Peroxisomal and Microsomal Lipid Pathways Associated with Resistance to Hepatic Steatosis and Reduced Pro-inflammatory State*

Accumulation of fat in the liver increases the risk to develop fibrosis and cirrhosis and is associated with development of the metabolic syndrome. Here, to identify genes or gene pathways that may underlie the genetic susceptibility to fat accumulation in liver, we studied A/J and C57Bl/6 mice that are resistant and sensitive to diet-induced hepatosteatosis and obesity, respectively. We performed comparative transcriptomic and lipidomic analysis of the livers of both strains of mice fed a high fat diet for 2, 10, and 30 days. We found that resistance to steatosis in A/J mice was associated with the following: (i) a coordinated up-regulation of 10 genes controlling peroxisome biogenesis and β-oxidation; (ii) an increased expression of the elongase Elovl5 and desaturases Fads1 and Fads2. In agreement with these observations, peroxisomal β-oxidation was increased in livers of A/J mice, and lipidomic analysis showed increased concentrations of long chain fatty acid-containing triglycerides, arachidonic acid-containing lysophosphatidylcholine, and 2-arachidonylglycerol, a cannabinoid receptor agonist. We found that the anti-inflammatory CB2 receptor was the main hepatic cannabinoid receptor, which was highly expressed in Kupffer cells. We further found that A/J mice had a lower pro-inflammatory state as determined by lower plasma levels and IL-1β and granulocyte-CSF and reduced hepatic expression of their mRNAs, which were found only in Kupffer cells. This suggests that increased 2-arachidonylglycerol production may limit Kupffer cell activity. Collectively, our data suggest that genetic variations in the expression of peroxisomal β-oxidation genes and of genes controlling the production of an anti-inflammatory lipid may underlie the differential susceptibility to diet-induced hepatic steatosis and pro-inflammatory state.

The current view of obesity-associated metabolic deregulations proposes that excessive accumulation of fat in adipose tissue initiates an inflammatory reaction characterized by cytokine production by adipocytes and infiltrating leukocytes (1,2). This leads to the development of insulin resistance in adipocytes and increased release of free fatty acids. Secreted cytokines and free fatty acids then induce insulin resistance in muscle and liver, through activation of serine/threonine kinases that phosphorylate essential signaling proteins such as the insulin receptor or its main substrates IRS1 and IRS2 (3). In the liver, insulin resistance leads to unrestrained glucose production causing hyperglycemia. However, insulin still normally activates the lipogenic transcription factor SREBP-1c leading to hepatosteatosis, increased secretion of VLDL, and elevated plasma triglycerides (4). Therefore, there is a correlation between the degree of inflammation in the adipose tissue and liver fat content and insulin resistance (5), and studies in human have demonstrated a correlation between the level of fat accumulation in the liver and various parameters of the metabolic syndrome (6,7).
Changes in liver metabolism may, however, also initiate a cascade of events that lead to generalized inflammation and insulin resistance. For instance, selective inactivation of the insulin receptor in mouse liver induces hepatic insulin resistance, dyslipidemia, and increased susceptibility to atherosclerosis (8). Also, liver-specific activation of the NF-B pathway (9) increases hepatic production of cytokines and insulin resistance both in the liver and in peripheral tissues. In contrast, inhibiting the NF-B pathway by transgenic expression of an IB super-repressor in hepatocytes prevents high fat diet-induced cytokine production and obesity (9). Thus, although there is strong support for an important role of fat tissue expansion during HFD 4 feeding as an initiator of systemic insulin resistance, many studies clearly indicate that alterations in liver metabolism and inflammatory state may also lead to general metabolic deregulation (10).
The presence of diverse mechanisms leading to metabolic deregulations is also supported by recent genome-wide association studies that identified many gene variants, each causing only a small increase in the risk of disease development (11). In addition, epigenetic modifications, such as DNA methylation or histone modifications (12)(13)(14), may also have an impact on gene expression and explain, for instance, the differential metabolic adaptation of genetically identical mice to high fat diet feeding (15,16).
Here, we studied C57Bl/6J (B6) and A/J mice, which are sensitive and resistant, respectively, to HFD-induced hepatosteatosis and obesity even though their food intake is the same (17)(18)(19). We confirmed these previous observations and further showed that A/J mice had a lower proinflammatory state as revealed by lower plasma concentrations of IL-1␤ and G-CSF and lower expression of these cytokines by Kupffer cells. By transcriptomic analysis, we found that A/J mouse livers had a coordinated up-regulation of several peroxisomal genes and increased peroxisomal ␤-oxidation capacity. They also expressed higher levels of the microsomal elongase Elovl5 and of the desaturases Fads1 and Fads2; lipidomic analysis showed this to be associated with higher concentrations of arachidonic acid-containing lipids, including 2-arachidonylglycerol, an agonist of the anti-inflammatory CB2 receptor that is highly expressed in Kupffer cells. Thus, our transcriptomic and lipidomic analysis reveals that elevated activity of lipid metabolic pathways may confer resistance to HFD-induced hepatosteatosis and reduce the pro-inflammatory state of A/J mice.

