Post-transcriptional control of cyclooxygenase-2 gene expression. The role of the 3'-untranslated region.

The cyclooxygenase (COX)-2 enzyme is responsible for increased prostaglandin formation in inflammatory states and is the major target of nonsteroidal anti-inflammatory drugs. Normally COX-2 expression is tightly regulated, however, constitutive overexpression plays a key role in colon carcinogenesis. To understand the mechanisms controlling COX-2 expression, we examined the ability of the 3'-untranslated region of the COX-2 mRNA to regulate post-transcriptional events. When fused to a reporter gene, the 3'-untranslated region mediated rapid mRNA decay (t(1/2) = 30 min), which was comparable to endogenous COX-2 mRNA turnover in serum-induced fibroblasts treated with actinomycin D or dexamethasone. Deletion analysis demonstrated that a conserved 116-nucleotide AU-rich sequence element (ARE) mediated mRNA degradation. In transiently transfected cells, this region inhibited protein synthesis approximately 3-fold. However, this inhibition did not occur through changes in mRNA stability since mRNA half-life and steady-state mRNA levels were unchanged. RNA mobility shift assays demonstrated a complex of cytoplasmic proteins that bound specifically to the ARE, and UV cross-linking studies identified proteins ranging from 90 to 35 kDa. Fractionation of the cytosol showed differential association of ARE-binding proteins to polysomes and S130 fractions. We propose that these factors influence expression at a post-transcriptional step and, if dysregulated, may increase COX-2 protein as detected in colon cancer.

H synthase (4). There are two isoforms of cyclooxygenase; the type 1 (COX-1) isoform is present under resting conditions in many cells, and presumably makes prostaglandins for physiological functions. The type 2 (COX-2) is not normally present under basal conditions, or present in very low amounts, but is rapidly induced by cytokines, growth factors, and tumor promoters to result in prostaglandin synthesis associated with inflammation and carcinogenesis (5)(6)(7)(8).
To further understand the molecular mechanisms that control the expression of the COX-2 protein, we examined the ability of the ARE-containing 3Ј-UTR of the COX-2 message to mediate post-transcriptional regulation of expression. The results presented here demonstrate that the 3Ј-UTR of COX-2 can influence mRNA stability along with controlling translation efficiency. Analysis of the sequences within the 3Ј-UTR identified a conserved 116-nucleotide AU-rich element responsible for rapid mRNA turnover and translational inhibition. Using RNA gel shift assays and label transfer analysis, we identified cytoplasmic proteins with apparent molecular masses ranging from 90 to 35 kDa that bound specifically to the COX-2 ARE.

EXPERIMENTAL PROCEDURES
Cell Culture and DNA Transfections-WI38, a human lung fibroblast cell line, was maintained in Eagle's basal medium, supplemented with 10% fetal bovine serum. Cells were grown to 70 -80% confluence in normal medium, then growth-arrested by incubation in Eagle's basal medium containing 0.5% fetal bovine serum for 48 h. Subsequent serum stimulation was initiated by the addition of fetal bovine serum to 20%. COS7 and HeLa cells were passed in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum. RNA half-life experiments were initiated by adding ActD (5 g/ml) or Dex (1 M) to the growth medium.
Stable transfections of COS7 cells were accomplished by plating cells at a density of 1 ϫ 10 5 cells per 60-mm culture dish 18 -20 h prior to transfection. Five g of luciferaseϩ3ЈUTR reporter cDNAs contained in the pcDNA3.1/Zeo expression vector (Invitrogen) were transfected using LipofectAMINE (Life Technologies, Inc.) following the vendor's instructions. Stably transfected cells were selected in normal growth medium containing 400 g/ml zeocin (Invitrogen) for 2-3 weeks. Several hundred zeocin-resistant colonies were pooled and subsequently used in the indicated experiments. Transient transfections of COS7 cells with full-length COX-2 or COX-2⌬3Ј-UTR expression plasmids were carried out similarly except 1 g of expression plasmid was cotransfected with 0.5 g of the pSV-␤-Gal (Promega) plasmid using 5 l of LipofectAMINE (Life Technologies, Inc.) in 6-well tissue culture plates. The transfection medium was replaced after 6 h with normal growth medium and the cells were incubated for 24 h before analysis. Transient transfections of lacZϩ3ЈUTR reporter cDNA constructs contained in the pcDNAI/Amp expression vector (Invitrogen) were done identically except transfections were carried out in 12-well tissue culture plates and all reagents were scaled accordingly.
Plasmid Construction-The full-length human COX-2 cDNA contained in the vector pBlueScript II (KSϩ) (Stratagene) was inserted into the EcoRI site of the expression vector pcDNA3. Deletion of the 3Ј-UTR of COX-2 was accomplished by standard PCR techniques using primers to amplify the 5Ј-UTR and coding region of COX-2. The PCR amplified product (COX-2⌬3Ј-UTR) was subcloned into the vector pCRII by T/A cloning (Invitrogen) and then inserted into the BamHI and XhoI sites of pcDNA3.
