Peroxidase activity of cyclooxygenase-2 (COX-2) cross-links β-amyloid (Aβ) and generates Aβ-COX-2 hetero-oligomers that are increased in Alzheimer's disease

Oxidative stress is associated with the neuropathology of Alzheimer's disease. We have previously shown that human Aβ has the ability to reduce Fe(III) and Cu(II) and produce hydrogen peroxide coupled with these metals, which is correlated with toxicity against primary neuronal cells. Cyclooxygenase (COX)-2 expression is linked to the progression and severity of pathology in AD. COX is a heme-containing enzyme that produces prostaglandins, and the enzyme also possesses peroxidase activity. Here we investigated the possibility of direct interaction between human Aβ and COX-2 being mediated by the peroxidase activity. Human Aβ formed dimers when it was reacted with COX-2 and hydrogen peroxide. Moreover, the peptide formed a cross-linked complex directly with COX-2. Such cross-linking was not observed with rat Aβ, and the sole tyrosine residue specific for human Aβ might therefore be the site of cross-linking. Similar complexes of Aβ and COX-2 were detected in post-mortem brain samples in greater amounts in AD tissue than in age-matched controls. COX-2-mediated cross-linking may inhibit Aβ catabolism and possibly generate toxic intracellular forms of oligomeric Aβ.

ies were from Amersham Biosciences. MagnaBind goat anti-mouse and rabbit IgG beads were from Pierce. The other reagents were obtained from Sigma unless otherwise noted.
Brain Samples-Frozen human post-mortem brain samples (superior temporal cortex, Brodmann area 22, 41/42) from moderately affected AD , n ϭ 5), severely affected AD (Braak stage 6, n ϭ 4), and control cases (n ϭ 4) were obtained from the Harvard Brain Tissue Resource Center (Belmont, MA). The profile of each case is summarized in Table I. There were no significant differences in the average age or post-mortem interval between the groups (p Ͼ 0.5, analysis of variance).
Preparation of Reactions-A␤ peptide stock solutions were prepared in HPLC grade water (Fisher) on the day of the experiment. The peptide preparation was then filtered through a Spin-X cellulose acetate filter unit (0.22 m; Corning Incorporated, Corning, NY). Concentrations of A␤ were determined by BCA assay (Pierce), which we have previously validated as an assay for A␤ concentration (53). A␤ was incubated with ovine COX-2 in a final volume of 100 l of Dulbecco's phosphatebuffered saline without calcium and magnesium at 37°C. Stock solutions of indomethacin (50 mM), ibuprofen (250 mM), and aspirin (250 mM) were prepared in ethanol. Where COX-2 inhibitors were used, the enzyme was preincubated with the reagent for 30 min on ice. For chemical modification of amino acids in A␤, the peptide was incubated with the reagent for 30 min at 25°C prior to adding COX-2 and hydrogen peroxide. The reaction was terminated by adding sample buffer for SDS-PAGE (Invitrogen) containing 5% 2-mercaptoethanol.
Immunoprecipitation-Solutions reacted in vitro were incubated with 2 l of anti-COX-2 antibody overnight at 4°C and with 30 l of MagnaBind anti-rabbit IgG beads for 1 h at 4°C subsequently. For co-immunoprecipitation of A␤ or COX-2 from tissue, control and AD brain samples were homogenized by a glass homogenizer in 10 ml/g of ice-cold phosphate-buffered saline containing protease inhibitor mixture (Roche Applied Science). The supernatant was separated by centrifugation at 20,000 ϫ g for 10 min and adjusted to a final protein concentration of 1.75 mg/ml with homogenization buffer. After pretreatment with 10 l of MagnaBind beads to decrease the nonspecific binding of proteins, the supernatant (0.8 ml) was incubated with 3 l of anti-A␤ (W0 -2) or COX-2 antibody overnight at 4°C and subsequently with 30 l of MagnaBind beads for 3 h at 4°C. For Western blot, the precipitates were washed three times with phosphate-buffered saline and boiled for 5 min at 95°C with sample buffer containing 5% 2-mercaptoethanol.
