Content uploaded by Uri Rozovski
Author content
All content in this area was uploaded by Uri Rozovski on Mar 13, 2016
Content may be subject to copyright.
Gene Expression Analysis Proposes Alternative Pathways for the
Mechanism by Which Celecoxib Selectively Inhibits the Growth
of Transformed but not Normal Enterocytes
Eyal Sagiv, Uri Rozovski, Diana Kazanov, Eliezer Liberman, and Nadir Arber
Abstract Purpose: Cyclooxygenase-2 inhibitor (celecoxib, Pfizer) is a promising chemopreventive agent,
yet its long-term use may be limited dueto increased cardiovascular toxicity.This study was aimed
to identify genes and pathways involved in colorectal tumorigenesis and affected by celecoxib.
Experimental Design: Normal rat enterocytes (IEC18 cells) and their Ra s-transformed deriva-
tives (R1) were exposed for 72 h or over 6 months to celecoxib and analyzed for gene expression
pattern using Genechip (RG-U34). Cluster and pathway analyses were done using GeneSpring
software and Gene Ontology database. Cyclin D1 was overexpressed in IEC18 cells using stable
transfection; cell cycle and prostaglandin synthesis were assessed.
Results: Five hundred thirty-eight genes were differentially expressed after transformation, and
70 and 126 genes, respectively, were affected by short and long treatments with celecoxib.
Clusters of expression showed different expression in the transformed cells that revert to normal
after treatment; they included Ras/Erk/Ral-B, Jagged2/Notch, calcineurin, lysyl-oxidase, etc.
Cyclin D1 is up-re gulate d under the Ras pathway and is down-regulated by celecoxib. Thus, we
showed that cyclin D1^ transformed cells are resistant to inhibition by celecoxib. Celecoxib was
also shown to work via cyclooxygenase-2 inhibition in transformed cells.
Conclusions: Celecoxib selectively affects transformed and not normal enterocytes by targeting
genes and pathways that are involved in the transformation. Thus, an alternative mechanism
is proposed for the cancer-preventive role of celecoxib other than the classic mechanism of
inhibiting prostaglandin synthesis, stressing mainly the role of cyclin D1. These data may help in
the development of safer and more effective preventive drugs.
Colorectal cancer, the second most prevalent cancer in the
developed world and the third most prevalent in developing
nations (1), is responsible worldwide for more than a million
new cases of cancer and half a million deaths annually (2).
Colorectal cancer develops through a stepwise process that
involves a variety of genetic and epigenetic changes that are
acquired over several years and culminate, eventually, in the
transformation of normal epithelium into neoplasm (3, 4). The
long latency period thus provides a window of opportunities
for preventive therapy, which has become a cornerstone in the
modern concept of health.
Up-regulation of cyclooxygenase-2 (COX-2) occurs in 40% to
50% of colorectal adenomatous polyps and in up to 85% of
carcinomas (5). The lack of COX-2 expression in normal
colonic mucosa, along with its increased expression in
malignant tissue, rationalizes the significant and selective
action of COX-2 inhibitors in both the primary and secondary
preventions of colorectal cancer. The association between
nonsteroidal anti-inflammatory drugs (NSAID) and colorectal
neoplasia has thus been studied extensively. Works by Arber
et al. (6) and Bertagnolli et al. (7) have recently shown the
effectiveness of the selective COX-2 inhibitor, celecoxib, in
preventing adenoma formation in the colorectal mucosa. Its
long-term use may be limited due to increased cardiovascular
system toxicity as, overall, three large-scale clinical studies
found a hazard ratio of 1.9 for cardiovascular events (8).
This study was aimed to find genes and pathways that are
unique to malignant transformed cells and are involved in the
chemopreventive action of NSAIDs, as former studies teaches
that a wide range of mechanisms are targeted by these family of
drugs. Cellular proliferation in the colonic mucosa may be
directly influenced by NSAIDs (9) by down-regulating prosta-
glandin E
2
(PGE
2
) synthesis, a mechanism that diverts the
arachidonic acid cascade into lipoxygenase metabolites by
inhibiting COX enzymes (10). Programmed cell death,
apoptosis, is another putative target for NSAIDs, as shown in
cell culture and animal studies (11 – 16). We seek to elucidate
the importance of these and novel alternative pathways for
NSAIDs in chemoprevention.
Microarray technology for gene expression profiling was
successfully used in the past to pin down genes that can
Cancer Therapy: Preclinical
Aut hors’ Affili ation: The Integrated Cancer Prevention Center,Tel Aviv Medical
Center and Tel Aviv University,Tel Aviv, Israel
Received 5/7/07; revised 7/1/07; accepted 7/20/07.
Gran t supp ort : Pfizer, Inc.
The costs of publication of this article were defrayedin part by the payment of page
charges. This article must therefore be hereby marked advertisement in accordance
with18 U.S.C. Section1734 solely to indicate this fact.
Requ es ts for re print s: NadirArber, Integrated Cancer Prevention Center, Tel Aviv
Medical Center, 6 Weizmann Street, Tel Aviv 64239, Israel. Phone: 972-3-697-
4968; Fax: 972-3-695-0339; E-mail: narber
@
post.tau.ac.il.
F2007 American A ssociation for Cancer Research.
doi:10.1158/1078-0432.CCR-07-1091
www.aacrjournals.org Clin Cancer Res 2007;13(22) November 15, 20 076807
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
be useful as targets for new treatment modalities and to
predict the response of the individual patient to chemother-
apy (17, 18). Two in vitro studies tested for the effect of non-
COX2–specific NSAIDs on colorectal cancer cell lines
detecting 130 and 140 genes that alter in sensitive cell lines
but not in a resistant cell line to sulindac and aspirin,
respectively (19, 20). In this study, we applied this method to
analyze the mechanism of action of celecoxib in a unique
in vitro model developed in our laboratory. The model
consists of normal, although immortalized, rat enterocytes
(IEC18 cells) and Ras-transformed IEC18 cells (R1; refs. 21–24);
these cells do not go through cell senescence, yet provide the
closest model for normal enterocytes grown in culture. The
IEC18 cells have a near diploid karyotype, without mutations
in ras, APC,orp53 . Moreover, they are contact inhibited; they
do not grow in soft agar; their plating efficiency is zero; and
they do not produce tumors when injected s.c. into nude mice.
