A biologically important human gene, cyclooxygenase-2 ( COX-2 ), has been proposed to be regulated at many levels. While COX-1 is constitutively expressed in cells, COX-2 is inducible and is upregulated in response to many signals. Since increased transcriptional activity accounts for only part of the upregulation of COX-2 , we chose to explore other RNA processing mechanisms in the regulation of this gene. We performed a comprehensive bioinformatics survey, the first of its kind known for human COX-2, which revealed that the human COX-2 gene has alternative polyadenylation (proximal and distal sites) and suggested that use of the alternative polyadenylation signals has tissue specificity. We experimentally established this in HepG2 and HT29 cells. We used an in vivo polyadenylation assay to examine the relative strength of the COX-2 proximal and distal polyadenylation signals, and have shown that the proximal polyadenylation signal is much weaker than the distal one. The efficiency of utilization of many suboptimal mammalian polyadenylation signals is affected by sequence elements located upstream of the AAUAAA, known as upstream efficiency elements (USEs). Here, we used in vivo polyadenylation assays in multiple cell lines to demonstrate that the COX-2 proximal polyadenylation signal contains USEs, mutation of the USEs substantially decreased usage of the proximal signal, and that USE spacing relative to the polyadenylation signal was significant. In addition, mutation of the COX-2 proximal polyadenylation signal to a more optimal sequence enhanced polyadenylation efficiency 3.5-fold. Our data suggest for the first time that alternative polyadenylation of COX-2 is an important post-transcriptional regulatory event.
Cyclooxygenases (COX) are the key and rate-limiting enzymes in the production of prostaglandins [reviewed in ( 1 ) and references therein]. The first steps in prostanoid synthesis are the release of arachidonic acid from membrane phosopholipids by phospholipases and conversion to prostaglandin H 2 by COX. Prostaglandins play a role in many biological processes, including but not limited to inflammation, bone formation, wound healing and pain perception. Inflammatory cells as well as other types of cells, including fibroblasts and epithelial cells produce prostaglandins ( 2 , 3 ).
Two separate COX genes have been identified, COX-1 and COX-2 ( 4 – 10 ). A spliced variant of COX-1 (COX-3) has also been identified ( 11 ). The proteins that these genes encode are ∼60% identical at the amino acid level. In contrast, the 3′-untranslated regions (3′-UTRs) of COX-1 and COX-2 are highly divergent. The most striking difference between these genes is in their regulation of expression; COX-1 is constitutively expressed while COX-2 is strongly induced in response to activation by hormones, pro-inflammatory cytokines, growth factors, oncogenes, carcinogens and tumor promoters ( 1 – 3 , 12 – 15 ).
The physiological or pathological outcomes of COX-2 activity depend upon its level of expression. COX-2 overexpression is associated with a number of conditions, including cancer, rheumatoid arthritis, seizures and inflammatory disorders [reviewed in ( 1 , 16 – 22 )]. In addition, COX-2 upregulation contributes to pain; indeed, inhibition of COX-2 enzymatic activities is responsible for the anti-inflammatory properties of aspirin, indomethacin, ibuprofen and related NSAIDs, such as Vioxx (Rofecoxib) and Celebrex (Celecoxib). Molecular events leading to overexpression of COX-2 have not been definitively characterized. Some studies have clearly demonstrated increased levels of COX-2 mRNA in colorectal adenomas, colon cancer cell lines, adenocarcinomas, gastric cancer, breast cancer, certain ovarian and prostate cancers, and non-small lung cancer ( 22 – 33 ). Enhanced COX-2 mRNA transcription may play a role but enhanced COX-2 protein expression most probably requires post-transcriptional gene regulation events, such as are mediated through mRNA processing and mRNA turnover. Indeed, the kinetics of transcriptional activation alone cannot account for the sustained induction of COX-2 mRNA by interleukin 1 in vascular endothelial cells ( 14 ).
The 3′ end of nearly every fully processed eukaryotic mRNA has a poly(A) tail, which has been suggested to influence mRNA stability, translation and transport (reviewed in ( 34 – 37 )]. Polyadenylation is a two-step process [reviewed in ( 38 – 41 ) and references therein], first involving specific endonucleolytic cleavage ( 42 ) at a site determined by binding of polyadenylation factors. The second step involves polymerization of the adenosine (A) tail to an average length of ∼200 residues. Most mammalian polyadenylation signals contain the consensus sequence AAUAAA or a close variant between 10 and 35 nt upstream of the actual cleavage and polyadenylation site. AAUAAA is associated with a frequency of ∼53% of all human polyadenylation signals and ∼59% of all mouse polyadenylation signals ( 43 ). This hexamer sequence serves as a binding site for the basal polyadenylation factor cleavage and polyadenylation specificity factor (CPSF) ( 38 – 41 ). Sequences ∼14–35 nt downstream of the polyadenylation signal are also known to be involved in directing polyadenylation by serving as a binding site for the basal polyadenylation factor cleavage stimulation factor (CstF) [( 38 – 41 , 44 – 52 ) and references therein]. Elements upstream of the AAUAAA sequences, known as upstream efficiency elements (USEs), have also been characterized that can enhance polyadenylation efficiency, and have been identified in viral and cellular systems [( 53 – 66 ) and references therein]. Spacing between the AAUAAA and the USE significantly influences USE efficiency in enhancing polyadenylation ( 65 ). Although USEs are polyadenylation efficiency elements, they may also provide additional functions in proper processing.