Immunoblotting-The samples in sample buffer were loaded onto a NuPAGE 4 -12% Bis-Tris gel and separated by electrophoresis at 200 V for 45 min. The gel was then transferred to polyvinylidene difluoride membrane (Bio-Rad) at 75 mA for 90 min. The membrane was blocked with 10% dry milk, 5% bovine serum albumin for anti-A␤ antibodies or 5% dry milk for anti-COX-2 antibody for 1 h at 25°C. The membrane was then incubated with a primary antibody overnight at 4°C followed by a horseradish peroxidase-conjugated secondary antibody for 1 h at 25°C. Antibody labeling was detected by LumiGLO chemiluminescent kit (Kirkegaard & Perry Laboratories, Inc., Gaithersburg, MD). The density of each band from the immunoblotting was analyzed by National Institutes of Health Image 1.62 software (National Institutes of Health, Bethesda, MD).

RESULTS
To study the effects of COX-2 on the oligomerization of A␤, we incubated purified ovine COX-2 (140 nM) with human synthetic A␤ 1-40 ("h40," 2.5 M) in the presence of H 2 O 2 (1 M) for 2 h at 37°C. We observed that the synthetic A␤ alone remained predominantly monomeric upon SDS-PAGE but that the combination of COX-2 and H 2 O 2 induced the conspicuous formation of an apparent A␤ dimer, as well as A␤ immunoreactive bands in the range of 50 -75 kDa (Fig. 1A). Incubation of A␤ with COX-2 alone induced far lower amounts of the apparent dimer and higher molecular weight A␤ immunoreactive products, but because the incubation was performed under aerobic conditions, small concentrations of H 2 O 2 may have been present because of the reduction and subsequent disproportionation of dissolved O 2 (26). This interpretation was indeed supported by the experimental effects of the H 2 O 2 scavenger catalase, which completely abolished all but monomeric A␤ immunoreactivity being detected upon co-incubation with A␤, COX-2, and H 2 O 2 (Fig. 1A). A␤ incubated with H 2 O 2 alone remained monomeric on SDS-PAGE (Fig. 1B).
These data established that H 2 O 2 acts in concert with COX-2 to convert monomeric A␤ into apparent dimeric and higher molecular weight A␤ immunoreactive products. Me 2 SO and mannitol, which scavenge OH . , did not have an inhibitory effect on the generation of these modified A␤ immunoreactivities ( Fig. 1A), suggesting that H 2 O 2 in the reaction is not inducing cross-linking through Fenton or Haber-Weiss chemistry. As a positive control, we also applied human A␤ reacted with horseradish peroxidase and hydrogen peroxide, which is known to induce dimeric dityrosine cross-linking of human A␤ (DT-A␤) (13). DT-A␤ co-migrated with the apparent A␤ dimers that we observed, and HRP, like COX-2 incubated under the same conditions, also appeared to form A␤ adducts (Fig. 1A). These data suggest that hydrogen peroxide might oxidize residues of both the COX-2-and A␤-inducing covalent cross-links, consistent with DT interstrand bridge formation for A␤ to A␤ and A␤ to COX-2.
Alzheimer neuropathology is only seen in animal species that have the human sequence of A␤ (27) and not in rats and mice, which have three amino acid substitutions (Arg 5 3 Gly, Tyr 10 3 Phe, and His 13 3 Arg) (28). The tyrosine substitution in the rat/mouse A␤ ("rat A␤") abolishes the possibility of DT interstrand bridge formation. To explore further whether COX-2 induces DT interstrand bridge formation, we repeated the experiment comparing the products of rat A␤ to human A␤ (Fig. 1B). Although incubation with COX-2/H 2 O 2 (but not H 2 O 2 alone) induced apparent dimer formation of human A␤ 1-40 as well as high molecular weight immunoreactivity, no such modifications of rat A␤ 1-40 were apparent even after 15 h of incubation (Fig. 1B).
The molecular mass of COX-2 is Ϸ70 kDa, and we determined that COX-2 itself did not cross-react with 4G8 antibody (Fig. 1B). Therefore, the A␤ immunoreactive band at Ϸ75 kDa when synthetic human A␤ was reacted with COX-2 and hydrogen peroxide (Fig. 1, A and B, asterisk) may indeed be a complex of human A␤ and COX-2. To confirm that this Ϸ75-kDa A␤ immunoreactivity reflects A␤ complex formation with COX-2, we immunoprecipitated the incubation products with a polyclonal anti-COX-2 antibody and immunoblotted the immunoprecipitates with anti-A␤ antibody (WO2). We again observed the 75-kDa band from the sample of human A␤1-40 reacted with COX-2 and hydrogen peroxide (Fig. 2, arrow) but not from the samples of COX-2, human A␤1-40 alone, or rat A␤1-40 incubated with COX-2 and hydrogen peroxide. Therefore, human A␤ may have adducted onto the COX-2 in this reaction. In addition, an extra band at Ϸ50 kDa was seen only in the

COX-2 Cross-links A␤
precipitate from the sample of human A␤ with COX-2 and hydrogen peroxide (Fig. 2, arrowhead), which may be a complex of human A␤ with a degraded COX-2 fragment. We next examined the effects of several peroxidase inhibitors on the cross-linking (Fig. 3A). Desferroxamine, an iron chelator, inhibited the apparent cross-linking completely, possibly by complexing iron from heme in COX-2. Sodium azide and sodium cyanide, inhibitors of peroxidase, also attenuated the cross-linking. The inhibition was more complete with cyanide compared with azide at the same concentration.
