Protection against β-Amyloid-induced Apoptosis by Peptides Interacting with β-Amyloid*

β-Amyloid peptide produces apoptosis in neurons at micromolar concentrations, but the mechanism by which β-amyloid exerts its toxic effect is unknown. The normal biological function of β-amyloid is also unknown. We used phage display, co-precipitation, and mass spectrometry to examine the protein-protein interactions of β-amyloid in normal rabbit brain in order to identify the biochemical receptors for β-amyloid. β-Amyloid was found to bind primarily to proteins involved in low density lipoprotein and cholesterol transport and metabolism, including sortilin, endoplasmic reticulum-Golgi intermediate compartment 2 (ERGIC2), ERGIC-53, steroid 5α-reductase, and apolipoprotein B. β-Amyloid also bound to the C-reactive protein precursor, a protein involved in inflammation, and to 14-3-3, a protein that regulates glycogen synthase kinase-3β, the kinase involved in tau phosphorylation. Of eight synthetic peptides identified as targets of β-amyloid, three were found to be effective blockers of the toxic effect of β-amyloid on cultured neuronal cells. These peptides bound to the hydrophobic region (residues 17–21) or to the nearby protein kinase C pseudo-phosphorylation site (residues 26–30) of β-amyloid, suggesting that these may be the most critical regions for β-amyloid effector action and for aggregation. Peptides or other small molecules that bind to this region may protect against β-amyloid toxic effect by competitively blocking its ability to bind β-amyloid effector proteins such as sortilin and 14-3-3.

␤-Amyloid peptide produces apoptosis in neurons at micromolar concentrations, but the mechanism by which ␤-amyloid exerts its toxic effect is unknown. The normal biological function of ␤-amyloid is also unknown. We used phage display, coprecipitation, and mass spectrometry to examine the proteinprotein interactions of ␤-amyloid in normal rabbit brain in order to identify the biochemical receptors for ␤-amyloid. ␤-Amyloid was found to bind primarily to proteins involved in low density lipoprotein and cholesterol transport and metabolism, including sortilin, endoplasmic reticulum-Golgi intermediate compartment 2 (ERGIC2), ERGIC-53, steroid 5␣-reductase, and apolipoprotein B. ␤-Amyloid also bound to the C-reactive protein precursor, a protein involved in inflammation, and to 14-3-3, a protein that regulates glycogen synthase kinase-3␤, the kinase involved in tau phosphorylation. Of eight synthetic peptides identified as targets of ␤-amyloid, three were found to be effective blockers of the toxic effect of ␤-amyloid on cultured neuronal cells. These peptides bound to the hydrophobic region (residues [17][18][19][20][21] or to the nearby protein kinase C pseudo-phosphorylation site (residues 26 -30) of ␤-amyloid, suggesting that these may be the most critical regions for ␤-amyloid effector action and for aggregation. Peptides or other small molecules that bind to this region may protect against ␤-amyloid toxic effect by competitively blocking its ability to bind ␤-amyloid effector proteins such as sortilin and 14-3-3.
␤-Amyloid (A␤) 2 is a toxic peptide produced by cleavage of amyloid precursor protein (APP) by ␤and ␥-secretases. Application of A␤ to cultured cells at micromolar concentrations causes apoptosis (1), and lower concentrations cause up-regulation of apoptosis markers caspase-3 (2) and annexin- (3). Biochemical effects of A␤ include activation of calcium channels (4,5), inactivation of potassium channels (6), production of free radicals (7), excitotoxicity through activation of N-methyl-D-aspartate receptors (8,9), inhibition of protein kinase C (10,11), and glutamate accumulation leading to increased Ca 2ϩ levels (12). Intracerebroventricular injection of A␤ produces impairment of spatial memory and non-spatial long term memory (13,14), reduction of protein kinase C activity (15), induction of apoptosis (13), and activation of astrocytes and microglia to release excessive amounts of inflammatory cytokines (16). Transgenic animals expressing human A␤ exhibit many of the pathologies of Alzheimer disease (AD), including cognitive deficits (17), age-related formation of amyloid plaques, activation of astrocytes and microglial cells, vascular amyloid pathology, degeneration of cholinergic nerve terminals, and reduced lifespan (18). However, transgenic mice expressing normal human APP do not exhibit the neurofibrillary tangles and significant neuronal loss characteristic of AD (18).
