Liver-secreted RBP4 does not impair glucose homeostasis in mice

Retinol-binding protein 4 (RBP4) is the major transport protein for retinol in blood. Recent evidence from genetic mouse models shows that circulating RBP4 derives exclusively from hepatocytes. Because RBP4 is elevated in obesity and associates with the development of glucose intolerance and insulin resistance, we tested whether a liver-specific overexpression of RBP4 in mice impairs glucose homeostasis. We used adeno-associated viruses (AAV) that contain a highly liver-specific promoter to drive expression of murine RBP4 in livers of adult mice. The resulting increase in serum RBP4 levels in these mice was comparable with elevated levels that were reported in obesity. Surprisingly, we found that increasing circulating RBP4 had no effect on glucose homeostasis. Also during a high-fat diet challenge, elevated levels of RBP4 in the circulation failed to aggravate the worsening of systemic parameters of glucose and energy homeostasis. These findings show that liver-secreted RBP4 does not impair glucose homeostasis. We conclude that a modest increase of its circulating levels in mice, as observed in the obese, insulin-resistant state, is unlikely to be a causative factor for impaired glucose homeostasis.

Circulating retinol is transported by its specific binding protein in serum, retinol-binding protein 4 (RBP4, also known as RBP) 3 (1). RBP4 is most highly expressed in liver, where it mobilizes hepatic retinyl ester stores for providing retinol to the periphery (1). Retinol-bound RBP4 circulates in a complex with transthyretin (TTR) and is taken up by cells expressing the RBP4 receptors stimulated by retinoic acid 6 (STRA6) (2) and RBPR2/STRA6L (3).
In 2005, increased levels of circulating RBP4 were linked to insulin resistance (4). Serum RBP4 levels were found elevated in genetic or dietary mouse models of insulin resistance, correlating well with increased expression that was detected in adipose tissue but not in liver (3). In humans, serum RBP4 in obese and obese/diabetic patients was found ϳ1.5-3-fold higher when compared with lean controls (4 -7), and this increase was mirrored by elevated expression of RBP4 in adipose tissue (8). Mice deficient in RBP4 were more insulin sensitive than control littermates, whereas mice with transgenic expression of RBP4 were prone to high-fat diet (HFD)-induced insulin resistance (4), although not all studies could reproduce these phenotypes (9,10). Interestingly, the sole injection of Escherichia coli-derived recombinant RBP4 into mice led to impaired glucose tolerance (4,11). These findings supported the notion of RBP4 working as adipokine, and it was hypothesized that an increased expression in adipose tissue of obese subjects leads to higher serum levels that, in turn, induce insulin resistance. So far, a variety of mechanisms of how RBP4 causes glucose intolerance and insulin resistance were reported (12), most prominently a RBP4-driven inflammatory response in adipose tissue involving the activation of antigen presentation and T cells (6,13,14).
A recent study showed that circulating RBP4 derives exclusively from hepatocytes because it was undetectable in serum of mice with a hepatocyte-specific deletion of RBP4 (15). Another study reported that transgenic expression of RBP4 in adipose tissue failed to increase circulating levels of RBP4 (16). Thus, RBP4 is unlikely to be an adipokine in the classical sense, suggesting that the elevation of serum RBP4 in obesity and diabetes is due to other effects, such as impaired renal clearance of RBP4 (7,17,18). It also proposes that RBP4 expressed in adipose tissue and other extra-hepatic sites is restricted to auto and/or paracrine effects (16).
RBP4-deficient mice (1), besides visual defects, suffer from behavioral abnormalities, neuronal loss, and some degree of gliosis (19) and may show developmental deficits due to the deletion of RBP4 in the germline. RBP4 transgenic mice express human RBP4 under the control of the mouse muscle creatine kinase promoter (20) and have 6 -10-fold increased total serum RBP4 (9), which is higher than what is usually observed in the obese/insulin-resistant state (4 -7, 21). These mice also suffer from retinal degeneration (9), and due to the transgenic expression of RBP4, accumulate retinol and retinyl esters in muscle (20). Whether these abnormalities interfere with the control of glucose homeostasis is unknown.
To test whether liver-secreted RBP4 impairs glucose homeostasis and to further delineate the underlying mechanisms, we opted to generate a system without the limitations of the current mouse models. We decided on a strategy that expresses mouse RBP4, and not the human protein, in a highly liver-specific manner in fully developed, adult mice. We used adenoassociated viruses (AAV) that express GFP or mouse RBP4 under the control of the synthetic and highly liver-specific LP1 promoter (22) to increase circulating levels of RBP4 by 2-3fold, similar to what is observed in the obese/insulin-resistant state (4 -7, 21). Surprisingly, mice with elevated levels of RBP4 failed to develop any metabolic phenotype when fed a normal chow (NC) or during a HFD challenge when compared with GFP controls. These findings show that liver-secreted RBP4 does not interfere with glucose homeostasis in mice. They also support the notion that the elevated levels of RBP4 in serum of obese/insulin-resistance individuals are not causative for impaired glucose homeostasis.

