Gene modulation by the cyclooxygenase inhibitor, sulindac sulfide, in human colorectal carcinoma cells: possible link to apoptosis.

The mechanisms underlying the anti-tumorigenic properties of cyclooxygenase inhibitors are not well understood. One novel hypothesis is alterations in gene expression. To test this hypothesis sulindac sulfide, which is used to treat familial adenomatous polyposis, was selected to detect gene modulation in human colorectal cells at physiological concentrations with microarray analysis. At micromolar concentrations, sulindac sulfide stimulated apoptosis and inhibited the growth of colorectal cancer cells on soft agar. Sulindac sulfide (10 microm) altered the expression of 65 genes in SW-480 colorectal cancer cells, which express cyclooxygenase-1 but little cyclooxygenase-2. A more detailed study of 11 genes revealed that their expression was altered in a time- and dose-dependent manner as measured by real-time RT-PCR. Northern analysis confirmed the expression of 9 of these genes, and Western analysis supported the conclusion that sulindac sulfide altered the expression of these proteins. Cyclooxygenase-deficient HCT-116 cells were more responsive to sulindac sulfide-induced gene expression than SW-480 cells. However, this response was diminished in HCT-116 cells overexpressing cyclooxygenase-1 compared with normal HCT-116 cells suggesting the presence of cyclooxygenase attenuates this response. However, prostaglandin E2, the main product of cyclooxygenase, only suppressed the sulindac sulfide-induced expression of two genes, with little known biological function while it modulated the expression of two more. The most likely explanation for this finding is the metabolism of sulindac sulfide to inactive metabolites by the peroxidase activity of cyclooxygenase. In conclusion, this is the first report showing sulindac sulfide, independent of cyclooxygenase, altered the expression of several genes possibly linked to its anti-tumorigenic and pro-apoptotic activity.

Colorectal cancer is the third leading cancer in the United States. Numerous review articles of population-based studies, animal studies, and in vitro studies with human colorectal carcinoma cells indicate that non-steroidal anti-inflammatory drugs (NSAIDs) 1 such as sulindac have anti-tumorigenic activity directed against colorectal cancer (1)(2)(3)(4). NSAIDs are known for their anti-inflammatory properties via inhibition of cyclooxygenase (Cox), which is responsible for the formation of prostaglandins (5). There are at least two distinct forms of Cox, the constitutively expressed Cox-1 and the inducible Cox-2.
Most of the studies demonstrating that Cox inhibitors have anti-cancer activity used classic NSAIDs that inhibit both Cox-1 and Cox-2. Moreover, classic Cox inhibitors such as sulindac sulfide and indomethacin are more potent inhibitors of Cox-1. However, more recent studies in animal models initiated in part due to the toxicity of non-selective Cox inhibitors suggest that specific Cox-2 inhibitors also inhibit the development of cancer. The anti-tumorigenic activity of NSAIDs may be dependent on the inhibition of Cox activity (6), but other results suggest that the anti-cancer activity of NSAIDs might be independent of Cox inhibition (7)(8)(9)(10)(11). Thus, it appears that both Cox-dependent and Cox-independent mechanisms could contribute to the anti-cancer effect of NSAIDs (12).
Of all the various NSAIDs reported to inhibit the development of tumors, sulindac appears to be very effective at inhibiting tumor development in experimental animals. Sulindac is not an inhibitor of Cox, but its metabolite sulindac sulfide is a potent non-selective inhibitor of Cox. Sulindac sulfide inhibits the growth of tumor cells in soft agar (13,14), inhibits the growth of tumors in xenograft mouse models (14,15), and is a effective stimulator of apoptosis under a number of experimental conditions (8, 16 -22). In addition, sulindac and, hence, its metabolite, sulindac sulfide, are currently used to suppress the development of adenomatous polyps in patients with familial adenomatous polyposis (23). However, understanding the mechanisms responsible for the anti-tumorigenic activity of sulindac sulfide and other NSAIDs is lacking.
This laboratory has proposed the hypothesis that the antitumorigenic activity of Cox inhibitors is mediated, in part, by altering gene expression either dependent or independent of Cox inhibition. Previously, our laboratory utilized subtractive hybridization to study changes in gene expression by NSAIDs and identified a novel member of the transforming growth factor-␤ superfamily with several names, including NAG-1 (NSAIDs-activated gene-1), PLAB, PTGFB, PDF, MIC-1, and HP00269 (24,25). Sulindac sulfide is one of the most potent NSAIDs both in terms of its ability to induce NAG-1 expression and its ability to induce apoptosis in Cox-deficient HCT-116 cells (25). Although NAG-1 has potent anti-tumorigenic activity as overexpression of NAG-1 induces apoptosis in cultured cells and suppresses xenograft growth in nude mice (25), it is likely only one of several genes involved in the pro-apoptotic and anti-tumorigenic effects of sulindac sulfide. In this report, we ask the following: Does sulindac sulfide alter the expression of other genes involved in either apoptosis, cell proliferation, and/or anti-tumorigenicity, and are these changes in gene expression dependent on Cox? We report here that sulindac sulfide increases and represses the expression of a number of genes with potential responsibility for the pro-apoptotic, antitumorigenic activity of sulindac sulfide. This is the first report showing alterations in gene expression in colorectal cancer cells at physiological concentrations of NSAIDs using microarray technology.

