Neprilysin-sensitive Synapse-associated Amyloid-β Peptide Oligomers Impair Neuronal Plasticity and Cognitive Function*

A subtle but chronic alteration in metabolic balance between amyloid-β peptide (Aβ) anabolic and catabolic activities is thought to cause Aβ accumulation, leading to a decade-long pathological cascade of Alzheimer disease. However, it is still unclear whether a reduction of the catabolic activity of Aβ in the brain causes neuronal dysfunction in vivo. In the present study, to clarify a possible connection between a reduction in neprilysin activity and impairment of synaptic and cognitive functions, we cross-bred amyloid precursor protein (APP) transgenic mice (APP23) with neprilysin-deficient mice and biochemically and immunoelectron-microscopically analyzed Aβ accumulation in the brain. We also examined hippocampal synaptic plasticity using an in vivo recording technique and cognitive function using a battery of learning and memory behavior tests, including Y-maze, novel-object recognition, Morris water maze, and contextual fear conditioning tests at the age of 13–16 weeks. We present direct experimental evidence that reduced activity of neprilysin, the major Aβ-degrading enzyme, in the brain elevates oligomeric forms of Aβ at the synapses and leads to impaired hippocampal synaptic plasticity and cognitive function before the appearance of amyloid plaque load. Thus, reduced neprilysin activity appears to be a causative event that is at least partly responsible for the memory-associated symptoms of Alzheimer disease. This supports the idea that a strategy to reduce Aβ oligomers in the brain by up-regulating neprilysin activity would contribute to alleviation of these symptoms.

A subtle but chronic alteration in metabolic balance between amyloid-␤ peptide (A␤) anabolic and catabolic activities is thought to cause A␤ accumulation, leading to a decade-long pathological cascade of Alzheimer disease. However, it is still unclear whether a reduction of the catabolic activity of A␤ in the brain causes neuronal dysfunction in vivo. In the present study, to clarify a possible connection between a reduction in neprilysin activity and impairment of synaptic and cognitive functions, we cross-bred amyloid precursor protein (APP) transgenic mice (APP23) with neprilysin-deficient mice and biochemically and immunoelectron-microscopically analyzed A␤ accumulation in the brain. We also examined hippocampal synaptic plasticity using an in vivo recording technique and cognitive function using a battery of learning and memory behavior tests, including Y-maze, novel-object recognition, Morris water maze, and contextual fear conditioning tests at the age of 13-16 weeks. We present direct experimental evidence that reduced activity of neprilysin, the major A␤-degrading enzyme, in the brain elevates oligomeric forms of A␤ at the synapses and leads to impaired hippocampal synaptic plasticity and cognitive function before the appearance of amyloid plaque load. Thus, reduced neprilysin activity appears to be a causative event that is at least partly responsible for the memory-associated symptoms of Alzheimer disease. This supports the idea that a strategy to reduce A␤ oligomers in the brain by up-regulating neprilysin activity would contribute to alleviation of these symptoms.
Accumulation of amyloid-␤ peptide (A␤) 4 is a triggering event leading to a decade-long pathological cascade of Alzheimer disease (AD) (1,2). A subtle but chronic alteration in metabolic balance between A␤ anabolic and catabolic activities could result in A␤ accumulation and change monomeric A␤ to pathogenic forms (3,4). Neprilysin is an A␤-degrading enzyme first identified as a major in vivo peptidase capable of hydrolyzing synthetic multiple-radiolabeled A␤ injected into rat hippocampus (5). A genetic deficiency in neprilysin results in an elevation of A␤ levels in the mouse brain (6). Transgenic or viral expression of neprilysin in the brains of amyloid precursor protein (APP) transgenic mice consistently leads to marked attenuation of A␤ pathology (7)(8)(9). The exposure of APP transgenic mice to an enriched environment is reported to result in pronounced deceleration in cerebral A␤ levels and amyloid deposits with a concomitant elevation of brain neprilysin activity (10). Neprilysin is a presynaptic membrane-associated ectoenzyme with an extracellular active site, and it is involved in A␤ degradation at presynaptic sites (9,11,12). A recent study showed that somatostatin causes selective reduction of A␤ 42 by promoting the surface appearance of neprilysin on the presynaptic membrane (13). Down-regulation of neprilysin in the hippocampus and cerebral cortex with aging (14,15) and at an early stage of AD development (15)(16)(17) suggests a close association of neprilysin with AD etiology and pathogenesis. This led us to hypothesize that a reduction in neprilysin activity may elicit a local elevation of A␤ concentration in the extracellular space close to synapses, possibly affecting the critical pathology during the course of AD development.
Dysfunction of synaptic integrity and plasticity is indeed a typical and early function-related event in the AD cascade (1, 18 -20). Levels of soluble A␤ rather than insoluble A␤ (A␤ plaque load) correlate well with the degree of synapse loss in AD brains (21,22). The same has been shown in several lines of APP transgenic mice (23,24). Increasing evidence shows that particular soluble forms of A␤, such as oligomers, protofibrils, A␤-derived diffusible ligands, and amylospheroids are more neurotoxic than soluble monomeric or insoluble fibrillar forms (25)(26)(27)(28). These results are supported by recent findings that the intracerebroventicular infusion of cultured cell-derived A␤ oligomers causes acute suppression of long-term potentiation (LTP) in the hippocampus in vivo as well as cognitive abnormality in the rat (29 -31). However, it is necessary to consider whether A␤ oligomers placed exogenously in ventricles diffuse sufficiently into hippocampal parenchyma and at which neuronal sites they act to influence synaptic and neuronal activities.
In the present study, to clarify a possible connection between a reduction in neprilysin activity and impairment of synaptic and cognitive functions, we cross-bred APP transgenic mice (APP23) (32) with neprilysin-deficient mice (33), analyzed A␤ accumulation biochemically and histologically in the brains of the cross-bred mice at the age of 13-16 weeks, and related the local elevation of A␤ oligomers with dysfunction of hippocampal synaptic plasticity by using an in vivo recording technique and with abnormalities in cognitive function by using a battery of learning and memory behavior tests.