EXPERIMENTAL PROCEDURES
Animals and Diet-Male C57BL/6J (Janvier, France) and A/J (The Jackson Laboratory via Charles River, Germany) mice were fed normal chow or high fat diet for different times as indicated in the text. Each experimental group included six mice per strain, treatment, and time point. HFD feeding of mice started at 5-6 weeks of age. HFD diet contained 58% fat, 26% carbohydrates, 16% proteins (Research Diet D12331; Normal Chow diet NAFAG 9333). Animals were housed six per cage in a temperature-controlled room (20 -22°C) with 12-h light/ dark cycles (light period 7 a.m. to 7 p.m.). Food and water were available ad libitum unless noted. For tissue collection, mice were killed in the fed state after isoflurane anesthesia. Tissue samples were rapidly dissected, snap-frozen in liquid nitrogen, and stored at Ϫ80°C. Animals were treated in accordance with our Institutional Guidelines and with the authorization of the Service Vétérinaire Cantonal.
Histology and Plasma Biochemical Analysis-Mice were perfusion-fixed with a 4% paraformaldehyde solution, and the livers were excised, placed on fixative overnight, and then equilibrated in sucrose solution overnight at 4°C. Lipid histological detection was performed with Sudan III staining of 5-m frozen sections.
Isolation and Purification of Liver Cells-Hepatic sinusoidal cells (stellate cells, sinusoidal endothelial cells, and Kupffer cells) from murine liver were isolated and cultured based on a method described previously (20). Briefly, murine livers were perfused with 0.02% collagenase, 0.04% Pronase and further digested in vitro with 0.04% collagenase, 0.04% Pronase, and 0.001% DNase for 20 min at 37°C under constant stirring. This suspension was filtered and centrifuged for 2 min at 50 ϫ g (4°C) to remove the remaining hepatocytes. The cells were then centrifuged for 8 min at 500 ϫ g (4°C) and resuspended in Hanks' balanced saline solution. After centrifuging the suspension for another 8 min at 500 ϫ g (4°C), the pellet was resuspended in buffer without Ca 2ϩ and with EGTA to avoid clumping of cells. The resulting cell suspension was then sorted on a FACSAria with a UV laser (355 nm). The 355-nm light excites the vitamin A contained in the lipid droplets of hepatic stellate cells, resulting in an autofluorescent signal between 425 and 475 nm. Sorting on this basis yielded a Ͼ98% pure hepatic stellate cell population. Sinusoidal endothelial cells were tagged with FITC-labeled anti-LSEC antibodies (Miltenyi, Bergisch Gladbach, Germany) prior to cell sorting; Kupffer cells were tagged with phycoerythrin-labeled antibodies to CD11b (Miltenyi).
RNA Preparation, cDNA Microarray Hybridizations, and Statistical Analysis-Frozen livers (50 -150 mg) were homogenized in 2 ml of PeqGOLD TriFast TM (Peqlab), and total RNA was isolated after chloroform/isoamyl alcohol (24:1) extraction. RNA was precipitated with isopropyl alcohol and purified by passage over RNeasy columns (Qiagen). RNA quality was checked before and after amplification with a Bioanalyzer 2100 (Agilent). RNA was reverse-transcribed, and RNA was amplified with MessageAmp TM kit (Ambion). RNAs were labeled by an indirect technique with Cy5 and Cy3 as published previously (15). Labeled RNAs were hybridized to microarrays containing 17,664 cDNAs prepared at the DNA Array Facility of the University of Lausanne. RNAs from each HFD-treated mouse (n ϭ 6 per time point) were separately co-hybridized with the labeled samples prepared from the RNA pooled from six NC-fed mice from the corresponding time point. Scanning, image acquisition, and quality control analyses were performed as published previously (15). Data were expressed as log2 intensity ratios (Cy5/Cy3), normalized with a print tip locally weighted linear regression (Lowess) method, and filtered based on spot quality and incomplete annotation.
Genes differentially expressed in HFD compared with NCfed mice were identified by fitting a linear model to the normalized data for each gene and calculating an empirical Bayes t statistic using the Bioconductor limma package in R (21). All p values were adjusted for multiple comparisons using Benjamini Hochberg correction. Genes were considered significantly upor down-regulated with an adjusted p value Յ0.01, which corresponds to a false discovery rate of 1%.