Establishing a novel model for liver-specific RBP4 overexpression in mice
We generated a model of liver-specific RBP4 overexpression by a nongenomic intervention in adult mice, thus excluding developmental compensations and potential unspecific effects due to genomic reorganization. AAV was chosen because of long-term expression of the transgene and low immunogenicity (23). AAV-mediated expression was under the control of a highly liver-specific promoter (LP1) (22) and first validated by LP1-driven GFP expression. In accordance with previous reports (23), we found that AAV serotype 2/2 was much inferior to AAV serotype 2/8 (AAV8cap) to express GFP in liver (Fig.  1A). AAV serotype 2/8 was then used throughout the study. LP1-driven GFP was detectable at significant amounts only in liver, when determined by UV microscopy (Fig. 1B) or by highly-sensitive mRNA expression analysis (Fig. 1C). We found that injecting 8E10 genomic copies of GFP or RBP4 expressing AAVs into the tail vein of mice led to a ϳ3-fold increase in serum RBP4, when analyzed 2 weeks later (Fig. 1D). RBP4 overexpression in this model was maintained for at least five months (data not shown).

Liver-specific overexpression of RBP4 induces elevated levels of RBP4 and retinol in serum
Seven-week-old male C57BL/6J mice were injected with respective AAV and fed NC for 6 weeks. Then, mice were switched to a HFD and kept for 18 weeks. A metabolic characterization was carried out as depicted in Fig. 2A. In accordance with earlier findings (4,21), serum RBP4 was elevated in HFDfed, obese mice by 2-fold (#, p Ͻ 0.05 between NC and HFD-fed AAV GFP control mice) (Fig. 2, B and C). Liver-specific overexpression of RBP4 increased serum RBP4 by 2-3-fold, irrespective of feeding NC (week 3) or HFD (week 24) (Fig. 2, B and C). Circulating RBP4 levels correlated well with serum retinol, suggesting that AAV-mediated RBP4 expression is functional, mobilizing hepatic retinyl ester stores (Fig. 2D). Interestingly, obese mice exhibited higher retinol levels in serum, which is in line with previous reports (16,24) and our finding that elevated RBP4 in mice upon HFD feeding is primarily retinol-bound holo-RBP4 (21). The underlying mechanisms that lead to increased holo-RBP4/retinol in serum of HFD-fed mice are still unclear but may involve alterations in hepatic RBP4 secretion, its renal clearance, and tissue retinol uptake. To further test whether AAV-mediated RBP4 expression is functional, we determined circulating levels of apo-and holo-RBP4 by nondenaturating PAGE. Consistent with increased retinol, liver-specific overexpression of RBP4 increased the abundance of circulating holo-RBP4 (Fig. 2, E and F). Retinol-free apo-RBP4, on the other hand, was highly variable in our mouse cohort and not affected by RBP4 overexpression (Fig. 2, E and F).

