Comparative transcriptional network modeling of three PPAR-α/γ co-agonists reveals distinct metabolic gene signatures in primary human hepatocytes.

Abstract

To compare the molecular and biologic signatures of a balanced dual peroxisome proliferator-activated receptor (PPAR)-α/γ agonist, aleglitazar, with tesaglitazar (a dual PPAR-α/γ agonist) or a combination of pioglitazone (Pio; PPAR-γ agonist) and fenofibrate (Feno; PPAR-α agonist) in human hepatocytes.
Gene expression microarray profiles were obtained from primary human hepatocytes treated with EC(50)-aligned low, medium and high concentrations of the three treatments. A systems biology approach, Causal Network Modeling, was used to model the data to infer upstream molecular mechanisms that may explain the observed changes in gene expression. Aleglitazar, tesaglitazar and Pio/Feno each induced unique transcriptional signatures, despite comparable core PPAR signaling. Although all treatments inferred qualitatively similar PPAR-α signaling, aleglitazar was inferred to have greater effects on high- and low-density lipoprotein cholesterol levels than tesaglitazar and Pio/Feno, due to a greater number of gene expression changes in pathways related to high-density and low-density lipoprotein metabolism. Distinct transcriptional and biologic signatures were also inferred for stress responses, which appeared to be less affected by aleglitazar than the comparators. In particular, Pio/Feno was inferred to increase NFE2L2 activity, a key component of the stress response pathway, while aleglitazar had no significant effect. All treatments were inferred to decrease proliferative signaling.
Aleglitazar induces transcriptional signatures related to lipid parameters and stress responses that are unique from other dual PPAR-α/γ treatments. This may underlie observed favorable changes in lipid profiles in animal and clinical studies with aleglitazar and suggests a differentiated gene profile compared with other dual PPAR-α/γ agonist treatments.

