A Cyclooxygenase-2-dependent Prostaglandin E2 Biosynthetic System in the Golgi Apparatus*

Background: When cyclooxygenases-1 and -2 (COXs-1 and -2) are co-expressed, COX-2 can function, whereas COX-1 is latent. Results: Significant amounts of COX-2, microsomal PGE synthase-1, and cytosolic PLA2α but not COX-1 are in the Golgi. Conclusion: Cytosolic PLA2α, COX-2, and microsomal PGE synthase-1 comprise a unique COX-2-dependent PGE2 biosynthetic system in the Golgi. Significance: COX-2 and COX-1 can function in different subcellular compartments. Cyclooxygenases (COXs) catalyze the committed step in prostaglandin (PG) biosynthesis. COX-1 is constitutively expressed and stable, whereas COX-2 is inducible and short lived. COX-2 is degraded via endoplasmic reticulum (ER)-associated degradation (ERAD) following post-translational glycosylation of Asn-594. COX-1 and COX-2 are found in abundance on the luminal surfaces of the ER and inner membrane of the nuclear envelope. Using confocal immunocytofluorescence, we detected both COX-2 and microsomal PGE synthase-1 (mPGES-1) but not COX-1 in the Golgi apparatus. Inhibition of trafficking between the ER and Golgi retarded COX-2 ERAD. COX-2 has a C-terminal STEL sequence, which is an inefficient ER retention signal. Substituting this sequence with KDEL, a robust ER retention signal, concentrated COX-2 in the ER where it was stable and slowly glycosylated on Asn-594. Native COX-2 and a recombinant COX-2 having a Golgi targeting signal but not native COX-1 exhibited efficient catalytic coupling to mPGES-1. We conclude that N-glycosylation of Asn-594 of COX-2 occurs in the ER, leading to anterograde movement of COX-2 to the Golgi where the Asn-594-linked glycan is trimmed prior to retrograde COX-2 transport to the ER for ERAD. Having an inefficient ER retention signal leads to sluggish Golgi to ER transit of COX-2. This permits significant Golgi residence time during which COX-2 can function catalytically. Cytosolic phospholipase A2α, which mobilizes arachidonic acid for PG synthesis, preferentially translocates to the Golgi in response to physiologic Ca2+ mobilization. We propose that cytosolic phospholipase A2α, COX-2, and mPGES-1 in the Golgi comprise a dedicated system for COX-2-dependent PGE2 biosynthesis.

immunity (1), reproduction (2), and renal and cardiovascular homeostasis (3,4). Two isoforms, prostaglandin-H synthases-1 and -2, catalyze the committed step in the biosynthesis of PGs from arachidonic acid (AA). These enzymes are commonly referred to as cyclooxygenases-1 and -2 (COXs-1 and -2), and this terminology is used in the present report. The formation of PGH 2 from AA involves two distinct steps occurring at separate COX and peroxidase active sites of prostaglandin-H synthases (for recent reviews see, Refs. 5-7). Briefly, after AA is mobilized from the sn2 position of membrane glycerophospholipids by one or more phospholipase A 2 (PLA 2 ) species (for recent reviews, see Refs. 8 -10), AA is oxygenated to form PGG 2 in the COX site. Newly formed PGG 2 then moves to the peroxidase site of the enzyme or another peroxidase where the 15-hydroperoxyl group of PGG 2 is reduced, yielding PGH 2 . Subsequently, PGH 2 is converted by a downstream prostaglandin synthase such as microsomal PGE synthase-1 (mPGES-1) into a biologically active prostanoid as detailed in a recent review (7).
Both isoforms catalyze the conversion of AA to PGH 2 with similar K m and V max values, and the critical catalytic residues are the same in COX-1 and COX-2 (5-7). Although both isoforms are sequence homodimers, they each function as conformational heterodimers comprising allosteric and catalytic monomers with only one of the two COX sites catalyzing a reaction at any given time (11)(12)(13)(14)(15)(16). Recent studies by our group have shown that the activities of both COX-1 and COX-2 are attenuated or enhanced, respectively, by non-substrate fatty acids such as palmitic acid that function allosterically to regulate COX activity (14,16,17).
COX-1 and COX-2 are N-glycosylated, ER-resident, homodimeric enzymes that exhibit about 60% sequence identity within a species. The most obvious sequence difference between COX-1 and COX-2 is the presence in COX-2 of a 19-amino acid sequence (Asn-594 to Lys-612) located just upstream of a C-terminal STEL ER retention signal.
Structurally, the mature monomers of COX-1 and COX-2 have three folding domains including sequentially an epidermal growth factor (EGF)-like domain, a membrane binding domain, and a catalytic domain (Fig. 1A). The function of the N-terminal EGF-like domain, part of which lies at the dimer interface, remains unclear. As first proposed by Garavito and co-workers (18), the membrane binding domains serve to anchor COXs to one face of a lipid bilayer. Both COX-1 and COX-2 have been shown to be bound to the luminal surfaces of the lipid bilayer of the ER and the contiguous inner membrane of the nuclear envelope (19 -23). COX-1 appears somewhat more concentrated in the ER, and COX-2 appears more concentrated on the nuclear envelope (24). The catalytic domain is located toward the C terminus, includes about 75% of the entire sequence, and contains the key COX and peroxidase active site residues. Despite their many similarities in structure and catalytic activities, COX-1 and COX-2 display temporally distinct patterns of gene expression and mRNA and protein turnover (25). The evolutionary forces that have driven many species to have two biochemically similar COX isomers associated with different physiologic events remain an enigma (26,27).
With respect to protein expression, COX-1 is a stable protein (28 -30) that is constitutively expressed in many but not all mammalian cells (31). PGs formed via COX-1 are involved in homeostatic housekeeping events such as thrombosis, parturition, and regulation of renal water balance. In contrast, COX-2 is usually expressed transiently upon treatment of cells with growth factors or proinflammatory stimuli (25) that promote cell division or differentiation. The expression of COX-2, although undetectable in most cells, can increase and decrease in a matter of hours often in association with inflammation. Regulation of COX-2 protein concentrations occurs at multiple transcriptional, post-transcriptional, translational, and posttranslational levels (25,29,32). COX-2 overexpression is associated with several forms of cancer (33,34).
