Microarray analysis of differential gene expression in the liver of
lean and fat chickens
Emmanuelle Bourneufa,⁎, Frédéric Héraulta, Céline Chicaultb, Wilfrid Carréa, Sirine Assafa,
Annabelle Monnierc, Stéphanie Mottierb, Sandrine Lagarriguea, Madeleine Douairea,
Jean Mosserb, Christian Diota
aUnité Mixte de Recherche Génétique Animale, INRA-Agrocampus Rennes, IFR140-GFAS, 65 rue de Saint-Brieuc, CS 84215, 35042 Rennes cedex, France
bUnité Mixte de Recherche 6061, CNRS-Université de Rennes 1, IFR140-GFAS, 2 avenue du Professeur L. Bernard, 35043 Rennes cedex, France
cTranscriptome platform, IFR140-GFAS, Unité Mixte de Recherche 6061, CNRS-Université de Rennes 1, IFR140-GFAS, 2 avenue du Professeur L. Bernard, 35043
Rennes cedex, France
Received 27 September 2005; received in revised form 20 December 2005; accepted 21 December 2005
Received by M. D'Urso
Excessive adiposity has become a major drawback in meat-type chicken production. However, few studies were conducted to analyze the liver
expression of genes involved in pathways and mechanisms leading to adiposity. A previous study performed by differential display on RNAs
extracted from chicken livers from lean and fat lines allowed us to isolate cDNA products of genes with putative differential expression. In this
study, a cDNA microarray resource was developed from these products together with cDNAs from genes involved in or related to lipid
metabolism. This resource was used to analyze gene expression in the liver from lean and fat chickens. Some genes were found with a difference
in expression between lean and fat animals and/or correlated to adipose tissue weight. Cytochrome P450 2C45, thought to play a role in
biotransformation of steroids and poly-unsaturated fatty acids, was more expressed in lean chickens whereas fatty acid synthase, stearoyl-CoA
desaturase, sterol response element binding factor 1 and hepatocyte nuclear factor 4, respectively involved in lipogenesis and its regulation, were
more expressed in fat chickens. These results indicate that mechanisms involved in the expression and regulation of lipogenic genes could play a
key role in fatness ontogenesis in chickens from lean and fat lines.
© 2006 Elsevier B.V. All rights reserved.
Keywords: Expression; Liver; Chicken; CYP2C; Lipogenesis
Meat-type chickens have been selected for many years on
their growth rate, leading to an excessive fattening. This trait
now represents a major drawback in poultry industry, as it
decreases the efficiency of breeding, especially the yield of lean
meat (Chambers et al., 1981; Leclercq, 1988). It would then be
of particular interest to better understand the physiological
pathways involved in this trait, and to know which genes are
responsible for this variability of fattening.
In order to further characterize mechanisms and genes
leading to adiposity, two lines of meat-type chickens have been
developed. These lean and fat lines (LL and FL respectively,
Leclercq et al., 1980) have been divergently selected on their
abdominal adipose tissue weight through 7 generations, and
provide a relevant material to decipher the mechanisms
involved in differential fattening. They have been extensively
studied, at physiological and genetic levels (for example, see
Leclercq et al., 1994; Assaf et al., 2004). As lipogenesis mainly
occurs in the liver in birds (Leveille et al., 1968; O'Hea and
Leveille, 1969), most of studies have been performed on hepatic
tissues, and some of them have demonstrated that the main
Gene xx (2006) xxx–xxx
GENE-35335; No of Pages 9
Abbreviations: AT, adipose tissue; CYP2C, cytochrome P450 2C; DD,
differential display; FL, fat line; GO, gene ontology; LL, lean line; SAM,
significance analysis of microarrays.
⁎Corresponding author. Tel.: +33 2 23 48 54 61; fax: +33 2 23 48 54 70.
E-mail address: Emmanuelle.Bourneuf@gmail.com (E. Bourneuf).
0378-1119/$ - see front matter © 2006 Elsevier B.V. All rights reserved.
ARTICLE IN PRESS
source of variability between the two lines comes from the
hepatic fatty acid metabolism (Legrand and Hermier, 1992).
We have previously shown that the expression of some genes
involved in fatty acid synthesis (Acetyl-CoA carboxylase, ATP-
citrate lyase, malic enzyme) or desaturation (stearoyl-CoA
desaturase 1) was more important in the liver of fat animals
(Douaire et al., 1992; Daval et al., 2000). However, genetic
linkage analyses on a 3-generation cross between FL and LL did
not allow to identify any of these genes as responsible for the
variability of the trait (Assaf et al., 2004). Moreover, the number
of genes analyzed was too low to globally characterize the
differences observed between the two lines.
Thus, we decided to analyze the liver expression of a larger
number of genes. About 200 genes were first selected by the
differential display method (Liang and Pardee, 1992) and found
with a differential amplification between the 2 lines (Carré et al.,
2001). However, this kind of method providing a lot of false-
positive results (Sompayrac et al., 1995), gene expression must
be further validated before drawing any conclusion. Methods
such as northern-blot or quantitative RT-PCR may not be very
suited to the expression analysis of hundreds of fragments. The
microarray technology allows to perform a simultaneous
analysis of the expression of thousands of genes, and was used
as a tool to evidence genes differentially expressed between lean
and fat chickens. A collection of cDNA coming from genes
involved in hepatic energy metabolism and regulation of ex-
pression was developed in parallel, and spotted onto microarray
resource jointly with the differential display fragments.