The transformed cells proliferate faster, form colonies in soft
agar, and have higher saturation density and plating efficiency.
Most importantly, they form tumors when injected s.c. into
nude mice.
Cyclin D1 is already a well-established oncogene. It
functions as an important modulator of cell cycle, overex-
pressed during the G
1
(25, 26). Expression of cyclin D1
stimulates DNA synthesis, accelerates cell proliferation, and can
by itself lead to a malignant transformation of enterocytes (27).
Overexpression of cyclin D1 is an early event at the
carcinogenesis of the gut and elevated nuclear expression levels
of cyclin D1 were also found in 67% of esophageal cancers
(both squamous cell and adenocarcinomas; refs. 28, 29), 48%
of gastric tumors (28), 35% of pancreatic adenocarcinomas
(30), and 43% of small (31) and 30% of large bowel (32)
tumors, respectively.
In recent preliminary studies, we found that celecoxib
inhibited cell growth and induced apoptosis in a time- and
dose-dependent manner, whereas rofecoxib did not inhibit cell
growth at all (22). With those cell lines, we carried out gene
expression profiling [using Affymetrix rat (RG-U34) Genechip]
after short (72 h) and long (6 months) durations of treatment
with celecoxib, looking for a differential gene expression profile
in the transformed enterocytes compared with the normal cell
line, which revert to normal after exposure to the drug.
Materials and Me thods
Cell cultures. IEC18 and IEC18-ras (R1) cells were maintained in
complete medium, DMEM, and were supplemented with 5% FCS, 1%
glutamine, and 1% antibiotics (Bet-Ha’Emek). R1 cells were produced
by cotransfection with the drug resistance selectable marker tk-neo and
the plasmid pMIKcys, which encodes a mini – human c-K-ras gene
(15 kb) that contains a cysteine mutation at codon 12 (24). To evaluate
the patterns of gene expression in celecoxib-sensitive and celecoxib-
resistant enterocytes, the IEC18 and R1 derivatives were established
from parent cells: Sensitive cells are cells that were treated with
celecoxib (20 Amol/L) for 72 h. Resistant cells were prepared by
exposing the cells to gradually increasing concentrations of celecoxib,
10% increase every four passages, starting at a concentration equal to
IC
20
. After 6 months, IEC18- and R1-resistant cells could tolerate
30 and 17 Amol/L of celecoxib, respectively.
Preparation of labeled RNA. Total RNA was extracted from these
cells while being active in mitosis, in 70% confluence using Tri Reagent
(Sigma-Aldrich). The quality and amount of total RNA were analyzed
both by using an agarose gel and a spectrophotometer. The RNA was
then used as a template for double-stranded cDNA synthesis with an
oligo-(dT)24 primer containing a T7 RNA polymerase promoter site
added to the 3¶end (Genset). The cDNA was extracted with phenol/
chloroform, ethanol precipitated, and used as a template for in vitro
Fig. 1. Nonsupervised analysis using 3,242 genes
that were present in at least three of the six
samples carried out by GeneSpring software.
A, Ras-transformedcells were clustered separately
from nontransformed cells: The three left signs
at the bottom stand for all three samples
consisting of IEC18 c ells and the three right for
those consisting of R1, Ras-transformed cells.
B, samples were not clustered according to their
treatment status: First and four th signs stand for
the two samples of RNA produced from cells that
were not exposed to celecoxib, the second and
sixth signs for resistant cells that were treated with
the drug for at least 6 mo as indicated, and the
third and fifthfor short exposure to celecoxib
(72 h) as indic ated.
Cancer Therapy: Preclinical
www.aacrjournals.orgClin Cancer Res 20 07;13(22) November 15, 2007 68 08
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
transcription (Ambion T7 Megascript system) with biotin-labeled
nucleotides (Enzo Diagnostics). Labeled cRNA was fragmented and a
hybridization mix was generated as recommended (Affymetrix, Inc.).
Hybridization of microarrays. Aliquots of each sample (10 Ag cRNA
in 200 mL hybridization mix) were hybridized to a rat (RG-U34)
Genechip (Affymetrix). After hybridization, each array was washed,
stained with streptavidin phycoerythrin (Molecular Probes), washed
again, hybridized with biotin-labeled anti – streptavidin phycoerythrin
antibodies, restained with streptavidin phycoerythrin (Molecular
Probes), and scanned.
Analysis of the Genechip data. The algorithm, implanted in
Affymetrix Suite Version 5.0 (MAS5), generates signal value (which
designates a relative measure of the abundance of the transcript),
a detection Pvalue (which indicates the reliability of the transcript
detection call), and detection call (present, absent, or marginal). The
detection calls were calculated based on detection Pvalue as follows:
probe sets with P>0.06weredesignatedasabsent,0.06>P>0.04as
marginal, and P< 0.04 as present. For interarray comparisons, the
data from each array was scaled using MAS5. The mean intensity for
each array was corrected by a scaling factor to a set target intensity
of 150.
The bioinformatics analysis was carried out using GeneSpring
version7 software (Silicon Genetics).
For the normalization procedure, values below 0.01 were set to 0.01.
Each measurement was divided by the 50th percentile of all measure-
ments in that sample (per chip normalization). Each gene was divided
by the median of its measurements in all samples (per gene
normalization). For filtering, genes that had a ‘‘present’’ detection call
in at least three of six samples were chosen, with 3,242 left for further
analysis.
Annotation analysis. NetAffex database was used to extract relevant
probe sets according to annotation demands. Functional classification
in Gene Ontology of list of genes that discriminates subclasses of
samples was examined to find annotation categories that are
overrepresented compared with the representation in the array.
Annotation analysis was preformed using ‘‘David’’.
1
This is an online
database hosting tools for annotation analysis, among them the EASE
software application. EASE provides statistical methods for discovering
enriched biological themes within gene lists using this; we used Fisher’s
exact probability to choose categories that were significantly overrep-
resented (P< 0.05).