EXPERIMENTAL PROCEDURES
Animals-All animal experiments were performed in compliance with the institutional guidelines for Animal Experiments of RIKEN and Nagoya University Graduate School of Medicine. The procedures involving animals and their care conformed to the international guidelines set out in Principles of Laboratory Animal Care (National Institutes of Health publication no. 85-23, revised 1985). APP transgenic mice with a heterozygous neprilysin-deficient background (NEP ϩ/Ϫ APP ϩ ) were produced by cross-breeding APP23 mice, which overexpress human-type APP carrying double mutations (K670N/ M671L) (32), and homozygous neprilysin-deficient mice (NEP Ϫ/Ϫ ) (33). NEP ϩ/ϩ APP ϩ and NEP Ϫ/Ϫ APP ϩ littermates were produced by breeding NEP ϩ/Ϫ APP Ϫ and NEP ϩ/Ϫ APP ϩ mice. All mice were on the same genetic background (C57BL/6J). In all experiments mice were analyzed at the age of 13-16 weeks. They were housed in plastic cages, received food (CE2, Clea Japan Inc., Tokyo, Japan) and water ad libitum, and were maintained on a 12/12-h light-dark cycle (lights on at 09:00, off at 21:00).
Western Blotting-The cerebral hemispheres were homogenized in 3 volumes (w/v) of 50 mM Tris-HCl buffer (pH 7.6) containing 150 mM NaCl, protease inhibitor mixture (Complete TM , Roche Diagnostics) supplemented with 0.7 g/ml pepstatin A (Peptide Institute, Osaka, Japan). The homogenates were centrifuged at 200,000 ϫ g and 4°C for 20 min using an Optima TL ultracentrifuge and a TLA100.4 rotor (Beckman, Palo Alto, CA). The clear supernatants were used for analyzing levels of secreted forms of APP (sAPP). The pellet (membrane fraction) was suspended with the above buffer and used for analyzing levels of murine and human full-length APPs. Protein concentrations were determined using a BCA protein assay kit (Pierce). We separated equivalent amounts of protein of the membrane or supernatant fraction in each lane by 5-20% gradient SDS-polyacrylamide gel electrophoresis and transferred them electrophoretically to polyvinylidene fluoride membranes (Hybond TM -P, Amersham Biosciences). To detect fulllength APP, sAPP (N-terminal fragments) and C-terminal fragments of APP (CTF-␤, -␣, and -␥/⑀), we probed the blots with 22C11 (0.2 g/ml; Chemicon, Temecula, CA), 6E10 (1.0 g/ml; Signet Laboratories, Inc., Dedham, MA), a rabbit polyclonal antibody against synthetic peptide CISEVNL, the 5 C-terminal amino acids of APP Swedish (␤NL; 1.0 g/ml), and a rabbit polyclonal antibody against the C-terminal of APP (A8717, Sigma) followed by horseradish peroxidase-conjugated antimouse or anti-rabbit IgG (Amersham Biosciences). An immunoreactive band on the membrane was visualized with an ECLplus kit (Amersham Biosciences), and the band intensity was determined with a densitometer, LAS3000 (Fuji Photo Film, Tokyo, Japan) using Science Lab 97 Image Gauge software (Version 3.0.1; Fuji Photo Film). Immunoreactive protein content in each sample was calculated based on a standard curve constructed with one of the samples. Each set of experiments was repeated at least three times to confirm the results. The protein levels of ␣-tubulin and ␤-actin as housekeeping genes or commonly expressed proteins were also measured by quantitative Western blotting using anti-␣-tubulin (DM 1A, Sigma) and anti-␤-actin (AC-15, Sigma) antibodies, respectively, to confirm that equal amounts of total protein were extracted.
To analyze monomeric or oligomeric status of A␤ by Western blot, homogenate from cerebral hemispheres was delipidized with chloroform:methanol (2:1) and chloroform:methanol:distilled water (1:2:0.8) sequentially to improve sensitivity, as described by Morishima-Kawashima et al. (34). The resultant protein fraction was evaporated and then solubilized with a sample buffer containing 4 M urea. Samples (20 -40 g) were separated by 5-20% gradient SDS-polyacrylamide gel electrophoresis and transferred electrophoretically to nitrocellulose membranes. The membranes were boiled in phosphate-buffered saline for 3 min to gain high sensitivity. The blot was probed with 6E10 followed by the ABC method (Vector Laboratories, Inc., Burlingame, CA). An immunoreactive band on the membrane was visualized with a Supersignal West Pico chemiluminescence kit (Pierce) according to the manufacturer's instructions, and the band intensity was determined as described above. Immunoreactive neprilysin content in each sample was calculated based on a standard curve constructed with synthesized A␤ 1-42 (Bachem AG, Bubendorf, Switzerland) that was linear in the range of 0.15-2.4 ng.
A␤ Quantitation-The cerebral hemispheres were homogenized in 3 volumes (w/v) of 50 mM Tris-HCl buffer (pH 7.6) containing 150 mM NaCl and the protease inhibitor mixture (TBS) with a Teflon-glass homogenizer and centrifuged at 200,000 ϫ g for 20 min at 4°C. The supernatant was defined as the soluble (TBS-extractable) fraction, and guanidine/HCl was added to give a final concentration of 0.5 M before ELISA. The pellet was solubilized by sonication in 6 M guanidine/HCl buffer containing the protease inhibitor mixture. The solubilized pellet was centrifuged at 200,000 ϫ g for 20 min at 4°C, after which the supernatant was diluted 12 times to reduce the concentration of guanidine-HCl and used as the insoluble fraction. The amounts of A␤ X-40 and A␤ X-42 in each fraction were determined by sandwich ELISA (BIOSOURCE, Camarillo, CA), respectively, according to the manufacturer's instructions.
