Neuroserpin Binds Aβ and Is a Neuroprotective Component of Amyloid Plaques in Alzheimer Disease*

Alzheimer disease is characterized by extracellular plaques composed of Aβ peptides. We show here that these plaques also contain the serine protease inhibitor neuroserpin and that neuroserpin forms a 1:1 binary complex with the N-terminal or middle parts of the Aβ1-42 peptide. This complex inactivates neuroserpin as an inhibitor of tissue plasminogen activator and blocks the loop-sheet polymerization process that is characteristic of members of the serpin superfamily. In contrast neuroserpin accelerates the aggregation of Aβ1-42 with the resulting species having an appearance that is distinct from the mature amyloid fibril. Neuroserpin reduces the cytotoxicity of Aβ1-42 when assessed using standard cell assays, and the interaction has been confirmed in vivo in novel Drosophila models of disease. Taken together, these data show that neuroserpin interacts with Aβ1-42 to form off-pathway non-toxic oligomers and so protects neurons in Alzheimer disease.

Alzheimer disease is the most common form of dementia. The pathological features are characterized by neurofibrillary tangles and extracellular A␤ plaques. The plaques are composed of 42 (A␤  )-and to a lesser extent 40 (A␤ 1-40 )-amino acid fragments of the amyloid precursor protein (1). Overproduction of the more aggregatory A␤  peptide is believed to cause neuronal dysfunction and death in most sporadic and familial forms of Alzheimer disease. Although insoluble plaques of A␤ 1-42 are a classic feature of the brains of patients with Alzheimer disease, these plaques are also present in some healthy, elderly individuals (2). This discrepancy, and the observation that soluble A␤ is a better marker of cognitive decline (3), has led to the proposal that mature amyloid plaques are an end stage aggregation product and that the directly neurotoxic species occur earlier in the aggregation pathway (4,5). The description of soluble oligomers of A␤ that can cause neuronal dysfunction and death (4 -6) has emphasized the importance of understanding the pathways and kinetics of A␤ aggregation. The presence of ancillary proteins that interact with A␤ may stabilize particular aggregation intermediates or seed-specific patterns of aggregation that have distinct toxic properties. Of particular interest in this regard are several proteins that are associated with ␤ amyloid plaques such as apolipoprotein E (7) and the serine protease inhibitor (serpin) ␣ 1 -antichymotrypsin (8). The role of these proteins is underscored by the finding that the E4 polymorphism of apolipoprotein E is the most powerful genetic risk factor for the development of sporadic Alzheimer disease (9,10), most likely because it accelerates the deposition of A␤ in the brain (11)(12)(13). ␣ 1 -Antichymotrypsin is also found in the majority of senile plaques (8), with cerebrospinal fluid concentrations being consistently raised in patients with Alzheimer disease (14). Depending on the relative molar ratio, ␣ 1 -antichymotrypsin may accelerate or inhibit the aggregation of the A␤ peptide in vitro (15,16).
We have recently shown that mutants of the neuron-specific serpin, neuroserpin, underlie a novel inclusion body dementia that we have called familial encephalopathy with neuroserpin inclusion bodies (17)(18)(19)(20)(21). Wild-type neuroserpin is expressed throughout the nervous system and inhibits the serine protease tissue plasminogen activator (tPA) 4 (19,22,23). tPA plays a critical role in neural development, synaptic plasticity, and memory (24), and its levels rise in animal models of both stroke and epilepsy (25)(26)(27). Neuroserpin expression is up-regulated in the neurons surrounding the ischemic core in experimental stroke, and its therapeutic administration in animal models reduces infarct size and apoptosis in the stroke penumbra (25). Neuroserpin administration has also been shown to attenuate seizure progression in animal models of epilepsy (27).
In view of the neuroprotective role of neuroserpin we have assessed whether it may be important in Alzheimer disease. We show here that neuroserpin is a plaque-associated protein in the brains of patients with Alzheimer disease and that it forms a specific binary complex with the A␤ peptide. The neuroprotective consequences of this interaction are demonstrated in cell culture and Drosophila models of disease.

