Two Distinct Pathways for Cyclooxygenase-2 Protein Degradation*

Cyclooxygenases (COX-1 and COX-2) are N-glycosylated, endoplasmic reticulum-resident, integral membrane proteins that catalyze the committed step in prostanoid synthesis. COX-1 is constitutively expressed in many types of cells, whereas COX-2 is usually expressed inducibly and transiently. The control of COX-2 protein expression occurs at several levels, and overexpression of COX-2 is associated with pathologies such as colon cancer. Here we have investigated COX-2 protein degradation and demonstrate that it can occur through two independent pathways. One pathway is initiated by post-translational N-glycosylation at Asn-594. The N-glycosyl group is then processed, and the protein is translocated to the cytoplasm, where it undergoes proteasomal degradation. We provide evidence from site-directed mutagenesis that a 27-amino acid instability motif (27-IM) regulates posttranslational N-glycosylation of Asn-594. This motif begins with Glu-586 8 residues upstream of the N-glycosylation site and ends with Lys-612 near the C terminus at Leu-618. Key elements of the 27-IM include a helix involving residues Glu-586 to Ser-596 with Asn-594 near the end of this helix and residues Leu-610 and Leu-611, which are located in an apparently unstructured downstream region of the 27-IM. The last 16 residues of the 27-IM, including Leu-610 and Leu-611, appear to promote N-glycosylation of Asn-594 perhaps by causing this residue to become exposed to appropriate glycosyl transferases. A second pathway for COX-2 protein degradation is initiated by substrate-dependent suicide inactivation. Suicide-inactivated protein is then degraded. The biochemical steps have not been resolved, but substrate-dependent degradation is not inhibited by proteasome inhibitors or inhibitors of lysosomal proteases. The pathway involving the 27-IM occurs at a constant rate, whereas degradation through the substrate-dependent process is coupled to the rate of substrate turnover.

multiply N-glycosylated ER-resident proteins that exist as homodimers and exhibit ϳ60% primary structure identity (1)(2)(3)(4)(5). They are also integral membrane proteins that insert into one face of the lipid bilayer and are largely compartmentalized in the ER lumen and the contiguous lumen of the nuclear envelope (6 -8). The mature forms of the COX isoforms are very similar in structure except that COX-2 has a unique 19-residue insertion (19-aa; residues Asn-594 to Lys-612) near its C terminus.
The principal endogenous fatty acid substrate of COX-1 and COX-2 is arachidonic acid (AA), which is mobilized from the sn-2-position of membrane phospholipids upon the activation of phospholipase A 2 s through the actions of bradykinin, thrombin, growth factors, calcium ionophore (A23187), or cytokines (1, 2, 9 -16). The COX isoforms have two catalytic activities that occur at two distinct active sites. At the COX active site, AA is oxygenated to form prostaglandin endoperoxide G 2 (1)(2)(3)17). Prostaglandin endoperoxide G 2 then moves to the peroxidase site, where its hydroperoxy group undergoes a two-electron reduction to form prostaglandin endoperoxide H 2 (1)(2)(3)17). Prostaglandin endoperoxide H 2 is a common substrate for downstream terminal prostanoid synthases that form various prostanoids.
COX-1 is a stable protein that is constitutively expressed in resting cells of many tissues, notably in platelets and vesicular gland (1, 18 -20). In contrast, COX-2 is a stimulus-inducible protein whose expression is short-lived in epithelial, endothelial, smooth muscle, and fibroblast cells (18,(21)(22)(23)(24)(25). The short half-life of COX-2 protein mimics that of COX-2 mRNA. Both COX-2 mRNA and protein have been found to possess instability elements that target them for rapid degradation. At the 3Ј-untranslated region of COX-2 mRNA are multiple AUUUA elements that target mRNAs for rapid exonuclease cleavage (21,26). COX-1 mRNA lacks these 3Ј-untranslated region AUrich elements, and COX-1 mRNA is stable in megakaryocytes and vascular endothelial cells (27)(28)(29). We have recently reported that the C-terminal 19-aa of COX-2 causes the enzyme to undergo proteasomal degradation via the endoplasmic reticulum-associated degradation (ERAD) pathway (18).
In the present study, we set out to examine what specific features of the C-terminal 19-aa, in addition to the Asn-594 glycosylation site, are critical for targeting the enzyme to the ERAD system. The role of the helical region upstream of the Asn-594 glycosylation site in regulating Asn-594 was also investigated. Our findings enabled us to identify a C-terminal 27-amino acid destabilizing motif  of COX-2 that regulates the glycosylation of the enzyme at Asn-594 and controls, at least in part, the timing and extent of its degradation.
In the course of our studies on COX-2 degradation, we discovered that N594A human COX (huCOX)-2, which lacks a functional 27-IM, was degraded rapidly when AA was added to cells expressing this mutant. This led us to examine the impact of COX catalysis on the protein stabilities of COX-1 and COX-2. We have found that the rate of COX-2 protein degradation is enhanced by AA in NIH/3T3 and HEK293 cells in a proteasome-independent manner. We provide evidence that substrate-dependent degradation of COX-2 requires a functional COX active site and proceeds from the substrate-induced inactive form(s) of the enzyme.
