Comprehensive Analysis of PPAR$\alpha$-Dependent Regulation of Hepatic Lipid Metabolism by Expression Profiling

2.1. Materials

Wy14643 was obtained from ChemSyn Laboratories (Lenexa, KS). Recombinant human insulin (Actrapid) was from Novo Nordisk (Copenhagen, Denmark). SYBR Green was from Eurogentec (Seraing, Belgium). DMEM, fetal calf serum, calf serum, and penicillin/streptomycin/fungizone were from Lonza Bioscience (Verviers, Belgium). Otherwise, chemicals were from Sigma (Zwijndrecht, The Netherlands).

2.2. Animals

Male pure-bred Sv129 and PPAR$\alpha$-null mice on a Sv129 background were used at 3–5 months of age (Jackson Laboratories, Bar Harbor, ME). Animals were fed normal laboratory chow (RMH-B diet, Arie Blok animal feed, Woerden, the Netherlands). Study 1: fed mice were killed at the end of the dark cycle. Fasting was started at the onset of the light cycle for 24 hours ($n=5$ per group). Studies 2 and 4: wild-type and PPAR$\alpha$-null mice were fed with Wy14643 for 5 days by mixing it in their food (0.1%, $n=5$ per group). Studies 2 and 4 were carried out independently and 2 years apart. Study 3: wild-type and PPAR$\alpha$-null mice fasted for 4 hours received a single dose of Wy14643 (400 $\mu$L of 10 mg/mL Wy14643 dissolved in 0.5% carboxymethylcellulose) and were killed 6 hours later ($n=5$ per group). Study 5: wild-type and PPAR$\alpha$-null mice at 2-3 months of age were given a high-fat diet (D12451, Research Diets, New Brunswick, NJ) for 20 weeks (composition available at http://www.researchdiets.com/pdf/Data%20Sheets/DIO%20Series.pdf). During the last week, half of the mice were given Wy14643 for 7 days by mixing it in their food (0.1%, $n=5$ per group). Livers were dissected and immediately frozen in liquid nitrogen.

All animal experiments were approved by the animal experimentation committee of Wageningen University and were carried out in conformity with the public health service (PHS) policy on humane care and use of laboratory animals.

2.3. Primary hepatocytes

Rat (Wistar) and mouse (sv129) hepatocytes were isolated by two-step collagenase perfusion as described previously [16]. Cells were plated on collagen-coated six-well plates. Viability was determined by Trypan Blue exclusion, and was at least 75%. Hepatocytes were suspended in William's E medium (Lonza Bioscience, Verviers, Belgium) supplemented with 10% (v/v) foetal calf serum, 20 m-units/mL insulin, 50 nM dexamethasone, 100 U/mL penicillin, 100 $\mu$g/mL of streptomycin, 0.25 $\mu$g/mL fungizone, and 50 $\mu$g/mL gentamycin. The next day, cells were incubated in fresh medium in the presence or absence of Wy14643 (10 $\mu$M) dissolved in DMSO for 24 hours, followed by RNA isolation.

Human hepatocytes and Hepatocyte Culture Medium Bulletkit were purchased from Lonza Bioscience (Verviers, Belgium). Human hepatocytes were isolated from a single donor. Cells were plated on collagen-coated six-well plates. Upon arrival of the cells, the medium was discarded and was replaced by Hepatocyte Culture Medium. The next day, cells were incubated in fresh medium in the presence or absence of Wy14643 (50 $\mu$M) dissolved in DMSO for 12 hours, followed by RNA isolation.

2.4. Affymetrix microarray

Total RNA was prepared from mouse livers and primary hepatocytes using TRIzol reagent (Invitrogen, Breda, The Netherlands). RNA was either pooled per group or treatment (studies 1 and 2, primary hepatocytes), or used individually (studies 3 and 4), and further purified using RNeasy micro columns (Qiagen, Venlo, the Netherlands). RNA integrity was checked on an Agilent 2100 bioanalyzer (Agilent Technologies, Amsterdam, the Netherlands) using 6000 Nano Chips according to the manufacturer's instructions. RNA was judged as suitable for array hybridization only if samples exhibited intact bands corresponding to the 18S and 28S ribosomal RNA subunits, and displayed no chromosomal peaks or RNA degradation products (RNA Integrity Number > 8.0). Ten micrograms of RNA were used for one cycle cRNA synthesis (Affymetrix, Santa Clara, CA). Hybridization, washing, and scanning of Affymetrix Genechip MOE430 (studies 1 and 2) or mouse genome 430 2.0 arrays (studies 3 and 4) was according to standard Affymetrix protocols.