As has been appreciated in recent years, 3′ end formation is interconnected to mRNA processing events, as well as to mRNA transcription and transcription termination ( 40 , 67 – 69 ). This interconnection and execution of a functional mRNA likely results from recognition and utilization of cis - and trans -acting signals.
The 3′-UTR of an mRNA can have a major influence on developmental and tissue-specific regulation of gene expression. In fact, the 3′-UTR has recently been called ‘a molecular hotspot for pathology’ ( 70 ). Regulation of gene expression through the 3′-UTR can include alternative polyadenylation, translational control and differential mRNA stability. Preliminary evidence suggests that regulated polyadenylation and mRNA stability may play a pivotal role in COX-2 expression.
The COX-2 gene is made up of 10 exons; the 3′-UTR is contained within exon 10 ( 71 ). The COX-2 3′-UTR is larger than average, encompassing ∼2.5 kb, and has many interesting features. It has several polyadenylation signals, only two of which are commonly used, resulting in mRNAs of ∼2.8 kb using the proximal (with regard to the stop codon) polyadenylation signal or ∼4.6 kb using the distal polyadenylation signal ( Figure 1 ). The proximal polyadenylation signal has a non-consensus CPSF binding site (AUUAAA) yet it is used. It is likely that regulation occurs here, resulting in two mRNAs with different RNA metabolism. Curiously, the mRNA that uses the proximal polyadenylation signal contains putative USEs which are similar in sequence and location to those we and others described previously ( 64 – 66 ). The 3′-UTR also has 22 repeats of an AU-rich motif resembling those known to be involved in regulation of mRNA stability ( 72 , 73 ). Recent studies have found that HuR, or a close variant, binds to these AU-rich elements (AREs) ( 74 – 76 ). Other studies have shown additional RNA binding proteins also may bind to the COX-2 3′-UTR ( 77 – 79 ). Additionally, it has been suggested, but not proven, that tissue specificity might play a role in COX-2 polyadenylation site choice ( 80 , 81 ). The significance of the different tissue distributions of these isoforms has not yet been addressed, and most probably will provide significant revelations with regard to COX-2 expression control.
Alternative polyadenylation of the COX-2 3′-UTR results in two mRNAs differing only in their 3′ ends. Therefore, the choice of a particular polyadenylation site likely involves specific cis - and trans -acting factors, resulting in RNAs with different metabolism, and ultimately will influence the cytoplasmic fate of the mRNA. This study examines the regulation of alternative 3′ end formation in the human COX-2 pre-mRNA. We demonstrate that alternative polyadenylation does occur in the COX-2 mRNA and that the two major, utilized polyadenylation signals (proximal and distal) are different in strength and composition. We also show that alternative polyadenylation of the COX-2 mRNA occurs in a tissue-specific fashion in cells that endogenously overexpress COX-2. The proximal polyadenylation signal contains auxiliary USEs that are important for its efficient usage. These experiments suggest that alternative polyadenylation of COX-2 mRNA is an important level of gene expression regulation, because polyadenylation signal choice will include or exclude additional regulatory elements that may influence the metabolic fate of the mRNA.
MATERIALS AND METHODS
Primers for PCR to amplify the proximal and distal polyadenylation signals of the human COX-2 3′-UTR were prepared by the Molecular Resource Facility of UMDNJ-NJMS, and had the sequences as listed in Supplementary Table 2. The COX-2 proximal and distal primers contained BamHI (forward) and PstI (reverse) restriction sites to allow insertion into appropriately digested vectors. Human genomic DNA was used as a template to amplify by PCR the appropriate COX-2 proximal and distal polyadenylation signals and flanking sequences for insertion into pGem4 and pCβS vectors. The pCβS vector (a gift from David Fritz, UMDNJ) ( 64 ) has a multiple cloning site downstream of the CMV promoter and upstream of the bovine growth hormone (BGH) polyadenylation signal. This vector also includes intron 1 of the rabbit β-globin gene accompanied by the splice donor and acceptor sites. Thus, pCβS-proximal and pCβS-distal were created.
