Brain Transcriptional Responses to High-Fat Diet in Acads-Deficient Mice Reveal Energy Sensing Pathways

Background How signals from fatty acid metabolism are translated into changes in food intake remains unclear. Previously we reported that mice with a genetic inactivation of Acads (acyl-coenzyme A dehydrogenase, short-chain), the enzyme responsible for mitochondrial beta-oxidation of C4–C6 short-chain fatty acids (SCFAs), shift consumption away from fat and toward carbohydrate when offered a choice between diets. In the current study, we sought to indentify candidate genes and pathways underlying the effects of SCFA oxidation deficiency on food intake in Acads−/− mice. Methodology/Principal Findings We performed a transcriptional analysis of gene expression in brain tissue of Acads−/− and Acads+/+ mice fed either a high-fat (HF) or low-fat (LF) diet for 2 d. Ingenuity Pathway Analysis revealed three top-scoring pathways significantly modified by genotype or diet: oxidative phosphorylation, mitochondrial dysfunction, and CREB signaling in neurons. A comparison of statistically significant responses in HF Acads−/− vs. HF Acads+/+ (3917) and Acads+/+ HF vs. LF Acads+/+ (3879) revealed 2551 genes or approximately 65% in common between the two experimental comparisons. All but one of these genes were expressed in opposite direction with similar magnitude, demonstrating that HF-fed Acads-deficient mice display transcriptional responses that strongly resemble those of Acads+/+ mice fed LF diet. Intriguingly, genes involved in both AMP-kinase regulation and the neural control of food intake followed this pattern. Quantitative RT-PCR in hypothalamus confirmed the dysregulation of genes in these pathways. Western blotting showed an increase in hypothalamic AMP-kinase in Acads−/− mice and HF diet increased, a key protein in an energy-sensing cascade that responds to depletion of ATP. Conclusions Our results suggest that the decreased beta-oxidation of short-chain fatty acids in Acads-deficient mice fed HF diet produces a state of energy deficiency in the brain and that AMP-kinase may be the cellular energy-sensing mechanism linking fatty acid oxidation to feeding behavior in this model.


Introduction
Fatty acid oxidation is implicated in the metabolic control of food intake and energy balance [1]. However, the signal(s) mediating the effects of fatty acid oxidation on food intake has not yet been identified [2][3]. A number of studies indicate that enzymes within fatty acid metabolic pathways may serve as targets for pharmacological tools to treat obesity [4], yet there is controversy over which aspect of fatty acid availability may inhibit food intake. We developed a novel approach for investigating the molecular and physiological links between fat intake and fat oxidation using an animal model with a genetic deficiency in short-chain acyl-CoA dehydrogenase (Acads), a mitochondrial enzyme that catalyzes the first reaction in the beta-oxidation of C4-C6 fatty acids. A spontaneous deletion in the structural gene for Acads results in the complete absence of functional short-chain acyl-CoA dehydrogenase [5] in the BALB/cByJ mouse inbred strain. We have shown that Acads-deficiency [6][7] suppressed voluntary fat-but not carbohydrate intake in mice [8]. Specifically, Acads2/2 mice shifted their food consumption away from fat and toward carbohydrate when offered a choice of diets, while total calorie intake remained unchanged. In addition we showed that during brief access lick tests, Acads2/2 mice showed similar licking responses to fat and carbohydrate stimuli compared with Acads+/+ controls suggesting that their fat avoidance is not mediated by an orosensory mechanism. In the current study, we tested the hypothesis that gene expression patterns in Acadsdeficient mice would reveal functionally important pathways mediating the behavioral feeding response to fat.
Energy balance is maintained by changes in behavior and metabolism which can be regulated by gene expression. To our knowledge, there are no reports of studies on the effects of dietary fat level on the brain transcriptome in genetic models of fatty acid oxidation deficiency. Microarray analysis of the effects of Acads deficiency on gene expression in brain is limited to a single investigation in which animals were fed standard chow and the number of genes/ESTs represented on the array (,5,000) was relatively small [9]. The purpose of the current study was to identify key transcriptional processes linked to fat intake, and ultimately to behavior, in a genetic model of impaired SCFA oxidation. We used microarray technology to compare the shortterm effects of high-or low-fat diet on gene expression in the brain of Acads2/2 and Acads+/+ mice. Tissues were harvested 2 d after initiating the experimental diets to coincide with the time-point at which fat avoidance begins [8]. We identified transcriptional responses to genotype and diet that that may be linked to feeding behavior in this model. We also investigated gene expression in hypothalamus because of its role in the regulation of metabolic and behavioral phenotypes.