Construction and transfection of the cyclin D1 expression plas-
mid. The 1.1-kb human cyclin D1 cDNA containing the entire coding
sequence was subcloned into the expression vector pMV7 as was
previously described (27). The resulting plasmid contained a Moloney
murine leukemia virus 5¶long terminal repeat and carried neomycin as
a selectable marker and a 3¶Moloney murine leukemia virus long
terminal repeat. The vector was transfected into IEC18 cells using
LipofectAMINE (Invitrogen), and resistant cells were selected in
complete medium with 0.2 mg/mL G418 for 3 weeks. Drug-resistant
(neo+) clones were isolated (designated D1-D10). Three clones
(D1-D3) were chosen for further expansion because they expressed
high levels of cyclin D1 (27) as shown in Western blot analysis (done as
described ref. 27); clone D1 was chosen for analysis in this study.
Assays for growth inhibition. Cells were plated at a density of 7 10
6
/
10-cm dish in complete medium. The next day, the medium was
replaced with a complete medium containing celecoxib (Pfizer),
dissolved in DMSO, at the indicated concentrations for 72 h. The
adherent and nonadherent cells were collected, during exponential
growth of the cells, and counted. Then, 1 10
6
to 2 10
6
cells were
washed in PBS, and the pellet was fixed in 3 mL of ethanol for 1 h at
4jC. Cells were pelleted and resuspended in 1 mL PBS and incubated
for 30 min with 0.64 mg/mL RNase at 37jC. Cells were stained with
45 Ag/mL propidium iodide, at least 1 h before analysis, by flow
cytometry using a standard protocol for cell cycle distribution and cell
size (27, 33). All experiments were repeated thrice with similar results.
Data acquisition was done on a FACScan and analyzed using CellQuest
software (Becton Dickinson Immunocytometry Systems). Debris was
eliminated from the analysis by using a forward-angle light scatter
threshold.
Measurement of PGE
2
concentration. IEC18, R1, and D1 cell lines
were treated for 72 h with 10 and 20Amol/L of celecoxib. PGE
2
concentration in the medium, as released by the cells, was determined
by a commercially available PGE
2
-specific enzyme-linked immunoassay
(R&D Biosystems) according to the protocol of the manufacturer.
Result s
Nonsupervised analysis differentiated transformed from non-
transformed cells. Nonsupervised hierarchical clustering, using
3,242 genes that were filtered on a nonparametric basis as
described in Materials and Methods, distinguished perfectly
between the enterocytes that had transformed and those that
Table 1. Genes that were differentially expressed
after malignant transformation according to false
discovery rate correction
GenBank Name
1 M31076 Tgf-a
2 AF005720 Clcn2
4 L26268 Btg1
5 AI013107 Kif3c
6 U03491 Tgf-h3
7 AA800190 Pygb
8 U46118 Cyp3a9
10 C07012 Ppicap
12 X63281 Mycn
14 U09793 Kras2
15 L32591 Gadd45a
16 M14656 Spp1
17 L02896 Gpc1
18 AI230712 Pace4
19 U86635 Gstm5
20 L32591 Gadd45a
21 U49953 Pak1
22 AA891068 Pam
26 AA859578 Pcsk5
28 AI112173 Atp1b1
29 D10699 Uchl1
30 AA891204 Sparc
31 AF063249 Ptprq
32 AF009329 Bhlhb3
33 AF031878 Prph1
35 U88958 Nrn
37 AI103957 Cd81
38 AI103957 Cd81
39 AF036548 Rgc32
40 X84210 Nfia
41 L20900 Ica1
43 U90261 Sh3kbp1
44 X15216 Rpl21
45 D64050 Ptprr
47 D90035 Ppp3ca
48 D83538 Pik4ca
50 D14014 Ccnd1
52 M83107 Tagln
NOTE: Data are gen es with common name only, numbered
according to their original place on the list.
1
‘‘Da vid’’ database and the E ASE software are available online (http://apps1. niaid.
nih.gov/david/) from the NIH (Bethesda, MD).
Gene Expression in Celecoxib-Treated Enterocytes
www.aacrjournals.org Clin Cancer Res 2007;13(22) November 15, 20 076809
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
had not (Fig. 1). Nonsupervised analysis did not differentiate
between cells that were and were not treated with celecoxib,
presumably because the normal enterocytes are barely affected
by the drug. This result serves as one validation to the quality of
the experiment.
Supervised analysis revealed a differential gene expression
profile in IEC18 cells after Ras transformation or after treatment
with celecoxib. A two-way ANOVA test was applied separately
for each of the 3,242 genes with two independent variables:
transformation (IEC18 versus R1 cells) and status of treatment
(no treatment, short treatment of 72 h and long exposure for
6 months that produced resistant cells). There were 538 genes
with altered expression between transformed and nontrans-
formed cells. Using correction for multiple comparisons, 52
genes of this group passed false discovery rate correction with
cutoff of 0.05 (Table 1). Among these, four genes also passed
the more strict Bonferroni correction: two are identified by their
Genbank codes as AA800711 and S74257 (the latter is a rat
gene described in Fos-transformed fibroblasts) and the other
two were trangelin and rat cyclin D1. (It should be noted than
an important group of genes are those which are affected by
transformation and returns to normal after treatment. Obvi-
ously, these genes are not represented in this list and will be
presented separately later.).
Seventy and 126 genes were differentially expressed in at least
3-fold difference between treated and nontreated cells after
short and long exposures to celecoxib, respectively. For 121
genes, the interaction between treatment and transformation
status was significant (Table 2). Among these genes, cyclin D1
had also passed false discovery rate correction for multiple
comparisons with a 0.05 criteria.
Although transformation clearly affected cyclin D1 and
resulted in overexpression of this gene relative to nontrans
formed cell, this effect was blunted by exposure to celecoxib,
which resulted in a much lower increase in expression relative
to nontreated cell.
