Retinoic Acid Receptor (cid:1) Stimulates Hepatic Induction of Fibroblast Growth Factor 21 to Promote Fatty Acid Oxidation and Control Whole-body Energy Homeostasis in Mice

Background: All- trans -retinoic acid ameliorates glucose intolerance and insulin resistance in diabetes. Results: Thegenetranscriptionofhepatic Fgf21 anditsmetaboliceffectsonlipidoxidation,ketogenesis,andwhole-bodyenergy expenditure are up-regulated by RAR (cid:1) activation. Conclusion: Hepatic RAR (cid:1) stimulates FGF21 production via the RARE. Significance: The hepatic RAR-FGF21 axis is a potential druggable target for treating metabolic syndrome. (cid:1) via a putative RARE site in the 5 (cid:2) -flanking region of the Fgf21 promoter. The recruitment of hepatic RAR (cid:1) to the putative RARE sequence of the Fgf21 gene may contribute to fasting-induced FGF21 production in vivo . Adenovirus-mediated overexpression of RAR (cid:1) in the liver persistently stimulates hepatic production and secretion of FGF21, which in turn enhances hepatic fatty acid oxidation and ketogenesis and increases whole-body energy expenditure in mice. The stimulatory effect of RA on fatty acid oxidation is diminished by siRNA-mediated knockdown of FGF21. These studies suggest that hepatic RAR (cid:1) plays an important role in regulating hepatic lipid metabolism and energy homeostasis, probably by inducing FGF21. These findings may provide the basis for a potential beneficial effect of pharmacological activation of RAR on metabolic disorders. ligand, induces expression of FGF21 in hepatocytes, which in turn simulates hepatic fatty acid oxidation and ketogenesis possibly through autocrine/paracrine actions of FGF21. The RAR/RXR heterodimer may sever as the main functional regulator transducing RA signaling to the transcriptional regulation of FGF21 via the putative RARE sequence. Hepatic FGF21 induction by RAR (cid:1) may result in enhanced systemic energy expenditure, possibly through increased hepatic (cid:1) -oxi- dation and/or some of endocrine effects of FGF21. The physiological regulation of FGF21 by RAR probably contributes to the adaptive response to nutrient deprivation.


Activation of retinoic acid receptor (RAR) with all-trans-retinoic acid (RA) ameliorates glucose intolerance and insulin
resistance in obese mice. The recently discovered fibroblast growth factor 21 (FGF21) is a hepatocyte-derived hormone that restores glucose and lipid homeostasis in obesity-induced diabetes. However, whether hepatic RAR is linked to FGF21 in the control of lipid metabolism and energy homeostasis remains elusive. Here we identify FGF21 as a direct target gene of RAR␤. The gene transcription of Fgf21 is increased by the RAR agonist RA and by overexpression of RAR␣ and RAR␤, but it is unaffected by RAR␥ in HepG2 cells. Promoter deletion analysis characterizes a putative RA-responsive element (RARE) primarily located in the 5-flanking region of the Fgf21 gene. Disruption of the RARE sequence abolishes RA responsiveness. In vivo adenoviral overexpression of RAR␤ in the liver enhances production and secretion of FGF21, which in turn promotes hepatic fatty acid oxidation and ketogenesis and ultimately leads to increased energy expenditure in mice. The metabolic effects of RA or RAR␤ are mimicked by FGF21 overexpression and largely abolished by FGF21 knockdown. Moreover, hepatic RAR␤ is bound to the putative RAREs of the Fgf21 promoter in a fastinginducible manner in vivo, which contributes to FGF21 induction and the metabolic adaptation to prolonged fasting. In addition to other nuclear receptors, such as peroxisome proliferator-activated receptor ␣ and retinoic acid receptor-related receptor ␣, RAR may act as a novel component to induce hepatic FGF21 in the regulation of lipid metabolism. The hepatic RAR-FGF21 pathway may represent a potential drug target for treating metabolic disorders.
All-trans-retinoic acid (RA), 3 a major active metabolite of vitamin A, plays essential roles in development, cellular differentiation, and survival through three retinoic acid receptors (RARs) of the nuclear receptor superfamily (1,2). RA is currently used for clinical cancer therapy (1). RAR is the ligand-dependent nuclear receptor that heterodimerizes with the retinoid X receptor (RXR) to transactivate its target genes by binding to gene promoters harboring retinoic acid response elements (RAREs) (1,2). RAR␤ is relatively more abundant in the liver compared with RAR␣ and RAR␥ (3). Recent studies indicate that hepatic retinoid signaling is impaired in humans with non-alcoholic fatty liver disease (4). Hepatic RAR receptor binding capacity is suppressed in high fat diet rats, accompanied by decreased expression of RAR␤ and unaltered expression of other RAR isoforms (5). Recently, several independent studies have demonstrated that administration of RA ameliorates obesity and glucose intolerance and suppresses adipose lipid stores in mouse models of obesity and diabetes (6 -8). However, the signaling molecules that mediate the metabolic actions of RAR remain largely unknown.
The recently discovered fibroblast growth factor 21 (FGF21) is a member of an atypical subfamily of FGFs, which includes FGF15/19 and FGF23, all of which circulate as hormones (9). FGF21 acts through cell surface receptors composed of classic FGF receptors complexed with ␤Klotho (10). In obese rodents or monkeys, pharmacological administration of FGF21 improves insulin sensitivity, normalizes plasma lipids levels, causes weight loss, and increases whole-body energy expendi-ture (11)(12)(13)(14). Thus, FGF21 is thought to be a potential drug candidate for treating metabolic disease. Hepatic FGF21 expression is highly induced in response to long term fasting, which in turn enhances the adaptive metabolic response (15). Although fasting-inducible hormone FGF21 has been previously shown to be regulated by the nuclear receptor peroxisome proliferator-activated receptor ␣ (PPAR␣) in the liver, it is important to note that fasting still causes a 5-fold increase in hepatic Fgf21 mRNA in PPAR␣ Ϫ/Ϫ mice (16), suggesting that a PPAR␣-independent mechanism may also contribute to the nutrient regulation of hepatic Fgf21 transcription. Recently, the nuclear receptor retinoic acid receptor-related receptor ␣ (ROR␣) is shown to play a role in regulating FGF21 in hepatocytes (17). FGF21 expression is also induced by feeding and PPAR␥ agonists in white adipose tissue as well as by cold exposure in brown adipose tissue (18,19). Therefore, the mechanisms underlying the transcriptional regulation of FGF21 are complex and not fully understood.