Experimental animals
The BALB/cByJ (Acads2/2) and BALB/cByKZ (Acads+/+) mice were obtained from breeding colonies maintained at the Pennington Biomedical Research Center (PBRC). The BALB/ cByJ mouse strain carries a spontaneous 278 bp deletion in the structural gene for Acads that occurred spontaneously sometime between 1981 and 1982 [5][6][7][10][11][12]. The BALB/cByJ mice are descendents of the BALB/cBy strain maintained originally by Donald Bailey at the Jackson Laboratory. The best Acads+/+ control line for BALB/cByJ mice is thought to be the BALB/ cBy [13]. The BALB/cByKZ.Acads+/+ substrain was separated from the research colonies at the Jackson Laboratory in 1996. Due to the length of time that these substrains have been separated, we cannot rule out undetected spontaneous mutations in genes other than Acads that could affect gene expression. All protocols were approved by the Institutional Animal Care and Use Committee of Pennington Biomedical Research Center.

Experimental protocol
Twelve-week old male mice were singly housed in filter-top cages and kept under 12 h light/12 h dark conditions at an ambient temperature of 22-23uC in a specific-pathogen free facility. Several days prior to the experiment, bedding was removed and replaced with stainless steel wire floor inserts, to permit recovery and measurement of food spillage. Polyvinylchloride nesting tubes (1K in. diameter) were provided to reduce time spent on wire flooring. The high-(HF) and low-fat (LF) experimental diets were equivalent for protein (16.4% of energy) with the balance of calories contributed by 58% fat and 25.5% carbohydrate in D12331 and by 10.5% fat and 73.1% carbohydrate in D12329 (Research Diets, Inc., New Brunswick, NJ) (Table S1).

Microarray analysis
Total RNA was isolated from whole brain using TRI Reagent (Molecular Research Center, Inc., Cincinnati, OH, USA), and the quality was determined using an Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA). One mg of total RNA was used to transcribe DIG-labeled cRNA using AB Chemiluminescent RT-IVT Kit v2.0. Microarray hybridization (using 10 mg of fragmented, DIG-labeled cRNA), processing, chemiluminescence detection, imaging, auto-gridding, and image analysis were performed according to AB protocols, using the 1700 Chemiluminescent Microarray Analyzer Software v. 1.0.3. The Applied Biosystems Mouse Genome Survey Microarray V2.0 was used which contains ,34,000 features, including a set of ,1000 controls. Each microarray uses 32,996 probes targeted to 32,381 curated genes representing 44,498 transcripts. The AB Expression system software was used to extract assay signal and signal-to-noise ratio values from the images. Signal intensities across microarrays were normalized according to the quantilequantile method based on R-script [14]; http://www. bioconductor.org). Features with a signal/noise value of $3.0 and quality flag value of ,5000 were considered ''detected'' and subjected to statistical analysis using AB array. The FDRcorrected t-statistic was calculated using the Benjamini-Hochberg procedure [15]. Differential expression was defined as a fold change $1.6 and a P value of #0.05. The microarray data, described according to MIAME guidelines, have been deposited in the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) repository. The accession number is GSE35180.

Acylcarnitine analyses
Acylcarnitines were measured by the Analytical Systems Laboratory of the Louisiana State University School of Veterinary Medicine. Plasma samples were analyzed by direct-injection electrospray tandem mass spectrometry, using a Quattro II LC-MS system (Waters-Micromass, Milford MA) equipped with an HTS-PAL autosampler (Leap Technologies, Carrboro, NC). An Agilent 1100 HPLC system with a binary pump (Agilent Technologies, Santa Clara, CA) was interfaced to the mass spectrometer. All data were acquired and analyzed using MassLynx software, version 4.0 (Waters-Micromass).

Ingenuity Pathway analysis (IPA)
Differentially expressed genes (.1.6-fold) with a comparison P value of ,0.05 were subjected to Ingenuity Pathways Analysis (www.ingenuity.com) (Ingenuity Systems, Mountain View, CA). The IPA is a manually curated database of molecular interactions and contains previously published findings from peer-reviewed publications. It should be noted that the interactions presented in the networks are not specific for brain tissue. Fischer's exact test was used to calculate P values determining the probability that the association between the genes in the dataset and pathway could be explained by chance alone. The significance score was expressed as the negative logarithm of the P value, indicating the likelihood that the focus genes were assembled into a pathway.