The USE mutants, the hexamer mutant and the non-specific mutant were made by PCR-based site-directed mutagenesis using the Stratagene QuikChange kit as per the manufacturer's protocols using the pCβS-proximal vector as the template. The USE and non-specific mutant were designed to create BglII restriction sites for ease in screening. The COX-2 proximal 5′ deletion (Δ) mutant was made by PCR using primers described in Supplementary Table 2. This construct is lacking 115 bases upstream of the COX-2 proximal polyadenylation signal.
The USE mutant 1 has the sequence AA GA T C AAA instead of the wild-type sequence AATTTGAA; the USE mutant 2 has the sequence AA GA TCTAA instead of the wild-type sequence AATTTCTAA; the USE mutant 3 has the sequence GA GA TCTTA instead of the wild-type GATTTCTTA. The USE mutants 1,3 and 2,3 had both USEs mutated as described above.
The pCβS-proximal–distal-BGH construct was made in the following manner. The COX-2 distal polyadenylation signal was removed from pCβS-distal by BamHI and PstI, then the ends were blunted and inserted into pCβS-proximal which had been digested with EcoRV. Therefore, pCβS-proximal–distal-BGH has three polyadenylation signals in tandem. The pCβS-proximal–distal construct was made by digesting pCβS-proximal–distal-BGH with XhoI and KpnI to remove the BGH polyadenylation signal. Therefore, pCβS-proximal–distal has both COX-2 polyadenylation signals in tandem.
All constructs were transformed into Escherichia coli XL1-Blue cells. Positive clones were sequenced and assayed for expression of the appropriately sized insert. Constructs were verified by sequencing (Molecular Resource Facility, UMDNJ-NJMS). SVL RNA was used as described previously ( 66 ).
Rapid site-directed mutagenesis using two PCR-generated DNA fragments
The USE double (USE mut 1,2) and triple (USE mut 1,2,3) mutations were generated by standard PCR using KOD polymerase and using the upstream (5′) wild-type COX-2 proximal primer containing a BamHI restriction site and a downstream primer (3′; USE 1,2 R) containing the respective site-directed mutagenic base pair changes. Then the upstream mutagenic primer (5′; USE 1,2 F) was used in standard PCR using KOD polymerase and the wild-type COX-2 proximal downstream primer (3′) containing a PstI restriction site. The USE 1,2 mut was prepared using COX-2 proximal wild-type DNA template in the PCR, while the USE 1,2,3 mut was prepared using USE 3 mut DNA template. The products of the above two sets of PCR steps were re-amplified using the 5′ and 3′ wild-type COX-2 proximal primers and the resulting 268 bp fragment was cloned into the BamHI/PstI sites of the pCβS vector.
Mammalian cell culture
HeLa, MDA-MD231 and HepG2 cells were maintained in DMEM (Life Technologies) supplemented with 10% fetal bovine serum (Sigma) and 1% penicillin–streptomycin (Life Technologies). HT29 cells were maintained in RPMI-1640 Medium (Life Technologies) supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin.
HeLa or MDA-MB-231 cells (7 × 10 5 cells/well) were seeded in 60 mm plates ∼12 h before transfection. When cells reached 80% confluency, they were transfected using LT-1 reagent (Mirus). Plasmid DNA (2.8 μg) was diluted in 180 μl of serum-free medium to which 6 μl of LT-1 was added and the mixture was incubated at room temperature for 20 min. Following the addition of 1 ml complete medium to the transfection mixture, the medium on the cells was removed and replaced with the entire transfection cocktail. After 24 h, cells were washed once with 1× phosphate-buffered saline (PBS). Cells were scraped and collected into 1 ml PBS and transferred into microcentrifuge tubes. Cells were then centrifuged at 7000 r.p.m. for 5 min (4500 g in an Eppendorf microcentrifuge). The PBS was aspirated and total RNA was extracted immediately from the cell pellet as described below.
Total RNA isolation and RNase protection assay
To determine polyadenylation signal use in vivo , total RNA from transfected cells was assayed by RNase protection. Total RNA was extracted from the cell pellet using either the TRIZOL method (Invitrogen) or the RNeasy Mini Kit (Qiagen) according to the manufacturer's spin protocol for isolation of total RNA from animal cells. For RNase protection assay reactions, 5 μg of total RNA was used per reaction. Probe RNA for the in vivo polyadenylation assay was prepared as described below. The reporter and/or endogenous RNA levels were determined by RNase protection using the RPAIII kit (Ambion Inc., Austin, TX). The RNA was then analyzed on 5% polyacrylamide–8 M urea gels as described previously ( 64 ). The ratio of the reporter assay and endogenous band detection was quantified using a Typhoon PhosphorImager and ImageQuant software.
