While feast and famine cycles illustrate that adipose tissue remodelling in response
to fluctuations in nutrient availability is essential for maintaining metabolic homeostasis,
the underlying mechanisms remain poorly understood
1,2
. Here, we identify fibroblast growth factor (FGF) 1 as a critical transducer in this
process and link its regulation to the nuclear receptor (NR) PPARγ, the adipocyte
master regulator and target of the thiazolidinedione (TZD) class of insulin sensitizing
drugs
3–5
. FGF1 is the prototype of the 22 member FGF family of proteins and has been implicated
in a range of physiological processes including development, wound healing and cardiovascular
changes
6
. Surprisingly, FGF1 knockout mice display no significant phenotype under standard
laboratory conditions
7–9
. We show that FGF1 is highly induced in adipose tissue in response to high-fat diet
(HFD) and that mice lacking FGF1 develop an aggressive diabetic phenotype coupled
to aberrant adipose expansion when challenged with HFD. Further analysis of adipose
depots in FGF1 deficient mice revealed multiple histopathologies in the vasculature
network, an accentuated inflammatory response and aberrant adipocyte size distribution.
Upon HFD withdrawal, this inflamed adipose tissue fails to properly resolve resulting
in ectopic expression of pancreatic lipases and extensive fat necrosis. Mechanistically,
we show that adipose induction of FGF1 in the fed state is regulated by PPARγ acting
through an evolutionarily conserved promoter proximal PPAR response element within
the FGF1 gene. This work describes the first phenotype of the FGF1 knockout mouse
and establishes the PPARγ-FGF1 axis as critical for maintaining metabolic homeostasis
and insulin sensitization.
As part of a directed screen to identify genes that respond to dietary cues in metabolic
tissues (muscle, liver, brown adipose (BAT) and white adipose tissue (WAT)), we observed
that FGF1 is selectively induced in visceral (i.e. gonadal) WAT (gWAT) in response
to a high-fat diet (HFD), pointing to a possible metabolic function (Fig. 1a, Supplementary
Fig. 1a, b). Subfractionation of adipose depots revealed that FGF1 was expressed in
the adipocyte fraction but not in the stromal vascular fraction (SVF) of gWAT, and
to a lesser extent in the adipocyte fraction of subcutaneous (i.e. inguinal) WAT (iWAT)
(Fig. 1b). Given that FGF1 gene transcription is directed by at least three distinct
promoters that are conserved across mammals
10
(Fig. 1c), we examined the tissue-specific expression patterns of each isoform. While
FGF1A and FGF1B were both expressed in gWAT (as well as other tissues) of mice fed
a standard chow diet (Fig. 1d–f), only FGF1A showed a striking and progressive 20-fold
induction between fasted, fed, and HFD exposure (Fig. 1g, h). While no metabolic role
has been attributed to FGF1, these results prompted us to reconsider this possibility.