Liver-specific RBP4 overexpression does not impair glucose homeostasis in NC-fed mice
GFP/RBP4-expressing mice did not differ in body weights when kept on NC (Fig. 3A). Blood glucose levels of ad libitumfed mice expressing RBP4 tended to be lower at 2 and 3 weeks after AAV injections but failed to reach statistical significance (Fig. 3B). Serum insulin levels were similar (Fig. 3C). Surprisingly, glucose tolerance and serum insulin levels 15 min after the intraperitoneal (i.p.) injection of glucose (2 g/kg) did not differ between groups (Fig. 3, D and E). We also determined daily food intake and found no differences between GFP and RBP4 overexpressing mice (Fig. 3F). These findings show were tail vein-injected with AAV serotypes 2/2 or 2/8 expressing GFP under the control of the liver-specific LP1 promoter. Two weeks later, GFP expression in liver was analyzed by qPCR. AAV2/8-LP1-GFP tail vein-injected mice were analyzed for the expression of GFP in various tissues using UV microscopy (B) and qPCR (C). D, mice were injected with AAV expressing GFP and RBP4 via the tail vein, and serum RBP4 was analyzed by immunoblotting (IB) 2 weeks later. ADIPOQ served as loading control. In A and C, data are represented as individual data points and mean Ϯ S.E.

Liver RBP4 and glucose homeostasis
that liver-secreted RBP4, when elevated to levels as seen in the obese, insulin-resistant state, does not impair glucose homeostasis.

Liver-specific RBP4 overexpression fails to aggravate the metabolic impairment upon HFD feeding
We hypothesized that healthy, NC-fed mice may compensate the detrimental effect of elevated RBP4. Therefore, NC-fed mice were switched to HFD and characterized as shown in Fig.  2A. Both total weights and % weight gain were similar between RBP4 and GFP mice (Fig. 4, A and B), and there was no difference in body fat content (Fig. 4C). Fasting blood glucose was robustly increased in HFD-fed compared with NC-fed mice, indicating that HFD-induced obesity results in insulin resistance. However, there were no differences between GFP and RBP4 mice (Fig. 4D). Also ad libitum-fed blood glucose levels were similar (Fig. 4E). Strikingly, liver-specific overexpression of RBP4 failed to affect glucose tolerance and serum insulin levels 15 min after the i.p. injection of glucose (0.5 g/kg) in these HFD-fed mice (Fig. 4, F and G). At week 24, serum nonesterified fatty acid levels, triglycerides, and ketone bodies were not changed (Table S2). Also liver triglycerides and glycogen levels were comparable (Table S3).

Liver-specific RBP4 overexpression has no effect on whole body energy metabolism and locomotor activity
To address energy metabolism in more detail, HFD-fed mice were placed in a TSE LabMaster system. As expected for this diet, on which fat rather than carbohydrates are being utilized, the respiratory exchange ratio (RER) was close to 0.7 and did not differ between groups (Fig. 5A). Energy expenditure was higher in the active (dark) phase of mice without showing differences between GFP and RBP4 mice (Fig. 5B). Similar results were obtained for locomotor activity (Fig. 5C). Also food intake did not differ between both groups (Fig. 5D). These findings suggest that liver-RBP4 overexpression and a ϳ2-fold increase in circulating RBP4 levels do not affect energy metabolism in mice.