In vitro transcription of RNA substrates
RNA transcripts for in vitro polyadenylation, in vivo polyadenylation assays and RNase protection assays were synthesized by use of SP6 or T7 RNA polymerase according to the supplier (Promega) in the presence of 50 μCi of [α- 32 P]UTP (Amersham Pharmacia Biosciences or Perkin Elmer Biosciences). RNAs were purified from 5% polyacrylamide–8 M urea–TBE gels by overnight crush elution in high salt buffer (0.4 M NaCl, 50 mM Tris, pH 8.0 and 0.1% SDS). Prior to use in reactions, eluted RNAs were ethanol precipitated and resuspended in water.
Linearization of all pCβS and pGEM4 DNA constructs at the BamHI site and transcription using T7 RNA polymerase as mentioned above generates antisense RNAs. Transcription of CβS COX-2 ‘proximal’ yielded a 610 base RNA, of COX-2 ‘distal’ an 854 base RNA for use in the in vivo polyadenylation assay. SVL was linearized with DraI as described previously ( 66 ).
In vitro polyadenylation assays
HeLa nuclear extracts were prepared as described [( 64 ) and references therein] using HeLa cells purchased from the National Cell Culture Center (Minneapolis, MN) or grown in our laboratory. In vitro polyadenylation assays using HeLa nuclear extract, the SV40 late polyadenylation signal pre-mRNA and COX-2 USE RNA or non-specific oligoribonucleotides were performed as described previously ( 64 ). Briefly, the in vitro polyadenylation reactions contain a final concentration of 58% (v/v) HeLa nuclear extract, 16 mM phosphocreatine (Sigma), 0.8 mM ATP (Amersham Biosciences), 2.6% polyvinyl alcohol, and 1 × 10 5 c.p.m. of 32 P-labeled SVL substrate RNA (∼50 fmol) in a total volume of 12.5 μl. These reactions are incubated at 30°C for 1 h. The COX-2 USE 3 RNA oligoribonucleotide was synthesized by Dharmacon Research, Inc. (Lafayette, CO) and had the sequence UUGUUUGAUUUCUUAAAGU. The non-specific RNA oligoribonucleotide was described previously ( 64 ).
Cell lysates were prepared by incubating 1.5 × 10 6 cells on ice for 30 min in a solution containing 1% NP-40, 150 mM NaCl and 50 mM Tris–HCl (pH 8.0), plus phenylmethylsulfonyl fluoride (50 μg/ml), leupeptin, aprotinin and pepstatin A (each at 1 μg/ml). Protein from cleared lysate (75 μg) was separated by 10% SDS–PAGE, then blotted onto nitrocellulose membrane and blocked in 5% non-fat dry milk. Primary antibody incubations were performed at a dilution of 1:100 in blocking solution for 3 h at room temperature. Mouse anti-human COX-2 monoclonal antibody and recombinant human COX-2 protein were purchased from Cayman Chemical (Ann Arbor, MI). Goat-antimouse horseradish peroxidase-conjugated secondary antibody (ICN) was used at a dilution of 1:5000 for 1 h at room temperature. Visualization of bound antibodies was accomplished through chemiluminescence using an ECL kit (Amersham Pharmacia Biosciences) and autoradiography.
Expressed sequence tag (EST) sequences corresponding to the human COX-2 gene and their tissue information were obtained from dbEST (March 2004 version; National Center for Biotechnology Information, NCBI), according to the UniGene database (March 2004 version; NCBI). Sequences were aligned to the human genome and polyadenylation sites were determined by using a method described previously [( 43 ) and references therein]. ESTs with poly(A) tail sequence were used to infer polyadenylation sites.
Results are expressed as ±SD of the mean, and analyses were performed by two-sample one-tailed Student's t -test. P -values <0.05 were considered significant.
EST surveys suggest existence of multiple polyadenylation signals for the human COX-2 gene with tissue specificity
Because alternative polyadenylation is potentially a key regulator of human COX-2 expression, we systematically investigated the mechanism controlling alternative 3′ end formation. Using a bioinformatics approach, we have surveyed the current EST database for human COX-2 mRNAs that would represent utilization of either the proximal or the distal COX-2 polyadenylation signal. Using ESTs with poly(A) tail sequence (see Materials and Methods; B. Tian and C. S. Lutz, unpublished data), we identified three COX-2 polyadenylation signals, named proximal 1, proximal 2 and distal, respectively ( Table 1 , top). The closed diamonds represent ESTs containing poly(A) tail sequence. The open circles denote ESTs that do not have poly(A) tail sequence. Since these open circles are located after proximal 2, they indirectly support the use of the distal polyadenylation signal, assuming there is no other poly(A) signal between proximal 2 and distal. The distance between proximal 1 and proximal 2 is ∼400 nt, and between proximal 2 and distal is a little less than 2000 nt. Since proximal 1 lacks a definable CstF binding site and previous northern blot data only support transcript sizes resulting from proximal 2 and distal signal usage ( 80 , 81 ), we focused on proximal 2, which we will refer to simply as ‘proximal’ throughout the rest of our study. The distance between proximal and distal from this EST survey is in good agreement with those reported in previous studies [( 80 , 81 ) and Figure 1 ]. Also in Table 1 are cDNA library IDs for the ESTs found (first column in Table 1 ) and tissue and organ information if known on the biological source of the cDNA libraries (last two columns in Table 1 ). Use of the COX-2 proximal and distal polyadenylation signals are noted by the numbers of ESTs that appeared from each library.