Scans of the Affymetrix arrays were processed using packages from the Bioconductor project [17]. Expression levels of probe sets were calculated using GCRMA [18], followed by identification of differentially expressed probe sets using Limma [19]. Comparison was between fasted wild-type and fasted PPAR$\alpha$-null mice (study 1) or between Wy14643-treated wild-type and Wy14643-treated PPAR$\alpha$-null mice (studies 2–4). $P$-values were corrected for multiple testing using a false discovery rate method [20]. Probe sets that satisfied the criterion of FDR < 1% ($q$-value $<0.01$) and fold-change $>1.5$ were considered to be significantly regulated. Functional clustering of the array data was performed by a method based on overrepresentation of Gene Ontology (GO) terms [21].

For the primary hepatocytes, expression levels were calculated applying the multichip-modified gamma model for oligonucleotide signal (multi-mgMOS) [22] and a remapped chip description file [23].

All microarray datasets were deposited to gene expression omnibus (GEO). The GEO series accession numbers are as follows: study 1: GSE8290, study 2: GSE8291, study 3: GES 8292, study 4: GSE8295, primary hepatocytes: GSE8302.

2.5. RNA isolation and Q-PCR

Total RNA was extracted from tissues with TRIzol reagent (Invitrogen, Breda, the Netherlands). 1 $\mu$g of total RNA was reverse-transcribed with iScript (Bio-Rad, Veenendaal, the Netherlands). cDNA was PCR-amplified with Platinum Taq DNA polymerase (Invitrogen) on a Bio-Rad iCycler or MyIQ PCR machine. Primers were designed to generate a PCR amplification product of 100–200 bp and were taken from Primerbank (http://pga.mgh.harvard.edu/primerbank). Specificity of the amplification was verified by melt-curve analysis and evaluation of efficiency of PCR amplification. The sequence of primers used is available upon request. The mRNA expression of all genes reported was normalized to 36B4 or cyclophilin gene expression.

2.6. In silico screening of putative PPREs using a PPRE classifier

Genomic sequences for mouse genes spanning 20 kbp centered at the transcriptional start site (TSS) were extracted from the Ensembl database (NBCI36) and screened for DR1-type REs with predicted binding strength of at least 1%. The binding strength prediction was based on a PPRE classifier that uses a database of in vitro binding data for PPARs to assign predicted binding strength according to a classification scheme (Matilainen et al. submitted). The conservation of the putative PPREs between mouse, human, dog, and rat were evaluated using the Vertebrate Multiz Alignment and Conservation track available from UCSC genome browser (NCBI releases for human and mouse genomes, hg18 and mm8, February 2006).

2.7. Histological examination of liver

5 $\mu$ sections were cut from frozen liver pieces. For oil red O staining, sections were air dried for 30 minutes, followed by fixation in formal calcium (4% formaldehyde, 1% CaCl₂). Oil red O stock solution was prepared by dissolving 0.5 g oil red O in 500 mL isopropanol. An oil red O working solution was prepared by mixing 30 mL oil red O stock with 20 mL dH₂O. Sections were immersed on working solution for 10 minutes followed by extensive washes in H₂O. Haematoxylin and eosin staining of frozen liver sections was carried out as described (http://www.ihcworld.com/histology.htm).