␤-Gal reporter expression constructs illustrated in Fig. 4 were created as follows. The p␤-Gal⌬ 3Ј-UTR expression construct containing the lacZ cDNA without a 3Ј-UTR was created by PCR amplification of the ␤-Gal coding region from pSV-␤-Gal; the 5Ј-end of the product was flanked by a HindIII site and the 3Ј-end contained a BamHI site adjacent to the stop codon. The fragment was subsequently cloned into the HindIII and BamHI sites of pcDNAI/Amp. The 3Ј-UTR of COX-2 was amplified by PCR from full-length COX-2 cDNA contained in pBlueScript II (KSϩ) using a forward primer to place a BamHI site at the 5Ј-end of the 3Ј-UTR and the M13 reverse primer used for the 3Ј-end. The product was digested with BamHI and XhoI and subcloned into pBlueScript (KSϩ). The full-length 3Ј-UTR was inserted into the BamHI and XhoI sites of p␤-Gal⌬3Ј-UTR to create p␤-Galϩ3ЈUTR. The p␤-Gal⌬234 and p␤-Gal⌬892 constructs were made by digesting p␤-Galϩ3ЈUTR with SphI to remove 216 bp or partial digestion with HindIII to remove 874 bp, respectively. Complete plasmids containing the 3Ј-UTR deletions were purified by agarose gel electrophoresis and religated. The plasmid p␤-Galϩ3ЈUTR was digested with HindIII and ScaI to generate a 3469-bp lacZϩARE fragment that was agarose gel-purified and inserted into the HindIII and EcoRV sites of pcDNAI/Amp to create p␤-GalϩARE. The construct p␤-Gal⌬ARE was made by digesting p␤-Galϩ3ЈUTR with XhoI and partial digestion with ScaI. The remaining 3Ј-UTR fragment (1392 bp) was inserted into the filled-in BamHI site and the XhoI site of p␤-Gal⌬3Ј-UTR.
Luciferase reporter expression constructs illustrated in Fig. 2 were prepared in the vector pcDNA3.1/Zeo(ϩ) containing the luciferase cDNA from pGL3-Basic (Promega) cloned into the HindIII and XbaI sites to yield pLuc⌬3Ј-UTR. Addition of the COX-2 3Ј-UTR was accomplished by PCR amplifying the COX-2 3Ј-UTR using XbaI-tailed primers and inserting it adjacent to the luciferase coding region to yield pLucϩ3ЈUTR. The constructs pLucϩARE and pLuc⌬ARE were prepared similarly as the p␤-GalϩARE and p␤-Gal⌬ARE, respectively. All DNA constructs were analyzed by restriction mapping and DNA sequencing. Plasmid DNAs were purified by two rounds of cesium chloride density gradient centrifugation (36).
In Vitro Transcription-The plasmid pBlueScript II (KSϩ) containing the 5Ј-end region of the COX-2 cDNA (bases 11-424; (8)) was linearized with XbaI; transcription from the T3 promoter yielded a 508-nucleotide antisense riboprobe. A 383-bp region of the neomycin resistance gene from pcDNA3 was cloned into pLitmus-29 (New England Biolabs) and a 381-bp region of the luciferase cDNA from pGL3-Basic was cloned into pBlueScript II (KSϩ); transcription from the T7 promoter on linearized plasmids yielded riboprobes of 495 and 425 nucleotides in length, respectively. Linearized plasmids used to generate riboprobes for human c-myc and GAPDH were purchased from Ambion. In vitro transcription reactions incorporating [␣-32 P]UTP (50 Ci) were performed using T7 or T3 RNA polymerase (Ambion) according to the manufacturer's instructions. Template DNA was removed by incubating the reaction with 2 units of RNase-free DNase I (Ambion) for 15 min at 37°C. Purification of radiolabeled riboprobe was accomplished using Elutip-R purification columns (Schleicher & Schuell) or TE Midi Select-D G-25 spin columns (5 Prime-3 Prime, Inc.) and quantitated by scintillation counting. The specific activity was typically 10 6 -10 7 cpm/g of RNA.
mRNA Analysis-Total RNA was isolated using TRIzol reagent (Life Technologies, Inc.) and chloroform extraction according to the vendor's RNA isolation protocol for use in RNase protection assay or reverse transcriptase-PCR (RT-PCR). RNase protection assay was done according to the RPA II kit (Ambion) with the following modifications (37). Total RNA samples (5 g) were incubated at 37°C in 1ϫ NEBuffer 2 buffer (10 mM Tris-HCl (pH 7.9), 10 mM MgCl 2 , 50 mM NaCl, 1 mM dithiothreitol) (New England Biolabs) containing 2 units of RNase free-DNase I for 15 min. 32 P-Labeled antisense riboprobe was added and both the sample RNA and probe were co-precipitated in ethanol. The RNA/riboprobe pellet was dissolved in hybridization buffer and incubated overnight at 45°C. Subsequent RNase A and T1 digestions were done and the protected riboprobe fragments were separated on a 5% denaturing polyacrylamide gel. The protected riboprobes yielded fragments of 413, 383, 316, and 250 nucleotides for COX-2, neo, GAPDH, and c-myc mRNAs, respectively. Quantitation of the relative mRNA levels was accomplished by PhosphorImager analysis using the STORM 860 system (Molecular Dynamics).
RT-PCR analysis of mRNA was accomplished as follows. One g of total RNA served as template for single strand cDNA synthesis in a reaction using oligo(dT) primers and SuperScript reverse transcriptase (Life Technologies, Inc.) under conditions indicated by the manufacturer. The sequences for the following PCR primers used were: luciferase sense, 5Ј-ACGGATTACCAGGGATTTCAGTC-3Ј and luciferase antisense, 5Ј-AGGCTCCTCAGAAACAGCTCTTC-3Ј; ␤-actin sense, 5Ј-G-AAAATCTGGCACCACACCTTC-3Ј and ␤-actin antisense, 5Ј-GCTCAT-TGCCAATGGTGATGAC-3Ј; GAPDH sense 5Ј-CCACCCATGGCAAAT-TCCATGGCA-3Ј and GAPDH antisense 5Ј-TCTAGACGGCAGGTCAG-GTCCACC-3Ј. PCR of cDNA samples was performed as described previously (8) with samples amplified for 30 cycles of denaturation at 94°C for 45 s, annealing at 55°C for 1 min, and extension at 72°C for 1 min. The amplified PCR products for luciferase had a size of 367 bp. PCR reactions from COS7 cell cDNA containing ␤-actin and GAPDH primers yielded products of sizes similar to reactions using human cDNA (514 and 600 bp, respectively). PCR products were analyzed by 1.8% agarose gel electrophoresis containing ethidium bromide and quantitated digitally using 1D Image Analysis Software (Kodak).