Full text

Page 1

Comparative Transcriptional Network Modeling of Three
PPAR-a/c Co-Agonists Reveals Distinct Metabolic Gene
Signatures in Primary Human Hepatocytes
Rene ´e Deehan1, Pia Maerz-Weiss2, Natalie L. Catlett1, Guido Steiner2, Ben Wong1, Matthew B. Wright2*,
Gil Blander1¤a, Keith O. Elliston1¤b, William Ladd1, Maria Bobadilla2, Jacques Mizrahi2,
Carolina Haefliger2, Alan Edgar{2
1Selventa, Cambridge, Massachusetts, United States of America, 2F. Hoffmann-La Roche AG, Basel, Switzerland
Abstract
Aims: To compare the molecular and biologic signatures of a balanced dual peroxisome proliferator-activated receptor
(PPAR)-a/c agonist, aleglitazar, with tesaglitazar (a dual PPAR-a/c agonist) or a combination of pioglitazone (Pio; PPAR-c
agonist) and fenofibrate (Feno; PPAR-a agonist) in human hepatocytes.
Methods and Results: Gene expression microarray profiles were obtained from primary human hepatocytes treated with
EC50-aligned low, medium and high concentrations of the three treatments. A systems biology approach, Causal Network
Modeling, was used to model the data to infer upstream molecular mechanisms that may explain the observed changes in
gene expression. Aleglitazar, tesaglitazar and Pio/Feno each induced unique transcriptional signatures, despite comparable
core PPAR signaling. Although all treatments inferred qualitatively similar PPAR-a signaling, aleglitazar was inferred to have
greater effects on high- and low-density lipoprotein cholesterol levels than tesaglitazar and Pio/Feno, due to a greater
number of gene expression changes in pathways related to high-density and low-density lipoprotein metabolism. Distinct
transcriptional and biologic signatures were also inferred for stress responses, which appeared to be less affected by
aleglitazar than the comparators. In particular, Pio/Feno was inferred to increase NFE2L2 activity, a key component of the
stress response pathway, while aleglitazar had no significant effect. All treatments were inferred to decrease proliferative
signaling.
Conclusions: Aleglitazar induces transcriptional signatures related to lipid parameters and stress responses that are unique
from other dual PPAR-a/c treatments. This may underlie observed favorable changes in lipid profiles in animal and clinical
studies with aleglitazar and suggests a differentiated gene profile compared with other dual PPAR-a/c agonist treatments.
Citation: Deehan R, Maerz-Weiss P, Catlett NL, Steiner G, Wong B, et al. (2012) Comparative Transcriptional Network Modeling of Three PPAR-a/c Co-Agonists
Reveals Distinct Metabolic Gene Signatures in Primary Human Hepatocytes. PLoS ONE 7(4): e35012. doi:10.1371/journal.pone.0035012
Editor: Zhi Xie, National Institutes of Health, United States of America
Received October 24, 2011; Accepted March 8, 2012; Published April 13, 2012
Copyright: ? 2012 Deehan et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by F. Hoffmann-La Roche AG (www.roche.com). The study design and conduct of the hepatocyte treatments, RNA isolation,
microarray hybridizations, and statistical analyses were performed in the laboratories of F. Hoffmann-La Roche, while data analysis was performed at Selventa
Laboratories (www.selventa.com). The decision to publish and manuscript preparation was made jointly by scientists and F. Hoffmann-La Roche and Selventa
Laboratories. Editorial assistance was funded by F. Hoffmann-La Roche AG.
Competing Interests: At the time of study design/implementation AE, CH, GS, JM, MBW and PM-W were employed by F. Hoffmann-La-Roche, Basel,
Switzerland, while BW, GB, KOE, NLC, RD and WL were employed by Selventa, Cambridge, United States of America. This work was supported by F. Hoffmann-La
Roche AG (www.roche.com). This does not alter the authors’ adherence to all the PLoS ONE policies on sharing data and materials.
* E-mail: matthew.wright@roche.com
¤a Current address: Department of Biology, Massachusetts Institute of Technology, Cambridge, Massachusetts, United States of America
¤b Current address: CHDI Foundation, Princeton, New Jersey, United States of America
{ Deceased.
Introduction
The peroxisome proliferator-activated receptors (PPARs) are
nuclear hormone receptors that function as transcription factors,
regulating the expression of genes involved in the metabolism of
fatty acids and carbohydrates. PPAR-a is highly expressed in the
liver, kidney and skeletal muscle. Fibrates, which are weak PPAR-
a agonists, alter hepatic energy metabolism that, in part, lowers
plasma triglyceride (TG) levels and slightly increases high-density
lipoprotein cholesterol (HDL-C) [1,2]. PPAR-c is highly expressed
in adipocytes and, to a lesser extent, in other tissues, including the
liver. PPAR-c agonists, such as pioglitazone and rosiglitazone,
improve insulin sensitivity, in part by reducing adipose TG
lipolysis [1]. PPAR-a and PPAR-c have also been shown to have
anti-inflammatory properties in insulin-resistant states, leading to
improvements in levels of cardiovascular (CV) risk markers [3–5].
Drugs that activate both PPAR-a and PPAR-c may be expected
to capture both the favorable changes in lipid profiles and the
glycemic benefits, and have been pursued for their therapeutic
potential to reduce CV risk associated with type 2 diabetes mellitus
(T2DM). However, several previous dual PPAR-a/c agonists were
discontinued due to compound-specific side effects: tesaglitazar
PLoS ONE | www.plosone.org1April 2012 | Volume 7 | Issue 4 | e35012