The aim of the present study is to test the hypothesis that hepatic RAR serves as a critical regulator that stimulates Fgf21 gene expression to control lipid and energy homeostasis, because anti-obesity phenotypes seen in ob/ob mice treated with RA, a natural RAR agonist (6), are markedly similar to those seen in transgenic mice expressing FGF21 in the liver (6 -8, 11). Here we identify FGF21 as novel target gene of RAR. Gene expression and promoter activity of Fgf21 are increased by RA, which is mimicked by overexpression of RAR␣ and RAR␤ but is unaffected by RAR␥ in HepG2 cells. RA stimulates the transcription of the Fgf21 gene in part through RAR␤ via a putative RARE site in the 5Ј-flanking region of the Fgf21 promoter. The recruitment of hepatic RAR␤ to the putative RARE sequence of the Fgf21 gene may contribute to fasting-induced FGF21 production in vivo. Adenovirus-mediated overexpression of RAR␤ in the liver persistently stimulates hepatic production and secretion of FGF21, which in turn enhances hepatic fatty acid oxidation and ketogenesis and increases whole-body energy expenditure in mice. The stimulatory effect of RA on fatty acid oxidation is diminished by siRNA-mediated knockdown of FGF21. These studies suggest that hepatic RAR␤ plays an important role in regulating hepatic lipid metabolism and energy homeostasis, probably by inducing FGF21. These findings may provide the basis for a potential beneficial effect of pharmacological activation of RAR on metabolic disorders.
Generation of Adenoviruses Expressing either FGF21 or RAR␤-Recombinant adenoviruses expressing mouse FGF21 and RAR␤ were generated using an AdEasy system (21). The cDNA clones for mouse FGF21 and RAR␤ were purchased from OriGene (catalog nos. MC204941 and MC205276). The full-length Fgf21 or RAR␤ cDNA was amplified by PCR and subcloned into the shuttle vector pShuttle-CMV, as described previously (22)(23)(24)(25). The resultant plasmids were linearized by the restriction endonuclease PmeI and purified using phenol/chloroform. The linearized plasmids were transformed into Escherichia coli strain BJ5183-AD1 competent cells (Stratagene) containing the supercoiled adenoviral vector pAd-Easy1 by electroporation (2.5 kV, 200 ohms, 25 microfarads), and recombinant bacteria colonies were selected by kanamycin resistance. Positive recombinant adenovirus colonies were characterized and selected by restriction endonuclease PacI digestion to release a small fragment of either 4.5 or 3 kb and a large fragment of a 35-kb adenovirus vector fragment. The correct recombinant adenoviral constructs were subsequently prepared and linearized with PacI and purified using phenol/chloroform. The linearized adenoviral vectors were transfected into a mammalian HEK293A packaging cell line with Lipofectamine 2000 (Invitrogen) for adenovirus production. Both cells and cultured media were harvested 7 days post-transfection, and viral lysates of transfected HEK293A cells were prepared by three freezethaw cycles alternatively using a 37°C water bath and a liquid nitrogen container. Samples were centrifuged at 500 ϫ g at 4°C to pellet the cell debris. Recombinant adenoviruses were further propagated in HEK293A cells. Briefly, cells with a confluence of ϳ60 -80% were reinfected by adding the transfected viral supernatants and then cultured in DMEM supplemented with 2% fetal bovine serum. When most of the infected cells were rounded up and approximately half of the cells were detached at 3-4 days following infection, the infected cells and cultured media were collected. Several rounds of amplification in HEK293A cells were performed. For the amplification and purification of high titer recombinant adenoviruses, large numbers of HEK293A cells were infected with viral supernatant at a multiplicity of infection of ϳ10 pfu/cell. The adenoviruses producing FGF21 or RAR␤, respective, were purified using the Adenovirus Purification Kit (Puresyn, Malvern, PA).
Animal Experiments-Male C57BL/6J mice at 8 weeks of age were purchased from Jackson Laboratory. The C57BL/6J mice were divided into two groups: fed and fasted. The fed group was fed ad libitum, and the fasted group of mice was fasted for 24 h. Feeding regimens were carried out in a staggered fashion, so that all mice were euthanized at the same time. All animal experiments were performed with the approval of the Institutional Animal Care and Use Committee at Boston University School of Medicine.
In Vivo Adenoviral Gene Transfer-Adenovirus-mediated overexpression of either FGF21 or RAR␤ in the liver of C57BL/6 mice was accomplished via intravenous injection, as described previously (25,26). One hundred microliters of adenovirus (5 ϫ 10 9 ϳ1 ϫ 10 10 pfu) per mouse were delivered into mice via tail vein injection. Two weeks postinjection, each group of mice was sacrificed under isoflurane anesthesia. Tissues were rapidly taken, freshly frozen in liquid nitrogen, and stored at Ϫ80°C until biochemical analysis.