Northern blot analysis was carried out according to Sheng et al. (38) with the following modifications. Total RNA (20 g) was extracted at the indicated time points and separated on 1.2% denaturing formaldehyde-agarose gels. The RNA was transferred to nylon membranes (Schleicher & Schuell) and fixed to the membrane by UV cross-linking. Membranes were prehybridized in NorthernMax prehybridization-hybridization buffer (Ambion) and then hybridized for 16 h with 1 ϫ 10 6 cpm/ml of a 32 P-labeled antisense riboprobe specific for luciferase at 65°C. Blots were then washed twice for 15 min at room temperature with 2 ϫ SSC, 0.1% SDS and 3 times at 65°C with 0.1 ϫ SSC, 0.1% SDS for 15 min prior to autoradiography and PhosphorImaging. 18 S rRNA signals were used as controls for RNA loading and integrity. Luciferase mRNA levels were quantitated by PhosphorImager analysis and normalized to signals for 18 S rRNA.
Protein Analysis-COS7 cells transfected with COX-2 expression constructs were lysed in 250 l of cell lysis buffer (39) and protein expression was determined by Western blot analysis after separation of 25 g of cell lysate on 10% SDS-PAGE. Protein content was determined using a BCA protein assay with bovine serum albumin as standard (Pierce). Detection of COX-2 protein was accomplished using a mouse monoclonal antibody against human COX-2 as described (39). Control ␤-Gal protein expression was detected on the same blot using a monoclonal antibody against ␤-galactosidase according to manufacturer's instructions (Promega). The blots were developed by the enhanced chemiluminescence system (ECL, Amersham Pharmacia Biotech) and exposed to Bio-Max MR film (Kodak); quantitation of the relative COX-2 and control ␤-Gal protein levels was accomplished by digital analysis using 1D Image Analysis Software (Kodak).
Cells transfected with ␤-Gal expression constructs were lysed in 150 l of 1ϫ reporter lysis buffer (Promega) and chemiluminescent detection of enzymatic activity was used to quantitate ␤-Gal protein expression. Aliquots (0.5 l) of cell extract were assayed for ␤-Gal activity using Galacto-Light (Tropix) and control luciferase activity using the Luciferase Assay System (Promega). ␤-Gal activity was normalized to luciferase activity and all results reported are the averages of three independent experiments done in triplicate.
Analysis of Protein-RNA Interactions-The p␤-Galϩ3ЈUTR plasmid was digested with BamHI and ScaI to release the 116-bp COX-2 ARE sequence, which was cloned into the BamHI and EcoRV sites of pBlue-Script (KSϩ). Complementary oligonucleotides containing the GM-CSF AU-rich element (15) were annealed and cloned into the BamHI site of pBlueScript (KSϩ). pTRI-CAT (Ambion) was used to create a control CAT RNA. In vitro transcription reactions incorporating [␣-32 P]UTP (50 Ci) were performed using T7 or T3 RNA polymerase as indicated above to yield sense RNAs for COX-2 and GM-CSF AREs of sizes 188 and 151 nucleotides, respectively; the control CAT RNA is 213 nucleotides in size. Unlabeled competitor RNAs were made using the Ribomax kit (Promega). Cytoplasmic cell lysates were prepared as indicated (22) with the following modifications. COS7 and HeLa cells were grown in p150 tissue culture dishes until confluent, and washed twice with phosphate-buffered saline before 4 ml of lysis buffer (25 mM Tris-HCl (pH 7.5), 0.5% Nonidet P-40) was added and the cells frozen at Ϫ70°C. Thawed cells were scraped from the plate, vortexed briefly (15 s), and centrifuged at 14,000 ϫ g for 10 min. The supernatant was assayed for protein concentration using a BCA protein assay with bovine serum albumin as standard and used immediately or snap frozen at Ϫ70°C. For native gel mobility shift assay, 5 g of COS7 cytoplasmic lysate was incubated with 1 ϫ 10 4 cpm of RNA in binding buffer (20 mM HEPES (pH 7.5), 3 mM MgCl 2 , 40 mM KCl, 1 mM dithiothreitol, 5% glycerol) in a total volume of 20 l. The mixture was then incubated for 15 min at room temperature, heparin was added to a final concentration of 5 mg/ml and incubation continued for 20 min at room temperature. Samples were electrophoresed in 4% polyacrylamide gels (60:1 acrylamide/ bisacrylamide) in 0.5 ϫ TBE (Tris borate-EDTA) buffer containing 5% glycerol, dried, and exposed overnight with Kodak Bio-Max MS film and an intensifying screen. For competition experiments, unlabeled amounts of the stated RNAs were added to labeled COX-2 ARE prior to the addition of 1 g of lysate. For UV cross-linking experiments, 10 g of lysate from COS7 or HeLa cells were incubated with 1 ϫ 10 5 cpm of labeled RNAs in a 40-l reaction as described above. Where indicated, cytoplasmic lysates were digested with proteinase K (50 g/ml) for 15 min at 37°C prior to the addition of RNA. Reaction mixtures were UV-irradiated in 96-well trays in a Stratalinker 2400 (Stratagene) for 5 min and then incubated with 10 g of RNase A and 5 units of RNase T1 (Ambion) for 30 min at 37°C. Laemmli buffer containing dithiothreitol was added and the samples were boiled for 3 min and electrophoresed in 10% denaturing SDS-PAGE with prestained or low-range molecular size markers (Bio-Rad). The 32 P-labeled proteins were visualized by autoradiography.