EXPERIMENTAL PROCEDURES
Materials-Chemicals and buffers, Dulbecco's modified Eagle's medium (DMEM), horse serum, fetal bovine serum, penicillin and streptomycin, B27 supplement, and thioflavin T were all from Sigma. PC12 (rat pheochromocytoma) cells were from the American Type Culture Collection (Manassas, VA). Unless otherwise stated, all experiments were carried out in PBS. The A␤ 1-40 peptides substituted with proline residues were obtained from the Keck Biotechnology Center, Yale University.
Immunohistochemistry-Immunostaining was performed on tissue fixed in 10% w/v buffered formalin for at least 2 weeks and embedded in paraffin before staining. Routine histological procedures were employed throughout, and staining used the streptavidin-biotin methodology and 3,3Ј-diaminobenzidine, with or without nickel enhancement, as the chromagen. The monoclonal 6F/3D (Novocastra Laboratories Ltd., UK) was used to detect the A␤ peptides, and neuroserpin was detected using an affinity-purified, polyclonal rabbit antiserum (23,28).
Expression and Purification of Neuroserpin-Wild-type recombinant neuroserpin was expressed with a 6-histidine tag at the N terminus in the pQE81L vector in Escherichia coli SG13009 (pREP4) cells and purified to homogeneity as described previously (19 -21). The resulting protein was dialyzed into PBS, or into DMEM for use in cell experiments, concentrated, and then stored at Ϫ80°C.
Assessing the Interaction between A␤ Peptides and Neuroserpin-The effect of A␤ 1-40 and A␤ 1-42 peptides on neuroserpin polymerization was determined by incubating 0.8 mg/ml recombinant neuroserpin at 45°C in the presence or absence of a 10-or 50-fold molar excess of A␤ 1-40 or A␤  . Neuroserpin polymerization was also assessed in the presence or absence of N-and C-terminal fragments of the A␤ 1-42 peptide and with mutants of the A␤ 1-40 peptide that contained proline substitutions (F20P, I31P, and M35P). Samples were taken at various time points and snap frozen before being assessed by 7.5% (w/v) non-denaturing PAGE.
The formation of an SDS-stable complex between neuroserpin and tPA was demonstrated by co-incubating 2 M neuroserpin with 2 M tPA (Calbiochem) in PBS for 5 min at room temperature. The effect of A␤ 1-42 on complex formation was assessed by incubating neuroserpin with or without a 10ϫ molar excess of A␤ 1-42 for 24 h at 37°C, followed by a 5-min incubation with tPA at room temperature. To control for any effect of A␤ 1-42 on tPA, 2 M neuroserpin, 2 M tPA, and 20 M A␤ 1-42 were incubated together for 5 min. The reaction between neuroserpin and tPA was stopped by the addition of 1,5-dansyl-Glu-Gly-Arg-chloromethylketone (1 mM final concentration, Calbiochem) to inhibit any residual free tPA (29). Following the final incubation in each condition, the samples were snap frozen in liquid nitrogen and stored at Ϫ80°C until the completion of the experiment. The samples were then mixed with SDS-PAGE loading buffer and boiled for 5 min, and the proteins were separated on a 10% w/v SDS-PAGE gel and visualized by silver staining.
N-terminal Amino Acid Sequencing-Recombinant neuroserpin (20 M) was incubated with a 40ϫ molar excess of A␤ 1-42 for 30 h at 37°C. The samples were then resolved by 7.5% (w/v) non-denaturing PAGE and electroblotted onto polyvinylidene difluoride (Immobilon, Millipore Corp.). Following transfer the polyvinylidene difluoride membrane was stained with Ponceau-S to visualize the bands. The band containing the neuroserpin-A␤ 1-42 complex was cut out and analyzed by quantitative N terminus amino acid sequencing.
Assessment of A␤ Aggregation with Thioflavin T-230 M A␤ 1-42 was incubated with a range of concentrations of neuroserpin (0, 2, 8, and 12 M) in PBS at 37°C. At various time points, up to 30 min, a 2-l aliquot from each incubation mixture was diluted into 298 l of 20 M thioflavin T in 50 mM glycine, pH 8.5. Fluorescence was measured in a 2-mm path length quartz cuvette in an LS50B luminescence spectrophotometer (PerkinElmer Life Sciences). The excitation and emission wavelengths were 450 and 485 nm, respectively, with a 2.5 nm band pass for excitation and 15 nm for emission.
