Catalysis of arachidonic acid (AA) by cyclooxygenase-2 (COX-2) gives rise to a single product that serves as a precursor for all prostaglandins, which are central mediators of inflammation. Rapid up-regulation of COX-2 expression in response to pro-inflammatory stimuli is a well-characterized means of generating the large pool of prostaglandins necessary for inflammation. However, an efficient inflammatory process must also terminate rapidly and thus requires cessation of COX-2 enzymatic activity and removal of excess protein from the cell. Previous studies showed that COX-2 that has not been exposed to AA (‘naive’) degrades in the cellular proteasome. However, continuous exposure to AA induces suicide inactivation of COX-2 and its elimination no longer occurs in neither the proteasomal nor lysosomal machineries. In the present study, we show that either overexpressed or endogenously induced COX-2 is secreted via exosomes through the endoplasmic reticulum–Golgi pathway. We further find that excretion of COX-2 is significantly enhanced by prolonged exposure to AA. Genetic or chemical inhibition of COX-2 enzymatic activity has no effect on its secretion in the absence of substrate, but prevents the additional activity-dependent secretion. Finally, transfer of COX-2 to target cells only occurs in the absence of AA stimulation. Together, these results suggest that exosomal secretion of AA-activated COX-2 constitutes a means to remove damaged inactive COX-2 from the cell.
The cyclooxygenase (COX) enzyme isoforms COX-1 and COX-2 catalyze the rate-limiting step in the conversion of arachidonic acid (AA) to prostaglandins (PGs), soluble ligands that regulate cardiovascular, immunological and neurological functions. Both COX isoforms are N-glycosylated homodimers that are associated with the luminal surface of the endoplasmic reticulum (ER) and the nuclear envelop, and share a high degree of structural and catalytic similarities . Contrary to COX-1 that is expressed at constitutive levels within most mammalian tissues, COX-2 expression is rapidly and transiently up-regulated in response to pro-inflammatory signals . Given the proven role of COX-2 in generating the necessary pool of prostanoids required for an effective inflammatory response within a short period of time, a tight control of its expression and activity is essential to prevent an ongoing response that may lead to complications of unresolved inflammation.
One of the important factors that limit COX-2 activity is an inherent property of the catalytic reaction itself. The conversion of AA to the final product that gives rise to all prostanoids occurs by a two-step catalysis of oxygenation of AA into PGG2, followed by peroxidation into PGH2 . Several studies found that oxidation of AA is achieved by a mechanism that generates free radicals, which ultimately cause structural damage to the protein and lead to suicide inactivation of COX-2 [3–5]. Multiple studies show that accumulation of COX-2 protein is characteristic of chronic inflammatory diseases , as well as several malignant tumors where its presence is highly correlated with poor prognosis [7–13]. However, despite the availability of potent inhibitors of COX activity, no significant effects are achieved by its inhibition. In the past decade, studies have elucidated that besides its enzymatic activity, COX-2 may have additional cellular roles that are independent of its catalytic activity , suggesting that removal of COX-2 after fulfilling its catalytic role is mandatory for the prevention of untoward effects.
While a multitude of studies focus on mechanisms that up-regulate COX-2 expression, surprisingly little is known about COX-2 degradation. Furthermore, most of the existing data was obtained for an enzyme that has not been subject to its substrate (naive COX-2 hereafter). In the absence of AA, COX-2 is N-glycosylated in the ER and then transported to the Golgi where its N594 glycan is trimmed prior to its retrograde transport back to the ER . From there, the mature COX-2 enters the ER-associated degradation (ERAD) pathway and is subsequently transported to the cytosol where it is ubiquitinated and degraded by the cellular proteasome [16–18]. Our previous studies showed that degradation of COX-2 in this pathway is enhanced by an interaction between COX-2 and certain G protein-coupled receptors including those that respond to the products of its catalysis (e.g. EP1 prostaglandin receptor) [19–21], thus providing an additional level of regulation for COX-2 expression.