Cell Culture and Transfection-NIH/3T3 fibroblasts at early passage (Ͻ6 passages) were cultured in DMEM supplemented with 10% bovine calf serum and 100 units/ml penicillin/streptomycin. To induce COX-2 expression, the cells were first made quiescent by serum starvation for 48 h in DMEM containing 0.2% bovine calf serum and thereafter treated with DMEM supplemented with 20% FBS for 4 h.
RAW 264.7 macrophage-like cells were cultured in DMEM supplemented with 10% FBS and 100 units/ml penicillin/streptomycin. To stimulate COX-2 expression, the cells were challenged with 200 ng/ml LPS for 12 h (30).
Enyzmatic Deglycosylation-For complete deglycosylation, HEK293 whole cell lysates were denatured by boiling in NuPAGE SDS sample loading buffer (Invitrogen) and then treated for at least 12 h with endoglycosidase H at a concentration of 0.4 milliunits/l.
Western Transfer Blotting-After the appropriate treatments, NIH/3T3, RAW 264.7, and HEK293 cells were scraped into ice-cold phosphate-buffered saline, pH 7.4, containing 5 mM EDTA and a mixture of protease inhibitors (Roche Applied Science) and lysed by sonication. Radioimmune precipitation lysis buffer (10 mM Tris, pH 7.2, 150 mM NaCl, 5 mM EDTA, 0.1% SDS, 1.0% Triton X-100, 1% deoxycholate) containing a protease inhibitor mixture (Roche Applied Science) was also used for cell lysis. Protein concentrations were determined using the BCA protein assay kit (Pierce). The NuPAGE system (Invitrogen) was used to resolve the proteins in the whole cell lysates on a 7% Tris acetate polyacrylamide gel. After transfer to a nitrocellulose membrane, immunoblotting was performed with the appropriate primary antibody. Horseradish peroxidase-conjugated anti-rabbit or anti-mouse IgG antibodies (Bio-Rad) were used as secondary antibodies. Immunodetection was performed using the Western Lighting Chemiluminescent kit (Amersham Biosciences) followed by exposure to x-ray film. Densitometry analysis was performed using Quantity One software (Bio-Rad).
Cyclooxygenase Inactivation Assay-Cultured Flp-In T-Rex HEK293 cells stably expressing native ovCOX-1 or His 6 muCOX-2 were incubated for 10 min with different concentrations of AA ranging from 2.5 to 50 M. After incubation, the cells were washed with 1ϫ phosphate-buffered saline and harvested by centrifugation at 2000 rpm for 3 min. Cell pellets were resuspended in 0.1 M Tris, pH 8.0, buffer containing 5 mM EDTA and a protease inhibitor mixture and lysed by sonication. Protein concentrations in the lysates were determined by BCA assay. COX activity assays were performed at 37°C by monitoring the initial rate of oxygen uptake using an oxygen electrode as described previously (18). Reactions were initiated by adding cell lysate to the assay chamber containing 3 ml of 0.1 M Tris-HCl, pH 8.0, 1 mM phenol, 5 M hematin, and 100 M AA. In some cases, microsomes were prepared from the cell lysates and used to quantify the synthesis of radiolabeled prostaglandins from [1-14 C]AA by radio thin layer chromatography using procedures described in detail previously (32).

Effects of Mutations of 19-aa on COX-2 Degradation-
The domain structures of COX-1 and COX-2 and the sequences of the C termini of native and mutant COXs are shown in Fig. 1A. We have previously shown that the Asn-594 glycosylation site is necessary, but not sufficient, to enable proteasomal degradation of COX-2 (18). Inserting the COX-2 19-aa cassette near the C terminus of ovCOX-1 also yields a mutant ins594 -612 ovCOX-1 that is unstable (t1 ⁄ 2 ϳ 3 h) (18). However, simply inserting a glycosylation site near the C terminus of ovCOX-1 (ins594 -596 ovCOX-1 or ins595-597 ovCOX-1) does not destabilize this COX isoform (Fig. 1, B and C). Both ins594 -596 ovCOX-1 and ins595-597 ovCOX-1 are variably glycosylated ( Fig. 1C) (18). Inhibition of protein synthesis causes a time-dependent accumulation of the more highly glycosylated forms of these two mutants, suggesting that glycosylation occurs posttranslationally at the additional N-glycosylation site (Fig. 1C). These observations indicate that C-terminal N-glycosylation of ins594 -596 ovCOX-1 or ins595-597 ovCOX-1 is insufficient to promote their degradation and suggest that in the case of COX-2, a segment of the 19-aa in addition to the Asn-594 glycosylation sequence is needed for degradation to occur.
