Degradation of amyloid beta-protein by a metalloprotease secreted by microglia and other neural and non-neural cells.

Amyloid β-protein (Aβ) is the major component of neuritic (amyloid) plaques in Alzheimer's disease, and its deposition is an early and constant event in the complex pathogenetic cascade of the disease. Although many studies have focused on the biosynthetic processing of the β-amyloid precursor protein and on the production and polymerization of Aβ, understanding the degradation and clearance of Aβ has received very little attention. By incubating the conditioned medium of metabolically labeled Aβ-secreting cells with media of various cultured cell lines, we observed a time-dependent decrease in the amount of Aβ in the mixed media. The factor principally responsible for this decrease was a secreted metalloprotease released by both neural and non-neural cells. Among the cells examined, the microglial cell line, BV-2, produced the most Aβ-degrading activity. The protease was completely blocked by the metalloprotease inhibitor, 1,10-phenanthroline, and partially inhibited by EDTA, whereas inhibitors of other protease classes produced little or no inhibition. Substrate analysis suggests that the enzyme was a non-matrix metalloprotease. The protease cleaved both Aβ1-40 and Aβ1-42 peptides secreted by β-amyloid precursor protein-transfected cells but failed to degrade low molecular weight oligomers of Aβ that form in the culture medium. Lipopolysaccharide, a stimulator of macrophages/microglia, activated BV-2 cells to increase their Aβ-degrading metalloprotease activity. We conclude that secreted Aβ1-40 and Aβ1-42 peptides are constitutively degraded by a metalloprotease released by microglia and other neural cells, providing a potential mechanism for the clearance of Aβ in brain tissue.

The defining pathological features of Alzheimer's disease (AD) 1 are extracellular deposits of amyloid ␤-proteins (A␤) that form senile plaques and amyloid angiopathy and intraneuronal deposits of modified tau proteins that form neurofibrillary tangles. A␤ are 40 -43-amino acid proteolytic fragments generated by unidentified proteases from the transmembrane glycopro-tein, ␤-amyloid precursor protein (␤APP) (1,2). The gene encoding ␤APP is on human chromosome 21q, and missense mutations in and around the A␤ coding region of this gene are a rare cause of familial AD (FAD). Moreover, trisomy 21 (Down's syndrome) is characterized by overexpression of ␤APP due to increased gene dosage, resulting in very early A␤ deposition followed by the gradual development of the classical neuropathological lesions of AD (3,4). Recently, mutations in the presenilin 1 and 2 genes, which cause severe early onset FAD (5)(6)(7), have been shown to selectively increase the production and cerebral deposition of the highly amyloidogenic 42residue form of A␤ (8,9). These and other findings provide strong evidence that disordered ␤APP metabolism can increase the production of A␤ peptides, particularly the A␤ 1-42 peptide, thereby initiating amyloid plaque formation and the pathological cascade of AD. In support of this model, transgenic mice overexpressing the V717F mutation or K670N/M671L mutation of ␤APP progressively develop A␤ deposits and plaqueassociated neuritic, microglial, and astrocytic pathology with age and may even show concomitant memory impairment (10 -12).
Much attention has been focused on the secretory and endocytic processing of ␤APP by cells in order to understand how A␤ is generated normally and in AD. In contrast, little is known about how A␤, once secreted, is degraded and cleared from tissues. At present, only the relatively rare forms of FAD linked to the ␤APP or presenilin genes are thought to involve overproduction of A␤. The excessive cerebral accumulation of A␤ that occurs in all other cases of the disease could be explained in part by a decrease in the ability of the brain to degrade and clear A␤. If specific A␤-degrading proteases can be shown to be released by neural cells, changes in the structure or activity of such proteases could be sought in as yet unexplained forms of FAD, and their up-regulation could represent a therapeutic approach to AD in general.
Here, we have screened certain neural and non-neural cell lines to ascertain whether they constitutively release proteases capable of degrading the A␤ peptides naturally secreted by cells. We have identified a secreted metalloprotease activity that efficiently degrades endogenous A␤ into fragments under cell culture conditions. Among the neural cell lines tested, a microglial line, BV-2, secretes the active protease robustly. Substrate analysis suggests that it is a non-matrix metalloprotease. Its production or activity is up-regulated by activating BV-2 cells. Interestingly, the metalloprotease is far more effective in degrading secreted A␤ 40 and A␤ 42 monomers than A␤ oligomers formed in cell culture. These findings have implications for the normal and pathological clearance of A␤ in brain.

