Fenretinide mediated retinoic acid receptor signalling and inhibition of ceramide biosynthesis regulates adipogenesis, lipid accumulation, mitochondrial function and nutrient stress signalling in adipocytes and adipose tissue

Graphical abstract Proposed mechanistic action of FEN in adipocytes. RA inhibits adipogenesis through early RAR activation, which PPARγ agonist ROSI cannot prevent. FEN inhibits adipogenesis with a delayed response in RAR signalling, however, inhibition is lost when combined with ROSI. The 4-OXO FEN catabolite cannot inhibit adipogenesis through RAR activation, but like FEN’s RAR-independent effects, displays increased phosphorylation of Akt, mild cellular stress/autophagy induction and decreased lipid accumulation. Moreover, FEN-mediated inhibition of DES-1 increases dihydroceramide levels in a RAR-independent manner, and is linked to a complete prevention of mitochondrial dysfunction and decreased adiposity in high-fat fed obese mice. Thus an alternative to the additive beneficial effects of FEN-mediated RAR-dependant and -independent signalling, 4-OXO FEN may be a novel therapeutic candidate to improve adipocyte hypertrophy via inhibition of ceramide biosynthesis and modulation of nutrient stress pathways.


Introduction
Retinol (vitamin A) and the retinoid metabolism pathway play an important role in body mass regulation and adipocyte biology [1][2][3][4][5][6]. Targeting retinoid homeostasis may therefore offer a therapeutic approach for obesity and type-2 diabetes. Vitamin A is a lipid soluble molecule which undergoes multiple steps of metabolism through a complex pathway of enzymes and transport proteins [7]. All-trans-retinoic acid (RA), which is the most active metabolite of retinol, has long been known to inhibit adipogenesis through the prevention of C/EBPb mediated transcription [1,8], More recently, RA has been shown to improve obesity and glucose homeostasis in vivo [4]. However, with prolonged exposure, naturally derived retinoid compounds such as RA and retinylacetate can lead to liver toxicity, which restricts their potential use as therapeutic agents [9].
Fenretinide (FEN), otherwise known as N-(4-hydroxyphenyl) retinamide or 4-HPR, is a structural derivative of RA with reduced toxicological profile [10,11]. FEN treatment leads to increased renal clearance and thus decreased serum levels of the retinol transport protein, serum retinol binding protein (gene name Rbp4) in both humans and mice [12][13][14][15][16]. This decrease in circulating RBP4 levels had been proposed to be the mechanism that FEN treatment led to prevention of insulin resistance associated with high-fat diet (HFD) induced obesity [15]. However, FEN has also been shown to reduce obesity and hyperleptinaemia in mice lacking RBP4, implying that the anti-obesity effects of FEN are most likely to be independent of its ability to decrease circulating RBP4 levels [16].
Thus, the beneficial effects of FEN appear to be through several different mechanisms including alterations to retinoid homeostasis in multiple tissues [6], increased hepatic lipid oxidation [17] and inhibition of ceramide biosynthesis leading to an increase in dihydroceramide in muscle and liver [18]. Studies in cancer cells have found that both FEN and dihydroceramide treatment are associated with the activation of cellular stress responses and autophagy induction [19][20][21]. Autophagy plays a crucial role in cellular homeostasis through the degradation and recycling of organelles such as mitochondria or ER and the regulation of intracellular lipid stores [22,23]. Since defective autophagy may also underlie impaired insulin sensitivity in obesity and upregulating autophagy may be a useful strategy to combat insulin resistance [24][25][26], it highlights the importance to further characterise the biological effects of FEN.
In addition to the beneficial effect of FEN in vivo, we recently demonstrated that FEN (similarly to RA) is able to inhibit 3T3-L1 adipocyte differentiation by blocking transcription of C/EBPa and PPARg, master regulators that synergistically coordinate adipogenesis and adipocyte biology and simultaneously increasing retinoid gene expression [6]. Interestingly, rosiglitazone (ROSI), a thiazolidinedione and PPARg agonist commonly used to stimulate adipogenesis, suppressed gene expression of all the retinoid metabolism markers examined in differentiating 3T3-L1 cells [6]. FEN also decreased lipid accumulation in fully differentiated 3T3-L1 adipocytes [6], but RA treatment did not [27,28]. Thus, FEN and RA appear to have divergent biological effects which may be due to unique activation of retinoic acid receptor (RAR)-dependent and -independent pathways. Since, it is unknown which pathway FEN requires for its biological effects in adipocytes we aimed to determine the mechanism(s) of FEN action in the 3T3-L1 model of pre-adipocyte differentiation and fully differentiated adipocytes. In addition, we have translated some of our new findings to adipose tissue of HFD-induced obese mice.