This study describes how microarray experiments were
conducted and how data were normalized and focuses on genes
that are now of particular interest concerning the fattening
variability of the chickens studied.
2. Materials and methods
Male chickens (Gallus domesticus) used in this study were
from lean and fat lines divergently selected for high and low
abdominal fat content (Leclercq et al., 1980). The animals were
kept in similar environmental conditions, food-restricted
overnight and fed ad libitum 4h before slaughter. They were
sacrificed at 8weeks, since the fatness variability is the highest
between lines at 8–9weeks of age.
2.2. Differential display products and specific chicken lipid
metabolism cDNA isolations
Differential display (DD) products (195) were selected and
were amplified thanks to the primers used for differential
display reactions, i.e. arbitrary forward primers L1AP1 to
L1AP10, and L2T12G, L2T12C and L2T12A as reverse
primers as previously described (Carré et al., 2001). For each
fragment, a PCR was performed in a final volume of 100μl
containing 20mM Tris–HCl pH 8.3, 50mM KCl, 1.5mM
MgCl2, 100μM dNTPs, 0.2μM forward and reverse primers
and 5U Taq Polymerase (Invitrogen/Life Technologies).
Reaction mixtures were incubated in a thermocycler
programmed to conduct one cycle (95°C for 2min), 40cycles
(95°C for 30s; 55°C for 30s and 72°C for 40s) and finally one
cycle (72°C for 10min). These amplified products were then
cloned into pGem-T Easy vector according to the manufac-
turer's recommendations (pGem-T Easy Vector System Kit,
Promega), and transformed into INVα-F′ Escherichia coli
competent cells. Plasmids were further extracted with the
NucleoSpin Plasmid kit (Macherey-Nagel) as recommended by
Specific chicken cDNAs (128) were prepared from total
RNA (1–2.5μg) extracted from the liver of commercial broilers
with 2bases anchored oligo-d(T) primers and SuperscriptII
reverse transcriptase (InVitrogen). They were then amplified by
PCR as described above with primers designed in the 3′ end of
coding region of chicken genes using Primer3 software (Rozen
and Skaletsky, 2000; http://www.broad.mit.edu/cgi-bin/primer/
primer3.cgi) when the chicken sequence was available, or
primers were designed in the most conserved regions of human
orthologue genes. Amplification products were then cloned in
pCR3.1 plasmid using TA cloning kit (Invitrogen).
2.3. Preparation of probes and spotting of microarrays
The cDNA probes were PCR-amplified using the forward
(5′-GTAAAACGACGGCCA GT-3′) and reverse (5′-CAG-
GAAACAGCTATGAC-3′) M13 primers. The PCR conditions
were: (94 °C for 30s, 50 °C for 30s, 72 °C for 1min 20s)
×35cycles. Amplification products were analyzed by agarose
gel electrophoresis, purified on NucleoFast 96 PCR plates
(Macherey-Nagel) according to the manufacturer's instructions,
suspended in 3× SSC/betaine, and transferred into 384-well
plates by MultiPROBE II (Perkin-Elmer—Packard). The
cDNAs were finally printed onto amino-silane coated slides
GAPS II (Corning Life Sciences) using MicroGrid II (Bior-
obotics). Fluorescent oligonucleotides (TagF and TagR, Euro-
gentec) control probes and 3 ready-to-spot PCR products from
Arabidopsis thaliana (corresponding to RUBISCO activase
(RCA), root cap 1 (RCP1) and papain-type cysteine endopep-
tidase (XCP2), Stratagene) were also spotted at 3 different
concentrations (10, 40 and 80ng/μl). We spotted all the probes
in triplicate, except control probes in duplicate, resulting in
1296 spots. Replicate spots were located within the same block
(i.e. spotted with the same pin) but were not next to each other in
2.4. Total and poly(A)+ RNA extraction
We extracted total RNA according to Chomczynski and
Sacchi (1987) from the liver of 20 male chickens from the lean
line and 20 chickens from the fat line. These 40 RNA samples
(250μg) were then pooled 4 by 4, in an order depending from
the adiposity of the corresponding chickens and leading to 10
different RNA pools (1mg), 5 lean and 5 fat pools (1–5 and 6–
10, respectively, Fig. 1). We extracted poly(A)+ RNA from
these 10 pools using the PolyATtract® mRNA isolation system
III (Promega) as recommended by the manufacturer. Each of
2E. Bourneuf et al. / Gene xx (2006) xxx–xxx
ARTICLE IN PRESS
these 10 poly(A)+ preparations corresponded to a test sample. A
reference sample was also prepared by mixing poly(A)+ RNA
from the 10 pools in a super-pool.
2.5. Target preparation and microarray hybridization
We reverse-transcribed 500ng of poly(A)+ RNA with
amino-allyl-dUTPs (CyScribe Post-Labeling Kit, Amersham
Biosciences). The amino-allyl modified cDNAs were then
purified on GFX columns (GFX Purification Kit, Amersham
Biosciences) and further labeled by incubation with cyanine
dyes (Cy3 or Cy5), according to the manufacturer's recom-
mendations (Amersham Biosciences). Labeled cDNAs were
also purified on GFX columns and mixed with hybridization
buffer (5× Denhardt's, 3.5× SSC, 0.3% SDS, 0.5μg/μl yeast
RNA, 0.5μg/μl poly(A) RNA and 50% formamide). Then, 10μl
of each solution (test sample labeled with Cy3 and reference
sample labeled with Cy5) were mixed and incubated 2min at
100 °C, and 30min at 37 °C before hybridization. Hybridiza-
tions were performed in hybridization chambers (Corning Life
Sciences) overnight at 42 °C and slides were washed in 2× SSC,
0.1% SDS, then in 1× SSC and finally twice in 0.2× SSC. We
performed hybridization in triplicate. Exceptions were for
targets 3 and 8 performed in duplicate, leading to a total number
of 28 slides hybridized.