EXPERIMENTAL PROCEDURES
Cell Culture-Chinese hamster ovary (CHO) cells stably transfected with ␤APP 770 cDNA containing the Val 3 Phe mutation at residue 717 (7PA2 cells) (13) were routinely cultured in Dulbecco's modified Eagle's * This work was supported by National Institutes of Health Grants AG 06173 and AG 12749 and grants from the Richard Saltonstall Charitable Trust and the Alder Foundation. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Assays of A␤-degrading Activity-To obtain conditioned media (CM) containing A␤-degrading activity, different cultured cells were washed 3 times with serum-free N2 medium (N2 supplement (Life Technologies, Inc.), 1% ovalbumin, 1 mM pyruvate in MEM) (N2). N2 was added, and cells were conditioned at 37°C for various times. CM were collected and centrifuged at 3,000 ϫ g for 30 min to remove cells. Some CM were concentrated 10 times in a Centricon 30 (Amicon) filter. To characterize A␤-degrading activity, confluent monolayers of 7PA2 ␤APP-transfected CHO cells in 10-cm dishes were preincubated for 30 min in methioninefree medium and labeled for 4 h with 300 Ci of [ 35 S]methionine. The labeled media were collected, combined, and centrifuged at 3,000 ϫ g for 30 min. An amount of labeled medium (3 ml) was mixed with an equal amount of cell conditioned medium from the various cells being tested or with unconditioned N2 medium as a control, and the mixtures were incubated at 37°C for 16 h. The amount of labeled A␤ remaining in each sample was assessed by immunoprecipitation with the high affinity A␤ antibody, R1282 (4). This antibody was generated to synthetic A␤  peptide and characterized similarly to our A␤ antibody, R1280, used in our previous study (15).
A␤ degradation in cultures was also quantified by incubating 7 l of conditioned or unconditioned medium with 125 I-labeled A␤1-40 (3.3 ϫ 10 4 cpm) (kind gift of Dr. John Maggio) (15) in 14 l of reaction buffer (187 mM NaCl, 0.02 M NaH 2 PO 4 , 10 mM Tris-HCl, pH 7.5) at 37°C for 16 h. After quenching the reaction with 7 l of glacial acetic acid and 5 l of 0.1% pyronin Y, the sample was analyzed by acid-urea-PAGE (15) and autoradiography.
Protease inhibition assays were conducted by adding different protease inhibitors at the indicated concentrations to the mixtures before incubating at 37°C for 16 h. All inhibitors shown in Table I were purchased from Sigma.
ELISA-Cold 4-h conditioned medium of 7PA2 cells was mixed with plain N2 or with CM of CHO or BV-2 cells. After incubating at 37°C for 16 h, total A␤ or specific A␤ 1-42 remaining in each sample was quantified by highly specific sandwich ELISA assays described previously (16). Two different ELISA assays were performed using distinct capture antibodies: 266 (raised against residues 13-28 of A␤), for total A␤ or 21F12 (raised against residues 33-42 of A␤), for A␤ peptides ending specifically at residue 42. The reporter antibody was 3D6 (to residues 1-5 of A␤) in both assays. The ratio of A␤ 1-42 over total A␤ in each sample was calculated.
LPS Treatment-Bacterial lipopolysaccharide (LPS) (Escherichia coli serotype O111:B4) was purchased from Calbiochem. Cells were either left unstimulated or stimulated with LPS (10 g/ml) in culture medium containing 10% FBS at 37°C for various times. Conditioned media were collected and centrifuged before assaying A␤-degrading activity as described above.