Cluster analysis detected genes that change expression after
transformation and reverts to normal after treatment with
celecoxib. K-means clustering analysis with four predetermined
clusterswasapplied.Ofthese,three(1,2,and4)are
characterized by a change in expression after transformation
(overexpression or underexpression) that revert to normal after
Table 2. Genes for which the interaction of
treatment and transformation status was
significant
Affymetrix ID Common name
3 rc_AI010357_at Pgrmc1
4 rc_AA891204_s_at Sparc
5 AJ007291_g_at SP22
6 L19699_g_at Ralb
8 AF029240_g_at RT1.S3
9 U70050_at Jag2
10 D84667_at Pik4cb
11 X85183_at Raga
12 AF061947_at Cain
13 rc_AI102103_at Pik4cb
14 AB016536_s_at Asl
15 D13120_s_at Atp5jd
16 U75932_at Prkar1a
18 U21662_at Mgat2
20 Z11995cds_at Lrpap1
24 rc_AA892649_at Gabarap
25 H32189_s_at Gstm1
26 X02610_g_at Eno1
31 AF091573_f_at Olr1504
34 U83896_at Pscd2
35 AF034218_at Hyal2
37 rc_AA894089_s_at Neurodap1
40 X75856_at Tegt
41 rc_AA799566_g_at LOC171124
42 D17614_at Ywhaq
44 AF004218_s_at Oprs1
45 rc_AI169005_at Clns1a
47 rc_AA945907_at Npr3
51 J05031_at Ivd
52 rc_AA944007_g_at Nucb
54 U57391_g_at Sh2bpsm1
55 rc_AA859975_at LOC64201
56 D37880_at Tyro3
57 rc_AA891880_g_at Loc65042
59 rc_AA956958_at Arl5
62 rc_AI012589_s_at Gstp2
65 X94185cds_s_at Dusp6
66 U53855_at Ptgis
71 M58364_at Gch
72 U34843_at D123
73 rc_AA859954_at Vmp1
74 L35767_at Vldlr
76 L23148_g_at Id1
77 J03914cds_s_at Gstm2
79 U67136_at Akap5
80 U59245_at Atp7a
81 U07181_g_at Ldhb
82 rc_AI171562_at LOC56769
83 U66471_at Cgr19
84 U52663mRNA#3_s_at Pam
86 U96130_at Gyg
87 rc_AI234060_s_at Lox
88 U16025_at RT1-M3
89 rc_AA875135_at Arl5
90 M25584_at Ins1
91 AF007583_at Colq
92 J03179_g_at Dbp
94 D26393exon_s_at Hk2
95 U38481_at Rock2
98 rc_AA875206_at Ubqln1
100 L15556_at Plcb4
101 X99477cds_at Pter
102 M61177_s_at Mapk3
103 rc_AI010083_at Prdx1
104 U41803_at Mfn2
105 Y13590_at Cast
106 X58375_at Irs1
Table 2. Genes for which the interaction of
treatment and transformation status was
significant (Cont’d)
Affymetrix ID Common name
107 M93257_s_at Comt
108 rc_AI231807_g_at Ftl1
109 X74834cds_s_at Chrng
110 rc_AI228599_at Top2a
111 D10699_at Uchl1
113 rc_AI170268_at B2m
114 rc_AA875132_at Tpm1
116 J02592_s_at Gstm2
118 rc_AA891107_at Nudt4
NOTE: Data are genes with common name only, numbered
according to their original place on the list.
Cancer Therapy: Preclinical
www.aacrjournals.orgClin Cancer Res 20 07;13(22) November 15, 2007 6 810
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
treatment with celecoxib (Fig. 2A, B, and D). These genes are
essentially what we are looking for while trying to elucidate the
preventive mechanism and downstream targets of celecoxib.
Cluster 3 describes genes without alteration in expression at all
among the six samples (Fig. 2C).
Next, we have done a functional annotation analysis using
the annotations of the Gene Ontology database for each of
these three clusters separately to detect specific biological
pathways that are involved in the mechanism of action of
the drug. In Table 3(A-C), the results for this analysis, done by
the EASE software as described, to clusters 1, 2, and 4,
respectively, are listed. Two pathways were particularly
dominant because the expression of genes that are included
in them was changed significantly in all three clusters that
differentiate the nontreated R1 cells of all others. The first is of
positive or negative regulators of nucleic acid metabolism that
results in a positive regulation of transcription in R1 cells. The
second is the carbohydrate metabolism pathway: Genes
involved in catabolism of hexose and glucose are transcribed
more in nontreated R1. Validation of these results confirmed
that the Ras pathway is hyperactive in R1 cells; it is also found
to be down-regulated by celecoxib (cluster 4). Catabolism of
hyaluronic acid is active in R1 cells and is also inhibited by
celecoxib.
Overexpression of cyclin D1 prevents the killing effect induced
by celecoxib in transformed cells. As a confirmation for our
results, we tested the effect of cyclin D1 overexpression in the
IEC18 cells under exposure to celecoxib. As cyclin D1 was
observed as the gene most affected by celecoxib, it was
predicted to be a crucial target gene for its mechanism. IEC18
cells overexpressing cyclin D1 were shown before to posses a
malignant phenotype (27) similar to that of the R1 cells.
Celecoxib (20 Amol/L) inhibits the growth of the R1 trans-
formed cells by induction of apoptosis (39.6F3.04%; Fig.
2B), whereas no effect was observed in the immortalized
IEC18 cells, or their cyclin D1–transformed derivatives (3.5F
1.45% and 3.4F0.45%,respectively).Furthermore,athigher
concentrations of celecoxib, the IEC18 cells were also affected
by the drug (9.1F3.2% at 40 Amol/L and 13.6F0.7% at
60 Amol/L), whereas the D1 cells were still barely affected
(2.9 F0.4% at 40 Amol/L and 9.4F1.3% at 60 Amol/L).
These results suggest and confirm that down-regulation of
cyclin D1 is a significant mechanism by which celecoxib
inhibits the growth of cancer cells, without affecting the
growth of normal enterocytes.
PGE
2
is oversynthesized in the transformed cells, inhibited by
celecoxib. Because COX-2 is overexpressed at cancer cells, we
tested the extent to which the classic mechanism of celecoxib,
COX-2 inhibition, acts in our cellular model. For that reason,
we determined the levels of PGE
2
secreted to the growth
medium, the end-product of the cyclooxygenases. Cells after
transformation, the R1 cells, showed f16-fold more PGE
2
than cells before transformation, IEC18 (Fig. 2D). Under
treatment with celecoxib, for 72 h, PGE
2
synthesis is reduced to
Fig. 2. K-means clustering of the 3,242 genes filtered on a nonparametric basis as described. A, 24 genes mainly down-regulated in transformed cells under no treatment.
B, 37 genes mainly up-regulated in transformed cells under no treatment. C, 35 genes that showed no change between the cells. D, 37 genes mainly up-regulate d in
transfected nontreated mice.