Electron Microscopy-A␤ 1-42 and neuroserpin were incubated as described above for the thioflavin T experiments. At various time points aliquots were removed and either diluted 1:1 with buffer and placed directly on copper grids or centrifuged at 10,000 ϫ g for 10 min, and the pellets were placed on the grids. The grids were negatively stained with 2% w/v uranyl acetate in water and then examined and photographed using a Hitachi H7100 microscope operated at 75 kV.
Cell Culture Experiments-Rat pheochromocytoma (PC12) cells were maintained in DMEM with 10% (v/v) heat-inactivated horse serum, 5% (v/v) fetal bovine serum, 100 unit/ml penicillin, 100 g/ml streptomycin, and 2 mM L-glutamate. The cells were plated at a density of 5000 -7000 cells/well in a 96-well plate at 37°C in 10% v/v CO 2 until 70 -80% confluency was achieved. Primary cortical neurons from E14 rat embryos were obtained as previously described (30) and cultured in B27 media (DMEM, 1% (w/v) penicillin/streptomycin, 2% v/v B27 supplement), plated at a density of 1000 cells/well in a 96-well plate and incubated at 37°C in 5% v/v CO 2 for 48 h. A␤  peptide (final concentration, 50 M) and a range of concentrations of neuroserpin (both in DMEM) were then added to the cell media of the PC12 cells and the primary cortical rat embryo neurons followed by further incubation at 37°C for 72 and 40 h, respectively. Mitochondrial activity of the cells was determined by the MTT assay. Quantification of cell death was performed using spectrophotometric measurement at 540 nm with a reference wavelength of 690 nm for the PC12 assays and 570 nm for the primary rat cortical neuronal culture assays.
Transgenic Flies-Wild-type neuroserpin with its human secretion signal peptide and the A␤ 1-42 peptide, fused to a secretion signal peptide from the Drosophila necrotic gene (31) were cloned into the GAL4-responsive pUAST expression vectors. The lines expressing wild-type neuroserpin had the transgene inserted on chromosomes 2 (NS16) and 3 (NS9). In the A␤ 1-42 -expressing lines the inserts were on chromosomes 2 (Alz1) and 3 (Alz2). Flies co-expressing neuroserpin and A␤  were generated by crossing the respective single transgenic lines. The lines containing the GAL4-responsive constructs were then crossed with various drivers (GAL4 da , GAL4 Act5c , GAL4 GMR , GAL4 sev ) to drive expression in various tissues (da and Act5c widespread, and GMR and sev primarily in the eye). Hatching frequency observation and scanning electron microscopy were both performed on flies cultured at 25°C and 29°C as described previously (32).

Neuroserpin Antibodies Selectively Label A␤ Deposits in
Alzheimer Disease-Neuroserpin was detected in the brains of seven patients with sporadic Alzheimer disease using a polyclonal antiserum to recombinant human neuroserpin. The relative distributions of neuroserpin and ␤-amyloid peptides were compared using the 6F/3D monoclonal antibody that detects A␤ peptides (Fig. 1). Both neuroserpin and A␤ were detected at high levels in the CA1 region of the hippocampus; A␤ was present as plaques (Fig. 1a); however, neuroserpin ( Fig. 1b) demonstrated a widespread intra-neuronal distribution with some puncta of peri-and intra-neuronal staining. This discrete staining for neuroserpin was only seen in association with plaques in tissue from patients with Alzheimer disease and was not seen in control tissue. Neuroserpin (Fig. 1c, brown stain) was found at the periphery of the amyloid core (arrow) of plaques, associated with dystrophic neurites and swollen axonal processes (arrowheads). Staining of the amyloid plaques for both A␤ and neuroserpin (Fig. 1d) demonstrated co-localization of the A␤ peptide (darker stain) with neuroserpin (lighter stain, arrowheads) within the plaque. The co-localization of neuroserpin to plaques was confirmed by staining consecutive sections of Alzheimer disease brain tissue for neuroserpin and A␤. Plaques were stained for neuroserpin (Fig. 1e, arrow and arrowhead), and the corresponding plaques were stained for A␤ in the adjacent section (Fig. 1f ). At higher power the presence of A␤ and neuroserpin within a single plaque was clearly seen (Fig. 1, e and f, insets). A␤  Abolishes the Protease Inhibitory Activity of Neuroserpin by the Formation of a 1:1 Molar Complex-To determine whether the localization of neuroserpin to ␤ amyloid plaques represents a specific interaction between the serpin and the A␤ peptide, we investigated the functional consequences of co-incubating pure solutions of protein and peptide. The proteinase inhibitory activity of neuroserpin, in common with the other serine proteinase inhibitors (serpins), requires the insertion of the reactive center loop (Fig. 2, a and b, yellow) into ␤-sheet A (Fig. 2, a and b, red) following the initial interaction with the proteinase. Reactive loop peptide insertion is energetically favorable and allows the subsequent deformation and inhibition of the proteinase (Fig. 2, a and b, blue and purple) (33). in the CA1 region of the hippocampus (asterisks) of brain tissue from patients with Alzheimer disease. Using the 6F/3D monoclonal antibody, A␤ (a, brown stain, main figure ϫ40 magnification, and inset ϫ200) was seen in amyloid plaques. Using a polyclonal antiserum against recombinant human neuroserpin (b, brown stain, main figure ϫ40, and inset ϫ200) widespread intra-neuronal staining was observed, particularly in dilated neurites (arrows), throughout the CA1 hippocampal region. At high power (c, ϫ300) we observed neuroserpin primarily at the periphery (arrowheads) of amyloid plaques (arrow). Double staining for both A␤ (d, dark brown) and neuroserpin (d, light brown, arrowhead) demonstrated co-localization of neuroserpin and A␤ within the amyloid plaques (two shown) of brain tissue from patients with Alzheimer disease. Consecutive sections from Alzheimer disease brain tissue were stained for neuroserpin (e, brown, ϫ50) or A␤ (f, brown, ϫ50). Two corresponding plaques are indicated (arrow and arrowhead) in both sections. The insets show a higher magnification of the arrow-labeled plaque (ϫ100).
Exogenous peptides such as a synthetic 12-mer peptide, corresponding to the reactive loop of the serpin (Fig. 2c, arrow) or the N terminus of the A␤ 1-42 peptide (Fig. 2d), may also anneal to ␤-sheet A, resulting in the loss of protease-inhibitory activity (34). Consequently we investigated the effect of A␤ peptides on the inhibitory activity of neuroserpin toward its cognate protease tPA. Normally 2 M neuroserpin (Fig. 3, lane 1) reacts with equimolar tPA (Fig. 3, lane 2) resulting in a high molecular mass SDS-resistant serpin-enzyme complex (Fig. 3, lane 3,  arrowhead). In contrast, preincubation of the neuroserpin with a 10ϫ molar excess of A␤ 1-42 at 37°C for 24 h before adding tPA abolished the serpin-enzyme complex (Fig. 3, lane 5). As a control we showed that tPA is not inhibited by A␤ 1-42 , because when tPA, neuroserpin, and a 10ϫ molar excess of A␤  were mixed simultaneously, the high molecular weight neuroserpin-tPA complex was still observed (Fig.  3, lane 4). These data show that A␤ 1-42 irreversibly inactivates neuroserpin. The migration profile of neuroserpin was also modified by preincubation with A␤ 1-42 , such that the tPA-cleaved neuroserpin species (Fig. 3, lanes 3 and 4, arrows) is no longer seen (Fig. 3, lane 5).
To determine the stoichiometry of the neuroserpin-A␤ 1-42 complex, 20 M neuroserpin was incubated with a 40ϫ molar excess of the A␤ 1-42 peptide for 30 h at 37°C. The samples were resolved by non-denaturing PAGE, and the neuroserpin band was analyzed by quantitative N-terminal amino acid sequencing. Two sequences were identified: 3.0 pmol of MRGSHH (consistent with the N-terminal end of the His-tagged neuroserpin protein) and 3.5 pmol of DAEFRH (the first six amino acids of the A␤ 1-42 peptide). Thus the neuroserpin-A␤ 1-42 complex has a 1:1 stoichiometry.
