A combination of gas chromatography (GC)-MS and direct infusion (DI)-MS combined with multivariate pattern recognition was used to profile metabolism within the white adipose tissue of ob/ob mice treated with either GW610742 (a selective PPARδ agonist), GW347845 (a selective PPARγ agonist) or a vehicle control. These analytical approaches provided coverage of total fatty acids and intact lipids and free fatty acids, respectively. The various spectra and chromatograms were interrogated using multivariate statistics comparing the dosed groups with the vehicle control.

Both PPARδ and PPARγ agonists induced large changes in the total fatty acid profile of white adipose tissue as measured by GC-MS of the fatty acid methyl esters and subsequent multivariate analysis (Figure 1a-c). Treatment with the PPARδ agonist induced decreases in the medium-chain fatty acids, while the concentration of the shorter chain fatty acids increased (Figure 1d). This was contrasted by the effect of PPARγ where the most profound change was an increase in activity of Δ-9 desaturase, increasing the concentrations of desaturated fatty acids, as well as an increase in the long chain fatty acid arachidate C20:0.

Analysis of the DI-MS negative mode ionization data of the organic phase was used to analyze changes in free fatty acids and a number of classes of intact lipids, with this approach distinguishing both adipose tissue from ob/ob mice treated with either of the agonists (Figure 1e, f). Interrogating the loadings plots of the multivariate models, both agonists stimulated increases in the constituents of the ω-6 fatty acid pathway, demonstrating that both agonists stimulate the activity of desaturases. However, the most profound difference between the two agonists was characterized by decreased concentrations of the long chain saturated fatty acids (C19:0, C20:0, C21:0 and C22:0) following PPARδ agonist treatment, while free palmitic acid, stearic acid and its desaturated forms (C18:1 and C18:2) increased in compensation.

Given the important role adipose tissue plays in modulating the lipid composition of blood serum, liquid chromatography (LC)-MS was used to profile intact lipids in blood serum (Figure 1g). While both agonists induced changes in polar lipids, the most dramatic contrast was apparent in changes in the TAG content. The PPARδ agonist induced increases in the concentration of a number of circulating TAG species containing C16:0, C18:0 and C18:1 fatty acids in blood serum, while the same TAG species were decreased following PPARγ agonist treatment (Figure 1g). Thus, stimulation of PPARδ increased the mobilization of TAGs, and PPARγ stimulation increased the sequestration of TAGs.

These metabolic changes demonstrate that while both agonists induced the activity of desaturases, the PPARδ agonist was characterized by a reduction in fatty acid chain length consistent with increased β-oxidation. However, because both agonists influence metabolism in a range of organs, this study was complemented with an analysis of 3T3-L1 adipocytes to examine adipocyte metabolism in isolation.

To profile total fatty acid changes, GC-MS of fatty acid methyl esters was again applied in conjunction with multivariate statistics (Figure 2a). The loadings plots of the partial least squares-discriminant analysis (PLS-DA) models of the data were again used to determine the key metabolic changes in total fatty acid profiles induced by the two agonists (Figure 2b, c). Similar to the changes detected in adipose tissue, both agonists affected the ω-6 fatty acid pathway, with the PPARδ agonist increasing the concentrations of a number of the later pathway intermediates, while the PPARγ agonist decreased γ-linolenate, and increased dihomo-γ-linolenate. PPARδ activation also stimulated an increase in the end products of the ω-3 fatty acid pathway (C20:5, C22:5 and C22:6).

However, the major difference between the two agonists was a general decrease in the concentration of fatty acids observed in PPARδ agonist-treated cells) while PPARγ stimulation induced a relative change in overall chain length characterized by decreases in the concentrations of the medium chain fatty acids (C13:0, C14:0, C15:0, C16:1, C17:0, C17:1, C18:1) and a concomitant increase in the steady state concentrations of the long chain fatty acids (C20:0 and C22:0). These changes in total fatty acids were also represented in the free fatty acid profile measured by DI-MS and modeled by multivariate analysis (Figure 2d, e).