Decrease of A␤ Is Mediated by the Conditioned Media of
Non-neural and Neural Cells-We examined several non-neural and neural cell lines for the secretion of an A␤-degrading protease activity: COS monkey kidney cells; Chinese hamster ovary (CHO) cells; the human neuroblastoma cell lines, M17 and SY5Y; and the murine microglial line, BV-2. All cultures were passaged by brief incubation with EDTA rather than trypsin, in order to avoid the formation of a serine protease/␣ 2macroglobulin complex in the medium, which we recently showed can occur during tryptic passage of cells and degrade secreted A␤ (15). To search for A␤-degrading activity, confluent cultures were washed and changed to the serum-free medium, N2, for further cultivation. After conditioning for 24 h, the media were collected and centrifuged to remove floating cells. CM from certain cell lines (CHO, COS, M17, and SY5Y) were concentrated 10-fold by Centricon filtration before assaying for proteolytic activity. To detect degradation of endogenously secreted A␤, CHO cells stably transfected with ␤APP 770 cDNA containing the V717F mutation (7PA2 cells; Ref. 13) were labeled with [ 35 S]Met in serum-free medium for 4 h, and the resultant medium (containing abundant labeled A␤ and p3 (17) peptides) was incubated for 16 h at 37°C with either unconditioned N2 medium or the CM of the aforementioned cell lines. Because no cells are present during the incubation, any loss of A␤ and p3 cannot be due to internalization of the peptides or any other cell-associated function. By immunoprecipitating the mixed media with a high titer A␤ antibody (R1282) and performing gel fluorography, we observed variable declines in the amounts of A␤ and p3 compared to those in plain N2 medium (Fig. 1). The amounts of total secreted proteins and APP s in the incubated CM changed little or not at all (data not shown). Among the cell lines we examined, CHO and BV-2 CM produced the largest loss of A␤ after incubation (Fig. 1).
To obtain quantitative information about A␤ degradation by these two cell types, we assayed the CM obtained from cultures of known cell densities grown in a standard volume of medium. The amount of A␤ remaining in the CM of 2.5 ϫ 10 7 BV-2 cells was consistently less than that in CM of this number of CHO cells ( Fig. 2A), indicating that the microglial cell line secreted more A␤-degrading activity. Conditioning these cell lines for increasing intervals revealed that the confluent CHO cells produced little or no A␤-degrading activity in the first 6 h and generated maximal activity by 24 h (Fig. 2B). In the case of BV-2 cells (1 ϫ 10 7 ), conditioning for 2 h already produced substantial activity, further suggesting that the microglial cell CM caused the most degradation among the lines we

characterized.
A␤ Degradation Is Mediated by a Secreted Metalloprotease in Both CHO and BV-2 Conditioned Media-To determine whether that the loss of A␤ in CM was caused by degradation by secreted protease(s), we examined the effects of several different protease inhibitors with an assay similar to that described above. Using concentrations known to cause maximal inhibition of other secreted proteases (18), various well characterized inhibitors were incubated for 24 h at 37°C with either N2 medium or CHO CM in the presence of synthetic A␤ 1-40 peptide (40 ng/ml) as a substrate, followed by a sandwich ELISA to detect remaining intact A␤. The inhibitor profile clearly demonstrated an A␤-degrading metalloprotease in the CHO medium (Table I). The degradation of cell-secreted A␤ and p3 by both CHO and BV-2 media was essentially abolished by 1,10-phenanthroline, whereas EDTA generally produced less inhibition (Fig. 3). The different degrees of inhibition of the degradation of endogenous (Fig. 3) versus synthetic (Table I) A␤ by EDTA are unexplained, but may be due to different conformations of these two substrates, and we are conducting experiments to address this issue. Although phosphoramidon is also a metalloprotease inhibitor, it did not inhibit the degradation of either synthetic or endogenous A␤ in this system (Fig.  3). These results indicate that the major A␤-degrading activity present in both CHO and BV-2 CM is a secreted metalloprotease. Another protease inhibitor, 4-(2-aminoethyl)-benzenesulfonyl fluoride (Pefabloc), produced slight inhibition of A␤ degradation (Fig. 3), suggesting that a serine protease may also make a small contribution to A␤ proteolysis by CHO and BV-2 media.
To detect the products of the A␤ degradation, acid-urea-PAGE (15) was used to analyze 125 I-labeled synthetic A␤ 1-40 after incubation in CHO or BV-2 CM for 24 h (Fig. 4). The amount of intact A␤ decreased markedly, and 3 peptide bands appeared that migrated differently from intact A␤ and were not produced by incubation in N2 medium alone. The production of these fragments was completely inhibited by 1,10-phenanthroline, confirming that A␤ is degraded by a secreted metalloprotease in both CHO and BV-2 CM and generates similar peptide products.