EXPERIMENTAL PROCEDURES
Chemicals-Chemicals were from Sigma Chemical Co. (St. Louis, MO) unless otherwise noted. Indomethacin was from Cayman Chemical Co. (Ann Arbor, MI). NSAIDs were dissolved in Me 2 SO and prepared fresh weekly. Prostaglandin E 2 was dissolved in ethanol.
Cell Lines and Reagents-Cell lines were purchased from ATCC (Rockville, MD) and were maintained at 37°C/5%CO 2 . Cell culture reagents were from Invitrogen (Rockville, MD). Human colorectal carcinoma SW-480 cells were maintained in Eagle's minimum essential medium supplemented with 15% fetal bovine serum (FBS), 10 mg/liter gentamicin, and 1 mM sodium pyruvate. Human colorectal carcinoma HCT-116 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% FBS and 10 mg/liter gentamicin.
Cell Culture Treatments-Cells were plated at 50% confluency in complete media for 24 h and treated in serum-free media containing vehicle or drug unless otherwise indicated for 0, 4, 8, or 24 h as indicated.
Cell Proliferation-Cell proliferation was measured using the MTS colorimetric assay by Promega (Madison, WI) as recommended by the manufacturer. Cells were treated with various concentrations of vehicle or sulindac sulfide in complete media containing 10% (HCT-116) or 15% (SW-480) FBS. Each experiment was carried out in quintuplet and repeated two times. Percent viability was calculated from mean OD of 490 Ϯ S.E. A representative experiment is shown.
Measurement of DNA Content and Apoptosis-The DNA contents for vehicle or sulindac sulfide-treated SW-480 and HCT-116 cells were determined by fluorescence-activated cell sorting (FACS). Cells were plated at 50% confluency in 6-well plates overnight then treated in SFM or media containing 2% FBS in the presence of various concentrations of sulindac sulfide or vehicle for various time points in triplicate repeated two or more times. Cells fixed in ethanol were stained with propidium iodine as previously reported (26), whereas living cells were stained with Annexin V and propidium iodine (Oncogene Research Products, San Diego, CA) according to the manufacturer's instructions. Both assays were analyzed by flow cytometry using a BD Biosciences FACSort sorter equipped with CellQuest TM software according to the manufacturer's instructions. Measurements are -fold increases over vehicle-treated time-matched controls.
Soft Agar Cloning Assay in the Presence of NSAIDs-Soft agar assays were performed to compare the clonogenic potential of colorectal cancer cells in semisolid medium. HCT-116 and SW-480 cells were resuspended at 6000 cells in 2 ml of warm media containing 0.35% agarose, and the final concentration of sulindac sulfide or vehicle was tested in appropriate media and plated on top of 1 ml of 0.5% agarose in six-well plates. Plates were incubated for 3 weeks at 37°C/5% CO 2 . Cell colonies were visualized following an overnight stain with 0.5 ml of p-iodonitrotetrazolium violet at room temperature then image-captured using a color charge-coupled device camera equipped with a personal computer loaded with Adobe Photoshop (San Jose, CA).
RNA Isolation-Following treatments, cells were rinsed twice with phosphate-buffered saline, then RNA was isolated using the Qiagen (Valencia, CA) RNeasy MIDI or MINI kit according to the manufacturer's instructions. Cell lysis was performed using an electric homogenizer for microarray or by centrifugation through a Qia-shredder (Qiagen) for real-time RT-PCR.
Microarray Hybridization and Analysis-A cDNA Human 12K Chip, developed in-house at the NIEHS Microarray Group, was used for gene expression profiling experiments. A complete listing of the genes on this chip is available at dir.niehs.nih.gov/microarray/chips.htm. cDNA microarray chips were prepared according to DeRisi et al. (27). The spotted cDNAs were derived from a collection of sequence-verified IMAGE clones that covered the 3Ј-end of the gene and ranged in size from 500 to 2000 bp (Research Genetics, Birmingham, AL). M13 primers were used to amplify insert cDNAs from purified plasmid DNA (pDNA) in a 100-l PCR reaction mixture. A sample of the PCR products (10 l) was separated on 2% agarose gels to ensure quality of the amplifications. The remaining PCR products were purified by ethanol precipitation, resuspended in ArrayIt buffer (Telechem, San Jose, CA), and spotted onto poly-L-lysine-coated glass slides using a modified, robotic DNA arrayer (Beecher Instruments, Bethesda, MD). Each total RNA sample (35 g) was labeled with cyanine 3 (Cy3) or cyanine 5 (Cy5)-conjugated dUTP (Amersham Biosciences, Piscataway, NJ) by a reverse transcription reaction using the reverse transcriptase, SuperScript (Invitrogen, Carlsbad, CA), and the primer, oligo(dT) (Amersham Biosciences). The fluorescent-labeled cDNAs were mixed and hybridized simultaneously to the cDNA microarray chip. Each RNA pair was hybridized to a total of four arrays employing a fluor reversal accomplished by labeling the control sample with Cy3 in two hybridizations and with Cy5 in the other two hybridizations. The cDNA chips were scanned with an Axon Scanner (Axon Instruments, Foster City, CA) using independent laser excitation of the two fluors at 532-and 635-nm wavelengths for the Cy3 and Cy5 labels, respectively. The raw pixel intensity images were analyzed using the ArraySuite v1.3 extensions of the IPLab image processing software package (Scanalytics, Inc., Fairfax, VA). This program uses methods that were developed to locate targets on the array, to measure local background for each target and subtract it from the target intensity value, and to identify differentially expressed genes using a probability-based method. We measured the pixel intensity level of "blank" spots comprised of spotting solution. The data were then filtered to provide a cut-off at the intensity level just above the blank measurement values to remove from further analyses those genes having one or more intensity values in the background range. After pixel intensity determination and background subtraction, the ratio of the intensity of the treated cells to the intensity of the control was calculated following normalization. A probability distribution was fit to the data and used to calculate a 99% confidence interval for the ratio intensity values. Genes having normalized ratio intensity values outside of this interval were considered significantly differentially expressed.
