, Volume 12, Issue 3, pp 221–229 | Cite as

Analysis of PPARα-dependent and PPARα-independent transcript regulation following fenofibrate treatment of human endothelial cells

  • Hiromitsu Araki
  • Yoshinori Tamada
  • Seiya Imoto
  • Ben Dunmore
  • Deborah Sanders
  • Sally Humphrey
  • Masao Nagasaki
  • Atsushi Doi
  • Yukiko Nakanishi
  • Kaori Yasuda
  • Yuki Tomiyasu
  • Kousuke Tashiro
  • Cristin Print
  • D. Stephen Charnock-Jones
  • Satoru Kuhara
  • Satoru Miyano
Original Paper


Fenofibrate is a synthetic ligand for the nuclear receptor peroxisome proliferator-activated receptor (PPAR) alpha and has been widely used in the treatment of metabolic disorders, especially hyperlipemia, due to its lipid-lowering effect. The molecular mechanism of lipid-lowering is relatively well defined: an activated PPARα forms a PPAR–RXR heterodimer and this regulates the transcription of genes involved in energy metabolism by binding to PPAR response elements in their promoter regions, so-called “trans-activation”. In addition, fenofibrate also has anti-inflammatory and anti-athrogenic effects in vascular endothelial and smooth muscle cells. We have limited information about the anti-inflammatory mechanism of fenofibrate; however, “trans-repression” which suppresses production of inflammatory cytokines and adhesion molecules probably contributes to this mechanism. Furthermore, there are reports that fenofibrate affects endothelial cells in a PPARα-independent manner. In order to identify PPARα-dependently and PPARα-independently regulated transcripts, we generated microarray data from human endothelial cells treated with fenofibrate, and with and without siRNA-mediated knock-down of PPARα. We also constructed dynamic Bayesian transcriptome networks to reveal PPARα-dependent and -independent pathways. Our transcriptome network analysis identified growth differentiation factor 15 (GDF15) as a hub gene having PPARα-independently regulated transcripts as its direct downstream children. This result suggests that GDF15 may be PPARα-independent master-regulator of fenofibrate action in human endothelial cells.


Endothelial cells Fenofibrate PPARα Transcriptome network 


Peroxisome proliferator-activated receptor alpha (PPARα) is a ligand-activated transcription factor belonging to the family of nuclear receptors. PPARα is highly expressed in liver, skeletal muscle, kidney, and heart and it regulates the transcription of genes involved in energy metabolism [1, 2]. Over the past decade, PPARα has been investigated as a therapeutic target and drugs targeting PPARα have been developed. Fenofibrate is one of the synthetic ligands of PPARα and has been widely used for the treatment of hyperlipidemia, type 2 diabetes, and cardiovascular diseases due to its lipid-lowering effects [3]. A reported molecular mechanism of the lipid-lowering effect is “trans-activation”, i.e., PPARα activated by fenofibrate forms a PPAR–RXR heterodimer complex, which binds to PPAR response elements (PPREs) in the promoter regions of genes involved in beta-oxidation and lipoprotein/cholesterol transport [2]. In addition, fenofibrate also has anti-inflammatory and anti-atherogenic functions, which are thought to be based on “trans-repression” mechanisms in endothelial cells, smooth muscle cells, and other vascular cells [4].

While the lipid-lowering molecular mechanism in the liver is well known, the anti-inflammatory mechanisms in vascular cells have not been fully investigated. In addition, there are some PPARα-independent drug effects in human endothelial cells. For example, fenofibrate has been shown to regulate the survival of cultured human retinal endothelial cells in PPARα-independent manner, since pretreatment with the PPARα antagonist, MK 886, did not alter this effect and since another selective agonist for PPARα, WY-14643, had no significant effect on cell survival [5]. Moreover, in human umbilical vein endothelial cells (HUVECs), fenofibrate has been shown to increase AMPK phosphorylation, but neither bezafibrate nor WY-14643 had the same effect [6]. Therefore, we speculate that fenofibrate has PPARα-independent actions in human endothelial cells.

The objectives of the present study are (1) to identify transcripts in HUVECs regulated by fenofibrate in a PPARα-dependent and -independent manner, (2) to construct dynamic Bayesian transcriptome networks to reveal PPARα-independent mechanisms of action of fenofibrate, and the master regulators of PPARα-independent transcripts, based on computational data analysis techniques. In order to do this, we treated HUVECs with fenofibrate after pretreatment (or not) with small interfering RNAs (siRNA) targeting PPARα, and then profiled mRNA abundance using microarrays. PPARα-dependent and -independent transcripts regulated by fenofibrate were identified from the microarray data. We also estimated a Bayesian transcriptome regulatory network showing gene–gene regulatory relationships that were activated by fenofibrate. Our transcriptome network analysis indicated that growth differentiation factor (GDF) 15 is a master regulator of fenofibrate action, which acts in a PPARα-independent manner. Based on our results, we conclude that effects of fenofibrate in HUVECs are largely PPARα-independent and GDF15 is a new candidate, which acts as a key mediator of fenofibrate action in endothelial cells.

Materials and methods

Cell culture

Human umbilical vein endothelial cells were isolated from umbilical cords by collagenase digestion and cultured at 37°C/5%CO2 in basal culture medium supplemented with a proprietary mixture of heparin, hydrocortisone, epidermal growth factor, FGF, and 2% fetal calf serum (FCS; EGM-2, Cambrex, Workingham, UK) as previously described [7].