Preparation of Polysomes-Confluent cultures of COS7 cells grown in p150 tissue culture dishes were washed twice with phosphate-buffered saline at 4°C. Cells were removed from dishes by scraping and pelleted at 200 ϫ g for 10 min at 4°C. Polysomes were isolated as described previously (40) with the following modifications. Crude cytoplasmic lysates were prepared by adding 2 ml of Buffer A (10 mM Tris-HCl (pH 7.5), 1 mM KCl, 1.5 mM MgCl 2 , 2 mM dithiothreitol) containing 0.5% Nonidet P-40 per p150 dish of cells and centrifuged (12,000 ϫ g) at 4°C for 10 min. The supernatant was layered on a sucrose cushion composed of Buffer A containing 30% sucrose and centrifuged 36,000 rpm for 135 min at 4°C in a Beckman SW55Ti rotor. The postribosomal supernatant (S130 fraction) was separated from the sucrose cushion and the polysome pellets were resuspended in Buffer A. The cytoplasmic lysate, S130 fraction, and polysome fraction was assayed for protein concentration using a BCA protein assay with bovine serum albumin as standard and used immediately or snap frozen at Ϫ70°C.

RESULTS
The AU-rich 3Ј-UTR of COX-2 Influences mRNA Stability-The expression of COX-2 has been demonstrated to occur through transcriptional activation that is induced by cytokines, growth factors, and tumor promoters (5)(6)(7)(8). Yet, growing experimental evidence has demonstrated that COX-2 expression is also regulated on a post-transcriptional level (38,(41)(42)(43). To determine if the AU-rich 3Ј-UTR of COX-2 mediates post-transcriptional regulation by changing mRNA stability, we measured COX-2 mRNA half-life (t[ifrax,1/2]) values in serum-stimulated human lung fibroblasts that also were treated with actinomycin D (ActD). Total RNA was isolated at various times after ActD treatment, and the stability of COX-2 mRNA was examined by RNase protection analysis. As shown in Fig. 1A, the level of COX-2 mRNA decayed significantly during the initial 2 h after ActD treatment; however, at later time points (Ͼ120 min) COX-2 mRNA stabilization was observed. This may have resulted from the ability of ActD to interfere with degradation (20, 44 -46). Using the observed initial decay values, we determined the half-life of the COX-2 mRNA to be approximately 90 min (Fig. 1B). This rate is greater than the inducible proto-oncogene of c-myc (t1 ⁄2 ϭ 27 min), but is similar to the half-lives of the induced cytokine mRNAs of interleukin-6 (t1 ⁄2 ϭ 67 min), interleukin-8 (t1 ⁄2 ϭ 44 min), and granulocyte macrophage-colony stimulating factor (GM-CSF; t1 ⁄2 ϭ 61 min) ( Fig.  1B and data not shown). The ability of the COX-2 mRNA to be rapidly degraded in the absence of ActD treatment was demonstrated using the glucocorticoid Dex (Fig. 1). In serum-stimulated fibroblasts rapid decay of COX-2 mRNA occurred following the addition of Dex (t1 ⁄2 ϭ 39 min), whereas Dex did not influence the turnover of c-myc (t1 ⁄2 Ͼ 240 min). The ability of Dex to influence COX-2 mRNA stability was also seen in phorbol ester-stimulated HeLa cells. However, transcription was necessary for Dex-mediated mRNA turnover since the addition of ActD attenuated COX-2 mRNA degradation. 2 These findings agree with other results that show a rapid turnover of COX-2 mRNA and that glucocorticoids increase the rate of degradation (47)(48)(49).
Within the COX-2 mRNA there are multiple copies of the AU-rich sequence element, AUUUA, in the 1455-nucleotide 3Ј-UTR (8). Growth-arrested human lung fibroblasts (WI38) were induced for 4 h with 20% fetal bovine serum. ActD or Dex then was added to 5 g/ml or 1 M, respectively, and total cellular RNA was prepared at the indicated time points. A, equal amounts of RNA (5 g) were analyzed by RNase protection of 32 P-labeled riboprobes specific for COX-2, c-myc, and the GAPDH internal standard. The protected mRNA for c-myc was detected as a doublet of equal intensities. B, summary of mRNA half-life data obtained from serum-stimulated human lung fibroblasts treated with ActD and Dex. RNase protection analysis was used to analyze mRNA for COX-2 (filled symbols) and c-myc (open symbols) under conditions of ActD (circles) or Dex (triangles). The radioactive signals were quantitated using a STORM Phos-phorImager with relative amounts of each mRNA normalized to the internal standard GAPDH and initial decay curves were plotted versus time. The data presented are representative of three experiments. mediates rapid degradation, we examined the decay of chimeric luciferase cDNA constructs containing the 3Ј-UTR of COX-2. Expression constructs containing luciferase with the fulllength COX-2 3Ј-UTR, a highly conserved AU-rich region of the 3Ј-UTR (48), or with the AU-rich region deleted from the fulllength 3Ј-UTR were stably transfected in COS7 cells to examine luciferase mRNA decay. ActD was added at the given times to pooled colonies of zeocin-resistant cells and the mRNA halflife was measured by RT-PCR and Northern blot analysis. The results shown in Fig. 2B demonstrate that both the full-length 3Ј-UTR (Lucϩ3ЈUTR) and conserved AU-rich region of the 3Ј-UTR (LucϩARE) confer instability on the reporter luciferase mRNA with half-lives of 30 and 29 min, respectively. Deletion of this AU-rich region within the 3Ј-UTR (Luc⌬ARE) or complete removal of the 3Ј-UTR (Luc⌬3ЈUTR) resulted in a stable mRNA, no decay was detected. Similar results were detected by Northern analysis of ActD-treated cells with the presence of the COX-2 ARE mediating rapid luciferase mRNA decay (t1 ⁄2 ϭ 49 min) whereas no decay was seen with the control luciferase mRNA or if cells were treated for 4 h with the protein synthesis inhibitor cycloheximide (Fig. 2C). COS7 cells similarly stably transfected with an expression construct containing c-myc showed rapid turnover of c-myc mRNA (t1 ⁄2 ϭ 55 min) comparable to previously reported findings (50) (data not shown). Taken together, these findings demonstrate that the AU-rich 3Ј-UTR of COX-2 can regulate COX-2 expression by increasing the rate of degradation and that this effect is mediated by a highly conserved AU-rich element (ARE) contained at the proximal end of the 3Ј-UTR (Fig. 5).