The present report focuses on the post-translational processing and degradation of COX-2 in the context of COX-2 trafficking and the subcellular localization and functioning of COX-2. We have reported that properly folded COX-2 can be degraded via two distinct pathways including AA turnover-dependent degradation and constitutive degradation involving the ER-associated degradation (ERAD) pathway (29,30). Others have shown that COX-2 can be ubiquitinated and degraded through the 26 S proteasome (32,35). ERAD involves a series of related but different events that include detecting misfolded proteins bound for degradation, translocating these proteins into the cytosol, and degrading the proteins via the 26 S proteasome (for recent reviews see, Refs. 36 -40). In general, the ERAD pathway functions as a quality control process to monitor protein folding and to remove aberrant proteins from the ER. However, in the case of COX-2, the ERAD pathway serves as one mechanism to control the level of active COX-2.
Previously, we identified a 27-amino acid instability motif involving residues Glu-586 to Lys-612 of COX-2 that acts as a degradation signal or "degron" (39). This degron includes the unique 19-amino acid sequence near the C terminus of COX-2 that is essential for regulating N-glycosylation of Asn-594 and mediating COX-2 entry into the ERAD pathway. Deletion of the 19-amino acid sequence Asn-594 to Lys-612 (Fig. 1A) of human (hu) COX-2 generates a degradation-resistant form, whereas inserting this sequence into the corresponding position in COX-1 markedly destabilizes the resulting COX-1 (28). The COX-2 degron appears to control the entry of COX-2 into the ERAD pathway. The effect parallels the function that the multiple AUUUA elements in the 3Ј-untranslated region of COX-2 mRNA play in its rapid cleavage (41).
There are four known N-glycosylation sites in COX-2 (42) (see Fig. 1A). N-Glycosylation of Asn-594 within the degron occurs post-translationally and is central to its ERAD (29,30,35). An N594A mutation in COX-2 blocks enzyme degradation. During the process of distinguishing between terminally misfolded proteins and properly folded intermediates, ␣-mannosidases remove mannoses to trim the Asn-linked Man 9 -(GlcNAc) 2 oligosaccharides, generating an appropriate ERAD signal (38,43,44). Apparently, this occurs with the N-glycosyl moiety linked to Asn-594 of COX-2. Importantly, N-glycosylation of Asn-594 is necessary but not sufficient for proteasomal degradation of COX-2; additional segments of the degron are involved (28,29). Accordingly, we searched for other factors involved in the ERAD of COX-2. During the course of these studies, we discovered that replacing the STEL sequence at the C terminus of COX-2 with KDEL, a more typical and robust ER retention signal (45), yielded a degradation-resistant COX-2. This led us to explore the role of protein trafficking in the degradation of huCOX-2. Data reported here indicate that a significant amount of huCOX-2 resides in the Golgi apparatus, whereas little or no COX-1 is detectable in the Golgi. Additionally, we found that one of the enzymes downstream of COX-2 in PG biosynthesis, mPGES-1, is also present in the Golgi and that COX-2 and mPGES-1 but not COX-1 and mPGES-1 efficiently couple to form PGE 2 from endogenously mobilized AA in intact cells. These findings in combination with a previous observation that cytosolic (c) PLA 2␣ undergoes a Ca 2ϩ -dependent translocation to the Golgi (46) led us to conclude that there is a COX-2-dependent PGE 2 biosynthetic system located in the Golgi apparatus.
Construction of COX Mutant Variants-Most constructs including those derived through site-directed mutagenesis, motif replacements, truncations, or deletions were generated according to the instructions of the QuikChange site-directed mutagenesis kit (Stratagene). The tetracycline-inducible vectors pCDNA5/FRT/TO harboring huCOX-1, huCOX-2, or ovine (ov) COX-1 (29) were used as templates for PCR. In constructing Golgi-⌬STEL huCOX-1 and Golgi-⌬STEL huCOX-2, the DNA fragments of Golgi-⌬STEL huCOX-2 were created from the DNA templates for Golgi-YFP and ⌬STEL huCOX-1 or ⌬STEL huCOX-2 by overlapping extension PCR and then subcloned into pcDNA5/FRT/TO using NotI sites. The correct sequences of constructs were confirmed by DNA sequencing by the University of Michigan DNA Sequencing Core.
Generation of Stable HEK293 Cell Lines-HEK293-derived cell lines stably and inducibly expressing various COX constructs were generated in tetracycline-inducible mammalian expression systems according to the manufacturer's instructions (Flip-in R T-Rex R , Invitrogen). Briefly, transfections of HEK293 cells in 60-mm cell culture dishes were performed using a calcium phosphate method with various COX-pCDNA5 vectors (2 g) and pOG44 helper vectors (2 g). After transfection overnight, the cells were trypsinized, collected, and transferred to complete DMEM (i.e. DMEM with 10% FBS) in the presence of 20 M blasticidin S and 50 M hygromycin for 2 weeks. Media containing the same amount of antibiotics for selection of clones were changed once after 1-week incubations. The surviving cells were collected and used as stable cell lines.
Measurements of COX Protein Degradation-To measure rates of protein degradation of huCOX-2 variants, HEK293 cells stably expressing each of the variants were grown to ϳ80% confluence, subjected to serum starvation for 24 h, and then treated with tetracycline (10 g/ml) for another 24 h in complete DMEM to induce the expression of COXs. Protein degradation experiments were then initiated by adding puromycin (50 g/ml) in the presence or absence of various trafficking inhibitors including MG132 (20 M) and H-89 (20 M) and/or other agents. Cells were harvested at various times, and COX and actin protein levels were determined by Western transfer blotting as described in the next section. In decay experiments for Golgi-⌬STEL huCOX-2 and ⌬STEL huCOX-2, the cells were pretreated with H-89 and MG132 for 1 h and then treated with puromycin (50 g/ml) in the presence of the same concentrations of inhibitors.
To measure rates of degradation of murine (mu) COX-2, murine NIH 3T3 fibroblasts were first subjected to serum starvation for 48 h and then treated with phorbol 12-myristate 13-acetate (1 M) for 4 h to stimulate the expression of muCOX-2. Before performing the protein degradation experiments, the cells were preincubated in the presence or absence of various inhibitors (MG132 and H-89) for 1 h at the concentrations described above for studies of huCOX-2 degradation. The degradation experiments with NIH 3T3 cells were then conducted for various times with cycloheximide (50 M) in the presence or absence of inhibitors.