Metabolic Cages-Mice injected with Ad-GFP or Ad-RAR␤ were housed individually in metabolic cages designed by Comprehensive Laboratory Animal Monitoring Systems (CLAMS) (Columbus Instruments, Columbus, OH). After 1 day of acclimation, metabolic data were collected automatically. On the third day of the experiment, food was removed for a 24-h fast. The rates of VO 2 and VCO 2 were expressed as average values measured every 18 min over a 12-h block of light and dark cycles. The energy expenditure (kcal/kg/h) was calculated with the formula, (3.815 ϩ 1.232 ϫ VCO 2 /VO 2 ) ϫ VO 2 ϫ 0.001, as described previously (27,28). Physical activity was measured on x and y axes using infrared beams to count the bean breaks in CLAMS cages. The values were then summed (x amb ϩ y amb ) over 12-h intervals of the light and dark cycles, respectively. These experiments were carried out in the Metabolic Phenotyping Core at Boston University School of Medicine as described previously (29,30).
Body Composition Analysis-Body composition was determined by an NMR system with the body composition analyzer Echo 900 (Echo Medical Systems, Houston, TX). Body fat, lean mass, body fluids, and total body water were measured in live conscious mice with ad libitum access to chow.
Dual Luciferase Activity Assays-Cells were transfected with 0.5 g of reporter plasmids, including human FGF21-Luc and 3XRARE-Luc, along with 40 ng of Renilla luciferase plasmid pRL-SV40 (Promega) as an internal control in a 12-well plate. Thirty-two hours post-transfection, cells were treated with either agonists or antagonists of RAR or PPAR␣ as described previously (24). Dual luciferase assays for firefly luciferase and Renilla luciferase activities were performed in duplicate according to the manufacturer's protocols (Promega). Luciferase activity was measured using an Infinite M1000 microplate reader (Tecan Group Ltd.). The firefly luciferase activity was normalized to the Renilla luciferase activity (firefly luciferase/ Renilla luciferase) and presented as relative luciferase activity.
Statistical Analysis-Values are expressed as mean Ϯ S.E. Statistical significance was evaluated using the unpaired twotailed t test. Differences were considered significant at the p Ͻ 0.05 level.

RESULTS
The Correlated Alterations of Hepatic RAR and FGF21 during the Metabolic Adaptation to Prolonged Fasting in Vivo-In response to nutritional challenges, metabolic gene transcription needs to be precisely controlled to maintain lipid homeostasis, which is associated with a network linking the nuclear receptor and hormonal signaling to the transcriptional regulation of metabolic pathways. Given that the computational screening results with the NUBIScan program (32) predict several possible RARE sites on the Fgf21 promoter, it is postulated that RAR might play a role in regulating hepatic FGF21 in response to physiological stimuli, such as nutrition deprivation. The alterations of RAR␤, FGF21, and fasting-related metabolic processes were compared in mice under both fed and fasted conditions. As shown in Fig. 1, hepatic mRNA levels of CRABP-II, the hallmark target gene of RAR (33,34), were significantly increased in mice after a 24-h fast. No significant changes in RAR␤ mRNA were noted in the fed and fasted mice. These results suggest that fasting stimulates RAR transcriptional activity, but it is not likely to be due to altered RAR expression. Hepatic and plasma levels of FGF21 were increased ϳ20-fold and ϳ7-fold in the fasted mice, compared with the fed mice, consistent with earlier studies (15). The fasting-inducible FGF21 might be predicted to result in enhanced fatty acid oxidation and ketogenesis via its autocrine/paracrine effects (15,19,35). In accordance with this notion, mRNA levels of CPT1␣, the rate-limiting enzyme of fatty acid oxidation, was increased over 7-fold in mice in the fed state. Expression of other enzymes involved in fatty acid oxidation, such as MCAD, was also increased. During a ketogenic process in response to prolonged fasting, mitochondrial 3-hydroxy-3-methylglytaryl-CoA synthase 2 (HMGCS2), the rate-limiting enzyme of ketogenesis, catalyzes the condensation of acetoacetyl-CoA with acetyl-CoA to synthesize HMG-CoA in hepatocytes, which is further converted into acetoacetate and acetyl-CoA by mitochondrial HMGCL. Acetoacetate is finally converted into ␤-hydroxybutyrate by ␤-hydroxybutyrate dehydrogenase 1 (36). Consistent with the induction of fatty acid oxidation, an over 4-fold increase in ketogenic genes (HMGCS2 and HMGCL) as well as an over 8-fold elevation in plasma ␤-hydroxybutyrate were observed in the fasted mice, compared with those of the fed mice. Our results suggest that RAR and FGF21 both are physiologically regulated by a feeding-fasting regimen in the maintenance of lipid homeostasis.
The Gene Transcription of Fgf21 Is Stimulated by the Natural RAR Ligand, RA, in HepG2 Cells-To determine the effect of pharmacological activation of RAR on FGF21 in hepatocytes, an FGF21(Ϫ2090/ϩ117) promoter-driven luciferase construct expressing a 2207-bp fragment of the 5Ј flanking region of the Fgf21 promoter was generated as described previously (24). Because HepG2 cells have strong endogenous RAR activity (37), this cell line was used to assess the reporter activity of the 3XRARE-driven luciferase reporter gene as well as the Fgf21 promoter-driven reporter gene. Cells were treated with RA at increasing concentrations (0.1-5 M), dosages that have been previously established in HepG2 cells and primary hepatocytes (8,38), and dual luciferase reporter assays were conducted. The transcriptional activity of the 3XRARE-driven reporter was robustly stimulated by RA. Consistently, Fgf21 promoter activity was enhanced by RA in a concentration-dependent manner (Fig. 2, A and B). Moreover, the transactivation of the Fgf21 promoter was significantly increased by other retinoids, such as 9-cis-retinoic acid, an RXR ligand (Fig. 2C). These results suggest that the transcription of Fgf21 is up-regulated by RA, the most active biological retinoid, and its 9-cis-isomer.