The 3Ј-UTR of the COX-2 mRNA Inhibits Protein Transla-tion-Previous work demonstrated the ability of AU-rich elements from various cytokine mRNAs to confer translational control (17,18,51,52). To determine if the 3Ј-UTR of COX-2 also functions at a post-transcriptional level to inhibit translation, we transfected COS7 cells transiently with cDNA expression constructs encoding either full-length COX-2 or COX-2 with the 3Ј-UTR removed, and protein expression was measured. We found a 2-3-fold increase in the amount of COX-2 protein when the 3Ј-UTR was absent ( Fig. 3A; COX-2⌬3ЈUTR). Co-transfection of a vector encoding the ␤-Gal protein was used to ensure that this effect was not owing to differences in transfection efficiency. The ability of the 3Ј-UTR to decrease COX-2 protein levels under these experimental conditions resulted from inhibition of translation since, in parallel transfections, the steady-state COX-2 mRNA levels showed similar amounts of full-length COX-2 and COX-2⌬3ЈUTR mRNAs (Fig. 3B). These results suggest that the COX-2 3Ј-UTR acts to primarily attenuate protein expression through translational inhibition when abnormally high mRNA levels are present during transient transfections and that this inhibition may directly influence rapid mRNA decay occurring through a co-translational mechanism (53). Similar findings were also seen in transfections of NIH3T3, Chinese hamster ovary, and HeLa cells (data not shown), demonstrating that this regulation is conserved among different cell types.
To determine if the AU-rich region that mediates rapid degradation could also confer translational inhibition, we used 3Ј-UTR deletion constructs fused to the reporter cDNA for lacZ to examine protein and mRNA levels in transiently transfected COS7 cells (Fig. 4). In agreement with the results shown in Fig.   FIG. 2. The COX-2 3-UTR mediates rapid mRNA decay. A, various deletions of the 1455-nucleotide COX-2 3Ј-UTR (open bars) were fused to the reporter gene luciferase (shaded bars) to create expression construct containing the luciferase cDNA fused to the full-length COX-2 3Ј-UTR (Lucϩ3ЈUTR), the putative COX-2 AU-rich element (LucϩARE), the AU-rich element deleted from the full-length 3Ј-UTR (Luc⌬ARE), or luciferase without a 3Ј-UTR (Luc⌬3ЈUTR). The filled circles represent AU-rich sequences, AUUUA, contained within the 3Ј-UTR; circles adjacent to one another indicate multiple repeat elements. B, pooled colonies of COS7 cells stably transfected with the indicated expression constructs were treated for various times with ActD (5 g/ml). Total RNA (1 g) was used to prepare cDNA by reverse transcription using oligo(dT) primers followed by PCR using primers specific for luciferase and the internal controls of ␤-actin or GAPDH as indicated. PCR products were resolved by electrophoresis on a 1.8% agarose gel containing ethidium bromide. The 90-min ␤-actin product from cells expressing the LucϩARE mRNA was omitted from this experiment. C, COS7 cells stably expressing LucϩARE or Luc⌬3ЈUTR mRNAs were treated for various times with ActD (5 g/ml) or for 4 h with cycloheximide (10 g/ml). Total RNA (20 g) was resolved by denaturing formaldehyde-agarose gel electrophoresis and analyzed by Northern blotting with a 32 P-labeled luciferase antisense riboprobe. The ethidium bromide-stained 18 S ribosomal RNA is shown below each respective autoradiogram. The results presented in B are representative of three experiments; the data shown in C are representative of duplicate experiments.
3, the level of ␤-Gal protein expression was decreased approximately 2-fold when the full-length COX-2 3Ј-UTR (labeled ϩ3ЈUTR) was present compared with the ⌬3ЈUTR construct. This inhibition was detected with successive deletions of the distal 6 AUUUA elements (constructs ⌬234, ⌬892, and ϩARE), whereas internal deletion of the proximal 6 elements (construct ⌬ARE) restored translational efficiency to that seen in the absence of the 3Ј-UTR. In the same experiment, the steadystate lacZ mRNA levels from each construct were approximately the same and mRNA half-life in transiently transfected cells was not influenced by the presence of any region of the COX-2 3Ј-UTR ( Fig. 4 and data not shown). Expression of COX-2 or luciferase protein from constructs containing analogous 3Ј-UTR deletions yielded protein levels similar to the analogous ␤-Gal constructs (data not shown). We conclude that the proximal 116-nucleotide ARE of the COX-2 3Ј-UTR regulates translational efficiency (Fig. 5).