Gene Expression in Celecoxib-Treated Enterocytes
www.aacrjournals.org Clin Cancer Res 2007;13(22) November 15, 20 076811
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
Table 3. Pathways of biological function active within the gene clusters (P< 0.025)
Category Genes in
category
% Genes in
category
Genes in
list in category
% Genes in
list in category
P
A. Cluster 1
GO: 45892, negative regulation of transcription, DNA-dependent 15 0.298 2 9.091 0.00185
GO:45934, negative regulation of nucleobase, nucleoside,
nucleotide, and nucleic acid metabolism
23 0.457 2 9.091 0.00436
GO:16481, negative regulation of transcription 23 0.457 2 9.091 0.00436
GO:44265, cellular macromolecule catabolism 85 1.688 3 13.64 0.00567
GO:9057, macromolecule catabolism 93 1.846 3 13.64 0.00728
GO:31324, negative regulation of cellular metabolism 31 0.615 2 9.091 0.00784
GO:6026, aminoglycan catabolism 2 0.0397 1 4.545 0.00872
GO:6027, glycosaminoglycan catabolism 2 0.0397 1 4.545 0.00872
GO:30214, hyaluronan catabolism 2 0.0397 1 4.545 0.00872
GO:30212, hyaluronan metabolism 2 0.0397 1 4.545 0.00872
GO:44262, cellular carbohydrate metabolism 106 2.104 3 13.64 0.0104
GO:9892, negative regulation of metabolism 38 0.754 2 9.091 0.0116
GO:6022, aminoglycan metabolism 3 0.0596 1 4.545 0.013
GO:30203, glycosaminoglycan metabolism 3 0.0596 1 4.545 0.013
GO:6066, alcohol metabolism 119 2.363 3 13.64 0.0143
GO:5975, carbohydrate metabolism 132 2.621 3 13.64 0.0188
GO:16052, carbohydrate catabolism 50 0.993 2 9.091 0.0197
GO:44275, cellular carbohydrate catabolism 50 0.993 2 9.091 0.0197
GO:7600, sensory perception 55 1.092 2 9.091 0.0235
B. Cluster 2
GO:42135, neurotransmitter catabolism 10 0.199 2 9.524 0.00073
GO:42133, neurotransmitter metabolism 16 0.318 2 9.524 0.00192
GO:6805, xenobiotic metabolism 17 0.338 2 9.524 0.00217
GO:9410, response to xenobiotic stimulus 17 0.338 2 9.524 0.00217
GO:41, transition metal ion transport 18 0.357 2 9.524 0.00243
GO:6605, protein targeting 30 0.596 2 9.524 0.00671
GO:50877, neurophysiologic process 191 3.792 4 19.05 0.00722
GO:189, nuclear translocation of MAPK 3 0.0596 1 4.762 0.0125
GO:1505, regulation of neurotransmitter levels 43 0.854 2 9.524 0.0135
GO:9912, auditory receptor cell fate commitment 4 0.0794 1 4.762 0.0166
GO:42127, regulation of cell proliferation 48 0.953 2 9.524 0.0166
GO:48730, epidermis morphogenesis 5 0.0993 1 4.762 0.0207
GO:9913, epidermal cell differentiation 5 0.0993 1 4.762 0.0207
GO:35315, hair cell differentiation 5 0.0993 1 4.762 0.0207
GO:42491, auditory receptor cell differentiation 5 0.0993 1 4.762 0.0207
GO:42490, mechanoreceptor differentiation 5 0.0993 1 4.762 0.0207
GO:42420, dopamine catabolism 5 0.0993 1 4.762 0.0207
GO:42424, catecholamine catabolism 5 0.0993 1 4.762 0.0207
GO:42402, biogenic amine catabolism 5 0.0993 1 4.762 0.0207
GO:42219, amino acid derivative catabolism 5 0.0993 1 4.762 0.0207
GO:50954, sensory perception of mechanical stimulus 5 0.0993 1 4.762 0.0207
GO:7605, sensory perception of sound 5 0.0993 1 4.762 0.0207
GO:6825, copper ion transport 5 0.0993 1 4.762 0.0207
GO:6725, aromatic compound metabolism 55 1.092 2 9.524 0.0215
GO:7600, sensory perception 55 1.092 2 9.524 0.0215
C. Cluster 4
GO:7264, small GTPase – mediated signal transduction 80 1.588 3 21.43 0.00124
GO:9057, macromolecule catabolism 93 1.846 3 21.43 0.00191
GO:7242, intracellular signaling cascade 352 6.988 5 35.71 0.00191
GO:15980, energy derivation by oxidation of organic compounds 99 1.965 3 21.43 0.00229
GO:44262, cellular carbohydrate metabolism 106 2.104 3 21.43 0.00279
GO:5975, carbohydrate metabolism 132 2.621 3 21.43 0.00518
GO:16358, dendrite morphogenesis 2 0.0397 1 7.143 0.00555
GO:46365, monosaccharide catabolism 48 0.953 2 14.29 0.00752
GO:19320, hexose catabolism 48 0.953 2 14.29 0.00752
GO:6007, glucose catabolism 48 0.953 2 14.29 0.00752
GO:6096, glycolysis 48 0.953 2 14.29 0.00752
GO:46164, alcohol catabolism 48 0.953 2 14.29 0.00752
GO:16052, carbohydrate catabolism 50 0.993 2 14.29 0.00814
GO:44275, cellular carbohydrate catabolism 50 0.993 2 14.29 0.00814
GO:6884, regulation of cell volume 3 0.0596 1 7.143 0.00832
GO:6006, glucose metabolism 62 1.231 2 14.29 0.0123
(Continued to the following page)
Cancer Therapy: Preclinical
www.aacrjournals.orgClin Cancer Res 20 07;13(22) November 15, 2007 68 12
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
normal level (Fig. 2D). A very mild effect was seen in IEC18
cells treated with celecoxib. The inhibition was even milder for
the cells that had undergone malignant transformation due to
overexpression of cyclin D1, which showed little synthesis of
PGE
2
to begin with.
Discussion
The main significance of the current study is that it suggests
some of the underlying mechanisms that make celecoxib such
an effective drug in the prevention of colorectal cancer. It
confirms our previous observations that celecoxib selectively
inhibits the growth of transformed, but not normal, intestinal
epithelial cells.