Cell culture
3T3-L1 pre-adipocytes were maintained and differentiated for 8 days (or 16 days where stated) as previously described [6]. C3H10T1/2 cells were similarly treated however penicillin/ streptomycin was omitted from the media. DMSO was used as vehicle control (VEH) and to dissolve all experimental compounds. FEN (Cilag AG, Schaffhausen, Switzerland), RA (Sigma-Aldrich, UK) and ROSI (Cayman Chemical, MI, USA) were used at 1 mM, 4-OXO (Santa Cruz, TX, USA) at 0.5 mM and ER50891 (Tocris Bioscience, Bristol, UK) at 10 mM and added at day 0 (or day 8 where indicated).
Cells were stained for neutral lipids with Oil Red O, images taken and then the stain was eluted and quantified at 520 nm.

Lipolysis assay
Basal glycerol levels were measured in media collected at day 16 of differentiation from 3T3-L1 cultures. The Triglyceride Liquid assay (Sentinel Diagnostics #17628) was used following the supplied protocol.

Gene expression and protein analysis
RNA isolation, cDNA synthesis and qPCR were performed at day 8 of differentiation (or as stated in the figure legends) as previously described [6]. Control reactions for contaminating DNA were performed routinely and relative expression calculated using the Pfaffl method [29]. The geometric mean of three stable reference genes (Nono, Ywhaz and Actb or as stated in the figure legend) were obtained from five commonly used sequences and used for normalisation. Primer sequences available on request, some of which were obtained from PrimerBank [30].

Global lipidomics analysis of adipocytes
Extraction of 3T3-L1 adipocyte lipids was performed according to the method described by Folch et al. [32]. The lipids were analysed by liquid chromatography-mass spectrometry (LC-MS) using a Thermo Orbitrap Exactive mass spectrometer (Thermo Scientific, Hemel Hempstead, UK), equipped with a heated electrospray ionization (HESI) probe and coupled to a Thermo Accela 1250 UHPLC system. All samples were analysed in both positive and negative ion mode over the mass to charge (m/z) range 200-2000. The samples were injected on to a Thermo Hypersil Gold C18 column (2.1 mm Â 100 mm, 1.9 mm). Mobile phase A consisted of water containing 10 mM ammonium formate and 0.1% (v/v) formic acid. Mobile phase B consisted of 90:10 isopropanol/acetonitrile containing 10 mM ammonium formate and 0.1% (v/v) formic acid. The initial conditions for analysis were 65%A/35%B. The percentage of mobile phase B was increased to 100% over 10 minutes and held for 7 min before re-equilibration with the starting conditions for 4 min. All solvents were LC-MS grade (Fisher Scientific, Loughborough, UK). The raw LC-MS data were processed with Progenesis CoMet v2.0 software (Non-linear Dynamics, Newcastle, UK) and searched against LIPID MAPS (www.lipidmaps.org) for identification.

Animals
Male C57BL/6 mice were randomised by body weight at three months of age and fed CHOW, HFD or FEN-HFD for 20 weeks. Analysed tissues were collected during previously performed experiments as described in [6], whereby FEN-HFD prevented obesity and factors associated with insulin resistance. Perigonadal white adipose tissue (PG-WAT) from ad libitum fed mice was rapidly dissected, frozen in liquid nitrogen, and stored at À80 C. Animal procedures were approved by the University of Aberdeen Ethics Review Board and performed under license (PPL60/3951) approved by the UK Home Office.