For gene set enrichment analysis (GSEA), gene lists at each time point ranked according to the t statistic were prepared. Enrichment of genes in GO cellular component gene sets was calculated using GSEA as described, with 1000 permutations (22). Hierarchical clustering of selected genes related to fatty acid oxidation was performed in R using Pearson correlation with complete linkage as a distance measure.
Quantitative RT-PCR was performed using the 7500Fast real time PCR ABI system (Applied Biosystems). All samples were run in duplicate. Relative expression levels were determined using a five-point serially diluted standard curve, generated from cDNA of all pooled samples. Gene expression was expressed in arbitrary units and normalized relative to the housekeeping gene, ␤ 2 -microglobulin (Fwd 5Ј-CACTGACCG-GCCTGTATGC-3Ј and Rev 5Ј-GGTGGCGTGAGTATACT-TGAATTTG-3Ј) or TATA box-binding protein (TBP Fwd 5Ј-ATCCCAAGCGATTTGCTGC-3Ј and Rev 5Ј-ACTCTTGG-CTCCTGTGCACA-3Ј) to compensate for differences in cDNA loading.
Western Blot-100 mg of frozen liver samples were homogenized with a Polytron in 1 ml of tissue buffer (Tris-HCl, pH 6.8, 80 mM EDTA, 5 mM SDS 5%, 4 mM Na 3 VO 4 , and 10 mM NaF, 1 tablet of protease inhibitor mixture (Sigma) per 50 ml of tissue buffer). Proteins were quantified using the BCA protein assay reagent kit (Pierce), and 25 g were separated on 10% SDS-PAGE and transferred onto nitrocellulose membrane by electroblotting. Membranes were incubated with the following primary antibodies: catalase (Abcam), L-PBE, acyl-coenzyme A oxidase (kindly provided by Mustapha Cherkaoui-Malki, Université de Bourgogne, Dijon, France), and ␤-actin (Sigma). The primary antibodies were revealed with a donkey anti-rabbit IgG antibody coupled to horseradish peroxidase (Amersham Biosciences) using the ECL Western blot detection kit (Amersham Biosciences). The bands were quantified by scanning densitometry (Quantity One software, Bio-Rad).
␤-Oxidation Measurements-Fresh livers were homogenized in 19 volumes of 0.25 M sucrose containing 2 mM EDTA and 10 mM Tris-HCl, pH 7.4, using a Potter-Elvehjem homog- enizer (Ikemoto Rika, Tokyo, Japan) and a tight-fitting Teflon pestle. Palmitate oxidation rate was measured using sealed vials in a medium, pH 7.4, containing the following: 25 mM sucrose, 75 mM Tris-HCl, 10 mM K 2 HPO 4 , 5 mM MgCl 2 , and 1 mM EDTA supplemented with 1 mM NAD ϩ , 5 mM ATP, 0.1 mM coenzyme A, 0.5 mM L-malate, 0.5 mM L-carnitine, and 25 M cytochrome c. After 5 min of preincubation at 37°C with shaking, the reaction was started by addition of 100 l of 600 M [1-14 C]palmitate (16:0) bound to albumin in a 5:1 molar ratio. The oxidation proceeded for 30 min at 37°C and was stopped by addition of 0.2 ml of 3 M perchloric acid. The released 14 CO 2 was trapped in 0.3 ml of ethanolamine/ethylene glycol (1:2 v/v) and measured by liquid scintillation counting in 5 ml of Ready Safe (Beckman Instruments, Fullerton, CA). After 90 min at 4°C, the acid incubation mixture was centrifuged for 5 min at 10,000 ϫ g, and the 0.5-ml supernatant containing 14 C-labeled perchloric acid-soluble products was assayed for radioactivity by liquid scintillation. Total palmitate oxidation rate was calculated from the sum of 14 CO 2 and 14 C-labeled acid-soluble products (23) and expressed in nanomoles of palmitate per g of wet tissue per min.
All assays were performed in duplicate and using three mice per condition. Oxidation rates were expressed in nanomoles of fatty acid per min per mg of wet tissue. Peroxisomal palmitate oxidation was determined in the presence of inhibitors of mitochondrial oxidation, i.e. antimycin A and rotenone (100 and 12.5 M final concentration, respectively).
Lipid Profiling-An aliquot (20 l) of an internal standard mixture (11 reference compounds at concentration level, 8 -10 g/ml), 50 l of 0.15 M sodium chloride, and chloroform/methanol (2:1) (200 l) was added to the tissue sample (20 -30 mg). The sample was homogenized, vortexed (2 min), left to stand (1 h for liver), and centrifuged at 10,000 rpm for 3 min. From the separated lower phase, an aliquot was mixed with 10 l of a labeled standard mixture (three stable isotope-labeled reference compounds at concentration level of 9 -11 g/ml), and a 0.5-1.0-l injection was used for LC/MS analysis.