Immunoelectron Microscopic (IEM) Study-IEM was carried out as previously described (35). The mice (NEP ϩ/ϩ APP ϩ , n ϭ 3; NEP Ϫ/Ϫ APP ϩ , n ϭ 3) were perfused with a fixative containing 4% paraformaldehyde and 0.2% glutaraldehyde after being deeply anesthetized with diethyl ether inhalation, then the brain tissue samples were immersed in a fixative containing 4% paraformaldehyde and 1% glutaraldehyde for 3 days, osmified, and embedded in epoxy resin. First, we cut semi-thin (0.5 m) sections of several samples from each subject to select appropriate areas, such as the CA1 region of the hippocampus and the outer molecular layer of the dentate gyrus (OML), for IEM study. Next, we cut serial ultrathin (100-nm thick) sections and placed the sections on nickel grids for IEM. Ultrathin sections on the nickel grid were etched with 1% H 2 O 2 for 1 min then 1% periodic acid solution for 1 min. After blocking with 5% normal goat serum, the sections were incubated overnight at room temperature with affinity-purified, oligomer-specific polyclonal antibody (4 g/ml; A11, BIOSOURCE), which has been well characterized and specifically recognizes the oligomeric state of A␤ (36). The sections were then incubated with the 5-nm gold-tagged goat anti-rabbit IgG (Fc) (1:20, Amersham Biosciences; RPN 420) for 2 h at room temperature. The sections were then fixed with 1% glutaraldehyde in 0.2 M phosphate buffer for 10 min. The EM sections on grids were stained briefly with conventional uranyl acetate and lead citrate. For the control study of IEM, the ultrathin sections were incubated with the normal rabbit IgG. There was no staining of the sections incubated with the normal rabbit IgG.
IEM sections were observed under an electron microscope (100CXII, JEOL Ltd., Tokyo, Japan). We randomly took photographs of each region of each subject at the same magnification (NEP ϩ/ϩ APP ϩ CA1, 63 photos; NEP Ϫ/Ϫ APP ϩ CA1, 62 photos; NEP ϩ/ϩ APP ϩ OML, 82 photos; NEP Ϫ/Ϫ APP ϩ OML, 83 photos). The magnification of photo prints (15.8 ϫ 11.4 cm) was ϫ30,000, and the actual size of each print was 19.8 m 2 . We identified the localization of oligomer immunoreactions (oligomer-IRs) in the neuropils, such as axons, axonal terminals, dendrites, glial processes and unidentified processes and counted the number of oligomer-IRs per photo print (field). We counted a site having 2 or more 5-nm gold particles as one oligomer-IR.
Electrophysiological Procedures-Mice were deeply anesthetized by intraperitoneal administration of 50 mg/kg sodium pentobarbital and placed on a stereotaxic device. Core body temperature was monitored and maintained at 35 Ϯ 0.5°C via a heating pad with a thermoprobe placed in the rectum of the mouse. The skull was exposed, and small holes were drilled at the sites where recording and stimulating electrodes were to be positioned. To maintain a constant depth of anesthetization, mice were given sodium pentobarbital (8 mg/kg) intramuscularly every 30 min beginning 40 min after the first injection. Glass recording electrodes (tip resistance of 1-3 megohms) were filled with saline. The electrode for the extracellular field potential recording was inserted into the stratum radiatum of the CA1 region of hippocampus (2.2 mm posterior to bregma, 1.25 mm lateral to midline, and 1.0 -1.5 mm ventral to the brain surface) or in the granular cell layer of the dentate gyrus (2.1 mm posterior to bregma, 1.5 mm lateral to midline, and a 1.8 -2.0-mm depth to the surface). For stimulation in the CA1 region, a platinum coaxial bipolar stimulating electrode with a tip diameter of 25 m was positioned in the stratum radiatum to stimulate the Schaffer collateral/commissural pathway (1.5 mm posterior to bregma, 2.0 mm lateral to midline, and 1.0 -1.5 mm ventral to the brain surface). For stimulation in the dentate gyrus region, the same electrode was positioned in the angular bundle (3.0 mm lateral to lambda and a 1.25-1.5-mm depth to the surface). A reference electrode was attached to the skin of the animal's head. For recordings in the CA1, stimulus intensity was adjusted to produce an average field excitatory postsynaptic potential (EPSP) amplitude of 1 mV, which was not larger than 1/3 of the maximum amplitude. For recordings in the dentate gyrus, we used a stimulation strength that evokes a population spike of 2-3 mV on the field EPSP. After 15-30 min of base-line stimulation at a frequency of 0.033 Hz, tetanic stimulation was delivered to each stimulation site. For LTP induction in the CA1 region, one set of four tetanal trains (50 pulses, 100 Hz) with a 10-s intertrain interval was used, and for the dentate gyrus region 1 tetanal train (100 pulses, 100 Hz) was used. Field EPSPs were monitored and analyzed using a computer-based data acquisition system. The magnitude of potentiation was expressed as the percentage of base line of the maximal slope of EPSPs in the CA1 and of the population spike amplitudes in the dentate gyrus, respectively. Before recording LTP, we generated input/output curves by gradually increasing the intensity of the stimulation in both the CA1 and the dentate gyrus. We also recorded paired pulse facilitation evoked with an interpulse interval of 30, 50, and 100 ms. The intensity of stimulation for paired pulse facilitation was adjusted to evoke ϳ1 mV of first field EPSP amplitude.
Behavioral Analysis-A battery of behavioral tests representing hippocampus-dependent learning and memory functions was carried out sequentially for the mice at the age of 13-16 weeks according to the experimental schedule (see supplemental Fig. 1), which was designed to minimize stress for the mice.
First, a Y-maze test was carried out using animals 13 weeks old. The maze was made of black painted wood; each arm was 40 cm long, 12 cm high, 3 cm wide at the bottom, and 10 cm wide at the top. The arms converged at an equilateral triangular central area that was 4 cm at its longest axis. Each mouse was placed at the center of apparatus and allowed to move freely through the maze during an 8-min session. The frequency of arm entries was recorded visually. Alternation was defined as successive entry into the three arms on overlapping triplet sets. The alternation behavior (%) was calculated as the ratio of actual alternations to possible alternations (defined as the number of arm entries minus two) multiplied by 100.