At the indicated times after different treatments, HEK293 cells or NIH 3T3 cells were harvested and resuspended in icecold PBS, pH 7.4 containing Complete protease inhibitor. After sonication, the protein concentration of the whole cell lysate was determined using a BCA protein assay kit (Thermo Scientific). The same amounts of protein sample were loaded into individual lanes for SDS-PAGE and resolved on a 7% tris acetate polyacrylamide gel based on the NuPAGE system (Invitrogen). After transfer to nitrocellulose membranes, the membranes were blocked in Tris-buffered saline with Tween 20 (i.e. 137 mM NaCl, 20 mM Tris-HCl, 0.1% Tween 20, pH 7.6) with 4% fat-free milk. Immunoblotting was performed with the appropriate primary anti-bodies in Tris-buffered saline with Tween 20 with 4% fat-free milk. Horseradish peroxidase-conjugated anti-rabbit or anti-mouse antibodies (Bio-Rad) were used as secondary antibodies. Immunodetection was performed using an ECL kit (Thermo Scientific) followed by exposure to x-ray film. Densitometry was performed using ImageJ software.
For Western blotting of native huCOX-2 and huCOX-2-derived mutants, either a rabbit polyclonal antibody previously raised against a peptide extending from Pro-583 to Asn-594 of huCOX-2 (29) or a commercial rabbit polyclonal antibody (Novus Biologicals, NB100-689) was used as the primary antibody. For Western blotting of samples from murine NIH 3T3 cells, only the commercial antibody from Novus Biologicals was used as the primary antibody. A rabbit polyclonal antibody raised against a peptide extending from Leu-272 to Gln-283 of ovCOX-1 was used to detect huCOX-1, ovCOX-1, and its mutants (29). A commercial anti-actin antibody raised in mouse was used to detect actin, which was used as the internal control.
Deglycosylation with Endoglycosidase H and Peptide-N-glycosidase F-For deglycosylation of COXs, HEK293 whole cell lysates (3 g of total protein) were heated at 100°C in denaturation buffer for 10 min and then treated with endoglycosidase H and/or peptide-N-glycosidase F at 37°C for 3 h. Afterward, SDS loading buffer was added to the reaction mixture for resolving the proteins by SDS-PAGE. Images were obtained after immunoblotting.
Cell Transfection with Sar1 and H79G Sar1-Transfections of HEK293 cells with Sar1 or dominant negative H79G Sar1 were performed using Lipofectamine 3000 according to the instructions of the manufacturer (Invitrogen). Briefly, 2 ϫ 10 6 HEK293 cells stably expressing native huCOX-2 were seeded into 6-well cell culture dishes 1 day before transfection. In the process of transfection, a preincubated mixture containing 250 l of Opti-MEM, 1 g of a DNA, 2 l of P3000 reagent, and 350 l of Lipofectamine 3000 was added dropwise to each well. After a 24-h incubation, the transfection reagent mixtures were removed, and the cells were incubated with tetracycline (10 g/ml) for another 24 h in fresh complete DMEM to induce COX-2 expression. Finally, protein degradation experiments were conducted by incubating the cells with puromycin (50 g/ml) for the indicated times, and COX-2 protein levels were determined by immunoblotting.
Immunocytofluorescence-HEK293 cell lines stably and inducibly expressing different COX variants were seeded on four-chambered slides in complete DMEM. The next day, the cells were subjected to a 24-h serum starvation regimen, and then COX expression was induced with tetracycline (10 g/ml) for 24 h in complete medium. For studies of murine NIH 3T3 fibroblasts and human dermal fibroblasts, the cells were seeded on four-chambered slides, subjected to serum starvation for 48 h, and then stimulated with phorbol 12-myristate 13-acetate (1 M) with or without 10% FBS for 4 h. For huCOX-1 in human CCL210 (ATCC) lung fibroblasts, after seeding the chambered slides and culturing for 1 day, the cells were used without further treatment. The remaining parts of the protocol were the same for all the immunocytofluorescence studies. After washing with PBS three times, the cells were fixed and permeabilized with 100% methanol at room temperature for 5 min. Prior to immunostaining, the fixed cells were rehydrated and then blocked with PBS containing 0.1% Triton X-100 supplemented with 10% goat serum. Two steps of co-immunostaining were then performed. First, the cells were co-stained with two primary antibodies against a COX and a cellular organelle, respectively. The primary antibodies used were as follows: rabbit polyclonal anti-calnexin (GeneScript), rabbit polyclonal antigiantin (Abcam), mouse monoclonal anti-COX-2 (Cayman Chemical Co.), and mouse monoclonal anti-COX-1 (Invitrogen) antibodies. In the second step, both Oregon Green antirabbit and Texas Red anti-mouse antibodies (Invitrogen) were used as secondary antibodies. After each step of immunostaining, three extensive washes with PBS were performed. Finally, the slides were dried and mounted using a Prolong antifade kit (Invitrogen). Microscopy was conducted in the University of Michigan Center for Live Cell Imaging on a Nikon Infinity confocal microscope at a magnification of 100ϫ. The resultant images were further processed and analyzed with Metamorph software to obtain correlation coefficient (CC) values. CC values for each COX sample were mean values collected from at least 15 cells. To assure a random selection of images, a second, control set of CC values was obtained by rotating one image from one camera 90°and testing for overlap with the image from the other camera; the differences between the observed and random control CC values were statistically different in all cases based on Student's t test (p Ͻ 0.05).
Cyclooxygenase Assays-COX assays were typically conducted in 100 mM Tris-HCl, pH 8. The supernatants were collected, and PGE 2 was quantified using a PGE 2 immunoassay kit (Cayman Chemical Co.) or a standard RIA using an anti-PGE 2 antibody from Sigma. All samples were stored at Ϫ80°C before RIAs were performed. Cell samples were collected for BCA protein assays and Western blotting.

RESULTS
Role of the C-terminal STEL Sequence in the Degradation of huCOX-2-Except for ovCOX-1, all mammalian COXs sequenced to date contain a STEL sequence at their C termini ( Fig. 1A) (7); ovCOX-1 has a PTEL sequence. The STEL sequence is a variant of the more typical, robust KDEL ER retention signal found in most mammalian ER-resident proteins. KDEL receptors localized in the cis-Golgi efficiently recognize and bind C-terminal KDEL-like motifs and rapidly transport proteins bearing this motif back to the ER. In contrast, stud-ies of various KDEL-like sequences in HeLa cells indicate that proteins carrying a STEL ER retention sequence are inefficiently transported from the Golgi to the ER (45).
We observed that replacing STEL with KDEL in huCOX-2 yields a form of the enzyme that is much more stable (t1 ⁄ 2 Ͼ 24 h) than native huCOX-2 (t1 ⁄ 2 ϳ 5 h) (Fig. 1, A-C). Thus, insertion of a strong ER retention signal slows the degradation of huCOX-2.