2.6. Microarray scanning and data acquisition
We read the microarrays with ScanArray® 4000 XL
Microarray Analysis system (Packard Biochip Technologies).
We analyzed the resulting images with the GenepixPro 4.0
software (Axon Instruments Inc.). The ratio is expressed as the
log (base 2) of the ratio of median pixel intensities. We filtered
the spots according to their quality, the criteria being the spot
shape, the median intensity of the spot and the homogeneity of
the local background pixels as well as the spot pixel intensities.
We thus eliminated spots having inadequate signals, as spots
with noisy backgrounds, spots with heterogeneous intensity for
the spot or the local background, and spots with more than 20%
of saturated pixels in both channels.
2.7. Data normalization and replicate filtering
Raw data were submitted to the MicroArray Data Suite of
Computed Analysis (MADSCAN; Le Meur et al., 2004; http://
madtools.org). A set of invariant genes between reference and
test samples is chosen by rank invariant method (Tseng et al.,
2001). These invariant genes are used to normalize gene
expressions, either globally, pin by pin or by a proximal
approach, according to the number of invariant spots in the
different array blocks. Finally, a print-tip group normalization is
performed by “Lowess fitness” (Dudoit et al., 2000) approach to
avoid intensity-dependent bias. Since we spotted all of the
probes in triplicate and reproduced the hybridization twice or
three times, we could filter them according to their reproduc-
ibility by the outlier detection median absolute deviation
(MAD) modified z-scored test (Burke, 2001; Le Meur et al.,
2004). Finally, data were expressed as Cy5/Cy3 ratio and log2-
2.8. Identification of differentially expressed genes
We identified “lean” versus “fat” differentially expressed
genes among normalized data using Significance Analysis of
Microarrays (SAM) algorithm (Tusher et al., 2001) implemen-
ted in TIGR MultiExperiment Viewer (Saeed et al., 2003, http://
www.tm4.org/mev.html). A Δ-value of 0.328 was chosen,
giving a reasonable cutoff of 1.66 and −1.81 in d-scores.
Expected values were generated at random after 252 data
permutations. We chose a two-class data method of calculation
and the lowest false discovery rate (FDR=5.2) and then
performed hierarchical clustering of differentially expressed
genes expression data. An euclidean distance metric and
average linkage were used for linkage to build the cluster of
genes. Student T-test p-values and correlation to adipose tissue
r-values were also calculated.
2.9. Gene ontology analyses
Gene ontology analyses of differentially expressed genes or
of all genes spotted on the arrays were performed using Gene
Ontology Tree Machine (GOTM) software (Zhang et al., 2004;
http://genereg.ornl.gov/gotm). Results were expressed as ob-
served values and were compared to results expected from gene
ontology analyses performed on human orthologues of
annotated chicken genes spotted on the array.
2.10. Real-time quantitative RT-PCR
Total RNA from pools 1 to 10 was treated with DNase I
using DNA-free kit (Ambion). Each sample (2.5μg) was
incubated during 5min at 65 °C, with 8μg oligo-d(T)25 and
10mM dNTPs, in a final volume of 12μl, before oligo-d(T)
annealing 5min on ice. First-strand cDNA synthesis was
performed during 50min at 42 °C, in the presence of 200 U
Fig. 1. RNA samples and adiposity of corresponding individuals.Chickens from
leanandfat lines were rankedaccording totheirabdominalfat weight,expressed
as percent of live body weight. RNA were extracted from the liver of these
animals and rankly pooled 4 by 4 in 5 lean (1–5) and 5 fat (6–10) test samples
for poly(A)+ RNA extraction and target preparations.
3 E. Bourneuf et al. / Gene xx (2006) xxx–xxx
ARTICLE IN PRESS
Superscript II, 0.1M DTT, first-strand buffer 1× and 40 U
RNasin® (Promega France). Reverse transcription was stopped
by a 15min incubation at 70 °C. Each primer pair was tested for
PCR amplification efficiency, with cDNA dilutions from 1/10 to
1/10,240. The efficiencies obtained were in a range of 99.7% to
101.2% (Table 1), and enabled us to choose an appropriate
cDNA dilution. A 1/80 dilution of each reaction was further
used for PCR. cDNA samples were mixed with 25μl SYBR®
Green PCR Master Mix (Applied Biosystems), and 300nM
reverse and forward primers for some genes (Table 1). Reaction
mixtures were incubated in an ABI Prism 7000 Sequence
Detection System (Applied Biosystems) programmed to
conduct one cycle (95 °C for 10min) and 40cycles (95 °C for
15s and 60°C for 60s). Results (fold changes) were expressed
as 2−ΔΔCtwith ΔΔCt=(Ct ij−Ct 18Sj)−(Ct i1−Ct 18S1),
where Ct ijand Ct 18Sjare the Ct forgene i and for 18S in a pool
or a sample (named j) and where Ct i1and Ct 18S1are the Ct in
pool 1 or sample 1, expressed as the standard. Student T-test p-
values between lean and fat pools and correlation to adipose
tissue weight were determined for each gene.