To our knowledge, this is the first data for COX-2 examined in this manner. Moreover, our data are comprehensive; all available ESTs corresponding to human COX-2 were used. We would like to highlight that (i) both COX-2 polyadenylation signals are used in vivo ; (ii) there are different tissue patterns/distributions and (iii) this is a much larger-scale analysis than has previously been reported. Importantly, these tissue-specific distribution patterns and the polyadenylation signals utilized closely match smaller scale, reported tissue specificity ( 80 , 81 ), with use of the proximal form notably predominating liver, while the distal form notably predominates in colon. Our data underscore the biological relevance of our study and highlight the need for understanding the mechanisms of regulation that are involved.
Endogenous COX-2 mRNAs are polyadenylated at different sites in cell lines representing different tissues
To further evaluate the biological usage of both polyadenylation signals with regard to the tissue specificity noted by our bioinformatics analysis, we chose two established human cancer cell lines, HT29 (colon) and HepG2 (liver). Like many colon cancer-derived cell lines ( 75 ), HT29 cells constitutively express COX-2 since endogenous COX-2 protein can be readily detected by western blotting in HT29 extracts ( Figure 2A , lane 2). The constitutive expression of COX-2 in hepatocellular carcinoma cells such as HepG2 has been debated in the literature. In our hands, COX-2 protein was detected in HepG2 extract indicating that COX-2 is expressing in this hepatocellular carcinoma cell line ( Figure 2A , lane 3). The levels of endogenous COX-2 protein appear to be lower in HepG2 cells than in HT29 cells. As a loading control, we probed the same western blot with an antibody specific for glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and found GAPDH to be equivalent in each lane (data not shown).
It has been reported that the distal COX-2 polyadenylation signal is used in primary colon samples and that the proximal polyadenylation signal is used in liver samples as assayed by northern blots, but this conclusion was based purely on transcript size ( 81 ). We therefore decided to perform our assay based on a more precise method of analyzing polyadenylation signal usage. This assay allows us to specifically identify use of the proximal and distal polyadenylation signals ( Figure 2B ; see also Figure 1 ). Total RNA was extracted from HT29 and HepG2 cells and this RNA was used in standard RNase protection assays using specific COX-2 proximal and distal probes. pGEM4-distal and pGEM4-proximal antisense probes protected the specific bases corresponding to the proximal and distal polyadenylation signals. The data in Figure 2B show that HT29 cells primarily use the COX-2 distal polyadenylation signal, giving a protected fragment of 235 bp. HepG2 cells utilize both the COX-2 proximal and distal signals, giving protected fragments of 168 and 235 bp, respectively ( Figure 2C ). The proximal probe is also able to bind to and protect a 268 bp fragment in the distal mRNA, which corresponds to the entire, uncleaved and unused proximal polyadenylation signal (lane 1 in Figure 2B and C ). The larger size reflects the fact that sequences downstream of the proximal cleavage site are contained in the distal mRNA. Our data reveal that there is regulated, differential, tissue-specific polyadenylation of COX-2.
In vivo polyadenylation of COX-2 using reporter assays
After confirming that both COX-2 polyadenylation signals are used in vivo , we next applied our in vivo polyadenylation assays ( 64 ) to address directly the mechanism controlling use of the COX-2 proximal and distal polyadenylation signals. We first addressed how the suboptimal AUUAAA proximal COX-2 polyadenylation signal behaved in our system. The apparent weakness of this polyadenylation signal was postulated based upon its divergence from the consensus AAUAAA CPSF binding site. We cloned a 268 bp fragment by PCR including the human COX-2 proximal polyadenylation signal and flanking sequences into our two polyadenylation signal reporter vector pCβS, diagrammed in Figure 3A . In this series of experiments, the proximal signal was inserted into the signal 1 site while the BGH signal occupied the signal 2 site. This pCβS-COX-2-proximal construct was then transformed into HeLa or MDA-MB-231 (breast cancer) cells. Total RNA was isolated after 24 h, RNase protection assays were performed using the antisense probe to the pCβS-COX-2-proximal construct ( Figure 3A ), and the results were quantified using a PhosphorImager and ImageQuant software. RNase protection assays will produce a protected fragment of 168 bases if the COX-2 proximal polyadenylation signal is used; utilization of the BGH signal will produce a protected fragment of 410 bases. We found that polyadenylation at the COX-2 proximal signal compared with that at the BGH signal occurred in a ratio of 1:3, indicating that the proximal COX-2 signal is indeed suboptimal ( Figure 3B , lane 4 and Figure 3C ).