For each of the four replicate arrays for each sample, lists of differentially expressed genes at the 99% confidence levels were created and deposited into the NIEHS MAPS data base (28). For each time point, a query of the data base yielded a list of genes that were differentially expressed in at least three of the four replicate experiments. Any of these genes that indicated fluor bias or high variation were not considered for further analysis. Mean values are from all valid replicates. Assuming that the replicate hybridizations are independent, a calculation using the binomial probability distribution indicated that the probability of a single gene appearing on this list when there was no real differential expression is negligible. The entire dataset is available at dir.niehs.nih.gov/microarray/datasets.
Reverse Transcription-For real-time RT-PCR assays, the RNA was treated with 1 unit of amplification grade deoxyribonuclease I (Life Technologies) per microgram of RNA to remove genomic DNA according to the manufacturer's instructions. 2 g of the RNA was reverse-transcribed using Superscript II reverse transcriptase then treated with RNase H according to the manufacturer's instructions. RT was performed using Qiagen's Omniscript reverse transcription kit (for realtime RT-PCR) according to the manufacturer's instructions. A negative control containing all of the RT reagents in the absence of RT enzyme (no RT control) was also routinely performed.
Primer Design-Primers were designed using PrimerExpress Software (Applied Biosystems, Foster City, CA). Primers were from Invitrogen and dissolved in 10  Traditional and Real-time RT-PCR with SYBR Green Detection-Traditional and real-time RT-PCR were performed as previously described (29) using an ABI Prism 7700 (Applied Biosystems). Real-time RT-PCR fluorescence detection was performed in 96-well plates using Quantitect SYBR Green buffer (Qiagen). Each 50-l PCR reaction contained cDNA, 0.5 unit of Amp Erase uracil-N-glycosylase (UNG, PerkinElmer Life Sciences, Boston, MA), forward and reverse primers, and the Passive Reference dye (ROX) to normalize the SYBR Green/ double-stranded DNA complex signal during analysis to correct for well-to-well variations. Primer concentrations were optimized to yield the lowest concentration of primers that yielded the same C t values as recommended by Applied Biosystems. A no RT control RNA sample was used with each real-time RT-PCR experiment containing human actin primers to verify no genomic DNA contamination. Amplification parameters were UNG incubation, for one cycle at 50°C for 2 min to prevent amplification of carryover DNA; denaturation/UNG inactivation at 94°C for 10 min; amplification, 40 cycles of 95°C/15 s and 60°C/30 s. Amplification products using SYBR Green detection were routinely checked using dissociation curve software (PerkinElmer Life Sciences) and by gel electrophoresis on a 1% agarose gel then visualized under UV light following staining with 0.05% ethidium bromide to confirm the size of the DNA fragment and that only one product was formed. Samples were compared using the relative (comparative) C t method to validate microarray results. The C t value, which is inversely proportional to the initial template copy number, is the calculated cycle number where the fluorescence signal emitted is significantly above background levels. The -fold induction or repression by real-time RT-PCR was measured in triplicate relative to time-matched vehicletreated controls and calculated after adjusting for actin using 2 Ϫ⌬⌬Ct , Sequence Confirmation-One microliter of a PCR reaction generated from each primer pair was cloned into Escherichia coli using the TOPO TA cloning kit (Invitrogen) in duplicate according to the manufacturer's instructions. The cloned cDNA inserts were then sequenced by dRhodamine and purified according to the manufacturer's instructions (Applied Biosystems). Sequences were determined following gel electrophoresis by the DNA sequencing facility at NIEHS (Research Triangle Park, NC). Results were verified using a nucleotide-nucleotide BLAST search on the NCBI website.
Northern Blot Detection-Sequence-verified clones provided by the NIEHS microarray group and purchased from Research Genetics (Invitrogen) were used to generate cDNA probes for Northern blots. They were grown on LB agar plus 50 g/ml ampicillin, and individual colonies were selected and grown in LB broth at 37°C for 16 h. The pDNA was isolated using the QIAprep Miniprep kit (Qiagen) according to the manufacturer's instructions. 10 g of pDNA was digested to release the cDNA insert using the appropriate restriction enzymes (Promega) according to the gene accession number online at www.ncbi.nlm.nih.gov. Following electrophoresis, cDNA inserts were excised and column-purified using a QIAquick Spin kit (Qiagen) according to the manufacturer's instructions. Probes were labeled with [␣-32 P]dCTP using a DECAprime II kit (Ambion) according to the manufacturer's instructions.