The novel-object recognition test (37) was started the day after the Y-maze test. Mice were individually habituated to an open-field box (30 ϫ 30 ϫ 35 (high) cm) for 3 days. During the training session two novel objects were placed into the open field, and the animals were allowed to explore for 10 min. The time spent exploring each object and total approach time were recorded. During retention sessions the animals were placed back into the same box 24 h after a training session, in which one of the familiar objects used during training was replaced by a novel object and allowed to explore freely for 10 min. A preference index, a ratio of the amount of time spent exploring any one of the two objects (training session) or the novel object (retention session) over the total time spent exploring both objects was used to measure recognition memory.
The Morris water maze test was conducted in a circular pool of a 1.2-m diameter and filled with water at a temperature of 22.0 Ϯ 1°C as described previously (38,39) with minor modifications. In a hidden platform test, a platform (7 cm in diameter) was submerged 1 cm below water level. Swimming paths were tracked with a camera fixed on the ceiling of the room and stored in a computer (Etho Vision system, Neuroscience⅐idea Co. Ltd., Osaka, Japan). The mice were given 2 trials (1 block) per day for 7 consecutive days during which the platform was left in the same position. The time taken to locate the escape platform (escape latency, s), swimming distance (cm), and swimming velocity were determined in each trial. Twenty-four hours after the last training trial the mice were given a transfer test without the platform, and they were given 60 s to search for the pool. In the visible-platform test, the black platform (7 cm in diameter) was located 1 cm below water level.
Contextual fear conditioning test (37,39) was carried out the day after the Morris water maze test. For measuring the basal level of freezing response (preconditioning phase) mice were individually placed in the conditioning cage (25 ϫ 31 ϫ 11 (high) cm) for 2 min. For training (conditioning phase) mice were placed in the conditioning cage, and then a 15-s foot shock of 0.6 mA was delivered through a shock generator (Neuroscience⅐idea Co. Ltd.). This procedure was repeated 4 times with 25-s intervals. One day after fear conditioning, mice were placed in the conditioning cage, and the freezing response was measured for 2 min.
Statistical Analysis-All data were expressed as means Ϯ S.E. For comparisons of the means between 2 groups, statistical analysis was performed by applying Student's t test after confirming equality between the variances of the groups. If the variances were unequal, Mann-Whitney U tests were performed. Comparisons of the means among more than three groups were done by analysis of variance (ANOVA) or repeated-measure ANOVA followed by Student-Newman-Keuls multiple range test. p values of less than 0.05 were considered to be significant.

Neprilysin Deficiency Affects Levels of A␤ but Not APP or Other
Fragments-In APP transgenic mice cross-bred with neprilysin-deficient mice (NEP Ϫ/Ϫ APP ϩ mice), although the quantities of APP or other APP fragments remained unaltered, there was no difference from control in the steady-state levels of full-length APP, sAPP␣ (the secreted form of APP generated by ␣-secretase), sAPP␤ (the secreted form of sAPP generated by ␤-secretase), the CTF-␤, CTF-␣, or CTF-␥/⑀ (Fig. 1), and the amount of A␤ was markedly and significantly elevated (Fig. 2, A  and B). Thus, cross-breeding APP transgenic mice with neprilysin-deficient mice enabled us to selectively elevate transgene-derived A␤ and to exaggerate the action of neprilysin-sensitive A␤. In addition to APP transgenic mice with neprilysin, a set of wild-type and neprilysin-deficient mice without the APP transgene serves as a relevant triple control on evaluating neuronal functions.
Interestingly, the predominant (more than 90% of total A␤) form of A␤ observed in the transgenic mouse brains was a dimer, the level of which was elevated nearly 3 times by neprilysin deficiency. We confirmed that the procedures used to extract A␤ from brain homogenates do not convert A␤ monomers to dimers (see supplemental Fig. 2). Furthermore, A␤ trimers and tetramers were also elevated by neprilysin deficiency although to a lesser degree than dimers (Fig. 2, A and B).
These results indicate that neprilysin regulates the quantities of A␤ monomers as well as oligomers in vivo, consistent with an earlier report demonstrating that recombinant human neprilysin can degrade synthetic A␤ oligomers including dimers, trimers, and tetramers in vitro (40). In addition, detection of the A␤ dimer as a major early A␤ form in APP transgenic mice is in agreement with a recent report by Kawarabayashi et al. (41) employing a distinct A␤ detection procedure (i.e. immunoprecipitation followed by Western blotting). Because neprilysin is also detected in the raft fraction (41), the A␤ dimer detected here might be derived from the raft fraction. We also used sandwich ELISA to quantify A␤40 and A␤42 and obtained consistent data indicating that A␤42 was more apt to be oligomerized than A␤40 (Fig. 2, C-F). Because the ELISA detects A␤ monomers, but not oligomers, as previously described (42).
A Local Elevation of A␤ Oligomers at the Synapses-We analyzed the density and distribution of A␤ oligomers by IEM analysis with an antibody, which recognizes A␤ oligomers with high molecular weights (36) in the CA1 stratum radiatum of the hippocampus (CA1) and the OML of NEP Ϫ/Ϫ APP ϩ and NEP ϩ/ϩ APP ϩ mice at the age of 13-16 weeks. Neprilysin is more abundant in the OML than in the CA1 (11), which may give rise to a difference in the levels of A␤ oligomers. Almost all of the immunoreactive oligomers were localized at cell processes, such as  (43). B-F, intensities of immunoreactive bands on blots shown in A were quantified as described under "Experimental Procedures." The levels of sAPP (D) and sAPP␤ (E) are expressed as relative amounts to that of ␣-tubulin. The intensity was normalized against the data from NEP ϩ/ϩ APP ϩ mice. Each column represents the mean Ϯ S.E. of eight mice. The expression and processing levels of APP were not changed between NEP ϩ/ϩ and NEP Ϫ/Ϫ mice regardless of the APP transgene. n.d., not detectable. axon terminals, dendrites, glial processes, and small unidentified processes and on axons (Fig. 3A), although the immunoreactions were not strong enough to be easily recognized at the light microscopic level. A␤ oligomers were observed as a cluster consisting of two or more gold particles, and we considered this cluster as one oligomer-IR. The total number of oligomer-IRs per field (density) was significantly greater in NEP Ϫ/Ϫ APP ϩ compared with that in NEP ϩ/ϩ APP ϩ mice, and this effect was larger in OML (a 4.1-fold increase) than in CA1 (a 2.2-fold increase) (Fig. 3B). The distribution pattern of oligomer-IRs localized in cell pro-cesses was significantly different between the two genotypes (Fig. 3C); the proportion of oligomer-IRs that appeared on the axon terminals and glial processes both in OML and in CA1 were more prominently increased in NEP Ϫ/Ϫ APP ϩ than in NEP ϩ/ϩ APP ϩ mice. Thus, the increase of the oligomers was reversely correlated with innate amounts of neprilysin (11). In the brains of both mouse genotypes at the age tested here we found no amyloid plaques and no structural abnormalities in cell processes containing oligomer-IRs.