3.1. Microarray hybridization and data normalization
In order to provide more information on genes involved in
chicken adiposity, we performed transcriptome analyses. A
microarray resource was developed with 195 probes selected
from differential display (DD) analysis and 128 probes selected
by RT-PCR from chicken genes expressed at least in the liver
and known to be involved in or related to lipid metabolism and
thought to play a role in lipid homeostasis. DD products
represent 173 different contigs (data not shown) as determined
by CAP3 analyses (Huang and Madan, 1999). Among them, 74
correspond to annotated genes. Selected chicken liver specific
probes correspond to genes involved in lipid (40), carbohydrate
(39) and ketone body (2) metabolisms, in the regulation of these
metabolisms by signal transduction (18) and by regulation of
transcription (27) and translation (2). Altogether, 192 DD
product and selected genes were similar to human annotated
genes involved in metabolism (cellular, macromolecule and
primary metabolisms) and especially in lipid metabolism, as
determined by GO tree machine analyses of gene ontology
biological processes associated to these genes (data not shown).
These 323 probes were spotted in triplicate on amino-silane
coated glass slides. The experiment consisted in hybridization
of labeled cDNAs prepared from liver RNA samples of lean and
fat chickens. These RNAwere pooled into 10 groups, according
to abdominal adiposity of the chickens, and poly(A)+ RNAs
were extracted directly from these RNA pools. Each of these
pools was considered as a test sample, and hybridized with the
reference sample, constituted of a mix of RNAs from all
samples. We used 28 slides in total, as most of the test samples
were hybridized 3times. After image analyses, raw data were
acquired. We normalized and filtered them using MADSCAN
(Le Meur et al., 2004) as described in Materials and methods.
Finally, 219 clones remained before significance analysis of
Data from this study were deposited as a whole in the GEO
database, with the series record GSE3867, and according to
MIAME 1.1 recommendations.
3.2. Identification of differentially expressed genes
We performed significance analysis of microarrays (SAM)
on normalized data as described by Tusher et al. (2001) in order
Primer pairs used to analyze gene expression by quantitative PCR, and the corresponding amplification efficiencies
Gene Forward primera
a,bForward and reverse primers used for quantification of specific expressed RNAs by real-time, quantitative RT-PCR. Sequences are indicated from 5′ end to 3′ end of
cG1-8B is the probe corresponding to the CYP2C45 gene.
4 E. Bourneuf et al. / Gene xx (2006) xxx–xxx
ARTICLE IN PRESS
to identify genes expressed differentially in the liver. The results
shown in Table 2 were obtained with a 0.328 delta value
corresponding to 1.66 and −1.81 upper and lower cutoffs in d-
scores, respectively, and to an FDR value of 5.2 (i.e.
approximately 5 genes can be expected as false-positive).
Student T-test p-values and coefficient of correlation to adipose
tissue weight r-values are also indicated. Sixteen products were
found with a d-score greater than 1.9 or smaller than −1.9.
Interestingly, most of these genes have a significant (p<0.05)
Student T-test p-value and have an expression pattern
significantly correlated to adipose tissue weight (r>0.632 or
r<−0.632 for α=0.05). Six additional clones with d-score
ranging from 1.66 to 1.85 and from −1.92 to −1.81 were found
differentially expressed with significant or near significant
(p<0.1) p-values and/or with an expression pattern correlated to
adipose tissue weight. Altogether, 22 products, corresponding
to 17 genes, were found differentially expressed, i.e. over- or
under-expressed in fat chickens, respectively, when compared
to expression in lean chickens (Table 2 and Fig. 2).
The 12 up-regulated clones (Fig. 2A, upper panel, d>1.66)
were clustered in a group (Fig. 2B) including acetyl-CoA
carboxylase (ACC), fatty acid synthase (FAS), stearoyl-CoA
desaturase 1 (SCD1 and SCD1-c), Apo(lipo)protein A1
(ApoA1), sterol regulatory element binding transcription factor
1 (SREBP1 and SREBP1-c) and mitochondrial malate dehy-
drogenase 2 (MDH2 and MDH2-c) genes. All these genes are
involved in synthesis and secretion of lipids, along with
transcription factor HNF4 also included in the cluster with C2-
3C, similar to hypothetical CGI-109 protein and the C5-10A
clone, with unknown function.
Genes showing a differential hepatic expression between lean and fat chickens,
after analysis by microarrays
Correlation of the expression with the adipose tissue weight of the animals.
aSREBP1 and SREBP1-c correspond to the same probe used at different
concentrations, as well as MDH2 and MDH2-c, SCD1 and SCD1-c, and G1-8B
and G1-8B-c (corresponding to CYP2C45).
bProbes with differential expression were selected by statistical analysis of
microarrays (SAM). Score d-values and false discovery rate q-values are
cStudent T-test p-values between lean and fat targets (see Figs. 1 and 2, pools
1–5 and 6–10, respectively): p-values,⁎p<0.05;⁎⁎p<0.01.
dCorrelation to adipose tissue r-values :⁎r>0.632 or r<−0.632 for α=0.05;
⁎⁎r>0.765 or r<−0.765 for α=0.01.