NSAIDs and aspirin were examined for effects on these peroxidative reactions (Fig. 3B). Indomethacin and ibuprofen, nonspecific inhibitors of COX-1 and COX-2, potently attenuated the apparent cross-linking of A␤ mediated by COX-2/H 2 O 2 . Aspirin, a more potent inhibitor of COX-1, had a lesser effect compared with the other inhibitors. Inactivation of COX-2 by heating at 95°C for 10 min abolished A␤ cross-linking (Fig.  3B), which excludes the possibility that iron in free heme itself might cause the cross-linking of A␤ by Fenton chemistry. Therefore, the oligomerization is caused by the enzymatic peroxidase activity of COX-2.
Arg 5 , Tyr 10 , and His 13 in human A␤1-40 are substituted in rat A␤1-40, which did not polymerize in the presence of COX-2 (Fig. 1B). Therefore, one of these three residues may be the site of dimeric cross-linking induced by COX-2/H 2 O 2 . As a preliminary analysis to identify the amino acid responsible for the cross-linking, we examined the effect of chemical modifiers for each amino acid (29) on the apparent cross-linking induced by COX-2/H 2 O 2 (Fig. 3C). Phenylglyoxal, an arginine modifier, did not have an apparent effect. N-Acetylimidazole, which acetylates tyrosine hydroxyl groups, partially inhibited the apparent cross-linking consistent with inhibiting the formation of dityrosine bridges. Diethylpyrocarbonate, a histidine modifier, did not inhibit the cross-linking but rather promoted the oligomerization of A␤. This result may indicate that the histidine residues play a role in minimizing radical reactivity or that the diethylpyrocarbonate is acting exceptionally in this system as a chemical cross-linker. Among the caveats with this approach is that the modifications are not completely selective for amino acids. Although future studies of amino acid substituted A␤ will be needed to consolidate these conclusions, these findings do not refute our earlier interpretation (Figs. 1 and 3, A and B) that DT cross-linking could be a consequence of the peroxidative activity of COX-2.
We next explored for evidence that COX-2 might react with A␤ in vivo. There is already abundant evidence for SDS-resistant A␤ dimers and polymers being enriched in AD-affected post-mortem brain tissue (2,30). Furthermore, there is evidence of elevated H 2 O 2 in AD-affected brain tissue (4). Our in vitro data indicated that whenever the combination of COX-2 and H 2 O 2 induced A␤ apparent dimeric cross-linking, a proportion of A␤ always adducted to COX-2 itself (Figs. 1-3). Therefore, if COX-2 indeed contributed to A␤ oligomeric cross-linking in vivo, the presence of such A␤-COX-2 complexes would be anticipated in post mortem human brain tissue as a biomarker. We investigated this possibility by co-immunoprecipitation of AD and age-matched control post-mortem human brain tissue with anti-A␤ and COX-2 antibodies, as in Fig. 1B. Two major bands were seen at approximately 70 and 50 kDa that were cross-immunoreactive with both antibodies (Fig. 4A). These bands were close to the molecular masses of the two major bands detected in anti-COX-2 immunoprecipitation of human A␤1-40 incubated with ovine COX-2 and H 2 O 2 , detected by A␤ immunoblot (Fig. 1B). Again, the 50-kDa band might be a degraded fragment of the complex possibly formed by the attack of free radicals. As a further negative control, we also probed these blots with antibody against human copper/zinc superoxide dismutase, which is widely expressed in the brain, and no immunoreactivity was detected (data not shown).