Consistent with previous reports, no metabolic or histological abnormalities, or major
gene expression changes were observed in FGF1−/−
mice fed a standard chow diet
7
(Supplementary Fig. 1c, 2, Supplementary Table 3). Similarly, when placed on a HFD,
FGF1−/−
and wild-type cohorts showed equivalent changes in weight (monitored for 24 weeks),
serum adipokines, cytokines (leptin, resistin, IL-6, TNFα, tPAI-1, total and HMW adiponectin)
as well as serum lipids (cholesterol, free fatty acids, and triglycerides) (Fig. 1i,
Supplementary Tables 1 and 2). However, FGF1−/−
mice developed an exaggerated diabetic phenotype, with increased levels of fasting
glucose and insulin (Fig. 2a), accompanied by severe insulin resistance (Fig. 2b,
c) and markedly enhanced serum MCP-1/CCL2 (48.0 ± 3.4 compared to 59.0 ± 3.4 pg/ml,
P < 0.01, in wild-type and FGF1−/−
mice, respectively), a marker of adipose macrophage infiltration and a causative factor
for peripheral insulin resistance
11,12
. To further investigate the role of FGF1 in insulin sensitivity, we performed hyperinsulinemic-euglycemic
clamp studies. The steady-state glucose infusion rate (GIR) during the clamp was about
40% lower in HFD-fed FGF1−/−
mice, reflecting decreased insulin responsiveness, and was accompanied by reductions
in whole body and insulin-stimulated glucose disposal rates (GDR and IS-GDR) indicating
pronounced peripheral insulin resistance (Fig. 2d–f). The ability of insulin to suppress
hepatic glucose production (HGP) was also significantly compromised in HFD-fed FGF1−/−
mice, revealing hepatic insulin resistance as well (Fig. 2g). Although HFD-fed FGF1−/−
mice had enlarged steatotic livers relative to wild-type controls (Fig. 2h, i), liver
function (based on serum ALT levels) and pancreatic function (based on islet histology
and insulin secretion) appeared normal (Supplementary Fig. 3, 4). On the other hand,
gWAT in FGF1−/−
mice failed to expand after HFD and exhibited pronounced structural abnormalities
as evidenced by hematoxylin and eosin staining (Fig. 2j, k, Supplementary Fig. 5).
As expected with higher circulating levels of MCP-1, we observed a dramatic increase
in macrophage infiltration, indicating a highly inflamed gWAT (Fig. 3a). Notably,
no defects were observed in iWAT (data not shown). This differential sensitivity of
WAT depots to HFD stress is consistent with the known association of the visceral
WAT with obesity-related pathologies including insulin resistance
13
.
To explore the possibility that the metabolic dysregulation observed in FGF1−/−
mice on a HFD was associated with defects in gWAT, we performed detailed histological
and molecular analyses of this tissue. Histochemistry with Masson’s trichrome stain
of gWAT from the HFD-fed FGF1−/−
mice revealed increased collagen deposition (blue staining) and a marked heterogeneity
in adipocyte size (Fig. 3b). Quantification of adipocyte cross-sectional areas showed
increases in the numbers of both small and large adipocytes in FGF1−/−
mice relative to wild-type (Fig. 3c). Next we examined the functional architecture
of the adipose vasculature by intravenous injection of fluorescent microbeads. Epifluorescence
microscopy of tissue sections from fluorescent bead-perfused mice revealed decreased
vascular density specifically in gWAT (Fig. 3d) but not in BAT or iWAT of HFD-fed
FGF1−/−
mice (Supplementary Fig. 6). Microarray analyses identified multiple transcriptional
changes induced by HFD in FGF1−/−
gWAT that were consistent with the observed phenotypes. The obesity and insulin resistance
markers RBP4 and CCL11 were upregulated by HFD, as was the angiogenic factor VEGF.
Interestingly, expression of PPARγ was also induced by HFD in FGF1−/−
gWAT, as were known PPARγ target genes (e.g. perilipin, DGAT1, DGAT2) (Supplementary
Table 3). However, the most dramatic inductions of gene expression were seen in multiple
fat necrosis-associated pancreatic lipases and the tissue remodelling factor elastase
1, which were confirmed by qPCR (Fig. 3e, Supplementary Table 3)
14–16
. Taken together, the observed histopathological and molecular changes in FGF1−/−
gWAT suggested a failure to execute the appropriate adipose remodelling program in
response to HFD stress. As adipose tissue needs to dynamically expand and contract
with fluctuations in nutrient availability, we postulated that FGF1 would also play
a role in its contraction capacity upon withdrawal from HFD. To test this, we re-adapted
HFD-fed wild-type and FGF1−/−
mice to 6 weeks of chow feeding. Gross examination of HFD-to-chow converted (HCC)
mice revealed features consistent with maladaptation resulting in disfigurement and
discoloration of gWAT from FGF1−/−
mice compared to control wild-type mice (Fig. 3f). Histological examination of the
HCC FGF1−/−
gWAT showed profound degeneration of adipose architecture and integrity of far greater
severity than the HCC wild-type gWAT (Fig. 3g). Indeed, HCC FGF1−/−
mice frequently presented with what appeared to be fragments of dissociated fat tissue
within the peritoneal cavity, which upon histological analysis were consistent with
a fat necrosis pathology (Fig. 3h). The observations from the HFD and HCC regimens
demonstrate a dynamic requirement for FGF1 in both adipose tissue expansion and contraction.