Formation of A␤ from APP is dependent on the intracellular transport system. APP is transported from the ER and Golgi to the cell surface membrane (19), where it may be cleaved by ␣-secretase. Protein kinase C-activated ␣-secretases also reside in the trans-Golgi network (20), which is a major site for ␤-secretase activity. Uncleaved APP is then internalized into endocytic compartments, where it is cleaved by ␤and ␥-secretase to produce A␤ (21,22). Both neurons and many non-neuronal cells also contain membrane ␤and ␥-secretases (23), which produce A␤ that is secreted into the extracellular space. In neurons, A␤ can also be produced in the endoplasmic reticulum (A␤- ) and the trans-Golgi network (A␤-(1-40)) (24,25). However, A␤ does not remain in these two compartments. A␤ has been found in a number of other compartments, including cytosol (26 -29), lysosomes (30,31), mitochondria (32,33), and even in cell nuclei (34). This widespread distribution may result in part from uptake of extracellular A␤ into neurons (28,35) and astrocytes (36). Because A␤ lacks an ER signal sequence, it may also be identified by the ER as a misfolded protein and translocated across the ER or Golgi membrane (37) by sec61 (38).
The normal functions of APP and A␤ are unknown. APP is an integral membrane protein with high affinity for copper (39,40). It has been suggested that APP is involved in neurodevelopment (41) and is essential for neuronal growth (42,43). Mutant mice in which APP has been knocked out develop reactive gliosis, weight loss, cognitive defects, and reduced levels of presynaptic marker proteins (44), indicating generalized central nervous system pathology. Down-regulation of APP inhibits neurite outgrowth (45), and anti-APP antibodies block memory formation in chicks (46). After being transported along nerve fibers, APP participates in synaptogenesis (47) and cell adhesion (48). Thus, both APP and A␤ may play an important role in normal synaptic plasticity and neuronal growth.
The A␤ peptide is not only produced in AD but has also been found in cerebrospinal fluid and blood plasma of normal patients (49). The production and secretion of A␤ is regulated by neuronal activity (50). Kamenetz et al. (50) found that APP reversibly depresses synaptic transmission by a mechanism mediated by activation of APP cleavage by N-methyl-D-aspartate receptors. This suggests that A␤ is normally produced by neurons and has one or more functions in normal cells. To identify the biochemical pathways in which A␤ participates, we performed phage display screenings and co-precipitation experiments to identify the binding partners of A␤ in normal rabbit brain. We also examined the ability of peptides based on these binding partners to block the toxic effects of A␤. A␤ was found to interact with apolipoprotein B and the C-reactive protein precursor, the transporter proteins sortilin, ER-Golgi intermediate compartment 2 (ERGIC2), and ERGIC-53, and the regulatory proteins 14-3-3⑀ and 14-3-3␥. Peptides based on these binding regions were effective blockers of A␤ toxicity.

EXPERIMENTAL PROCEDURES
Materials-5Ј-Nucleotidase was purchased from Biomol International (Plymouth Meeting, PA). Phosphodiesterases 3A and 3B were obtained from EMD Biosciences (San Diego CA). Antibodies against 14-3-3␤, -␦, -␥, and -were obtained from Upstate (Charlottesville, VA). Anti-ERGIC2 was obtained from Aviva Systems Biology (San Diego, CA). Kidins antibody was obtained from Orbigen (San Diego, CA). Sortilin antibody was obtained from Abcam Inc. (Cambridge, MA). Other antibodies were obtained from Santa Cruz Biotechnology (Santa Cruz, CA). All ␤-amyloid and ␤-amyloid derivatives were purchased from Anaspec (San Jose, CA). UltraLink protein A and protein G-agarose, sequencing grade trypsin, and Pro-Ject protein transduction reagent were purchased from Pierce. Caspase-3 substrate (Ac-DEVD-amidomethylcoumarin) was purchased from American Peptide Co. (Sunnyvale, CA). Rabbit and mouse brains were obtained from Pel-Freez (Rogers, AR). Culture media were obtained from Invitrogen. Other chemicals were obtained from Sigma.