Degradation of A␤ 40 , A␤ 42 and A␤ Oligomers by the Metalloprotease-Because A␤ peptides ending at residue 42 (A␤  ) are the initially deposited species in AD and normal aged brain tissue, we examined the substrate specificity of the metalloprotease on endogenous A␤ 1-40 versus A␤ 1-42 peptides. Confluent 7PA2 cells were washed and cultured in serum-free N2 medium for 4 h. Equal amounts of this CM were incubated with plain N2, CHO CM, or BV-2 CM for 24 h, followed by sandwich ELISAs measuring either total A␤ (including A␤ 1-40 and A␤  or A␤ 1-42 alone. The amounts of total A␤ and A␤ 1-42 remaining in CHO and BV-2 media were decreased approximately 75% compared to those in N2 (Fig. 5, A and B). As a result, the ratios of A␤ 1-42 to total A␤ showed only a slight and insignificant increase (Fig. 5C).
We also characterized the specificity of the metalloprotease in the 24-h conditioned media of CHO and BV-2 cells for A␤related species found in the culture medium of our ␤APP-  Fig. 1. B, confluent CHO cells were grown in N2 medium at 37°C for 2, 6, 16, or 24 h. These CM and unconditioned N2 medium (lane 1) were incubated with the labeled 7PA2 medium, followed by immunoprecipitation with R1282. C, BV-2 cells (1 ϫ 10 7 ) were grown in N2 medium for 0.5, 1, 2, 6, 16, or 24 h. The CM and unconditioned N2 medium (lane 1) were incubated with labeled 7PA2 medium followed by R1282 immunoprecipitation.

TABLE I Inhibition profile of A␤-degrading enzyme in the CHO
conditioned medium Amounts of exogenously added A␤ 1-40 remaining undegraded after incubation in the CHO conditioned medium in the presence of different protease inhibitors were determined using a sandwich ELISA assay. The amount of A␤ 1-40 after incubation in the unconditioned medium, N2, was 4.10 ng/ml, while the amount of A␤ 1-40 remained in the CHO conditioned medium without the protease inhibitors is 0.37 ng/ml. Therefore, the A␤-degrading activity in the original CHO conditioned medium was set as 0% inhibition; recovery of A␤ 1-40 level to that in N2 medium represents 100% inhibition. TPCK, N-tosyl-L-phenylalanine chloromethyl ketone; DFP, diisopropyl phosphofluoridate.  expressing 7PA2 cells. A␤ was substantially decreased by 4 h of incubation, whereas p3 showed relative stability for the first 8 h of incubation (Fig. 6). Importantly, the endogenous 6 -12-kDa oligomeric A␤ species that we previously documented in the CM of these V717F ␤APP-transfected CHO cells (13) were completely resistant to proteolysis (Fig. 6).

FIG. 3. Degradation of A␤ by CHO or BV-2 conditioned media is blocked by certain metalloprotease inhibitors. A␤ secreted by 7PA2 CHO cells was analyzed as in
Regulation of the Secreted Metalloprotease-During the course of these studies, we noticed that high A␤-degrading activity in CM (Ͼ90% decrease of A␤ compared to that in N2) occurred when cells were conditioned in the absence of serum.
In contrast, only a 10 -15% decrease in A␤ occurred when 10% FBS was present during conditioning (data not shown). To clarify the role of serum in the production and activity of the A␤-degrading protease, we conditioned CHO cells in N2 medium containing increasing concentrations of FBS for 24 h. These CM were then incubated with labeled CHO 7PA2 medium at 37°C for 16 h, followed by the A␤ immunoprecipitation assay described above. As a control, 10% FBS was added to a sample of medium conditioned without serum just prior to the incubation. CHO CM containing 1% FBS degraded A␤ as much as CHO CM without FBS, whereas the amount of A␤ in CHO CM containing 10% FBS appeared virtually unchanged from that in plain N2 (Fig. 7). Increasing the serum concentration between 1% and 10% led to a graded decrease in the amount of A␤-degrading activity (Fig. 7). Using the ELISA assay, we were similarly able to detect only a 10 -20% decrease in A␤ levels in medium conditioned in 10% FBS (data not shown). These results indicate that increasing the percentage of serum during conditioning results in a corresponding decrease in A␤ proteolysis. On the other hand, adding 10% FBS solely during the incubation period did not decrease the strong A␤-degrading activity generated by CHO cells conditioned without serum. Serum also decreased the metalloprotease activity released from BV-2 cells but to a lesser degree than CHO cells (Fig. 8B).
Thus, serum appears to decrease the cellular production of the A␤-degrading metalloprotease or increase the secretion of a metalloprotease inhibitor rather than directly inhibiting the protease.