The Transcription of the Fgf21 Gene Is Up-regulated by RAR␣ and RAR␤ but Not by RAR␥ in HepG2 Cells-RAR consists of three isotypes (RAR␣, -␤, and -␥) (1, 2). To gain direct insight into whether the transcription of Fgf21 is modulated by different RAR isotypes, the effect of overexpressed FLAG-tagged mouse RAR␤ on FGF21 was initially examined in human HepG2 cells. As shown in Fig. 2, D-F, the 3XRARE luciferase promoter activity was dose-dependently stimulated by RA treatment and potentiated by overexpressed FLAG-RAR␤. Compared with control cells, mRNA expression of endogenous FGF21 was stimulated ϳ3-fold in cells transfected with FLAG-RAR␤. Next, overexpression experiments with isotype-specific mouse RAR␣, RAR␤, and RAR␥ were performed to assess the specific role of the three isotypes in FGF21 regulation. To this end, three plasmids encoding each of the different mouse RAR isotypes were transfected into HepG2 cells, and they were approximately equally expressed, as confirmed by elevated mRNA levels of each mouse RAR isotype. It was worth noting that RAR␤ was a more potent inducer of FGF21, although mRNA levels of Fgf21 were significantly increased by both exogenous RAR␣ and RAR␤. No significant changes in FGF21 were noted in cells transfected with mouse RAR␥ (Fig. 2, G and  H). These results suggest that RAR␣ and -␤, but not RAR␥, appear to regulate Fgf21 transcription. To further test a possible role of RAR␥ in RA-induced FGF21, luciferase reporter assays showed that RA-stimulated FGF21 activity appeared to be unaffected by RAR␥-selective antagonist MM11253 (Fig. 2I). Together, these results indicate that RAR␣ and RAR␤ play a major role in regulating FGF21. Because RAR␤, an inducible isotype of RAR (39), more potently stimulates FGF21 expression, RAR␤ was the main focus of all subsequent studies.
RAR␤ Plays a Critical Role in Inducing FGF21 in HepG2 Cells in Response to RA and under Conditions of Fasting-To further test the specific role of RAR␤, the effect of RAR␤ knockdown on FGF21 was characterized in HepG2 cells. To rule out possi-ble off-target effects of a small interfering RNA against RAR␤, siRNAs that targeted distinct areas of RAR␤ mRNA were used. As shown in Fig. 3, A-C, the mRNA levels of RAR␤ were effectively suppressed by siRNA RAR␤#7, with the greatest knockdown efficiency being ϳ60%, and, to a lesser extent, by siRNA RAR␤#8. Consistent with early studies showing that RAR␤ is greatly inducible by RA in both HepG2 cells and human livers with no significant changes in RAR␣ and RAR␥ (39), RA treatment caused a ϳ3-fold increase in mRNA amounts of RAR␤, the bona fide target gene of RA (40). Knockdown of RAR␤ by siRNA resulted in a profound reduction in the basal and RAstimulated RAR␤ expression, suggesting that RA is a more efficacious RAR ligand in HepG2 cells. Similarly, RA treatment also caused a ϳ2-fold increase in endogenous FGF21 in cells transfected with control siRNA. The stimulatory effect of RA was almost abolished by RAR␤ knockdown. Similar results were obtained with Fgf21 promoter activity. Given that the induction of FGF21 by RA is mimicked by overexpression of RAR␤ and is completely abrogated by knockdown of RAR␤, our studies using molecular and biochemical approaches indicate that RAR␤ is sufficient and necessary for the up-regulation of FGF21 in response to RA.
To understand the functional importance of RAR␤ on FGF21 and ketone body production under conditions of fasting, an in vitro model of nutrient deprivation that mimics in vivo fasting was generated as described previously (41), and Fgf21 promoter activity and ketone bodies were determined in HepG2 cells transfected with control siRNA or RAR siRNA. As shown in Fig.  3, D and E, Fgf21 luciferase activity was significantly increased ϳ2-fold in cells incubated in serum-free medium for 16 h and was sustained for 24 h. In concert with the increase in FGF21, ␤-hydroxybutyrate levels in cultured medium were increased by over 2-fold, consistent with increased ketone body production in primary hepatocytes in a similar starvation state (41). RAR␤ knockdown by siRNA caused a ϳ50% reduction of fasting-induced Fgf21 promoter activity. Consequently, fasting-induced ketone body production was largely abolished, suggesting that FGF21 induction by fasting may stimulate the ketogenic process in an RAR␤-dependent manner. Taken together with the in vivo findings (Fig. 1), these data suggest that fasting-induced FGF21 expression and ketogenesis are probably mediated through RAR␤.
Fgf21 Transcription Is Up-regulated by RAR␤ Independent of PPAR␣ in HepG2 Cells-Recent studies have implicated other nuclear receptors, such as PPAR␣ and ROR␣, as critical regulators of FGF21 in hepatocytes (15,17). Consistent with previous studies (17), mRNA abundance and promoter activity of Fgf21 were dose-dependently elevated by increased expression of mouse ROR␣ in HepG2 cells (supplemental Fig. 1), similar to those of RA treatment and RAR␤ overexpression (Fig. 2). As shown in Fig. 4, A-C, mRNA levels of Fgf21 were significantly increased by overexpression of PPAR␣. In parallel, FGF21 expression was also stimulated in cells treated with the synthetic PPAR␣ agonist GW7647. Additional reporter gene experiments showed that the transcriptional activity of Fgf21 was stimulated by a synthetic PPAR␣ agonist, Wy14643. Strikingly, the promoter activity of Fgf21 was dose-dependently increased by RA treatment, and this induction was potentiated by GW7647, suggesting that RAR␤ and PPAR␣ may synergize to induce Fgf21 transcription. Together, these findings indicate that RAR␤ appears to be one of the critical nuclear receptors that result in induction of FGF21 expression in hepatocytes.
Two pieces of data with a small interfering RNA approach further addressed the distinct role of RAR␤ and PPAR␣ in FGF21. First, in cells in which PPAR␣ was effectively suppressed by siRNA-mediated knockdown (Fig. 4D), FGF21 expression could be stimulated by RA treatment. In contrast, effective down-regulation of RAR␤ by siRNA caused a profound reduction in RA induction of FGF21, to the nearly normal levels seen in untreated cells (Fig. 4E). Second, similar experiments were performed to examine the role of PPAR␣ in regulating FGF21, and it was found that the induction of FGF21 by GW7647 was completely abolished in cells with effective knockdown of PPAR␣, but it was not affected in cells with siRNA-mediated suppression of RAR␤ (Fig. 4F). These results provide direct evidence that the Fgf21 transcription is independently up-regulated by PPAR␣ and RAR␤.