Identification of Cytoplasmic Proteins that Specifically Interact with the COX-2 AU-rich Element-Various cytoplasmic pro-teins bind specifically to the AREs of mRNAs that are controlled on a post-transcriptional level. Therefore, we sought to determine if these or other unidentified proteins recognize the ARE region of the COX-2 mRNA. To investigate this we incubated COS7 cytoplasmic lysates with in vitro transcribed, 32 Plabeled RNAs of similar lengths composed of the COX-2 ARE or a positive control AU-rich region from the GM-CSF mRNA 3Ј-UTR known to confer mRNA instability and translational inhibition (15,17). A region of the CAT mRNA of similar length was used as a negative control. Protein binding was analyzed by a shift in the electrophoretic mobility of the labeled RNA during electrophoresis. We detected discrete protein-RNA complexes with COX-2 and GM-CSF elements, but no binding to the control CAT RNA (Fig. 6A). High level binding of the COX-2 ARE by factors present in the cell lysates was observed with nearly all of the RNA being bound. This binding was saturated at approximately 1 g of cytoplasmic protein (data not shown). A protein-RNA complex of similar mobility was detected for the COX-2 element using a cytoplasmic lysate from HeLa cells (data not shown). Complex formation with the GM-CSF element was less dramatic (Fig. 7A). The two unbound RNA species seen with the GM-CSF element did not result from differences in size of the labeled RNA since only one RNA species was detected by denaturing PAGE (data not shown). This suggests the GM-CSF element is able to form different RNA conformations and that the protein binding preferred the RNA species of slower mobility. Cytoplasmic protein binding to the COX-2 ARE generated a broad range of complexes.
The specificity of the protein binding to the COX-2 ARE was assessed by adding increasing concentrations of unlabeled competitor CAT RNA or COX-2 ARE to the radiolabeled COX-2 ARE prior to the addition of the COS7 lysate (Fig. 6B). We found that complex formation was progressively inhibited by increasing amounts of unlabeled COX-2 ARE (right panel) but that at least 20-fold more of the CAT RNA was required to achieve a similar inhibition (left panel). Competition of COX-2 ARE binding was also detected using unlabeled GM-CSF element, although approximately 2-fold more competitor GM-CSF RNA was needed to inhibit binding compared with unlabeled COX-2 ARE as a competitor (data not shown).
The number of proteins involved in forming stable complexes with the COX-2 ARE was assessed by cross-linking radioactive labeled RNA to the RNA-binding proteins through UV light irradiation. Cytoplasmic lysates from COS7 or HeLa cells were incubated with radiolabeled COX-2 ARE, irradiated with UV light, and treated with RNases A and T1 to remove the exogenous unprotected RNA prior to electrophoresis on SDS-PAGE. COS7 cells were transiently transfected with the expression vector pcDNA3, or with cDNAs for full-length COX-2 (COX-2) or COX-2 with the 3Ј-UTR deleted (COX-2⌬3ЈUTR). Transfection efficiency was controlled by either co-transfection of the pSV-␤-Gal expression vector encoding the ␤-Gal protein or expression of the neomycin resistance mRNA (neo) contained on the pcDNA3 vector. COX-2 protein expression (A) was determined by SDS-PAGE and Western blot analysis of 25 g of cell lysate using a monoclonal antibody specific for human COX-2 and ECL detection. ␤-Galactosidase protein expression was detected on the same blot using a monoclonal antibody specific for ␤-Gal. Steady-state COX-2 mRNA levels (B) were detected by RNase protection as described in the legend to Fig. 1. The detection of control lacZ mRNA in this experiment shows similar results (not shown). The mobility of undigested riboprobes is shown in lane 1 and the substitution of yeast RNA demonstrates total digestion of unprotected riboprobes (lane 2). The yield of expressed mRNA in transiently transfected COS7 cells is approximately 10-fold greater than in stably transfected cells such as in Fig. 2 (data not shown).  Fig. 2. The designation for each deletion construct defines the region of 3Ј-UTR removed as indicated in the illustrated construct. The ␤-Gal reporter constructs were transiently transfected in COS7 cells along with vectors expressing luciferase enzyme (pGL2-Control) or CAT mRNA (pcDNAI/Amp-CAT) to control for transfection efficiency. ␤-Gal protein expression was determined by assaying for ␤-Gal enzymatic activity and values were normalized for luciferase enzymatic activity. Steady-state lacZ mRNA levels were normalized to CAT mRNA expression and detected by RNase protection assay. All percentages listed are based on expression of ␤-Gal protein or lacZ mRNA from the construct containing no 3Ј-UTR (⌬3ЈUTR). These results are the averages of three independent experiments done in triplicate; the standard deviation within each set was less than 5% except for the ⌬ARE construct that was 9.5%. As shown in Fig. 7A, several distinct proteins or protein complexes with molecular masses ranging from 90 to 35 kDa crosslinked to the COX-2 ARE. No label was transferred to protein when UV light irradiation was omitted or when the lysate was pretreated with proteinase K. No proteins containing transferred label were detected using the control CAT RNA (data not shown). The similar sizes of bound proteins seen with cytoplasmic extracts from COS7 and HeLa cells suggest that these ARE-binding proteins are conserved between these cell types.