Liver RBP4 and glucose homeostasis
expression was previously shown to be increased in RBP4 injected mice (4). However, both gluconeogenic genes were unchanged in mice with liver-specific RBP4 overexpression (Fig. 6A). Also cytochrome P450 family 26 subfamily a member 1 (Cyp26a1), retinoic acid receptor ␤2 (Rarb2), cellular retinolbinding protein 1 (Crbp1), and lecithin-retinol acyltransferase (Lrat), representing well-characterized RAR␣ target genes (21,25,26), were not affected in RBP4 mice (Fig. 6B). mRNA expression of STRA6L, the RBP4 receptor highly expressed in liver (3), was slightly reduced (Fig. 6C), suggesting a compensatory down-regulation upon exposure to high levels of circulating RBP4. We then determined liver retinoid content. Due to reduced hepatic retinol mobilization, RBP4 deficiency in mice induces an accumulation of liver retinoids (1). Conversely, a strong overexpression of hepatic RBP4 for 6 days by adenoviral gene transfer led to a depletion of hepatic retinoids (21). Unexpectedly, we found that liver-specific overexpression of RBP4 for a total duration of more than 4 months had no effect on liver retinoid content (Fig. 6D).

Acute hepatocyte-specific depletion of TTR induces RBP4 accumulation in liver
We next asked why RBP4 is secreted from liver but not adipose tissue. RBP4 stoichiometrically assembles with TTR in the endoplasmic reticulum before its loading with retinol and secretion by hepatocytes (27). We depleted TTR expression in hepatocytes by expression of miR-155 that targets TTR through an inserted RNAi cassette (miTTR). Reducing TTR expression in liver had no effect on RBP4 mRNA expression (Fig. 7A). TTR protein was detectable by immunoblotting only after a harsh thermal denaturation, shown exemplarily for mouse serum (Fig. 7B). Hepatocyte-specific depletion of TTR strongly reduced TTR levels in the circulation, suggesting that its liver expression is the major determinant for its concentration in blood (Fig. 7C). Also serum RBP4 was reduced (Fig. 7C), which is consistent with the notion that TTR is required for complex formation that prevents renal filtration (28). In liver, however, depletion of TTR resulted in an accumulation of RBP4 protein, indicating that TTR facilitates the secretion of RBP4 (Fig. 7D). Strikingly, we noted that the robust expression of RBP4 in liver is matched with even higher expression of TTR (Fig. 7E, calculation based on qPCR amplifications with equal efficiencies). In epididymal adipose tissue (eWAT), RBP4 expression was less than 5% of that in liver and the expression of TTR expression even less (Fig. 7E). This indicates that RBP4's low expression combined with the lack of adequate TTR co-expression may prevent adipose tissue-expressed RBP4 to enter the circulation.