In contrast to degradation of naive COX2, exposure of COX-2 to AA elicits activity turnover-dependent degradation that is manifested by a reduction in the levels of the full-length protein. However, this type of degradation does not seem to occur in neither the proteasome nor the lysosome and its whereabouts are unknown . A recent study showed that exposure of lung cancer cells to a selective COX-2 inhibitor elevates the intracellular levels of COX-2 and causes its secretion and transfer to target recipient cells by exosomes . In the present study, we tested whether extracellular secretion of COX-2 may also be used as an unrecognized means to remove the catalytically damaged protein from the cell.
Materials and methods
Arachidonic acid, (S)-MG132 and NS-398 were from Cayman Chemical (Ann Arbor, MI, U.S.A.). Tunicamycin (TM), cycloheximide, chloroqine and bacterial LPS were from Sigma–Aldrich (Israel). Brefeldin A was from Frementek (U.K.). All other reagents were standard laboratory grade.
Cell culture and transfection
HEK293 cells were grown in Eagle's MEM; RAW 264.7 and H9C2 were grown in Eagle's DMEM and THP-1 were grown in RPMI. All media were supplemented with 10% heat-inactivated fetal bovine serum and 100 U/ml penicillin and streptomycin. Transient transfections were carried in sub-confluent (70–80%) monolayers using PolyJet (SignaGen Laboratories) at a ratio of 1 : 3 cDNA : PolyJet, according to the manufacturer's instructions. Cell viability and apoptosis were determined by flow cytometry (BD FACSCanto II) using annexin V and the propidium iodide staining kit (MBL International Corporation).
pcDNA5/FRT/TO encoding for huCOX-2, N549A-huCOX-2, huCOX-1, G533A-COX-2, Golgi-ΔSTEL huCOX-2 and KDEL-huCOX-2 were generous gifts from Prof. William L. Smith, University of Michigan.
Goat polyclonal anti-COX-2 (C-20), goat polyclonal anti-COX-2 AC-conjugated beads, mouse monoclonal anti-actin (C-2), mouse monoclonal anti-Alix (1A12), rabbit polyclonal anti-CD9, mouse monoclonal anti-CD81 and goat polyclonal anti-GAPDH were obtained from Santa Cruz Biotechnology (Santa Cruz, CA, U.S.A.). Mouse monoclonal COX-1 was from Cayman Chemical Company (Ann Arbor, MI, U.S.A.). Horseradish peroxidase-conjugated bovine anti-goat IgG, goat anti-rabbit IgG and goat anti-mouse IgG were obtained from Jackson ImmunoResearch Laboratories (West Grove, PA, U.S.A.).
Isolation of COX-2 from conditioned media
HEK293 cells (5.5 × 106) were seeded in 150-mm dishes and transfected with 20 µg of the different cDNA constructs for 8 h, after which they were washed once with warm PBS and media were replaced with 10 ml of MEM without serum (MEM−). RAW 264.7 and H9C2 (5.5 × 106) were seeded in 150 mm dishes and starved in DMEM− for 60 min after which they were treated with 100 nM LPS. Sixteen hours later, the media from all experiments were collected into 50 ml tubes, centrifuged at 1200×g for 10 min at 4°C and the supernatant was transferred to new tubes and either lyophilized overnight (VirTis BenchTop), or separated by Amicon Ultra-15 centrifugal filters (Merck Millipore, Ireland). Samples for western blot detection were resuspended in 1 ml of RIPA/SDS [50 mM Tris (pH 8), 150 mM NaCl, 5 mM EDTA, 1% (v/v) NP-40, 0.5% (w/v) deoxycholic acid, 0.1% (w/v) SDS and 10 µM NaF] and COX-2 was immunoprecipitated overnight as described previously .