Probes were column-purified using a G-50 column (Amersham Biosciences). 20 g of RNA from vehicle or sulindac sulfide-treated SW-480 cells was subjected to Northern blots analysis as previously described (30). The signal was detected by autoradiography using Biomax MS film and an intensifying screen for several hours as appropriate (Kodak, Rochester, NY). Each membrane was also hybridized with a glyceraldehyde-3-phosphate dehydrogenase probe (Ambion) to verify equal loading.
Tissue Collection and Sample Preparation-Surgically resected colorectal tumor samples (adenocarcinomas) and matched adjacent normal tissues were obtained from the University of North Carolina at the Chapel Hill Comprehensive Cancer Center. Tissue was obtained with the approval of the University of North Carolina and NIEHS/National Institutes of Health, and local boards governing research on human subjects as previously described by our laboratory (31).
Western Blotting-Proteins from tissue culture cells (20 g) were isolated as previously described (26), and human tissues (50 g) also isolated as previously described (31) were separated by SDS-PAGE, transferred onto nitrocellulose membranes, and stained with Ponceau S to verify equal loading. Blots were blocked overnight with 10% skim milk in TBS containing 0.1% Tween 20 (TBS-T) and probed for 1 h at room temperature in TBS-T with 5% milk containing Cox-1, Cox-2 (Cayman), ATF3, MAD2, C/EBP␤, MSX1, or actin (Santa Cruz Biotechnologies, Santa Cruz, CA). Blots were washed in TBS-T then treated with the appropriate horseradish peroxidase-conjugated secondary antibody (Santa Cruz Biotechnologies) for 1 h at room temperature in TBS-T containing 5% milk and washed several times in TBS-T. The signal was detected by enhanced chemiluminescence purchased from (Amersham Biosciences) and followed by autoradiography. Where necessary, blots were stripped of antibody before reuse while sealed in a plastic bag containing a solution of 62.5 mM Tris-HCl (pH 6.8), 2% (w/v) SDS, and 100 mM ␤-mercaptoethanol for 30 min with constant agitation in a 50°C water bath followed by washing in TBS-T.
Statistical Analyses-Real-time RT-PCR was performed in triplicate with individual time-matched vehicle-treated controls for each gene tested. Statistical significance was determined according to a one-sided t test with a 0.025 level of significance on C t values following adjustment for actin, because only gene modulation in the same direction as the microarray data were considered significant.
Densitometry Measurements-Autoradiograms from Western blots were scanned using a Umax TM Powerlook III TM scanner equipped with a transparency adapter and scanning software. Bands were quantitated using Scion Image TM beta version 4.0.2 and cut to size for publication without modification and labeled using Adobe Photoshop version 5.0.

Sulindac Sulfide Induces Apoptosis and Inhibits Clonogenic
Growth-The human colorectal cell line (32) SW-480 was selected to evaluate gene expression by the traditional NSAID, sulindac sulfide, because this cell line expresses Cox-1 and a little Cox-2. The expression of Cox-1 and Cox-2 in each of these cell lines was evaluated at the mRNA level by traditional PCR and at the protein level by Western analysis using Cox-1-and Cox-2-specific antibodies as previously reported and data not shown (30). In addition, we selected HCT-116 cells that are devoid of Cox expression, which was confirmed by high-performance liquid chromatography analysis for metabolites of arachidonic acid (30). This cell line is reported to undergoes apoptosis when exposed to sulindac sulfide (8,33,34) and other Cox inhibitors (7). Both cell lines are commonly used to study the anti-tumorigenic activity of Cox inhibitors (22,(33)(34)(35)(36). To determine if sulindac sulfide alters cell growth or was toxic, cells were treated with various concentrations of sulindac sulfide. Cell proliferation was not inhibited at concentrations at or below 100 M, nor was toxicity observed below these concentrations (data not shown). Apoptosis was then measured by FACS analysis after incubating the cells with various concen-trations of sulindac sulfide or vehicle. As shown in Fig. 1 (A and  B), sulindac sulfide induced apoptosis at 20 -40 M in SFM following a 30-h treatment. This observation was verified using Annexin-V fluorescein isothiocyanate, which measures live cells following treatment in media containing 2% serum for 30 h in HCT-116 cells but required a 48-h treatment in SW-480 cells. This is consistent with previous results where sulindac sulfide inhibited cell growth at an IC 50 of 50 M and induced apoptosis at an EC 50 of 65 M in the presence of serum in HT-29 colon cancer cells following a 48-h treatment with similar results in SW-480 cells (37). HCT-116 cells were more susceptible to sulindac sulfide-induced apoptosis at lower doses, and both cells were more susceptible in SFM.
In addition, the ability of sulindac sulfide to inhibit the growth of cells on soft agar, an assay for anti-tumorigenic activity, was performed. As shown in Fig. 2, higher concentrations of sulindac sulfide were required to inhibit the growth of HCT-116 on soft agar than were required to induce apoptosis. This finding is in agreement with previous reports in the literature and may be related to the presence of 10% serum in the incubation mixture that would bind up the sulindac sulfide thus requiring a higher concentration to elicit a response (38) similar to that seen here by FACS analysis (Fig. 1). Based upon these findings the major focus was to search for genes potentially linked to apoptosis. A physiological (10 M) concentration of sulindac sulfide was used based upon the reported plasma levels of sulindac sulfide obtained in human subjects (39). This concentration is related to clinical usage of the drug and should reduce any general toxic responses and nonspecific downstream targets.