Discussion
The here presented study investigated whether liver-secreted RBP4 impairs glucose homeostasis. We took advantage of a liver-specific AAV expression system in adult mice that allows for elevating RBP4 to serum levels that are observed in the obese, insulin-resistant state. Surprisingly, we found that these mice failed to develop signs of glucose intolerance and insulin resistance when compared with their controls. Even when feeding HFD, an established metabolic stress challenge, RBP4 overexpression had no effect on the worsening of glucose control. We were surprised by these findings because prior studies showed a very clear impairment of glucose homeostasis in mice with muscle-specific overexpression of RBP4 and in mice that were injected with recombinant RBP4 of human origin (4,11). Mouse sex is unlikely to account for this because previous studies also used male mice (4,9,10). The obvious differences of how RBP4 gain-of-function was achieved as well as its magnitude may explain the discrepancies with our study. However, we believe that our model, with a modest 2-3-fold increase in circulating RBP4 in adult mice, is closest to the observed changes in serum RBP4 in obese, insulin-resistant humans. Endogenous RBP4 protein levels may follow a different expression/secretion pattern or degree of retinol binding than the overexpressed protein and thus differ in its effects. Although we cannot rule out this possibility, it seems rather unlikely because also the overexpressed protein should be subject to, e.g. regulated secretion by the hepatocyte. Because elevated RBP4 in the circulation of obese/insulin-resistant mice is primarily holo-RBP4 (21), we would also argue that our model mimics these changes closely. Thus, our study does not support a causative role for elevated levels of circulating RBP4 in the development of insulin resistance. This is further supported by the finding that mice lacking RBP4 in hepatocytes, and therefore in the circulation, are not protected from developing insulin resistance upon feeding a HFD (15). Of note, those mice were characterized on a mixed C57BL/6J ϫ 129Sv background that may prevent the development of a robust metabolic phenotype. A characterization on a cleaner C57BL/6J background will be required to draw final conclusions, as it is stated in the particular study (15).
If liver-secreted RBP4 and its circulating levels are rather irrelevant for its metabolic actions, what is the link between  RBP4 and impaired glucose homeostasis, and how could apparently contradicting results be reconciled? We believe that a potential clue comes from mice with adipocyte-specific overex-pression of RBP4. Although adipocyte-RBP4 did not contribute to RBP4 levels in blood, these mice still developed a strong metabolic phenotype, including impaired glucose tolerance, dyslipidemia, and hepatic steatosis (16). This could imply an important, yet unknown role of adipocyte-RBP4 in controlling lipid handling in an autocrine or paracrine manner. Although RBP4 seems to be restricted to adipocytes/adipose tissue in this mouse model, it could induce an exaggerated release of lipids   . Liver-specific overexpression of RBP4 has no effect on gluconeogenic gene expression and hepatic retinoid content. After feeding HFD for 16 weeks, mice were sacrificed and hepatic expression of Rbp4 and gluconeogenic genes (A), retinoid-responsive genes (B), and the hepatic RBP4 receptor Stra6l (C) was determined by qPCR. D, liver retinoid levels were analyzed by HPLC. Data are represented as individual data points and mean Ϯ S.E. and *, p Ͻ 0.05 between AAV GFP and AAV RBP mice.

Liver RBP4 and glucose homeostasis
that may deteriorate glucose homeostasis in other organs, including liver and muscle. In accordance with this is the observed increase of nonesterified fatty acids in plasma and elevated uptake of those into the liver in mice with adipocytespecific RBP4 overexpression (16). Intriguingly, a recent crystallographic study showed that RBP4 can indeed bind certain saturated fatty acids (29). In this scenario, liver-secreted RBP4, whether modestly elevated as shown here or lacking (15), does not affect local RBP4 functions in the adipose tissue compartment and therefore fails to affect glucose homeostasis. On the other hand, a mouse model of very strong overexpression or the i.p. injection of recombinant RBP4 may interfere with adipocyte/adipose tissue RBP4 and thus impair glucose control. Further studies are needed to test these hypotheses and to understand the role of RBP4 in lipid metabolism, especially in regard to lipid handling/mobilization in adipocytes.
Another interesting observation is that our long-term overexpression of RBP4 in liver did not deplete hepatic retinoid stores or affect RAR␣ target gene expression. We previously showed that an adenoviral overexpression of RBP4 in liver led to ϳ25% reduction in total liver retinyl esters during a 6-day period (21). This suggests that increased retinol mobilization by liver-secreted RBP4 can be compensated in the long-term. Interestingly, liver-specific deletion of RBP4 did not increase hepatic retinoids (15) as observed in whole body knockout mice (1), suggesting that extra-hepatic RBP4 may be involved in the compensational response.
Finally, we show that hepatocyte-specific depletion of TTR in adult mice increases RBP4 protein levels in liver. Because RBP4 mRNA is unchanged, this suggests that TTR is involved in RBP4 secretion and that its depletion leads to an accumulation of RBP4 in liver. Interestingly, elevated RBP4 levels in liver were also observed in mice with germline deletion of TTR (30). Our findings extend this by providing evidence for a hepatocyteautonomous mechanism independent of potential developmental effects caused by the deletion of TTR (19,31). Strikingly, elegant follow-up experiments showed that despite the loss of TTR, RBP4 secretion by the liver and cultured hepatocytes was not impaired (28). This indicates that liver RBP4 does not require TTR for its secretion. Whether or not very low TTR expression in adipose tissue is the underlying reason for why adipose RBP4 fails to enter the circulation will have to be tested in future experiments.
In summary, we characterized a novel RBP4 gain-of-function mouse model and show that an elevation of liver-secreted RBP4 to serum levels as observed in the obese, insulin-resistant state does not impair glucose homeostasis in mice. Thus, although increased levels of circulating RBP4 associate with glucose intolerance and insulin resistance (12), they are unlikely to be a causative factor. However, RBP4 may exhibit local functions in organs such as adipose tissue whose investigation could help to reconcile the divergence of previous reports.