A␤  Prevents the Loop-sheet Polymerization of Neuroserpin-The patency of ␤-sheet A is also required for a competing biophysical process, termed loop-sheet polymerization. As with other serpins, neuroserpin will form homopolymers when incubated at elevated temperatures (19 -21) (Fig. 4, a and b). The insertion of an exogenous peptide into ␤-sheet A of serpins blocks the formation of loop-sheet polymers; however, to date, this has not been demonstrated for neuroserpin (19). Consequently we co-incubated a 50ϫ molar excess of A␤   (Fig. 4a,  arrowhead) or A␤   (Fig. 4b, arrowhead) with 0.8 mg/ml neuroserpin at 45°C and assessed the effect on the rate of neuroserpin polymerization using non-denaturing PAGE (Fig. 4, a  and b) with densitometry (Fig. 4, c and d). Both A␤ 1-40 and A␤ 1-42 significantly inhibited neuroserpin polymerization. At  all time points there was more monomeric neuroserpin (arrows) and less polymer in the presence of A␤ compared with neuroserpin alone (Fig. 4, a and b, ϩ lanes versus Ϫ lanes, and c  and d, squares versus triangles). Polymerization was also inhibited by a 10ϫ molar excess of A␤ 1-42 , but the degree of inhibition was less marked (data not shown). The specificity of the A␤ 1-42 -neuroserpin interaction was confirmed by coincubating A␤ 1-42 with another serpin, ␣ 1 -antitrypsin. In contrast the polymerization of ␣ 1 -antitrypsin was not inhibited by a 50ϫ molar excess of A␤ 1-42 at 50°C and 55°C (data not shown).
Neuroserpin Modulates A␤ Polymerization-The effect of neuroserpin on A␤ aggregation was assessed using the thioflavin T assay (37) that gives a fluorescence signal in the presence of cross-␤ structures and not with monomeric A␤ or amorphous aggregates. The co-incubation of neuroserpin with the A␤ 1-42 peptide at 37°C increased the fluorescence signal over a 30-min incubation period (Fig. 6a). The increased thioflavin T fluorescence was dependent on the concentration of neuroserpin over a range of 0 and 12 M. Control incubations with monomeric or polymeric neuroserpin alone did not give a fluorescence signal (data not shown), an expected result because serpin polymerization involves native-fold proteins and does not generate cross-␤ structure (38,39). Confirmation that neuroserpin accelerates the rate of aggregation of A␤ was seen on non-denaturing PAGE gels as the more rapid loss of A␤ (Fig. 5, A␤  band, lane 4 versus 5). Unexpectedly, electron microscopic inspection of the aggregates revealed that 12 M neuroserpin markedly suppressed the generation of A␤ fibrils over a range of incubation times and sample preparation protocols (1 mg/ml A␤, at 60 min for Fig. 6, b and c, applied directly to the grid; at 60 min for Fig. 6, d and e, centrifuged pellet applied to the grid). Thus neuroserpin dramatically alters the course of A␤ polymerization so that the dominant product, although thioflavin T-positive, appears amorphous and not fibrillar.
Neuroserpin Reduces the Cytotoxicity of A␤  in Cell Culture-The toxicity of A␤ is dependent on the size and conformation of its aggregated species. Mature amyloid is typically composed of unbranched, 10 nm diameter fibrils; however, oli- gomeric A␤ has a variety of conformations, including beadson-a-string, spheres, and rings (4 -6, 40 -42). Our finding that neuroserpin accelerates the aggregation of A␤ did not immediately predict either a neuroprotective or neurotoxic role for neuroserpin. Consequently, we used two cell culture models to determine the effect of neuroserpin on A␤ toxicity. Firstly, we treated cultures of PC12 rat pheochromocytoma cells with A␤ and used the MTT colorimetric assay to demonstrate impairment in metabolic activity. In the presence of 50 M A␤, PC12 cells showed a 40% reduction in the MTT signal; however, when we co-incubated the PC12 cells with 50 M A␤ and increasing concentrations of neuroserpin (4 -16 M) we observed a concentration-dependent reduction in cytotoxicity (Fig. 7a). At a concentration of 16 M neuroserpin the majority of the toxicity of A␤ was reversed ( p Ͻ 0.01). In contrast neuroserpin alone had a mild toxic effect. We then confirmed the protective role of neuroserpin by repeating the experiments using primary cultures of E14 rat embryonic cortical neurons (Fig.  7b). In this system 50 M A␤ resulted in a 60% reduction in the MTT signal and neuroserpin was able to rescue the cells in a concentration-dependent manner with 40% of the toxicity being reversed following treatment with 18 M neuro-serpin ( p Ͻ 0.01). Neuroserpin alone was not toxic to rat embryonic cortical neurons.