Impairment of Synaptic Plasticity-We then performed in vivo electrophysiological analyses in the molecular layer of the dentate gyrus (DG) and the CA1 stratum radiatum of hippocampus (CA1) to examine possible alterations of neuronal functions in the mice. We first recorded field EPSPs evoked by stimulation of the perforant path and the Schaffer collaterals across a range of stimulus intensities in the mice (see supplemental Fig. 3). We found no significant difference between genotypic groups in the size and shape of the field EPSPs or input-output curves in each region.
Next we analyzed LTP using tetanus stimulation protocols in vivo to examine synaptic plasticity in DG (Fig. 4) and CA1 (Fig. 5). High frequency stimulation in DG caused stable potentiation of population spike amplitudes in the mice without the APP transgene. The initial rise of population spike amplitudes was potentiated to nearly 400% and was significantly larger than that in mice with the APP transgene. LTP recorded at the DG of NEP Ϫ/Ϫ APP ϩ mice was significantly suppressed by 35% and practically lost until 120 min after the tetanus stimulation compared with the other groups (Fig. 4B). LTP at the DG of NEP ϩ/ϩ APP ϩ mice was also significantly inhibited (by 72% compared with the mice without the APP transgene; Fig. 4C). No significant alteration in LTP between the mice without APP transgene (between NEP ϩ/ϩ APP Ϫ and NEP Ϫ/Ϫ APP Ϫ mice) was observed. In CA1, although all mice showed an initial response rise immediately after tetanus of up to 2-2.5-fold of the base line (Fig. 5B), LTP observed in NEP ϩ/ϩ APP ϩ and NEP Ϫ/Ϫ APP ϩ mice but not the other groups gradually decreased until 60 min after tetanus stimulation (Fig. 5C). Other differences in LTP among each genotypic group were essentially similar to that obtained from DG. However, the suppression of LTP in NEP Ϫ/Ϫ APP ϩ mice was more prominent in DG than in CA1, reflecting a greater increase in the quantity of A␤ oligomers.
In addition we delivered several pairs of pulses to examine short-term potentiation. Under the stimulations of paired pulses with different interpulse intervals, all groups of mice showed an enhancement of the second response relative to the first one, but there was no significant difference in the responses from either the dentate granular cells or the CA1 pyramidal neurons in all genotypic groups (see supplemental Fig.  4). Thus, although basal synaptic transmission and short-term potentiation were not affected by APP overexpression, neprilysin-deficiency, or both, LTP was prominently suppressed.
Impairment of Cognitive Function-We examined using several learning and memory paradigms to determine whether neprilysin deficiency in APP transgenic mice affects cognitive functions. In a Y-maze test representing spatial working memory classified as short term and hippocampus-dependent memory, only the NEP Ϫ/Ϫ APP ϩ mice showed a significant impairment of spontaneous alternation behavior compared with NEP ϩ/ϩ APP Ϫ mice (Fig. 6A). However, two-way ANOVA showed significant main effects of APP transgene (p Ͻ 0.05) and neprilysin deficiency (p Ͻ 0.05). Locomotion abilities and exploratory activities in the maze were normal in all mice (number of total arm entries did not differ among genotypes; data not shown).
Next we tested visual recognition memory classified as long term and hippocampus-dependent memory using a novel-object recognition FIGURE 2. Effect of neprilysin deficiency on A␤ levels in APP transgenic mice. A, A␤ extracted from mouse brains (13-16 weeks old) was analyzed by Western blotting using 6E10 antibody. We confirmed that the extraction procedures do not convert synthetic monomer A␤ into oligomers in the presence or absence of the brain homogenates. A␤ monomers, dimers, trimers, and tetramers are indicated by the arrows. The asterisk shows nonspecific band. B, data shown in A were quantified by densitometry. Synthetic A␤ 1-42 (0.15-2.4 ng) run on the same gel as indicated in A was used for calibration within a linear range. *, p Ͻ 0.01 (NEP ϩ/ϩ APP ϩ versus NEP Ϫ/Ϫ APP ϩ ). A␤ levels of more than trimer (trimer and tetramer) are significantly different between NEP ϩ/ϩ APP ϩ and NEP Ϫ/Ϫ APP ϩ (*, p Ͻ 0.05). A␤x-40 (C and E) and A␤x-42 (D and F) levels in both soluble (C and D) and insoluble (E and F) fractions of the mouse brains were determined by sandwich ELISA at the ages of 13-16 weeks. Each column with a bar represents the mean Ϯ S.E. of eight mice. *, p Ͻ 0.05, significantly different from NEP ϩ/ϩ APP ϩ (Student-Newman-Keuls multiple range test).
task. During the training session there was no significant difference in exploratory preference for two objects (Fig. 6B, left) and total exploratory time (data not shown), suggesting that all mice had similar levels of motivation, curiosity, and interest in exploring novel objects. As a retention session, we observed a significant shortened exploratory preference time for the novel object of NEP Ϫ/Ϫ APP ϩ mice than that of any other FIGURE 3. Density and distribution of oligomer-IRs localized in the cell processes of the hippocampal formation. A, a typical distribution pattern of oligomer-IRs detected at axons (AX), axonal terminals (AT), dendrites (Den), glial process (Glia) in the outer molecular layer of the dentate gyrus of NEP Ϫ/Ϫ APP ϩ mice. B, densities of oligomer-IRs at the processes either in the CA1 of the hippocampus or the outer molecular layer of the dentate gyrus were significantly higher in NEP Ϫ/Ϫ APP ϩ mice than those in NEP ϩ/ϩ APP ϩ mice (*, p Ͻ 0.01, Mann-Whitney U test). C, distribution patterns in both CA1 of the hippocampus and the molecular layer of dentate gyrus were significantly different between NEP ϩ/ϩ APP ϩ and NEP Ϫ/Ϫ APP ϩ mice (p Ͻ 0.01, 2 for independence test).
group and for that of NEP ϩ/ϩ APP ϩ mice than NEP ϩ/ϩ APP Ϫ mice (Fig.  6B, right). However, there was no difference in the total exploratory time in all genotypes (data not shown). Two-way ANOVA showed significant main effects of APP transgene (p Ͻ 0.001) and neprilysin deficiency (p Ͻ 0.05) on exploratory preference at a retention session.