RAR Acts as a Transcriptional Activator of FGF21 in Vitro-To further delineate a molecular basis for the transcriptional regulation of FGF21 by RAR, experiments for structure-function analysis of the Fgf21 promoter were designed to identify a potential DNA element responsible for RAR using the fulllength FGF21(Ϫ2090/ϩ117) promoter as the "wild-type" promoter. A series of 5Ј-deletion FGF21 promoter vectors was generated by introducing various lengths of the proximal regulatory region of the Fgf21 promoter into the pGL3-Basic reporter vector, each driving production of firefly luciferase. As shown in Fig. 5, the full-length FGF21(Ϫ2090/ϩ117) promoter showed a ϳ2-fold increase in FGF21 activity in HepG2 cells in response to RA. Progressive 5Ј-flanking deletion reporters, including FGF21(Ϫ1500/ϩ117), FGF21(Ϫ1290/ϩ117), FGF21(Ϫ1040/ϩ117), and FGF21(Ϫ850/ϩ117), caused a reduction in basal transcriptional activity but permitted a robust response to RA, to an extent similar to that of the "wild type" FGF21(Ϫ2090/ϩ117) reporter. Conversely, each of the 5Ј-flanking deletion reporters, including FGF21(Ϫ600/ϩ117), FGF21(Ϫ200/ϩ117), and FGF21(Ϫ70/ϩ117), revealed a decrease in basal FGF21 activity and led to an elimination of almost all of the RA responsiveness (Fig. 5A). These promoter mapping studies suggest that a functional RARE is possibly located in the nucleotide sequence between Ϫ850 and Ϫ600 on the Fgf21 promoter. To precisely define a putative RARE site, further analysis of FGF21 deletion reporters, including FGF21(Ϫ800/ϩ117), FGF21(Ϫ750/ϩ117), FGF21(Ϫ700/ ϩ117), and FGF21(Ϫ650/ϩ117), showed a maximal response to RA, comparable with that of the FGF21(Ϫ850/ϩ117) reporter. In contrast, the FGF21(Ϫ600/ϩ117) deletion reporter led to the elimination of nearly all of the RA responsiveness (Fig.   5B), suggesting that the regulatory region located between Ϫ650 and Ϫ600 is responsible for RA-induced FGF21. The nucleotide sequences in this regulatory region were analyzed to identify one putative RARE sequence separated by one base (DR1) in the human Fgf21 gene as well as two putative RARE sequences spaced by five nucleotides (DR5) in the mouse Fgf21 promoter (Fig. 5C). Furthermore, the functional importance of this RARE-like nucleotide sequence within the FGF21(Ϫ650/ ϩ117) promoter was assessed by mutagenesis studies. The ability of RA to stimulate Fgf21 transactivation was abolished by the point mutation in the promoter containing a disrupted RARE when compared with that of the same reporter harboring the wild type RARE (Fig. 5B), strongly suggesting that induction of FGF21 by RA is dependent on a proximal RARE. These studies clearly establish FGF21 as a bona fide target gene of RAR.
The Recruitment of Hepatic RAR␤ to the Fgf21 Promoter Contributes to the Induction of Hepatic FGF21 during Fasting-To further define whether hepatic RAR␤ activity is physiologically relevant to FGF21 induction in vivo, ChIP experiments were performed to examine whether RAR␤ might bind to the RARE- Primers against the RARE-containing region (Ϫ699 to Ϫ500) and its upstream region (Ϫ5200 to Ϫ5000) of the mouse Fgf21 promoter as indicated were designed for ChIP-quantitative PCR. D, the occupancy of endogenous RAR␤ on the Fgf21 gene in mouse livers is increased by fasting. In vivo quantitative ChIP assays were performed using either RAR␤ antibody or control IgG. The specificity of the ChIP signal at the Fgf21 loci was confirmed by minimal binding that occurred in the distal upstream region or with IgG immunoprecipitation. The value obtained from fed mice with control IgG was set to 1, and -fold enrichment relative to this value was presented as the mean Ϯ S.E. (error bars) (n ϭ 4 -5); *, p Ͻ 0.05 versus fed mice.
like DNA sequence of the mouse Fgf21 gene. As shown in Fig.  5D, significantly less association of RAR␤ with this proximal region of the Fgf21 promoter was noted in livers of the fed mice. Strikingly, hepatic RAR␤ was recruited and bound to the putative RARE sequence in a fasting-inducible manner, which functionally contributes to the induction of hepatic FGF21 in these mice (Fig. 1A). The specificity of the occupancy of hepatic RAR␤ on the Fgf21 gene in vivo was evidenced by less binding of RAR␤ with negative controls using 1) a nonspecific PCR primer that amplified a genomic sequence located at over 5 kb upstream of the mouse Fgf21 promoter from the transcriptional start site and 2) an immunoprecipitation with a nonspecific IgG in lieu of RAR␤ antibody. These in vivo studies illustrate that the RAR␤/RARE is responsible for fasting-induced Fgf21 transcription in mouse livers.