Fig. 2. Up- and down-regulated genes. (A) Expression profiles from up-
(d>1.66, upper panel) and down-regulated genes (d<−1.81, lower panel) in
lean (1–5) and fat (6–10) targets prepared from chickens ranked according to
adipose tissue (AT) weight. (B) Hierarchical clustering of expression profiles of
up- and down-regulated genes. Expressions values are expressed as a color code
(bar color chart on top, red up- and green down-regulation) corresponding to the
log2 ratio of expression value in the test poly(A)+ sample on the value in the
reference super-pool of the 10 poly(A)+ samples. Missing values are shown in
grey. The cladogram on the left summarizes the relatedness of gene expression
across the 10 targets. SREBP1 and SREBP1-c are the same probe used at
different concentrations, as well as MDH2 and MDH2-c, SCD1 and SCD1-c,
and G1-8B and G1-8B-c. C2-3C, C5-10A, G1-8B/CYP2C45, G1-12B, G1-6Y,
G8-10A, A4-10B and C10-3A are probes originally from the differential display
5 E. Bourneuf et al. / Gene xx (2006) xxx–xxx
ARTICLE IN PRESS
The 10 down-regulated clones (Fig. 2A, lower panel, d<
−1.81) were clustered in a group (Fig. 2B) including DECR2
peroxisomal 2,4-dienoyl CoA reductase 2 (DCR2) involved in
peroxisomal beta-oxidation of fatty acids, activating transcrip-
tion factor 4 (ATF4), corresponding to the cAMP-response
element binding protein 2 (CREB2), pyruvate carboxylase
(PC), G1-12B, G1-8B and G1-8B-c, that are homologues to
cytochrome P450 2C45 (CYP2C45, Baader et al., 2002), G1-
6Y similar to ras homolog gene family member F (RHOF), G8-
10A similar to human Ran GTPase activating protein 1
(RANGAP1), A4-10B similar to cytochrome b (CYTB) and
C10-3A similar to hepatic alpha amylase (AMY2A).
3.3. Analysis of expression by quantitative RT-PCR
In order to validate the differential analyses of microarray
data performed using SAM, we analyzed the expression of
some genes by SYBR® Green (Applied Biosystems) quantita-
tive RT-PCR. Quantitative RT-PCR were performed on 9 genes
found up-regulated by SAM method (FAS, SCD1, C5-10A,
MDH2, SREBP1, C2-3C, ApoA1, ACC and HNF4), on 8found
down-regulated (G1-8B/CYP2C45, DCR2, G8-10A, G1-6Y,
A4-10B, PC, C10-3A and ATF4), on 3 invariant genes (G8-5A,
X3-3C and Phosphatidic acid phosphatase 2C (PAP2C)) and on
18S rRNA as internal standard (Tables 3 and 4 and Fig. 3). As
expected, FAS, SCD1, ACC and SREBP1 were found
expressed with significant difference (p<0.05) between lean
and fat samples and up-regulated in fat chickens (Table 3).
HNF4, despite of a non-significant Student T-test, shows the
same trend as this lipogenic genes. The invariant genes selected
(fold change <1.35) were found expressed with non-significant
difference (p>0.05, Table 4), were used as internal standards
and showed equivalent results compared to the use of 18S RNA
(data not shown). Some genes identified by SAM method as
differentially expressed (C5-10A, MDH2, DCR2 and G8-10A)
were not found differentially expressed by RT-PCR (Table 3).
Surprisingly, two genes (PC and CYTB, the A4-10B clone) that
were considered as up-regulated in lean animals in microarray
data show a significant up-regulation in fat animals according to
the quantitative PCR results. Moreover, the CYP2C45 (G1-8B)
differential expression was not confirmed although found
correlated to adipose tissue (−0.72). As observed in Fig. 3,
CYP2C45 fat data were found with a great standard error
(standard error on mean ratio near 2.6). This could result from
Comparison of microarrays and real-time RT-PCR analyses of up- and down-
aQuantitative RT-PCR validations were performed on up- (FAS, SCD1, C5-
10A, MDH2, SREBP1 and HNF4) and down- (G1-8B (CYP2C45), DCR2, G8-
10A and ATF4) regulated genes.
bFold changes (F/L: mean of fat sample values on mean of lean sample
values). Microarray F/L fold changes are taken from Table 2.
cStudent T-test p-values (lean versus fat target).⁎p<0.05,⁎⁎p<0.01.
Comparison of microarrays and real-time RT-PCR analyses of invariant genes
and 18S rRNA control
aQuantitative RT-PCR validations were performed on invariant genes (G8-
5A, X3-3C and PAP2C) and 18S rRNA control.
bFalse discovery rate q-values from SAM analysis.
cStudent T-test between lean and fat samples p-values.
Fig. 3. Real-time RT-PCR validation of microarray data. Expression levels of
genes found up-regulated (FAS, SCD1, C5-10A, MDH2, SREBP1, C2-3C,
ApoA1, ACC and HNF4), down-regulated (G1-8B/CYP2C45, DCR2, G8-10A,
G1-6Y, A4-10B, PC, C10-3A and ATF4) and invariant (G8-5A, X3-3C and
PAP2C) by SAM method were determined in the 10 lean and fat samples. The
figure represents the RNA levels (2−ΔΔCtunits) of each gene in the lean and fat
pools, considering the mean value of the lean pools as standard. Exception is for
G1-8B/CYP2C45, where the fat pools mean is used as a standard. The dotted
line indicates a 2-fold change threshold in RNA levels, and the stars point to
genes with a significant Student T-test p-value between fat and lean animals
6E. Bourneuf et al. / Gene xx (2006) xxx–xxx
ARTICLE IN PRESS
an experimental artifact and explain the p=0.17 calculated.