Changes in the composition of complex lipids was observed in the PPARδ activated 3T3-L1 adipocytes. An increase in the concentration of a number of glycerophosphocholine and phosphatidylcholine (PCs) species was ascertained and this was accompanied by a decrease in the concentration of specific TAGs (Table 1). Unlike the shift from TAGs to phospholipids induced by PPARδ, the activation of PPARγ produced a more complex remodeling of TAGs with an increase in longer chain and desaturated fatty acids, which dominated the resultant PLS-DA model (Table 1). In addition, a range of PCs, glycerophosphocholines, glycerophosphoethanolamines and glycerophosphoinsoitols decreased in concentration while the concentration of several cholesterol esters increased (data not shown).

A combination of both NMR spectroscopy and GC-MS analysis of aqueous metabolites readily distinguished the action of the two PPAR receptors. PPARδ activation increased the concentration of the peroxisomal oxidation product adipic acid, while PPARγ stimulation decreased the concentration of carnitine, the main transporter of fatty acids across the mitochondrion. The PPARδ agonist also decreased the concentration of glucose and other carbohydrates in adipose cells, as well as increased the concentration of citrate and glutamate (the latter in fast exchange with 2-oxogluturate). While PPARγ stimulation also decreased the concentration of glucose, it also decreased the concentrations of the later tricarboxylic acid (TCA) cycle metabolites. Changes in the steady state concentrations of specific metabolites and corresponding metabolic pathways in 3T3-L1 adipocytes treated with either the PPARδ or PPARγ agonist are summarized in Figure 2f, g.

To assess how metabolism in the 3T3-L1 cells influenced their environments, metabolite changes in the media were investigated using a combination of GC-MS, probing fatty acid export from the cells, and ¹H NMR spectroscopy, determining changes in aqueous phase metabolites. While the PPARδ agonist did not affect fatty acid export compared with cells treated with the vehicle control, PPARγ reduced the export of fatty acids, particularly of saturated fatty acids (palmitate, P = 0.01, 23% reduction; stearate, P = 0.04, 19% reduction). However, PPARδ activation markedly reduced the concentrations of amino acids in the PPARδ cell culture media compared with both the control group and cells treated with the agonist, in particular the BCAAs leucine (P < 0.0001), isoleucine (P = 0.002) and valine (P = 0.005) (Figure 2h). Concomitantly, the steady state intracellular concentration of valine was increased in PPARδ agonist-treated cells (P < 0.05)

In order to identify the metabolic mechanisms associated with PPARδ and PPARγ activation in white adipose tissue and 3T3-L1 adipocytes, the ¹³C-labeled substrates 1-¹³Cglucose and U-¹³C-palmitate were used to monitor flux through glycolytic and fatty acid oxidative pathways.

The use of 1-¹³C glucose and GC-MS readily distinguished the two agonists. Examination of the aqueous phase by GC-MS revealed that lactate, glutamate (readily labeled from the TCA cycle from labeled 2-oxoglutarate) and succinate from PPARδ agonist-treated cells were enriched with ¹³C when compared to control (Figure 3a). In contrast, the PPARγ agonist caused a reduction in labeling of lactate, succinate and glutamate compared to the vehicle-treated cells (Additional file 1). The PPARδ agonist also decreased labeling of the medium chain fatty acid palmitate from 1-¹³C glucose, while the PPARγ agonist increased the labeling of the long chain fatty acid arachidate. In addition, ¹³C NMR spectroscopy of the organic fraction of control and PPARδ agonist-treated adipocytes incubated in media containing 1-¹³C glucose showed that glycerol and esterified glycerol from PPARδ agonist-treated cells had reduced enrichment compared with the control group (Additional file 2).

Similarly, the labeled substrate U-¹³C palmitate readily distinguished the two agonists. Assessment of the aqueous phase by GC-MS indicated that several TCA cycle intermediates from PPARδ agonist-treated adipocytes were enriched compared to control cells (Figure 3b). Investigation of the organic phase by GC-MS demonstrated that the fatty acids downstream of palmitic acid in the β-oxidation pathway showed greater ¹³C enrichment in PPARδ agonist-treated cells; as did the Δ-9 desaturation product of palmitic acid. Simultaneously, the enrichment of fatty acids upstream of palmitic acid in the fatty acid synthesis pathway was reduced in PPARδ agonist-treated cells (Figure 3b).