We next compared polyadenylation using the COX-2 distal polyadenylation signal to BGH by the same type of assay. We cloned a 454 bp fragment by PCR containing the COX-2 distal polyadenylation signal into the pCβS vector, with the COX-2 distal signal occupying the polyadenylation signal 1 site in the reporter vector. Use of the COX-2 distal polyadenylation signal in the RNase protection assay will protect a fragment of 235 bases. Because the distal polyadenylation signal has a consensus AAUAAA, we predicted that this signal would be stronger than the proximal polyadenylation signal of COX-2. Indeed, the COX-2 distal polyadenylation signal was stronger than the COX-2 proximal signal, and was used ∼45% of the time compared with polyadenylation at the BGH signal of ∼55% ( Figure 3B , lane 5 and Figure 3D ). In this set of experiments, the band corresponding to BGH migrated slightly faster than expected. This result was reproducible in this assay, and has been observed previously using the same assay and same vector ( 64 ). These assays were performed in HeLa or MDA-MB-231 cells where there is no detectable endogenous COX-2 protein or mRNA (data not shown). These data demonstrate that although both COX-2 polyadenylation signals are used, the COX-2 proximal signal is indeed weak and the COX-2 distal signal is stronger as compared with the internal control.
Since the COX-2 polyadenylation signals are normally in tandem in the COX-2 3′-UTR, and therefore may be in competition with each other, we measured their relative use in a similar assay as described above. Here, the COX-2 proximal and distal polyadenylation signals were both placed in their respective order in the pCβS reporter plasmid while keeping the BGH polyadenylation signal intact (pCβS-proximal-distal-BGH). After transfection into HeLa cells, RNase protection assays were performed as described above. In this context, the relative use of COX-2 proximal signal was 14.3%, the relative use of the COX-2 distal signal was 53.5%, and the relative use of the BGH polyadenylation signal was 32.2%, giving a ratio of proximal/distal/BGH of 1:3.8:2.3 (data not shown). When the BGH polyadenylation signal was removed to produce a tandem construct (pCβS-proximal-distal), transfected into HeLa or MDA-MB-231 cells, RNA isolated and RNase protection assays performed, the ratio of proximal/distal polyadenylation signal usage was 1:3, not appreciably different from when the BGH was present ( Figure 3B , lane 7 and Figure 3E ). Note that the creation of the pCβS-proximal-distal construct caused the resulting protected fragments to differ in size slightly from the results obtained using either the pCβS-proximal or the pCβS-distal alone. These data demonstrate that the COX-2 proximal polyadenylation signal does not compete significantly with the distal polyadenylation signal in tandem.
Effect of mutation of core and auxiliary polyadenylation elements in the suboptimal COX-2 proximal polyadenylation signal
Since the COX-2 proximal core polyadenylation signal is AUUAAA, and therefore differs from the consensus AAUAAA, we showed that it was suboptimal in vivo ( Figure 3B , lanes 4 and 7). Previous in vitro studies in the Wickens and Shenk laboratories had mutated the consensus AAUAAA to every base at each position ( 82 , 83 ). These studies had shown in vitro that while AUUAAA worked better than any other derivative of AAUAAA, it still did not perform as well as the consensus AAUAAA. To further prove this in vivo , we mutated the A U UAAA to the consensus sequence A A UAAA using site-directed mutagenesis (see Figure 4A for diagram). We then measured polyadenylation signal use (relative to BGH polyadenylation signal use) by RNase protection assays as described above. Here, we have calculated fold-induction at the proximal polyadenylation signal; so utilization at the wild-type proximal signal was set at a value of 1. We found that this single change, A U UAAA to A A UAAA mutation, enhanced polyadenylation by 3.5-fold over the level of polyadenylation at the COX-2 proximal polyadenylation signal ( Figure 4B ), thus acting as a positive mutant. Similar results were found using MDA-MB-231 cells (data not shown). These results confirmed directly for the first time in vivo that the difference between an ‘optimal’ AAUAAA and a ‘suboptimal’ AUUAAA core polyadenylation signal is operationally real.