Quantitative analysis of ceramides and dihydroceramides in adipose tissue
Lipids were extracted from murine adipose tissue according to the method of Bligh and Dyer [33]. The ceramides and dihydroceramides were then isolated by silica solid phase extraction chromatography. C17:0 ceramide and C12:0 dihydroceramide (Avanti Polar Lipids, Alabaster, AL, USA) were included in the experimental system as internal standards (ISTD). LC-MS/MS analyses were performed in positive ion mode on a Thermo TSQ Quantum Ultra triple quadrupole mass spectrometer equipped with a HESI probe and coupled to a Thermo Accela 1250 UHPLC system. The ceramides and dihydroceramides were separated on a Kinetex 2.6 mm C8 column (100 Â 2.1 mm) (Phenomenex, Macclesfield, UK). Mobile phase A consisted of 90% H 2 O, 10% acetonitrile with 0.1% formic acid and mobile phase B consisted of acetonitrile with 0.1% formic acid. The gradient was held at 80% B for 1 min initially, increased to 100% B at 15 min, held at 100% B for 1 min and then re-equilibrated to starting conditions with a total run time of 20 min. The flow rate was 500 ml/min with a column temperature of 40 C. All solvents were HPLC grade or above (Fisher Scientific, Loughborough, UK). The data were acquired and processed using Xcalibur software v2.1 (Thermo Scientific). The concentration of the ceramide and dihydroceramide molecular species was determined by comparison to calibration curves generated with C16:0 and C24:1 standards (Avanti Polar Lipids, Alabaster, AL, USA). Total ceramide and dihydroceramide concentrations were calculated from the summed concentrations of all the monitored molecular species. All values were normalised to wet weight of PG-WAT.

Metabolomic profiling of adipose tissue
Metabolomic profiling was carried out on a ZICpHILIC column (150 Â 4.6 mm, 5 mm, HiChrom, Reading, UK) and an Orbitrap Exactive MS using conditions described previously [34]. Data extraction and data base searching were also carried out as described previously [34].

Statistics
Data represents the mean AE SD and n indicates the number of biological replicates. Data were analysed using one-way ANOVA with Tukey's multiple-comparison post-hoc test (or unpaired Student's t-test where stated) with P < 0.05 considered significant.

PPARg agonist driven adipocyte differentiation is inhibited by RA but not FEN
FEN completely inhibited adipocyte differentiation in the 3T3-L1 cell line [6] and in C3H10T1/2 pluripotent stem cells indicating that inhibition of adipogenesis is a common characteristic of FEN treatment (Fig. 1a). We examined whether FEN could inhibit adipogenesis in 3T3-L1 cells stimulated with ROSI. RA was able to completely inhibit differentiation ( Fig. 1b) with ROSI present (RA + ROSI). FEN was not able to inhibit adipogenesis in the presence of ROSI, but decreased lipid accumulation when compared to VEH or ROSI treatments (Fig. 1b). Moreover, RA + ROSI was able to suppress levels of C/ebp alpha but FEN + ROSI was unable to replicate this suppression (Fig. 1c). FEN + ROSI could not suppress C/EBPa-PPARg target genes, Fabp4, Adiponectin   in the presence of ROSI in terms of lipid accumulation and C/ebp alpha induction, gene expression of terminal markers of adipogenesis (Fabp4 and Adiponectin) appeared unaffected in contrast to treatment with RA alone. These findings highlight that several markers should be used to draw conclusions when assessing adipocyte differentiation. RA + ROSI induced retinoid homeostasis genes Crbp1, Raldh1 and Rar gamma, although the increases in expression were partially suppressed by ROSI (Fig. 1d). FEN + ROSI did not induce these retinoid genes. These results suggest that induction of RAresponsive genes is required for retinoid-mediated inhibition of adipogenesis since PPARg stimulation can block this inhibition.

RARa antagonism leads to loss of FEN effects on inhibition of adipogenesis
To account for the differences between FEN and RA observed during ROSI induced differentiation, we hypothesized that FEN induced retinoid signalling is less potent compared to RA. To test this, we measured retinoid-responsive gene expression following FEN/RA treatment during the early stages of adipogenesis (Fig. 2a) and in response to increasing concentrations ( Fig. 2b and c). We determined that FEN induced retinoid signalling is 2-3 times less potent than RA. Therefore, next we tested this RAR-dependence by blocking RAR ligand-mediated signalling with a selective RARa antagonist (ER50891). ER50891 blocked RA-induced inhibition of adipogenesis and allowed accumulation of lipid (Fig. 3a). Similarly, ER50891 blocked FEN-induced inhibition of adipogenesis (Fig. 3a). In addition, RARa antagonism prevented inhibition of adipocyte genes (Fig. 3b) and inhibited Crbp1 and Raldh1 by !50% (Fig. 3c). Overall, these findings strongly demonstrate that the mechanism of FEN action to inhibit 3T3-L1 adipogenesis is mediated by ligandinduced activation of RARa signalling and genes involved in retinoid metabolism.