Adenoviral Overexpression of RAR␤ in the Liver Is Sufficient to Enhance Production and Secretion of Hepatic FGF21 and to Promote Hepatic Fatty Acid Oxidation and Ketogenesis in Mice-
To gain important insight into the physiological relevance of hepatic RAR␤ to FGF21 actions in the regulation of lipid metabolism, overexpression and siRNA-mediated knockdown of RAR␤ were utilized. Adenovirus-mediated expression of RAR␤ in the liver was achieved by tail vein injection of adenovirus expressing RAR␤ (Ad-RAR␤) into C57BL/6 mouse livers. As shown in Fig. 6, A-H, expression of recombinant RAR␤ was confirmed by a 2-3-fold increase in the liver of mice injected with Ad-RAR␤ at 2 weeks postinjection. Hepatic overexpression of RAR␤ caused a ϳ2-fold increase in hepatic mRNA and protein levels of FGF21, consistent with the induction of FGF21 by RA treatment and RAR␤ overexpression in HepG2 cells (Fig.  2). Consequently, hepatic secretion of FGF21 was markedly increased, as reflected by an elevation of circulating FGF21 levels, suggesting that hepatic overexpression of RAR␤ for 2 weeks persistently increases hepatic and circulating FGF21 levels in mice. To explore the functional consequence of hepatic RAR␤-FGF21 axis, mRNA levels of key components of fatty acid oxidation and ketogenesis were determined by quantitative RT-PCR. Hepatic overexpression of RAR␤ resulted in the stimulation of fatty acid oxidation, as shown by elevations in CPT1␣ and MCAD, some of which were consistent with adaptive metabolic response to fasting (Fig. 1). Additionally, hepatic overexpression of RAR␤ significantly enhanced expression of HMGCS2 and HMGCL as well as elevated plasma ␤-hydroxybutyrate (Fig. 6, F-H). These studies suggest that hepatic RAR␤ positively regulates FGF21 at the transcriptional level and recapitulates some of physiological metabolic adaptation to fasting. To further determine whether RAR␤ mediates the metabolic effect of RA in vitro, expression of the key enzyme CPT1␣ was increased over 3-fold in HepG2 cells exposed to RA, comparable with that of FGF21 induction (Fig. 2F). This effect of RA was reduced by ϳ40% in cells transfected with RAR␤ siRNA (Fig.  6I). In agreement with increased gene expression of fatty acid oxidation, secreted ␤-hydroxybutyrate levels in the cultured medium were significantly elevated by RA treatment and reduced by RAR␤ knockdown (Fig. 6J), which was well correlated with the alterations of FGF21 and CPT1␣ (Figs. 2F and 6I). These findings indicate that RAR␤ preferentially transduces the RA signaling to fatty acid oxidation in hepatocytes.
The in Vivo Metabolic Effects of Hepatic RAR␤ Is Mimicked by Adenoviral Overexpression of FGF21 in the Liver-To better understand the in vivo functions of FGF21 in the regulation of hepatic lipid homeostasis, adenoviral gene delivery was accomplished by tail vein injection of adenoviruses expressing either GFP or FGF21 into C57BL/6 mouse livers. As shown in Fig. 7, hepatic mRNA levels of Fgf21 were dramatically elevated over 20-fold in mice injected with Ad-FGF21, compared with con-trol mice. Commensurate with hepatic induction of FGF21, circulating concentrations of FGF21 were ϳ6-fold increased, which is consistent with previous observations in FGF21 transgenic mice (42). The concept that FGF21 induces a metabolic state mimicking the liver's adaptive response to fasting was reinforced by an increase in hepatic fatty acid oxidation enzymes (CPT1␣ and MCAD) and ketogenic enzymes (HMGSC2 and HMGCL) as well as an elevation in plasma ␤-hydroxybutyrate levels in mice expressing FGF21 in the liver, a phenotype shared by RAR␤-expressing mice (Fig. 6). These data provide evidence that the metabolic outcomes of elevated RAR␤ or FGF21 in the liver can mimic most of the adaptive metabolic response to fasting, probably through enhanced hepatic fatty acid oxidation and substrate availability for ketogenesis.
FGF21 Is Required for RAR to Stimulate Fatty Acid Oxidation and Ketogenesis in HepG2 Cells-To rigorously define whether FGF21 acts downstream of RAR in the regulation of fatty acid oxidation and ketogenesis, HepG2 cells were transfected with a scrambled siRNA control or an siRNA targeting FGF21 and treated with RA. The ability of RA to up-regulate expression of CPT1␣ was almost abolished by FGF21 knockdown, suggesting that the stimulatory effect of RAR on fatty acid oxidation is mediated by FGF21 (Fig. 7, H and I). In agreement with altered CPT1␣, the effect of RA on the key ketogenic enzyme, HMGCS2, and secreted ␤-hydroxybutyrate in the cultured medium were significantly diminished by FGF21 knockdown (Fig. 7, J and K), indicating that RA-activated RAR controls fatty acid homeostasis in an FGF21-dependent manner. Together with in vivo overexpression studies, these results suggest that hepatic RAR␤ augments fatty acid utilization and supplies an energy source in part through the induction of hepatic FGF21.
Adenoviral Overexpression of RAR␤ in the Liver Is Sufficient to Enhance Energy Expenditure in Mice-Altered rates of hepatic fatty acid oxidation can contribute to enhanced energy expenditure and body weight loss (43). Whereas pharmacological activation of RAR by RA protects against diet-induced obesity and insulin resistance (7), little is known about a role of hepatic RAR␤ in the control of systemic energy balance. Because FGF21 has emerged as a critical regulator of energy balance in mouse models of obesity and diabetes (12), the direct functional impact of RAR␤ on systemic energy metabolism in vivo was explored. A cohort of Ad-RAR␤-injected mice, along with Ad-GFP-injected mice, was subjected to indirect calorimetry to measure oxygen consumption (VO 2 ) and carbon dioxide production (VCO 2 ) as an assessment of energy expenditure. As shown in Fig. 8, A-E, in both fed and fasted states, the rates of VO 2 and VCO 2 were significantly increased during light and dark cycles in mice expressing RAR␤ in the liver, compared with those of control mice. The calculated energy expenditure was also increased. Due possibly to the increase in both VO 2 and VCO 2 , the respiratory quotient (RQ ϭ VCO 2 /VO 2 ) appeared unchanged in mice expressing RAR␤, compared with control mice in fed and fasted states (data not shown). The RQ was not affected in mice expressing RAR␤, which possibly reflects the relatively equal use of carbohydrates versus lipids as a source of energy in whole body, as was described previously (45). Furthermore, circadian locomotor activity was similar in mice under the fed and fasted conditions because horizontal and vertical movements were comparable in mice expressing either hepatic GFP or FGF21 during light and dark cycles. Moreover, body composition by NMR measurement showed that a small but insignificant reduction in fat mass was noted in mice expressing RAR␤, compared with control mice. The lean mass and body weight were indistinguishable between the two groups of mice. These findings illustrate that hepatic induction of FGF21 by RAR␤ may contribute to whole-body energy homeostasis possibly through increased hepatic fatty acid oxidation or other endocrine effects.