3.1. Global analysis of PPAR$\alpha$-dependent gene regulation

We analyzed the data from 4 independent microarray studies to obtain a comprehensive picture of PPAR$\alpha$-dependent upregulation of gene expression in mouse liver. In the first study, mRNA was compared between livers of 24-hour fasted wild-type and PPAR$\alpha$-null mice. In the second study, mRNA was compared between liver of wild-type mice and PPAR$\alpha$-null mice fed the PPAR$\alpha$ agonist Wy14643 for 5 days. In these two studies, RNA was pooled from 4-5 mice and hybridized to Affymetrix MOE430A GeneChip arrays. Since no biological replicates were analyzed, only a fold-change threshold criteria could be applied. Using a cutoff of 1.5-fold, expression of a total of 1847 probesets was lower in 24-hour fasted PPAR$\alpha$-null mice compared with 24-hour fasted wild-type mice (Figure 1(a)) (http://nutrigene.4t.com/microarray/ppar2007). Using the same cutoff, 2234 probesets were at least 1.5-fold lower in the livers of PPAR$\alpha$-null mice fed Wy14643 compared to wild-type mice fed Wy14643 (http://nutrigene.4t.com/microarray/ppar2007). The number of probesets that overlapped between the two groups was 569. A large proportion of these genes, which are thus under control of PPAR$\alpha$ under pharmacological and physiological conditions, may represent target genes of PPAR$\alpha$.

In the third study, mRNA was compared between livers of wild-type mice and PPAR$\alpha$-null mice treated with Wy14643 for 6 hours, while in the fourth study mRNA was compared between livers of wild-type mice and PPAR$\alpha$-null mice fed Wy14643 for 5 days. Study 4 was carried out independently of study 2 in a different set of mice. For these two studies, biological replicates (4-5 mice per group) were run using Affymetrix mouse genome 430 2.0 GeneChip array, enabling statistical analysis of the data which was not possible for studies 1 and 2. Applying a false discovery rate of 0.01 and a 1.5-fold cutoff, 1679 probesets were lower in the livers of PPAR$\alpha$-null mice compared to wild-type mice 6 hours after treatment with Wy14643, and 2207 probesets after 5 days of feeding Wy14643 (Figure 1(b)) (http://nutrigene.4t.com/microarray/ppar2007). While the majority of genes regulated by PPAR$\alpha$ after 6 hours of Wy14643 treatment were also, and generally more significantly, regulated after 5 days of Wy14643 treatment (overlap of 1001 probesets), many genes were specifically or more significantly regulated after 6 hours, including the direct PPAR target G0S2 and the EL gene, respectively. The complete set of data from studies 2 and 4, which includes up- and downregulated genes, has been submitted to the Peroxisome Proliferators compendium assembled by Dr. J.C. Corton (US EPA, Research Triangle Park, USA). They will be analyzed in conjunction with numerous other microarray experiments involving peroxisome proliferators to obtain the “peroxisome proliferator transcriptome.” In addition, the datasets have been submitted to GEO.

3.2. Pathway analysis of PPAR$\alpha$-dependent gene regulation

Functional clustering analysis of the microarray data by Gene Ontology classification indicated that numerous Gene Ontology classes were overrepresented among the genes that were $>1.5$-fold upregulated in 24-hour fasted wild-type compared to 24-hour fasted PPAR$\alpha$-null mice. The same was true for the comparison between wild-type and PPAR$\alpha$-null mice treated with Wy14643 for 5 days. Among the overrepresented Gene Ontology classes, we found many classes that are known to be governed by PPAR$\alpha$, including fatty acid beta-oxidation, acyl-CoA metabolism, leukotriene metabolism, and peroxisome organization and biogenesis (http://nutrigene.4t.com/microarray/ppar2007). Interestingly, we also noticed that numerous Gene Ontology classes were specifically upregulated by PPAR$\alpha$ under fasting conditions or by Wy14643 feeding. The data suggest, for example, that pyruvate metabolism and posttranslational protein targeting to membrane are specifically regulated in a PPAR$\alpha$-dependent manner by Wy14643 but not by fasting. Indeed, it is clear that some genes (e.g., Acot2 and Cd36) are PPAR$\alpha$-dependently regulated by Wy14643 and much less so by fasting, whereas others (e.g., Gpam, Hmgcs2) are PPAR$\alpha$ dependently regulated by fasting and much less so by Wy14643. However, it is important to emphasize that the ErmineJ Gene Ontology classification, as any functional clustering analysis, needs to be interpreted carefully.