Small interference RNA

Small interfering RNAs treatment was performed through lipid-based transfection method [8] with either PPARα siRNA (Cat. No. M-00343400) or control siRNA (Cat. No. D-001206-13), using SmartPool reagents (Dharmacon, Lafayette CO, USA) according to the manufacturers’ instructions. HUVECs from 10 individuals were pooled, plated at 2.2 × 105 cells in six-well plates, and allowed to recover for 24 h at 37°C/5%CO2, at which time they were ~80% confluent. For transfection, 5 μl/well of 20 μM RNAi (PPARα or control), 42 μl/well of siFECTAMINE (ICVEC, UK no longer available), and 500 μl/well of serum-free medium were incubated at room temperature for 5 min. The relevant transfection mix was then added to each well containing 500 μl of fully supplemented media and incubated at 37°C/5%CO2 for 3 h, after which the transfection medium was replaced with 3 ml of fresh fully supplemented medium. About 24 h post transfection, relevant wells were treated with 25 μmol fenofibrate (Sigma–Aldrich, UK) dissolved in DMSO (Sigma–Aldrich, UK). Cells were incubated at 37°C/5%CO2 for the time points in transcriptional profiling.

Transcriptional profiling using microarrays

Here, time 0 indicates the start point of observation immediately prior to exposure to the 25 μmol fenofibrate. In order to obtain fenofibrate time-course data of experiments designated N, A, B, and C in Fig. 1, total RNA was prepared from fenofibrate-treated or -untreated HUVECs using Trizol reagent (Invitrogen) and RNA quality was confirmed using an Agilent 2100 bioanalyser. Biotin-labeled complex cRNAs were prepared and hybridized to CodeLink UniSet I Human 20K according to the manufacturer’s protocols (GE Healthcare, Amersham, UK). Three fenofibrate-treated biological replicates (designated as A in Fig. 1), four untreated biological replicates (designated as N in Fig. 1), and single fenofibrate-treated HUVEC in the presence of siRNA (designated as B or C in Fig. 1) were generated using HUVECs pooled from 10 different individuals each time. The quality of the expression data from all microarrays was confirmed using CodeLink expression analysis Software (v. 4.1). In the pre-processing stage, missing values were estimated by the LSimpute ( and data normalization was performed with the loess as previously described [9]. The time-course microarray data sets are available through National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) and accessible through GEO series accession number GSE 15494.
Fig 1

Schematic view of this study. N non-treated HUVECs (n = 4), A HUVECs treated with 25 μmol fenofibrate (n = 3), B HUVECs adding control siRNA prior to the treatment of fenofibrate (n = 1), C HUVECs adding PPARα siRNA prior to the treatment with fenofibrate (n = 1)

Microarray analysis

We first compared fenofibrate-treated cells in data set A in Fig. 1 with untreated control cells in data set N (comparison 1). We also compared PPARα siRNA-treated cells in data set C with control siRNA-treated cells in data set B (comparison 2). In the comparison 1, we identified fenofibrate-regulated transcripts based on the Significant Analysis of Microarray (SAM) statistical test, which takes multiple testing into account by estimating a false discovery rate, “q value” [10]. If a transcript was up- or down-regulated 1.7-fold or more in data set A rather than in data set N, and had a SAM q value of ≤0.05, we regarded that transcript as a fenofibrate-regulated transcript. In the comparison 2, if a transcript was up- or down-regulated 1.7-fold or more in data set C rather than in data set B, we regarded that transcript as a PPARα-regulated transcript. From two comparisons, if a transcript was oppositely regulated by fenofibrate and PPARα siRNA, we regarded that transcript as a fenofibrate-regulated transcript through PPARα-dependent mechanisms. The remaining fenofibrate-regulated transcripts that did not meet the criterion in the comparison 2 were considered PPARα-independently regulated transcripts.

In Gene Ontology analysis of the comparison 1 by MetaGP [11], SAM P value of each gene at each time point is used to evaluate significance of Gene Ontology annotations (biological process). The Benjamini and Hochberg’s method [12] is used for the adjustment of multiplicity.

Transcriptome network estimation

For time-course microarray data, various computational techniques have been developed in bioinformatics for estimating transcriptome networks that show causal dependencies between expression data of genes in terms of Granger causality. Among them, we use dynamic Bayesian network (DBN) and non-parametric regression method [13], because this method does not require any data discretization and can capture non-linear dependencies between transcripts. Also, we use the node-set separation method [9], in order to find dynamical changes of activities of transcriptome networks by dosing with fenofibrate. From this combination, we can find which networks are activated over our observed time course. In this study, we estimated five transcriptome networks based on node-set separation method [9]. Each node-set consists of regulated transcripts at (1) 2 h, (2) 2 or 4 h, (3) 4 or 6 h, (4) 6 or 8 h, and (5) 8 and 18 h, respectively.

We briefly introduce our model, the DBN and non-parametric regression, as follows: let x(t) be the transcript expression value vector observed at time t (t = 1,…,T). The DBN model can be represented by statistical density functions as f(x(1),…,x(T)) = ΠtΠjfj(xj(t)|pj(t − 1), θj), where pj(t) is the expression value vector of the parent genes of jth gene and θj is the parameter vector. The essential point of the node-set separation method is to define a node-set for an observed time. In order to create the node-set, the first-order Markov process of the DBN is used, i.e., a node-set Nt at time t is defined by the union of the sets of active transcript sets At and At−1, where At is the active transcript set defined as the set of transcript whose expression values are significantly different from the control expression profiles. By using the DBNs, we estimated the transcriptome network Gt on Nt, and the final transcriptome network can be obtained as G = G1∪…∪GT. The final transcriptome network allows us to identify which transcripts are important depending on time and which pathways are activated by the drug. In the trascriptome network estimation, a non-redundant gene set was used, so if a transcript has multiple probes on microarray, the probe with the highest intensity was used.