Therefore, RT-PCR analysis of CYP2C45 expression was
repeated and on larger series of samples corresponding to the 20
lean and 20 fat chicken liver individual RNAs included in the
pooled samples (Fig. 4). The same was done for SREBP1, as
control. Significant correlation to adipose tissue was observed
for CYP2C45 and SREBP1 expressions (r=−0.58 and 0.48,
respectively; threshold value 0.4) and differences in expression
between lean and fat animals were found significant (p=0.004
and 0.0002, respectively) clearly indicating differential expres-
sion for these 2 genes.
3.4. Functional classification of differentially expressed genes
Functional classification of differentially expressed genes as
determined by SAM was generated by GO Tree Machine at the
5th annotation level under biological process and with a specific
reference gene list corresponding to annotated chicken genes
spotted on the array (Fig. 5). Some biological processes were
more or less expressed than expected. The more striking
discrepancies correspond to (i) organic acid metabolism,
cellular lipid and lipid metabolisms, and especially the afferent
6th or 7th level fatty acid biosynthesis process (data not shown)
and (ii) cellular macromolecule metabolism and cellular
catabolism, respectively more and less expressed than expected.
Microarray analyses were performed in order to analyze
differences in gene expression from chicken livers of lean and
fat lines divergently selected for low or high abdominal fat
weight (Leclercq et al., 1980) and correlation of these
expressions to abdominal adipose tissue weight. The microarray
resource developed in this study includes 195 probes selected
from DD analysis, and expected to include in some extent true
differential genes, and 128 specific genes selected for their
involvement in lipid homeostasis. These probes are similar to
192 different human annotated genes involved in metabolism
(cellular, macromolecule and primary metabolisms) gene
ontology, and especially in lipid metabolism, as determined
by GO tree machine (data not shown). Labeled targets were
prepared from pooled poly(A)+ RNAs prepared from the liver
of lean and fat chickens. Pooling was not performed at random
from lean and fat samples but taking into account abdominal
Fig. 4. Real-time RT-PCR in individuals. Real-time RT-PCR analysesof CYP2C
(A) and SREBP1 RNA (B) levels were performed on lean (n=20) and fat
(n=20) individuals. Results are expressed as relative RNA levels (2−ΔΔCt) (see
Fig. 3) and shown as a scatter-plot from adipose tissue weight (x) and 2−ΔΔCt
values (y). The black squaresrepresent the dataobtainedwith the fat animals and
the white squares the data from lean animals.
Fig. 5. Gene ontology analysis. Number of up- and down-regulated probes
annotated with level 5 categories under biological process in gene ontology are
indicated (Observed, black bars) and are compared to the number of genes
spotted on the array and annotated with these categories (Expected, grey bars) as
determined by GO tree machine (see Materials and methods).
7 E. Bourneuf et al. / Gene xx (2006) xxx–xxx
ARTICLE IN PRESS
adipose tissue weight in corresponding animals in order to make
accurate correlation analysis between levels of gene expression
and adipose tissue weight. Thus, 10 pools were realized and
were considered as 5 lean and 5 fat pools when differential
expression was analyzed but also as 10 different pools when
correlation to adipose tissue was determined. As reported by
Peng et al. (2003), RNA sample pooling, often conducted to
reduce scale and cost of microarray experiments, can be
statistically valid and efficient. Some indications are given here
by quantitative RT-PCR analyses that statistically significant
differences can be found, even after reduction of scale by
pooling. However, it is also observed that finding of significant
differences (when really existing) needs in some cases larger
series of samples (illustrated by G1-8B data). This further
suggests that microarray analyses should be considered as a first
screening of genes of interest before more specific or accurate
methods of investigation.
Before analyses, we normalized the expression data by print-
tip Lowess method (Dudoit et al., 2000; Yang et al., 2002) based
on ranked invariant genes (Tseng et al., 2001) and filtered the
outliers according to z-scores. After filtering and normalization,
expression data from about 220 probes (70%) were retained. We
then analyzed normalized data by SAM algorithm (Tusher et al.,
2001) and selected 22 (10%) products with differential
expression. Most of these products (corresponding to 17
different genes) were also found with a significant difference
of expression by Student T-test (p<0.05). One of the major
questions when microarray expression data are produced is to
globally address the validity of the results (Chuaqui et al., 2002
for review). Rajeevan et al. (2001) have evaluated real-time RT-
PCR to validate differentially expressed genes identified by
microarrays. Their data indicate (i) that genes with strong
hybridization signals and at least 4-fold difference are likely to
be validated by real-time RT-PCR, (ii) that validation is not
consistently achieved and is necessary for genes identified with
a 2- to 4-fold difference in expression and (iii) that genes
identified with less than a 2-fold difference in expression are not
likely to be validated. Similar results are found in our study,
although the threshold level should be lowered to 1.6.
Validation by real-time RT-PCR was performed on 9 RNAs
found up-regulated, 8 found down-regulated and 4 found non-
significant by SAM. Most of the genes identified with less than
1.6-fold difference on the arrays were not found with significant
difference (p>0.05) by RT-PCR, or found with a different
expression pattern. Altogether, these validation results indicate,
similarly to Rajeevan et al. (2001), that genes with fold changes
<1.7 by microarray analysis and/or <2 by real-time RT-PCR are
difficult to confirm and are likely to be false positives.