In Vivo Demonstration of an Interaction between Neuroserpin and A␤  in D. melanogaster-We confirmed that the A␤ 1-42 -neuroserpin interaction occurs in vivo using transgenic Drosophila melanogaster. Flies expressing the wild-type human neuroserpin gene under the control of the GAL4-UAS system were generated and characterized. Expression of wild-type neuroserpin under control of the strong GAL4 Act5c (strong ubiquitous expression) and GAL4 daughterless (moderate ubiquitous expression) drivers was lethal for developing embryos. Likewise, expression of wild-type neuroserpin in the retina using GAL4 GMR at 29°C resulted in a rough eye phenotype associated  Samples were diluted 1:1 with buffer and placed directly onto the electron microscopy grid (b and c) or centrifuged at 10,000 ϫ g for 10 min, and the pellet placed on the grid (d and e). The morphology of the resulting protein species was examined by transmission electron microscopy. A␤ 1-42 incubated in the absence of neuroserpin formed amyloid fibrils (b and d), whereas co-incubation A␤ 1-42 and neuroserpin abolished fibril formation and promoted the formation of smaller species (c and e). Scale bar, 100 nm. Neuroserpin alone did not form species that could be visualized under the electron microscope (data not shown).
with reduction in the number of ommatidia and malformation of the cornea. The most strongly affected eyes show a novel phenotype of deformed, necrotic intrusions that are distinct from more healthy tissue (Fig. 8a). Expression in a limited range of photoreceptors using the GAL4 sevenless driver at 25°C, however, permitted some flies to hatch. We hypothesized that the lethality of wild-type neuroserpin was due to toxic proteinase inhibition. To create a non-toxic form of neuroserpin we mutated the critical P1-P1Ј residues in the reactive center loop from Met-Arg to Pro-Pro (PP-neuroserpin). The PP-neuroserpin was no longer toxic to the flies, and experiments using recombinant PP-neuroserpin expressed in E. coli demonstrated that, despite being correctly folded, it was no longer an inhibitor of its cognate serine protease, tPA (data not shown). Having determined that the embryonic lethality of wild-type neuroserpin was due to its protease inhibitory activity we then performed experiments to determine the effect of co-expressing neuroserpin with A␤ 1-42 .
Flies expressing A␤ 1-42 as a secreted peptide were generated and characterized (32). Flies expressing A␤ 1-42 under the control of GAL4 sevenless did not exhibit embryonic lethality, and so we determined the effect of co-expressing A␤ 1-42 with neuroserpin. An interaction with neuroserpin would be expected to block the proteinase inhibitory activity of wild-type neuroserpin (Fig. 3) and consequently rescue the embryonic lethality. We established combinatorial crosses in which GAL4 sevenless was used to drive expression of A␤ 1-42 and wild-type neuroserpin. By counting the genotypes of newly eclosed flies we determined the fractional lethality of expressing either A␤ 1-42 or neuroserpin alone, assuming no lethality in flies expressing neither transgene. Assuming that each protein was acting independently to cause lethality we calculated the predicted eclosion (hatching) rates for flies, expressing both transgenes. We observed that co-expression of A␤ 1-42 with wild-type neuroserpin using the GAL4 sevenless driver resulted in a 5ϫ increase in the number of adult flies eclosing as compared with expected (Fig. 8b, bar NSϩA␤). Similarly, co-expression of the neuroserpin 12-mer reactive loop peptide, which inserts into the neuroserpin ␤-sheet A, had a similar rescuing effect (Fig. 8b, bar NSϩNS-RLP). In contrast, co-expression of neuroserpin with the reactive center peptide from antithrombin, a peptide that does not interact with neuroserpin in vitro (19), did not rescue the embryonic lethal phenotype (Fig. 8b, bar NSϩAT-RLP). Coexpression of another serpin, ␣ 1 -antitrypsin, which does not interact with A␤ 1-42 in vitro but which exhibits partial embryonic lethality under the GAL4 Act5C driver, did not demonstrate rescue of lethality when co-expressed with A␤ 1-42 (Fig. 8b, bar A.TrypϩA␤). These data demonstrate that the embryonic lethal phenotype of wild-type neuroserpin expression can be rescued by reducing the anti-proteinase activity of neuroserpin by co-expression of the A␤ 1-42 peptide.