Analysis of Gene Expression-Following treatment of SW-480 cells with 10 M sulindac sulfide, 65 genes (0.5%) were identified as significantly induced or repressed by microarray at 4, 8, and/or 24 h at the 99% confidence interval in at least three of four hybridizations (Tables I and II) several genes significant at multiple time points resulting in a total of 65 genes. The microarray data set was further inspected with regard to the greatest -fold induction/repression, if the induction or repression occurred at multiple time points, and the potential biological function particularly as related to cell growth and apoptosis. Of the 65 genes, 14 were ESTs (at the beginning of the study), whereas several others related to growth and differentiation or cell proliferation such as MAD2, and/or were transcription factors such as ATF-3, MSX1, and C/EBP␤. Additionally, several of the genes have previously been reported as regulated by NSAIDs in SW-480 colorectal cancer cells by microarray analysis such as ATF3, Stanniocalcin 2, MAD2, and C/EBP␤ albeit at much higher doses and using the precursor form of sulindac sulfide, sulindac (34). Subsequently, this gene set was empirically reduced to 16 genes (8 induced, 8 repressed) for further study based on these factors. These genes are illustrated in boldface in Table I.

Measurement of Gene Expression by Real-time RT-PCR-
Real-time RT-PCR was then used to verify changes in the expression of these 16 genes using the same RNA from the original microarray experiment. Changes in gene expression at one or more time points of that determined by microarray were verified in 15 genes (Table III). The exception was dynein, likely due to low mRNA levels evidenced by a high C t value. In general, real-time RT-PCR detected larger changes in gene expression than microarray analysis. This difference was anticipated, because microarray often underestimates changes in gene expression and is poor at measuring differences at low mRNA concentrations. For example, stanniocalcin was induced 1.6-fold by microarray analysis but 4.5-fold by real-time RT-PCR. Similarly, MAD2 was repressed 0.83-fold by microarray analysis but 0.58-fold by real-time RT-PCR. Changes in gene expression by sulindac sulfide in nine of the most avidly induced or repressed genes were selected for further study and are indicated in boldface in Table II. Interferon-related developmental regulator 1 was not selected, because it was repressed at the other time points tested.
Gene Modulation Is Time-dependent-To replicate the previous results and to determine if the induction or repression of these nine genes selected for more detailed study occurred in a time-dependent fashion according to real-time RT-PCR, SW-480 cells were incubated with vehicle or 10 M sulindac sulfide, and RNA was isolated at various time points from new experiments. Additionally, NAG-1 and NRG-1, which are induced and repressed, respectively, by NSAIDs in vitro, were used as controls for further experiments. Sulindac sulfide did not modulate NAG-1 and NRG-1 according to microarray analysis at the 99% confidence level. However, NAG-1 was not modulated by any treatments and a defective spot for NAG-1 was present on these chips that was later confirmed by the Microarray group, whereas NRG-1 was repressed 0.83-fold at 8 h, albeit not significantly. Real-time RT-PCR was used to measure the time dependence of induction or repression for each gene. Most genes were consistently induced or repressed by sulindac sulfide according to real-time RT-PCR in a time-dependent manner (Fig. 3, A and  B). The highest induction and greatest repression was, in general, observed following an 8-or 24-h treatment. However, modulation generally was significant at both 8 and 24 h, which is prior to the induction of apoptosis. Interestingly, in addition to ATF3, NAG-1, which can induce apoptosis, was consistently one of the most highly induced genes, whereas NRG-1 was one of the most repressed indicating they make convenient and reliable controls in this study. At 4 h, ATF3 and C/EBP␤, which are transcription factors, were induced 2.2-and 1.4-fold, respectively.    Northern Blot Analysis-Northern blots were performed for these genes to ascertain more information on them such as transcript size, basal levels of transcription, relative abundance following induction or repression, and to confirm the most significant time of induction or repression seen by microarray and real-time RT-PCR. With the exception of ATF3 and nucleoporin, all the induced genes were detected by Northern analysis in SW-480 cells following treatment with sulindac sulfide (Fig. 4A). Induction of ATF3 was confirmed in HCT-116 cells (data not shown). All of the repressed genes (MSX1, IN-SIG1, MAD2, and NRG-1) were detected by Northern analysis in SW-480 cells (Fig. 4B). These results confirm the findings by both microarray and real-time RT-PCR. These findings support the conclusion that the Cox inhibitor, sulindac sulfide, significantly alters gene expression in SW-480 human colorectal cells, and their expression levels are significant using physiological doses of sulindac sulfide. Other genes such as ESTs and recently discovered genes have little or no known function. In an attempt to find out more about one such EST (R34224), we cloned a 1.1-kb fragment and sequenced it from HCT-116 cells in which it was highly induced. 2 Measurement of Protein Expression-Measurement of protein expression was limited by availability of antibodies. Whole cell protein lysates were isolated from SW480 cells after a 24-h treatment with vehicle, 10, or 50 M sulindac sulfide and Western blots performed on four genes, MAD2, ATF3, MSX1, and C/EBP␤, which have antibodies commercially available. A significant increase in protein expression by treatment of the cells with sulindac sulfide was observed for ATF3 and C/EBP␤ (Fig.  5A), whereas a decrease in expression was observed for MAD2 and to a lesser extent MSX1 (Fig. 5B). Thus a good correlation was observed between the genes measured by Northern and Western analysis. Changes in protein expression for these genes are concentration-dependent, and an increase of ϳ8-fold in ATF3 and C/EBP␤ while a 60 -90% decrease in MSX1 and MAD2 expression was observed at 50 M sulindac sulfide. Sulindac sulfide also induces NAG-1 protein expression in SW-480 cells (data not shown) and in HCT-116 cells (26).