We used site-directed mutagenesis to identify which of the 16 amino acids downstream of the consensus N-glycosylation sequence (Asn-594 to Ser-596) of the C-terminal 19-aa are involved in COX-2 degradation. The data on rates of protein degradation are summarized in Fig. 1B. Two C-terminal deletion mutants of huCOX-2 (i.e. del602-612 huCOX-2 and del607-612 huCOX-2) and an insertion mutant of ovCOX-1 (i.e. ins594 -601 ovCOX-1) that has the region of 19-aa identical to that of del602-612 huCOX-2 were not degraded at a detectable rate (Fig. 1B). Thus, removal of the last 6 residues of 19-aa prevents COX-2 degradation. The fact that COX-2 degradation was prevented by trimming the 19-aa raised the possibility that it was the length of the 19-aa cassette rather than its primary structure that was important in COX-2 degradation. To test this possibility, we prepared two mutants of COX-2 whose 19-aa sequences had been scrambled (Fig. 1B). Sins597-612 huCOX-2 has an intact Asn-594 glycosylation site, but the rest of the 19-aa has been scrambled. Sins595-612huCOX-2 is not only mutated at Asn-594 but also has a randomized version of the remainder of the 19-aa. As summarized in Fig. 1B, both mutants were much more stable (t1 ⁄ 2 Ͼ 24 h) than native huCOX-2 (data not shown). These results suggested that there is primary structure information within the C-terminal 16amino acid segment of 19-aa, perhaps within the last six residues of the insertion, that is essential for mediating COX-2 protein degradation.
To identify which of the last 6 amino acids of 19-aa were important for COX-2 degradation, we performed alanine scanning mutagenesis of this region. The mutants P607A/T608A huCOX-2 and V609A huCOX-2 had a t1 ⁄ 2 of ϳ8 h, which was somewhat longer than that of native enzyme (t1 ⁄ 2 ϳ 5 h) (Fig. 1, D and E). Mutation of the highly conserved Lys-612 residue at the C terminus of 19-aa did not prevent COX-2 protein turnover (t1 ⁄ 2 ϳ 3 h) (Fig. 1, D and E). In contrast, both L610A huCOX-2 and L611A huCOX-2 were degraded at significantly slower rates than native COX-2; the t1 ⁄ 2 for both mutants was Ͼ24 h (Fig.  1, D and E). Collectively, these results suggest that the two adjacent large hydrophobic residues near the C terminus of the 19-aa of COX-2 are important for enabling COX-2 degradation.
Glycosylation of Asn-594 Correlates with COX-2 Protein Degradation-KIF inhibits ERAD by preventing N-glycan processing by ER ␣1,2 mannosidase I. We previously reported that KIF stabilizes huCOX-2 heterologously expressed in HEK293 and causes the appearance of a less mobile, alternatively glycosylated form of the native enzyme (18). N594A huCOX-2 and native huCOX-2 have identical electrophoretic mobilities on SDS-PAGE, but a more highly glycosylated form seen with native huCOX-2 is not observed after KIF treatment of HEK293 cells expressing N594A huCOX-2 (18). A parallel phenomenon is seen with NIH/ 3T3 and RAW 264.7 cells even after protein translation has been inhibited ( Fig. 2A). These results are consistent with the idea that KIF stabilizes a form of the enzyme that is post-translationally N-glycosylated at Asn-594.
The fact that treatment of cells expressing native COX-2 with KIF stabilizes the Asn-594-glycosylated form of COX-2 suggests that KIF can also be used to assess Asn-594 glycosylation of mutant forms of the enzyme. Accordingly, experiments were performed to determine if various deletion mutants (i.e. del597-612 huCOX-2, del602-612 huCOX-2, and del607-612 huCOX-2) and the scrambled mutant Sins597-612 huCOX-2 are glycosylated at Asn-594. As noted above, all of these mutations prevent COX-2 degradation (Fig. 1B). The deletion mutant del597-612 huCOX-2 has the same electrophoretic mobility as del595-612 huCOX-2, which lacks the Asn-594 N-glycosylation sequence, and KIF treatment of cells expressing del597-612 huCOX-2 did not affect the electrophoretic mobility of this mutant (data not shown); this sug-  gests that del597-612 huCOX-2 is not glycosylated at Asn-594. Similarly, treatment of HEK293 cells expressing del602-612 huCOX-2 or del607-612 huCOX-2 with KIF failed to lead to a higher molecular mass species (data not shown). Were del602-612 huCOX-2 and del607-612 huCOX-2 already glycosylated at Asn-594 in the absence of KIF treatment, their molecular masses would be expected to be ϳ1 and ϳ1.5 kDa higher than the native huCOX-2, respectively. However, in the absence of KIF treatment, neither del602-612 huCOX-2 nor del607-612 huCOX-2 had an electrophoretic mobility slower than the native enzyme. Thus, neither of these mutants appears to be N-glycosylated at Asn-594. The scrambled mutant Sins597-612 huCOX-2 carrying an intact Asn-594 glycosylation site had the same electrophoretic mobility as Sins595-612 huCOX-2 with or without KIF treatment (data not shown). Collectively, these results indicate that mutations of the 19-aa that confer stability to COX-2 also prevent glycosylation of Asn-594. The mutants L610A huCOX-2 and L611A huCOX-2, which are degraded more slowly and to a lesser extent than native huCOX-2, are also glycosylated less efficiently at Asn-594 in KIF-treated cells (Fig. 2B). However, mutations of the 19-aa that did not greatly enhance the stability of huCOX-2 (i.e. P607A/T608A huCOX-2, V609A huCOX-2, and K612A huCOX-2) were glycosylated to a similar extent as native huCOX-2. These results show that at least a portion of the C-terminal 16-amino acid segment of 19-aa affects the rate of N-glycosylation of Asn-594. This is one mechanism by which the 19-aa cassette could mediate COX-2 protein degradation.