In order to identify transcripts regulated by fenofibrate in HUVECs in a PPARα-dependent or -independent manner, we generated two types microarray data (Fig. 1). First, we selected HUVEC transcripts regulated by fenofibrate drug treatment by comparing time-course microarray data from fenofibrate-treated cells (data set A in Fig. 1) with untreated control cells (data set N in Fig. 1). Next, PPARα-dependently or -independently regulated genes were identified by comparing time-course microarray data from cells treated with fenofibrate in the presence or absence or anti-PPARα siRNAs, i.e., by the combination of experiments N and A and experiments B and C, as shown in Fig. 1. Then, we constructed dynamic Bayesian transcriptome networks to further analyze fenofibrate’s mode of action, especially focusing on PPARα-independent actions.

Fenofibrate treatment time-course microarray data on HUVECs

There are ~20,000 transcripts on the CodeLink Human Uniset I gene arrays. The expressed transcripts are defied as those having a flag value of “Good” as assigned by the CodeLink Expression analysis software in at least one of three replicates at each time point. In this criterion, about 17,600 (88%) transcripts at each time point are expressed. We identified transcripts significantly regulated by fenofibrate from microarray data at 2, 4, 6, 8 and 18 h time point, respectively (1.7-fold or more with SAM q value ≤ 0.05). The number of regulated transcripts at each time point is shown in Table 1 and this increases as time advances. Five percent or fewer of expressed transcripts are significantly regulated by fenofibrate based on our criteria (see “Materials and methods”). There are transcripts encoding several anti-inflammatory proteins, which have been already reported as regulated by fenofibrate, including CEBPB, NOS3, and THBS1 [14, 15, 16]. PPARα is also upregulated at both 8 and 18 h. We used the MetaGP tool ( [11] to detect significantly enriched Gene Ontology annotations (biological process) based on SAM P values of all transcripts at each time point. MetaGP analysis shows that the transcript sets are significantly enriched for transcripts encoding intracellular transport, splicing, and cell cycle related proteins at early time points, 2 or 4 h. However, at the middle or late time points, 6 or 8 h, transcripts annotated with angiogenesis is most significant (P value < 1.0 × 10−16). This is not unexpected, since fenofibrate has an angiogenesis regulation function in endothelial cells [16]. The detailed results of the MetaGP analysis are shown in Supplementary Tables S1–S5.
Table 1

The number of transcripts regulated by fenofibrate (1.7-fold or more with SAM q value ≤ 0.05)

Time point (h)

Up regulation

Down regulation
















PPARα siRNA knocked down

We profiled HUVECs mRNA abundance following fenofibrate treatment in the presence and absence of PPARα or control siRNA (Fig. 1). In order to confirm whether PPARα and control siRNA have on- or off-target effects, we looked at the intensity signals for PPARα and other two PPARs, PPARβ/δ and PPARγ (Fig. 2). PPARα transcript abundance in the control siRNA-treated or untreated cells were very similar (Pearson correlation coefficient = 0.74). However, the abundance of PPARα mRNA in the PPARα siRNA-treated cells decreased more than twofold relative to control at 8 and 18 h (Fig. 2). In contrast, the abundances of PPARβ/δ and PPARγ mRNAs were not affected by either PPARα or control siRNA (Fig. 2). Taken together, we conclude that PPARα siRNA specifically and effectively inhibited PPARα mRNA and control siRNA is unlikely to have appreciable off-target effects.
Fig 2

On- and off-target effects of PPARα and control siRNA in vitro. The mRNA encoding PPARα, PPARβ/δ, and PPARγ are detected in all microarray data. The level of PPARα mRNA is effectively reduced by PPARα siRNA, but other two PPAR mRNAs are unaffected. Each PPAR mRNA shows similar expression pattern in microarray data from control siRNA and untreated cells. The solid line represents experiments N and A, dashed line represents experiment B, and dot-dashed line represents experiment C in Fig. 1. The probe IDs of CodeLink are shown in the parentheses

Identification of PPARα-dependent and -independent transcriptional regulation by fenofibrate on HUVECs

The primary objective in this study was to identify PPARα-dependent and -independent transcriptional regulation by fenofibrate. According to our criteria described in “Materials and methods”, the number of PPARα-dependent fenofibrate-regulated transcripts are 0, 13, and 164 (0, 7, and 26% of significantly regulated transcripts) at the 4, 8, and 18 h time points, respectively (Tables 1, 2). This result was unexpected in spite of the fact that PPARα is a target transcript of fenofibrate and implies that PPARα may play a minor role in the action of fenofibrate in HUVECs. The list of PPARα-dependent transcripts is shown in the Supplementary Table S6.
Table 2

The number of transcripts oppositely regulated by fenofibrate and PPARα siRNA-mediated knock-down