Total lipid extracts were analyzed on a Waters Q-Tof Premier mass spectrometer combined with an Acquity Ultra Performance LC (UPLC). The column, which was kept at 50°C, was a BEH C18 10 ϫ 50 mm column with 1.7-m particles. The binary solvent system (flow rate 0.200 ml/min) included solvent A, water (1% 1 M NH 4 Ac, 0.1% HCOOH), and solvent B, LC/MS grade (Rathburn) acetonitrile/isopropyl alcohol (5:2, 1% 1 M NH 4 Ac, 0.1% HCOOH). The gradient started from 65% solvent A, 35% solvent B, reached 100% solvent B in 6 min, and remained there for the next 7 min. The total run time per sample, including a 5-min re-equilibration step, was 18 min. The temperature of the sample organizer was set at 10°C.
Mass spectrometry was carried out on Q-Tof Premier (Waters) run in ESI ϩ mode. The data were collected over the mass range of m/z 300 -1,200 with scan duration of 0.2 s. The source temperature was set at 120°C, and nitrogen was used as desolvation gas (800 liters/h) at 250°C. The voltages of the sam- Values are for 30 days NC and HFD mice. Data represent mean Ϯ S.D., n ϭ 5-6. Cytokines IL-17, MIP-1a, IL-2, IL-4, IL-3, IL-12, GM-CSF, and IL-13. Eoxatin, MCP-1, MIB-1b, and IL-9 could not be detected. RANTES is regulated on activation normal T cell expressed and secreted.

B6
A pling cone and capillary were 39 V and 3.2 kV, respectively. Reserpine (50 g/liter) was used as the lock spray reference compound (5 l/min; 10-s scan frequency). Data processing was performed using the MZmine software (24). Identification was performed based on an internal reference data base of lipid species (25) or, alternatively, utilizing the tandem mass spectrometry. Tandem mass spectrometry was used for the identification of selected lipid species. MS/MS runs were performed by using ESI ϩ mode, collision energy ramp from 15 to 30 V, and mass range starting from m/z 150.
Partial least squares discriminant analysis was utilized as a modeling method for clustering and discrimination (26). Partial

TABLE 2 Genes regulated by HFD and involved in lipid metabolism
Genes for fatty acid oxidation were selected from regulated genes identified by microarray analysis, by Western blot analysis for catalase, and by qRT-PCR for Fads2 and Abcd3. least squares discriminant analysis is a pattern recognition technique that correlates variation in the dataset with class membership (26,27). The resulting projection model gives latent variables that focus on maximum separation ("discrimination"). The random subsets cross-validation method (28) and Q 2 scores were used to develop the models (28). Partial least squares discriminant analyses were performed using Matlab,

Physiological Analysis of High Fat-fed A/J and B6
Mice-To confirm that A/J mice were resistant to high fat feeding, we fed A/J and B6 mice an HFD or an NC and measured body weight gain and hepatic lipid accumulation. Fig. 1A shows that B6 mice fed an HFD gained significantly more weight than NC-fed mice and developed hepatic steatosis (Fig. 1B). In contrast, A/J mice fed an HFD did not gain more weight than NC-fed mice and did not develop hepatosteatosis (Fig. 1B). This was confirmed by measuring liver triglyceride concentrations (Fig. 1C).
Blood biochemical analysis showed different adaptation of both strains to HFD (Table 1). Fed and fasted blood glucose levels were higher in NC-fed B6 than A/J mice and further increased upon HFD feeding in the B6 but only slightly in the A/J mice. This was associated with a tendency to higher plasma insulin levels in the B6 mice suggesting lower insulin sensitivity. Plasma leptin concentrations increased more upon HFD in the A/J than in the B6 mice. Resistin levels increased similarly in both strains, and adiponectin showed decreased plasma concentration in the HFD-fed A/J mice but no change in the B6 mice. Upon HFD feeding, the A/J mice showed lower increase in plasma cholesterol and lower levels of VLDL and LDL and a higher HDL/LDL ratio. Analysis of plasma cytokines levels showed significantly lower concentrations of IL-1␤ and G-CSF in the A/J as compared with B6 mice and a trend to lower plasma levels of IL-6 and TNF␣ in HFD-fed A/J mice.
Together the above data confirmed that A/J mice were resistant to high fat diet-induced obesity and hepatosteatosis. This resistance was associated with lower glycemia and higher HDL/LDL ratio. In addition, we showed that A/J mice had a lower pro-inflammatory state both at base line and after HFD feeding.
Transcriptional Analysis of A/J and B6 Mouse Livers-To identify transcriptional programs associated with resistance to HFD-induced hepatosteatosis and obesity, we performed microarray analysis of the transcripts expressed in the livers of A/J and B6 mice fed an NC or HFD for 2, 10, or 30 days.