Cell Culture-Rat hippocampal H19-7/IGF-IR cells (ATCC, Manassas, VA) were plated onto poly-L-lysine-coated plates and grown at 35°C in DMEM/10% fetal calf serum for several days until Ϸ50% coverage was obtained. The cells were then induced to differentiate into a neuronal phenotype by replacing the medium with 5 ml of N2 medium containing 10 ng/ml basic fibroblast growth factor at 39°C and grown in T-75 flasks at 37°C.
␤-Amyloid Oligomerization-Human A␤ (Anaspec 20276, 1 mg) was dissolved in 3 ml of deionized water by the addition of a minimum volume of 1% ammonium hydroxide and was incubated for 3 days at 37°. The solution was then reduced in volume to 0.886 ml by lyophilization and incubated for an additional 48 h at 37°C. The oligomerized A␤ was then stored at Ϫ80°C.
Characterization of Oligomerized A␤-A 10-l sample of A␤ was injected onto a Macrosphere GPC100 7-m size-exclusion high performance liquid chromatography column (250 ϫ 4.6-mm inner diameter) and eluted with 20 mM ammonium carbonate. Absorbance was monitored at 210 nm. The oligomerized A␤ used in each experiment was tested to ensure that a similar degree of oligomerization had occurred.
Synthetic Peptides-A␤-binding peptides were synthesized by Genscript (Piscataway, NJ). All peptides were dissolved in phosphate-buffered saline at a concentration of 1 mM, except for peptide #3 (VSVGMLWC), which was dissolved at 0.1 mM. Peptides were then sterilized by filtration before use. One vial of Pro-Ject protein transduction reagent (Pierce) was then dissolved in 250 l of CHCl 3 . Four l of Pro-Ject solution were transferred to 1.5-ml polypropylene centrifuge tubes and evaporated with nitrogen. Peptide was diluted in phosphate-buffered saline to 9.1 g/ml, and 0.22 g of peptide was added. The vial was incubated for 5 min, then vortexed, and the volume was brought to 0.5 ml with serum-free DMEM.
Cells were grown in 12-well plates containing 1 ml of N2 culture medium. When the cells reached 75-80% confluence, they were washed with serum-free DMEM, and peptide/Pro-Ject mixture dissolved in 0.5 ml of serum-free DMEM was added. After 4 h of incubation at 37°C, 0.5 ml of DMEM containing 20% fetal bovine serum was added. After 18 h, oligomerized A␤ was added. Cells were monitored daily. Under these conditions some cell death is visible within 48 -72 h. The number of visibly apoptotic cells continued to increase for several more days. Therefore, we stopped the experiment after 7 days (6 days after adding A␤). The medium was removed, and the cells were washed twice with 1ϫ phosphate-buffered saline. A 100-l aliquot of phosphate-buffered saline was added, and the cells were removed by gentle scraping. The cells were homogenized by sonication and stored at Ϫ80°C.
␤-Galactosidase-We tested for senescence using the hydrolysis by ␤-galactosidase at pH 6 of X-gal, a commonly used ␤-galactosidase substrate. Under these conditions, ␤-galactosidase is easily detectable in senescent cells but undetectable in quiescent, immortal, or tumor cells (51). To measure ␤-galactosidase, cell homogenate (20 l) was incubated in 100 l of 0.5 M Tris-HCl, pH 6.8, containing 0.1 mg/ml X-gal. After 24 h at 37°C, the samples were diluted to 1 ml, and absorbance at 610 nm was measured.
pBad-Bad (Bcl-x L /Bcl-2-associated death promoter) is a member of the Bcl-2 family and regulates the survival signal (52). Unphosphorylated Bad dimerizes with Bcl-2 and Bcl-x L , which neutralizes their anti-apoptotic activity. Activation of the phosphoinositol 3-kinase pathway ultimately leads to activation of Akt, which phosphorylates Bad on serine 136. Activation of mitogen-activate kinase pathways results in phosphorylation of Bad on serine 112. Phosphorylated Bad is sequestered from its proapoptotic role by binding with 14-3-3 protein (53). Thus, a decrease in Bad phosphorylation indicates apoptosis (54 -60). pBad was measured by densitometry of Western blots stained with phospho-Bad antibody. Image quantitation and molecular weight estimation were done on 16-bit images using the Unix-based image analysis program imal. The background value for each band was calculated by fuzzy k-means clustering analysis of the appropriate region of the image (61)(62)(63)(64). Densitometry results are expressed as units of relative staining.