Third, we examined the mice for the ability of reference memory classified as long term and hippocampus-dependent memory by the Morris water maze test. The NEP ϩ/ϩ APP Ϫ , NEP Ϫ/Ϫ APP Ϫ , and NEP ϩ/ϩ APP ϩ mice managed to learn the hidden-platform task, but only the NEP Ϫ/Ϫ APP ϩ mice needed a significantly longer time (Fig. 6C) and swimming distance (data not shown) to reach the platform than other mouse groups. In the transfer test after 7 days, the NEP ϩ/ϩ APP Ϫ , NEP Ϫ/Ϫ APP Ϫ , and NEP ϩ/ϩ APP ϩ mice again searched preferentially in the trained quadrant, but NEP Ϫ/Ϫ APP ϩ mice did not (Fig. 6D). This impaired ability of the NEP Ϫ/Ϫ APP ϩ mice reflected neither a dysfunction of locomotion nor poor swimming ability; the mice in all groups showed a similar speed of movement both in the hidden-platform task and in the visible-platform task (data not shown). In addition, the escape  (13-16 weeks old). A, the response traces of field EPSPs from NEP ϩ/ϩ APP Ϫ , NEP Ϫ/Ϫ APP Ϫ , NEP ϩ/ϩ APP ϩ , and NEP Ϫ/Ϫ APP ϩ mice (13-16 weeks old) before (black lines) and 120 min after (gray lines) tetanic stimulation were superimposed. After 15-30 min of base-line stimulation at a frequency of 0.033 Hz, tetanic stimulation was delivered to the perforant path. One set of one tetanal trains (100 pulses, 100 Hz) with a 10-s intertrain interval was used to induce LTP. B, the averaged time course of population spike amplitude values taken from NEP ϩ/ϩ APP Ϫ (n ϭ 15), NEP Ϫ/Ϫ APP Ϫ (n ϭ 8), NEP ϩ/ϩ APP ϩ (n ϭ 13), and NEP Ϫ/Ϫ APP ϩ (n ϭ 18) mice was illustrated. Each point indicates the mean Ϯ S.E. Repeated-measure two-way ANOVA of the magnitude of LTP in the last 10 min showed a significant main effect of genotype (F (3,1000) ϭ 15.632; p Ͻ 0.0001). Post hoc analysis revealed that the potentiation of population spike amplitude in NEP Ϫ/Ϫ APP ϩ or NEP ϩ/ϩ APP ϩ mice was significantly different from the other groups (*, p Ͻ 0.05; NEP Ϫ/Ϫ APP ϩ versus the other groups. #, p Ͻ 0.05; NEP ϩ/ϩ APP ϩ versus the other groups), but a significant difference between NEP ϩ/ϩ APP Ϫ and NEP Ϫ/Ϫ APP Ϫ was not observed. C, population spike amplitude values taken 120 min after tetanic stimulation. Each column represents the mean Ϯ S.E. *, p Ͻ 0.05; NEP Ϫ/Ϫ APP ϩ versus the other groups; #, p Ͻ 0.05; NEP ϩ/ϩ APP ϩ versus the other groups (Student-Newman-Keuls multiple range test). APP transgenic mice (13-16 weeks old). A, The response traces of field EPSPs from NEP ϩ/ϩ APP Ϫ , NEP Ϫ/Ϫ APP Ϫ , NEP ϩ/ϩ APP ϩ , and NEP Ϫ/Ϫ APP ϩ mice (13-16 weeks old) before (black lines) and 60 min after (gray lines) tetanic stimulation were superimposed. After 15-30 min of base-line stimulation at a frequency of 0.033 Hz, tetanic stimulation was delivered to the Schaffer collateral/commissural pathway. One set of four tetanal trains (50 pulses, 100 Hz) with a 10-s intertrain interval was used to induce LTP. B, the averaged time course of field potential slope values taken from NEP ϩ/ϩ APP Ϫ (n ϭ 10), NEP Ϫ/Ϫ APP Ϫ (n ϭ 9), NEP ϩ/ϩ APP ϩ (n ϭ 10), and NEP Ϫ/Ϫ APP ϩ (n ϭ 15) mice was illustrated. Each point indicates the mean Ϯ S.E. Repeated-measure two-way ANOVA of the magnitude of LTP in the last 10 min showed a significant main effect of genotype (F (3, 800) ϭ 11.018; p Ͻ 0.0001). Post hoc analysis revealed that the potentiation of field EPSP slopes in NEP Ϫ/Ϫ APP ϩ or NEP ϩ/ϩ APP ϩ mice was significantly different from the other groups (*, p Ͻ 0.05; NEP Ϫ/Ϫ APP ϩ versus the other groups; #, p Ͻ 0.05; NEP ϩ/ϩ APP ϩ versus the other groups), but a significant difference between NEP ϩ/ϩ APP Ϫ and NEP Ϫ/Ϫ APP Ϫ was not observed. C, field potential slope values taken 60 min after tetanic stimulation. Each column represents the mean Ϯ S.E. *, p Ͻ 0.05, NEP Ϫ/Ϫ APP ϩ versus the other groups; *, p Ͻ 0.05, NEP ϩ/ϩ APP ϩ versus the other groups by Student-Newman-Keuls multiple range test.