The main proof lies within the cluster analysis presented
(Fig. 3), where at all the clusters (besides the one that consists of
genes that were equally expressed throughout the experiment)
no changes were seen after exposures of the normal immortal-
ized cells to the drug, but a reversion toward normal expression
was clearly seen while treating the Ras-transformed cells. These
results are in alignment with the recent clinical observations, in
large clinical trials, that celecoxib can prevent colorectal
neoplasia with no side effect. The bioinformatic analysis allows
only a limited interpretation of the results. This is mainly
because of the small number of samples relative to the number
of comparisons that makes a correct statistical hypothesis testing
with proper correction for multiple comparisons not entirely
applicable; the presented data are thus mainly descriptive. This
work is therefore a hypothesis-generating technique as a first
step toward a deeper investigation of the discussed chemo-
preventive agent. However, validation to the results lies within
the annotation analysis that among the differentially expressed
genes revealed pathways that are interesting for the biologist in
the context of the study framework, promising interesting
directions for further studies. These results were biologically
confirmed in other works by reverse transcription-PCR
(21–24, 33).
The power of these results is derived out of the validation of
the in vitro system of transformation that is in use in this
experiment. We showed in former publications that the IEC18
cells went through a successful malignant transformation after
transfection with a mutated k-Ras oncogene (21 –24). In the
current study, it is shown in nonsupervised analysis that
there are unique gene expression patterns that distinguish the
three samples of R1 cells from their comparative normal
parental IEC18 cells. A validation of our results was obtained by
confirming similar changes in the protein expression of tumor-
related genes such as cyclin D1, COX 2, survivin, and bak
(21–24, 33).
Using pathway analysis available online, we were able to
trace cellular functions that are affected by the transformation
and driven back to normal activity after treatment with
celecoxib and thus offer alternative pathways by which the
drug serves in the prevention of tumor formation. Genes that
were included among the three clusters mentioned above are
potentially involved in pathways that are related to tumor
formation and, hence, are targets for celecoxib. An important
pathway that is represented in all three clusters is negative
regulation of nucleic acid metabolism. This finding conforms
to a conclusion of a former publication of our group that
celecoxib reduces cell proliferation rates by inducing a G
2
-M
arrest in human colorectal cancer cell lines (33). A second
pathway that seems to be significant is the carbohydrate
metabolism, including the genes insulin 1 and insulin 2.We
showed that the transformed cells are capable of a higher
metabolism that is suppressed by celecoxib. The fact that glucose
consumption is elevated in correlation with tumor cell
density and activity is also well established, and is the basis
for the positron emission tomography scanning methodology
(34, 35).
Furthermore, single genes which activity is normalized after
treatment are evident and are suggested to be the alternative
target of celecoxib besides affecting prostaglandin synthesis.
Especially important as a drug target is cyclin D1,which
Table 3. Pathways of biological function active within the gene clusters (P< 0.025) (Cont’d)
Category Genes in
category
% Genes in
category
Genes in
list in category
% Genes in
list in category
P
GO:5996, monosaccharide metabolism 68 1.35 2 14.29 0.0147
GO:19318, hexose metabolism 68 1.35 2 14.29 0.0147
GO:6092, main pathways of carbohydrate metabolism 68 1.35 2 14.29 0.0147
GO:9056, catabolism 194 3.851 3 21.43 0.015
GO:902, cellular morphogenesis 75 1.489 2 14.29 0.0177
GO:30308, negative regulation of cell growth 7 0.139 1 7.143 0.0193
GO:45792, negative regulation of cell size 7 0.139 1 7.143 0.0193
GO:271, polysaccharide biosynthesis 7 0.139 1 7.143 0.0193
GO:9250, glucan biosynthesis 7 0.139 1 7.143 0.0193
GO:5978, glycogen biosynthesis 7 0.139 1 7.143 0.0193
GO:43284, biopolymer biosynthesis 7 0.139 1 7.143 0.0193
GO:9581, detection of external stimulus 8 0.159 1 7.143 0.022
GO:9583, detection of light stimulus 8 0.159 1 7.143 0.022
GO:7602, phototransduction 8 0.159 1 7.143 0.022
GO:9314, response to radiation 8 0.159 1 7.143 0.022
GO:9416, response to light stimulus 8 0.159 1 7.143 0.022
GO:44265, cellular macromolecule catabolism 85 1.688 2 14.29 0.0225
GO:51606, detection of stimulus 9 0.179 1 7.143 0.0248
GO:9582, detection of abiotic stimulus 9 0.179 1 7.143 0.0248
Abbreviations: GO, Gene Ontology; MAPK, mitogen-activated protein kinase.
Gene Expression in Celecoxib-Treated Enterocytes
www.aacrjournals.org Clin Cancer Res 2007;13(22) November 15, 20 076813
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
passed false discovery rate correction under strict criteria for
its expression pattern, being overexpressed after transforma-
tion and down-regulated by exposure to the drug. Thus, we
confirmed the power of our results by overexpressing it in the
same IEC18 cell–based system of transformation, and
showed that this alone is sufficient to abolish the growth
inhibition induced by celecoxib.
Cyclin D1 was shown before to be overexpressed in
cancer cells harboring a hyperactive Ras pathway, and also
to be down-regulated by the nonspecific COX inhibitor
sulindac sulfide (12, 23). Also, cyclin D1 is vastly overex-
pressed in human colorectal cancer already at an early stage
of the carcinogenesis process (32), which further ration-
alizes the use of celecoxib in vivo in the setting of colorectal
cancer. This result is confirmed by a former study
that had shown underexpression of three other cyclins in
SW480 human colorectal cancer cells after exposure to
sulindac (19).
Nonetheless, although this study is aimed to search for
alternative pathways by which celecoxib affects cancer cells, we
were also interested in confirming the level to which the classic
mechanism of inhibition of prostaglandin synthesis acts in our
model. We showed that transformed cells produce much more
PGE
2
than normal cells. Celecoxib inhibits synthesis down to
normal-level PGE
2
. This pathway was shown to be nonrelated
to those we define in the study, as cyclin D1 – transformed cells
do not express a high level of PGE
2
.