Caspase-3 Assay-Members of the caspase family of cysteine aspartyl proteases are related to the Caenorhabditis elegans CED-3 death protein. Caspases-8, -9, and -10 are activated by receptor clustering and are known as "initiator caspases" (65). Caspases-3, -6, and -8 are activated by changes in mitochondrial permeability that are associated with apoptosis and are known as "effector caspases." Effector caspase-3 proteolyzes a number of substrates, including DNA fragmentation factor/inhibitor of caspase-activated deoxyribonuclease (DFF45/ICAD), poly(ADP-ribose) polymerase, gelsolin, and nuclear lamins (66). Proteolysis of DFF45/ICAD liberates the DNase subunit of DFF to cause chromatin degradation (67). Proteolysis of poly-(ADP-ribose) polymerase has been used by many researchers as a marker for apoptosis. We measured caspase-3 activity fluorometrically using the commercially available substrate Ac-Asp-Glu-Val-Asp-NH 2 -methylcoumarin, which contains the poly-(ADP-ribose) polymerase cleavage site (68). Hydrolysis of this peptide yields a species that fluoresces at 440 -460 nm.
Phosphodiesterase Assay-Phosphodiesterase was measured by the method of Alvarez and Daniels (69) with slight modifications. Fines were removed from acidic alumina (Brockmann grade 1) by resuspending several times in water. Bio-Rad Poly-Prep columns were packed with a slurry containing 1.3 g of alumina to a height of 3 cm. Columns were precycled with 0.1 M ammonium acetate to remove any cAMP, then equilibrated with water. In a 0.5-ml polypropylene centrifuge tube, a volume equivalent to 0.15 l of rat brain homogenate was diluted to 100 l with water and 10 l of buffer (1 M Tris-HCl, pH 7.5, plus 2 mM MgSO 4 ), and 0.5 l of [ 3 H]cAMP were added. After 10 of min incubation at 30°C, the samples were boiled and cooled, and 0.1 l of 5Ј-nucleotidase was added (Biomol, 500 kilounits/ ml). The samples were incubated for 30 min at 30°C and applied to the column. The adenosine was eluted with 4 ml of 5 mM HCl, mixed with scintillation fluid, and counted in a scintillation counter.
Western Blots-Protein was measured using the Bradford dye binding technique (71). Samples were boiled in SDS sample buffer and loaded on a 4 -20% polyacrylamide gel. Identical amounts of protein were applied to each gel. The samples were subjected to SDS-polyacrylamide gel electrophoresis, nitrocellulose blotting, and antibody staining using commercial anti-bodies and alkaline phosphatase-conjugated secondary antibody as described previously (72).
Co-precipitation-One rabbit brain was homogenized by sonication in 3 volumes of 10 mM Tris-HCl, pH 7.4, containing 50 mM NaF and 1 mM phenylmethylsulfonyl fluoride. The homogenate was centrifuged for 20 min at 100,000 ϫ g. The supernatant was then re-centrifuged to produce the cytosolic fraction, and the original pellet was sonicated in the original volume of 10% N-lauroyl sarcosine or 10% CHAPS and centrifuged for 20 min at 100,000 ϫ g. The detergent extracts and cytosol were divided into 4-ml aliquots and incubated in 17 ϫ 100-mm polypropylene tubes (Falcon 2059) with 10 g of biotin-LC-␤-amyloid-(1-42) (Anaspec) for 1 h at room temperature. Avidin-agarose (50 l) was added, and the incubation was continued on an orbital shaker (800 rpm) for an additional 60 min. The mixture was cooled on ice, transferred to a Bio-Rad Poly-Prep column, and rapidly washed with 2 ml of ice-cold 10 mM Tris-HCl plus 100 mM NaCl using pressure. The co-precipitated proteins were eluted with 2 ml of 4 M NaCl and 2 ml of 0.2 M glycine-HCl, pH 2.2. The eluted fractions were concentrated and desalted in Centricon-3 ultrafiltration units, then mixed with SDS sample buffer for Western blotting. The non-eluted proteins were eluted from the agarose beads by boiling with SDS sample buffer. Samples were applied to a 4 -20% SDS-polyacrylamide gel and stained with Coomassie Blue.