First, for each strain, we identified transcripts that were significantly up-or down-regulated in the livers of HFD and NC fed mice at each time point as described under "Experimental Procedures." From these gene lists, we sought to identify specific categories of up-or down-regulated genes that might explain the resistance of A/J mice to HFD-induced hepatic steatosis. For this, we decided to concentrate first on genes that were up-regulated in livers of A/J mice fed an HFD but not in B6 mice, because these genes might confer such resistance.
To perform the analysis, we created ranked lists of differentially expressed genes for each strain, at each time point, and performed gene set enrichment analysis (GSEA) against gene ontology (GO) categories (22). The result showed enrichment of peroxisome genes in A/J mice fed for 10 days an HFD (GO category, GO:0005777; false discovery rate ϭ 0.047) although B6 mice under the same conditions showed no enrichment ( Fig.  2A). Closer examination of the peroxisome-enriched genes revealed several genes, including Pex11a and Ehhadh, that are involved in peroxisomal biogenesis and ␤-oxidation of fatty acids (see supplemental Table 1 for a full list of enriched genes). Pex11a is a peroxisomal membrane protein whose expression regulates peroxisome biogenesis (29,30); Ehhadh encodes the inducible L-PBE, which catalyzes the second step of peroxisomal ␤-oxidation (31). This led us to hypothesize that resistance to hepatic steatosis in A/J mice could be due, at least in part, to increased peroxisomal ␤-oxidation of fatty acids.
To further explore this hypothesis, we investigated additional genes that could be involved in fatty acid ␤-oxidation but were not detected by the GSEA due to lack of appropriate gene sets. We thus selected genes that were significantly up-regulated in HFD fed A/J or B6 mice at any time point and related to fatty acid ␤-oxidation and microsomal lipid metabolism; this identified a total of 17 genes. These genes were then clustered based on their expression profile across the three time points in both mouse strains to identify common patterns of differential expression as described under "Experimental Procedures." The result of the clustering (Fig. 2B) shows several distinct clusters of genes that were differentially expressed between A/J and B6 mice. The two largest clusters (clusters A and B in Fig. 2B) both contain five genes that show up-regulation in A/J mice at 10 days HFD (cluster A) and up-regulation in B6 mice at 2 days HFD (cluster B). Cluster A corresponds closely to genes involved in peroxisomal ␤-oxidation, thus confirming the previous observation from GSEA, and cluster B contains genes involved in mitochondrial ␤-oxidation (see Table 2 for a description of the genes). These data therefore suggest that upon HFD feeding, A/J mice increase preferentially peroxisomal ␤-oxidation, and B6 mice increase preferentially mitochondrial ␤-oxidation.
Peroxisomal Gene Expression and ␤-Oxidation in A/J and B6 Mouse Livers-To confirm the higher expression of peroxisomal genes in A/J mice, we first assessed by qRT-PCR the level of several of the peroxisomal genes of Table 2. Fig. 3 shows that Ehhadh, Pex11a, Alas1, and ABCD3 (a long chain fatty acid transporting ATP-binding cassette protein) were indeed expressed at a higher level in the liver of A/J than B6 mice at base line and further increased at 10 and 30 days of HFD. Western blot analysis of catalase and Ehhadh expression in the liver of NC or HFD fed A/J and B6 mice also confirmed that both proteins were expressed at higher levels in the A/J liver at base line and following HFD feeding (Fig. 4A). The ratelimiting step in peroxisomal ␤-oxidation is catalyzed by acyl-CoA oxidase. Acyl-CoA oxidase (Acox) is produced as a 72-kDa Acox-A precursor that is cleaved into the active 51-kDa Acox-B form (32). Although FIGURE 4. A, Western blot analysis of catalase and L-PBE protein levels in liver of B6 and A/J mice fed an HFD or an NC for 10 days. Left, Western blots; right, quantitation of the Western blot data expressed as fold increase over the value for expression in the liver of NC-fed B6 mice. Data are expressed as mean Ϯ S.D., n ϭ 6, *, p Ͻ 0.05 versus B6 mice. B, Western blot analysis of acyl-coenzyme A oxidase (Acox) protein in livers of B6 and A/J mice fed an HFD or NC diet for 10 days. Acyl-coenzyme A oxidase is detected as two bands as follows: the inactive AcoxA (72 kDa) precursor and the proteolytically activated AcoxB (51 kDa). Right, quantitation of the ratio of AcoxB to AcoxA. HFD feeding is associated with higher conversion or acyl-coenzyme A oxidase into its active form in AJ/than in B6 mice. Data are mean Ϯ S.D., n ϭ 6. *, p Ͻ 0.05 versus B6 mice. C, peroxisomal palmitate oxidation in liver homogenates of B6 and A/J mice. Liver homogenates from mice fed an NC or an HFD for 30 days were used for measurements of [ 14 C]palmitate oxidation in the presence of the oxidative phosphorylation inhibitors antimycin A and rotenone. HFD induced a significant increase in ␤-oxidation in the livers of A/J but not of B6 mice, and upon HFD feeding ␤-oxidation was higher in A/J than in B6 mouse livers. Data are mean Ϯ S.D., n ϭ 3 liver, each measurement was made in triplicate. OCTOBER 1, 2010 • VOLUME 285 • NUMBER 40

JOURNAL OF BIOLOGICAL CHEMISTRY 31017
Acox mRNA was not differentially expressed between the two strains, we nevertheless evaluated its protein expression level and conversion efficiency. Fig. 4B shows that upon HFD feeding, there was a more important conversion of Acox-A into the active Acox-B form in the A/J mouse livers.