Thus, we believe that this study is powered enough to
single out other genes as targets of COX-2 inhibitors. Among
these are genes that are highly involved in oncogenic signal
transduction and were found to be overexpressed in R1, for
example, Erk1, a proproliferative kinase of the mitogen-
activated protein kinase family, which is also a downstream
effector of the Ras pathway; Ral-B expression was interestingly
suppressed by celecoxib, and also in a former study (19), by
aspirin, Jagged 2, involved in the Notch signaling pathway
for differentiation. Calcineurin binding protein 1 interacts
with and inhibits the protein phosphatase calcineurin-
mediated signal transduction that leads to transcription of
genes, among which interleukin-2 and interleukin-8 (36),
which were shown to contribute to cancer progression
under the Ras pathway (37). Other genes among these
clusters are potentially involved in cell morphology and
motility and the relation with the extracellular matrix such
as hyaluronidase 2, TYRO3 protein tyrosine kinase 3, and
a-tubulin. These data prove that a mutation in oncogene
change the protein and behavioral milieu of the cell by
cross-talking with many other pathways in which changes
are crucial. However, other genes that are not yet known
in relation to cancer are proposed by this study to have
a potential role in tumorigenesis and chemoprevention,
such as lysyl oxidase tyrosine 3-monooxygenase, very
low density-lipoprotein receptor, and more. The nuclear
factor GADD45a, related to maintenance of genomic
stability and DNA repair, was up-regulated at transforma-
tion and was also shown in a former study to be down-
regulated in SW480 colorectal cancer cells after exposure to
sulindac (19).
Fig. 3. A, Western blot analysis confirmed that the clone designated D1overexpresses cyclin D1as against a very weak expression in the normal cells, IEC18, and a
vector control, as previously described (28). B, a scheme of the differential expres sion of cyclin D1 in the microarray analysis : An elevation in expression is shown after
Ras transformation, which is significantly inhibited by short andlong treatments with celecoxib. C, IEC18 cells, and their transformed derivative overexpressing cyclin D1 (D1)
or k-Ra s (R1), were analyzed through fluorescence-activated cell sorting after staining with propidium iodide for their cell cycle properties under 72 h treatments with the
indicated concentrationsof celecoxib. D, IEC18,D1, and R1were treated with celecoxib for72 h at selected doses.PGE
2
levels inthe culture medium weremeasured byenzyme
immunoassay as described. Columns, mean of two experiments; bars, SD.
Cancer Therapy: Preclinical
www.aacrjournals.orgClin Cancer Res 20 07;13(22) November 15, 2007 68 14
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
References
1. Pisani P, Bray F, Parkin DM. Estimates of the world-
wide prevalence of cancer for twent y-five sites in the
adult population. Int J Cancer 20 02;97:72 ^ 81.
2. Ferlay J, Bray P, Pisani P, Parkin DM. GLOBOCAN
2002 : Cancer incidence, mortality and prevalence
worldwide. Lyon: IARC Press ; 2004.
3. Vogelstein B, Fearon ER, Hamilton SR, et al. Genetic
altera tions during colore ctal-tumor de velopment .
N Engl J Med 198 8;319:525 ^ 32 .
4. Bos JL, Fearon ER, Hamilton SR, et al. Prevalence of
Ras gene mut ations in human color ectal ca ncers.
Nature1987;327:293^ 7.
5. Tsujii M, DuBois RN. Alterations in cellular adhesion
and apoptosis in epithelial cells overexpressing pros-
taglandin endoperoxide synthas e 2. Cell 199 5;83:
493^ 501.
6. Arber N, Eagle CJ, Spicak J, et al. Celecoxib for the
prevention of colorectal adenomatous polyps. N Engl
J Med 200 6;355:885 ^ 95.
7. Bertagnolli MM, EagleC J, ZauberAG, et al. Celecoxib
for the prevention of sporadic colorectal adenomas.
N Engl J Med 200 6;355:873 ^ 8 4.
8. Solomon SD, Pfeffer MA, McMurray JJ, et al. Effect
of celecoxib on cardiovascular events and blood pres-
sure in two trials for the prevention of colorectal ade-
nomas. Circulation 20 06;114:1028 ^ 35.
9. Wolfe MM, Lichtenstein DR, Singh G. Gastrointestinal
toxicity o f nonste roidal antiinflammator y drugs. NEngl
J Med 1999;340:1888^ 99.
10. Eberhart CE, DuBois RN. Eicosanoids and the gastro-
intestinaltract. Gastroenterology1995;109:285 ^ 301.
11. Shiff SJ, Oiao L,Tsai LL, Rigas B. Sulindac sulfide, an
aspirin-like compound, inhibits proliferation, causes
cell cycle quies cence, and induce s apoptos is in
HT-29 colon adenoca rcinoma cells. J Clin Inves t
1995;96:491^ 503.
12 . Goldberg Y, Nassif II, Pit tas A, et al. The anti-
proliferative effect of sulindac and sulindac sulfide
on HT-29 colon cancer cells: alterations in tumor
suppressor and cell cycle-regulatory proteins.
Oncogene 1996;12:893 ^ 901.
13 . Piazza GA, Rahm AL, Krutzsch M, et al. Antineo-
plastic drugs sulindac sulfide and sulfone inhibit cell
growth by inducing apoptosis. Cancer Res 1995;55:
3110 ^ 6.
14. DugganDE,HookeKF,HwangSS.Kineticsofthe
tissue distributions of sulindac and metabolites: rele-
vance to sites and rate s of bioactivation. Drug Metab
Dispos 19 80;8:241 ^ 6.
15. HanifR,PittasA,FengY,etal.Effectsofnonsteroi-
dal anti-inflammatory drugs on p roliferation and on
induction ofapoptosis in colon cancer cells by a pros-
taglandin-independent pathway. Biochem Pharmacol
19 9 6 ; 5 2 : 2 37 ^ 4 5 .
16. Hixson LJ, Alberts DS, Krutzsch M, et al. Antiproli-
ferative effect of nonsteroidal antiinflammatory drugs
against human colon cancer cells. Cancer Epidemiol
Biomarkers Prev 1994;3:433^ 8.
17. Arango D, Wilson AJ, Shi Q, et al. Molecular mech-
anisms of action and prediction of response to oxali-
platin in colorectal cancer cells. Br J Cancer 2004;91:
19 31 ^ 4 6 .