Gene Expression in HCT-116 Cells-Subsequently, another colorectal carcinoma cell line was selected to confirm that changes in gene expression by sulindac sulfide were not restricted to SW-480 cells. To determine if their modulation was dependent on Cox, we selected HCT-116 that are devoid of both Cox-1 and Cox-2 that was confirmed by Western analysis (data not shown) and high-performance liquid chromatography analysis of metabolites (30). The cells were treated with several concentrations of sulindac sulfide and RNA isolated at 8 and 24 h. The expression of these genes plus the two control genes was measured by real-time RT-PCR. In HCT-116 cells, there was a similar pattern of gene expression, which occurred in at similar time points (data not shown). The fold change was far more dramatic following treatment with sulindac sulfide than that seen in SW-480 cells at equimolar concentrations (Table  IV). For example, the EST, NAG-3 was induced 20-fold compared with a 3.4-fold induction in SW-480 cells. However, nucleoporin and INSIG1 were not modulated in HCT-116 cells. The induction by sulindac sulfide occurred in a concentrationdependent manner, and the two repressed genes, INSIG1 and MSX1, which were not modulated at 10 M, were both modulated at higher does. (Fig. 6, A and B).
In general, gene modulation in HCT-116 cells was greater than that in SW-480 cells at 10 M doses (Table III). One possible explanation for the different response is the presence of Cox-1 in the SW-480 cells. To test for this possibility HCT-116 cells that stably express Cox-1 were incubated with sulindac sulfide, and then sulindac sulfide-induced alteration in gene expression was compared directly between the normal HCT-116 and Cox-1-expressing HCT-116 cells (30). As shown in Table III, the presence of Cox-1 in the cells significantly attenuated the expression of most genes in response to sulindac sulfide according to real-time RT-PCR. RT-PCR followed by gel electrophoresis confirmed the expression of Cox-1/2 in SW-480 cells, the expression of Cox-1 in the Cox-1-overexpressing HCT-116 cells, and the absence of Cox-1/2 in normal HCT-116 cells (data not shown). Western blotting further confirmed the presence of Cox-1 in SW-480 and HCT-116-overexpressing cells (data not shown).
Other Cox Inhibitors and Gene Expression-The prodrug sulindac, the metabolite of sulindac sulfide, sulindac sulfone, and indomethacin were incubated with HCT-116 cells to deter-mine if these drugs also altered gene expression. At equal molar concentrations, sulindac and sulindac sulfone did not appear to alter the expression of these genes at comparable doses to sulindac sulfide (Table V). However, indomethacin did appear to produce a modest change in gene expression at low concentrations. Unlike sulindac sulfide, at concentration up to 20 M, sulindac and sulindac sulfone did not alter the expression of most of these genes, but indomethacin gave a concentration-dependent change in the expression of these genes as reported in Fig. 7. At the 100 M concentration of indomethacin, which is in excess of its ability to inhibit Cox, the changes in expression were essentially equivalent to the changes observed for 10 M sulindac sulfide, and thus indomethacin is ϳ10 times less potent as sulindac sulfide but has a similar IC 50 for the inhibition of Cox-1 (40).
Effect of Prostaglandins on Gene Expression-The expression of Cox-1 in HCT-116 cells attenuated sulindac sulfideinduced changes in gene expression. Two potential explanations for this result is the metabolism of sulindac sulfide to the inactive metabolites sulindac and sulindac sulfone (41,42) or that prostaglandins could have effects on gene expression. The most likely explanation is metabolism of sulindac sulfide, because their appears to be a general reduction in the response to sulindac sulfide and under the experimental protocol incubation of the cells in serum free media would be devoid of arachidonic acid and hence result in poor prostaglandin formation. However, to determine if prostaglandin does directly alter the expression of these genes, cells were incubated with 200 ng/ml PGE 2 , the major metabolite produced from arachidonic acid in Cox-1-overexpressing HCT-116 cells (30), in the presence and absence of sulindac sulfide, and expression was measured by real-time PCR. Overall, gene modulation by sulindac sulfide in the presence of PGE 2 was minimally affected, suggesting the modulation of these genes is prostaglandin-independent. Only the sulindac sulfide-induced expressions of myozenin and ATF3 were attenuated 2-fold or more, whereas the expression of NAG-3 and myozenin were increased about 2-fold by the addition of prostaglandin to the cells (Table VI). Thus the role prostaglandins have on gene expression is complex and depends on the regulation of the individual gene but is likely to be of minor significance in the global pattern of gene expression. However, this does not rule out that other metabolites of arachidonic acid may affect these genes.