FEN acts through RAR-independent mechanisms to decrease lipid levels in mature adipocytes
Next, we examined whether RAR-dependent pathways were responsible for the FEN-induced decrease in lipid accumulation in mature 3T3-L1 adipocytes [6]. Antagonism of RARa with ER50891 failed to prevent this effect of FEN ( Fig. 4a and b). Moreover, FEN did not affect media glycerol levels when compared to VEH controls suggesting that the basal rate of lipolysis is not altered by FEN (Fig. 4c). ROSI increased media glycerol levels indicating a putative increase in lipolysis consistent with an increase in ATGL expression [6,35,36]. These data suggest that the FEN-induced decrease in lipid accumulation in mature 3T3-L1 adipocytes is via an RAR-independent mechanism(s).

Lipidomics analysis of FEN treated 3T3-L1 adipocyte cells
Previous studies have identified alterations in ceramide biosynthesis and autophagy induction associated with FEN treatment [19,37], however it is unknown if FEN can alter these pathways in adipocytes. We performed global lipidomics analysis in mature adipocytes to determine whether ceramide lipid species (i.e. sphingolipids) were altered. FEN specifically led to a  (Table 1). Interestingly, both FEN and RA exposure resulted in a complex remodelling of triacylglyceride species, with similar increases (from 2-to 18-fold) in longer chain and desaturated fatty acids ( Table 2). FEN and/or RA also decreased a number of triacylglyceride species (Tables 1 and 2). However, FEN decreased many more triacylglyceride lipid species (from 2 to 11-fold) and by a greater amount compared to RA. These findings are in agreement with a FEN-specific (RAR-independent mechanism) to decrease lipid accumulation in mature adipocytes ( Fig. 4 and Ref. [6]).

FEN completely normalises HFD induced impairments to mitochondrial function in vivo
Since excess ceramide synthesis has been negatively associated with mitochondrial function and insulin sensitivity, we performed global metabolomics analysis of PG-WAT from mice fed HFD +/À FEN treatment. HFD led to an accumulation of numerous acylcarnitines, indicative of impaired b-oxidation (Table 3). FEN completely normalised all of these elevations to levels measured in CHOW-fed mice (Table 3). FEN also completely prevented the rise in tricarboxylic acid (TCA)-cycle intermediates and oxidative stress markers (Table 3). This strongly suggests that FEN-treatment in vivo can alleviate disturbances to adipose tissue mitochondrial function resulting from HFD feeding.

FEN can induce markers of cellular stress and autophagy in mature adipocytes
To determine whether FEN's inhibition of ceramide biosynthesis was associated with the induction of cellular stress and survival responses as reported in cancer cell lines, we examined the levels of phospho-eIF2a, phospho-p38 and markers of autophagy. FEN increased levels of phospho-eIF2a (downstream of the PERK stress response pathway) and phospho-p38 (a mitogen activated protein kinase) (Fig. 6a). FEN also increased protein expression of Beclin1 and LC3B II which are required for autophagosome formation (Fig. 6a). FEN did not increase the expression of Atf4, the spliced isoform of Xbp1 or Chop. Surprisingly, FEN decreased gene expression levels of Atf4 and Chop, which may indicate the induction of pro-survival pathways rather than progression to apoptosis in 3T3-L1 cells (Fig. 6b).