To evaluate whether the changes in mRNA and protein levels were associated with increased peroxisomal ␤-oxidation, we measured fatty oxidation in homogenates of liver from both strains of mice. Total palmitic acid oxidation was similar in both strains of mice fed an NC or HFD. However, peroxisomal ␤-oxidation, measured after poisoning mitochondria with antimycin A and rotenone, was significantly increased in A/J mice by HFD and was higher than in the liver of control or HFD-fed B6 mice (Fig. 4C). Together, the above data indicated that the higher expression of 10 peroxisomal genes in the A/J as compared with B6 mouse liver was accompanied by increased peroxisomal ␤-oxidation.
Microsomal Lipid Modification and Lipidomic Analysis-In addition to genes for peroxisomal and mitochondrial ␤-oxidation, the microarray analysis identified several genes encoding microsomal elongases and fatty-acid desaturases that were differentially expressed upon HFD between A/J and B6 mice ( Fig.  2B and Table 2). We measured the expression of Fads1, Fads2, Elovl5, and Elovl6 by quantitative PCR (Fads2 was included as an additional candidate gene that was not present on the microarray) and found that all genes except Elovl6 showed markedly higher expression in the liver of A/J as compared with B6 mice at all time points (Fig. 5).
As these enzymes are involved in the elongation and desaturation of lipids of the n-6 and n-3 series leading to, respectively, arachidonic acid (C20:4) or eicosapentaenoic acid (C20:5) and docosahexaenoic acid (C22:6), which have various pro-or antiinflammatory actions and may regulate transcription factors such as peroxisome proliferator-activated receptors (33), we analyzed by ultra-performance liquid chromatography/mass spectroscopy the lipid profiles of the A/J and B6 mouse liver. Analysis by partial least square/discriminant analysis of ϳ150 lipid species identified in each condition allowed clustering of the mice in a two-dimensional space (Fig. 6A). There was a striking segregation in separate regions of the graphs of the A/J and B6 mice fed an NC and following 2, 10, and 30 days of HFD. The lipid species that most strongly contributed to discriminate A/J and B6 mice were long chain fatty acid-containing triglycerides (Fig. 6B) and arachidonic acid-containing lyso-PC and 2arachidonylglycerol (2-AG), which were present at higher concentrations in the liver of A/J mice at baseline but also upon HFD feeding (Fig. 6C).
These results were consistent with the increased level of expression of the microsomal elongases and desaturases in the livers of A/J mice and provided two interesting pieces of information. First, stimulated peroxisomal activity is associated with a selective increase in lyso-PC in the peroxisomal membrane (34). Thus, the increase in lyso-PC in the livers of A/J mice under NC or HFD diets further supports an increased hepatic peroxisomal activity in these mice. Second, 2-AG is an agonist of the cannabinoid receptors, which can act as an anti-inflammatory agent through its binding to the CB2 receptor (35). 2-AG is produced from lyso-PC(20:4) by the action of phospholipase C or from diacylglycerol by the action of diacylglycerol lipase ␣ or ␤. The mRNAs of these enzymes were indeed significantly increased in the livers of A/J mice (Fig.  7). Thus, increased expression of phospholipase C, in the presence of higher lyso-PC content, may explain the increased hepatic concentration of 2-AG in A/J mice.
To evaluate which cannabinoid receptor was expressed in the liver, we performed qRT-PCR analysis of the CB1 and CB2 mRNAs in the liver of NC-fed A/J and B6 mice. Whereas CB1 receptor mRNA was at the limit of detection, a robust amplification signal for CB2 receptor was found in both strains (Fig.  8A). When analyzed in cell sorter-purified liver cell subpopulations, the CB2 mRNA was undetectable in endothelial cells, present at a low level in stellate cells, and expressed at a high level in Kupffer cells and total liver (Fig. 8B). As Kupffer cells represent a small fraction of the total liver cells, and hepatocytes contribute greater than 90% of total liver mRNA, this indicates that both Kupffer cells and hepatocytes express the CB2 receptor.