GC-MS analysis of the aqueous phase of cells incubated in U-¹³C palmitate indicated that the early TCA cycle intermediates exhibited decreased ¹³C enrichment in PPARγ agonist-treated adipocytes when compared to control adipocytes (Additional file 1). Assessment of the organic phase by GC-MS indicated that the ¹³C enrichment of the long chain fatty acid arachidate was increased in the PPARγ agonist-treated cells when compared to control (Additional file 1). Concurrently, the ¹³C enrichment of the shorter chain fatty acid myristate was decreased.

To further characterize the PPARδ induced upregulation of oxidative pathways in adipocytes, the oxygen consumption of PPARδ agonist-treated and control 3T3-L1 cells was measured both when using fatty acid as substrate and during isolated electron transport chain complex IV oxidation using in situ studies in a Clarke type oxygen electrode. Both complex IV and fatty acid oxidation were significantly increased in the adipocytes exposed to the PPARδ agonist when compared to control adipocytes (Figure 4a, b). This was accompanied by a profound decrease in TAGs as measured by Oil Red O staining of neutral lipids (Figure 4c).

The combination of steady state metabolomic changes in adipose tissue and adipocytes and isotope labeling studies indicated a profound upregulation of glucose and fatty acid oxidation following PPARδ activation. To investigate these changes in more detail, we moved focus to the transcriptome using microarray analysis of PPARδ activation in adipocytes. Of the 45,281 probes utilized, 13,718 were expressed above the background defined by the negative control probe. From these, 2,349 were determined to be differentially expressed with a 95% confidence level between PPARδ agonist-treated and control 3T3-L1 adipocytes. In addition to the univariate analysis, multivariate models were also built using the total normalized data (Figure 5a). The 6% of transcripts most responsible for separation in the multivariate models were then examined (3% with the highest positive contribution to principal component 1 and 3% with the highest negative contribution to principal component 1 in PPARδ agonist-treated cells as identified in the multivariate models). The multivariate analysis indicated that the mRNA of genes involved in a number of key metabolic pathways was altered following PPARδ activation. The Reactome Skypainter tool was then utilized to determine which pathways and reactions were statistically overrepresented by the 3% most increased and 3% most decreased transcripts in PPARδ agonist-treated cells identified in the multivariate models (Table 2; Figure 5b) 13 .

The expression of genes encoding proteins involved in the mitochondrial β-oxidation pathway and the peroxisomal fatty acid β-oxidation pathway was increased in PPARδ agonist-treated cells. Alongside these changes were an increase in the transcription of genes involved in both mitochondrial and peroxisomal biogenesis and maintenance. The transcription of several genes whose products play a role in the glycolytic metabolic pathway and the TCA cycle was also upregulated in PPARδ agonist-treated cells. In addition, there was a detected increase in the concentrations of expressed mRNA for components of the electron transport chain, and genes involved in fatty acid desaturation (Table 2; Additional file 3).

The results from the steady state metabolomic experiments in adipose tissue and adipocytes and the isotope labeling studies suggest PPARγ activation has a pronounced effect on glucose utilization and fatty acid synthesis and metabolism in adipocytes. Transcriptional changes were investigated by DNA microarrays to further define the changes associated with PPARγ activation in adipocytes. Of the 45,281 probes utilized, 13,755 were expressed above the background defined by the negative control probe. From these, 3,282 were determined to be differentially expressed with a 95% confidence limit. Multivariate models were then built using the total normalized data (Figure 5c). The 6% of transcripts most responsible for separation in the multivariate models (that is, those contributing most to the total variance of the multivariate model) were then examined using a combination of multivariate analysis and the Reactome Skypainter tool as described above 13 . The pathways and reactions that were statistically overrepresented by the 3% most increased and 3% most decreased transcripts in PPARγ agonist-treated cells identified in the multivariate models are shown in Table 3, Figure 5c, and Additional file 4.