FEN metabolite 4-OXO cannot inhibit 3T3-L1 adipogenesis
FEN can activate both RAR-dependent and -independent signalling pathways in 3T3-L1 cells. Previous studies have determined that 4-oxo-N-(4-hydroxyphenyl)retinamide, an oxidised FEN catabolite hereafter termed 4-OXO [41], operates in a RAR-independent manner, can manipulate ceramide biosynthesis similarly to FEN and is more potent than the parent compound [42,43]. We tested whether 4-OXO was responsible for the inhibition of adipogenesis and/or decreased lipid accumulation in 3T3-L1 cells. 4-OXO was unable to inhibit adipogenesis in 3T3-L1 cells, however, lipid accumulation was less than in VEH treated  (Fig. 7a). This effect was not due to toxicity as concentrations up to 10 mM were well tolerated (data not shown). Unexpectedly, 4-OXO resulted in increased expression levels of adipogenic markers (Fig. 7b). Retinoid genes were down regulated by differentiation and 4-OXO was unable to induce any of these genes above the levels observed in VEH (Fig. 7c). Thus, the catabolism of FEN to 4-OXO does not contribute to the inhibition of 3T3-L1 differentiation, however it appears 4-OXO may have nonretinoid, insulin-sensitizing properties. To test this further, insulin stimulated phosphorylation of Akt (phospho-Akt) was examined. VEH treated cultures responded poorly to insulin stimulation with only small increases in phospho-Akt levels observed when compared to saline controls (Fig. 7d). This is likely to be a result of the development of insulin resistance through chronic insulin treatment, which has been described previously [44]. 4-OXO increased phospho-Akt levels by almost 3-fold compared to VEH controls in the absence of insulin. 4-OXO further increased phospho-Akt levels with insulin stimulation indicating that treatment with 4-OXO results in increased insulin signalling in differentiated 3T3-L1 adipocytes (Fig. 7d).
Next we investigated whether 4-OXO exhibited other RAindependent signalling properties as observed with FEN treatment. 4-OXO led to increased levels of phospho-p38, phospho-eIF2 a and LC3B II indicating the induction of cellular stress responses and autophagy induction (Fig. 8). Thus, the RAindependent effects of FEN treatment are also exhibited by 4-OXO in 3T3-L1 adipocytes.