Comparison of Gene Expression in Normal versus Tumor
Tissue-MSX1 and MAD2 are repressed by sulindac sulfide, whereas C/EBP␤ and ATF3 are linked to apoptosis and are highly induced by sulindac sulfide, and these are the only genes with commercially available antibodies at the time of this study. To determine if these proteins were expressed in human colorectal tissues and if the expression was altered in adjacent tumors, Western analysis was done on two sets of four pairs of human colorectal tumors and adjacent normal tissue. MSX1 and MAD2 were not detected in any of the eight samples (data not shown). Expression of C/EBP␤ was down-regulated in tumors relative to matching normal tissue from four pairs of samples but was not detected in normal or tumor tissue in four other pairs (Fig. 8). The expression of ATF3 was down-regulated in the tumors from five pairs of tissue, up-regulated in the tumors from two pairs, and very poorly expressed in one pair of tumors relative to their normal adjacent tissue. These data support the notion of lower expression of anti-tumorigenic proteins such as ATF3 and C/EBP␤ in tumors but whose expression could be subsequently increased by sulindac sulfide restoring their normal expression. DISCUSSION Sulindac sulfide, an NSAID that inhibits both Cox-1 and Cox-2 with known pro-apoptotic, anti-tumorigenic, and antiinflammatory activity, is used to treat familial adenomatous polyposis. Using physiological concentrations based on the blood level in patients treated with the drug (39), we report that sulindac sulfide modulated the expression of a number of genes that can be linked to its anti-tumorigenic and pro-apoptotic properties. This approach contrasts with other investigations concerned with alternation in gene expression by Cox inhibitors, using high, non-physiological concentrations of the drugs (34). In an attempt to find genes that may be responsible for the anti-tumorigenic and other effects of sulindac sulfide, the human colorectal carcinoma cell line SW-480 was incubated with sulindac sulfide and changes in gene expression were examined using microarray technology. Of the 12,000 genes on the chip, 65 genes were detected as being either significantly induced or repressed by the sulindac sulfide treatment. Fifteen of 16 were confirmed by real-time RT-PCR using the same RNA. Nine genes were considered suitable candidates for further investigations based on their -fold induction or repression, multiple times of induction or repression, and biological function (i.e. possible link to apoptosis, regulation of cell growth, tumor formation, and/or function as transcription factors). Changes in their expression occurred relatively early and prior to apoptosis and in a time-dependent manner according to real-time RT-PCR analysis of expression. Two additional genes (NAG-1 and NRG-1) modulated by sulindac sulfide and other NSAIDs were included in subsequent studies as controls. NAG-1 was identified in our laboratory using subtractive hybridization of indomethacin-treated Cox-deficient HCT-116 cells (25). It is induced in vitro and down-regulated in tumors in vivo (31). NAG-1 is induced by a variety of compounds (26,43,44), including sulindac sulfide (34) and has pro-apoptotic activity. NRG-1 was discovered by Z. Zhang et al. (45) using subtractive hybridization the expression of which is repressed by treatment of colorectal carcinoma cells with the Cox-2 inhibitor NS-398. NRG-1 is up-regulated in tumors in vivo, but its biological activity is not well characterized.
A second cell line, HCT-116, which lacks both Cox isoforms, was also tested to further validate these results, as a comparison, to determine if the changes seen were cell line-dependent, and to test for importance of Cox activity. The magnitude of changes in expression in HCT-116 cells was far more dramatic after sulindac sulfide treatment and occurred in a concentration-and time-dependent manner. HCT-116 cells also showed a greater degree of apoptosis at low concentrations of sulindac sulfide in serum-free media and at a somewhat higher dose in the presence of serum compared with SW-480 cells, which also required longer incubation times to exhibit a response. The greater response seen in HCT-116 cells may be the result of cell line differences or could be due to the differences in Cox expression. The response on gene regulation to sulindac sulfide was also measured in HCT-116 cells stably transfected with Cox-1, and the response was diminished as compared with the wild-type HCT-116 cells, suggesting attenuation of this response by Cox-1. One explanation for this observation is that Cox-1 or the prostaglandins that are formed as a result of its enzymatic activity may have an effect on sulindac sulfidealtered gene expression. However, the addition of PGE 2 , the major metabolite of HCT-116 cells overexpressing Cox-1, only appeared to attenuate the sulindac sulfide-induced expression of myozenin and NAG-3, two relatively unknown genes. Thus, it does not appear likely that prostaglandins significantly effect the expression of these genes even at high concentrations. Furthermore, indomethacin altered the expression of these genes but only at higher concentration than that required to inhibit Cox activity. Indomethacin was ϳ10 times less effective than sulindac sulfide in altering the expression of these genes in HCT-116 cells but is equal at inhibiting Cox. Understanding the regulation of gene expression will require a detailed investigation of each gene.
A second possible explanation for the reduced response by sulindac sulfide in the Cox-1-expressing cells is the metabolism of sulindac sulfide to sulindac by the peroxidase activity of Cox (41,42). It is well established that the peroxidase activity of Cox-1 and presumably Cox-2 converts sulindac sulfide to sulindac, which is not a Cox inhibitor, and sulindac is reported to be a weaker inducer of apoptosis in vitro (25). Sulindac and sulindac sulfone over a range of concentrations did not alter the expression of these genes, thus the peroxidase activity of Cox would convert sulindac sulfide to inactive metabolites illustrated by the lack of gene modulation by these compounds. Similarly, the pro-apoptotic and anti-tumorigenic effects of sulindac sulfide are diminished in colorectal carcinoma cells by increasing levels of Cox-2 overexpression (8,46). Thus, the increased expression of Cox-2 observed in human tumors could contribute to a diminished response of tumors to the apoptotic and anti-tumorigenic activity of sulindac sulfide by its metabolism to sulindac that is inactive. Furthermore, previous studies in our laboratory examining the regulation of NAG-1 by NSAIDs revealed that, although sulindac sulfide induces the apoptotic protein, sulindac was ineffective at altering NAG-1 expression and apoptosis (25). Thus, the peroxidase activity of Cox-1 may effectively decrease the concentration of the active metabolite, sulindac sulfide.