The expression of genes encoding proteins involved in the glycolytic metabolic pathway were upregulated in PPARγ agonist-treated cells. In addition, the expression of the gene encoding the TCA cycle enzyme isocitrate dehydrogenase was identified as decreased, suggesting that citrate was being channeled to fatty acid synthesis rather than being metabolized by the TCA cycle. PPARγ activation was also discerned to significantly affect the transcription of genes responsible for the remodeling and metabolism of lipids. Genes for fatty acid desaturases (Scd2 and Fads3) were increased in expression following PPARγ activation. The transcripts of a number of genes that favor conditions of fatty acid synthesis were also increased in concentration in the adipocytes following treatment with the PPARγ agonist. Concomitantly, the expression of genes encoding enzymes that catalyze the hydrolysis of medium and long chain acyl-CoAs to free fatty acids and coenzyme A (CoA) was upregulated in the treated adipocytes (Acot7 and Nudt19). In addition, the transcription of an insulin responsive fatty acid transporter gene (Slc27a1) responsible for the import of long chain fatty acids into adipose tissue undergoing high levels of TAG synthesis was increased.

Several genes involved in the restructuring and remodeling of complex lipids were also affected by PPARγ activation. There was an increase in transcription of genes encoding enzymes responsible for conversion of lysophospholipids to phospholipids, favoring polyunsaturated fatty acyl-CoAs as acyl donors (lysophosphatidylcholine acyltransferase 3 acyltransferase). In addition, mRNA transcripts of genes encoding products that regulate lipolysis, alongside other metabolic processes, including gluconeogenesis, were increased in the PPARγ agonist-treated cells (platelet activating factor acetylhydrolase 2 lipase, angiopoietin-related protein 4 and nuclear receptor corepressor 1). Additionally, transcription of the PPAR transcriptional coactivator gene chromodomain helicase DNA binding protein 9 was upregulated in the PPARγ agonist-treated adipocytes.

Several transcripts were increased in both PPARδ and PPARγ agonist-treated cells, principally involved with glycolysis and lipid metabolism. However, PPARδ activation was unique in its effect on the citric acid cycle, the electron transport chain and fatty acid β-oxidation (Tables 2 and 3).

All animal studies were performed within the relevant local legislation. Two-month-old male ob/ob mice (Jackson Labs, Bar Harbor, ME, USA) were fed standard laboratory chow ad libitum under controlled temperature, lighting and humidity. During the studies, body weight and food consumption (cage average) were recorded. The ob/ob mice were weight matched (mean weight 46 ± 1 g), assigned to three groups of eight and dosed orally daily with vehicle control, the PPARδ agonist GW610742 (30 mg/kg) or the PPARγ agonist GW347845 (5 mg/kg). Serum was collected via cardiac stick under isoflourane anesthesia at completion of the study on day 15. White adipose tissue (gonadal fat pad) was rapidly dissected (< 60 s post mortem), snap frozen in liquid nitrogen and stored at -80°C until extraction.

3T3-L1 preadipocytes were grown in T75 flasks and maintained in DMEM (high glucose 4.5 g/l; Sigma-Aldrich, Gillingham, Dorset, UK) supplemented with 10% (v/v) new born calf serum (Sigma-Aldrich), 50 units/ml penicillin, and 50 μg/ml streptomycin (Sigma-Aldrich) in a humidified 5% CO₂ incubator at 37°C. At 2 days post-confluence cells were induced to differentiate with DMEM supplemented with 10% (v/v) fetal bovine serum (FBS; Invitrogen, Paisley, Renfrewshire, UK), 1 μM dexamethasone (Sigma-Aldrich), 0.5 mM isobutylmethylxanthine (Sigma-Aldrich), 100 nM insulin (Sigma-Aldrich), 50 units/ml penicillin, and 50 μg/ml streptomycin. The cells were maintained in this media for 72 h as this was found to improve the reproducibility of differentiation between flasks. After 72 h the medium was replaced with DMEM supplemented with 10% FBS, 100 nM insulin, 50 units/ml penicillin, and 50 μg/ml streptomycin. The medium was subsequently changed for DMEM supplemented with 10% FBS, 50 units/ml penicillin, and 50 μg/ml streptomycin every 48 h 32 .