Previous findings have demonstrated proto-oncogene and cytokine mRNA decay is influenced by cytosolic factors that associate with polysomes or remain unassociated (40,54). To further characterize the factors identified to bind the COX-2 ARE from cytoplasmic lysates, we performed UV light crosslinking of radiolabeled ARE to proteins contained in polysomes and S130 postribosomal supernatants from COS7 cells. As shown in Fig. 7B, polypeptides with masses ranging from 90 to 35 kDa contained within the crude cytoplasmic lysate were detected to bind the COX-2 ARE. However, after centrifugation though a sucrose pad to isolate polysomes from the S130 postribosomal supernatant fraction, differential localization of ARE-binding proteins was observed. The protein doublet of 90/88 kDa was localized to the S130 fraction and the 35-kDa polypeptide was primarily polysome associated. The factors with masses of 66 and 64 kDa appeared to primarily bind only in the crude lysate with approximately 30% detected in the S130 and Ͻ10% in the polysome fractions. Nearly identical results were detected with the GM-CSF ARE, thus we conclude that several cytoplasmic proteins that bound to the COX-2 ARE are likely to be the same as those described in Jurkat cells and human peripheral blood leukocytes that bind to the GM-CSF element and consensus AU-rich RNA sequences (22,24,40). DISCUSSION The molecular mechanisms controlling the expression of COX-2 are not completely understood but is important because this is a crucial component of inflammation and a rate-limiting step in colon carcinogenesis. In this report we demonstrate that a major regulatory point of COX-2 gene expression occurs at the post-transcriptional level. This control is mediated by the 3Ј-UTR of the COX-2 mRNA through a conserved AU-rich sequence element contained within the 3Ј-UTR; the context of these AUUUA motifs within the 3Ј-UTR strongly implicates the involvement of this region in rapid decay of mRNA (16). We showed that cytoplasmic proteins specifically bind to the COX-2 ARE and we postulate that this interaction regulates the stability of COX-2 mRNA. Thus, the levels of COX-2 protein are determined by post-transcriptional regulation as well as by transcriptional mechanisms. This level of complexity is consistent with the requirement for tight control of the enzymatic action of COX-2, which has pathogenic effects if its expression is unregulated.
Many observations demonstrate that proto-oncogene and cytokine mRNAs are rapidly degraded and that this is mediated by AU-rich elements. Here we show that the inducible COX-2 gene encodes a transcript that can be degraded, and that the ARE-containing 3Ј-UTR of COX-2 mediates this destabilization. It is unclear why the serum-induced COX-2 message is not completely degraded in the presence of the transcriptional inhibitor ActD. One possibility is that the transcriptional inhibitor ActD may have inhibited a component of the rapid degradation pathway, which has been described (20,46). The effects of ActD may work through a mechanism to influence mRNA stability indirectly by allowing the accumulation of sequencespecific RNA-binding proteins in the cytoplasm. These proteins conceivably could bind the COX-2 ARE and protect it from degradation (20, 44 -46) as demonstrated with the ARE-binding protein HuR (33,34). Along these same lines, alterations in ARE binding by ActD may directly inhibit mRNA translation since ARE-dependent mRNA turnover is proposed to be coupled to translation (53,(55)(56)(57). We also demonstrate the antiinflammatory glucocorticoid dexamethasone causes rapid degradation of the COX-2 mRNA (Fig. 1). This effect may simply reside in the ability of Dex to inhibit serum-induced COX-2 transcription (58), thereby allowing degradation to ensue. Others have found that glucocorticoids regulate COX-2 expression FIG. 5. The COX-2 AU-rich element. The representation of the COX-2 mRNA is not to scale and is described in the legend to Fig. 2. The 116-nucleotide sequence of the COX-2 ARE in uppercase letters contains six AU-rich sequence motifs (AUUUA). The COX-2 termination codon is shown in lowercase letters.
FIG. 6. Detection of protein complexes that specifically bind to the COX-2 AU-rich element. A, 32 P radiolabeled in vitro transcribed RNAs containing control CAT RNA, the COX-2 ARE, or GM-CSF ARE were incubated with cytoplasmic lysate (5 g) from COS7 cells at room temperature prior to addition of heparin and electrophoresis on native low-ionic strength polyacrylamide gels. Detection of protein-RNA complexes bound to the COX-2 ARE are shown by the bracket with the arrow indicating the mobility of the major discrete species. Unbound RNAs are shown on the left; the resulting two bands seen with the GM-CSF element is presumably due to different GM-CSF RNA conformations. B, the specificity of COX-2 ARE binding was determined by adding increasing amounts of unlabeled nonspecific competitor CAT RNA or specific COX-2 ARE RNA to 32 P radiolabeled COX-2 ARE prior to the addition of 1 g of COS7 lysate. The arrow on the right indicates the position of the main band-shifted complex. The results presented are representative of experiments done in triplicate. on a post-transcriptional level. The effects elicited by Dex presumably occur through the induction or repression of factors required for COX-2 mRNA turnover (48,49). However, this appears to be independent of the COX-2 ARE since expression of reporter genes containing the COX-2 3Ј-UTR and ARE binding activity are not altered by the presence of Dex. 2 The ability of AU-rich sequences in interferon-␤, GM-CSF, and tumor necrosis factor mRNAs to confer translational control similar to the COX-2 ARE has been shown in Xenopus oocytes, rabbit reticulocyte lysate extracts, and transfected RAW 264.7 cells (17,18,51,52). In these studies the mRNA levels of the respective genes were virtually unaffected, demonstrating that the AU-rich sequences act solely as translation inhibitory elements in these experimental systems. Although we cannot specifically account for the observed differences in mRNA decay detected in stably and transiently transfected cells, one possible explanation may reside in an inability of the cells to degrade abnormally high levels of 3Ј-UTR containing mRNA when transiently transfected because the ARE-binding proteins are present in limiting amounts. Findings illustrating this effect include sequestration of the ARE-binding protein AUF1 by the heat shock protein hsp70, which results in stabilization of ARE-containing mRNAs (59). These studies also demonstrate the physical interaction of ARE-binding proteins with translation initiation factors. Thus, sequestration of AREbinding factors bound to the eIF4G and poly(A)-binding proteins may be responsible for the observed inhibition of translation initiation seen here and this inhibition may directly influence ARE-dependent mRNA decay that is coupled to ongoing translation (53,55,60). With regard to this, polysome profile studies have demonstrated the ability of AREs to block mRNA translation by inhibiting polysome association (19,56).