Together our findings show that when challenged with a HFD or HCC, FGF1−/−
mice are unable to remodel visceral adipose tissue in response to dietary changes.
This suggests that defects in adipose plasticity, attributable to the loss of FGF1,
are causally linked with a series of peripheral pathologies including hepatic steatosis
and systemic insulin resistance. These results establish FGF1 as a transducer of adipose
remodelling in response to nutrient fluctuations, and identify an indispensible role
for FGF1 in defending the body against metabolic disease.
As PPARγ expression was induced in FGF1−/−
gWAT by HFD, and HFD elevates the levels of circulating PPAR ligands
17,18
, we postulated that the HFD-induction of FGF1A may be regulated by a PPAR family
member. Luciferase reporter assays indeed revealed a robust induction of the human
FGF1A promoter by the PPARs, of which PPARγ was the most potent (Fig. 4a). Furthermore,
in this system the PPAR induction of FGF1 appears isoform specific, as PPARs did not
induce FGF1B, the other FGF1 isoform expressed in WAT (Fig. 4b). Examination of the
promoter region of FGF1A revealed a highly conserved (98% conservation) region ~100
bp proximal to the transcription start site (TSS) (Supplementary Fig. 7a) containing
a conserved putative PPAR response element (PPRE) at −60 bp relative to the TSS (Fig.
4c). Inactivation of this PPRE using site directed mutagenesis resulted in a complete
loss of its response to PPARγ (Fig. 4c, d: compare human vs ΔPPRE). To examine the
functional conservation of FGF1 regulation by PPARγ, we assayed the native human and
mouse FGF1A promoters along with reporter constructs containing orthologous PPREs
(rat, horse, opossum) introduced into the human FGF1A promoter. PPARγ activation was
observed for all promoters except for the more distantly related opossum PPRE (Fig.
4d). Chromatin immuno-precipitation (ChIP) experiments confirmed that PPARγ does indeed
bind to the identified PPRE in 3T3-L1 adipocytes (Fig. 4e, Supplementary Fig. 7b).
Data mining of a published genome-wide PPARγ ChIP-Seq study indicates that the PPARγ-FGF1A
interaction may be specific to WAT, as PPARγ binding was seen in 3T3-L1 cells, but
not in macrophages
19
. Together, these findings show that the adipocyte PPARγ-FGF1 axis is functionally
conserved in a wide range of mammals.
To confirm the physiological relevance of the PPARγ induction of FGF1, we determined
the expression of the FGF1A transcript in mice in response to the potent and specific
PPARγ ligand rosiglitazone (TZD). We found that oral administration of rosiglitazone
(5 mg/kg for 3 days) significantly increased the mRNA levels of FGF1A in gWAT, in
fed but not fasted states (Fig. 4f). FGF1B and FGF1D expression was unchanged by rosiglitazone
in gWAT and liver (Fig 4g; Supplementary Fig. 8) whereas FGF1A and FGF1D were undetectable
in liver and gWAT, respectively (data not shown). Further, adipocyte-specific PPARγ
knockout mice
20
displayed decreased levels of FGF1 in the adipocyte fraction, without compensatory
changes in the closely related FGF2, which was only detected in the SVF (Fig. 4h).
In conclusion, we have discovered an unexpected metabolic role for FGF1 as a critical
transducer of PPARγ signalling that mediates the proper coupling of nutrient storage
to adaptive remodelling of adipose tissue. We found that HFD results in potent and
selective induction of FGF1 in adipose tissue and that its transcription is controlled
by PPARγ and its insulin sensitizing ligands. Loss of FGF1 leads to systemic metabolic
dysfunction and insulin resistance, revealing an indispensable role for FGF1 in metabolic
homeostasis (Fig. 4i). Importantly, we show that FGF1−/−
mice are unable to both properly expand and contract their adipose tissue in response
to dietary changes, revealing the dynamic requirement of FGF1 for adipose tissue remodelling.