Time points

4 h

8 h

18 h

Fenofibrate up regulation-PPARα siRNA down regulation




Fenofibrate down regulation-PPARα siRNA up regulation




Dynamic transcriptome network

In order to further analyze the mechanism of action of fenofibrate in PPARα-dependent or -independent manner, we estimated a transcriptome network by the DBN method using non-parametric regression and node-set separation based on active transcript sets for each time point [9]. Here, we mainly focused on the transcriptome network with transcript sets, which are significantly regulated at 8 or 18 h, because PPARα is regulated at the same time points. We initially evaluated our transcriptome network predictions by reference to our siRNA experiments in which PPARα was knocked down. We were reassured that 14 out of 28 PPARα child transcriptomes in the transcriptome network show PPARα-dependent manner in our siRNA experiments (hypergeometric test P value < 0.01) (Table 3). This result suggests that our transcriptome network methods can capture at least a subset of PPARα regulated genes. Adenosine A2a receptor (ADORA2A) is a PPARα-dependently regulated gene and is upregulated at 18 h. Adenosine is a potent inhibitor of inflammatory process, and previously increasing gene expression of ADORA2A has been shown to suppress multiple inflammatory responses by reducing E-selectin expression in HUVECs. The expression of ADORA2A has also been shown to inhibit NF-κB translocation to the nucleus independently of any effect on IκB degradation [17]. Phosphoinositide-3-kinase class 2 beta polypeptide (PIK3C2B) is one of phosphoinositide 3-kinase (PI3K) family. Fenofibrate has been shown to inhibit VEGF-induced cell migration in HUVECs through PI3K-Akt pathway [18].
Table 3

Gene list having PPARα-dependently (a) and -independently (b) regulated transcripts as its child in transcriptome network with statistical significance (P value < 0.01)

Gene name

Gene description



P value

(a) Gene list having PPARα-dependently regulated transcripts


RecQ protein-like 5





Dickkopf homolog 1





Regulatory solute carrier protein family 1 member 1





Chromosome 1 open reading frame 24










URB2 ribosome biogenesis 2 homolog





Peroxisome proliferator-activated receptor alpha




Gene name

Gene description



P value

(b) Gene list having PPARα-independently regulated transcripts


Solute carrier family 1 member 4





Chromosome 4 open reading frame 34










Hypothetical protein LOC148189





Growth differentiation factor 15




A The number of child transcripts in transcriptome network

B The number of PPARΑα-dependently regulated transcripts in “A” of (a)

C The number of PPARα-independently regulated transcripts in “A” of (b)

Hub genes in transcriptome networks or protein–protein interaction networks are recognized as master regulators in E. coli [19], in yeast [20], and in mammalian cells [7, 21, 22]. Therefore, we focused on the hub transcripts in our network. We defined hub transcripts as those transcripts with the top 5% of numbers of direct children in the network (Supplementary Table S7). Reassuringly, PPARα, a direct target of fenofibrate, is listed as a hub transcript. This seems reasonable because a drug targeted molecule is likely to be important for downstream gene–gene regulation relationships.

We also identified transcripts whose children in transcript network were formed due to significant enrichment of PPARα-independently regulated transcripts (Table 3). There are five transcripts including one EST or two hypothetical proteins, which meet significant P value < 0.01. Among them, solute carrier family 1 member 4 (SLC1A4), shows the highest significance (P value < 1.2E-03). This gene is a glutamate/neutral amino acid transporter and mediates the efflux of l-serine from glial cells and its uptake by neurons [23], but its relevance to human endothelial cells has not been reported previously. On the othert hand, growth differentiation factor 15/macrophage inhibitory cytokines 1 (GDF15/MIC-1) is a very interesting gene. This gene is a transforming growth factor beta family related protein that exerts multiple effects on cell fate such as on cell growth, differentiation, and, inflammatory and apoptotic pathways [24] and is regulated by several anti-tumor agents [25, 26]. GDF15 inhibits endothelial cell migration and decreases matrix metallopeptidase 2 (MMP2) activity produced by the HUVECs in a concentration-dependent manner [25]. These effects are very similar to fenofibrate’s effects [18, 27]. GDF15 is also listed as a hub transcript, therefore we focused on the 34 downstream children of GDF15 and 11 out of these 34 transcripts are related to apoptosis and cell death (Table. 4).
Table 4

The child genes of GDF15 in transcriptome network

Gene name

Gene description


ADAM metallopeptidase with thrombospondin type 1 motif, 1


Adrenergic, beta-2-, receptor, surface


Ariadne homolog 2 (drosophila)


Asparagine synthetase


Chromosome 1 open reading frame 24


Chromosome 14 open reading frame 139


Cyclin b1


Cdc6 cell division cycle 6 homolog (S. cerevisiae)


Cyclin-dependent kinase-like 3


Centrosomal protein 72 kDa


Ectodermal-neural cortex (with btb-like domain)