Real-time quantitative PCR
cDNA preparation for RT-PCR. RNA quality was assessed using the Agilent 2100 bioanalyzer (Agilent Technologies, Santa Clara, CA, USA), and quantity was determined in the NanoDrop ND-1000 Spectrophotometer (NanoDrop Technologies, Rockland, DE, USA). Complementary cDNA was obtained by reverse transcription (SuperScript III First-Strand Synthesis System for RT-PCR, Invitrogen, Carlsbad, CA, USA) of 4 mg of RNA from each sample. This reaction used both Oligo(dT) 20 and Random Hexamers as primers; all further steps were done following the supplier's instructions (Invitrogen, Carlsbad, CA, USA). Purification of cDNA was performed using QIAquick PCR purification columns (Qiagen, Valencia, CA, USA). For quantitative RT-PCR, 6-12 samples each were tested for the Acads2/2 and Acads+/+ strains using a combination of the microarray samples (3 per strain and condition) and others.
Primers for amplification were designed using Primer Express 3.0 software from Applied Biosystems (AB; Foster City, CA, USA) and synthesized by IDT (Integrated DNA Technologies, Coralville, IA, USA). The locations and sequences of primer are listed in Table S2. Primers were designed, when possible, to span an exon/ exon junction to avoid amplification from potentially contaminating genomic DNA. Primer sequences were subjected to BLAT analysis (http://genome.ucsc.edu) to confirm their specificity. The ENSEMBLE genome browser contains the transcript and exon information for the genes investigated in this study (Table S2 with accession numbers given).
The MultiPROBE II PLUS HT EX robot (Perkin Elmer, Shelton, CT, USA) was programmed to pipette 10 ml reactions into an AB 384 well plate. The robot adds 3 ml template (2 ng cDNA) and 7 ml Master Mix (5 ml iTaq SYBR Green Supermix with ROX, 0.1 ml forward primer 10 mM, 0.1 ml reverse primer 10 mM, 1.8 ml NanoPure water) per reaction. The iTaq SYBR Green Supermix with ROX (26) was obtained from Bio-Rad Laboratories (Hercules, CA, USA). Gene expression levels were measured using the ABI PRISM 7900HT Sequence Detection System with SDS 2.3 software version (AB). Individual samples were run in triplicate. As endogenous reference for normalization, we used the measurement of Cyclophilin (Ppia) cDNA in the same sample; Ppia levels were unaffected by diet genotype on the arrays.
Relative quantification was carried out using the comparative C T method (AB: User Bulletin #2: Relative quantification of gene expression. P/N 4303859). A two-tailed Student's t-test was performed on delta Ct values to evaluate strain differences in gene expression. P values of #0.05 were considered statistically significant.

Protein analysis by immunoprecipitation and immunoblotting
Total protein was isolated from hypothalamus using Lysis buffer (25 mM HEPES pH 7.8, 1% NP-40, 50 mM KCl, 125 mM DTT, 1 mM PMSF, 5 mM NaPPi, 1 mM EDTA, 50 mM NaF including protease inhibitor cocktail). Protein concentration was determined with the BCA assay (Pierce Biotechnology, Rockford, IL). Electrophoretic separations were carried out on Mini-Protean II electrophoresis cells (Bio-Rad Laboratories, Inc., Hercules, CA) and run at 100V. 45 mg of proteins were separated on a 10% SDS-PAGE and transferred onto PVDF membranes (Bio-Rad) using a Mini Trans-Blot (Bio-Rad). The membranes were blocked in 5% milk in TBST and then immunoblotted overnight at 4 deg C with rabbit anti-phospho-AMPK alpha1, 2 (Thr 172 ) or rabbit anti-AMPK (Abcam Inc., Cambridge, MA). After extensive washing, the membranes were incubated with goat anti-rabbit horseradish peroxidase-linked secondary antibody (Abcam). Membranes were washed again and proteins were visualized using the Pierce enhanced chemi-luminescence Western blotting detection system. Mouse monoclonal antibodies to beta Actin (HRP) (Abcam) were used as loading control. Densitometry of Western blots was performed using Image J Software [16].

Statistical analyses
Acylcarnitine data were analyzed using an analysis of variance approach (ANOVA) with Tukey's adjustment for pair-wise comparisons. The testing procedure focused on the main effects of diet and strain, as well as the diet by strain interaction. The level for statistical significance was set at P,0.05. The SAS version 9.2.1 statistical software package was used.

Results
The present study was designed to identify the brain transcriptional responses to genotype and diet that may be linked to feeding behavior in this model of Acads deficiency. To further clarify the microarray results in total brain (see Text S1), we investigated gene expression in the hypothalamus, because of its role in the regulation of metabolic and behavioral phenotypes.