Page 2

due to data suggesting irreversible impairment of renal function
[6,7] and muraglitazar due to concerns regarding increased CV
risk [8]. The lessons emerging from these experiences are that each
PPAR-targeted drug is different and that agonist profile alone may
not reflect the underlying biology that defines the balance between
efficacy and safety of specific molecules. In support of this notion, a
recent report has shown that phosphorylation of PPAR-c by
cyclin-dependent kinase 5 is an important determinant of adipose
insulin sensitivity [9]. Selective recruitment of co-activators or co-
repressors may provide additional mechanisms whereby different
PPAR-targeted drugs can differentially activate target pathways.
Aleglitazar is a novel PPAR-a/c agonist that was designed to
provide balanced activity on both receptors. Preclinical studies
have shown that aleglitazar potently decreases blood glucose and
insulin resistance and improves plasma lipids in animal models
[10,11]. In a Phase II clinical trial in patients with T2DM,
aleglitazar provided effective glycemic control, improved lipid
parameters, blood pressure and markers of CV risk, and was well
tolerated [12]. Aleglitazar is currently under investigation in a
large Phase III outcomes trial (ALECARDIO) evaluating its effect
on CV mortality and morbidity in post-acute coronary syndrome
patients with T2DM [13].
Since emerging evidence suggests that each PPAR drug has a
unique efficacy and safety profile, we investigated the molecular
effects of two different dual PPAR-a/c agonists (aleglitazar and
tesaglitazar) and a combination of a PPAR-c and PPAR-a agonist
(pioglitazone plus fenofibrate [Pio/Feno]) in human hepatocytes
using whole-genome microarray studies in conjunction with a
systems biology approach, Causal Network Modeling (CNM).
Gene-expression profiling studies provide a broad, unbiased view
of the transcriptional response to drug treatment, but the large
amounts of data generated often require systems-based analytics
for comprehensive interpretation. One common approach is to
identify differentially expressed genes whose corresponding gene
products act in biologic pathways of interest and extrapolate the
biologic changes that may result from the changes in gene
expression. Enrichment statistics may also be used to determine
over-representation of genes involved in specific pathways in the
set of differentially expressed genes [14]. This approach relies on
the assumption that changes in gene expression directly correlate
with changes in protein activity and is thereby limited by the
validity of this assumption, as well as the inability to reason
downstream of genes of unknown function.
CNM is a complementary approach that uses reverse-
engineering principles to infer the activity states of signaling
mechanisms that lead to the observed changes in gene expression.
These inferences, called ‘‘hypotheses’’, are then evaluated
statistically and those that meet predefined statistical criteria are
assessed for biological integrity. Examples of this vetting process
include the following interrogations: determining whether an
inferred change in the activity state of a biological mechanism or
pathway is consistent with what would be expected in the context
of the experiment under investigation, and examining whether the
inferred mechanism is known to be expressed in the tissue under
study. Causal links between different signaling mechanisms are
also investigated and used to build models for disease states or
therapeutic responsiveness, discover mechanisms of action, assess
biomarkers for patient stratification or, as done here, differentiate
the molecular effects induced by multiple treatments. This data-