Family with sequence similarity 46, member a


4-Hydroxyphenylpyruvate dioxygenase-like


3-Hydroxymethyl-3-methylglutaryl-coenzyme a lyase


Heparan sulfate (glucosamine) 3-o-sulfotransferase 1


Heat shock 70 kDa protein 1B




Hypothetical protein loc157562


Hypothetical protein loc256273


Minichromosome maintenance complex component 3


Minichromosome maintenance complex component 10


NEDD4 binding protein 2-like 2


n-myc downstream regulated gene 1


Protocadherin alpha subfamily c, 2


Solute carrier family 5, member 6




Tbc1 domain family, member 2


Transforming growth factor, beta 2


Tubulin tyrosine ligase-like family, member 1


Uridine monophosphate synthetase


Vasoactive intestinal peptide


X-ray repair complementing defective repair in chinese hamster cells 2


Zinc finger protein 251




Fenofibrate is a synthetic ligand of a nuclear receptor PPARα that has been widely used for the treatment of metabolic disorders over the past decade [2, 3]. This drug has lipid-lowering action on liver, and in addition anti-inflammatory and anti-atherogenic action on endothelial cells and smooth muscle cells. The lipid-lowering actions of fenofibrate are thought to be based on PPARα’s binding to the PPREs within promoter regions of genes involved in lipid, lipoprotein, and glucose metabolism—a “trans-activation” mechanism [1, 2]. The anti-inflammatory and anti-atherogenic actions in vascular cells are thought to be mediated in part by PPARα-ligand complexes suppressing inflammatory signaling pathways through interaction with nuclear factor-κB (NF-κB) and activated protein 1 (AP-1) and through up regulation of IκBa [1, 4]. Nevertheless, the molecular mechanisms of fenofibrate actions in vascular cells remain poorly understood. For example, there are some reports suggesting that fenofibrate acts in PPRAα-independent manner on human endothelial cells [5, 6]. Revealing the mechanisms of fenofibrate’s actions on endothelial cells may lead to new drug development for diseases related to endothelial dysfunction, such as atherosclerosis, thrombosis, and hypertension.

In this study, we used microarray analysis of HUVECs treated with fenofibrate in the presence or absence of pretreatment with siRNA directed against PPARα (Fig. 1). We first identified transcripts that were significantly regulated by fenofibrate. The number of regulated transcripts gradually increased with time of fenofibrate exposure (Fig. 1) and fenofibrate’s effects were first clearly observed after 6 h of treatment which is consistent with previous report [15]. Interestingly, the transcript encoding PPARα was not detected to be upregulated until after 8 h of treatment. Then, we used the MetaGP tool to biologically interpret microarray data at each time point. MetaGP evaluated the significance of Gene Ontology terms based on a P value at each time point. This analysis suggested that inflammation related genes annotated by inflammatory response (GO:0006954), angiogenesis (GO:0001525), and cell adhesion (GO:0007155) were highly significant at middle or late time points (6, 8, and 18 h).

Next, we identified PPARα-dependently and -independently regulated transcripts by using microarray analysis in presence and absence of anti-PPARα siRNA prior to the treatment of fenofibrate (Fig. 1). Contrary to our expectations, data for most fenofibrate-regulated transcripts were consistent with PPARα-independent fenofibrate mechanisms of action, in spite of the fact that PPARα is a target of fenofibrate (Table 2). It is interesting to speculate about the possible explanations for this observation. Fenofibrate and other PPARα ligands appear to predominantly modulate the expression of genes that are traditionally thought to be activated by inflammatory pathways. These genes may not been significantly regulated by fenofibrate responses in HUVECs without a prior inflammatory stimulus, as was the case in our study. This lead to our observation of mainly PPARα-independent regulation of transcript abundance by fenofibrate. Another possible explanation is that fenofibrate might mainly act through PPARα-independent mechanisms in endothelial cells. Unlike the mechanism of lipid-lowering actions based on the binding PPREs of promoter regions of PPARα targeting genes, the anti-inflammatory mechanisms may be very complicated.

In order to address PPARα-dependent and -independent mechanisms of fenofibrate action in human endothelial cells, we estimated a transcriptome network by the DBN and node-set separation model [9]. In this network, PPARα has a significant number of transcripts as children, which we also showed were regulated by siRNA targeted towards PPARα (P value < 0.01) (Table 3). This suggests that our network estimation method may be capable of capturing gene–gene regulation relationships mediated by fenofibrate through PPARα. Despite this, a limitation of our network estimation method is the small number of time points of microarray data—this method might prove even more informative with more time points of microarray data.

Our microarray data show that many transcripts are regulated by fenofibrate in HUVECs in a PPARα-independent manner. Therefore, we focused on the PPARα-independently regulated transcripts in transcriptome networks in an attempt to reveal fenofibrate’s PPARα-independent molecular mechanisms. Some network hubs had significant numbers of PPARα-independently regulated transcripts as children (P value < 0.01) (Table 3). Based on the relevance of their biological functions and drug efficacy of fenofibrate, we focused on GDF15/MIC-1 belonging to the transforming growth factor beta (TGF-beta) superfamily. Previous work has shown that in HUVECs, GDF15 regulated the MMP-2 activity and cell migration that contributes to atherosclerotic lesion formation and increases the risk of plaque hemorrhage [25]. It is important to note that fenofibrate appears to affect similar pathways to GDF15 in HUVECs [18, 27]. GDF15 also induces apoptosis in tumor cells [28] as does fenofibrate [16, 29]. In fact, there are 11 apoptosis or cell death related genes (ADRB2, ASNS, CCNB1, CDC6, CDKL3, HSPA1B, MCM10, NDRG1, TGFB2, VIP, and XRCC2) downstream of GDF15 in our transcriptome network (Table 4). Moreover, although there is no study that reports the regulation of GDF15 expression by fenofibrate in human endothelial cells, several groups reported that PPARγ agonists induce GDF15 expression through PPARγ independent pathway in cancer cell lines [30, 31, 32]. Two molecular mechanisms are proposed in this regulation; transcriptional and post-transcriptional regulation. In the former, PPARγ agonists induce the expression of early growth response 1 (EGR-1) transcription factor and binding to the promoter regions of GDF15 [30, 31]. In the latter, PPARγ agonist inhibits GDF15 RNA degradation, thereby increasing GDF15 expression [32]. In these two proposed mechanisms, the phosphorylation of ERK1/2 by PPARγ agonists plays very important role [30, 32, 33]. Fenofibrate also induces the phosphorylation of ERK1/2 in adult rat ventricular myocytes [34]. Taken together, we hypothesize that GDF15 might be a novel target point of fenofibrate via the ERK pathway independent of PPARα activation in human endothelial cells.