Plasma acylcarnitine levels
Tandem mass spectrometry was used to analyze 33 independent acylcarnitine species ranging in size from 2 to 22 carbons, in the plasma of Acads2/2 and Acads+/+ mice fed HF (58%) or LF (11%) fat diet for 2 days. Plasma levels of seven acylcarnitines showed a main effect of Diet and three acylcarnitines showed a main effect of Strain (Table S3). Specifically, on HF diet the plasma content of C4 and C5 was 13-fold (P,0.0001) and 2-fold (P,0.05) higher, respectively, in Acads2/2 compared with Acads+/+ mice (Fig. 1). On LF diet, a 2.5-fold increase in circulating C4 (P,0.01) was observed in Acads2/2 compared with Acads+/+ mice. The substantial increase in plasma acylcarnitine levels in SCAD-deficient mice points to the location of the metabolic block and demonstrates an effect of short-term HF diet. Only two acylcarnitines showed a significant Diet x Strain interaction: C4 [F(3,21) = 9.55, P,0.01] and C18-OH [F(3,21) = 7.6, P,0.05]. The physiological significance of the altered C18-OH levels is unknown.

Gene expression analysis
Expression levels of 32,381 transcripts in the brain were compared between genotypes on HF or LF diet, and between diets for Acads2/2 or Acads+/+ respectively. Only transcripts whose expression levels showed a fold change of $1.6 with a corresponding P value of ,0.05 were analyzed. Transcripts were also sorted based on gene ontologies assigned in the MGI database version 3.0 (Fig. S1).

Effects of high-or low-fat diet on gene expression
Strikingly, the expression of nearly four thousand genes (3879) was significantly deregulated in wildtype controls after 2 d of HF diet feeding, and more than one-half of these genes showed increased expression (Table 1). By contrast, only 281 genes were differentially expressed as an effect of diet in Acads-deficient animals and most of those were decreased. These results clearly indicate that the lack of functional SCAD enzyme diminished the animals' ability to respond to HF diet. Forty-one (41) genes were regulated by diet in both Acads2/2 and Acads+/+ mice ( Figure 2A)  Table S3 for a summary of ANOVA results for these variables. doi:10.1371/journal.pone.0041709.g001 and of these, eighteen (18) were differentially expressed in opposite direction. Specifically, we identified diet-regulated genes involved in processes such as regulation of cell matrix adhesion (Plekha2), ATP catabolic process (Atp7a), protein folding (Dnajb9), development (Pax7), adrenal gland function (Nr3c1), RNA binding (Oas1c), and copper ion homeostasis (Aplp2). These genes were strongly down-regulated by dietary fat (.2-fold) in Acads-deficient animals, but were up-regulated by HFD in controls.

Effects of Acads genotype on gene expression
With HF diet, the expression of 3917 genes was altered as an effect of genotype (Acads2/2 vs Acads+/+), compared with 275 genes with LF diet (Table 1). Thirty-four (34) genes were regulated by genotype under both HF and LF diet conditions ( Fig. 2A), and of these, sixteen (16) genes were changed in opposite direction. For example, we identified transcripts that were strongly down-regulated in Acads-deficient mice and upregulated in controls under both diet conditions, including genes involved in processes such as AMP-activated protein kinase activity (Pkig, Prkag1, Stk38, Ccdc88a), protein folding (Dnajb9), cell adhesion (Pcdhb8), ATP binding (Ddx6), and regulation of transcription (Carf).   Table 1 for the total number of differentially expressed genes for each of the four comparisons. Within each main comparison (e.g., separate colors in B), the overlap from sub-categories was subtracted from the total number of diet-or genotype-regulated transcripts. Only six transcripts were found in common among all four comparisons. doi:10.1371/journal.pone.0041709.g002 Effects of Acads genotype and diet on gene expression:

4-way Venn diagram
A four-way Venn diagram was constructed to illustrate overlapping transcripts among all four comparisons (Fig. 2B). Within each main comparison, the overlap from sub-categories was subtracted from the total number of diet-or genotyperegulated transcripts (totals shown in Table 1). Only six transcripts were found in common among all four comparisons and of these, only three were annotated: Dnajb9 (DnaJ homolog, subfamily B, member 9), Oas1c (29-59 oligoadenylate synthase 1C), and Prlhr (prolactin releasing hormone receptor).
Gene expression in HF-fed Acads2/2 mice resembles that of Acads+/+ mice on LF diet When we compared Acads+/+ to Acads2/2 mice on HF diet, 3917 genes were differentially expressed based on our study criteria (fold-change above 1.6 and P,0.05) ( Table 1). A comparison of HF to LF diet in Acads+/+ controls revealed 3879 differentially expressed genes. An examination of both sets of results revealed that they share 2551 genes or approximately 65% in common (Fig. 3A). On further evaluation, we discovered that 2550 of these genes were expressed in opposite direction in one result set versus the other, and with similar magnitude (Fig. 3B). For example, genes whose expression increased in wildtype controls as an effect of diet (HF vs. LF), were decreased in HF diet as an effect of genotype (Acads2/2 vs. wildtype). This finding demonstrates that the response to HF diet in Acads-deficient mice yields transcriptional responses similar to those of wildtype Acads+/ + animals on LF diet, i.e., that Acads2/2 mice show a diminished response to HF diet.
An alternative hypothesis is that Acads2/2 on LF diet show a HFD-like transcriptome. However, an examination of the result sets for two comparisons: Acads2/2 HF to LF diet (281 genes) and Acads+/+ HF to LF (3879) genes revealed only 41 genes in common (results not shown). Of these, only 23 genes showed increased expression in both wildtype and mutant animals as a result of HF diet, lending minimal support to this hypothesis.