Cloning, production, and purification of recombinant AAV
Murine RBP4 or coral GFP cDNAs were amplified by PCR and cloned downstream of a Kozak sequence into the pds-AAV plasmid (In-Fusion HD cloning kit, Clontech). These vectors were used to generate self-complementary AAV vectors, containing the synthetic and highly hepatocyte-specific LP1 promoter (22). For liver-specific depletion of TTR, vectors encoding pre-miR with flanking regions of miR-155 (pcDNA 6.2-GW/miR) using the antisense target sequence 5Ј-TCACC-ACAGATGAGAAGTTTG-3Ј of mouse TTR or a nontargeting sequence 5Ј-AAATGTACTGCGCGTGGAGAC-3Ј were cloned (BLOCK-iT TM Pol II miR RNAi kit, ThermoFisher). miControl and miTTR cassettes were then transferred into the pds-AAV plasmid for generation of self-complementary AAV vectors. AAVs of the indicated serotypes were produced by cotransfection of pds-AAV plasmids with appropriate helper plasmids expressing E2a, E4-orf6, VARNA, AAV2rep, and AAV2cap or AAV8cap in 293T cells by standard calcium phosphate precipitation. AAV particles were isolated after 48 -72 h by cell lysis using 1% Triton X-100 and repeated freeze-thaw cycles, followed by Benzonase digestion (250 units/ml) for 45 min, and purification by FPLC (AVB-Sepharose column, GE Healthcare 28-4112-11). For loading to the AVB-Sepharose column, freeze-thaw lysates were diluted with an equal volume Figure 7. Acute hepatocyte-specific depletion of TTR induces RBP4 accumulation in liver. A, mice were injected with AAV expressing miControl or miTTR that depletes endogenous TTR expression, and liver mRNA expression of Ttr and Rbp4 determined by qPCR 3 weeks later. B, TTR immunoblots with unboiled or boiled (20 min) mouse serum samples and an arrow indicating denatured TTR monomers at 15 kDa. C, mice were injected with AAV expressing miControl or miTTR and serum TTR and RBP4 was analyzed by immunoblotting 3 weeks later. ADIPOQ served as loading control. D, livers of mice described in C were analyzed for protein expression of TTR and RBP4 by immunoblotting. E, mRNA expression of Rbp4 and Ttr in liver and epididymal white adipose tissue (eWAT) was determined by qPCR. In A and E, data are represented as individual data points and mean Ϯ S.E. with *, p Ͻ 0.05 between AAV miControl and AAV miTTR mice.

Liver RBP4 and glucose homeostasis
of PBS-MK buffer (PBS with 1.0 mM MgCl 2 and 2.5 mM KCl). Unbound proteins were washed out with 30 column volumes of PBS-MK and AAV vectors were eluted with 0.5 M NaCl, 0.1 M sodium acetate, pH 2.5, under immediate neutralization with 1/10 volume of 1.0 M Tris, pH 8.5. Peak fractions with purified AAV were dialyzed against sterile PBS-MK buffer (Slide-A-Lyzer, number 66380, ThermoFisher) and titers were determined by qPCR, amplifying a fragment of the LP1-promoter (Table S1).