The Helical Region Upstream of Asn-594 Negatively Regulates N-Glycosylation-Inserting the COX-2 19-aa cassette near the C terminus of COX-1 yields an unstable mutant ins594 -612 COX-1 that degrades with a t1 ⁄ 2 of ϳ3 h (18). Ins594 -612 ovCOX-1 has a molecular mass ϳ2 kDa higher than that of its stable, Asn-594-mutated homolog ins594 -612 (N594A) ovCOX-1, indicating that all of the ins594 -612 ovCOX-1 is N-glycosylated at Asn-594 (18) (Fig. 3); moreover, KIF treatment of HEK293 cells expressing ins594 -612 ovCOX-1 significantly stabilized this mutant without eliciting the appearance of a higher N-glycosylated form. Thus, although both native huCOX-2 and ins594 -612 ovCOX-1 carry the same C-terminal 19-aa, ins594 -612 ovCOX-1 is more easily N-glycosylated at Asn-594. The observation that the t1 ⁄ 2 of ins594 -612 ovCOX-1 is ϳ2 h shorter than that of native huCOX-2 (18) is also consistent with the concept that the efficiency of Asn-594 glycosylation positively correlates with the rate of protein degradation. Native huCOX-2 and ins594 -612 ovCOX-1 differ substantially in amino acid sequence in the region immediately upstream of the Asn-594 glycosylation site (Fig. 4A, i). In the muCOX-2 x-ray crystal structure, the Asn-594 glycosylation site is near the C-terminal end of a 9-residue ␣-helical region (Helix A; Fig. 4A, ii and iii). Helix A is connected to a nearby helix (Helix B) by a long 15-residue loop; four of the residues in the loop are not resolved in the crystal structure. The amide side chain of Asn-594 appears to be pointed upward in the direction of Helix B, which is at a distance of ϳ4.5 Å. The length of two GlcNAc residues of an N-glycan group is ϳ8 Å. In short, Helix B is stacked against Helix A in a way that would be expected to impede the transfer of an N-glycan group to the amide of Asn-594. A disulfide bridge formed between Cys-569 of Helix B and Cys-575 of the intervening loop may help constrain the Helix A-loop-Helix B region in a conformation in which Asn-594 cannot become glycosylated. The disulfide bridge can also be seen in the x-ray crystal structure of ovCOX-1, indicating that it is conserved in both COX isoforms. The last resolved residue in the ovCOX-1 structure is Pro-583, which is probably part of a loop. This suggests that the region that is immediately upstream of Asn-594 in ins594 -612 ovCOX-1 (Asp-584 to Val-593) is flexible. Furthermore, the presence of proline and glycine residues at two consecutive positions, 591 and 592, respectively, suggests that the region upstream of Asn-594 in ins594 -612 ovCOX-1 lacks helical structure (Fig. 4A, i). Thus, ins594 -612 ovCOX-1 is probably fully glycosylated at Asn-594, because this N-glycosylation site is situated in a loop that is long enough to be very flexible even in the presence of the disulfide linkage.