At day 11 post-induction the medium on the cells was replaced with DMEM supplemented with 10% FBS, 100 nM insulin, 50 units/ml penicillin, and 50 μg/ml streptomycin containing DMSO (control; n = 6), the PPARδ selective agonist GW610742 (n = 6 at 100 nM and 1 μM) or the PPARγ selective agonist GW347845 (n = 6 at 10 nM and 100 nM) for 2 days prior to cell collection and metabolite extraction. These doses were based on the specific affinities of the compounds for their respective receptors.

Cells were collected by removing the medium and washing each T75 flask with 10 ml of phosphate-buffered saline. Cells were then washed with 1.5 ml trypsin-EDTA solution (5 BAEE units trypsin/ml, 1.8 μg EDTA/ml; Sigma-Aldrich) for 2 minutes at 37°C to remove the cells from the surface of the flask. DMEM (8.5 ml) supplemented with 10% (v/v) new born calf serum, 50 units/ml penicillin, and 50 μg/ml streptomycin was added to each flask. The DMEM containing the cells was transferred to a falcon tube and centrifuged at 200 g for 2 minutes to pellet the cells. The remaining medium was removed and the cells washed with physiological saline (0.9% NaCl) solution, and 2 ml of media was stored for further analysis.

Metabolites were extracted from white adipose tissue and 3T3-L1 cells using a modified Bligh and Dyer method 33 . Frozen white adipose tissue (approximately 100 mg for NMR and approximately 50 mg for GC-MS analysis) was pulverized with liquid nitrogen. Methanol-chloroform (2:1, 600 μl) was added to the white adipose tissue, serum (50 μl), and to 5 mg cell pellets (3T3-L1 cells) and the samples were sonicated for 15 minutes. Chloroform-water (1:1) was then added (200 μl of each). Samples were centrifuged (16,100 g, 20 minutes) and the organic and aqueous phases were separated and stored at -80°C until analysis. Of these fractions, 100 μl of the organic phase was used for LC-MS, and the remaining organic phase was used for GC-MS. Prior to analysis the organic fractions were dried in a fume hood. For the aqueous phase, 100 μl of the aqueous phase was taken for GC-MS analysis, and the remaining aqueous phase sample was used for ¹H NMR spectroscopy.

Dried extracts were dissolved in 600 μl of D₂O and buffered in 0.24 M sodium phosphate (pH 7.4) containing 1 mM TSP (sodium-3-(trimethylsilyl)-2,2,3,3-tetradeuteriopropionate; Cambridge Isotope Laboratories, Andover, MA, USA) and 0.02% sodium azide. Samples were analyzed using a DRX Avance II+ spectrometer interfaced to a 5-mm TXI ATMA probe (Bruker BioSpin GmbH, Rheinstetten, Germany) at a proton frequency of 500.13 MHz. A presaturation pulse sequence for water suppression based on a one-dimensional nuclear Overhauser effect spectroscopy pulse sequence was used to saturate the residual water proton signal (relaxation delay = 2 s, t_{1 =} 4 μs, mixing time = 50 ms). We collected 128 and 256 transients for white adipose tissue extracts and 3T3-L1 cell extracts, respectively, into 64 K data points over a spectral width of 8,000 Hz at 300 K. NMR spectra were processed in ACD 1D NMR Manager (Advanced Chemistry Development Inc., Toronto, Canada). The NMR spectra were integrated using 0.04 ppm integral regions between 0.2 and 9.56 ppm (excluding water resonance between 4.20 and 5.08 ppm). Spectra were normalized to total integrated area to account for differences in concentration between samples and assigned by comparison with previous literature and Chenomx NMR suite 5.0 libraries.

Dried aqueous phase samples were derivatized using methoxyamine hydrochloride solution (20 mg/ml in pyridine; Sigma-Aldrich) and 30 μl of N-methyl-N-trimethylsilyltrifluoroacetamide (Macherey-Nagel, Duran, Germany) using the method described by Gullberg et al. 34 .