Although it is clear that apoptosis plays a critical role in the anti-carcinogenic effects of NSAIDs, the mechanisms remain unclear and are complicated. Evidence is mounting that gene regulation may play a part in the anti-carcinogenic effect of NSAIDs (8,10,14,19,25,47). Several of the genes discovered by microarray and verified by our methods make candidates for the anti-tumorigenic and anti-inflammatory action of sulindac sulfide in that they function as transcription factors, are modulators of growth and proliferation, and modulate apoptosis. Furthermore, changes in gene expression by sulindac sulfide occur prior to the induction of apoptosis supporting the conclusion that these genes are part of the apoptotic events rather than the result of apoptosis. Comparison of the protein expression of these genes in normal adjacent versus resected colorectal tumor tissues indicates that the expression of two important genes C/EBP␤ and ATF3 are altered during tumorigenesis, suggesting the effect of NSAIDs on these genes may  also have an affect on tumorigenesis.
C/EBP␤ belongs to a diverse family of transcriptional regulators and, like ATF3, are members of the basic leucine zipper family and are considered immediate early genes. C/EBP␤ is induced by sulindac, aspirin (48), and the antioxidants pyrrolidinedithiocarbamate and vitamin E (49), followed by an induction of apoptosis in colorectal cancer cells in vitro. C/EBP␤ binds to the promoter of p21 WAF1/CIP1 , which is a powerful cell cycle inhibitor, thereby inducing its expression (49). Sulindac and sulindac sulfide modulate the anti-tumorigenic proteins p21 WAF1/CIP1 and p34cdc2 in human colorectal HT-29 cells in a manner independent of the apoptotic response but consistent with and anti-tumorigenic role for these compounds (50). ATF3 is also a transcription factor that forms heterodimers with C/EBP␤ regulating the expression of Gadd153 and several other growth-regulating cellular promoters and even heterodimerizes with Gadd153, resulting in down-regulation of ATF3 and C/EBP␤-mediated gene regulation (51). ATF3 is induced by camptothecin and etoposide, agents known to induce apoptosis (44,52,53). Furthermore, ATF3, which is highly expressed after sulindac sulfide treatments, may play an important role in sulindac sulfide-induced apoptosis and antitumorigenic activity. Tetracycline-inducible overexpression of ATF3 suppresses cell growth and slows down cell cycle progression from G 1 to S phase (54). In addition to NSAIDs, the garlic oil constituent diallyl disulfide, which has anti-tumorigenic properties and induces NAG-1 (26), also modulated ATF3, C/EBP␤, and NRG-1 in these studies indicating that these genes may be responsive to a variety of anti-cancer compounds independent of Cox-inhibition (data not shown).
MSX-1 is a transcription factor that is repressed by sulindac sulfide. MSX1 is known to be involved in cell proliferation and apoptotic signal pathways downstream of NF-B, and induction of NF-B represses MSX1 (55). Furthermore, MSX-1 gene expression correlates with the degree of apoptosis in models for apoptosis, and reduced levels of MSX-1 gene expression are correlated with an induction in apoptosis (56). MAD2 is a mitotic checkpoint protein up-regulated between non-recurrent primary prostate cancers and highly proliferative metastatic prostate cancers (57), and embryonic cells of MAD2 knockout mice lack accurate chromosome assembly, undergo apoptosis, and are non-viable (58). In addition, we found three genes that were ESTs (nucleoporin, NAG-3, and R33154) with uncharacterized biological activity. Characterization of biological activity of these ESTs may provide additional clues to better understand how NSAIDs exert their anticancer activity. Thus, several of these genes have defined roles in the process of apoptosis or appear to be associated with the cell growth, proliferation, and anti-tumorigenicity, whereas others have no known biological activity.
In conclusion, the results presented here suggest that a multitude of genes, both induced and repressed, could be important in mediating the pro-apoptotic effect of sulindac sulfide. These findings clearly demonstrate that sulindac sulfide modulates gene expression at physiological concentrations. Gene modulation may, in part, explain the pro-apoptotic and anti-tumorigenic effects of sulindac sulfide, and possibly other NSAIDs through the modulation of these and likely other genes and their downstream targets. Furthermore, sulindac sulfideinduced gene expression appears to be independent of Cox inhibition, whereas the expression of Cox appears to attenuate the effect of sulindac sulfide on gene expression. Recently, Zhu et al. (59) examined the molecular and structural requirements for the induction of apoptosis by Cox-2-specific inhibitors in prostate cells. They concluded that the induction of apoptosis was independent from the structural requirements for Cox inhibition. It would be of interest to determine if similar structural requirements are required for the gene modulation seen in this investigation. The anti-tumorigenic activity of sulindac sulfide may arise from the concerted action of multiple mechanisms, including apoptosis, dependent or independent of Cox. Further studies are required to determine the biological significance of the genes found in this investigation and how they may influence the specific mechanisms involved in sulindac sulfide and NSAID-mediated anti-tumorigenicity.