Discussion
The data presented here demonstrates that FEN can increase dihydroceramide and dihydrosphingomyelin lipid species independently of RAR-signalling, which is associated with activation of cellular stress responses and decreased lipid content in mature 3T3-L1 adipocytes. Moreover, our translational findings from mouse adipose tissue suggest that the manipulation of ceramide biosynthesis is also linked to normalisation of mitochondrial function in association with FEN-mediated inhibition of HFDinduced obesity and insulin resistance. This is in contrast to the mechanism of FEN action to inhibit 3T3-L1 differentiation, which we have demonstrated is mediated by ligand-induced activation of RARa signalling and induction of genes involved in retinoid metabolism (summarised in the graphical abstract).
Whereas RA was able to completely inhibit adipogenesis in the presence of ROSI, FEN was only able to partially prevent lipid accumulation. ROSI can stimulate adipogenesis via increased PPARg occupancy at nearby target genes but can also repress some genes that appear to be more dependent on C/EBPa [45]. Our results suggest that the ROSI-induced down-regulation of retinoid homeostasis genes can be overcome by RA via co-ordinate  Table 3 Metabolomics analysis of PG-WAT from HFD mice treated with FEN. Levels of carnitine, TCA cycle intermediates and oxidative stress markers in PG-WAT following global metabolite analysis by LC-MS. For each group n = 5 mice. All data represents mean and significance **P < 0.01 and ***P < 0.001 vs CHOW by one-way ANOVA with Tukey post hoc test. inhibition of C/EBPa and early induction of retinoid responsive genes such as Crbp1 and Rar gamma. Under these conditions, the less potent FEN could not repress C/EBPa or induce retinoid responsive genes and thus could not prevent adipogenesis. Although liganded RAR has been known to inhibit adipogenesis through the prevention of C/EBPb mediated transcription, recent studies have also pointed to an inhibitory role for CRBP1 in adipogenesis and triacylglyceride accumulation [3].
The ability of FEN treatment to decrease adipocyte lipid content in mature adipocytes was found to be RARa independent, consistent with previous reports that RA is unable to replicate this effect [6,27,28]. Moreover, dihydroceramide lipid species increased specifically in FEN-treated adipocytes, similarly to reported effects in carcinoma cells [37,46]. This effect appears to occur via direct inhibition of Des1 activity and mRNA, which catalyses the final step in de novo ceramide synthesis [43]. Fig. 8. 4-OXO induces cellular stress and autophagy markers. Western blot analysis of ER stress and autophagy markers in 3T3-L1 adipocytes differentiated with indicated compounds. Right panels are quantification of western blots in left panels. Proteins were normalised to total p38 or eIF2a levels, n = 6. Significance *P < 0.001 vs VEH by t-test. However, this is the first report that FEN can cause such alterations or in addition, inhibit Cers6 gene expression in adipocytes or adipose tissue. Thus, the complete normalisation of impaired mitochondrial b-oxidation and TCA cycle flux in adipose tissue suggests that in vivo, FEN may inhibit obesity and insulin resistance via inhibition of ceramide-induced insulin resistance and attenuation of mitochondrial dysfunction.
Accumulation of numerous acylcarnitines and ceramide linked to excess lipid supply and impaired mitochondrial fatty acid b-oxidation in skeletal muscle is associated with muscle insulin resistance [47,48]. Moreover, S. Summers and co-workers have shown FEN to directly prevent lipid induced insulin resistance in cultured myotubes and isolated muscles strips in association with increases in dihydroceramide levels [18]. However, insulin sensitization after treatment with the thiazolidinedione class of PPARg agonists is associated with increased adipose mitochondrial capacity and expression of adipose genes essential for branchedchain amino-acid oxidation, fatty acid b-oxidation, TCA-cycle and oxidative phosphorylation pathways [49,50]. Thus overall, FEN appears to correct these impairments in lipid homeostasis pathways in multiple insulin sensitive tissues in vivo [51][52][53][54].
The FEN metabolite 4-OXO [41] has been identified as biologically active and inhibits Des1 more potently than FEN [43]. Thus, 4-OXO may be partly responsible for some of the effects observed in response to FEN treatment [42]. We have revealed that 4-OXO is poor at inducing RAR-signaling and thus unable to inhibit 3T3-L1 adipogenesis. However, 4-OXO increased adipogenic biomarkers and insulin sensitivity. Thus, elevated p-AKT levels are probably due to the non-RAR signaling characteristics of FEN and FEN-like compounds, such as inhibition of ceramide synthesis [37,43]. These findings identify that 4-OXO may be a novel and previously unexplored therapeutic candidate to improve adipocyte insulin sensitivity without negatively impacting upon adipocyte differentiation or other non-beneficial effects of retinoid treatment.
We found that both FEN and 4-OXO, (but not RA), led to increases in markers of cellular stress and autophagy without inducing apoptotic cell death. Increased dihydroceramide species presents a potential non-RAR mechanism of FEN/4-OXO action linked to modulation of nutrient stress pathways. Regulation of these pathways has been shown to be impaired and contribute to the pathogenesis of diabetes [22,23]. Induction of ER stress and autophagy has been observed in variety of different cells types after FEN or dihydroceramide supplementation [19][20][21] and may account for the decrease in adipocyte lipid content in mature 3T3-L1 adipocytes. A recent study indicated that hypertrophic adipocytes fail to process autophagasomes through reduced autophagic flux [55]. This suggests defective autophagy may be associated with or even responsible for increases in adiposity and states of obesity. FEN and 4-OXO may increase autophagic flux and alleviate cellular disturbances that arise in hypertrophic adipocytes as a result of diet-induced obesity. However, further investigations will be required to determine the importance of this finding in vivo.
Overall, our findings indicate that FEN appears to display the unique ability to inhibit adipocyte differentiation and hypertrophy in vitro and these effects are mediated through distinct mechanistic pathways that are translated in vivo. We have confirmed that like RA, FEN inhibits adipocyte differentiation in vitro through RARa-dependent signaling. However, the non-RAR effects of FEN and its catabolite 4-OXO to modulate ceramide synthesis appear to be linked to improved mitochondrial function and modulation of cellular stress responses and autophagy. Future experiments will be required to demonstrate a causal molecular link between the alterations in ceramide synthesis and the observed metabolic outcomes. Our findings suggest that this unique combination of biological effects may be responsible for the low toxicity and beneficial effects of FEN treatment to inhibit obesity and insulin resistance.

Conflict of interest
No potential conflicts of interest relevant to this article were reported.
Author's contribution GDM and NM made contributions to all areas of the submitted work including study conception and design, acquisition, analysis and interpretation of data and drafting/revision of the work for intellectual content and context. SRT, BHM, LG, MKD, DGW, MD and PDW contributed to acquisition, analysis and/or interpretation of data. MD also contributed the study conception and design. NM takes overall responsibility for the work including final approval.