FIGURE 5. Effect of neprilysin deficiency on LTP in CA1/Schaffer collaterals of
latency or swimming distance of NEP Ϫ/Ϫ APP ϩ mice in the visible-platform task was similar to that of other genotypic mice, indicating that visual functions were normal (data not shown).
Last, the mice were tested for associative memory in a hippocampusdependent contextual fear conditioning task. All mice hardly showed a freezing response in the preconditioning phase (before shock), so there was no difference in basal levels of freezing response (Fig. 6E, left). Then, after electric foot shock training, all the mice learned context-associated fear. NEP ϩ/ϩ APP Ϫ , NEP ϩ/ϩ APP Ϫ , and NEP ϩ/ϩ APP ϩ mice maintained a marked contextual freezing response 24 h after fear conditioning, and FIGURE 6. Performance in learning and memory tasks. A battery of behavioral tests representing cognitive functions was carried out sequentially for the mice at the age of 13-16 weeks. Each column indicates the mean Ϯ S.E. (NEP ϩ/ϩ APP Ϫ , n ϭ 13; NEP Ϫ/Ϫ APP Ϫ , n ϭ 14; NEP ϩ/ϩ APP ϩ , n ϭ 11; NEP Ϫ/Ϫ APP ϩ , n ϭ 11). A, Y-maze task. The alternation behavior (%) was calculated as the ratio of actual alternations to possible alternations. Two-way ANOVA showed significant main effects of APP transgene (F (1,45) ϭ 4.19; p Ͻ 0.05) and neprilysin deficiency (F (1,45) ϭ 5.01; p Ͻ 0.05) with no interaction between APP transgene and neprilysin deficiency. Post hoc analysis revealed that the NEP Ϫ/Ϫ APP ϩ mice were significantly impaired successive entry into the three arms compared with NEP ϩ/ϩ APP Ϫ mice (*, p Ͻ 0.05). B, novel object recognition task. A retention session was carried out 24 h after the training, and exploratory preference during a 10-min session was measured. All genotypic mice except NEP Ϫ/Ϫ APP ϩ mice after training showed significantly higher exploratory preference for the novel object than during training (**, p Ͻ 0.01). After training, exploratory preference for the novel object of NEP Ϫ/Ϫ APP ϩ mice was significantly lower than that of any other group ( ## , p Ͻ 0.01), and the exploratory preference of NEP ϩ/ϩ APP ϩ mice was significantly different from that of NEP ϩ/ϩ APP Ϫ mice ( §, p Ͻ 0.05); two-way ANOVA showed significant main effects of APP transgene (F (1,45) ϭ 14.67; p Ͻ 0.001) and neprilysin deficiency (F (1,45) ϭ 6.96; p Ͻ 0.05) with no interaction between APP transgene and neprilysin deficiency. C, reference memory in a Morris water maze task. Escape latency in each block of the hidden platform during a 60-s session was measured. Repeated-measure two-way ANOVA showed a significant main effect of genotype (F (3,210) ϭ 3.69; p Ͻ 0.05). Post hoc analysis revealed that the escape latency of NEP Ϫ/Ϫ APP ϩ mice was significantly different from the other groups (*, p Ͻ 0.01). D, spatial memory for a platform location performed after training of reference memory of the Morris water maze task. Percent search time for each quadrant during a 60-s session was measured. In all genotypes except NEP Ϫ/Ϫ APP ϩ mice, the time spent in the trained quadrant was significantly longer than in any other quadrant (*, p Ͻ 0.05), but NEP Ϫ/Ϫ APP ϩ mice spent an almost equal duration of time in all quadrants (one-way ANOVA, F (3,44) ϭ 1.380, p ϭ 0.2613). E, contextual conditioned-fear task. Retention session was carried out 24 h after the conditioning. All genotypic mice after conditioning showed a significantly longer freezing time than preconditioning (*, p Ͻ 0.05; **, p Ͻ 0.01). In the freezing time after conditioning, two-way ANOVA showed significant main effects of neprilysin deficiency (F (1,45) ϭ 17.80; p Ͻ 0.01), no significant effects of APP transgene (F (1,45) ϭ 0.96; p Ͻ 0.0001), and a significant interaction between APP transgene and neprilysin deficiency (F (1,45) ϭ 0.96; p Ͻ 0.0001). Post hoc analysis revealed that in this task freezing time in NEP Ϫ/Ϫ APP ϩ mice was significantly shorter than that in any other group after training (#, p Ͻ 0.01). no significant difference in the three groups was observed (Fig. 6E,  right). However, NEP Ϫ/Ϫ APP ϩ mice presented less freezing responses in the contextual tests compared with the other three groups (Fig. 6E,  right). No alteration of nociceptive response was found between the mouse groups; the minimal current required to elicit flinching/running, jumping, or vocalization was similar in all mice (data not shown).
Thus, the NEP Ϫ/Ϫ APP ϩ mice displayed obvious cognitive abnormalities consistently in different learning and memory paradigms, and the neprilysin deficiency in the APP transgenic background exaggerated the action of APP transgene-derived A␤ oligomers on all of the learning and memory tasks tested here.

DISCUSSION
An aberrant increase in A␤ at synapses has been suggested to give rise to synaptic dysfunction (1, 18 -24). A␤ generated from axonal-transported APP is released from presynaptic sites and subsequently accumulates close to the nerve terminal (43,44). The similarity of neprilysin with APP/A␤ in the spatial and topological terms makes this peptidase stand out among other A␤-degrading peptidases (3,4,45) as a candidate regulator of synapse-associated A␤. However, much less is understood about the possible relationship between a local increase in synaptic A␤ and a reduction of A␤-catabolic activity involving neprilysin in the brain; nevertheless, it has been observed that neprilysin in the hippocampus and cerebral cortex is down-regulated upon normal aging (14,15) and from an early stage of AD development (15)(16)(17). The present study demonstrated that dysregulation of the metabolic balance of A␤, resulting from reduced activity of neprilysin, leads to synaptic dysfunction observed as an early change of AD as well as cognitive impairments through an increase in A␤ levels. IEM clearly showed a striking increase in A␤ oligomers at the synapses in the hippocampus and the dentate gyrus, where an obvious difference in the increase of the oligomers was observed in a reverse correlation with innate amounts of neprilysin (11). Thus, A␤ oligomers that escape degradation by neprilysin at the presynapses may diffuse to the synaptic cleft or the extracellular space and may play a critical role in functional alteration of the synaptic plasticity and cognitive ability. The presence of neprilysin at the presynaptic sites is of particular importance in protecting synapses from oligomer-derived synaptotoxicity. The quantity of oligomers at the synapses can be regulated by neprilysin, as schematically demonstrated in Fig. 7.