To further characterize the regulation of the metalloprotease in the microglial line, E. coli LPS, a general stimulator of macrophage and microglial cells (14), was applied to the cells during conditioning in the presence of 10% FBS, since the latter generally decreased the amount of A␤-degrading activity. LPS stimulated the BV-2 cells to release more A␤-degrading protease, but failed to do so as expected in the CHO fibroblast line (Fig. 8A). As the conditioning time in LPS was prolonged, the stimulation of A␤-degrading activity in BV-2 CM rose. The released protease was specifically blocked by 1,10-phenanthroline (Fig. 8B), suggesting that the protease activity stimulated by LPS in the presence of serum represented the same metalloprotease observed constitutively in the absence of serum.

DISCUSSION
The degree of A␤ deposition seen in AD and aged normal brains is determined by the rates of both A␤ production and A␤ removal. Therefore, it is important to understand the contribution of A␤ removal under normal and pathological circumstances, identify its mechanisms, and search for methods to enhance clearance. In this report, we describe a metalloprotease secreted by both non-neural and neural cells that is capable of efficiently cleaving A␤ and p3 but not oligomeric A␤ species found endogenously in the medium of ␤APP-expressing cells.
Our studies using [ 35 S]methionine-labeled CM of ␤APPtransfected CHO cells as a source of A␤ and related species and the CM of different cell lines as a source of proteases show that several non-neural and neural cell lines secrete A␤-degrading proteases, although the levels of activity vary substantially among cell types. Because the two culture media were incubated together in the absence of cells, our paradigm excludes the possibilities of cell-mediated internalization of A␤ (19,20) or the degradation by a cell surface protease, if such exists. By using EDTA to lift the cells during passage, we also excluded the possibility that the serine protease/␣ 2 -macroglobulin complex we previously described in vitro (15) was responsible for A␤ degradation. Two different methods, immunoprecipitation and ELISA, were employed to assess the amounts of A␤ remaining in the media after incubation and to confirm unequivocally that it was A␤ itself that was the substrate. We consistently observed the degradation of A␤ by proteases secreted by two non-neural cell lines, CHO and COS, and three neural cell lines, M17 and SY5Y neuroblastoma and BV-2 microglial cells (Fig. 1). Among them, BV-2 and CHO showed the strongest activity, followed in decreasing order by COS, M17, and SY5Y. BV-2, a murine microglial cell line, released the most A␤degrading activity per cell (Fig. 2), suggesting that microglial cells could play an important role in the clearance of A␤ from the extracellular space of brain. Furthermore, unlike the CHO fibroblast cells, which released little A␤-degrading activity into medium in the first 6 h of conditioning, BV-2 cells immediately and continually secreted proteolytic activity after being placed in serum-free medium (Fig. 3).
The metalloprotease inhibitors 1,10-phenanthroline and EDTA substantially blocked the degradation of endogenous A␤ by the CM of both CHO and BV-2 cells (Fig. 3). Phosphoramidon, another metalloprotease inhibitor, which can only block certain metalloproteases (18), lacked the ability to inhibit the protease we describe. Selected inhibitors of the cysteine and aspartyl classes likewise produced little or no detectable inhibition of A␤ degradation under the conditions of our experiments. However, Pefabloc, a broad spectrum serine protease inhibitor, modestly inhibited the catabolism of A␤ by the conditioned media we examined (Fig. 3), suggesting that a serine protease may contribute to A␤ degradation in our cultures, although it plays a minor role. Naidu et al. (22) have reported that a mixture of inhibitors of all four major protease classes (leupeptin, pepstatin, EDTA, and phosphoramidon) inhibited A␤ degradation by CHO cells. In the cell lines we used, a metalloprotease is responsible for the major A␤ degradation, as confirmed when the specific A␤-derived proteolytic products seen in acid-urea gels were completely abolished by 1,10-phenanthroline (Fig. 4). This inhibitor was not tested by Naidu et al.