The Gene Ontology classification analysis of the comparison wild-type versus PPAR$\alpha$-null mice treated with Wy14643 for 6 hours (study 3) was almost identical to the analysis for mice treated with Wy14643 for 5 days (study 4), suggesting that most of the gene expression changes elicited by Wy14643 treatment are fast transcriptional responses in correspondence with direct regulation of gene expression by PPAR$\alpha$. One notable exception was the class representing the acute phase response, which was regulated by 5-day but not 6-hour treatment with Wy14643.

3.3. Comprehensive list of PPAR$\alpha$-targets involved in lipid metabolism

Using these lists of genes that are upregulated by PPAR$\alpha$ in mouse liver, we were able to create a comprehensive picture of PPAR$\alpha$-regulated genes connected with lipid metabolism. Genes in bold are PPAR$\alpha$ dependently regulated by Wy14643 and during fasting, representing a conservative list of PPAR$\alpha$ targets (Figure 2). Genes in normal font are PPAR$\alpha$ dependently regulated in any of the four studies included. From this picture, it is evident that rather than merely regulating the rate limiting enzyme in fatty acid oxidation, PPAR$\alpha$ appears to regulate virtually every single step in the peroxisomal and mitochondrial fatty acid oxidation pathway. Furthermore, many genes involved in fatty acid binding and activation, lipid transport, and glycerol metabolism were controlled by PPAR$\alpha$. What is remarkable is that PPAR$\alpha$ also governs the expression of numerous genes involved in the synthesis of fats, which runs counter to the idea that PPAR$\alpha$ mainly regulates fat catabolism. Several genes belonging to the lipogenic pathway have previously been recognized as PPAR$\alpha$ targets, including Mod1 and Scd1, yet the extent of regulation by PPAR$\alpha$ is unexpected [24]. Regulation of lipogenesis by PPAR$\alpha$ was mainly observed after Wy14643 treatment, and to a much lesser extent after fasting.

3.4. Novel putative targets of PPAR$\alpha$ involved in lipid metabolism

In addition to providing an overview of PPAR$\alpha$-dependent gene regulation, we were interested in identifying novel PPAR$\alpha$-regulated genes that are implicated in lipid metabolism. To that end, we went through the array data from studies 1 and 2 on the one hand, and studies 3 and 4 on the other hand, and selected a number of genes to generate a heat map showing their PPAR$\alpha$-dependent upregulation by fasting and/or Wy14643 (Figure 3). To our knowledge, none of the genes shown, all of which are involved in hepatic lipid metabolism, has yet been reported to be regulated by PPAR$\alpha$. This includes phosphatidylcholine transfer protein (lipoprotein metabolism), glycerol-3-phosphate acyltransferase (triglyceride synthesis), very low-density lipoprotein receptor, choline phosphotransferase (phosphatidylcholine synthesis), and leptin receptor. Since all of these genes, except Abcg5, Abcg8, and Lipe, were upregulated 6 hours after Wy14643 treatment, they possibly represent novel direct target genes of PPAR$\alpha$ in liver, although PPREs have yet to be identified in their respective gene promoters.