Ingenuity canonical pathway analysis
Ingenuity Pathway Analysis (IPA) was used to further organize the differentially expressed genes into functionally annotated pathways and networks. The most significant pathways affected by either genotype or dietary fat are shown in Fig. 4. Two comparisons (Acads2/2HF vs Acads+/+HF and Acads+/+HF vs Acads+/+LF) shared the same three top-scoring pathways including oxidative phosphorylation, mitochondrial dysfunction, and CREB signaling in neurons. The oxidative phosphorylation pathway includes mitochondrial complex genes such as NADH dehydrogenases, succinate dehydrogenase, cytochrome c oxidases (COX) and ATP synthases, as well as ubinquinol-cytochrome c reductases and ATPases. The mitochondrial dysfunction pathway contains glutathione, glutaredoxin, and glutathione reductase. The CREB signaling in neurons pathway ranked 3 rd highest, encompassing differentially expressed genes such as adenylate cyclases, cAMP responsive element binding protein 5 (CREB5), guanine nucleotide binding proteins, glutamate receptors, MAPK, phosphoinositide kinases, and protein kinase C. In total, 142 pathways were significantly affected by genotype in HF diet, compared with only 2 in LF diet. Three pathways were significantly affected by fat level in Acads2/2 mice (androgen and estrogen metabolism; riboflavin metabolism; oxidative phosphorylation), in contrast to 114 pathways in Acads+/+ mice.

AMP-activated protein kinase (AMPK) related genes
Pathway analysis of our microarray results revealed that AMPK signaling was significantly (5.93610 24 ) altered in two key experimental comparisons: HF Acads2/2 vs. HF Acads+/+ and Acads+/+ HF vs. LF Acads+/+. A set of 18 significantly regulated genes known to act in concert with AMPK are shown in Table 2. For example, gene expression of Stk32c, an upstream activator of AMPK as well as Prkag1, the protein kinase AMPactivated gamma 1 non-catalytic subunit for AMPK, were increased in HF Acads2/2 compared to HF Acads+/+ but were decreased in HF Acads+/+ vs. LF Acads+/+. The activation of AMPK stimulates fatty acid oxidation by inactivating acetyl-CoA carboxylase, and we observed that Acaca (acetyl-CoA carboxylase alpha) expression was significantly decreased in HF Acads2/2 vs. HF Acads+/+. These gene expression changes suggest that in Acads-deficient animals, a perceived deficit in the amount of energy obtained from dietary fat activates the cellular energy sensor AMP-activated protein kinase pathway [17]. By contrast, each of these genes in wildtype controls (HF vs. LF) is expressed in opposite direction demonstrating once again that HF Acads2/2 mice respond to HF diet in a manner similar to wildtype controls on LF diet.

Increased hypothalamic pAMPK in Acads2/2 mice fed HF diet
The integration of extracellular hormonal and nutrient signals involves hypothalamic AMPK activity, a key signal in the response to changes in energy balance [17][18] including consumption of high-fat diet [19]. Immunoblot analysis of hypothalamic tissue showed enhanced AMP kinase activity as evidenced by a 20% increase in (Thr-172) phosphorylation of the AMPK protein in Acads-deficient mice fed HF diet for 2 d, compared with controls (Fig. 5).

Neuropeptide and neurotransmitter genes
Pathway analysis uncovered statistically significant changes in expression of neuropeptide and neurotransmitter genes known to play a role in the regulation of food intake ( Table 3). Most of these transcripts were expressed in opposite direction and with similar magnitude, i.e., expression was increased when comparing HF-fed mutant to HF-fed wildtype, and decreased in wildtype controls on LF (relative to wildtype on HF). For example, gene expression of the orexigenic peptides Npy (neuropeptide Y) and Gal (galanin) was increased in HF-fed Acads2/2 mice compared with Acads+/+ ( Table 3). The increased expression of Cart (cocaine-and amphetamine-regulated transcript) in HF-fed Acads2/2 mice is surprising, but is consistent with evidence for an orexigenic role of this neuropeptide [20][21][22][23]. In addition, other genes involved in the control of food intake were differentially expressed such as Nts (neurotensin), Gabbr1 (gamma-aminobutyric acid B receptor, 1), several genes related to GABA-A receptor activity (7 genes), dopamine (6 genes), and opioids (3 genes) ( Table 3). Results for Agrp and Ghrelin were not obtained in this microarray analysis because they did not meet the signal/noise ratio specification.