As the plasma levels of IL-1␤ and G-CSF were lower in A/J than in B6 mice (Table 1), we evaluated the level of expression of their mRNAs in the livers of both strains. Fig. 8C shows a tendency for lower expression of the IL-1␤ mRNA in the liver of A/J than of B6 mice at base line and a significantly lower expression after 30 days of HFD; the level of G-CSF mRNA was lower in A/J than B6 livers at 30 days of HFD. Analysis of these cytokine mRNA expressions in the liver subpopulations revealed that they were expressed by Kupffer cells (Fig. 8D). Together, these data suggest that elevated endogenous production of 2-AG may promote a lower proinflammatory state in the liver of A/J mice through its action on the Kupffer cell CB2 receptor.

DISCUSSION
The liver plays an important role in metabolic control and in the pathogenesis of the metabolic syndrome. In particular, increased fat accumulation and pro-inflammatory cytokine production can lead to hepatic and whole body insulin resistance. In this study, we found that the resistance to hepatic steatosis and obesity of HFD-fed A/J mice was associated with a coordinated increased expression of peroxisomal genes and a higher peroxisomal ␤-oxidation rate. In addition, the increased expression of a microsomal elongase and of two desaturases may lead to increased production of the anti-inflammatory lipid 2-AG and the elimination of possibly harmful fatty acids (see Fig. 9).
Peroxisomal ␤-Oxidation-Our gene expression analysis revealed a striking differential expression of genes related to fatty acid oxidation. In B6 mice, HFD feeding was associated with increased expression of three genes involved in mitochondrial ␤-oxidation, including Cpt1, which encodes the rate-limiting enzyme in this process. In contrast, in the livers of HFDfed A/J mice, there was a coordinated increase in the expression of 10 genes involved in peroxisome structure (Pex11a and Pex1), peroxisome very long chain fatty acid uptake (Abcd3), peroxisomal ␤-oxidation (Slc27a2, Acsl4, Ehhadh, Ech1, and Crat), and genes for detoxification of H 2 O 2 produced during peroxisomal ␤-oxidation (Cat and Alas1). These changes in gene expression were correlated with increased expression of L-PBE, Abcd3, and catalase and also with increased processing of acyl-coenzyme A oxidase to its active form and with increased peroxisomal ␤-oxidation activity.
Thus, HFD feeding stimulates different pathways for fatty acid ␤-oxidation in both strains of mice. The up-regulation of peroxisomal ␤-oxidation in A/J mice may contribute to the resistance to hepatosteatosis, in particular because very long chain fatty acids depend on the peroxisomes for their catabolism. In contrast, the increased expression of the mitochondrial ␤-oxidation gene in B6 mice is clearly not sufficient to prevent hepatosteatosis.
The molecular basis for the differential expression of peroxisomal and mitochondrial genes in both strains of mice is not known. A differential activity of peroxisome proliferator-activated receptor ␣ (PPAR␣) could potentially explain these differences because treatment of mice with PPAR␣ activators increases peroxisome proliferation and can correct diet-induced obesity (36,37). However, an involvement of PPAR␣ in the strain-specific adaptation to high fat diet is unlikely. First, we did not find any change in expression level of this transcription factor between the two strains of mice. Second, even though some of the differentially expressed peroxisomal and mitochondrial genes are PPAR␣ targets (Ehhadh and Cpt1, for instance), their up-regulation is found either in A/J or B6 mouse livers. Therefore, these different patterns of gene expression cannot be explained by simple variations in PPAR␣ activity in both strains.
A possible explanation for the increased expression of peroxisomal ␤-oxidation genes is that the higher expression of Elovl5, Fads1, and Fads2 may increase the conversion of potentially harmful fatty acids such as palmitate, stearate, or oleate into unsaturated long chain fatty acids. These may then increase peroxisomal gene expression and lead to degradation of these lipid species.
Lipidomic Analysis-Lipidomic analysis revealed a pattern of lipid expression in the liver of A/J and B6 mice that allowed clustering of the different mouse groups by partial least square/ discriminant analysis. The lipid species that most significantly contributed to the discrimination of these groups were triglycerides enriched in long chain fatty acids, arachidonic acid-containing lyso-PC, and 2-AG. These data are in agreement with the increased expression of Fads1, Fads2, and Elovl5, which are involved in the elongation and desaturation of n-6 fatty acids to arachidonic acid (20:4). Lyso-PC has been reported to be a marker (34) and possibly be required for activated peroxisomes, further supporting that increased peroxisomal activity is a characteristic of the A/J livers.