To test our concepts regarding the effect of the region upstream of Asn-594 on N-glycosylation, we prepared and stably transfected huCOX-2 mutants bearing various modifications of the 8-residue sequence (UPSTRM8) immediately upstream of Asn-594 (Fig. 4B). V591P/T592G huCOX-2 has both proline and glycine substitutions of the UPSTRM8 designed to disrupt the helix that is present in this region of huCOX-2. This mutation was also designed to mimic the corresponding region of ovCOX-1. A second mutant was made in which the UPSTRM8 region of huCOX-2 was replaced with the corresponding region of ovCOX-1. The initial rates of degrada-  Fig. 1. B, protein expression was induced by treating HEK293 cells stably and inducibly expressing the indicated huCOX-2 variant proteins with 10 g/ml tetracycline (tet) for 24 h. The cells were incubated for an additional 12 h in normal culture medium with or without 25 M KIF. As indicated, cell lysates were prepared and treated with or without endoglycosidase H as described under "Experimental Procedures." Proteins were resolved by SDS-PAGE and subjected to Western blotting. The arrows indicate differences in molecular mass due to N-glycosylation. tion of V591P/T592G huCOX-2 (t1 ⁄ 2 ϳ 5 h) and ovCOX-1 UPSTRM8 huCOX-2 (t1 ⁄ 2 ϳ 3 h) were similar to that of the native enzyme (t1 ⁄ 2 ϳ 5 h). Overall, both mutants appeared to be degraded to a greater extent than native huCOX-2 (Table 1). A more highly glycosylated form of V591P/T592G huCOX-2 was detected even in the absence of KIF (Fig. 4B).  ii, amino acid sequences of the C-terminal regions of muCOX-2 and huCOX-2 that are located upstream of the Asn-594 glycosylation sequence (underlined). Residues that are colored blue are conserved between the two species. Segments that form helices in the muCOX-2 x-ray structure are highlighted. The region marked with an asterisk is not resolved in the muCOX-2 x-ray structure and is very likely to be a continuation of the resolved loop. The disulfide linkage between Cys-569 and Cys-575 is also shown. iii, ribbon diagrams depicting the structure of the C-terminal region of muCOX-2, the sequence of which is shown in ii. The last resolved residue in the muCOX-2 x-ray structure is Ser-596, the final amino acid of the Asn-594 glycosylation site. Asn-594 is situated in Helix A, which is positioned under an adjacent Helix B. The amide chain of Asn-594 appears to be pointed upward in the direction of Helix B. The position of the disulfide linkage relative to Asn-594 is shown. B, amino acid sequences of the C termini of upstream 8-residue ␣-helix huCOX-2 mutants that were stably transfected into HEK293 cells. The underlined sequence is the Asn-594 glycosylation site. The expression of the native or mutant proteins was induced by treating the appropriate HEK293 cell lines with tetracycline for 24 h. After protein induction, cells were incubated for an additional 12 h in normal culture medium with or without 25 M KIF. Cell lysates were prepared, treated with or without endoglycosidase H, resolved by SDS-PAGE, and subjected to Western blotting. The arrows indicate differences in molecular mass due to N-glycosylation.
The observations with UPSTRM8 mutants suggested that disrupting Helix A facilitated Asn-594 glycosylation and, consequently, enhanced the overall degradation of the protein. Accordingly, replacing the UPSTRM8 region of huCOX-2 with that of muCOX-2 should not change the glycosylation pattern at Asn-594, which is what is observed. Very little of the Asn-594-glycosylated form of muCOX-2 UPSTRM8 huCOX-2 was detected in the absence of KIF (Fig. 4B). The conserved helical secondary structures of UPSTRM8 regions of muCOX-2 and huCOX-2 appear to negatively regulate Asn-594 glycosylation. As with native huCOX-2, the degradation of all three UPSTRM8 huCOX-2 mutants is inhibited by MG132 and KIF (data not shown), indicating that they are degraded via the ERAD system.
COX Inhibitors Retard COX-2 Degradation in NIH/3T3 Cells-During the course of our studies, we observed that the degradation of COX-2 protein in serum-stimulated quiescent murine NIH/3T3 cells is inhibited by nonselective nonsteroidal anti-inflammatory drugs, such as flurbiprofen (Fig. 5A) as well as by COX-2-selective inhibitors, such as NS-398 (shown below), but without altering the levels of COX-2 mRNA levels (7) (data not shown). In contrast, constitutive COX-1 protein levels were unchanged (data not shown). There is no evidence that nonsteroidal anti-inflammatory drugs directly inhibit the 26 S proteasome, and because proteasome inhibitors partly inhibit COX-2 degradation in NIH/3T3 cells (18), the findings with FB suggest that there is a second pathway for COX-2 protein degradation that is proteasome-independent.
In contrast to NIH/3T3 cells, COX-2 protein degradation in HEK cells stably expressing native COX-2 was not significantly affected by 20 M NS-398 or 100 M FB (Fig. 5B). Moreover, tetracycline-inducible COX-2 expression in HEK293 cells was not augmented by treating the cells with 20 M NS-398 (data not shown).
Substrate-dependent Degradation of COX-2-Serum stimulation of quiescent 3T3 cells results in the induction of COX-2, the mobilization of free AA due to activation of phospholipases, and the formation of prostaglandins (33). During catalysis, COX-2 undergoes suicide inactivation (1), a process that could initiate the degradation of the enzyme. Resting HEK293 cells lack detectable phospholipase A 2 activity, so heterologously expressed COX-2 is normally not functioning (34,35). This may explain why COX inhibitors retard COX-2 protein degradation in NIH/3T3 cells but not HEK293 cells.