Acid-catalyzed esterification was used to derivatize the organic phase samples. Chloroform-methanol (1:1, 0.25 ml) and BF₃-methanol (10%, 0.125 ml) was added to the organic phase and incubated at 90°C for 90 minutes. Water (0.15 ml; ultrapurified to resistivity 18.2 MΩ cm) and hexane (0.3 ml) were added and the samples vortex mixed for 1 minute and left to form a bilayer. The aqueous phase was discarded and the organic layer evaporated to dryness prior to reconstitution in analytical grade hexane (200 μl) before GC-MS analysis. All GC-MS analyses were made using a Trace GC Ultra coupled to a Trace DSQ II mass spectrometer (Thermo Scientific, Hemel Hempstead, UK). Derivatized aqueous samples were injected splitless onto a 30 m × 0.25 mm 5% phenylpolysilphenylene-siloxane column with a 0.25 μm ZB-5 ms stationary phase (Phenomenex, Macclesfield, Cheshire, UK) as described in Roberts et al. 32 . Full-scan spectra were collected using three scans/s over a range of 50 to 650 m/z.

The derivatized organic samples were injected with a split ratio of 60 for white adipose tissue and 8 for 3T3-L1 cells onto a 30 m × 0.25 mm 70% cyanopropyl polysilphenylene-siloxane 0.25 μm TR-FAME stationary phase column (Thermo Scientific) as described above and by Roberts et al. 32 .

GC-MS chromatograms were processed using Xcaliber (version 2.0; Thermo Scientific). Each individual peak was integrated and then normalized. Overlapping peaks were separated using traces of single ions. Peak assignment was based on mass fragmentation patterns matched to the National Institute of Standards and Technology (USA) library and to previously reported literature. Identification of metabolites from organic phase GC-MS analysis was supported by comparison with a FAME standard mix (Supelco 37 Component FAME Mix; Sigma-Aldrich).

Chromatography was performed using an ACQUITY UPLC^® system (Waters Corporation, Centennial Park, Elstree, Hertfordshire) equipped with an Acquity UPLC 1.7 μm Bridged Ethyl Hybrid (BEH) C8 column (2.1 × 100 mm Waters), which was kept at 65°C and coupled to a Micromass QTof-Ultima™ with a Z-spray™ electrospray source as described by Roberts et al. 32 . The binary solvent system used was solvent A comprising HPLC grade water (Sigma-Aldrich), 1% 1 M ammonium acetate (NH₄Ac; Sigma-Aldrich) and 0.1% formic acid (Sigma-Aldrich) and solvent B comprising analytical grade acetonitrile (chromosolv, Sigma-Aldrich)/isopropanol (Fisher Scientific, Loughborough, Leicestershire, UK) 5:2, 1% 1 M NH₄Ac, and 0.1% formic acid 35 . Mass spectrometric data were collected in full scan mode from 100 to 1,350 m/z for adipose and 100 to 1,500 m/z for 3T3-L1 cells from 0 to 14 minutes with a scan duration of 0.5 s and an interscan delay of 0.1 s.

For tissue extracts, the column mobile phase was held at 85% solvent B for 0.5 minutes followed by an increase from 85 to 100% solvent B over 0.5 to 8 minutes. The mobile phase was then held at 100% solvent B for 4 minutes. Between 12 and 12.25 minutes the mobile phase was returned to 85% solvent B held for 1.75 minutes to re-equilibrate the column. For serum extracts, the column mobile phase was held at 70% solvent B for 0.5 minutes followed by an increase from 70 to 100% solvent B over 0.5 to 6.5 minutes. The mobile phase was then held at 100% solvent B for 3.5 minutes. Between 10 and 10.25 minutes the mobile phase was returned to 70% solvent B held for 3.75 minutes to re-equilibrate the column. For 3T3-L1 cell organic phase metabolites, the column mobile phase was held at 50% solvent B for 0.5 minutes followed by an increase from 50 to 100% solvent B over 0.5 to 6.5 minutes. The mobile phase was then held at 100% solvent B for 3.5 minutes. Between 10 and 10.25 minutes the mobile phase was returned to 50% solvent B held for 3.75 minutes to re-equilibrate the column. The total ultra performance liquid chromatography (UPLC) cycle was 14 minutes and the eluent flow rate was 600 μl/minute for both methods.