driven approach has been successfully applied to confirm or
propose novel hypotheses, to identify molecular mechanisms
involved in diverse biological processes in vitro [15–17] and in vivo
[18,19], and including discovery of potential biomarkers of clinical
response to therapeutic intervention [19]. Example applications of
CNM include models built that describe pulmonary proliferative
and cellular stress processes [16,17], hypoxia-induced hemangio-
sarcoma [20], the role of sirtuin-1 in human keratinocyte
differentiation [16] and response to drug treatment [15,18,19].
The aim of the present CNM study was to compare the effects
of aleglitazar with tesaglitazar and with Pio/Feno in a primary
human hepatocyte model system to elucidate the molecular
pathways modulated by aleglitazar and comparator treatments.
Our study revealed that aleglitazar has caused a distinct gene
signature profile compared with other dual PPAR-a/c agonist
treatments that likely underlies the observed favorable changes in
lipid profiles in animal and clinical studies [10–12].
Methods
Cell culture
Primary human hepatocytes (Hepacult GmbH, Regensburg,
Germany) were seeded at a density of 0.156106/cm2in six-well
plates. They were then incubated for 6 hours at 37uC in a 5%
carbon dioxide-humidified atmosphere in 1.5 ml of William’s
Medium E + 2 mM L-glutamine (Sigma-Aldrich Corp., St Louis,
MO, USA) containing aleglitazar (0.013, 0.064 or 0.32 mM),
tesaglitazar (1.44, 7.2 or 36.0 mM), Pio/Feno (6.0/0.11, 30.0/0.56
or 150.0/2.8 mM) or vehicle (0.1% dimethyl sulfoxide). The
concentrations were chosen based on in vitro transactivation data
providing EC50-aligned low, medium and high concentrations of
each of the three treatments, such that there was similar receptor
activation across treatments at each concentration level.
RNA preparation
Total RNA was extracted with TRI ReagentH (Molecular
Research Center, Inc., Cincinnati, OH, USA) according to the
manufacturer’s instructions and RNA samples were stored at
280uC. Total RNA was quantified on a NanoDropTMspectro-
photometer (NanoDrop Technologies Inc., Wilmington, DE,
USA) and RNA integrity was determined on an Agilent 2100
Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA).
Gene-expression microarray analysis
Total RNA (1 mg/sample) was converted to complementary
DNA (cDNA) using the low-RNA input fluorescent linear
amplification kit (Agilent Technologies). cDNA was converted to
cRNA and biotinylated using biotin-LC vinyl CTP. Amplified and
biotinylated cRNA was quantified and quality checked by
formaldehyde agarose gel electrophoresis. After statistical ran-
domization of sample order, labeled cRNA (1.5 mg/sample) was
used for array hybridization using the Whole-Genome Gene
Expression Direct Hybridization Assay and Illumina Human-6
Expression BeadChips version 3 arrays (Illumina Inc., San Diego,
CA, USA). Staining and washing steps were performed as
suggested by the manufacturer. BeadChips were scanned using
an Illumina BeadArrayH reader (Illumina Inc.) and raw data were
extracted using BeadStudio? software (version 3.1.3.0) with Gene
Expression Module 3.4.0 (Illumina Inc.).
Gene-expression profiles were obtained using EC50-aligned low,
medium and high concentrations of the three treatments as
specified above, such that there was similar receptor activation
across treatments at each concentration level. Transcript data
were analyzed using five replicates of each treatment concentra-
tion in order to identify significant changes in gene expression
between the treatment and the vehicle control (RNA state
changes).
Unique Gene Signature Profile of PPAR-a/c Agonists
PLoS ONE | www.plosone.org2April 2012 | Volume 7 | Issue 4 | e35012