In conclusion, there are several reports of PPARα-independent actions of fenofibrate, including experiments in human endothelial cells. Here, we used microarray analysis of HUVECs treated with fenofibrate with or without siRNA targeted towards PPARα and identified transcripts regulated by fenofibrate through putative PPARα-dependent and -independent mechanisms. Unexpectedly, most fenofibrate-regulated genes appeared to be regulated in a PPARα-independent manner. In order to further reveal PPARα-independent mechanisms, we estimated a transcriptome network by using DBN model. This network suggested that GDF15 may be an important regulator of PPARα-independent fenofibrate action in HUVECs. Previous studies have reported that in HUVECs treated by GDF15, cell migration is inhibited and MMP-2 activity is suppressed. Therefore, GDF15 may provide a novel mechanism for fenofibrate’s PPARα-independent actions in human endothelial cells.



The authors would like to thank two reviewers for their constructive comments and suggestions. Computation time was provided by the Super Computer System, Human Genome Center, Institute of Medical Science, University of Tokyo.

Supplementary material

10456_2009_9142_MOESM1_ESM.xls (450 kb)
(XLS 450 kb)
10456_2009_9142_MOESM2_ESM.xls (450 kb)
(XLS 451 kb)
10456_2009_9142_MOESM3_ESM.xls (450 kb)
(XLS 451 kb)
10456_2009_9142_MOESM4_ESM.xls (450 kb)
(XLS 451 kb)
10456_2009_9142_MOESM5_ESM.xls (450 kb)
(XLS 450 kb)