The capacity of adipose tissue to remodel is crucial for accommodating changes in
energy availability in fasted and fed states but is not unlimited, and can become
perturbed in obesity and related pathologies. Previous reports have shown that FGF1
can signal through FGFRs to pre-adipocytes
21–23
. Recently, several endocrine FGF family members (FGF15/19, 21) have been linked to
metabolic homeostasis through NR regulation
24–26
. We now expand this NR-FGF interface to include the paracrine FGF1. Our discovery
of the PPARγ-FGF1 axis leads us to consider the therapeutic potential of FGF1 in potentially
mediating insulin sensitization without provoking the full range of adverse events
associated with PPARγ activation.
Methods summary
FGF1−/−
mice and age and sex-matched wild-type controls (>99% C57BL/6 genetic background)
received a standard diet or high fat (60%) diet (F3282, Bio-Serv) and water ad libitum.
For the HFD-to-chow conversion diet regimen (HCC), mice were fed HFD starting from
6 weeks of age. After 9 months of HFD, the diet was converted back to standard laboratory
chow for 6 weeks. Glucose tolerance tests (GTT) and Insulin tolerance tests (ITT)
were conducted after overnight and 5 hour fasting, respectively. Glucose (1g/kg i.p.)
or insulin (0.5 U insulin/kg i.p.) was injected and blood glucose monitored. Serum
analyses were performed on blood collected by tail bleeding either in the ad libitum
fed state or following overnight fasting. Free fatty acids, triglycerides, cholesterol
and ALT were measured using commercial enzymatic colorimetric kits. Serum insulin
levels and total and high-molecular weight (HMW) adiponectin levels were measured
by ELISAs using commercial kits. Plasma adipokine levels were measured using a Milliplex™
MAP kit. Histological and immunohistochemical analyses were performed on sections
of fixed tissues according to standard procedures.
Vasculature was visualized by tail vein injection of fluorescent microbeads (0.1 μm
FluoSpheres®, Molecular Probes). Subsequently, mice were anesthetized and perfused
through the heart with additional fluorescent microbeads. Tissues were dissected,
embedded and 10 μm frozen sections analyzed for blood vessel density using fluorescence
microscopy.
Luciferase reporter assays were performed in CV-1 cells treated overnight with or
without ligands (PPARα, 1 μM WY14643; PPARγ, 1 μM Rosiglitazone (TZD); PPARδ, 100
nM GW1516). Site-directed mutagenesis of the PPRE in the human promoter was performed
using a QuikChange II kit.
Chromatin immunoprecipitation assays were performed on differentiated 3T3-L1 cells.
Sheared chromatin generated from cross-linked cell pellets by sonication was incubated
with PPARγ antibody or control rabbit IgG, and precipitated with pre-blocked protein
A-agarose beads.
Methods
Animals
FGF1−/−
mice
7
and age and sex-matched wild-type controls (>99% C57BL/6 genetic background) received
a standard chow diet (MI laboratory rodent diet 5001, Harlan Teklad) or high fat (60%)
diet (F3282, Bio-Serv) and water ad libitum. PPARγfl/fl mice were crossed with aP2-Cre
mice to generate aP2-Cre; PPARγfl/fl
mutant mice as previously described
20
and received standard chow up to analysis at five months of age. All mice used for
studies were male unless otherwise noted.
Reporter assays
Luciferase reporter assays were performed in CV-1 cells treated overnight with or
without ligands (PPARα, 1 μM WY14643; PPARγ, 1 μM Rosiglitazone (TZD); PPARδ, 100
nM GW1516). Site directed mutagenesis of the PPRE in the human promoter was performed
using a QuikChange II kit (Stratagene, La Jolla, CA).