Real-time RT-PCR validation of microarray results in hypothalamus
Using quantitative RT-PCR in hypothalamus, we examined a subset of fatty acid metabolism genes that were differentially expressed in the microarrays, as well as some neuropeptide genes (Table 4). Several genes involved in fatty acid beta-oxidation were coordinately down-regulated in HF-fed Acads-deficient mice compared to wildtype. Significantly decreased expression was found for Cox8b (cytochrome c oxidase, subunit VIIIb), an enzyme of the mitochondrial respiratory chain; UCP3 a mitochondrial uncoupling protein; Acox1 and Cpt1c encoding beta-oxidation enzymes; Acaca (acetyl-CoA carboxylase alpha), and Fasn (fatty acid synthase), and Mlycd (malonyl-CoA decarboxylase) which are  involved in fatty acid biosynthesis. Creb5 is a cAMP-responsive transcription factor with a newly described role in the regulation of hypothalamic metabolism [24] and energy regulation [25]. Approximately one-half of the qPCR results confirmed those of the microarray analysis (Table 4). Although changes in RNA expression do not necessarily correspond to changes in protein expression, the overall pattern of these related genes suggests a functional change.
The downstream effects of phosphorylated AMPK may be expressed through changes in neuropeptide signaling, therefore we examined orexigenic agouti-related peptide (Agrp) and neuropeptide Y (Npy) and anorexigenic pro-opiomelanocortin (Pomc) and cocaine-and amphetamine-regulated transcript (Cart). Using realtime RT-PCR, we observed significant reductions in the anorexigenic peptide Pomc in HF fed Acads2/2 compared with HF Acads+/+ mice (P,0.05) ( Table 4), suggesting a compensatory response to negative energy balance in Acads-deficient mice. This result contrasts with the significantly decreased expression of orexigenic Agrp in wildtype animals (P,0.05) ( Table 4), which may represent a compensatory response to the positive energy balance associated with acute exposure to HF diet [26].