DISCUSSION
The present study characterizes the hepatocyte-derived hormone FGF21 as a novel target gene of RAR␤. First, mRNA expression of FGF21 is selectively up-regulated by RAR␣ and -␤ but not by RAR␥ in HepG2 cells. Second, the gene expression and transcriptional activity of Fgf21 are stimulated by RA at least in part through RAR␤. In cells with suppression of PPAR␣, FGF21 is still inducible by RA. Third, functional dissection of the Fgf21 promoter illustrates that a putative RARE site of the Fgf21 promoter is responsible for RAR␤-stimulated Fgf21 tran-scription. Fourth, hepatic overexpression of RAR␤ by the adenoviral gene transfer approach persistently simulates expression of FGF21 and recapitulates much of FGF21 metabolic actions, such as enhanced hepatic fatty acid oxidation and ketogenesis, as well as increased whole-body energy expenditure in mice. Finally, the ability of RA or RAR␤ to stimulate fatty acid oxidation and ketogenesis is mimicked by FGF21 overexpression and diminished by FGF21 knockdown. Although recent studies implicated PPAR␣ and ROR␣ in the regulation of FGF21 in hepatocytes (15,17), the present study provides an alternative mechanism by which gene transcription of Fgf21 is mediated through hepatic RAR␤ via the putative RARE sequence in mice under conditions of nutrient deprivation. This additional layer of the transcriptional regulation of FGF21 by RAR␤ could further enhance the long term adaptive response to metabolic stresses. As depicted in Fig. 8F, hepatic RAR activation, particularly that of RAR␤, can up-regulate FGF21, leading to enhanced metabolic adaptation to fasting. This study may provide a rationale for the therapeutic potential of RAR-selective agonists for humans with cancers and metabolic disorders, especially because RA has been clinically used for cancer therapy in humans (1,2).

The Interplay of RAR and FGF21 in the Liver Plays a Role in Enhancing Hepatic Fatty Acid Oxidation and Ketogenesis-An
interesting finding of the present study is that RA/RAR simulates gene expression of lipid oxidation and ketogenesis in an FGF21-dependent manner. Our results also highlight a potential role of FGF21 in stimulating ␤-oxidation and ketogenesis in hepatocytes in response to nutrient deprivation. This mechanism would allow hepatocytes to increase mitochondrial fatty acid oxidation and ketogenesis and meet the energetic requirement in the circumstance of energy stress. Multiple studies have established that the treatment of RA, the active vitamin A metabolite, induces weight loss and attenuates insulin resistance in obesity and diabetes (6,7). One critical mechanism of the actions of RA is that activation of RAR may reduce hepatic lipid accumulation and normalize serum triglyceride levels (6,46). Unfortunately, the currently described RAR isotype knockout models demonstrate severe developmental defects; they display congenital malformations with the fetal and postnatal vitamin A deficiency syndrome (47), making it impossible to evaluate the impact of RAR on energy metabolism. Therefore, in vivo adenoviral gene transfer is used to generate mice with a moderate overexpression of RAR␤ in the liver, which displays an elevation of hepatic and plasma levels of FGF21, induction of fatty acid oxidation, and an increase in energy expenditure. There is a strong overlap between hepatic lipid metabolic genes that are regulated by enforced overexpression of FGF21 and those controlled by overexpression of RAR␤ in vivo. Because the phenotypic changes seen in mice expressing RAR␤ or FGF21 in the liver, such as elevation of enzymes for fatty acid oxidation (CPT1␣ and MCAD) and ketogenesis (HMGS2 and HMGCL), recapitulate much of the adaptive response to nutrient deprivation, understanding the hepatic RAR␤-mediated production of FGF21 may provide insight into a molecular basis for reprogramming metabolic gene transcription during physiological, nutrient stresses. Supporting this notion, FGF21 induction can be explained by the strong binding of hepatic RAR␤ to the putative RARE sequence on the Fgf21 promoter in the fasted mice. Future studies with generation of tissue-specific knock-out mice of isotype-specific RAR will be important for establishing the relative contribution of individual RAR isotype to the regulation of FGF21 and lipid homeostasis in physiological and pathological states.
The identification of additional components of the adaptive fasting pathway possesses potentially clinical implications. The fundamental fasting response has recently gained considerable attention as a potential therapeutic avenue for the treatment of metabolic dysfunction. Based on reduced hepatic fat accumulation in RA-treated obese mice (6, 7), the stimulation of fatty FIGURE 8. Adenoviral overexpression of RAR␤ in the liver is sufficient to enhance whole-body energy expenditure in mice. A-C, the effect of hepatic RAR␤ on the metabolic rate in mice. The rates of VO 2 (A), VCO 2 (B), and energy expenditure (C) are measured by comprehensive metabolic monitoring over a 24-h period with food and over a 24-h fast in mice expressing either GFP or RAR␤ in the liver. The top panels represent circadian changes in energy metabolic parameters in mice in the fed state during the dark cycle, and the bottom panels represent metabolic parameters in mice in fed and fasted states during both light and dark cycles. D, the locomotor activity is expressed as total counts, as measured by summing x and y beam breaks. The top panel represents the circadian locomotor activity of two groups. The bottom panel represents the average locomotor activity over a 12-h block of the light and dark phases. E, body composition analysis of mice expressing either GFP or RAR␤ in the liver. Bar graphs are presented as the mean Ϯ S.E. (error bars) (n ϭ 4). *, p Ͻ 0.05 versus Ad-GFP-injected mice. F, a proposed model for hepatic RAR as a novel regulator of FGF21. Activation of RAR by retinoid acid, the natural ligand, induces expression of FGF21 in hepatocytes, which in turn simulates hepatic fatty acid oxidation and ketogenesis possibly through autocrine/paracrine actions of FGF21. The RAR/RXR heterodimer may sever as the main functional regulator transducing RA signaling to the transcriptional regulation of FGF21 via the putative RARE sequence. Hepatic FGF21 induction by RAR␤ may result in enhanced systemic energy expenditure, possibly through increased hepatic ␤-oxidation and/or some of endocrine effects of FGF21. The physiological regulation of FGF21 by RAR probably contributes to the adaptive response to nutrient deprivation.