We surveyed the relative abundance of A␤-COX-2 complexes in post-mortem human brain samples obtained from patients with AD and age-matched controls. In moderately affected AD, A␤-COX-2 complexes were significantly increased compared with age-matched controls (p Ͻ 0.05, paired t test; Fig. 4B) in the COX-2 blot following A␤ immunoprecipitation but not in A␤ blots following COX-2 immunoprecipitation. The increase A␤-COX-2 complexes was more conspicuous in severely affected AD, where both COX-2 and A␤ blots revealed greater amounts of the immunoprecipitated complexes (p Ͻ 0.05, paired t test). This indicates that the A␤-COX-2 complexes become more abundant as the disease progresses. DISCUSSION Our data indicate that the peroxidative activity of COX-2 induces the dimerization of human A␤ by a H 2 O 2 -mediated mechanism, and the enzyme itself also cross-links with human A␤ directly. These cross-links were attenuated by a scavenger of H 2 O 2 and by peroxidase inhibitors. Our evidence, for the first time, implicates COX-2 activity in the direct oxidation of A␤, to generate SDS-resistant oligomeric A␤ forms that resemble forms that have previously been reported to be increased in the disease and proposed to mediate pathophysiology.
A␤ is extracted from AD brains as toxic water-soluble, SDSresistant oligomers (30,31). Soluble A␤ levels, including the SDS-resistant oligomers, are correlated to the disease severity of AD (32). Diffusible A␤ oligomers are deleterious to hippocampal neurons in vitro (33) and in vivo (34). Recent data have proposed that A␤ oligomers are generated intracellularly and secreted subsequently to form extracellular amyloid fibrils (35). Evidence suggests that the subcellular origin of these oligomers may be microsomes (35), the same compartment that contains COX. To date, none of the reports correlating SDS- FIG. 4. A␤-COX-2 complexes are elevated in AD-affected brain tissue. A, detection of A␤-COX-2 complexes by co-immunoprecipitation in human brains. Brain tissues from superior temporal area were homogenized, and the supernatants were immunoprecipitated with anti-COX-2 or A␤ (W0 -2) antibody. The precipitates were blotted with anti-A␤ (4G8) or COX-2 antibody, respectively. The background data of the cases are shown in Table I resistant apparent oligomers of A␤ with neurotoxicity have established the basis of the SDS resistance. A␤ will form SDSresistant oligomers to some extent in the absence of any apparent oxidative modification (Fig. 3A). However, our data establish that oxidation will lead to an apparent covalent crosslink that leads to similar SDS resistance. The possibility that oxidized A␤ SDS-resistant oligomers (e.g. produced by COX-2 activity) are more neurotoxic than the nonoxidized oligomers must be considered.
COX possesses both cyclooxygenase and peroxidase activities. The cycle of the peroxidase reaction can occur independently of cyclooxygenase activity, utilizing hydroperoxides such as hydrogen peroxide (15). During the peroxidase reaction, ferric heme of the resting enzyme is oxidized to form ferryl iron and a porphyrin radical (Intermediate I). The porphyrin radical is reduced to form a tyrosyl radical at Tyr 385 by an intramolecular electron transfer (Intermediate II). In the presence of reducing co-substrates, Intermediates I and II are returned to resting state (36). Phenolic compounds such as phenol (15), luminol (37), and dopamine (38,39) are preferred as reducing co-substrates. We hypothesized that the sole tyrosine residue of human A␤ could be a target for the peroxidative activity of COX-2 and that tyrosyl radicals of human A␤ formed by the reduction of oxidized COX-2 will cross-link each other to generate DT (40), emulating the reaction of HRP with A␤ (13). The A␤-COX-2 complex that we observed may also be formed from the bridging of tyrosyl radicals of oxidized COX-2 and A␤.
Human A␤ is a strongly redox active peptide that reduces copper and iron ions (8,12). The same properties of the peptide may lead to redox reactions with COX-2. The redox activity of A␤ also fosters the catalytic generation of hydrogen peroxide (6,8,9,12), which may supply the substrate to COX-2 for peroxidative cross-linking. A␤ 1-40 and A␤ 1-42 have high affinity for Cu 2ϩ (41) and are bound to the metal ion in AD-affected brain tissue (9,11). Therefore, A␤ accumulating in the ADaffected neocortex is very likely to be a source of H 2 O 2 . H 2 O 2 is freely permeable across lipid boundaries, which may contribute to its potency in mediating oxidative attack to neocortical cells (4,6). Contributions to increased H 2 O 2 in AD may also come from failing mitochondrial metabolism and from microglial activation. H 2 O 2 generated extracellularly (e.g. from A␤ or microglia) and intracellularly (e.g. from mitochondria) could migrate into microsomal compartments and then act as a cosubstrate for COX-2-mediated cross-linking of A␤. Pharmacological removal of excess copper and iron or scavenging of hydrogen peroxide may therefore be therapeutic strategies to prevent A␤ oxidation by COX-2. Such a mechanism may contribute to the efficacy of clioquinol, which blocks hydrogen peroxide production by A␤ by binding to copper and iron at its active site (42).