Polyadenylation efficiency can be enhanced by the presence of auxiliary elements located upstream of the CPSF binding site, known as USEs ( 64 – 66 ). Sequences upstream of the COX-2 proximal polyadenylation signal that we identified bear striking similarity to the consensus UAU 2–5 GUNA USE sequence previously studied in our laboratory. The COX-2 USEs are located 58, 49 and 6 bases upstream of the proximal polyadenylation signal ( Figure 4A ). We predicted that these putative USEs would affect polyadenylation efficiency of the suboptimal COX-2 proximal polyadenylation signal. In order to experimentally address this prediction, we mutated the USEs in the context of our in vivo polyadenylation construct. Mutants were prepared by PCR-based site-directed mutagenesis in each of the three putative USEs, as well as a non-specific mutant in a non-USE region upstream of the COX-2 proximal polyadenylation signal ( Figure 4A ). These constructs were then transfected into HeLa (or MDA-MB-231 cells; data not shown), total RNA was isolated and RNase protection assays were performed as described above. The results are shown in Figure 4C . Mutation of each USE was increasingly more deleterious as the USE mutation approached the suboptimal core polyadenylation hexamer element AUUAAA, with USE 3 mutation alone decreasing in vivo polyadenylation to approximately half of the wild-type level ( Figure 4C ). Mutation of the USEs in duplicate (USE 1,2 mut; USE 2,3 mut; USE 1,3 mut) resulted in approximately the same level of reduced polyadenylation efficiency as the USE 3 mut alone ( Figure 4C ). However, the triple USE mutation (USE 1,2,3 mut) resulted in a significant decrease in polyadenylation signal utilization ( Figure 4C ). Deletion of the entire USE-containing region in the 5′Δ mutant almost abolished polyadenylation at the COX-2 proximal polyadenylation signal ( Figure 4C ). Taken together, these data suggest that the USEs may have additive or synergistic effects. These data also suggest that the USEs are important polyadenylation efficiency elements in the COX-2 proximal polyadenylation signal, and perhaps ensure that the proximal signal is utilized.
USE competition for polyadenylation suggests that a trans -acting factor could be influencing polyadenylation at signals containing USEs
To further define a functional role of the USEs as auxiliary efficiency elements, we performed in vitro polyadenylation assays using specific competitors. Previously, we have shown that in vitro polyadenylation reactions containing an SV40 late polyadenylation signal substrate RNA (SVL) could be inhibited specifically by oligoribonucleotides representing the SV40 USE motifs ( 66 ), and that polyadenylation in vitro of human COL1A2 substrate RNA can be inhibited by a COL1A2 USE oligoribonucleotide ( 64 ). We now prepared an oligoribonucleotide that represents COX-2 USE 3, and added this oligoribonucleotide to in vitro polyadenylation reactions containing HeLa nuclear extract and a 32 P-labeled SVL prepared by in vitro transcription using SP6 polymerase. This substrate RNA was chosen because it works well for in vitro polyadenylation, and because the USEs in SV40 are similar to those found in the proximal COX-2 polyadenylation signal. The data are shown in Figure 5A and B . Figure 5A demonstrates graphically that the COX-2 USE oligoribonucleotide specifically inhibited SVL polyadenylation in vitro , whereas a non-specific oligoribonucleotide had no significant effect on SVL polyadenylation. An additional non-specific oligonucleotide also had no effect (data not shown) ( 64 ). Figure 5B shows representative experiments. Lane 1 in Figure 5B , both upper and lower panels, represents a reaction performed in the absence of competitor oligoribonucleotides and demonstrates that the SVL RNA was efficiently polyadenylated in the in vitro system. Interestingly, 50 pmol of the COX-2 USE 3 oligoribonucleotide specifically inhibited polyadenylation to <10% of the control reaction ( Figure 5B , upper panel). This amount is similar to other USE oligoribonucleotide inhibition reactions ( 64 , 66 ). The non-specific oligo did not inhibit in vitro polyadenylation ( Figure 5B , lower panel). In vitro polyadenylation of a non-USE containing polyadenylation signal, AAV, was also not inhibited by the addition of the COX-2 USE 3 at similar concentrations as expected ( Figure 5C ). Polyadenylation of AAV was also not affected by addition of the non-specific oligonucleotides as expected ( Figure 5C ). Taken together, these data demonstrate that the COX-2 USE 3 oligoribonucleotide specifically binds and sequesters a putative common factor(s) important for polyadenylation, and suggest that the similarity to the SV40 USE motifs is functionally significant.
Polyadenylation is a useful and underexplored mechanism of regulating gene expression, and this regulation is partially implied by the complex nature of this process. Alternative polyadenylation can generate mature transcripts that only differ in their 3′ ends. Efficiency of processing is another level at which regulation can occur, since most pre-mRNAs in the cell are not efficiently processed, and therefore, even small changes in the overall processing efficiency of a particular pre-mRNA may have a great effect on ultimate protein levels. Experimental evidence has demonstrated that poly(A) signal strength directly influences the amount of mature, exported mRNA ( 84 , 85 ).