Co-expression of A␤ 1-42 also reduced the prevalence of corneal micro-necrotic spots (Fig. 8c) that are seen when wild-type neuroserpin is expressed under control of GAL4 GMR at 25°C. Micro-necrotic spots were significantly ( p Ͻ 0.001) less frequent in flies co-expressing wild-type neuroserpin and the A␤ 1-42 peptide (1 spot per 62 eyes) as compared with flies expressing wild-type neuroserpin alone (22 spots per 82 eyes). Taken together these findings show that, despite the limitation of the toxic proteinase inhibitory phenotype of neuroserpin in the fly, there is clear evidence for an in vivo interaction between A␤ 1-42 and neuroserpin.

DISCUSSION
The aggregation of the A␤ peptide is widely regarded as the primary toxic insult in Alzheimer disease. A␤ aggregates are composed of peptides arranged in a cross-␤ structure (43,44) with the individual peptides stacked in a parallel orientation (45,46). The process of assembling an ordered aggregate is a highly co-operative process requiring a nucleation event but also permitting propagation of specific patterns of aggregation through time. The significance of propagating folding patterns is seen most clearly in the clinical consequences of the various prion strains (47). Despite being generated from a protein with identical primary structure the different strains cause different clinical consequences (48). The concept of aggregate strains is also relevant to Alzheimer disease, because propagation of differentially folded aggregates of A␤, with varying cytotoxic potencies, has also been observed (49). In vitro the favored pathways of ordered protein aggregation may be altered by changing simple physicochemical parameters such as solvent composition or pH (50), however, in vivo other factors such as interacting proteins are likely to be involved.
We show here that neuroserpin is present in close proximity to the A␤ peptide in plaques of patients with Alzheimer disease. We show in vitro that there is a specific molecular interaction between the A␤ peptide and neuroserpin (Fig. 2d). Support for this interaction comes from the finding of a 1:1 stoichiometry in the A␤-neuroserpin complex. The results show that the interaction fills the ␤-sheet A of neuroserpin, because, after co-incubation with A␤, neuroserpin is no longer able to inhibit tPA and to undergo loop-sheet polymerization, both of which are dependent on a patent ␤-sheet A (Fig. 2, a and b). The interaction with neuroserpin is likely to occur with the N-terminal and middle fragments of A␤, because A␤ 1-16 , A␤ [15][16][17][18][19][20][21] , and A␤ [22][23][24][25][26][27][28][29][30][31][32] and A␤ 40 peptides with proline substitutions in their mid and C-terminal portions were equally effective at preventing neuroserpin loop-sheet polymerization as the full-length A␤. Serpins are known to undergo promiscuous interactions with exogenous peptides, for example the reactive loop peptide from antithrombin will insert into both antithrombin and ␣ 1 -anti-trypsin. Indeed the antithrombin peptide inserts more rapidly into ␤-sheet A of ␣ 1 -antitrypsin than does the ␣ 1 -antitrypsin reactive loop peptide itself (51). Furthermore a five-residue peptide can inactivate the serpin plasminogen activator inhibitor-1 by the tandem insertion of two peptides into ␤-sheet A (52). These characteristics make it difficult to precisely define the sequence of A␤ that anneals to ␤-sheet A of neuroserpin. However, our data demonstrate that the binding residues are likely to be within the N-terminal or middle regions of the A␤ peptide.