Page 3

Determination of significant RNA expression changes
RNA expression data were analyzed using the ‘‘affy’’ and
‘‘limma’’ packages of the Bioconductor suite of microarray analysis
tools available for the R statistical package [21,22]. Robust
microarray analysis background correction and quantile normal-
ization were used to generate microarray expression values. Prior
to analysis of changed genes, data were visualized using principal-
components analysis and hierarchical clustering to identify outlier
samples. An overall linear model was fitted to the data for all
sample groups and specific contrasts of interest were evaluated to
generate raw p-values for each probe set on the expression array
[23]. The Benjamini–Hochberg false discovery rate method [24]
was then used to correct for multiple testing effects. Probe sets
were considered to have changed qualitatively in a specific
comparison if an adjusted p-value of 0.05 was obtained, their
average expression intensity was .150 in any treatment group and
they had an absolute change .1.3-fold. Genes represented by
multiple probe sets were considered to have changed if at least one
probe set was observed to change. Gene-expression changes that
met these criteria were termed ‘‘statistically significant RNA state
changes’’ and had the directional qualities of ‘‘up’’ or ‘‘down’’ (i.e.
they can be upregulated or downregulated in response to
treatment). We confirm that the microarray data is MIAME
compliant and has been deposited in the GEO database
(#GSE33152)http://www.ncbi.nlm.nih.gov/geo/query/acc.
cgi?acc=GSE33152.
Evaluation of signaling networks using Causal Network
Modeling
The CNM platform uses a reverse engineering approach to
automatically generate hypotheses that are statistically significant
explanations of RNA state changes. The substrate for this
approach is a knowledgebase comprised of qualitative causal
relationships in a computable format, e.g. the transcriptional
activity of PPAR-c increases the RNA expression of insulin-
induced gene 1 (INSIG-1) [25]. These cause and effect relation-
ships are derived from the knowledgebase at the time of
completion/analysis of the present study. That version contained
biological interactions derived from 50,827 unique PubMed
entries encompassing studies in human, mouse and rat. The
human-specific assembly of this knowledgebase includes relation-
ships defined both in human and in rodent studies with gene and
proteins mapped to the human homologue. The current knowl-
edgebase contains more than 100,000 concepts and molecular
entities, and more than 400,000 causal relationships. Each
hypothesis is ranked according to two probabilistic scoring metrics,
richness and concordance, which examine distinct aspects of the
probability of a hypothetical cause explaining a given number of
RNA state changes. Richness is based on hypergeometric
distribution and represents the probability that the number of
observed RNA state changes connected to a given hypothesis
could have occurred by chance alone. Concordance is based on
binomial distribution and represents the probability that the
number of observed RNA state changes that match the
directionality of the hypothesis (e.g. increased or decreased kinase
activity for a kinase and increased or decreased transcriptional
activity for a transcription factor) could have occurred by chance
alone. Stronger concordance indicates high confidence in the
inferred directionality of the hypothesis. Those hypotheses with
concordance p,0.1 as well as richness p,0.1 were considered to
be statistically significant. Application of these significance values
to randomly generated data produces less than 5% of the number
of hypotheses meeting both significance criteria than are observed
for experimental data with similar numbers of RNA state changes.
Hypotheses were further investigated and prioritized by evaluation
of their biologic relevance to the experimental context, whether
they could be causally linked to phenotypes observed and
processes relevant to PPAR-a/c biology or metabolism, as
evidenced in the literature, and if they were causally downstream
of the experimental treatments. Hypotheses that were determined
to be biologically relevant for the model were assessed for their
relationship to other hypotheses of interest and then ordered
causally into small models based on causal relationships in the
knowledgebase or additional relevant knowledge identified in the
literature. Several examples of applications of this technology and
associated methodology have been published [15–19,26].
Explanation of hypothesis tables
As illustrated in Figure 1, green and red boxes indicate observed
increases or decreases in RNA expression of a given gene,
respectively. Gene expression is depicted as ‘‘exp’’ followed by the
National Center for Biotechnology Information gene symbol in
parentheses. Yellow and blue boxes indicate biologic processes or
protein activities that are inferred by reverse causal reasoning to
increase or decrease upon treatment, respectively. Darker shades
of yellow or blue indicate stronger concordance. Numbers in the
boxes indicate the number of gene-expression state changes that
support the inference. Negative numbers indicate an inferred
decrease, while positive numbers indicate an inferred increase.
Results
State change analysis
Aleglitazar, tesaglitazar and Pio/Feno treatment of primary
human hepatocytes resulted in both shared and distinct RNA
expression state changes (Figure 2). At the high drug dosage, only
82 out of 1,238 (6.6%) total state changes observed were induced
by all three agonist treatments (Figure 2B). There were only 29
(2.3%) and 36 (2.9%) state changes that were shared by
tesaglitazar and Pio/Feno or aleglitazar and tesaglitazar treat-
ments, respectively (Figure 2B). CNM was then used to generate
statistically significant inferences (i.e. those with both richness and
concordance p-values,0.1) of the molecular signaling networks
underlying these RNA expression changes, and to provide
explanations for the RNA expression changes in each treatment
group. Figure 3 shows the inferred biologic pathways affected by
each treatment. The 2,000 hypotheses/molecular predictions
represented in the knowledgebase were scored across all treatment
groups; 280 unique predictions were identified and 276 of these
met the criteria for statistical significance in at least one of the
three high drug dosage treatments (Figure 3) and since a consistent
direction of effect was inferred these were considered to be
hypotheses. The molecular pathways represented by these
hypotheses were then clustered under broader biologic processes,
e.g. PPAR signaling, metabolic pathways, inflammation and stress,
insulin signaling, cell proliferation and mitogen-activated protein
kinase pathways (Figure 4). The following sections detail the
biologic signatures that were inferred to be most robustly modified
by the three treatments.
PPAR-a signaling
In the liver, PPAR-a activation modulates lipid metabolism and
lipoprotein synthesis and secretion, affecting TGs, HDL-C
synthesis and regulation of bile acid synthesis and detoxification
[27]. CNM identified qualitatively similar PPAR-a signaling when
comparing the medium concentrations of all three treatments
(Figure 5A), with increased PPAR-a transcriptional activity
supported by broadly similar gene expression changes (94, 66
Unique Gene Signature Profile of PPAR-a/c Agonists
PLoS ONE | www.plosone.org3April 2012 | Volume 7 | Issue 4 | e35012