To test the hypothesis that COX fatty acid substrates promote COX-2 protein degradation, we analyzed the effect of exogenous AA on the degradation of native huCOX-2, del595-612 huCOX-2, and N594A huCOX-2 in HEK293 cells. The basal degradation of native huCOX-2 was enhanced by treatment with levels of AA as low as 5 M AA for 4 h (Fig. 6A, i). The mutant del595-612 huCOX-2, which is stable under basal conditions, was degraded upon treatment with AA (Fig. 6A, ii). Therefore, substrate-induced degradation of COX-2 can occur independent of the C-terminal 19-aa. Because removal of the 19-aa does not affect COX-2 catalytic activity (18), our results imply that substrate-induced degradation is due to substrate turnover at the COX active site.
We next examined the effects of COX inhibitors on substrate-induced COX-2 degradation in HEK293 cells. FB and NS-398 attenuated AA-induced degradation of huCOX-2 (Fig.  6B). These inhibitors blocked that part of huCOX-2 degradation that was substrate-dependent without affecting basal degradation of the enzyme (Fig. 6B). AA-induced degradation was not attenuated to a significant extent by the proteasome inhib-

Relationship between the extent of Asn-594 glycosylation and the efficiency of protein degradation
Each mutation is shown in gray type in the sequence. The extent of protein clearance is a measure of how much COX protein is remaining at the end of a 24-h protein degradation experiment as determined by immunoblotting and densitometry. Glycosylation at Asn-594 in the underlined N-glycosylation sequence was assessed by Western blotting of HEK293 cells expressing native or mutant proteins that have been treated with (w/) or without (w/o) KIF. Y, yes; N, no.
We also determined the time dependence of AA-induced degradation in HEK293 cells stably expressing native huCOX-2. The cells were exposed to 20 M AA for either 5 min or 4 h, washed to remove residual AA, and then treated with puromycin with or without FB for an additional 4 h. Regardless of the duration of AA treatment (5 min or 4 h), COX-2 degradation was enhanced to a similar extent (data not shown); moreover, FB added after the 5-min treatment with AA did not protect the enzyme from degradation as well as when inhibitor was added together with AA.
G533A huCOX-2 Is Refractory to Substrate-dependent Degradation-G533A huCOX-2 has less than 5% of the specific COX activity of native enzyme with AA as substrate (5,36). The fact that G533A COX-2 has at least some COX activity with AA indicates that the mutant enzyme binds AA but usually in an orientation that is unfavorable for catalysis.
We stably expressed G533A huCOX-2 in HEK293 cells and analyzed its degradation profile. Under basal conditions G533A huCOX-2 degraded with a half-life that was similar to that of the native huCOX-2 (data not shown). However, unlike native huCOX-2, AA did not cause G533A huCOX-2 to be degraded (Fig. 7). These findings support the concept that a functional COX active site is required for substrate-induced degradation of COX-2.
Substrate-dependent COX-2 Degradation in NIH/3T3 Cells-Exogenous AA enhanced COX-2 degradation in murine NIH/ 3T3 cells expressing COX-2 in response to serum stimulation (Fig. 8A). FB significantly inhibited both the endogenous degradation of COX-2 (Fig. 5) and AA-induced degradation of the protein (data not shown). Experiments were also conducted to determine if stimulating the release of endogenous AA at levels greater than obtained with serum treatment alone would enhance COX-2 degradation. Challenge with 10 M A23187 increased COX-2 degradation in NIH/3T3 cells to about the same extent as treatment with 20 M AA (Fig. 8B). The enhancement in COX-2 degradation by either A23187 or exogenous AA was inhibited by NS-398 (Fig. 8B). In contrast, 10 M bradykinin treatment of NIH/3T3 cells expressing COX-2 did not increase the rate of COX-2 protein degradation (Fig. 8B). A23187 causes a greater enhancement of endogenous AA release than does bradykinin, and we assume that the level of AA mobilization elicited by bradykinin was insufficient to enhance COX-2 degradation in serum-treated NIH/3T3 cells.
Substrate-dependent Degradation of COX-2 Is Not Proteasome-or Lysosome-dependent-We tested an inhibitor of the 26 S proteasome and selective inhibitors of lysosomal degradation in an attempt to identify the pathway responsible for substratedependent COX-2 protein turnover. The 26 S proteasome inhibitor MG132 failed to inhibit substrate-induced COX-2 degradation in HEK293 cells (Fig. 6, B and C). Similarly, lysosomal degradation inhibitors (E64 and leupeptin) and an inhibitor of ER to Golgi trafficking (brefeldin A) failed to prevent AA-induced COX-2 degradation in HEK293 cells expressing native huCOX-2, del595-612 huCOX-2, or N594A huCOX-2 (data not shown). Therefore, substrate-dependent degradation of COX-2 does not appear to require the proteolytic activity of the proteasome or lysosome; nor does it appear to require trafficking between the ER and the Golgi. Efforts to prepare intact, right side-out microsomes from HEK293 cells expressing del595-612 huCOX-2 or N594A huCOX-2 were unsuccessful. Thus, we were unable to establish whether a resident ER protease is involved in substrate-induced COX-2 degradation.