Finally, this study indicates that genes involved in lipo-
genesis, and especially in fatty acid biosynthesis and in
regulation of lipogenic gene expression are up-regulated in fat
animals and/or positively correlated to adipose tissue. These
data suggest that expression level of these genes could play
important roles in the regulation of adiposity in lean and fat
chickens. Indeed, we have previously shown that SCD1
expression affect lipogenic activity of chicken LMH hepatoma
cells in culture (Diot et al., 2000). Concerning SREBP1, this
suggestion is also in agreement with results from transgenic and
knock-out mice that overexpress or lack single component of
the SREBP pathway (Shimano, 2001 and Horton et al., 2002 for
reviews) and showing respectively dramatic increased or
decreased expression of genes involved in lipogenesis in
correlation with liver fatty acid synthesis.
Other genes involved in lipogenesis and spotted on the array,
ATP citrate lyase, acyl-CoA synthases and malic enzyme for
example, were not found up-regulated in this study. Interest-
ingly, all these genes were expressed with less than a 1.6-fold
difference (data not shown), under the threshold value
previously discussed. Previous studies from our laboratory
(Daval et al., 2000; Assaf et al., 2004) globally lead to the same
conclusion, all lipogenic genes studied were not constantly
validated and some discrepancies were observed in genes found
differential. The fact that most of these genes were found with
less than a 1.6- to 2-fold difference probably explain most of
FAS, SCD1, ACC and SREBP genes were not found
genetically linked to adipose tissue weight in another study
(Assaf et al., 2004), excluding the possibility that differential
expressions observed are genetically linked to polymorphism in
regulatory elements of these genes in chickens from lean and fat
lines. The exact mechanism responsible for differential
expression in these lines remains to be determined but could
be linked to the regulation of lipogenic gene expression, as
suggested by SREBP1 and HNF4 differential expression.
As expected (Carré et al., 2002), G1-8B or namely
CYP2C45 (Baader et al., 2002) expression was found inversely
correlated to adipose tissue in lean and fat chickens. Chicken
CYP2C45 and mammalian CYP2C sub-family members are
well known for xenobiotics detoxification role (Baader et al.,
2002) but also play a role in bio-transformation of important
biological regulators such as steroids and poly-unsaturated
fatty acids. As the latter are thought to play key roles in the
regulation of genes involved in lipogenesis (Jump and Clarke,
1999; Clarke, 2001), it is tempting to speculate that this
activity is linked to the differential expression observed.
However, this and the mechanisms involved remain to be
Finally the results reported here strongly suggest that
regulation of adiposity in chickens from lean and fat lines is
linked to regulation of genes involved in lipogenesis. It is
however important to take into account that many of the
probes spotted on the array are involved in metabolism and
especially in lipid metabolism. Thus, it is not very surprising to
find most of differential genes involved in these processes,
although they appear more expressed than expected as
determined by gene ontology analyses. Gain in better under-
standing of adiposity, and other processes than those investi-
gated here with our “metabolic-specific” array, will need the use
of genome-wide microarrays, including larger number of
probes, involved in more biological processes. These resources
are now available for the study of gene expression in chickens,
and will be used in order to find more genes, and possibly other
pathways involved in the fatness variability of fat and lean
8 E. Bourneuf et al. / Gene xx (2006) xxx–xxx
ARTICLE IN PRESS
This research was supported by the Institut National de la
Recherche Agronomique (INRA), the Conseil Régional de
Bretagne and by Direction Générale de l'Enseignement et de la
Recherche of the Ministry of Agriculture.
W. Carré and E. Bourneuf were doctoral fellows supported
by INRA and by the Direction Générale de l'Enseignement et
de la Recherche of the Ministry of Agriculture.
We greatly acknowledge J.J. Léger and colleagues and
particularly N. Le Meur (INSERM U533, Ouest-Genopole) for
help in data normalization and analyses using Madscan, and
IFR40/OuestGenopole for the use of transcriptome facility.
Assaf, S., et al., 2004. Genetic linkage and expression analysis of SREBP and
lipogenic genes in fat and lean chicken. Comp. Biochem. Physiol., Part B
Biochem. Mol. Biol. 137, 433–441.
Baader, M., Gnerre, C., Stegeman, J.J., Meyer, U.A., 2002. Transcriptional
activation of cytochrome P-450 CYP2C45 by drugs is mediated by the
chicken xenobiotic receptor CXR interacting with a phenobarbital-response
enhancer unit. J. Biol. Chem. 277, 15647–15653.
Burke, S., 2001. Missing values, outliers, robust statistics and non-parametric
methods. Statistics and data analysis, LC.GC Europe online Supplement.
Carré, W., et al., 2001. Development of 112 unique expressed sequence tags
from chicken liver using an arbitrarily primed reverse transcriptase-
polymerase chain reaction and single strand conformation gel purification
method. Anim. Genet. 32, 289–297.
Carré, W., Bourneuf, E., Douaire, M., Diot, C., 2002. Differential expression
and genetic variation of hepatic messenger RNAs from genetically lean and
fat chickens. Gene 299, 235–243.
Chambers, J.R., Gavora, J.S., Fortin, A., 1981. Genetic changes in meat-type
chickens in the last twenty years. Can. J. Anim. Sci. 61, 555–563.
Chomczynski,P., Sacchi, S., 1987. Single-step method of RNA isolation by acid
guanidium thiocyanate–phenol–chloroform extraction. Anal. Biochem.
Chuaqui, R.F., et al., 2002. Post-analysis follow-up and validation of microarray
experiments. Nat. Genet. 32, 509–514.