Exosomes were isolated from conditioned media of ∼24 × 106 cells (four 150-mm dishes per point) by differential centrifugation, as previously described . Briefly, following transfection, media were collected and centrifuged at 2000×g for 10 min at 4°C. The supernatant was transferred to fresh tubes and centrifuged at 12 000×g for 30 min at 4°C. The resulting supernatant was filtered through 0.22 µm Whatman filters and centrifuged again at 100 000×g for 90 min at 4°C. The exosomal pellet was resuspended in 30 ml of PBS and recentrifuged at 100 000×g for 60 min. The pellet was resuspended in 80–100 µl of PBS for density light scattering (DLS) experiments (see below) or in 30 µl of PBS for western blot detection.
Exosome preparations were diluted in 80 µl of PBS and DLS measurements were performed at room temperature using the Vasco DLS system (Cordouan Technologies) at The Russell Berrie Nanoparticle and Nanometric Systems Characterization Center Analysis at the Technion. Analysis of the data was performed using the nanoQ software (Cordouan Technologies).
Media transfer experiments
HEK293 cells were transfected overnight with either Mock or COX-2 cDNA. Treatment with 20 µM AA was given on the next day for 1 h, after which media were collected, centrifuged at 1000×g for 10 min at room temperature, treated with DNAse and then applied on to HEK293 acceptor cells pretreated with 2.5 µM cycloheximide (CHX) to prevent de novo protein synthesis. Samples from the acceptor cells were collected 24 h after transfer and COX-2 was immunoprecipitated as described above.
Results are presented as mean ± SD of the indicated number of biological repeats. Statistical significance was determined by Student's t-test and P-values <0.05 were considered significant.
The overall objective of the present study was to determine the mechanism of elimination of substrate-activated COX-2 from the cell. A previous study by Mbonye et al. had shown that exposure of COX-2 to AA causes a reduction in the levels of the full-length COX-2, which is not sensitive to either proteasome or lysosome inhibitors . To begin studying our hypothesis that damaged COX-2 is removed from the cell by means of excretion, we first tested the effect of AA stimulation on both COX isoforms. HEK293 cells expressing COX-1 or COX-2 were treated with CHX to prevent de novo protein synthesis, followed by exposure to high levels of AA for 4 h. As shown in the left panel of Figure 1A, stimulation of COX-1 with AA had no effect on its levels. In contrast, prolonged exposure of COX-2 to AA caused a significant reduction in the levels of the full-length protein (Figure 1A, right panel) . However, we have also noticed that stimulation with AA caused the appearance of 1–2 lower molecular mass immunoreactive bands between 35 and 25 kDa (Figure 1A, arrows in the right column). Although their amino acid sequence is yet to be determined, the consistent de novo appearance of these bands following AA treatment was found to be a reliable indicator for AA-activated protein (as will be shown next). Consistent with the previous report, the reduction in mature COX-2 levels due to AA stimulation was not affected by treatment with inhibitors of the proteasome or lysosome pathways , which confirmed that disappearance of activated COX-2 from the cell occurs via a different mechanism of action.
COX-2 is found in the extracellular compartment.
We next sought to determine whether mere elevation of intracellular COX-2 levels causes it to appear in the extracellular media. First, we transfected HEK293 cells that do not express COX-2 endogenously, with either empty vector or COX-2 cDNA, replaced the overlaying media 7 h after transfection with fresh serum-free media and examined its content for the presence of COX-2 16 h later (see Materials and Methods for details). As depicted in the left panel of Figure 1B, overexpression of COX-2 caused the appearance of a single band at ∼72 kDa, consistent with the full-length protein. Immunoprecipitation of COX-2 from the extracellular media also revealed the presence of the full-length protein, which was present only in the media of COX-2-transfected cells (Figure 1B,C and Supplementary Figure S1 for full gel depiction). Since serum starvation in itself may cause cell death, we measured cell viability under serum-free and serum-supplemented conditions and found no difference in cell survival (Supplementary Figure S2A). These data suggest that the presence of COX-2 in the extracellular compartment is not due to disruption of cell integrity due to apoptosis.