PANTHER analysis identified diet-regulated chemosensory processes
We also used PANTHER Analysis to identify processes that were overrepresented in GO Biological Process categories, but not portrayed by pathway analysis tools ( Table 5). The reported P value was based on the number of differentially expressed genes obtained versus the number that would be expected. Calculation of the number of genes expected was based on 281 genes in Acads2/2 and 3917 genes in Acads+/+ mice. Of note, dietregulated biological processes such as chemosensory perception and olfaction were under-represented in Acads+/+ mice and by contrast, over-represented in the Acads2/2 mice, including Gprotein signaling, the the primary sensing mechanism for vision, taste and smell. Sensory organs provide information that allows animals to effectively identify foods that have been associated with both positive and negative internal consequences. One possible interpretation of these results is that chemosensory pathways were invoked in Acads2/2 mice to support associative learning about the negative internal consequences of HF diet, i.e., incomplete oxidization and reduced energy availability.
Our data have identified the brain molecular pathways and genes that respond to fat intake in mice with a genetic inactivation of Acads, encoding short-chain acyl-CoA dehydrogenase (SCAD). The lack of functional SCAD enzyme in Acads2/2 mice results in the impaired oxidation of shortchain fatty acids, as demonstrated by the significantly higher levels of C4 and C5 acylcarnitines in plasma of SCAD deficient mice (Fig. 1). Using a whole genome array, changes in gene expression were evaluated in mutant or wildtype mice at 2 d after initiating experimental HF or LF diets. The top three pathways affected by diet or genotype were oxidative phosphorylation, mitochondrial dysfunction, and CREB signaling in neurons. For the majority of genes that map to these pathways, HF-fed Acads2/2 mice diet displayed transcriptional responses similar to those of wildtype Acads+/+ mice fed LF diet (Fig. 3). Specifically, genes involved in energy sensing, mitochondrial function and the peptidergic control of feeding behavior followed this pattern. In particular, an examination of the gene expression profile for energy sensing and AMPKrelated genes suggests that the decreased beta-oxidation of short-chain fatty acids in Acads-deficient mice fed HF diet produces a condition of energy deficiency in the brain ( Table 2). Further investigation in hypothalamus using quantitative RT-PCR confirmed the dysregulation of genes in these pathways. Western blotting showed increased hypothalamic AMP-kinase, a key protein in an energy-sensing cascade that responds to depletion of ATP. Previously we demonstrated that in a diet choice paradigm Acads2/2 mice, compared to Acads+/+, shifted consumption away from fat-and toward carbohydrate-containing diets, thus effectively preventing a reduction in total calories [8]. In particular, Acads2/2 mice significantly reduced their selfselected fat intake on day two of diet selection providing the rationale for examining gene expression changes on day two after diet initiation in the present study. We hypothesized that gene expression changes at this time-point would reveal functionally important pathways mediating the behavioral feeding response to dietary fat.
The LF and HF experimental diets used in the current study are composed mainly of C12-C18 fatty acids (86-91%) and thus provide plenty of long-chain fatty acids for beta-oxidation. SCAD deficiency blocks only the beta oxidation of C4-C6 substrates, resulting in nearly normal amounts of acetyl-CoAs along with the accumulation of butyric acid. We speculate that the incomplete oxidation of short-chain fatty acids due to SCAD deficiency may have produced a small mitochondrial oxidative deficiency, consistent with a previous report of 20% lower acetyl-CoA levels in the brain of Acads2/2 compared with Acads+/+ mice [27].
Results from the microarray revealed the significant downregulation of genes involved in the activation of AMP-kinase in Acads2/2 mice compared with Acads+/+ on HF diet, suggesting that impaired SCFA oxidation resulted in energy deficiency. Because hypothalamic AMP-kinase is decreased in mice fed highfat diet [19], we also examined the effects of Acads deficiency on hypothalamic AMPK in Acads2/2 mice. We hypothesized that the reduced cellular energy leading to expression changes in AMPK-related and neuropeptide genes could ultimately be responsible for the altered macronutrient diet selection observed previously in Acads2/2 mice [8].
The brain possesses a high density of mitochondria which are needed to provide energy to maintain cell functions [28]. Although the brain relies primarily on glucose metabolism for energy, it can also use ketones formed in the liver under conditions of low carbohydrate availability. Normally, cytosolic fatty acyl-CoAs taken up by mitochondria undergo betaoxidation or are converted into various structural and signaling lipid molecules [29]. Inside the mitochondria, fatty acids undergo beta-oxidation in a process whereby acyl-CoA dehydrogenases, differing in their substrate specificity based on the fatty acid chain length, cleave two carbons at a time from acyl-CoA molecules starting at the carboxyl end, e.g., ACADS catalyzes the first step in the beta-oxidation of C 4 -C 6 fatty acids. The two-carbon units of acetyl-CoA then enter the Krebs cycle Figure 5. Effects of diet, genotype and day on pAMPK in the hypothalamus. On low-fat diet, there was no effect of genotype on pAMPK level in hypothalamus (A, C). On Day 2 of high-fat diet, Acads2/2 mice showed increased pAMPK levels compared to Acads+/+ mice (B, D). Representative immunoblots for total AMPK, pAMPK, and beta actin are shown. n = 2-3 per genotype and day. *P,0.05 vs. Acads+/+. doi:10.1371/journal.pone.0041709.g005 and provide substrates for gluconeogenesis and ketogenesis. Although fatty acid oxidation in the brain is modest, it is important in determining the metabolic fate of fatty acids that have entered cerebral tissue [29][30].
Whether or not the observed transcriptional effects in the brain of Acads2/2 mice were the result of Acads deficiency within or outside of the brain remains in question. The evidence for a primary role of fatty acid oxidation in ATP production in brain cells is quite scant. Nevertheless it remains possible that a very small amount of fatty acid beta-oxidation in the brain may function to generate energy [29,[31][32][33] in addition to other functions such as the synthesis and/or degradation of unique cellular lipids for neuronal function [34]. The fact that both ACADS mRNA and protein are expressed in several distinct brain nuclei [9] and that MCAD is highly expressed in neuronal axons and dendrites [34] suggests a primary role for beta-oxidation in ATP production in these cells. Moreover, fatty acid oxidation enzyme activity, although low, is measurable in whole brain; for example in SCHAD (short chain 3-hydroxyacyl-CoA dehydroge-nase) knockout mice, enyzme activity was lost in an allele-specific manner with heterozygotes showing 50% of the activity of wildtype animals (Dr. Michael J. Bennett, personal communication). However, it is not yet known which cell types in the brain express these enzymes. Therefore clear evidence that brain cells use fatty acids as an oxidative source of energy production awaits future studies.
Pathway analysis identified oxidative phosphorylation as one of the top three pathways that were significantly activated by genotype in HF-fed mice (Fig. 4). The oxidative phosphorylation and electron transport carried out by the respiratory chain in the mitochondria is composed of five complexes that lead to the production of ATP. In our microarray study, expression of genes representing all five mitochondrial complexes was significantly altered by SCAD deficiency and HF diet. For example in HF Acads2/2 vs. HF Acads+/+ mice, we found significantly decreased expression of ATP synthase genes in mitochondrial complex 5 (Fig. 6A), thus providing evidence for decreased energy availability. Mitochondrial oxidative phosphorylation and resulting ATP levels in the brain play a key role in the control of feeding. The observation of significant changes in gene expression of neuropeptides and transmitters in whole brain led us to examine hypothalamic expression of selected genes involved in the metabolic control of food intake. For example, proopiomelanocortin is the precursor of melanocortin peptides synthesized in the arcuate nucleus of the hypothalamus and is an important regulator of food intake. The reduced hypothalamic Pomc expression in Acads-deficient animals fed HF diet is reminiscent of the response to fasting [35] providing further evidence of decreased energy availability.
The gene expression results of the current study revealed an over-representation of the biological processes chemosensory perception and olfaction in Acads2/2 mice as an effect of diet (Table 5). It is tempting to speculate that these pathways are involved in associative learning, e.g., that gene expression in the CNS was altered when the post-oral consequences of the diet were associated with their sensory characteristics in a memory forming process [36]. Learning about the energy consequences of food  PANTHER Analysis was used to identify processes that were overrepresented in Gene Ontology Biological Process categories. The reported P value was based on the number of differentially expressed genes obtained versus the number that would be expected. Calculation of the number of genes expected was based on 281 genes in Acads2/2 and 4014 genes in Acads+/+ mice. doi:10.1371/journal.pone.0041709.t005 enables animals to detect and distinguish between food sources in subsequent encounters, both in the wild and in laboratory settings. We propose that the behavior of Acads2/2 mice to avoid a fatbased diet and increase consumption of a carbohydrate-rich diet [8] may involve such learning.