Most interesting, however, is the observation of increased concentrations of 2-AG in the liver of A/J mice. 2-AG can bind to both the CB1 and CB2 receptors, and our qRT-PCR analysis revealed that, in both strains of mice, the CB2 receptor was the predominant cannabinoid receptor. This suggests that 2-AG may stimulate the CB2 receptor, which is expressed on immune cells, including Kupffer cells, and has a general anti-inflammatory action (38,39). In liver, CB2 receptor agonists protect against ischemia/reperfusion injury by attenuating oxidative stress and inflammatory response (40,41). In man, CB2 receptor agonists reduce inflammation and fibrosis in patients with chronic liver disease (42). Furthermore, administration of low doses of cannabinoids reduces the progression of atherosclerosis by a mechanism that requires CB2 receptor activation and reduced macrophage activation (43). Thus, the increased production of 2-AG in the liver of A/J mice may be responsible for the lower pro-inflammatory state of these mice and the lower expression of IL-1␤ and G-CSF after HFD feeding. The fact that we found high CB2 receptor, as well as IL-1␤ and G-CSF expression in Kupffer cells, suggests the presence of a local regulatory circuit in which the production by hepatocytes of 2-AG may act on neighboring Kupffer cells to regulate their activation and production of the inflammatory cytokines. These, in turn, may influence hepatocyte metabolism through IL-␤-induced modulation of the NF-B pathway (9) and recruitment and activation by G-CSF of inflammatory leukocytes (Fig. 9) (44). On the other hand, as CB2 is also expressed by hepatocytes, an autocrine action of 2-AG may influence their own metabolism.
It should be noted, however, that a recent study of CB2 knock-out mice showed lower hepatic inflammation and reduced triglyceride accumulation when the mice were fed a high fat diet (45). These results seem to contradict our proposal that CB2 agonism may protect against inflammation. However, studying mice with a systemic knock-out of the CB2 receptor may lead to erroneous interpretation of the physiological role of this receptor. For instance, local production of 2-AG by hepatocytes may have paracrine anti-inflammatory action on neighboring Kupffer cells that cannot be inferred from the study of knock-out mice. One possible difficulty is that CB2 receptors are involved in chemotaxis and migration of inflammatory cells (46,47). Thus, absence of this receptor may lead to altered cellular composition of the liver causing the particular response of CB2 knock-out mouse livers to HFD feeding. In addition, it is known that CB2 agonists may have pro-or anti-inflammatory properties depending on the cellular context (39). Similarly, the systemic injection of CB2 agonist JW133, which increases liver triglyceride content (45), appears to contradict our hypothesis. But systemic administration and activation of CB2 receptors may lead to an effect on liver inflammation and triglyceride accumulation that is opposite to that regulated by intraparenchymal paracrine actions. Clearly, more work is needed to sort out the differences between paracrine and systemic actions of CB2 agonists on liver function.
In summary, our data demonstrate that resistance to dietinduced hepatosteatosis and obesity in A/J mice is associated FIGURE 8. A, quantitative RT-PCR analysis of CB1 and CB2 expression in total liver mRNA from A/J or B6 mice. B, expression of the CB2 mRNA (log scale) in liver cell subpopulations from B6 mice. C, quantitative RT-PCR analysis of IL-1␤ and G-CSF in total liver from A/J or B6 mice. D, IL-1␤ and G-CSF mRNAs expression in liver cell subpopulations. Data are mean Ϯ S.E. for triplicates measurements made in two separate experiments.

Pathways Associated with Resistance to Hepatic Steatosis
with changes in the expression of genes for two sets of lipidmodifying genes, one controlling peroxisomal ␤-oxidation and the other increasing the capacity to convert essential unsaturated fatty acids from the n-6 series into arachidonic acid-containing lyso-PC and the cannabinoid agonist 2-AG. These data are compatible, on the one hand, with increased oxidation of exogenous lipids, in particular of the very long chain fatty acids that need peroxisomes for their degradation. On the other hand, increased 2-AG production may lead to the reduced proinflammatory state of the A/J mice. Thus, slight variations in the level of expression of hepatic genes may have protective effects against hepatosteatosis by a dual action on lipid metabolism and the activity of the liver endogenous inflammatory system. The molecular basis for the differences in gene expression observed here is not known. The coordinated increase in expression of 10 peroxisomal genes and 3 microsomal genes suggests, however, a common mechanism driving their differential expression. Finally, these data also suggest that these genes and their corresponding pathways represent attractive candidates for metabolic syndrome targets. Hence, it will be interesting to evaluate whether human orthologs of the genes identified in this study may present polymorphisms that are associated with increased risks of the metabolic syndrome or of some specific parameters of this disease.