Clarke, S.D., 2001. Polyunsaturated fatty acid regulation of gene transcription: a
molecular mechanism to improve metabolic syndrome. J. Nutr. 131,
Daval, S., Lagarrigue, S., Douaire, M., 2000. Messenger RNA levels and
transcription rates of hepatic lipogenesis genes in genetically lean and fat
chickens. Genet. Sel. Evol. 32, 521–531.
Diot, C., et al., 2000. Stearoyl-CoA desaturase 1 coding sequences and antisense
RNA affect lipid secretion in transfected chicken LMH hepatoma cells.
Arch. Biochem. Biophys. 380, 243–250.
Douaire, M., Le Fur, N., el Khadir-Mounier, C., Langlois, P., Flamant, F.,
Mallard, J., 1992. Identifying genes involved in the variability of genetic
fatness in the growing chicken. Poult. Sci. 71, 1911–1920.
Dudoit, S., Yang, Y.H., Callow, M.J., Speed, T.P., 2000. Statistical methods for
identifying differentially expressed genes in replicated cDNA microarray
experiments. Technical Report 578, Department of Statistics. UC Berkeley,
Horton, J.D., Goldstein, J.L., Brown, M.S., 2002. SREBPs: activators of the
complete program of cholesterol and fatty acid synthesis in the liver. J. Clin.
Invest. 109, 1125–1131.
Huang, X., Madan, A., 1999. CAP3: A DNA sequence assembly program.
Genome Res. 9, 868–877.
Jump, D.B., Clarke, S.D., 1999. Regulation of gene expression by dietary fat.
Annu. Rev. Nutr. 19, 63–90.
Le Meur, N., et al., 2004. A dynamic, web-accessible resource to process raw
microarray scan data into consolidated gene expression values: importance
of replication. Nucleic Acids Res. 32, 5349–5358.
Leclercq, B., 1988. Genetic selection of meat-type chickens for high and low
abdominal fat content. In: Leclercq, B., Whitehead, C.C. (Eds.), Leanness in
Domestic Birds. INRA Butterworths, London, England, pp. 25–40.
Leclercq, B., Boyer, J.-P., Blum, J.-C., 1980. Selecting broilers for low or high
abdominal fat: initial observations. Br. Poult. Sci. 21, 107–113.
Leclercq, B., Chagneau, A.M., Cochard, T., Khoury, J., 1994. Comparative
responses of genetically lean and fat chickens to lysine, arginine and non-
essential amino acid supply. I. Growth and body composition. Br. Poult. Sci.
Legrand, P., Hermier, D., 1992. Hepatic delta 9 desaturation and plasma VLDL
level in genetically lean and fat chickens. Int. J. Obes. Relat. Metab. Disord.
Leveille, G.A., O'Hea, E.K., Chakrabarty, K., 1968. In vivo lipogenesis in the
domestic chicken. Proc. Soc. Exp. Biol. Med. 128, 398–401.
Liang,P., Pardee,A.B.,1992.Differential displayof eukaryoticmessengerRNA
by means of the polymerase chain reaction. Science 257, 967–971.
O'Hea, E.K., Leveille, G.A., 1969. Lipid biosynthesis and transport in the
domestic chick (Gallus domesticus). Comp. Biochem. Physiol. 30,
Peng, X., Wood, C.L., Blalock, E.M., Chen, K.C., Landfield, P.W., Stromberg,
A.J., 2003. Statistical implications of pooling RNA samples for microarray
experiments. BMC Bioinformatics 4, 26.
Rajeevan, M.S., Vernon, S.D., Taysavang, N., Unger, E.R., 2001. Validation of
array-based gene expression profiles by real-time (kinetic) RT-PCR. J. Mol.
Diagn. 3, 26–31.
Rozen, S., Skaletsky, H., 2000. Primer3 on the WWW for general users and for
biologist programmers. Methods Mol. Biol. 132, 365–386.
Saeed, A.I., et al., 2003. TM4: a free, open-source system for microarray data
management and analysis. Biotechniques 34, 374–378.
Shimano, H., 2001. Sterol regulatory element-binding proteins (SREBPs):
transcriptional regulators of lipid synthetic genes. Prog. Lipid Res. 40,
Sompayrac, L., Jane, S., Burn, T.C., Tenen, D.G., Danna, K.J., 1995.
Overcoming limitations of the mRNA differential display technique.
Nucleic Acids Res. 23, 4738–4739.
Tseng, G.C., Oh, M.K., Rohlin, L., Liao, J.C., Wong, W.H., 2001. Issues in
cDNA microarray analysis: quality filtering, channel normalization, models
of variations and assessment of gene effects. Nucleic Acids Res. 29,
Tusher, V.G., Tibshirani, R., Chu, G., 2001.Significance analysis of microarrays
applied to the ionizing radiation response. Proc. Natl. Acad. Sci. U. S. A. 98,
Yang, Y.H., et al., 2002. Normalization for cDNA microarray data: a robust
composite method addressing single and multiple slide systematic variation.
Nucleic Acids Res. 30, e15.
Zhang, B., Schmoyer, D., Kirov, S., Snoddy, J., 2004. GOTree Machine
(GOTM): a web-based platform for interpreting sets of interesting genes
using Gene Ontology hierarchies. BMC Bioinformatics 5, 16.
9E. Bourneuf et al. / Gene xx (2006) xxx–xxx
ARTICLE IN PRESS