Conclusions
We found altered brain expression of genes involved in energy production and energy sensing. Remarkably, the subset of genes deregulated in HF-fed Acads2/2 mice showed an expression pattern that strongly resembled that of Acads+/+ mice on LF diet, in particular, genes involved in both oxidative phosphorylation and AMP-kinase signaling. The differential expression of brain peptide and transmitter genes known to control food intake suggest possible downstream targets leading to changes in feeding behavior (Fig. 7). Therefore these genes could be responsible for the altered feeding behaviors observed in Acads2/2 mice. Figure 6. IPA analysis reveals oxidative phosphorylation pathway. Shown are the results from the IPA functional annotation after analysis of all the genes with a fold change value of equal or greater than 1.6 and P value of ,0.05. Up-regulated genes are colored in red, down-regulated genes in green. The diamond shapes with both colors contain gene families in which some genes are up-regulated, and some are down-regulated. Overall, the oxidative phosphorylation pathway was the most significantly activated pathway in the brain of HF Acads2/2 (ratio 0.414) vs. HF Acads+/ +, and of HF Acads+/+ vs. LF Acads+/+ (ratio 0.393) mice. Notably in HF Acads2/2 vs. HF Acads+/+ mice, we found significantly decreased expression of ATP synthase genes in mitochondrial complex 5 (top panel), consistent with a state of decreased energy availability. doi:10.1371/journal.pone.0041709.g006 Figure S1 Functional categories of genes with altered expression as a function of genotype or dietary fat. Transcripts were sorted based on gene ontologies assigned in the MGI database version 3.0. The majority of diet-regulated genes in Acads-deficient mice (54.2%) fell into four categories: unknown (21.4%), signal transduction (13.9%), protein metabolism (10.0%), and developmental processes (8.9%), while those altered in Acadsreplete mice (42.2%) were: transport (11.8%), developmental processes (10.3%), signal transduction (10.1%), and protein metabolism (10.0%). The majority of genes regulated by genotype on HF diet fell into categories of: transport (12%), developmental processes (10.6%), signal transduction (10.3%), and protein metabolism (10.1%). Those categories most affected by genotype with LF diet consisted of: unknown (36.4%), signal transduction (11.3), transport (8.0%), and developmental processes (5.8%). (TIFF)  Table S4 Quantitative real-time RT-PCR validation in whole brain of microarray data for selected genes in the experimental comparison: Acads2/2 HF vs. Acads+/+ HF.

(DOC)
Text S1 Quantitative real-time RT-PCR validation of microarray data for selected genes. (DOC)