Western analysis
Total cell lysates prepared in 50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% Triton-X100,
0.1% SDS, 2 mM sodium azide and protease inhibitor cocktail (Complete, Roche), were
resolved by SDS-PAGE and probed using primary antibodies to FGF1, FGF2 (Santa Cruz)
and ERK1/2 (Cell Signaling Technology).
Serum analysis
Blood was collected by tail bleeding either in the ad libitum fed state or following
overnight fasting. Free fatty acids (Wako), triglycerides (Thermo), cholesterol (Thermo)
and ALT (Thermo) were measured using enzymatic colorimetric methods. Serum insulin
levels (Ultra Sensitive Insulin, Crystal Chem) and total and high-molecular weight
(HMW) adiponectin levels (ALPCO) were measured by ELISAs. Plasma adipokine levels
were measured using a Milliplex™ MAP kit (Millipore).
Histological analysis and immunohistochemistry
4 μm sections of fixed tissues were stained with hematoxylin and eosin according to
standard procedures. For immunohistochemistry, tissues were deparaffinized in xylene
and rehydrated. Slides were incubated with 5% normal donkey serum for 30 min, followed
by overnight incubation with primary and secondary antibodies (F4/80, Abcam, 1:100).
Adipocyte Size Analysis
Adipocyte cross-sectional area was determined from photomicrographs of gonadal, mesenteric,
and inguinal fat pads using ImageJ
27
.
Adipose tissue fractionation
Adipose tissues were excised and finely minced with a razor blade. Minced tissue was
digested in adipocyte isolation buffer (100mM HEPES pH7.4, 120mM NaCl, 50mM KCl, 5mM
glucose, 1mM CaCl2, 1.5% BSA) containing 1mg/ml collagenase at 37°C with constant
slow shaking (~120rpm) for 2 hours. During the digestion period, the suspension was
gently mixed several times. The suspension was then passed through a 200μm mesh and
100μm successively. The flowthrough was allowed to stand for 15 min to separate the
floating adipocyte fraction and infranatant containing the stromal vascular fraction.
The infranatant was removed and saved while minimally disturbing the floating adipocyte
fraction. Both fractions were centrifuged at 500xg for 10 minutes and further washed
twice in DMEM/Ham’s F-12 media before further manipulation.
Metabolic studies
Glucose tolerance tests (GTT) and Insulin tolerance tests (ITT) were conducted after
o/n and 5 hour fasting, respectively
28,29
. Glucose (1g/kg i.p.) or insulin (0.5 U insulin/kg i.p.) was injected and blood glucose
monitored using a OneTouch Ultra glucometer (Lifescan Inc).
Hyperinsulinemic-euglycemic clamp
Mouse clamps were performed as previously described
27,28
. Briefly, mice implanted with dual jugular catheters 3 days prior were fasted for
6 hr, then equilibrated with tracer (5.0 μCi/h, 0.12 ml/h [3-3H]D-glucose, NEN Life
Science Products) for 90 min. A basal blood sample was then drawn via tail vein to
calculate basal glucose uptake. The insulin (8 mU/kg/min at 2 μl/min, Novo Nordisk)
plus tracer (5.0 μCi/h) and glucose (50% dextrose at variable rate, Abbott) infusions
were initiated simultaneously, with the glucose flow rate adjusted to reach a steady
state blood glucose concentration (~120 min). Steady state was confirmed by stable
plasma tracer counts during the final 30 min of clamp. Blood was taken at 110 and
120 minutes for the determination of tracer-specific activity. At steady state, the
rate of glucose disappearance or the total GDR is equal to the sum of the rate of
endogenous or HGP plus the exogenous GIR. The IS-GDR is equal to the total GDR minus
the basal glucose turnover rate.
Gene Expression Analysis
Total RNA was isolated from mouse tissue and cells using TRIzol reagent (Invitrogen).
cDNA was synthesized from 1 μg of DNAse-treated total RNA using SuperScript II reverse
transcriptase (Invitrogen). mRNA levels were quantified by QPCR with SYBR Green (Invitrogen).