Several groups have detected polypeptides that specifically interact with AREs from rapidly degraded mRNAs. These regulatory trans-acting factors include several cytoplasmic mRNA-binding proteins proposed to be involved with the destabilization (27,(61)(62)(63)(64), stabilization (21,22,29), or mRNA processing and nucleocytoplasmic transport (23,24,65). While a number of RNA-binding proteins that recognize AU-rich elements have been reported, the mechanism by which these factors mediate mRNA degradation or translational inhibition is unknown. Insight into the molecular events involved in posttranscriptional control has been facilitated through the identification of these proteins. Several ARE-binding proteins have been identified and show a wide variety of activities ranging from pre-mRNA processing, developmental control, and metabolic catalysis (24 -29, 31). More importantly, the AUBF, HuR, TTP, and AUF1 ARE-binding proteins have been shown to directly effect ARE-mediated mRNA decay (32-34, 40, 59). Investigation into the role these proteins play in post-transcriptional control of COX-2 is currently in progress.
A number of observations suggest that genetic alterations of AU-rich sequences play a role in neoplastic transformation of cells. When AU-rich elements are removed from the protooncogenes c-fos and c-myc there is a correlation with increased oncogenicity (66,67) and mast cells show enhanced tumorgenicity when transfected interleukin-3 lacking the normal AREcontaining 3Ј-UTR is overexpressed (68). These observations are consistent with our finding of increased levels of COX-2 protein in cells transfected with a 3Ј-UTR deletion construct (Fig. 3). Additionally, a variety of human tumor cells show enhanced mRNA stability of ARE-containing cytokine genes (69) and a reporter gene containing the 3Ј-UTR of GM-CSF is stable in mouse monocytic tumor cells (70). This suggests that AU-rich sequences may not function properly in tumor cells because of alterations in ARE-binding regulatory proteins. These findings, taken together, suggest that post-transcriptional regulation mediated by AU-rich sequences is vital for maintaining normal cellular growth and the removal or defective recognition of these elements results in enhanced tumorgenicity.
The molecular events leading to the overexpression of the COX-2 protein in colon cancer are not totally understood. Kutchera et al. (71) demonstrated constitutive activation of the COX-2 promoter in colon cancer cell lines, suggesting that the increased levels of COX-2 mRNA seen in colorectal adenomas, adenocarcinomas (71,72), and colon cancer cell lines (71) occurs at the transcriptional level. Yet, it is interesting to note that concomitant expression of COX-2 protein was not detected in all adenomas or adenocarcinomas (73,74) and increased levels of COX-2 protein was not seen in all colon cancer cell lines shown to overexpress COX-2 mRNA (75). 2 This apparent discrepancy between COX-2 mRNA and protein expression appears to be limited to the earlier stages of adenoma-carcinoma FIG. 7. Identification of cytoplasmic proteins interacting with the COX-2 AU-rich element. A, cytoplasmic lysates from COS7 or HeLa cells were incubated with radiolabeled RNAs as shown in Fig. 6 and described under "Experimental Procedures." The bound protein was cross-linked to 32 P-labeled RNA by UV light irradiation and then treated with RNases A and T1 to remove the exogenous unbound RNA. The cross-linked products were resolved by electrophoresis on SDS-PAGE. Distinct proteins or protein complexes from COS7 or HeLa cells are detected to bind to the COX-2 ARE. No radiolabeled proteins were seen when UV light irradiation was omitted or when the lysate was pretreated with proteinase K. B, distribution of ARE-binding proteins from crude cytoplasmic lysate, S130, and polysomes isolated from COS7 cells and cross-linked to COX-2 and GM-CSF AREs. Molecular weight protein standards (kDa) are listed on the right. Electrophoresis of the gel shown in B, was carried out approximately 1.5 times longer than A. The results presented in A are representative of five experiments; the data shown in B are representative of duplicate experiments. development (73) and COX-2 protein expression is enhanced with increasing size of small intestinal and colonic polyps in mice (76). These findings, with the results presented here, suggest that loss of post-transcriptional regulation of COX-2 may be a crucial step in colon carcinogenesis and complements the genetic evidence to showing that induction of COX-2 protein is a rate-limiting step for adenoma formation (3,76). These events could result from a lack of COX-2 3Ј-UTR recognition by a trans-acting regulatory factor and not a deletion or modification of the COX-2 ARE since the COX-2 3Ј-UTR was normal in a number of colon cancer tissues and cell lines (77). We propose that this loss of regulation occurs through mutation of the proteins that specifically interact with the COX-2 ARE. This results in unregulated expression of COX-2 protein and presumably other ARE-containing early-response genes that are detected in the later stages of adenoma development.