  1. 1.
    Lefebvre P, Chinetti G, Fruchart JC, Staels B (2006) Sorting out the roles of PPARα in energy metabolism and vascular homeostasis. J Clin Invest 116:571–580. doi:10.1172/JCI27989 PubMedCrossRefGoogle Scholar
  2. 2.
    Gervois P, Fruchart JC, Staels B (2007) Drug insight: mechanisms of action and therapeutic applications for agonists of peroxisome proliferator-activated receptors. Nat Clin Pract Endocrinol Metab 3:145–156. doi:10.1038/ncpendmet0397 PubMedCrossRefGoogle Scholar
  3. 3.
    Staels B, Maes M, Zambon A (2008) Fibrates and future PPARα agonists in the treatment of cardiovascular disease. Nat Clin Pract Cardiovasc Med 5:542–553. doi:10.1038/ncpcardio1278 PubMedCrossRefGoogle Scholar
  4. 4.
    Zandbergen F, Plutzky J (2007) PPARα in atherosclerosis and inflammation. Biochim Biophys Acta 1771:972–982PubMedGoogle Scholar
  5. 5.
    Kim J, Ahn JH, Kim JH, Yu YS, Kim HS, Ha J, Shinn SH, Oh YS (2007) Fenofibrate regulates retinal endothelial cell survival through the AMPK signal transduction pathway. Exp Eye Res 84:886–893. doi:10.1016/j.exer.2007.01.009 PubMedCrossRefGoogle Scholar
  6. 6.
    Murakami H, Murakami R, Kambe F, Cao X, Takahashi R, Asai T, Hirai T, Numaguchi Y, Okumura K, Seo H, Murohara T (2006) Fenofibrate activates AMPK and increases eNOS phosphorylation in HUVEC. Biochem Biophys Res Commun 341:973–978. doi:10.1016/j.bbrc.2006.01.052 PubMedCrossRefGoogle Scholar
  7. 7.
    Affara M, Dunmore B, Savoie C, Imoto S, Tamada Y, Araki H, Charnock-Jones DS, Miyano S, Print C (2007) Understanding endothelial cell apoptosis: what can the transcriptome, glycome and proteome reveal? Philos Trans R Soc Lond B Biol Sci 362:1469–1487. doi:10.1098/rstb.2007.2129 PubMedCrossRefGoogle Scholar
  8. 8.
    Spagnou S, Miller AD, Keller M (2004) Lipidic carriers of siRNA: differences in the formulation, cellular uptake, and delivery with plasmid DNA. Biochemistry 43:13348–13356. doi:10.1021/bi048950a PubMedCrossRefGoogle Scholar
  9. 9.
    Tamada Y, Araki H, Imoto S, Nagasaki M, Doi A, Nakanishi Y, Tomiyasu Y, Yasuda K, Dunmore B, Sanders D, Humphries S, Print C, Charnock-Jones DS, Sanders D, Tashiro K, Kuhara S, Miyano S (2009) Unraveling dynamic activities of autocrine pathways that control drug-response transcriptome networks. Pac Symp Biocomput 14:251–263Google Scholar
  10. 10.
    Tusher VG, Tibshirani R, Chu G (2001) Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA 98:5116–5121. doi:10.1073/pnas.091062498 PubMedCrossRefGoogle Scholar
  11. 11.
    Gupta PK, Yoshida R, Imoto S, Yamaguchi R, Miyano S (2007) Statistical absolute evaluation of gene ontology terms with gene expression data. Proceedings of the 3rd international symposium on bioinformatics research and applications. Lecture note in Bioinformatics, vol 4463. Springer-Verlag, pp 146–157Google Scholar
  12. 12.
    Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc B 57:289–300Google Scholar
  13. 13.
    Kim S, Imoto S, Miyano S (2004) Dynamic Bayesian network and nonparametric regression for nonlinear modeling of gene networks from time series gene expression data. Biosystems 75:57–65. doi:10.1016/j.biosystems.2004.03.004 PubMedCrossRefGoogle Scholar
  14. 14.
    Gervois P, Vu-Dac N, Kleemann R, Kockx M, Dubois G, Laine B, Kosykh V, Fruchart JC, Kooistra T, Staels B (2001) Negative regulation of human fibrinogen gene expression by peroxisome proliferator-activated receptor alpha agonists via inhibition of CCAAT box/enhancer-binding protein beta. J Biol Chem 276:33471–33477. doi:10.1074/jbc.M102839200 PubMedCrossRefGoogle Scholar
  15. 15.
    Goya K, Sumitani S, Xu X, Kitamura T, Yamamoto H, Kurebayashi S, Saito H, Kouhara H, Kasayama S, Kawase I (2004) Peroxisome proliferator-activated receptor α agonists increase nitric oxide synthase expression in vascular endothelial cells. Arterioscler Thromb Vasc Biol 24:658–663. doi:10.1161/01.ATV.0000118682.58708.78 PubMedCrossRefGoogle Scholar
  16. 16.
    Panigrahy D, Kaipainen A, Huang S, Butterfield CE, Barnés CM, Fannon M, Laforme AM, Chaponis DM, Folkman J, Kieran MW (2008) PPARα agonist fenofibrate suppresses tumor growth through direct and indirect angiogenesis inhibition. Proc Natl Acad Sci USA 105:985–990. doi:10.1073/pnas.0711281105 PubMedCrossRefGoogle Scholar
  17. 17.
    Sands WA, Martin AF, Strong EW, Palmer TM (2004) Specific inhibition of nuclear factor-kappaB-dependent inflammatory responses by cell type-specific mechanisms upon A2A adenosine receptor gene transfer. Mol Pharmacol 66:1147–1159. doi:10.1124/mol.104.001107 PubMedCrossRefGoogle Scholar
  18. 18.
    Goetze S, Eilers F, Bungenstock A, Kintscher U, Stawowy P, Blaschke F, Graf K, Law RE, Fleck E, Gräfe M (2002) PPAR activators inhibit endothelial cell migration by targeting Akt. Biochem Biophys Res Commun 293:1431–1437. doi:10.1016/S0006-291X(02)00385-6 PubMedCrossRefGoogle Scholar
  19. 19.
    Cooper TF, Morby AP, Gunn A, Schneider D (2006) Effect of random and hub gene disruptions on environmental and mutational robustness in Escherichia coli. BMC Genomics 7:237. doi:10.1186/1471-2164-7-237 PubMedCrossRefGoogle Scholar
  20. 20.
    Zotenko E, Mestre J, O’Leary DP, Przytycka TM (2008) Why do hubs in the yeast protein interaction network tend to be essential: reexamining the connection between the network topology and essentiality. PLOS Comput Biol 4:e1000140. doi:10.1371/journal.pcbi.1000140 PubMedCrossRefGoogle Scholar
  21. 21.
    Abdollahi A, Schwager C, Kleeff J, Esposito I, Domhan S, Peschke P, Hauser K, Hahnfeldt P, Hlatky L, Debus J, Peters JM, Friess H, Folkman J, Huber PE (2007) Transcriptional network governing the angiogenic switch in human pancreatic cancer. Proc Natl Acad Sci USA 104:12890–12895. doi:10.1073/pnas.0705505104 PubMedCrossRefGoogle Scholar
  22. 22.
    Basso K, Margolin AA, Stolovitzky G, Klein U, Dalla-Favera R, Califano A (2005) Reverse engineering of regulatory networks in human B cells. Nat Genet 37:382–390. doi:10.1038/ng1532 PubMedCrossRefGoogle Scholar
  23. 23.
    Yamamoto T, Nishizaki I, Nukada T, Kamegaya E, Furuya S, Hirabayashi Y, Ikeda K, Hata H, Kobayashi H, Sora I, Yamamoto H (2004) Functional identification of ASCT1 neutral amino acid transporter as the predominant system for the uptake of l-serine in rat neurons in primary culture. Neurosci Res 49:101–111. doi:10.1016/j.neures.2004.02.004 PubMedCrossRefGoogle Scholar
  24. 24.
    Ago T, Sadoshima J (2006) GDF15, a cardioprotective TGF-beta superfamily protein. Circ Res 98:294–297. doi:10.1161/01.RES.0000207919.83894.9d PubMedCrossRefGoogle Scholar
  25. 25.
    Ferrari N, Pfeffer U, Dell’Eva R, Ambrosini C, Noonan DM, Albini A (2005) The transforming growth factor-beta family members bone morphogenetic protein-2 and macrophage inhibitory cytokine-1 as mediators of the antiangiogenic activity of N-(4-hydroxyphenyl) retinamide. Clin Cancer Res 11:4610–4619. doi:10.1158/1078-0432.CCR-04-2210 PubMedCrossRefGoogle Scholar
  26. 26.
    Huang CY, Beer TM, Higano CS, True LD, Vessella R, Lange PH, Garzotto M, Nelson PS (2007) Molecular alterations in prostate carcinomas that associate with in vivo exposure to chemotherapy: identification of a cytoprotective mechanism involving growth differentiation factor 15. Clin Cancer Res 13:5825–5833. doi:10.1158/1078-0432.CCR-07-1037 PubMedCrossRefGoogle Scholar
  27. 27.
    Lin R, Liu J, Gan W, Yang G (2004) C-reactive protein-induced expression of CD40-CD40L and the effect of lovastatin and fenofibrate on it in human vascular endothelial cells. Biol Pharm Bull 27:1537–1543. doi:10.1248/bpb.27.1537 PubMedCrossRefGoogle Scholar
  28. 28.
    Martinez JM, Sali T, Okazaki R, Anna C, Hollingshead M, Hose C, Monks A, Walker NJ, Baek SJ, Eling TE (2006) Drug-induced expression of nonsteroidal anti-inflammatory drug-activated gene/macrophage inhibitory cytokine-1/prostate-derived factor, a putative tumor suppressor, inhibits tumor growth. J Pharmacol Exp Ther 318:899–906. doi:10.1124/jpet.105.100081 PubMedCrossRefGoogle Scholar
  29. 29.
    Holland CM, Saidi SA, Evans AL, Sharkey AM, Latimer JA, Crawford RA, Charnock-Jones DS, Print C, Smith SK (2004) Transcriptome analysis of endometrial cancer identifies peroxisome proliferator-activated receptors as potential therapeutic targets. Mol Cancer Ther 3:993–1001PubMedGoogle Scholar
  30. 30.
    Baek SJ, Kim JS, Nixon JB, DiAugustine RP, Eling TE (2004) Expression of NAG-1, a transforming growth factor-beta superfamily member, by troglitazone requires the early growth response gene EGR-1. J Biol Chem 279:6883–6892. doi:10.1074/jbc.M305295200 PubMedCrossRefGoogle Scholar
  31. 31.
    Chintharlapalli S, Papineni S, Baek SJ, Liu S, Safe S (2005) 1,1-Bis(3′-indolyl)-1-(p-substitutedphenyl)methanes are peroxisome proliferator-activated receptor gamma agonists but decrease HCT-116 colon cancer cell survival through receptor-independent activation of early growth response-1 and nonsteroidal anti-inflammatory drug-activated gene-1. Mol Pharmacol 68:1782–1792PubMedGoogle Scholar
  32. 32.
    Yamaguchi K, Lee SH, Eling TE, Baek SJ (2006) A novel peroxisome proliferator-activated receptor gamma ligand, MCC-555, induces apoptosis via posttranscriptional regulation of NAG-1 in colorectal cancer cells. Mol Cancer Ther 5:1352–1361. doi:10.1158/1535-7163.MCT-05-0528 PubMedCrossRefGoogle Scholar
  33. 33.
    Baek SJ, Wilson LC, Hsi LC, Eling TE (2003) Troglitazone, a peroxisome proliferator-activated receptor gamma (PPAR gamma) ligand, selectively induces the early growth response-1 gene independently of PPAR gamma. A novel mechanism for its anti-tumorigenic activity. J Biol Chem 278:5845–5853. doi:10.1074/jbc.M208394200 PubMedCrossRefGoogle Scholar
  34. 34.
    Duhaney TA, Cui L, Rude MK, Lebrasseur NK, Ngoy S, De Silva DS, Siwik DA, Liao R, Sam F (2007) Peroxisome proliferator-activated receptor alpha-independent actions of fenofibrate exacerbates left ventricular dilation and fibrosis in chronic pressure overload. Hypertension 49:1084–1094. doi:10.1161/HYPERTENSIONAHA.107.086926 PubMedCrossRefGoogle Scholar