The impact of the KDEL substitution on huCOX-2 degradation is comparable with that of an N594A substitution or deletion or modifications of certain other residues in the C-terminal degron of huCOX-2 (28,29). As shown in Fig. 1B, inhibition of KDEL huCOX-2 protein synthesis by puromycin leads to a time-dependent accumulation of a more slowly migrating form of the enzyme; when treated with endoglycosidase H and then subjected to Western blotting, only a single band (M r ϳ 65,000; data not shown) was observed, indicating that the slower mobility species seen with KDEL huCOX-2 in Fig. 1B resulted from post-translational changes in N-glycosylation.
Based on previous studies of the degradation of native COX-2 in the presence of puromycin plus the ␣-1,2-mannosidase inhibitor kifunensine (29), we suspected that the time-dependent formation of the upper band seen with KDEL huCOX-2 was due to N-glycosylation of Asn-594. As a test of this concept, we analyzed an N594A KDEL huCOX-2. This mutant was also degradation-resistant, but no time-dependent accumulation of a less mobile form was apparent. Finally, we found that the formation of the slowly moving species observed with KDEL huCOX-2 was retarded by tunicamycin (Ն5 M), which by inhibiting N-acetylglucosamine-1-phosphate transferase retards N-glycosylation (48) (Fig. 1D). These results indicate that the less mobile form of KDEL huCOX-2 results from the N-glycosylation of Asn-594. The results also establish that in the case of KDEL huCOX-2 Asn-594 glycosylation and proteasomal degradation are not directly coupled. This is consistent with our previous conclusion that glycosylation of Asn-594 is necessary but not sufficient to enable proteasomal degradation of COX-2 (29). The present results and those reported previously (29) also suggest that glycosylation of Asn-594 of both native and KDEL huCOX-2 occur in the ER lumen in a relatively slow manner. Importantly, restricting the trafficking of huCOX-2 by introducing a strong ER retention signal markedly slows or prevents the degradation of huCOX-2 via the ERAD pathway.
Another mutant denoted Golgi-⌬STEL huCOX-2 ( Fig. 1A) was engineered to examine the effects of protein localization on the degradation of huCOX-2. ␤-1,4-Galactosyltransferase consists of a short N-terminal cytoplasmic domain, a transmembrane domain as well as a stem region, and a C-terminal catalytic domain that localize to the lumen side of Golgi complex. The intact transmembrane domain is responsible for its subcellular localization to the medial-and trans-Golgi complex (49). As shown in Fig. 1A, we replaced the catalytic domain of ␤-1,4galactosyltransferase with the huCOX-2 sequence lacking both its signal peptide and C-terminal STEL sequence. This construct was degraded much more rapidly (t1 ⁄ 2 ϳ 30 min) than native huCOX-2 (Fig. 1, B and C). As discussed below, a proteasome inhibitor (MG132) and lysosomal inhibitors (chloroquine and NH 4 Cl) attenuated the degradation of Golgi-⌬STEL huCOX-2. This suggests that the degradation of Golgi-⌬STEL huCOX-2 involving the proteasome may not involve the same route as that of native huCOX-2. Degradation of native huCOX-2 in HEK293 cells was not inhibited either by the lysosomal cysteine protease inhibitor E-64 or by NH 4 Cl or chloroquine treatments (Fig. 1E) consistent with previous results indicating that native huCOX-2 does not undergo lysosomal degradation (28,29).
The Effects of Protein Trafficking Inhibitors on Degradation of COX-2-We further investigated the role of protein trafficking in the degradation of huCOX-2 using protein trafficking inhibitors. Trafficking between the ER and Golgi is mainly mediated by two sets of cytosolic proteins. COPII-coated vesicles are utilized in anterograde ER to Golgi transport, and COPI-coated vesicles mediate retrograde Golgi to ER transport (50). H-89 and AlF 4 Ϫ disrupt protein trafficking by impeding the functions of different components of COPI and COPII vesicles. H-89, a potent PKA inhibitor, is thought to prevent Sar1 from interacting with membranes and blocks anterograde trafficking (51). AlF 4 Ϫ is a general activator of G proteins and inhibits the Golgi to ER protein trafficking by preventing dissociation of COPI coat proteins from the Golgi membrane (52).
H-89 and AlF 4 Ϫ each slowed the rate of degradation of native huCOX-2 heterologously expressed in HEK293 cells ( Fig. 2A). Following induction of huCOX-2 expression and subsequent incubation with puromycin for 8 h, the percentage of huCOX-2 remaining was 30% in the absence of drugs, 80% with H-89, and 60% with AlF 4 Ϫ . These trafficking inhibitors were as effective as MG132, a proteasome inhibitor, in retarding COX-2 degradation. As noted above (Fig. 1E), degradation of native huCOX-2 in HEK293 cells was not blocked by inhibitors of lysosomal proteolysis.
The inhibitory effects on huCOX-2 degradation in HEK293 cells by trafficking inhibitors were dose-dependent (data not shown). The effective concentration for H-89 was 20 M, which is about 3 orders of magnitude greater than the K i value of H-89 with PKA (0.048 M) (51), suggesting that the target(s) of H-89 in this process is not PKA. These pharmacologic data suggest that the trafficking of huCOX-2 in HEK293 cells is essential for its efficient degradation via the ERAD pathway and that both COPII-and COPI-coated vesicles are involved in the trafficking of huCOX-2.
Next, we examined the effects of inhibitors on degradation of endogenously expressed muCOX-2 in NIH 3T3 cells using a commercial antibody (Novus Biologicals). The results clearly show that H-89 and MG132 block the degradation of muCOX-2 (Fig. 2B).
We also tested the effect of H-89 on the degradation of KDEL huCOX-2 and Golgi-⌬STEL huCOX-2 expressed heterologously in HEK293 cells. H-89 did not affect the degradation of KDEL huCOX-2 (data not shown). Similarly to what was observed with native huCOX-2, MG132 partially stabilized Golgi-⌬STEL huCOX-2 as did the lysosomal inhibitors chloroquine and NH 4 Cl (data not shown). These latter results suggest that Golgi-⌬STEL huCOX-2 is a substrate for both proteasomal and lysosomal degradation. In summary, evidence from COX-2 in the Golgi Apparatus FEBRUARY 27, 2015 • VOLUME 290 • NUMBER 9 studies with several relatively nonspecific protein trafficking inhibitors indicates that organellar trafficking is involved in the degradation of COX-2.