Although A␤ blocks the loopsheet polymerization of neuroserpin, neuroserpin has the converse effect on A␤, causing an increase in thioflavin T fluorescence. This may result from accelerated A␤ aggregation in the presence of neuroserpin as suggested by the faster loss of monomeric A␤ on non-denaturing gels (Fig. 5, A␤ band, lane 4 versus 5). Alternatively the novel neuroserpin-induced A␤ aggregates could have a greater specific fluorescence than conventional aggregates, an idea supported by the higher fluorescence observed at 30 min in the presence of neuroserpin (Fig. 6a). Notably neuroserpin was able to abolish A␤ fibrillization, favoring instead the formation of small amorphous aggregates (Fig. 6b). Previous studies have shown that oligomeric forms of A␤ are toxic to neurons (53); therefore, we undertook two independent cell experiments to assess the cytotoxicity of A␤-neuroserpin aggregates. To our surprise we found a concentration-dependent cytoprotective effect of neuroserpin on both PC12 cells and primary embryonic rat neurons. This protective effect argues against neuroserpin accelerating the formation and stabilization of a toxic oligomer and argues in favor of neuroserpin changing the pathway of A␤ aggregation toward a less toxic end point.
The specific interaction between neuroserpin and the A␤ peptide was then shown to be significant in vivo using a D. melanogaster model system. The ubiquitous expression of wild-type neuroserpin in the fly during development was lethal. Expressing a mutant form of neuroserpin with a non-functional reactive loop abolished the toxicity, demonstrating that lethality was mediated by inappropriate protease inhibition. The formation of an inactive complex between ␤-sheet A of neuroserpin and A␤ explains the rescue of the serpin lethality. This was underscored by the demonstration that the lethality could only be rescued by peptides that insert into the ␤-sheet A of neuroserpin, specifically A␤ and the peptide representing the neuroserpin 12-mer reactive loop peptide. In contrast, the reactive loop 12-mer peptide from another serpin, which does not interact with neuroserpin (19), did not rescue the lethality.  (black patch, a). Co-expressing A␤ 1-42 rescued neuroserpin lethality, resulting in 5ϫ more flies hatching as compared with the number predicted (b, bar NSϩA␤, 370 flies eclosed; ***, p Ͻ 0.001 by Chi-square, representative of triplicate experiments using two independent fly lines for both A␤ 1-42 and neuroserpin). Similarly, the expression of another peptide that would be predicted to interact with neuroserpin (the neuroserpin reactive loop 12-mer) resulted in a 6.5ϫ increase in the number of flies hatching (b, bar NSϩNS-RLP, 140 flies eclosed; ***, p Ͻ 0.001 by Chi-square, representative of quadruplicate experiments using two independent fly lines for NS-RLP and one fly line for neuroserpin). In contrast the co-expression of neuroserpin with a non-reactive peptide (the antithrombin reactive loop 12-mer peptide) did not suppress neuroserpin toxicity (b, bar NSϩAT-RLP). The toxicity of a control serine protease inhibitor, ␣ 1 -antitrypsin, was not reduced when co-expressed with A␤ 1-42 (b, bar A.TrypϩA␤). When flies were cultured at 25°C the eye phenotype was less severe, consisting of micro-necrotic spots (c, bar ϭ 20 m). Micro-necrotic spots were significantly ( p Ͻ 0.001) less frequent in flies co-expressing wild-type neuroserpin and the A␤ 1-42 peptide (1 spot per 62 eyes) as compared with flies expressing wild-type neuroserpin alone (22 spots per 82 eyes).
Our data suggest that neuroserpin incorporates the N-terminal part of A␤ as the sixth strand within ␤-sheet A. Solid-state NMR studies of conventional A␤ aggregates are consistent with an unstructured N terminus (residues 1-10) that is followed by two areas of ␤-structure (residues 12-24 and 30 -40) separated by a turn (residues 25-29) (Fig. 9a) (45, 46, 54 -56). We hypothesize that forcing the N terminus of A␤ into a ␤ conformation will change the way in which the peptide folds (Fig. 9b) and result in the propagation of an off-pathway aggregate of reduced toxicity. In this hypothesis the binary complex of neuroserpin and A␤ seeds the aggregation of further A␤ into a non-canonical conformation with greater N-terminal ␤-structure. This conformation is predicted to be less stable, breaking frequently and seeding further non-canonical aggregates. This in turn prevents mature fibril formation and renders the A␤ peptide less toxic to neuronal cells in Alzheimer disease.