To reconstitute and analyze the human pathology experimentally in vivo, transgenic mice overexpressing APP have been widely used. However, there has always been an intrinsic difficulty in interpreting data obtained from such mouse models. Because other APP fragments are overproduced, not only A␤, the effects of these non-A␤ fragments on the pathological phenotypes cannot be excluded (4). The neurotrophic activity of sAPP (46,47), for example, enhancement of LTP (48), and a recent finding that APP intracellular domain up-regulates neprilysin in the brain (49) further makes the interpretation complex. We overcame these drawbacks by utilizing the activity of neprilysin (3)(4)(5)(6)(7)(8)(9). Crossbreeding APP transgenic mice with neprilysin-deficient mice resulted in selective elevation of A␤, especially oligomeric forms, at the presynaptic terminals without altering APP metabolism and concomitant impairment of hippocampal LTP and cognitive behaviors. The NEP Ϫ/Ϫ APP ϩ mice also showed functional impairments from an early age (3-4 months) before senile plaque formation and neuronal destruction, clearly segregating the effects of soluble A␤ oligomers from those of insoluble A␤ deposits or indirect effects of inflammatory responses caused by the A␤ deposits.
The LTP recorded in DG and CA1 was significantly suppressed in NEP Ϫ/Ϫ APP ϩ mice. The LTP in DG was more strongly suppressed than that in CA1, reflecting a greater increase in the quantity of A␤ oligomers. There was also a significant difference among the APP transgenic and two non-transgenic mice, the former showing moderate suppression of LTP. This difference appears to be attributable to the presence of A␤ oligomers in the transgenic mice. However, it remains a possibility that A␤ monomer may be involved in the suppression of LTP because the A␤ monomer was significantly increased in the brains of NEP Ϫ/Ϫ APP ϩ mice even though the amount was only 3.7% of total A␤.
Because neprilysin-deficient mice without the APP transgene did not differ from wild-type littermate controls, the suppressed LTP caused by neprilysin deficiency in APP transgenic mice was due to elevation of the FIGURE 7. Schematic representation of neuronal metabolism of A␤ by neprilysin. Neprilysin, after synthesis and maturation in endoplasmic reticulum and Golgi apparatus, undergoes axonal transport and, consequently, resides inside secretory vesicles and on the surface of presynaptic terminals (9,11,14). Likewise, APP is axonal-transported and cleaved by ␤and ␥-secretases to produce A␤ presynaptically (43,44). This concomitant localization and movement of neprilysin and APP/A␤ in the neuron suggests that neprilysin is particularly involved in regulating the A␤ levels associated with synapses and synaptic clefts. On the other hand, at the postsynaptic sites, A␤ binds to the ␣-7 nicotinic receptor and appears to regulate synaptic activity through membrane trafficking and destabilization of N-methyl-D-aspartate and ␣amino-3-hydroxy-5-methyl-4-isoxazolepropionate receptors, respectively (53,54). Furthermore, glial cells internalizes oligomeric A␤ as well as fibrillar A␤ (55,56) and may be critically involved in the removal of oligomeric A␤ from synaptic clefts. To our knowledge no other A␤-degrading enzyme candidates (45) are localized to presynapses. Our results indicate that neprilysin regulates the quantities of not only A␤ monomers but also dimers, trimers, and tetramers.
transgene-derived A␤ oligomers, not to altered metabolism of potential endogenous neprilysin substrates. We found that neprilysin deficiency does not elevate in vitro"neprilysin substrate" neuropeptides, such as somatostatin, cholecystokinin, and substance P, in mouse brains. 5 Saria et al. (50) also reported that the levels of Leu-and Met-enkephalins were not increased in the brains of the neprilysin-deficient mice compared with that of wild-type mice. In clear contrast to LTP, there was no significant difference between genotypes in basal synaptic transmission or short-term potentiation. It is notable that, whereas short-term potentiation is known to involve amplification of the presynaptic transmitter release mechanism (51), LTP is more closely associated with postsynaptic signal transduction cascades (52). It is likely that the A␤ oligomers have exerted their effects mainly via postsynaptic processes. This notion is supported by recent findings that A␤ regulates synaptic activity through membrane trafficking and destabilization of N-methyl-D-aspartate and ␣amino-3-hydroxy-5-methyl-4-isoxazolepropionate receptors, respectively (53,54).
The NEP Ϫ/Ϫ APP ϩ mice displayed obvious cognitive abnormalities consistent in different learning and memory paradigms. The neprilysin deficiency in the APP transgenic background exaggerated the action of APP transgene-derived A␤ oligomers on the spatial working memory in the Y-maze tasks, visual recognition memory in the novel-object recognition task, and acquisition and retention of reference memory in the water maze task. It is known that the spatial memory in these learning and memory paradigms requires integrative control function of the hippocompal formation. These results indicate that neprilysin-sensitive A␤ oligomers could cause impairments of synaptic integrity and plasticity as well as cognitive function.
Our findings indicate that a strategy using neprilysin activity (3, 4, 45) as well as secretase inhibitors (1) may have a chance of improving or alleviating the memory-associated symptoms of AD by reducing A␤ oligomers at the synapses even before plaque deposition. Recently, we found that somatostatin is capable of reducing the levels of A␤, preferentially A␤ 42 , in the brain by up-regulating neprilysin activity (3,4,13). Therefore, blood-brain barrier-permeable, non-peptidic agonists selective to the somatostatin receptor(s) localized in AD-affected regions might be a new promising target, particularly for lowering A␤ 42 levels.