Samples were run in technical triplicates and relative mRNA levels were calculated
by using the standard curve methodology and normalized against 36B4 mRNA levels in
the same samples.
Microarrays
500 ng of total RNA extracted from gWAT (24 wk old wild-type and FGF1−/−
mice on chow or HFD) using Trizol reagent (Invitrogen) was reverse transcribed into
cRNA and biotin-UTP labeled using the Illumina TotalPrep RNA Amplification Kit (Ambion).
cRNA was hybridized to the Illumina mouseRefseq-8v2 Expression BeadChip using standard
protocols (Illumina). Image data was converted into unnormalized Sample Probe Profiles
using the Illumina BeadStudio software and analyzed by VAMPIRE. Stable variance models
were constructed for each experimental conditions (n=2). Differentially expressed
probes were identified (unpaired VAMPIRE significance test with a 2-sided, Bonferroni-corrected
threshold of α
Bonf = 0.05) and the significance of apparent differences between 2 experimental conditions
determined. Lists of altered genes were mapped to pathways (VAMPIRE tool Goby) to
determine KEGG categories overrepresented (Bonferroni error threshold of α
Bonf = 0.05).
Chromatin immunoprecipitation assay (ChIP)
Adipocytes differentiated from 3T3-L1 cells
29
(ATCC) were sequentially cross-linked with 2mM Disuccinimidyl glutarate (30 min at
RT) and 1% formaldehyde (10 min at RT). Crosslinking was quenched with glycine, and
the washed cell pellet frozen at −80°. Cell pellets were lysed and centrifuged at
12,000 × g for 1min at 4°C. Sheared chromatin generated from cell pellets by sonication
was incubated with PPARγ antibody (2μg sc-7196, overnight at 4°C) or control rabbit
IgG (sc-2027, Santa Cruz Biotechnologies). The immuno-complexes were precipitated
with 20μl pre-blocked protein A-agarose beads (1h at 4°C) and washed extensively
30
.
Input DNA isolation, DNA de-crosslinking, purification and analysis
Sheared input chromatin was ethanol precipitated and Chelex 100 (100μl of 10% slurry,
Bio-Rad) added to both input DNA and washed ChIP samples. Samples were vortexed, boiled
for 10min and then centrifuged (12,000 × g for 1min). Samples were treated with proteinase
K (1μl of 20mg/ml, 55°C for 30min), heat deactivated, and the DNA purified (Qiagen
MinElute PCR Purification Kit) prior to qPCR analysis using the following primers:
FGF1 Fw 5′-AGAGTAGGGCACAGACACAGC-3′
FGF1 Rev 5′-TGGATTAGACACGCAGGCTA-3′
aP2 Fw 5′-ATTTGCCTTCTTACTGGATCAGAGTT-3′
aP2 Rev 5′-TTGGGCTGTGACACTTCCAC-3′
Angpl4 Fw 5′-CCAGCCAGGGAAAGTAGGAGA-3′
Angpl4 Rev 5′-CAGAAAGTGCCTGCATGCC-3′
36b4 Fw 5′-GCCAATAGACGCGCATGTTT-3′
36b4 Rev 5′-TGGTTCCATCGACTGTCCTG-3′
Fluorescent microbead perfusion for vasculature studies
Mice were tail vein injected with 150 μl of PBS fluorescent microbeads (0.1 μm red
fluorescent microbeads, Invitrogen), anesthetized 5 minutes later, then perfused through
the heart with 6 ml of 1:10 PBS dilution of fluorescent microbeads. Tissues were then
dissected and embedded in Tissue-Tek®OCT compound (Sakura) and 10μm frozen sections
were mounted in Vectashield® medium (Vector Laboratories) for analysis of blood vessel
density using fluorescence microscopy
31
.
Statistical analysis
All values are given as means ± standard errors. The two-tailed unpaired Student’s
t-test was used to assess the significance of difference between two sets of data.
Differences were considered to be statistically significant when P < 0.05.
Supplementary Material
1
2