Effects of a Dominant Negative Sar1 on COX-2 Protein Degradation-As an additional test of the role of protein trafficking in COX-2 degradation, we examined the degradation of huCOX-2 in HEK293 cells transiently transfected with the dominant negative H79G Sar1 mutant. The core layer of COPII involved in anterograde transport consists of Sar1 and Sec23-24 and Sec13-31 complexes. These are sequentially recruited to the ER membrane to form COPII vesicles. Sar1, a small GTPase, plays a central role in assembling and disassembling the COPII coats. In the studies depicted in Fig. 3, transiently expressed VSV-G-tagged Sar1 and H79G Sar1 were examined for their abilities to affect COX-2 degradation. Dominant negative H79G Sar1 is a constitutively active, GTP-bound form of Sar1. Although efficiently recruited to ER membrane to form COPII vesicles, the inability of H79G Sar1 to hydrolyze GTP hampers the disassembly of COPII vesicles. This results in inhibition of ER to Golgi transport and disruption of ERGIC and Golgi compartments (53). Compared with the control experiment with empty plasmid, overexpression of native Sar1 had relatively little effect on COX-2 degradation (Fig. 3). More than 60% of the huCOX-2 was degraded within 8 h after transfection with vector or native Sar1. There was slightly less COX-2 at 0 h in cells treated with native Sar1, suggesting that excess Sar1 can promote COX-2 degradation. In cells treated with dominant negative H79G Sar1, the rate of huCOX-2 degradation was slowed significantly, and about 80% of the huCOX-2 remained after 8 h. Similar levels of VSV-G-labeled Sar1 and VSV-G-labeled H79G Sar1 were expressed under the conditions of the experiment. This experiment indicates that disrupting protein anterograde ER to Golgi transport affects the degradation of huCOX-2 occurring via the ERAD pathway and that trafficking of huCOX-2 utilizes the COPII coat system. Ϫ are statistically significant (p Ͻ 0.05) from zero time to 24 h. The AlF 4 Ϫ -treated group differs from the control group at 8 and 24 h (p Ͻ 0.05). These differences are denoted with asterisks. B, effects of inhibitors on the degradation of endogenous native muCOX-2 in murine NIH 3T3 fibroblasts. NIH 3T3 cells were cultured to express COX-2 as detailed under "Experimental Procedures." After the cells were treated with inhibitors for 1 h, cycloheximide (50 M) was added to block translation along with the amounts of inhibitors indicated in A above. Cells were collected at the indicated times, and proteins were subjected to Western blotting. The densitometry measurements of muCOX-2 are based on two separate experiments. The primary antibody used for the experiments with murine NIH 3T3 cells was a commercial antibody from Novus Biologicals. Data are shown as mean Ϯ S.D. (error bars). Based on repeated measures of ANOVA (IBM SPSS Statistics 21), the differences between the control group without drug treatment and the MG132-and H-89-treated groups are statistically significant (p Ͻ 0.05) from 0 to 24 h. These differences are denoted with asterisks.

Subcellular Location of COXs and mPGES-1 as Determined by Confocal
Microscopy-To examine the concept that COX-2 can migrate to and reside in the Golgi apparatus, the subcellular location of COX-2 was investigated by confocal fluorescence microscopy. Giantin, a highly conserved cis-and medial-Golgiresident protein, was used as the Golgi marker, and calnexin was used as the ER marker. For studies with HEK293 cells, we tested COX-2 mutants that we expected to be primarily localized to either the ER or the Golgi for comparison. HEK293 cells are not ideal for immunocytofluorescence studies, and so we performed parallel studies with several fibroblast lines that endogenously express COX-1 and/or COX-2 and in which the subcellular structures appear in better definition.
As shown in Fig. 4A, image overlay data indicate that a part of native huCOX-2 expressed heterologously in HEK293 cells is localized in the Golgi apparatus. Similarly, in murine NIH 3T3 fibroblasts, part of the endogenous muCOX-2 also co-localized with giantin closely neighboring the nucleus; the pattern of intense perinuclear staining is quite similar to that observed previously using normal phase immunocytofluorescence (24). Golgi-⌬STEL huCOX-2 in HEK293 cells was also largely if not exclusively co-localized with giantin, whereas neither KDEL huCOX-2 nor N594A huCOX-2 co-localized with giantin (Fig.  4A). The distribution of huCOX-1 and ovCOX-1 expressed in HEK cells and endogenous huCOX-1 in CCL210 cells was also investigated by confocal microscopy (Fig. 4B). The COX-1 forms examined did not exhibit significant overlap with giantin but did overlap with the ER marker calnexin. We emphasize that the majority of both COX-1 and COX-2 staining was associated with the ER and nuclear envelope consistent with much previous work (19 -23). This is illustrated in the bottom panel of Fig. 4B that shows widespread staining throughout the cell with considerable overlap between calnexin staining and ovCOX-1 staining in HEK293 cells.
CCs were determined for giantin staining and the staining of various forms of endogenous and heterologously expressed COX-1 and COX-2 (Fig. 4C). Native muCOX-2 in 3T3 cells and native huCOX-2, Golgi-⌬STEL huCOX-2, and KDEL huCOX-2 in HEK293 cells exhibited CCs of 0.33, 0.31, 0.42, and Ϫ0.10, respectively (Fig. 4C). These data indicate that there is a significant difference between the subcellular locations of native huCOX-2 and Golgi-⌬STEL huCOX-2 compared with KDEL huCOX-2 in HEK cells and that native huCOX-2 expressed heterologously and native muCOX-2 expressed endogenously were similarly localized. ⌬STEL huCOX-2 with a CC of 0.09 does not closely co-localize with the giantin marker. However, only a low level of fluorescent staining was observed with this variant. We suspect that ⌬STEL huCOX-2 can move to the Golgi but, lacking both an ER retention signal and the Golgi anchoring sequence of Golgi-⌬STEL huCOX-2, then moves on through the secretory pathway. CCs for the Golgi marker with ovCOX-1 and huCOX-1 in HEK cells were 0.03 and 0.05, respectively, whereas the value for endogenously expressed huCOX-1 in CCL210 cells was 0.02. Again, these numbers are significantly lower for COX-1 than those for native huCOX-2 and muCOX-2 and occurred across several cell types with heterologously expressed and endogenous COX-1.