Page 4

and 75 for aleglitazar, tesaglitazar and Pio/Feno, respectively)
(Figure 5A).
Lipid metabolic pathways
All three treatments may promote activation of signaling
pathways related to a decrease in plasma TG levels (Figure 5).
Aleglitazar, tesaglitazar and Pio/Feno each upregulated RNA
expression of adipose differentiation-related protein (ADFP),
angiopoietin-like 4 (ANGPTL4) and apolipoprotein A-V (Apo-
AV) to similar degrees. In addition, CNM infers that the
abundance of ANGPTL4 itself is increased. Transcriptional
control of all three of these genes is coordinately modulated by
PPAR-a [28–30]. ADFP is a lipid droplet coat protein present in
non-adipose cells that do not express perilipin [28]. It is broadly
distributed in non-adipogenic liver and muscle tissues and plays a
surfactant role at the lipid droplet surface, mediating triacyl-
glycerol packaging [31]. ANGPTL4 and Apo-AV affect plasma
TG metabolism in part by either inhibiting or potentiating the
activity of lipoprotein lipase, which mediates TG hydrolysis [32–
35].
In contrast to the similar treatment effects observed in pathways
related to TG metabolism, CNM inferred that aleglitazar increases
expression of signaling pathways related to HDL-C metabolism to
a greater extent than tesaglitazar or Pio/Feno. In general,
aleglitazar treatment induced more changes in gene expression
related to HDL-C metabolism than tesaglitazar and Pio/Feno,
particularly at the high concentration (Figure 5A). ATP-binding
cassette transporter proteins A1 (ABCA1) and G1 (ABCG1)
promote HDL formation by increasing cholesterol efflux to pre-b
and a HDL particles [36]. Aleglitazar significantly increased
Figure 1. Key for figures and hypothesis tables. Green and red boxes=observed increases and decreases, respectively, in RNA expression of a
given gene. Gene expression is depicted as ‘‘exp’’ followed by the National Center for Biotechnology Information gene symbol in parentheses. Yellow
and blue boxes=biologic processes or protein activities that are inferred by the CNM approach to increase or decrease upon treatment, respectively.
Darker shades of yellow or blue=stronger concordance. Numbers in the blue or yellow boxes indicate the number of gene expression state changes
that support the inference: negative numbers indicate an inferred decrease and positive numbers indicate an inferred increase. Numbers in the green
or red boxes indicate the log fold change.
doi:10.1371/journal.pone.0035012.g001
Figure 2. Treatment-induced RNA and gene expression changes. (A) Total number of RNA state changes (up- or downregulation) in human
hepatocytes treated with aleglitazar, tesaglitazar or Pio/Feno. (B) Unique and shared gene expression following treatment with aleglitazar,
tesaglitazar or Pio/Feno.
doi:10.1371/journal.pone.0035012.g002
Unique Gene Signature Profile of PPAR-a/c Agonists
PLoS ONE | www.plosone.org4 April 2012 | Volume 7 | Issue 4 | e35012