Because the major class of secreted metalloproteases is the matrix metalloproteases, we have recently characterized both CHO and BV-2 CM by chromatography on gelatin-Sepharose, a FIG. 6. p3 and low molecular weight A␤ oligomers are relatively resistant to degradation by the metalloprotease. Overnight labeled 7PA2 medium was incubated with unconditioned N2 medium (lanes 1, 4, 7, 10, and 13), CHO CM (lanes 2, 5, 8, 11, and 14) or BV-2 CM (lanes 3, 6, 9, 12, and 15) at 37°C for the indicated reaction times. Labeled A␤ and p3 (arrows) were then precipitated with R1282 and analyzed as in Fig. 1. Labeled oligomeric A␤ bands (8,12, and 16 kDa), which were fully characterized previously (13), are shown above the monomeric A␤.  10% (lane 7). These CM were then incubated with the labeled 7PA2 medium for another 16 h, and the remaining amounts of A␤ analyzed as in Fig. 1. As controls, unconditioned N2 medium (lane 1) and medium from cells conditioned without serum to which 10% FBS was then added (lane 8) were also incubated with the labeled 7PA2 medium.

FIG. 8. Stimulation of BV-2 cells by LPS enhances A␤-degrading activity.
A, RPMI and 10% FBS alone (lanes 1 and 2) or with CHO cells (lanes 3 and 4) or with BV-2 cells (lanes 5 and 6) were conditioned at 37°C for 5 h in the absence (lanes 1, 3, and 5) or presence (lanes 2, 4, and 6) of LPS. The CM were collected and A␤-degrading activity detected as in Fig. 1 matrix metalloprotease binding gel (23). The flow-through fractions of this column showed no loss of A␤-degrading activity. Moreover, p-aminophenylmercuric acetate, a matrix metalloprotease activator, did not change the level of A␤ degradation of CHO and BV-2 CM, whereas it increased the degree of gelatin degradation by these CM, as shown by gel zymography. 2 These results suggest that the protease we have characterized belongs to a class of metalloprotease different than the principal matrix metalloproteases.
Because accumulating studies have shown that A␤ 42 peptides are the initial species involved in the formation of amyloid deposits in AD and Down's syndrome (4,8,24,25) despite the fact that cells secrete much more A␤ 40 than A␤ 42 (8), we characterized the specificity of the metalloprotease for secreted peptides ending at A␤ 40 and A␤ 42 . Both CHO and BV-2 CM decreased the amounts of total A␤ and A␤ 42 to similar levels and left the ratio of A␤ 42 to total A␤ largely unchanged (Fig. 5), suggesting that the metalloprotease has equal avidity for endogenous A␤ 1-40 and A␤ 1-42 peptides. We found that the p3 peptide (A␤ 17-40/42 ) was more resistant to degradation by the metalloprotease than A␤ (Fig. 6). Interestingly, the low molecular weight A␤ oligomers that can be detected in CHO medium (13) and that are composed of both A␤ 40 and A␤ 42 peptides 3 were unchanged after 48 h incubation in conditions that allowed virtually complete degradation of monomeric A␤ and p3. Therefore, we speculate that the threshold concentration of monomeric A␤ needed to allow aggregation could be reached by increased production and/or impaired degradation of A␤. After A␤ 40 and A␤ 42 peptides are oligomerized, they will apparently be cleared inefficiently by secreted proteases, even those released by activated microglial cells in senile plaques. These oligomers could then accumulate and ultimately form high molecular weight assemblies, including the potentially neurotoxic amyloid fibrils characteristic of AD.
Because the clearance of A␤ should help determine the levels of extracellular A␤ and thus the rate of fibril formation, methods to stimulate A␤ degradation represent one approach to slowing the development of AD neuropathology. During our experiments, we noticed that the activity of A␤-degrading metalloprotease decreased in parallel with increasing levels of serum in the cultures (Fig. 7), suggesting that the enzyme is probably regulated by extracellular factors under physiological conditions. It is currently unclear whether this loss of activity is caused by the increased secretion of a protease inhibitor by the cells or by decreased release of the protease itself. We also found that LPS, a general stimulator of macrophage/microglia, activated BV-2 cells to increase A␤-degrading metalloprotease activity in their media under the proteolytically adverse condition of culturing in 10% serum (Fig. 8). It is possible that the activated microglia which are consistently found in mature (neuritic) plaques but less frequently present in immature (diffuse) plaques (21) release proteases such as that described here. Our data lead to the hypothesis that at early stages of AD, microglial cells could play an active role in slowing the development of A␤ deposits and amyloidosis; however, at later stages there could be an adverse effect, because the released metalloprotease cannot efficiently degrade oligomeric A␤ but would be available to act upon other cellular and extracellular substrates, thereby potentially aggravating the inflammatory and cytotoxic events in and around the mature plaque. Full purification of the metalloprotease should allow the development of antibodies and probes useful for examining the role of this and related proteases in the normal and abnormal biology of A␤.