Interestingly, the CC for N594A huCOX-2 and giantin in HEK293 cells is 0.06, indicating that N594A huCOX-2 is not localized in the Golgi. This observation indicates that posttranslational glycosylation of Asn-594 occurs in the ER lumen, that glycosylated Asn-594 is an ER exit signal, and that variant forms of COX-2 having mutations that prevent N-glycosylation of Asn-594 become lodged in the ER. Overall, our confocal microscopy data support the hypothesis that a significant fraction of COX-2 but not COX-1 resides in the Golgi apparatus.

p C D N A 3 S a r 1 -p C D N A 3 H 7 9 G S a r 1 -p C D N A 3 Time (h)
A.

JOURNAL OF BIOLOGICAL CHEMISTRY 5613
We speculate that because COX-1 lacks an efficient ER to Golgi trafficking signal (i.e. glycosylated Asn-594) this isoform resides primarily in ER.
Finally, we examined the subcellular location of human mPGES-1 expressed heterologously in HEK293 cells and endogenous mPGES-1 in human dermal fibroblasts (Fig. 4D).   Microscopy was conducted on a Nikon Infinity confocal microscope at a magnification of 100ϫ. B, co-localization of huCOX-1 in HEK293 cells, ovCOX-1 in HEK293 cells, and huCOX-1 in CCL210 human lung fibroblasts with Golgi (giantin) and ER (calnexin) markers. For CCL210 cells, the slides were directly fixed and prepared 1 day after seeding the cells. The protocols used for staining the HEK293 cells were as described in A above. The primary antibodies used were mouse monoclonal anti-COX-1 and either rabbit polyclonal anti-giantin or rabbit polyclonal anti-calnexin. C, CC values for co-localization of different COX-1 and COX-2 variants with the Golgi marker giantin were determined as detailed under "Experimental Procedures". Based on the one-way ANOVA for multiple comparisons (GraphPad Prism), a difference between CC values for huCOX-2 in HEK293 cells, muCOX-2 in 3T3 cells, and Golgi-⌬STEL huCOX-2 in HEK293 cells and the CC values for COX-1 and other COX-2 variants are statistically significant (p Ͻ 0.05) as denoted with asterisks (*). No significant difference was observed between huCOX-2 in HEK293 cells and muCOX-2 in 3T3 cells. D, co-localization of mPGES-1 with the Golgi marker giantin and the ER marker calnexin in human dermal fibroblasts (HDFn). Human dermal fibroblasts were cultured to express endogenous huCOX-2, and immunocytofluorescence was performed as described under "Experimental Procedures". A mouse monoclonal antibody to mPGES-1 was used as the primary antibody.
The patterns of staining are similar to those seen for huCOX-2 in HEK293 cells and muCOX-2 in NIH 3T3 cells. Specific Activities of Native huCOX-2, KDEL huCOX-2, and Golgi-⌬STEL huCOX-2-Although COX-2 protein was present in different organelles, it was not clear whether the mutant COX-2 variants were functional. To address this question, COX assays were performed by incubating 18 M [1-14 C]AA with lysates prepared from HEK293 cells expressing native huCOX-2, KDEL huCOX-2, or Golgi-⌬STEL huCOX-2. These particular HEK293 cell lines have very low levels of mPGES-1. After formation via COX activity, PGH 2 rearranges non-enzymatically to a number of products. These products were separated by HPLC and quantified by scintillation counting. PGH 2derived products (i.e. PGs and the mono-oxygenated product 17-hydroxyeicosatetraenoic acid) were generated by lysates prepared from each cell line, and product formation was inhibited by ibuprofen, a common COX inhibitor. COX-2 protein levels in the lysates were estimated using Western blotting and densitometry comparing the staining intensities of the COX-2 in cell lysates with those of known amounts of purified recombinant huCOX-2 (16). Specific activities for huCOX-2, KDEL huCOX-2, and Golgi-⌬STEL huCOX-2 were 13, 12, and 17 mol of AA consumed/min/mg of COX-reactive protein, respectively. These values, which were determined with 18 M AA, agree reasonably well with the V max value of ϳ21 mol AA/min/mg determined for purified, recombinant huCOX-2 given that the K m for purified huCOX-2 is ϳ10 M (16). Importantly, Golgi-⌬STEL huCOX-2, most if not all of which appears to be localized to the Golgi apparatus (Fig. 4A), exhibited high levels of COX activity. Our quantitative measures of COX-2 activity and protein indicate that the membrane environment of the different organelles does not change the overall structure of COXs and importantly that COX-2 present in the Golgi apparatus is functional.
As seen in the Western transfer blot shown in Fig. 5A, the five lines tested co-expressed cPLA 2 and mPGES-1 along with huCOX-1, Golgi-⌬STEL huCOX-1, huCOX-2, KDEL huCOX-2, or Golgi-⌬STEL huCOX-2. cPLA 2 was expressed at similar levels by all the cell lines. mPGES-1 was expressed by all the lines but at a somewhat lower level in the cell line expressing native huCOX-2. Different antibodies were used to detect COX-1 and COX-2 so no direct comparisons of the amounts of COX-1 and COX-2 protein can be made from comparing the Western blots  FEBRUARY 27, 2015 • VOLUME 290 • NUMBER 9

COX-2 in the Golgi Apparatus
in Fig. 5A. We also examined the PG product profiles for each of the five cell lines. When incubated with exogenous [1-14 C]AA (12 M) for 10 min, the only significant PG product in each case was PGE 2 as determined using HPLC with PG standards (14) (data not shown). This indicates that there is sufficient mPGES-1 activity in each cell line to convert any COX-derived PGH 2 to PGE 2 . A 10-fold range of COX-specific activities was observed with the different lines (Fig. 5B): KDEL huCOX-2 Ͼ huCOX-1 Ͼ huCOX-2 Ͼ Golgi-⌬STEL huCOX-2 Ͼ Golgi-⌬STEL huCOX-1. These differences reflect differences in the amounts of catalytically active COX protein because COX-1 and COX-2 have similar kinetic constants with AA as substrate (7).