To confirm that the above observation is not an artifact of heterologous expression, we searched for the presence of COX-2 in the extracellular media of cells that express COX-2 endogenously. For this, we used two cell lines, RAW 264.7 murine macrophages (Figure 1D,E) and H9C2 myoblasts (Figure 1F,G), that express detectable levels of COX-2. Stimulation of RAW 264.7 (Figure 1D) and H9C2 cells (Figure 1F) with lipopolysaccharide (LPS) caused a significant increase in the intracellular levels of COX-2. This also resulted in a parallel increase in the presence of extracellular full-length COX-2 (Figure 1E,G). Of note is the appearance of lower COX-2 immunoreactive bands in RAW264.7 cells. In these cells, serum starvation in itself did lower cell viability, but this was reversed by treatment with LPS (Supplementary Figure S2B). In contrast, COX-2 secretion was enhanced in the presence of LPS, suggesting that the presence of COX-2 outside the cell is not due to cell disruption.
Having established that COX-2 is present in the extracellular compartment, we sought to determine the pathway by which it is secreted. Following its translation in the ribosome, COX-2 is directed from the ER to the Golgi apparatus, where it was shown to delay prior to its retrograde transport back to the ER . To test whether COX-2 is also secreted via the ER–Golgi pathway, we disrupted the integrity of the Golgi using low concentrations of Brefeldin A (BFA) that do not interfere with the endosomal system . Flow cytometry analysis confirmed that treatment with BFA under these conditions has no effect on cell viability (Supplementary Figure S2A). As shown in Figure 2A, BFA treatment caused complete disappearance of COX-2 from the extracellular media. We next tested whether N-glycosylation of COX-2, which is required for its maturation, is also required for COX-2 excretion. For this, we treated COX-2-overexpressing cells with tunicamycin (TM), which blocks N-linked glycosylation, and tested the effect on COX-2 secretion. As shown in Figure 2B, treatment with TM prevented N-glycosylation of intracellular COX-2, causing the appearance of a lower molecular form of the protein. An examination of the extracellular media showed that unglycosylated COX-2 was not secreted to the overlaying media. However, a COX-2 mutant lacking the ability to undergo N-glycosylation at only one of its four potential sites (N549-COX-2)  was present outside the cell to the same extent of the wild type (Figure 2D). Therefore, the effect of TM on COX-2 secretion may either involve the additional N-glycosylation sites on COX-2 or reflect an indirect effect of additional proteins that may be involved in the process.
COX-2 is secreted by the ER–Golgi pathway.
Previous studies have shown that the delay of COX-2 in the Golgi apparatus is largely due to a significantly less efficient ER retention signal compared with COX-1 (STEL vs. a KDEL motif, respectively) and that swapping the STEL of COX-2 with a KDEL motif expedites its return to the ER and stabilizes it . Secretion experiments done using the KDEL-COX-2 mutant showed significantly lower presence in the extracellular compartment. An additional mutant, Golgi-ΔSTEL, which confines COX-2 to the Golgi , had a similar effect (Figure 2D). Together, the data presented in Figure 2 support the secretion of COX-2 via the ER–Golgi pathway.
To confirm that extracellular COX-2 is found in exosomes , we used a protocol of differential centrifugation to isolate the exosomal fraction  of HEK293 cells transfected with either empty or COX-2 cDNA and used both biophysical and biochemical methods to confirm the presence of COX-2 within exosomes. DLS experiments showed that while no particles were present in samples from cells transfected with empty vector (Mock) (Figure 3A, blue line), the supernatant of cells expressing COX-2 showed a major peak of 175 ± 57 nm consistent with exosomes. A much smaller peak of 463 ± 28 nm consistent with the presence of larger microvesicles was also evident (Figure 3A, red line). Probing of samples isolated by the same method using western blot revealed the presence of COX-2 only in the fraction obtained from COX-2-transfected cells, together with the exosome marker CD81 (Figure 3B). Note the absence of CD81 in the mock-transfected cells, indicating that overexpression of COX-2 is sufficient to induce exosome formation.
COX-2 is found within exosomes.