Tandem mass spectrometry (MS/MS) was used for the identification of selected lipids. Data were processed using Micromass Markerlynx Applications Manager (Waters Corporation).

Mass spectrometric analysis was also performed using a Thermo Finnigan LTQ equipped with a Finnigan Surveyor pump and Finnigan Micro AS Autosampler Thermo Finnigan, Hemel Hempstead, Hertfordshire, UK.

The 3T3-L1 organic phase samples for DI-MS were reconstituted in 500 μl methanol:tetrahydrofuran (2:1, v/v). Samples were analyzed in triplicate using both positive and negative mode. The scan range was set at 100 to 1,100 m/z in profile for both positive and negative mode. DI-MS chromatograms were processed using Xcaliber (version 2.0; Thermo Electron). The mass data were summed from the chromatogram for the period of sample injection and the exact masses were exported; the data points were summed between M and M+1, normalized to total metabolite concentration and integrated. Tandem mass spectrometry data were collected for identification purposes.

At 2 days post-differentiation medium was removed from the T75 flasks and replaced with DMEM (10% (v/v) fetal bovine serum, 50 units/ml penicillin, and 50 μg/ml streptomycin) and either 4.5 g/l unlabeled glucose with DMSO control (n = 6), 1 μM GW610742 (n = 7) or 1 μM GW347845 (n = 7), or 4.5 g/l 1-¹³C-glucose with DMSO control (n = 7), 1 μM GW610742 (n = 7) or 1 μM GW347845 (n = 7). After 2 days cells were collected and metabolites extracted as previously described.

Palmitate was solubilized using a dialyzed albumin solution. At 2 days post-differentiation medium was removed from the T75 flasks and replaced with DMEM (serum free, 50 units/ml penicillin, and 50 μg/ml streptomycin) and either 70 μM unlabeled palmitate with DMSO control (n = 6), 1 μM GW610742 (n = 7) or 1 μM GW347845 (n = 7), or 70 μM U-¹³C labeled palmitate with DMSO control (n = 6), 1 μM GW610742 (n = 7) or 1 μM GW347845 (n = 7). After 2 days cells were collected and metabolites extracted as previously described.

Dried organic phase extracts were dissolved in 600 μl of deuterated chloroform. Samples were analyzed using a DRX Avance II+ spectrometer interfaced to a 5-mm TXI ATMA probe. Analysis was performed using two-dimensional H-1/X correlation via double inept transfer with sensitivity improvement. Spectral widths of 10.00 ppm and 160 ppm were used in the F2 (1H) and F1 (13C) dimensions, respectively, with an offset of 75.00 ppm. Spectra were acquired using 96 scans with a relaxation delay of 1.0 s. Datasets were zero-filled and multiplied by sine bell squared functions prior to Fourier transformation.

Analysis of organic and aqueous phases was carried out as previously described above. Enrichment of metabolites was identified by calculating isotope ratios of the M and M+1 ions for the parent ion of the fragmentation pattern in the case of ¹³C-glucose metabolism analysis and TCA cycle intermediates originating from ¹³C-palmitate oxidation. For fatty acid synthesis and desaturation products from ¹³C-labeled palmitate an ion ratio of M+16/M was used, and for fatty acids originating from oxidation of ¹³C-labeled palmitate an ion ratio of M+n/M was used, where n = the carbon chain length of the fatty acid. Statistical analysis was performed using a univariate Student t-test.