acid oxidation and energy expenditure by RAR␤ overexpression may explain most of the beneficial effects of RA and provide a rationale for pharmacological intervention with RARselective agonists to counteract the detrimental effects of fatty liver disease and metabolic syndrome.
FGF21 Is a Direct Target Gene of RAR in Hepatocytes-The major finding of the present study is the identification and characterization of RAR as one of the major nuclear receptors that regulate hepatic FGF21 and lipid metabolism. Strikingly, cells with knockdown of PPAR␣ remain responsive to FGF21 induction by RAR agonists, suggesting that additional FGF21 regulators such as RAR may modulate its transcription. We have demonstrated that both RAR␣ and RAR␤, but not RAR␥, can induce Fgf21 transcription. We notice that the three isoforms contain a highly conserved DNA binding domain that allows them to bind to a RARE site on a target gene promoter. However, the variability of the modulating domains may determine the specificity of the interaction of each RAR isotype with different co-regulators or cofactors on target gene promoters and may result in a different potential for the transcription of various target genes.
Our results indicate that hepatic RAR␤ is sufficient and necessary for the induction of FGF21 because Fgf21 transcription is stimulated by pharmacological activation of RAR and overexpression of RAR␤ as well as suppressed by RAR␤ knockdown. This cross-talk raises the possibility that RAR␤ may directly regulate FGF21. Functional analysis of deletion promoter reporters indicates that the putative RARE sequence present in the proximal region of the Fgf21 promoter is responsible for RAR␤-stimulated FGF21 in hepatocytes. Previous observations that deletion of PPAR␣ in mice cannot fully abolish the fastinginducible FGF21 suggest that other regulators may contribute to the induction of hepatic FGF21. One such mechanism described in this study is that FGF21 induction is possibly mediated through the recruitment of RAR␤ to the putative RAREcontaining region of the Fgf21 promoter in mice under the well established, physiological adaptive response to prolonged fasting. The data presented here show that mRNA levels of CRABP-II, a known target gene of RAR, are increased by fasting, which reflects a net effect of increased activity of RAR isotypes, despite no significant changes in RAR␤ mRNA levels being seen in the fasted mice. The nutrient sensing and/or hormonal inputs that regulate RAR activity, however, remain largely unknown. Accumulated evidence demonstrates that post-translational modifications of RAR, such as protein kinase A (PKA) phosphorylation of RAR and NAD-dependent deacetylase SIRT1-mediated deacetylation of RAR, may also affect the transcription of RAR or its target genes (48,49). Given that expression and/or activity of SIRT1 and cAMP/PKA are stimulated by fasting (50), nutrient sensor SIRT1 and other hormonal signaling linked to PKA may possibly be involved in the regulation of RAR␤ activity during fasting. The cross-talk between nutrient sensing pathways, the nuclear receptor, and FGF21 in the regulation of lipid homeostasis remains to be further investigated.
Hepatic RAR␤ Is Implicated in the Control of Whole-body Energy Balance-An intriguing observation is that increasing RAR␤ in the liver causes a moderate increase in whole-body oxygen consumption and energy expenditure, and these changes are probably attributable to the induction of hepatic FGF21 and fatty acid oxidation. Dysregulation of total energy balance/expenditure contributes to the pathogenesis of metabolic disorders, such as obesity and type 2 diabetes (44). Despite extensive studies, the pathogenesis and treatment of human obesity have not been fully elucidated. Obesity results from a positive energy balance, where energy intake exceeds energy expenditure. Manipulation of energy expenditure has long been pursued in the development of new therapies for obesity. For instance, thyroid hormone stimulates energy expenditure and promotes weight loss, but it also produces side effects, such as loss of lean mass and production of cardiac toxicity. Thus, the identification of a novel and druggable signaling pathway that controls energy metabolism could create many pharmacological opportunities. The present study uncovers a previously uncharacterized link between selective activation of RAR, a major development pathway, as well as FGF21, a nutrient sensor, in the regulation of lipid homeostasis and energy balance. Although elevated hepatic FGF21 levels and enhanced energy expenditure have been seen in mice expressing RAR␤ in the liver, future studies are needed to determine whether selective activation of RAR␤ in the liver can effectively reduce individuals susceptible to obesity-induced diabetes and energy imbalance.
In conclusion, the present study clearly establishes FGF21 as a novel target gene of RAR␤ in hepatocytes. Hepatic overexpression of RAR␤ stimulates FGF21, which in turn increases ␤-oxidation and ketogenesis and enhances energy expenditure in vivo. During prolonged fasting, hepatic RAR␤ is recruited to the putative RARE sequences of the Fgf21 promoter, leading to the induction of Fgf21 transcription in vivo. Because of the associated weight loss and systemic insulin sensitivity seen in diabetic mice treated with RA (6,8), targeting the RAR-FGF21 axis potentially represents a novel therapeutic avenue for treating fatty liver disease and metabolic syndrome.