We next performed similar experiments in H9C2 cells that express COX-2 endogenously and elevate its expression in response to LPS stimulation. As depicted in Figure 3C, in the absence of LPS, there is a small peak of 116 ± 54 nm that probably represents the presence of a small population of vesicles forming in these cells at baseline conditions. LPS stimulation caused narrowing of this band to 108 ± 15 nm, reflecting a more uniform vesicle population and amplified it more than 6-fold, consistent with the induction of exosome secretion. Two additional peaks of 462 ± 15 and 804 ± 26 nm were not affected by LPS treatment. As opposed to the HEK293 cells where exosome formation can be attributed to COX-2 overexpression, LPS stimulation may also cause secretion of cargo that does not include COX-2. However, probing of the extracellular media showed that COX-2 is indeed present within these exosomes (Figure 3D).
Having established that COX-2 is secreted to the extracellular compartment by exosomes, we next tested our hypothesis that suicide-inactivated COX-2 is eliminated by means of secretion. As depicted in the left panels of Figure 4A,B, stimulation of COX-2-expressing HEK293 cells with AA caused a reduction in the intracellular levels of the full-length protein and also the de novo appearance of the lower molecular mass immunoreactive bands that indicate AA activity (Figure 1A). Examination of the supernatant of the same samples revealed that AA stimulation also augmented the presence of COX-2 in the extracellular compartment (Figure 4A,B, right panels). Next, we sought to determine whether enzymatic activity is required for the additional secretion of COX-2. For this, we used a catalytically impaired protein, which binds AA but cannot process it into product (G533A-COX-2) . Examination of the effect of AA stimulation on the extracellular levels of COX-2 showed that in contrast to the wild-type protein, G533A-COX-2 that has not been exposed to substrate is secreted from the cell, but the addition of AA does not further increase its secretion (Figure 4C,D).
AA stimulation of COX-2 increases its secretion.
We next tested how AA stimulation affects the secretion of COX-2 in RAW 264.7 and H9C2 cells that express the protein endogenously (Figure 4E,F, respectively). In both cell lines, stimulation with LPS caused the expected increase in intracellular and extracellular levels of COX-2 (as seen also in Figure 1D–F). In contrast to HEK293 cells, stimulation with AA did not cause a significant reduction in the levels of the full-length protein. However, AA activation was evident by the appearance of the lower molecular mass immunoreactive bands. In the extracellular compartment, stimulation of AA in addition to LPS caused further elevation in the levels of COX-2 outside the cell (Figure 4E,F, right panels). Importantly, chemical inhibition of COX-2 activity by NS-398 resulted in disappearance of the lower molecular bands in the intracellular samples and in decreased AA-mediated secretion to the same level obtained by LPS alone (Figure 4F). This result is in accordance with the data obtained for AA-stimulated G533A-COX-2 (Figure 4C,D) whereby catalytic activity of COX-2 is necessary for its additional secretion.
A recent study by Kim et al. found that elevation of COX-2 in lung cancer cells enables its transfer to neighboring cells, where it is catalytically active . Here, we tested whether AA-activated COX-2 is also transferred in the same way. HEK293 cells expressing COX-2 were treated with or without AA, and media were collected and transferred on to recipient cells (see details in the Materials and Methods section). As shown in Figure 4G,H, acceptor cells that received the media of COX-2-overexpressing cells showed the presence of COX-2 within them. However, while exposure to AA elevated extracellular COX-2 levels, there was significantly less COX-2 present in the acceptor cells.
Given its central role in the inflammatory process, regulation of COX-2 expression is critical for maintaining homeostasis and prevention of unresolved inflammation. Whereas the mechanisms that control COX-2 synthesis are well characterized, less is known of processes that dispose of the protein once its activity is ceased. Previous studies have shown that COX-2 that has not been exposed to its substrate (‘naive’) is continuously formed and degraded by ERAD [15,16], possibly as a means of maintaining a reservoir of protein for immediate use. Interestingly, while the ERAD pathway is employed mainly to promote clearance of misfolded or unfolded proteins , COX-2 is degraded in this pathway, only in its naive form but not under conditions that generate a damaged protein . In the present study, we also confirm that naive COX-2 is secreted by exosomes  and provide evidence that a surplus of naive COX-2 protein, whether artificially overexpressed or endogenously induced, exits the cell through the ER–Golgi pathway. Indeed, disruption of the Golgi, or swapping the weak ER retention signal to a more robust one, significantly attenuates secretion, as does inhibition of general N-glycosylation.