Multivariate data analysis was performed using SIMCA-P⁺ 11.0 (Umetrics AB, Umeå, Sweden). NMR, DI-MS and UPLC-MS data sets were mean-centered and Pareto-scaled prior to analysis. GC-MS data sets were scaled to unit variance (UV) as only manually fitted peaks were analyzed. Data sets were analyzed using principal components analysis and PLS-DA. Metabolite changes responsible for clustering or regression trends within the pattern recognition models were identified by interrogating the corresponding loadings plot. Metabolites identified in the variable importance in projections/coefficients plots were deemed to have changed globally if they contributed to separation in the models with a confidence limit of 95%.

Cells were grown, treated with either vehicle control, the PPARδ agonist or the PPARγ agonist for 2 days prior to collection into respiration medium (100 mM KCl, 50 mM MOPS (3-(N-morpholino)propanesulfonic acid), 1.0 mM KH₂PO₄, 1.0 mg/ml defatted bovine serum albumin, pH 7.4). Respiratory rates of in situ permeabilized 3T3-L1 cells were measured using a Clark-type oxygen electrode (Strathkelvin Instruments Ltd, Glasgow, UK) 36 . Respiration rates were recorded and quantified using 782 Oxygen System v3.0 software (Strathkelvin Instruments). Oxygen concentrations were measured continuously in 0.5 ml respiration medium containing 250,000 cells in a respiration chamber maintained at 37°C for 40 minutes. The cells were initially permeabilized with the addition of Digitonin (25 μg/ml), before malate (5 mM) plus palmitoyl-carnitine (0.04 mM) were added as respiratory substrates to measure the fatty acid oxidation rates. Respiration was stimulated by the addition of a saturating concentration of ADP (2 mM) plus MgCl₂ (0.6 mM) and subsequently measured. Antimycin (5 μM) was then added to inhibit complex III of the electron transport chain, and respiration ceased. Complex IV respiration was stimulated by addition of the artificial substrates TMPD (N, N, N', N'-tetramethyl-p-phenylenediamine dihydrochloride, 0.5 mM) and ascorbate (2 mM). Finally, respiration was terminated by addition of the complex IV inhibitor sodium azide (3 mM).

3T3-L1 adipocytes were cultured and then treated with either the PPARδ agonist, the PPARγ agonist or vehicle control as described above (n = 6 independent individual hybridizations for each treatment group). RNA was extracted using RNeasy (Qiagen GmbH, Hilden, Germany). Approximately 5 mg of cells was used per sample for RNA isolation. Procedures were carried out according to the manufacturer's instructions. Extracted RNA was quantified and its purity assessed using a Nanodrop ND-1000 Spectrometer (Nanodrop Technologies Inc., Wilmington, NC, USA) to measure the absorbance at 260 nm and the A₂₆₀/A₂₈₀ ratio, respectively. Illumina Infinium Gene Expression BeadArrays (Illumina Inc., San Diego, CA, USA) were used to perform transcriptomics. A mouse WG6 array platform was used with 45,281 probes. Analysis was performed with R/BioConductor version 2.5. The R package lumi 37 was used with the detection P-value threshold set to 0.01. Probes were required to be successfully detected (P-value < 0.01 in Lumi) in at least one sample to pass the selection. The data were transformed using variance stabilization 38 and then normalized using quantile normalization. Gene expression was compared using the R package limma 39 with a 95% confidence interval. The selected and normalized data were then analyzed using Simca-P+. The 6% of transcripts most responsible for separation in the multivariate models were then examined (3% most increased and 3% most decreased in treated cells as identified in the multivariate models). The microarray data have been deposited with the Gene Expression Omnibus and have the accession number [GSE26207].

The Reactome Skypainter tool was used to determine which pathways were statistically significant in terms of key perturbations 13 . From a given set of genes participating in a pathway, the total genes for Mus musculus and the submitted genes (genes increased in PPARδ- or PPARγ-activated cells) of which N genes participate in a pathway, the probability of observing at least N genes from a pathway if that pathway is not overrepresented in the submitted list of genes is calculated using Fisher's exact test. A P-value smaller than the significance level suggests the pathway is significantly represented.

Univariate analysis was performed using an unpaired Student's t-test with a significance level set to P < 0.05. An F-test was also utilized to compare the variance of two distributions. All univariate analysis was conducted in GraphPad Prism (version 4).