Finally, we compared PGE 2 production by each of the cell lines following treatment with excess exogenous AA (12 M) versus a relatively low concentration of the Ca 2ϩ ionophore A23187 (2 M) to mobilize a low concentration of endogenous AA (Fig. 5B). A23187 elicits mobilization of endogenous AA for PG production by activating Ca 2ϩ -dependent PLA 2 s including cPLA 2 . Also enumerated in Fig. 5B are the ratios of PGE 2 formed from cells treated with 2 M A23187 versus cells treated with exogenous 12 M AA (ϫ10 Ϫ3 ). Because similar amounts of cPLA 2 are present in the cells (Fig. 5A), these values indicate the relative abilities of the different COX forms to generate PGE 2 from endogenous AA relative to exogenous AA. Cells expressing mPGES-1 in combination with native huCOX-2, Golgi-⌬STEL huCOX-2, or Golgi-⌬STEL huCOX-1 all exhibited significantly higher levels of PGE 2 formation from endogenous versus exogenous AA than COX variants that are concentrated in the ER (i.e. huCOX-1 or KDEL huCOX-2). It should be noted that treatment of the cell lines with the COX inhibitor flurbiprofen 1 h prior to either the addition of exogenous 12 M AA or treatment with 2 M A23187 caused in each case Ͼ95% inhibition of PGE 2 formation (data not shown). Moreover, treatment with the cPLA 2␣ inhibitor pyrrophenone 1 h prior to the addition of 2 M A23187 caused 80% inhibition of PGE 2 formation; however and as expected, pyrrophenone did not interfere with PGE 2 formation from exogenous AA. The results in Fig. 5 indicate that PGE 2 formation from endogenous AA occurs more efficiently with COX variants present in the Golgi apparatus than in the ER. The results suggest that cPLA 2␣ , COX-2, and mPGES-1 in the Golgi provide an efficient system for PGE 2 formation.

DISCUSSION
Overall, the results presented in this report suggest that there is a dedicated system for COX-2-mediated PGE 2 synthesis localized in the Golgi apparatus of cells. This system could account for the longstanding enigma of how COX-2 can function under conditions in which COX-1 is relatively inoperative when both COX isoforms are expressed in cells (54). Our conclusions emerged from studies of COX-2 protein degradation, so we first discuss the work in the context of COX-2 as an unusual substrate of ERAD.
Degron-mediated ERAD of COX-2 Involves ER to Golgi to ER Trafficking-The ERAD pathway is often regarded as a quality control mechanism for disposing of misfolded, wrongly glycosylated, or unassembled ER proteins (for recent reviews, see Refs. 36 -40). However, ERAD can also play a role in the disposal of correctly folded ER proteins (39). For instance, to avoid detection by the immune system, human cytomegalovirus uses US11 and US2 proteins as adaptors to deliver correctly folded MHC class I heavy chains to the ERAD system (55). Another example is the elegant regulation of 3-hydroxy-3-methylglutaryl-CoA reductase by physiological sterol concentrations. In mammalian cells, high sterol levels elicit structural changes in 3-hydroxy-3-methylglutaryl-CoA reductase, and the then pseudo-unfolded enzyme can bind to adaptor proteins Insig-1 and Insig-2 to promote ERAD (56,57).
The aforementioned ERAD target proteins are transmembrane proteins. In contrast, COXs are integral membrane proteins that are embedded in a single layer of the lipid bilayer through a membrane binding domain composed of four amphipathic helices (7). A large, soluble catalytic domain extends into the vesicle lumen. The degradation of apparently intact COX-2 via ERAD involves a modular degron located near the C terminus of the catalytic domain (28,29). COX-2 with its intrinsic degron represents a relatively new category of ERAD substrates in mammalian proteostasis.
Central conclusions of our study are illustrated in Fig. 6, which depicts degron-coupled ERAD of COX-2. Key features of this model are as follows. Newly translated and folded COX-2 is post-translationally glycosylated at Asn-594 in the ER lumen and then deglucosylated by glucosidases I and II to generate an ER departure signal (36,39), presumably the Man 9 (GlcNAc) 2 -Asn-594 moiety (58)(59)(60)(61). We propose that this Asn-594-linked oligosaccharide binds to a member of the Ca 2ϩ -dependent L-type family of lectins such as ERGIC-53, VIPL (vesicular-integral membrane protein 36-like protein), or VIP-63 in conjunction with an appropriate adapter protein such as MCFD-2 (60), leading in turn to packaging COX-2 into COPII vesicles (36). The structure of the COX-2 ER outing signal is not known but presumably involves the Asn-594-linked carbohydrate and perhaps an exposed ␤-hairpin loop between Val-572 and Thr-588 (29) analogous to that of procathepsin Z (58). COX-2 then undergoes anterograde transport to the Golgi apparatus. DatasupportingERtoGolgitraffickingofCOX-2include(a)previous evidence for post-translational N-glycosylation of Asn-594 (28,29) and now the demonstration that post-translational N-glycosylation of KDEL huCOX-2 occurs in the ER, (b) the inhibition of COX-2 degradation by a dominant negative Sar1 and H-89, and (c) the detection of significant levels of COX-2 in the Golgi by confocal immunofluorescence. This latter result is also consistent with a recent report of COX-2 in the Golgi apparatus visualized using a fluorescent active site probe for COX-2 (62). We envision that the weak ER retention sequence STEL (45) of COX-2 that is present in all mammalian COX-2s sequenced to date (7) provides for a relatively extended residence time in the Golgi. This permits ER mannosidase-1 in the Golgi (43) time to remove ␣1,2-linked mannose residues from the B chain of the Man 9 (GlcNAc) 2 -Asn-594 moiety, generating a Man 8 (GlcNAc) 2 -Asn-594 species; we assume that this has no effect on enzyme activity as N594A COX-2 mutants have the same activity as native COX-2 (42). Eventually, the modified but still catalytically active COX-2 undergoes retrograde transport to the ER via a COPI vesicle in a process inhibited by AlF 4 Ϫ . We speculate that once returned to the ER the Man 8 (GlcNAc) 2 -Asn-594 group is trimmed on the C-chain perhaps through the action of EDEM3 to Man 7 (GlcNAc) 2 -Asn-594 (38,44). Finally, in events involving EDEM, ERp57, OS-9, XTP3-B, and the HRD1-SEL1L ubiquitin ligase complex (39,44), a trimmed and unfolded COX-2 is translocated to the cytosol for ubiquitination and proteasomal degradation (25,28,29,32,35). The slow glycosylation observed with KDEL huCOX-2 suggests that the rate-limiting step in ERAD of COX-2 is post-translational glycosylation of Asn-594 in the ER. The rate of post-translational N-glycosylation of Asn-594 of COX-2 could be subject to control, which could in turn regulate the rate of both COX-2 ERAD and COX-2-mediated PGE 2 formation in the Golgi.
In the model proposed in Fig. 6, the proper folding and posttranslational N-glycosylation of COX-2 in the ER is spatially separate from the ER mannosidase-1-mediated event in the Golgi that eventually leads to ERAD. This spatial segregation eliminates competition between COX-2 protein folding and degradation. The model can explain the observation that N-glycosylation of Asn-594 is necessary but not sufficient for ERAD of COX-2 (29). The process differs from that involving co-translational N-glycosylation, which is most commonly associated with glycoprotein degradation via ERAD.