Our objective in this study was to accurately define and characterize alternative polyadenylation as it applied to the human COX-2 3′-UTR. Interestingly, the size and scheme of the 3′-UTR is conserved among human, chimpanzee, mouse and rat COX-2 genes, along with the two polyadenylation signals, suggesting that the importance of alternative polyadenylation plays a key regulatory function (B. Tian, H. Zhang, and C. S. Lutz, unpublished data). Indeed, the sequences for USEs 1 and 2 are exact matches in the 3′-UTR of mouse COX-2 homolog (B. Tian, H. Zhang, and C. S. Lutz, unpublished data). Choice of a particular polyadenylation signal may result in mRNAs with different stability or different translation efficiency, which has direct consequences for how much protein product is ultimately produced. Therefore, polyadenylation signal choice can be a key step in mRNA maturation. Our work examined the relative strength and composition of the COX-2 proximal and distal polyadenylation signals. We have shown that the COX-2 proximal and COX-2 distal polyadenylation signals are utilized in an in vivo polyadenylation assay. We further demonstrated that the use of these COX-2 polyadenylation signals is regulated in a tissue-specific manner. We propose that while basal polyadenylation of these signals does occur experimentally in cell types that do not normally express COX-2 mRNA (HeLa and MDA-MB-231), the cell-type specific utilization of COX-2 proximal and distal polyadenylation signals may require additional as yet unidentified factors.
Additionally, we have demonstrated that a suboptimal CPSF binding site (AUUAAA) is truly weaker than the canonical CPSF binding site, AAUAAA. Taken together, the data suggest a working hypothesis, shown in Figure 6 , for polyadenylation signal choice, as well as for possible correlations between polyadenylation signal choice and resulting mRNA stability as postulated ( 81 ).
Our model ( Figure 6 ) suggests that use of the COX-2 proximal polyadenylation signal results in a shorter mRNA, while use of the COX-2 distal polyadenylation signal results in a longer mRNA. Use of the proximal polyadenylation signal would exclude a portion of the 3′-UTR, which may contain important regulatory elements, for example, elements that regulate mRNA stability. The best characterized 3′-UTR sequences that cause mRNA instability are the AREs found in many short-lived mRNAs such as cytokines, growth factors and proto-oncogenes ( 72 , 73 ). The 3′-UTR of COX-2 is complex and has many AREs ( Figure 1 ). Currently, data in the literature conflict with regard to which form of COX-2 is more stable. It has been shown that removal of some sequences upstream of the COX-2 proximal polyadenylation signal containing some of the AREs confers increased stability to a heterologous reporter ( 86 ), while other data have shown that this same region can transfer instability to a heterologous reporter ( 87 , 88 ). Additionally, the longer COX-2 mRNA has been reported to be the more unstable form as revealed by northern blotting of the COX-2 mRNA while the shorter form is generally thought to be the more stable form ( 81 ). While our study has not addressed these questions of stability directly, our study underscores the importance of understanding how alternative polyadenylation of COX-2 is regulated.
We have also demonstrated here that the USEs present in the suboptimal COX-2 proximal polyadenylation signal have a significant effect on in vivo polyadenylation of that signal, since mutation or deletion of the USEs has a substantial effect ( Figure 4C ). We suggest that the presence of the USEs in the COX-2 proximal polyadenylation signal ensure its usage, possibly by supporting and enhancing assembly of the basal polyadenylation machinery on the polyadenylation signal, or by serving as a trans -acting factor binding site which may recruit or enhance interactions with the basal polyadenylation machinery. These explanations are not mutually exclusive. Additionally, tissue-specific factors may be involved since we have shown tissue-specific utilization occurs at the COX-2 proximal polyadenylation signal. These details remain to be explored.
Supplementary Material is available at NAR Online.
Top: genomic sequence of COX-2 3′-UTR containing three polyadenylation signals is depicted schematically. The position of proximal 1 is arbitrarily set as 0. Closed diamonds denote the locations of poly(A) signals supported by ESTs with poly(A) tail sequence. Open circles are the locations of the 3′ end of ESTs without poly(A) tail sequence. Bottom: LIB_ID corresponds to cDNA library ID in the dbEST database. P1, proximal 1; P2, proximal 2; D, distal; and D1, distal supported by ESTs without poly(A) tail sequence (explained in the text).
We thank the members of the Lutz, O'Connor, Rogers, Bellofatto, Modak and Wilusz laboratories for helpful experimental suggestions and comments on this manuscript. Special thanks to David Fritz for helpful suggestions and Matias Elijovich for technical assistance. This work was funded by American Cancer Society Grant RPG-00-265-01-GMC and NSF award MCB-0426195 to C.S.L. Funding to pay the Open Access publication charges for this article was provided by NSF award MCB-0426195.
Conflict of interest statement . None declared.