Copyright information

© Springer Science+Business Media B.V. 2009

Authors and Affiliations

  • Hiromitsu Araki
    • 1
    • 7
  • Yoshinori Tamada
    • 2
  • Seiya Imoto
    • 2
  • Ben Dunmore
    • 3
  • Deborah Sanders
    • 3
  • Sally Humphrey
    • 3
  • Masao Nagasaki
    • 2
  • Atsushi Doi
    • 1
    • 7
  • Yukiko Nakanishi
    • 1
    • 7
  • Kaori Yasuda
    • 1
    • 7
  • Yuki Tomiyasu
    • 1
    • 7
  • Kousuke Tashiro
    • 5
  • Cristin Print
    • 6
  • D. Stephen Charnock-Jones
    • 3
    • 4
  • Satoru Kuhara
    • 5
  • Satoru Miyano
    • 2
  1. 1.Systems Pharmacology Research InstituteGNI LtdFukuokaJapan
  2. 2.Human Genome Center, Institute of Medical ScienceThe University of TokyoTokyoJapan
  3. 3.Department of Obstetrics and GynaecologyUniversity of CambridgeCambridgeUK
  4. 4.Cambridge National Institute for Health Research Biomedical Research CentreCambridgeUK
  5. 5.Graduate School of Genetic Resources TechnologyKyushu UniversityFukuokaJapan
  6. 6.Department of Molecular Medicine and Pathology, School of Medical SciencesThe University of AucklandAucklandNew Zealand
  7. 7.Cell Innovator Inc.FukuokaJapan

Personalised recommendations