Finally, with respect to degradation, we note that in addition to ERAD COX-2 is subject to substrate turnover-induced degradation (29,30). This latter process does not require N-glycosylation of Asn-594. We speculate that substrate turnover-induced degradation is a backup degradation system for the COX-2 that accumulates in the ER but is (a) not N-glycosylated for transport to the Golgi for PGE 2 synthesis and/or (b) not needed for synthesizing other PG products in the ER. Substrate turnover-induced degradation of COX-2 would be expected to occur in the presence of excess substrate, a condition favoring COX-1 activity. In cells co-expressing both isoforms, COX-2 is typically in much lesser abundance than COX-1 (63).
A Golgi cPLA 2␣ /COX-2/mPGES-1 System for PGE 2 Synthesis-Reddy and Herschman (54) were the first to demonstrate that when COX-1 and COX-2 are co-expressed in cells that COX-2 can function with endogenously mobilized AA, whereas COX-1 is effectively latent; however, when treated with a bolus of exogenous AA, both COX-1 and COX-2 are operative. Several explanations have been proffered for these observations. One is simply that the effective concentration of AA released endogenously is significantly less than that when cells are treated with a bolus of exogenous AA and that COX-2 has a slightly lower K m than COX-1 for AA (64). A related possibility is that COX-2 operates within an endogenous pool of non-substrate fatty acids that through allosteric mechanisms favors COX-2 over COX-1 activity (14 -17). Additionally or alternatively, COX-2 has a lesser hydroperoxide requirement than COX-1 for its activation (6), and the concentration of activating hydroperoxides present following COX-2 induction may be too low to activate COX-1. These kinetic explanations are difficult to test in intact cells, and there may be other explanations for the preferential use of endogenous AA by COX-2. Our studies suggest one alternative: that COX-1 and COX-2 can operate independently in cells because they function at least in part in geographically separate compartments.
Our results suggest that ER to Golgi transport provides COX-2 to a compartmentalized cPLA 2␣ /COX-2/mPGES-1 system for FIGURE 6. Model of COX-2 synthesis, trafficking and ERAD that provides for COX-2-mediated PGE 2 biosynthesis in the Golgi apparatus. A, translation and entry of COX-2 into the lumen of the ER during which COX-2 is co-translationally N-glycosylated at Asn-68, Asn-144, and Asn-410. B, post-translational N-glycosylation of Asn-594. C, trimming by glucosidases I and II to generate an ER outing signal. D, incorporation of COX-2 into a COPII vesicle, anterograde transport to the cis-Golgi apparatus, and transfer to the lumen of the cis-Golgi. E, trimming of the Asn-594-linked carbohydrate moiety by ER mannosidase-1 (ERMan-1) to yield a COX-2 available for retrograde transport. F, retrograde transport of COX-2 to the ER involving the KDEL receptor interacting with the weak ER retention sequence STEL at the C terminus of COX-2. G, further processing of the Asn-594-linked group of modified COX-2 perhaps to Man 7 (GlcNAc) 2 -Asn-594. H, ubiquitination and transport to the cytosol for proteasomal degradation. I, COX-1 localized to the lumen of the ER. J, formation of a COX-2-mediated PGE 2 synthetic system involving cPLA 2␣ translocated in a Ca 2ϩ -dependent manner from the cytosol to the surface of the Golgi, COX-2 in the Golgi apparatus, and Golgi-localized mPGES-1. FEBRUARY 27, 2015 • VOLUME 290 • NUMBER 9 PGE 2 synthesis. We found COX-2 but little or no COX-1 in the Golgi apparatus. The reason is that only COX-2 has an ER to Golgi trafficking signal. Furthermore, we found that mPGES-1 is also located in the Golgi apparatus. Finally, another key enzyme, cPLA 2 , which plays a central role in regulating AA release for PGE 2 synthesis in response to intracellular Ca 2ϩ mobilization, binds to Golgi in preference to ER membranes at lower cytosolic Ca 2ϩ concentrations (46). According to our model (Fig. 6), part of the cellular COX-2 resides in the Golgi and would have access to this endogenous pool of AA when it is mobilized by cPLA 2 at low intracellular Ca 2ϩ levels (i.e. when only low levels of endogenous AA are being released). In contrast, COX-1 resides primarily if not exclusively in the ER where AA is not accessible to COX-1 at lower cytosolic Ca 2ϩ concentrations. PGE 2 formed in the Golgi would be expected to be produced at low levels relative to that produced via a burst of COX-1 action in the ER (54). This is because there is almost always less COX-2 than COX-1 (63), and as noted earlier and discussed further below, most of the COX-2 appears to be in the ER rather than the Golgi. Accordingly, we speculate that PGE 2 produced via COX-2 in the Golgi would appear at low concentrations extracellularly. This would favor autocrine rather than paracrine actions of PGE 2 .

COX-2 in the Golgi Apparatus
We note that there are previous reports that cPLA 2␣ and COX-2 co-localize with caveolae in fibroblasts (65) and in lipid droplets in many cell types (66). Additionally, there is a significant body of evidence indicating that cPLA 2␣ and COX-2 function coordinately (67)(68)(69). COX-2 and mPGES-1 are also functionally related (for a recent review, see Ref. 7). There is a previous report of perinuclear mPGES-1 immunocytofluorescence in human chondrocytes that partially overlaps with COX-2 fluorescence (68), and this could result from the staining of mPGES-1 in the Golgi.
It is important to recognize that most of the COX-2 staining observed in our studies was present in the ER and not the Golgi apparatus (Fig. 4). Unfortunately, there are no robust methods for quantifying the relative amounts of COX-2 (or cPLA 2␣ or mPGES-1) in the Golgi versus the ER of cultured cells. COX-2 appears to be only slowly glycosylated on Asn-594, and thus, trafficking of COX-2 to the Golgi from the ER must be equally slow. This is an important issue because COX-2 may play important roles in both the ER and the Golgi. COX-2 in the ER could serve to augment COX-1-mediated PG formation at relatively higher cytosolic Ca 2ϩ concentrations when cPLA 2 is translocated to the ER as well as the Golgi (46,70). COX-2 in the ER and nuclear envelope also could be involved in generating another PG product(s) (e.g. PGI 2 ). Further studies are needed to examine the relationship between cell activation in, for example, immune cells and COX-2 trafficking to the Golgi.