COX-1 Is Resistant to Substrate-induced Protein Turnover-COX-1 is a very stable protein (18,23), and NIH/3T3 cells that have been made quiescent by serum deprivation for 24 -48 h continue to retain high levels of this isoform (18,23). Incubating NIH/3T3 cells with FB did not lead to any detectable change in COX-1 protein levels (data not shown). Similarly, treatment of NIH/3T3 or HEK293 cells stably expressing ovCOX-1 with 20 M AA did not elicit COX-1 degradation (data not shown).  COX-1 and COX-2 Suicide Inactivation-Both COX-1 and COX-2 undergo irreversible, mechanism-based "suicide" COX inactivation during in vitro catalysis (2,17). It is possible that COX-2 suicide inactivation leads to the formation of an inactivated protein form that is susceptible to degradation. As shown in Fig. 9, membranes from HEK293 cells expressing native ovCOX-1 or His 6 muCOX-2 were both found to lose COX activity when the cells were treated with AA under conditions that lead to enhanced COX-2 (but not COX-1) degradation. Related rate experiments performed with [1-14 C]AA also indicated a loss of COX activity and no change in the pattern of products formed by the activity remaining.

DISCUSSION
The studies reported here indicate that there are two independent pathways for COX-2 protein degradation. One involves ER-associated degradation (ERAD) with proteolysis mediated by the 26 S proteasome. This process occurs at a relatively constant rate and is coupled to post-translational N-glycosylation of Asn-594. A second degradative pathway is initiated by fatty acid substrate oxygenation, and the rate of degradation via this pathway correlates with the formation of suicide-inactivated COX-2. The biochemical steps involved in substrate-induced COX-2 degradation are not resolved, but the process appears not to require the 26 S proteasome, lysosomal proteases, or trafficking of COX-2 between the ER and the Golgi apparatus.
There are four N-glycosylation sites in COX-2. Previous studies have suggested that Asn-67, Asn-144, and Asn-410 are primarily co-translationally N-glycosylated and that glycosylation of these sites facilitates COX-2 maturation in the ER (37) through a process that probably involves the ER chaperones calnexin and/or calreticulin (38). Asn-594, the fourth N-glycosylation site of COX-2, is variably glycosylated (37,39,40) and does not need to be glycosylated for the enzyme to achieve a catalytically competent conformation (37). Instead, we have found that post-translational N-glycosylation of Asn-594 is important for the degradation of huCOX-2 expressed in HEK293 cells (18). Here we have provided evidence suggesting that Asn-594 is post-translationally N-glycosylated in murine NIH/3T3 cells and RAW264.7 cells and that N-glycosylation of Asn-594 is regulated by residues both upstream and downstream. As discussed in more detail below, we suggest that an interaction between an unfolded C-terminal segment downstream of Asn-594 and the upstream 8-residue ␣-helix containing Asn-594 exposes this residue to post-translational N-glycosylation, which, in turn, initiates entry of COX-2 into the ERAD system (Fig. 10). The ER ␣1,2-mannosidase I catalyzes the committed step in the ERAD pathway (41)(42)(43)(44). This enzyme is able to cleave up to 4 mannose residues from an N-linked Man 9 GlcNAc 2 (43,44). The effect of this trimming is to prevent glucosylation of N-linked, mannose-containing oligosaccharides by a UDP-glucose-glycoprotein glucosyltransferase (45)(46)(47)(48). Lacking the appropriate terminal glucose, irreversibly misfolded glycoproteins are no longer able to re-enter the calnexin/calreticulin folding cycle. Mannose-binding proteins, such as Yos9p or the ER degradation mannosidase I-like protein (EDEM), will bind Man n GlcNAc 2 groups and facilitate the delivery of the glycoprotein, now having a processed N-linked glycan to a membrane retrotranslocon for export from the ER to the cytosol for proteasomal degradation (49 -55). The identity of this membrane retrotranslocon has not been resolved. KIF inhibition of ER ␣1,2-mannosidase I prevents retrotranslocation and proteasomal degradation of misfolded glycoproteins by causing them to be retained in the ER (43-46, 55-58). Because KIF inhibits  [AA] Specific COX Activity (nmol/min/mg lysate) FIGURE 9. Effect of exogenous AA on the COX activity of HEK293 cells expressing COX-1 or COX-2. Following induction of ovCOX-1 or His 6 muCOX-2, the HEK293 cells stably expressing these isoforms were incubated with the indicated concentrations of AA for 10 min. Afterward, the cells were washed, collected, and lysed. The whole cells lysates were assayed for COX activity using an O 2 electrode assay as detailed under "Experimental Procedures." COX-2 degradation and leads to the accumulation of the Asn-594-glycosylated form of the enzyme, we reason that COX-2 degradation requires Asn-594 glycosylation and subsequent processing of the attached N-glycan group by ER ␣1,2-mannosidase I. Although ERAD is generally considered to be an ER quality control mechanism for the elimination of aberrantly folded glycoproteins, COX-2 appears to be degraded from a properly folded conformation. Sifers and co-workers (41) have proposed that glycoprotein ERAD (GERAD) requires both an N-glycan component and some nonnative protein structure. Their model is based on the observation that inhibiting the N-glycosylation of misfolded proteins in the ER prevents their degradation (59). However, it is still unclear how GERAD quality control is able to selectively eliminate terminally defective glycoproteins while sparing folding glycoproteins that have yet to achieve their native protein structure. It is possible that nascent folding glycoproteins are protected from degradation by association with ER-resident chaperones such as hsp70 BiP.