The tyrosine of human A␤ is thought to coordinate the metals that interact with the peptide (43,44), and we hypothesize that it may inappropriately coordinate with the metal active site of COX-2 (Fig. 5). The DT bond that may be formed in this reaction is chemically stable and resistant to proteolytic cleav-age, and so the catabolism of DT cross-linked proteins or peptides is inhibited (45). The total DT content is elevated in the regions affected by A␤ pathology in AD brains, in accordance with the regions where COX-2 expression is up-regulated (46). A polyclonal antibody raised against what may have been a DT-A␤ antigen detected highly abundant immunoreactivity in neuritic plaques (14).
The lack of tyrosine in rat (and mouse) A␤ might explain the scarcity of cerebral amyloid deposits in these animals (47). Transgenic mice that overexpress human COX-2 in neurons did not exhibit cerebral amyloid pathology (25), which would be consistent with the inability of endogenous mouse A␤ (lacking tyrosine) to cross-link. Supporting an essential role for the tyrosine residue in the human A␤ sequence causing amyloid pathology, cerebral A␤ deposition in double mutant presenilin 1/human amyloid precursor protein transgenic mice was markedly exaggerated upon crossing with the same line of human COX-2 transgenic mice (48).
Me 2 SO and mannitol, scavengers of hydroxyl radicals, did not inhibit the cross-linking of A␤ (Fig. 1A). This is consistent with the oxidative reaction of the peroxidase, which is not mediated by the production of hydroxyl radicals. Both sodium azide and sodium cyanide are general inhibitors of peroxidase. However, cyanide is known to have much stronger affinity to the COX peroxidase site than azide (15). Consistent with that, cyanide a showed stronger inhibitory effect on the cross-linking of A␤ (Fig. 3A).
Interestingly, nonselective NSAIDs attenuated the A␤ dimerization that was induced by COX-2 and COX-2 complex formation with human A␤ (Fig. 3B). NSAIDs inhibit cyclooxygenase activity more than the peroxidase activity of COX. However, NSAIDs such as indomethacin and flurbiprofen specifically suppress the radical formation at Tyr 385 of COX (36), suggesting that NSAIDs inhibited COX-2-A␤ cross-linking (Fig. 3B) by decreasing tyrosyl radicals in COX-2. Alternatively, this inhibitory effect of NSAIDs may be a consequence of the direct interactions of NSAIDs with A␤ peptide that have been described (49,50). The concentrations of indomethacin and ibuprofen used in this study are similar to those in the report by Weggen et al. (50), who demonstrated that some NSAIDs decrease the production of human A␤1-42. The prevention of A␤ cross-linking would facilitate A␤ clearance and may be a novel mechanism for the proposed protective effect of NSAIDs on AD.
We detected the possible complexes of A␤ and COX-2 from human brains by co-immunoprecipitation with antibodies against each component (Fig. 4). These complexes are increased in AD brains, especially in severely affected cases (Fig.  4). This may be due to the increased activity of COX-2 promoting more cross-linking of COX-2 and A␤ or due to diminished clearance of the complexes in AD. The 50-kDa band of fragmented COX-2 that we observed (Fig. 4A) has also previously been detected in AD brains (21). This is consistent with our current data (Fig. 2) indicating that reaction with A␤ may lead to the fragmentation of COX-2 into such a Ϸ50-kDa species that is adducted to A␤, whose concentrations appear to be increased in AD brain (Fig. 4).
The significance of the COX-2-A␤ complex is not clear, although we hypothesize that abnormal A␤ reaction with COX-2 may interfere with COX-2 integrity, function, or metabolism. More importantly, we noted that COX-2 mediates the production of A␤ dimers, accompanied by the invariable generation of such COX-2-A␤ complexes (Figs. 1-4). Therefore, the detection of COX-2-A␤ complexes in human tissue is evidence that these two proteins are in biochemical proximity in vivo and that therefore the peroxidation of A␤ by COX-2 in AD may indeed

COX-2 Cross-links A␤
contribute to the A␤ oligomeric forms that are increasingly considered to mediate the disease process. If the intracellular oligomerization of A␤ by COX-2 is an early event in the pathology of AD, its biochemical intervention may be a treatment strategy to halt or delay the disease progression at early stages.