Page 5

ABCA1 expression at all concentrations and ABCG1 expression at
the medium and high concentrations (Figure 5A). In contrast,
tesaglitazar increased ABCA1 expression at all concentrations but
did not have a significant effect on ABCG1, while Pio/Feno
increased ABCA1 and ABCG1 expression only at the highest
concentrations. The greater effect of aleglitazar on ABCA1 and
ABCG1 versus tesaglitazar and Pio/Feno suggests that aleglitazar
may increase HDL-C to a greater extent, with a superior effect on
reverse cholesterol transport than tesaglitazar or Pio/Feno.
The three treatments used in this study were also inferred to
have differential effects on pathways related to low-density
lipoprotein (LDL) synthesis, as exemplified by their effects on
sterolregulatory elementbinding
(SREBF1A) activity and fatty acid desaturase 1 and 2 (FADS1/2)
mRNA expression [37]. Aleglitazar and Pio/Feno were inferred to
significantly decrease SREBF1A activity (at the medium and high
concentrations of aleglitazar and the low and high concentrations
of Pio/Feno), while tesaglitazar decreased SREBF1A activity at
the medium concentration and increased it at the high
concentration (Figure 5A). Aleglitazar also significantly decreased
the expression of both FADS1 and FADS2 at medium and high
concentrations, while Pio/Feno decreased FADS2 expression only
at the low concentration. Tesaglitazar had no effect. These data
suggest that aleglitazar treatment may more strongly decrease
LDL than tesaglitazar.
Based on the above inferences, it was possible to develop a
model that represents how aleglitazar may affect a number of
molecular signaling pathways involved in lipid metabolism
(Figure 5B).
transcription factor1A
Cellular stress response
CNM inferred a relatively weak cellular stress response
following aleglitazar treatment compared with tesaglitazar and
Pio/Feno (Figure 6). The activity of nuclear factor (erythroid-
derived 2)-like 2 (NFE2L2), a transcription factor activated by, and
involved in, mediating oxidative and xenobiotic stresses [38], was
inferred to increase for the medium and high concentrations of
tesaglitazar and Pio/Feno (Figure 6A). These increases are
supported by 22 and 23 (tesaglitazar) and 38 and 40 (Pio/Feno)
gene expression changes, respectively. In contrast, statistically
significant changes in NFE2L2 activity were inferred only for the
low concentration of aleglitazar, while medium or high concen-
trations of aleglitazar had no effect. This weaker stress response to
aleglitazar was further supported by the absence of an inferred
effect of aleglitazar on nelfinavir-responsive genes (a proxy for
unfolded protein response and cellular stress) [39] at any
concentration, which was evident with the high concentrations
of both tesaglitazar and Pio/Feno. In addition, only aleglitazar was
inferred to decrease dimethylnitrosamine-responsive genes (a
proxy for liver damage) [40], consistent with the notion that
aleglitazar induces a minimal stress response. These inferences
allowed the modeling of molecular signaling pathways that could
lead to stress response, and of the potential effects of aleglitazar
treatment on these molecular mechanisms (Figure 6B).
Insulin signaling and cellular proliferation pathways
CNM inferred that aleglitazar would decrease expression of
insulin-responsive genes at all three concentrations (Figure 7A),
while tesaglitazar would have differing effects at the medium and
high concentrations (decreasing and increasing insulin-responsive
genes/pathways, respectively) and Pio/Feno would have no
significant effect at any concentration. In addition, CNM inferred
that all three treatments would affect multiple pathways involved
in decreased cell proliferation and survival signaling. All three
concentrations of aleglitazar were inferred to decrease the activity
of phosphoinositide-3-kinase (PI3K) and Yin-Yang 1 (YY1)
proteins, which leads to proliferation and survival [41,42], while
tesaglitazar had no significant inferred effect on PI3K at any
concentration and Pio/Feno decreased PI3K at the low and
medium concentrations. Similarly, only the low and medium
concentrations of tesaglitazar and Pio/Feno had inferred effects on
YY1 activity. All three treatments were also inferred to increase
the activities of signaling that lead to inhibition of proliferation,
namely phosphatase activity of phosphatase and tensin homolog
(PTEN) and transcriptional activity of forkhead box protein
(FOXO1) and tumor protein 53 (p53) [43]. Consistent with the
inferred changes in insulin signaling, there was also an inferred
increase in cyclic AMP and cAMP-responsive element-binding
protein 1 with all treatments, although the high concentration of
tesaglitazar failed to have any effect. Additionally, both aleglitazar
and Pio/Feno increased cyclin-dependent kinase inhibitor 1A
(CDKN1A) expression, which leads to cell cycle arrest [44], while
tesaglitazar had no significant effect at any concentration. The
molecular signaling pathway that could result in inhibition of
proliferation by these treatments is shown in Figure 7B and is
colored for measurements and inferences from the medium
concentration of aleglitazar.
Discussion
Analysis of gene transcript data from human hepatocytes using
CNM revealed that the balanced dual PPAR-a/c agonist
aleglitazar exerts molecular effects that are distinct from the
effects of tesaglitazar (a dual PPAR-a/c agonist) or combination
Pio/Feno (a PPAR-c agonist/PPAR-a agonist treatment), partic-
ularly with regard to lipid metabolism and the cellular stress
response.
All three treatments led to qualitatively similar PPAR-a
signaling, and to increases in genes related to decreases in TG
levels. The observed increase in ADFP mRNA expression is
consistent with published literature, which has demonstrated that
PPAR agonists can upregulate ADFP mRNA in vitro and in vivo
[28,45]. Although the exact function of ADFP is unclear, it is
Figure 3. Molecular pathways modulated in response to
treatment with aleglitazar, tesaglitazar or Pio/Feno. Number of
RNA expression changes observed following treatment with aleglitazar,
tesaglitazar or Pio/Feno. Of the 2,000 unique mechanisms represented
in the knowledgebase, 280 were considered statistically significant in at
least one treatment condition and are designated as ‘‘hypotheses’’.
doi:10.1371/journal.pone.0035012.g003
Unique Gene Signature Profile of PPAR-a/c Agonists
PLoS ONE | www.plosone.org5 April 2012 | Volume 7 | Issue 4 | e35012