Based on the criteria for GERAD and our present findings, we provide a model for the initiation of COX-2 degradation by Asn-594 glycosylation (Fig. 10). In this model, COX-2 synthesis and maturation in the ER involve co-translational N-glycosylation of Asn-67, Asn-144, and Asn-410, disulfide bond formation, incorporation of heme, assembly into a homodimer, and membrane insertion. We propose that Asn-594 of newly formed, properly folded COX-2 is not easily accessible for N-glycosylation, because it is situated in Helix A with the amide group of Asn-594 shielded by the adjacent Helix B about 4.5 Å distant. We suggest that a conformational change involving Helix A-loop-Helix B exposes Asn-594 for post-translational N-glycosylation; this conformational change may be initiated by an interaction of Leu-610/Leu-611 with the Helix A-loop-Helix B region. Finally, we propose that the resulting N-glycosylation of Asn-594 constitutes the signal for COX-2 GERAD.
To our knowledge, the only other report of post-translational N-glycosylation of a native protein is that of the secretory glycoprotein human coagulation factor VII (60). Factor VII has two consensus glycosylation sites and, like COX-2, exists as two alternatively glycosylated forms of 54 and 56 kDa. Because COX-2 is a homodimer, post-translational glycosylation of one of its monomers at Asn-594 could be sufficient to induce the degradation of the dimer complex. This may explain why KIF treatment stabilizes the smaller glycosylated form of COX-2; it is possible that the processing of the Asn-594 N-glycan moiety by ER ␣1,2-mannosidase I precedes the disassembly of COX-2 that would have to occur in order for the protein to be removed from the ER.
We have found that COX inhibitors do not inhibit proteasome-dependent degradation of COX-2 either in NIH/3T3 cells where the enzyme is expressed endogenously or in HEK293 cells expressing COX-2 heterologously. This suggested that in addition to the ERAD pathway, there is a second pathway for COX-2 protein degradation that is proteasomeindependent. Degradation of native COX-2 via this second pathway can be stimulated by endogenous or exogenous AA, but treatment of G533A COX-2 that binds AA but has a very slow rate of substrate turnover does not augment degradation of this mutant. Finally, a brief 5-min treatment of COX-2-expressing cells with AA is sufficient to enhance COX-2 degradation. Collectively, these findings indicate that the second pathway is substrate turnover-dependent.
We propose that substrate-dependent degradation of COX-2 occurs as a result of structural damage to the enzyme that occurs as a consequence of suicide inactivation. COX-1 and COX-2 undergo irreversible suicide inactivation during COX catalysis under cell-free conditions (2,17,61,62) and perhaps in cultured RAW264.7 cells that have been challenged with LPS and IFN-␥ (63). In vitro COX inactivation is accompanied by significant changes in the structure of the enzyme (61). We find that both COX-1 and COX-2 become inactivated in intact HEK293 cells upon exposure to exogenous AA. It is unclear why COX-1 undergoes catalytic inactivation but is resistant to substrate-dependent degradation.
We have not identified the protease(s) responsible for the substrate-induced COX-2 degradation. Studies with various inhibitors suggest that the proteolytic activities of the 26 S proteasome and the lysosome and trafficking between the ER and Golgi are not involved. Proteases known to be associated with or present in the ER include the signal peptidase, the signal peptide peptidase, ER aminopeptidases-1 and -2 (ERAP-1 and -2) (64 -67), a 90-kDa plasminogen-related protease (68,69), and a yet to be identified serine protease (70). Recently, Donoso et al. showed that degradation of a misfolded mutant of BiP is initiated in the ER lumen by this latter protease. A proteolytic system may exist in the ER that serves as an alternative or a complement to the ERAD-proteasome pathway in the quality control of structurally defective ER-associated proteins. A catalytically inactivated form(s) of COX-2 could be degraded by this putative ER protease system.
Although our results establish that COX-2 protein can be degraded via two distinct pathways, it is not known which pathway is most important in vivo or whether different cell types may favor one pathway over the other. We have recently engineered a del595-612 knock-in mouse line that we expect to be useful in addressing these issues. N-Glycosylation of Asn-594 is necessary for COX-2 degradation. A hydrophobic interaction between the C-terminal region of 19-aa and the upstream 8-residue ␣-helix may expose Asn-594 for post-translational N-glycosylation. Subsequent processing of the attached N-glycan group by ER ␣1,2-mannosidase I and other glycosidases and glycosyl transferases could initiate the translocation of the protein from the ER for ubiquitination and proteasomal degradation.