Eight genes (shown in bold, Figure 3) were selected for more detailed investigation of PPAR$\alpha$-dependent gene regulation. Three of these genes are expected to be involved in the breakdown of hepatic triglycerides towards fatty acids: adipose triglyceride lipase (Pnpla2), hormone sensitive lipase (Lipe), and monoglyceride lipase (Mgll). Recent studies suggest that this threesome of genes is responsible for adipose tissue lipolysis [25–27]. In addition, we selected endothelial lipase (EL, Lipg), a recently identified member of triglyceride lipase gene family that is a major determinant of plasma HDL cholesterol [28–30], and electron transferring flavoprotein dehydrogenase (Etfdh) and electron transferring flavoprotein $\beta$ polypeptide (Etfb), which are components of the electron transport chain and accept electrons from at least nine mitochondrial matrix flavoprotein dehydrogenases [31, 32]. Finally, we selected phosphatidylcholine transfer protein (Pctp), which is involved in lipoprotein metabolism, and thioredoxin interacting protein (Txnip), which was recently identified as a major regulator of the hepatic response to fasting, similar to PPAR$\alpha$. The selection of these genes was based entirely on perceived novelty and potential functional importance of the observed regulation. Using real-time quantitative PCR (Q-PCR), we confirmed that the expression of all 8 genes in liver was increased by Wy14643 feeding in a PPAR$\alpha$-dependent manner (Figure 4(a)). In addition, we measured regulation of expression of this set of genes by PPAR$\alpha$ during the course of fasting (Figure 4(b)). Expression of all 8 genes went up during fasting which, except for Pnpla2, was PPAR$\alpha$-dependent. However, the pattern of expression was remarkably different between the various genes, suggesting for each gene a complex and unique interplay between several fasting-dependent transcription factors, including PPAR$\alpha$.

3.5. PPAR$\alpha$-dependent regulation in primary hepatocytes

To examine whether the PPAR$\alpha$-dependent regulation of the set of genes shown in Figure 3 was not an indirect consequence of metabolic perturbations elicited by the experimental challenge, we studied the effect of PPAR$\alpha$ activation in primary mouse, rat, and human hepatocytes. Gene expression was first analyzed by microarray (Figure 5(a)), followed by targeted analysis of the selected 8 genes by Q-PCR (Figure 5(b)). Expression levels were calculated by applying a multichip modified gamma model for oligonucleotide signal (multi-mgMOS) [22] and a remapped chip description file [23] to allow for parallel analysis of the same gene within different species. Expression of almost every gene studied was highly upregulated by Wy14643 in mouse and rat hepatocytes, compared to a more modest or no induction in human hepatocytes. For reasons that are not completely clear, in human hepatocytes, data from Q-PCR and microarray did not always perfectly align. Overall, the data indicate that the PPAR$\alpha$-dependent regulation observed in vivo can be reproduced in primary hepatocytes. Furthermore, the data suggest that expression of 6 genes is governed by PPAR$\alpha$ in human as well.

3.6. In silico screening of putative PPREs

To evaluate whether the selected eight genes represent possible direct PPAR target genes, the (mouse) genes were analyzed for the presence of putative PPREs using an in silico screening method (Figure 6). Ten kbp up- and downstream of the TSS were examined. For each putative PPRE identified, the predicted PPAR subtype specific binding strength was determined. For each gene, at least one PPRE was identified that was conserved among rat, dog, and human. The Etfdh and Txnip genes were characterized by the presence of two very strong putative PPREs that were conserved in human. Up to six putative PPREs were identified in the Mgll gene, only one of which was conserved in human. A similar picture was found for Pnpla2. The putative PPREs located in the EL gene were weak and generally not conserved. Interestingly, a strong putative PPRE was identified in the Pctp gene, which however was not conserved in human. Conversely, the human Pctp gene contained several putative PPREs that were not conserved in mouse (data not shown).

3.7. PPAR$\alpha$ activation prevents hepatic lipid storage after fasting

Our data extend the role of PPAR$\alpha$ in hepatic lipid metabolism and suggest that PPAR$\alpha$ may govern triglyceride hydrolysis. To find out whether activation of the triglyceride hydrolysis pathway by PPAR$\alpha$ is associated with a decrease in hepatic triglyceride stores, we compared wild-type and PPAR$\alpha$-null mice fed an HFD for 20 weeks, followed by treatment for one week with Wy14643. Numerous studies, including ours [33], have shown that chronic HFD increases hepatic triglyceride stores. In wild-type mice fed the HFD, treatment with Wy14643 markedly decreased hepatic lipids (Figures 7(a) and 7(b)), as shown by smaller lipid droplets, which was paralleled by significant induction of expression of Pnpla2, Lipe, and Mgll (Figure 7(c)). These data suggest that induction of the triglyceride hydrolysis pathway may contribute to the overall reduction in liver triglycerides elicited by PPAR$\alpha$ activation.