Inflammatory and pathological signals that cause a rapid and substantial increase in COX-2 levels in physiological systems also initiate activation of PLA2  that liberates AA and ultimately lead to substrate-mediated suicide inactivation of COX-2. Mbonye et al.  demonstrated that suicide inactivation and subsequent substrate-dependent degradation occur in HEK293 transfected with COX-2 and in RAW 264.7 cells treated with LPS. Using the same experimental conditions in our study, we were able to show that the presence of COX-2 in the extracellular compartment is significantly increased when it is exposed to AA, suggesting that at least part of the damaged protein is secreted by exosomes. The complete disappearance of intracellular COX-2 was not observed even under conditions of inhibition of protein synthesis by cycloheximide, possibly due to complete engagement of the cellular pool of exosomes by excessive overexpression of COX-2. Interestingly, while naive COX-2 was found to transfer to target cells, AA treatment significantly reduced the amount of COX-2 in recipient cells. Whether this is because damaged proteins are immediately degraded by the host cells or due to a mechanism that prevents their entry into target cells altogether remains to be determined.
In their study of COX-2 degradation pathways, Mbonye et al.  proposed that disappearance of COX-2 following exposure to AA may be the result of its cleavage by unknown proteases. Here, we show that such proteolysis seems to exist. While exposure to AA causes a reduction in the quantity of the full-length protein in HEK293 cells that overexpress COX-2, a much smaller effect of AA is seen in systems that express COX-2 endogenously. However, AA treatment always causes the de novo appearance of one or two lower molecular immunoreactive COX-2 species that reliably reflect proteolytic activity. Unfortunately, the experimental conditions that require detection of extracellular COX-2 by means of immunoprecipitation mask our ability to detect these lower fragments.
The use of exosomes as means of transferring functional proteins to target cells has been established in several studies . In the case of COX-2, lung cancer-derived cells were shown to secrete and transfer the protein to the THP-1 cells, where it maintained its catalytic activity . We have carefully followed these protocols in our quest to discover whether secreted COX-2 is active. Experiments using large volumes of cells and positive controls confirming our ability to detect activity within the range of the standard curve failed to detect activity of COX-2 in the extracellular compartment (data not shown). The reasons for this may be the lack of a sufficient amount of prostaglandin synthases that convert the single COX-2 PGH2 product into PGE2 that is detected in the assay, or possibly preparation procedures that harm COX-2.
Elevated levels of COX-2 are a hallmark of chronic inflammation, as well as several malignancies where its presence is indicative of poor prognosis [30–32]. Furthermore, the great majority of COX-2-overexpressing tumors are refractory to treatment with non-steroidal anti-inflammatory drugs [33–35]. It is possible that COX-2 expression in these conditions reflects excess suicide-inactivated damaged protein that is not cleared properly. Accumulation of the full-length protein or its fragments within the cells may cause untoward effects, some of which may be independent of catalytic activity. Therefore, understanding the pathway of COX-2 secretion may reveal additional means of treating COX-2-overexpressing conditions in the future.
E.S. designed, executed and analyzed experiments presented in Figures 1, 2, 3B,D and 4. S.T. designed, executed and analyzed experiments presented in Figure 3A,C. L.B.-H. designed experiments, prepared the figures and wrote the manuscript. E.S. and S.T. had input into writing the manuscript and generating the figures.
This work was supported by the Israel Science Foundation [grant no. 1445/14 to L.B.H.].
The authors thank Dr. Hila Toledano for critical reviewing of the manuscript.
The Authors declare that there are no competing interests associated with the manuscript.