Mice and tail vein injection of AAV
Animal procedures were in accordance with institutional guidelines and approved by the "Landesamt für Gesundheit und Soziales" in Berlin. Sample size estimate for mouse experiments was based on power analyses performed using the analysis platform at the University of Muenster. C57BL/6J mice were housed under standard 12-h light/12-h dark cycles. Equal titers of AAV2/8 (8E10 genomic copies of RBP4 and GFP AAV) were injected via the tail vein into 7-week-old male mice, after dilating the tail vein in a 42°C water bath for 1 min. Injected mice were fed a standard chow diet (9% kcal fat, 25 IE vitamin A/g, diet R/M-H from ssniff Spezialitäten GmbH, Soest, Germany) for 6 weeks before switching to a HFD (60% kcal fat, 5.17 IE vitamin A/g, D12492, Research Diets) for 18 weeks. For the depletion of hepatic TTR, 7-week-old male mice fed a standard chow diet were tail vein-injected with equal titers (3E10 genomic copies) of AAV2/8 miControl and miTTR and analyzed 3 weeks later.

Metabolic characterization of mice and analysis of serum and liver parameters
After adaptation for 16 h, physical activity was analyzed for 48 h using a multidimensional IR light beam system of a TSE LabMaster monitoring system. The respiratory exchange ratio (RER) was calculated as the ratio of CO 2 produced and O 2 consumed. Energy expenditure was normalized to the body mass to the power of 0.75. Body fat content was determined by NMR (Bruker's MiniSpec MQ10), and insulin levels by ELISA (Rat Insulin ELISA, CrystalChem). Glucose tolerance was determined by repeated measurements of blood glucose at the indicated times after an intraperitoneal injection of 2 g/kg (NC-fed) or 0.5 g/kg of glucose (HFD-fed) after an overnight fast for 16 h. Food intake of single-housed mice was determined over a period of 48 h. Serum metabolites were determined by commercially available kits for triglycerides (Triglyceride FS, DIASYS Diagnostic Systems), ␤-hydroxybutyrate (LiquiColor, Stanbio), and nonesterified fatty acids (HR Kit, WAKO Chemicals). Liver triglycerides were analyzed as described previously (32). Liver glycogen was determined by measuring glucose (Glucose (HK) Assay Kit, Sigma) after homogenization of liver tissue and acidic hydrolysis with 1.25 N HCl at 100°C for 1 h and neutralization with NaOH.

Protein isolation and immunoblotting
Liver protein was prepared from shock-frozen tissue by homogenization in RIPA buffer that contained protease inhibitors using an ultraturrax and subsequent sonication. Protein concentrations of liver extracts were determined by a BCA assay (Thermo Scientific). Liver protein and mouse serum samples (diluted 1:25) were denatured for 5 min at 95°C and separated by SDS-PAGE and blotted to polyvinylidene difluoride membranes. After incubation with antibodies for RBP4 (A0040, Dako, Germany), TTR (A0002, Dako, Germany), GAPDH (Cell Signaling), or ADIPOQ (sc-26497, Santa Cruz), a secondary horseradish-coupled antibody was added and chemiluminescence detected (Thermo Scientific). For immunoblots visualizing denatured TTR monomers, liver/serum samples were incubated in boiling water for 20 min. Nondenaturated PAGE of mouse serum for the detection of holo-and apo-RBP4 was performed as previously described (21). Densitometric analysis was performed using ImageJ (33).

RNA isolation and analysis of gene expression by qPCR
Total RNA of mouse tissues was purified, translated to cDNA, and analyzed by qPCR as described previously (21). Primer sequences are listed in Table S1.

Quantification of liver and serum retinoid levels
Retinol and retinyl esters were quantified by HPLC-photodiode array detector as described previously (21).

Statistical analyses
Data are displayed as individual data points and mean Ϯ S.E. Significance was determined by the two-tailed Student's t test or 2-way analysis of variance (Tukey's post hoc test) analysis by GraphPad Prism as appropriate, and p Ͻ 0.05 deemed significant.