-type (n 5 6)
bar, light cycle starting at ZT 4; black bar, dark cycle. b, Ppard and Bmal1 and LPPARDKO (n 5 7) mice. f, Comparison of liver and serum lipidomes.
expression in dexamethasone-synchronized primary hepatocytes (n 5 3, each g, Column purification of serum lipids (See methods for detail). IPA, isopropyl
time point). Circadian time, hours after dexamethasone treatment. c, Gene alcohol; MeOH, methanol; HOAc, acetic acid. Data presented as mean 6 s.e.m.
expression in wild-type and LPPARDKO livers under daytime restricted

©2013 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Figure 3 | Identification and characterization of wild-type and LPPARDKO serum. Bottom panels, EIC of m/z 5 788.6 in
PC(18:0/18:1), or SOPC. a, Heat map of identified features in wild-type and LACC1KD serum and adPPARd livers. f, Normalized PC(36:1) intensity in
LPPARDKO serum under daytime feeding (n 5 3, each time point). White bar, wild-type and LPPARDKO mouse serum (n 5 4) under ad libitum or daytime
light cycle starting at ZT 0; black bar, dark cycle; red bar, time when food restricted feeding (DF). g, Top, multiple reaction monitoring (MRM)
was available. b, Dendrogram of serum samples under daytime restricted parameters for identification of acyl-chain composition of PC(36:1). Bottom
feeding. c, Principal component analysis (PCA) of positive mode features in left, co-elution of the PC (18:0/18:1) standard with m/z 5 788.6. Bottom right,
wild-type, LPPARDKO, Scramble and LACC1KD serum under ad libitum PC(36:1) acyl-chain composition determined by tandem mass spectrometry
feeding. Top, score plot of the first three PCs representing 53.2% of the total running in the MRM mode. h, Top panels, lipid levels in mice intraperitoneally
variation. Bottom, score plot of PC1 and PC3. Circle, 95% confidence interval. injected with various doses of PC(18:0/18:1) (n 5 4). Bottom, in vivo FA uptake
d, Loading plot of the PCA. The putative identities of 11 features identified in soleus muscle (left) and serum PC(36:1) enrichment (right) 4 h after
in Fig. 3d are shown in red. Additional top features contributing to the PC(18:0/18:1) injection at 5mg kg21 body weight. *P , 0.05 (t-test), data
segregation are highlighted in blue. e, Top panels, EIC of m/z 5 788.6 in presented as mean 6 s.e.m.

©2013 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Figure 4 | Requirement of hepatic PPARd and muscle with 100 mM PC(18:0/18:1). f, Heat map showing serum phospholipid changes
PPARa for the inter-organ communication mediated by PC(18:0/18:1)/ between ZT 20 and ZT 8 in 7-month-old male C57BL/6J mice on chow (n 5 3)
SOPC. a, Cd36 gene expression in muscle of wild-type and LPPARDKO mice or high-fat diet (HFD for 4 months, n 5 5) from targeted metabolomics.
under daytime restricted feeding (n 5 3, each time point). Red bar, time when g, Serum PC(36:1) concentrations under chow or HFD. h, Blood glucose levels
food was available. #P , 0.05 (ANOVA). b, Effects of GW501516 on serum TG of ad libitum fed db/db mice measured between ZT 0 and ZT 3 before daily
and muscle FA uptake in wild-type and LPPARDKO mice (n 5 5). c, Cd36 and lipids injections (vehicle, n 5 4; PC(18:0/18:1), n 5 5). i, Model for the role of
Fabp3 gene expression in C2C12 myotubes treated with vehicle or 25 mM PPARd–PC(18:0/18:1)–PPARa signalling in FA synthesis and use in the liver–
PC(18:0/18:1) (n 5 3). d, FA uptake in control or stable Cd36 knockdown muscle axis. j, Top panel, in vivo fatty acid uptake in soleus and gastrocnemius
C2C12 myotubes pretreated with indicated lipids. e, The mammalian one- muscle 4 h after vehicle or 5 mg kg21 PC(16:0/18:1) injection though the tail
hybrid assay (diagram shown on the top) to determine the transactivation vein (n 5 6); bottom panel, muscle Cd36 and Fabp3 gene expression after
activity of the PPAR ligand binding domain (LBD) (n 5 3). Left panel, relative PC(16:0/18:1) injection (n 5 4). k, Top panel, activities of a PPRE-containing
luciferase unit (RLU, presented as fold change) indicative of the reporter luciferase reporter in PPARa-expressing C2C12 cells treated with vehicle,
activity regulated by Gal4 DNA binding domain (DBD)-PPARaLBD fusion 50 mM PC(18:0/18:1) or PC(16:0/18:1) and 1 mM GW7647 (a PPARa synthetic
protein (Gal4-PPARaLBD) in 293 cells treated with indicated phospholipids at ligand). Bottom panel, Cd36 expression in C2C12 myotubes. *P , 0.05, (t-test),
100 mM. Right panel, RLU of Gal4-PPARdLBD and Gal4-PPARcLBD treated data presented as mean 6 s.e.m.

©2013 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Figure 5 | Validation of metabolomics analyses. a, The normalization with median centering and inter-quartile range (IQR) scaling.
reproducibility of the untargeted metabolomics platform was validated from The resulting data show equal distribution among different groups of samples.
two separate runs of 6 serum samples. The Spearman’s rank correlations are White bar represents samples obtained in the light cycle and black bar for those
between 0.9 and 0.94. The duplicate pair with the lowest correlation in the dark cycle.
(Spearman’s r 5 0.90) is shown. b, The raw intensity of samples was subject to

©2013 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Figure 6 | Flow chart of metabolomics data analysis showing the positive-mode metabolites. See methods for detailed description.

©2013 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Table 1 | List of metabolites recognized by Metaboanalyst program and subsequently used for MSEA analysis

©2013 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Table 2 | Metabolic characteristics of mouse models

(I) Wild-type and LPPARDKO mice under normal chow diet (male, 8–10 weeks old, n 5 6); (II) vehicle- (n 5 4) and PC(18:0/18:1)-treated (n 5 5) FVB/NJ-db/db mice. *P , 0.05, two-tailed t-test. Data presented as
mean 6 s.e.m.

©2013 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Table 3 | List of animal cohorts used for this study

ED, Extended Data.

©2013 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Table 4 | List of primers used for qRT–PCR and oligonucleotides for short hairpin RNA constructs

©2013 Macmillan Publishers Limited. All rights reserved