18. Mariadason JM, Arango D, Shi Q, et al. Gene
expression profiling-based prediction of response of
colon carcinoma cells to 5-fluorouracil and campto-
thecin. Cancer Res 200 3;63:8791^ 812.
19. Iizaka M, FurukawaY,Tsunoda T, Akashi H, Ogawa
M, Nakamura Y. Expression profile analysis of colon
cancer cells in response to sulindac or aspirin. Bio-
chem Biophys Res Commun 2002;292:498^ 512.
20. Hardwick JC,Van Santen M, van den Brink GR, van
Deventer SJ, Peppelenbosch MP. DNA array analysis
of the effect s of aspirin on colon cancer cells: involve-
ment of Rac1. Carcinogenesis 2004;25:1293^ 8.
21. Harris N. C-K-ras transformed ratenterocytes (R1) are
more sensitive than normal immortalized enterocytes
(IEC18) to growth inhibition a nd apopt osis-induc ed by
sulindac sulfone and nimesulid. Gastroenterology1998;
114 : 6 12 A .
22. Kazanov D, Dvory-Sobol H, Pick M, etal. Celecoxib
but not rofecoxib inhibits the growth of transformed
cells in vitro . Clin Cancer Res 200 4;10:267 ^ 71.
23. Arber N, SutterT, Miyake M, et al. Increased expres-
sion of cyclin D1 and the Rb tumor suppres sor gene in
c-K-ras transformed rat enterocytes. Oncogene 1996;
12 : 19 0 3 ^ 8 .
24. Arber N, Han EK, Sgambato A, et al. A K-ras onco-
gene increases resistance to sulindac-induced apo-
ptosis in rat enterocytes. Gastroenterology 1997;113:
18 92 ^ 9 0 0 .
25. Won KA, Xiong Y, Beach D, Gilman MZ. Growth
regulated expression of D-type cyclin genes in human
diploid fibroblasts. Proc Natl Acad Sci U SA 1992;89:
9910 ^ 4.
26. Sewing A, Burger C, Bruss elbach S, Schalk C,
Lucibello FC, Muller R. Human cyclin D1 encodes a
labile nuclear protein who se synth esis is directly
induced by growth factor and suppressed by cyclic
AMP. J Cell Sci 19 93;104:54 5 ^ 54.
27. Ka zanov D, Shapira I, Pick M, et al. Oncogenic
transformation of normal enterocytes by overexpres-
sionofcyclinD1.DigDisSci2003;48:1251^61.
28. Arber N, Gammon MD, Hibshoosh H, et al.Overex-
pression of cyclin D1 occurs in both squamous carci-
nomas andadenocarcinomas of the esophagus andin
adenocarcinomas of the stomach. Hum Pathol 1999;
30:1087^ 92.
29. Jiang W, Zhang YJ, Kahn SM, et al. Altered expres-
sion of the cy clin D1 and retinoblastoma genes in
human esophageal cancer. Proc Natl Acad Sci U S A
19 93;90:9026 ^ 30.
30. Kornma nn M, Ishiwata T, Ar ber N, B eger HG, Korc
M. Increased cyclin D1 expression in chronic pancrea-
titis. Pancreas 1998;17:158 ^ 62.
31. Arber N, Hibshoosh H, Yasui W, et al. Abnormalities
in the expre ssion of cell cycle related proteins in
tumors of the small bowel. Cancer Epidemiol Bio-
markers Prev1999;12:1101 ^ 5.
32. Arber N, Hibshoosh H, Moss SF, et al. Increased
expres sion o f cyclin D1 is an e arly event in multistage
colorectal carcinogenesis. Gas troenterology 1996;
110:669^ 74.
33. Dvory-Sobol H,Cohen-Noyman E, Kazanov D, et al.
Celecoxib leads to G
2
/Marrest by induction of p21and
down-regulation of cyclin B1 expre ssion in a p53 -
independent manner. EurJ Cancer200 6;42:422^ 6.
34. Minn H, Joensu H, Ahonen A, Klemi P. Fluorodeox-
yglucose imaging : a method to assess the proliferative
activity of human cancer in vivo . Comparison with
DNA flow c ytom etry in head a nd neck tumours .
Cancer1988;61:1776 ^ 81.
35.Young H, Baum R, Cremerius U, et al. Measurement
of clinical and s ubclinical tumour response using
[
18
F]fluoro deoxyglucose and positron emission
tomography. Eur J Cancer1999;35:1773 ^ 82.
36. Yu Y, De Waele C, Chadee K. Calcium-dependent
interleukin-8 gene expression inT84 human colonic
epithelial cells. Inflamm Res 2001;50:220 ^ 6.
37. Sparmann A, Bar-Sagi D. Ras-induced interleukin-
8 expression plays a critical role in tumor growth and
angiogenesis. Cancer Cell 200 4;6:447^ 5 8.
Gene Expression in Celecoxib-Treated Enterocytes
www.aacrjournals.org Clin Cancer Res 2007;13(22) November 15, 20 076815
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
2007;13:6807-6815. Clin Cancer Res
Eyal Sagiv, Uri Rozovski, Diana Kazanov, et al.
the Growth of Transformed but not Normal Enterocytes
for the Mechanism by Which Celecoxib Selectively Inhibits
Gene Expression Analysis Proposes Alternative Pathways
Updated version
http://clincancerres.aacrjournals.org/content/13/22/6807
Access the most recent version of this article at:
Cited articles
http://clincancerres.aacrjournals.org/content/13/22/6807.full.html#ref-list-1
This article cites 36 articles, 10 of which you can access for free at:
Citing articles
http://clincancerres.aacrjournals.org/content/13/22/6807.full.html#related-urls
This article has been cited by 1 HighWire-hosted articles. Access the articles at:
E-mail alerts related to this article or journal.Sign up to receive free email-alerts
Subscriptions
Reprints and
.pubs@aacr.orgDepartment at
To order reprints of this article or to subscribe to the journal, contact the AACR Publications
Permissions
.permissions@aacr.orgDepartment at
To request permission to re-use all or part of this article, contact the AACR Publications
Cancer Research.
on January 14, 2016. © 2007 American Association forclincancerres.aacrjournals.org Downloaded from