Co-precipitation Using Protein A-Agarose-Rabbit brain or N-lauroyl sarcosine extracts (3 ml) were incubated with 1 g of A␤-(1-42) for 1 h at room temperature. Anti-A␤ antibody (Calbiochem 171604, 9.09 g) was added. After 45 min of shaking, 30 l of protein A-agarose was added, and the samples were incubated with shaking for 30 min at room temperature. Samples were then transferred to disposable plastic columns and rapidly washed with 1 ml of 10 mM Tris-HCl, pH 7.4, containing 150 mM NaCl and 1 mM phenylmethylsulfonyl fluoride. The agarose beads were transferred to a centrifuge tube, proteins were eluted with SDS sample buffer, and the proteins were identified by electrophoresis and Western blotting.
Because phage display screening using low M r peptides can be inefficient, we also performed a solution panning using mouse anti-A␤-(18 -30) antibody (Calbiochem 171587), which was determined to give the broadest reactivity for all forms of A␤ of the four antibodies we tested. Phage were purified using Ultralink protein A-agarose alternating with Ultralink Protein G-agarose to avoid purifying phage that had affinity for protein A or protein G. Three rounds of solution panning were performed.
Identifying A␤ Targets by Mass Spectrometry-One rabbit brain was sonicated in 3 volumes of buffer (10 mM Tris-HCl pH 7.4, 50 mM NaF, 1 mM phenylmethylsulfonyl fluoride, 0.3 mM leupeptin, 0.125 mM pepstatin, and 5% CHAPS) and centrifuged at 100,000 ϫ g. The supernatant was aliquoted into four 17 ϫ 100-mm Falcon 2059 tubes and incubated for 1 h at room temperature in the presence of 10 M CuCl 2 or 1 M EDTA with 100 g of fresh biotin-LC-␤-amyloid-(1-42) or biotin-LC-␤-amyloid-(1-42) that had been preincubated at 37°C for 4 days. After incubation, 0.5 ml of avidin-agarose was added, and incubation was continued with shaking for 30 min. The sample was cooled to 4°C, transferred to a disposable column, and rapidly washed with 4 ml of ice-cold homogenization buffer. The proteins with affinity for A␤ were then sequentially eluted with 2 ml of 0.5 M NaCl, 1.5 M NaCl, 4 M NaCl, 1 M Tris-HCl, pH 9, or 0.2 M glycine HCl, pH 2.2. All buffers contained 0.1% CHAPS plus 1 mM phenylmethylsulfonyl fluoride. All NaCl solutions also contained 10 mM Tris-HCl, pH 7.4. The eluates were concentrated and desalted in Centricon-3 ultrafiltration units. The proteins were then separated by SDS-polyacrylamide gel electrophoresis followed by Coomassie Blue staining.
Mass Spectrometry-After excision, Coomassie-stained protein bands were destained by washing with acetonitrile/ acetic acid/water (1:1:1), reduced, alkylated, and digested with 120 ng of trypsin using the in-gel method (73,74). Digestion with trypsin was carried out overnight at 37°C. Peptides were twice extracted from the gel into 50% acetonitrile, 5% formic acid. The extracts were pooled, the volume was reduced by vacuum centrifugation, and the final volume was brought up to 6 l with 25 mM ammonium bicarbonate. The peptides from the tryptic digests were analyzed by liquid chromatography/tandem mass spectrometry (LC-MS/MS). Liquid chromatography was performed using an LC Packings UltiMate NanoLC at 250 nl/min using a PepMap C18 reverse-phase 100-Å pore column. Peptides were separated using a linear gradient from 95% A plus 5% B to 5% A plus 95% B (A ϭ H 2 O plus 0.1% formic acid; B ϭ 80% acetonitrile plus 0.1% formic acid) over 38 min followed by 95% B for 17 min. The LC effluent was electrosprayed directly into the sampling orifice of an LTQ mass spectrometer (Thermo Finnigan) using a nanospray interface. The LTQ was operated to collect MS/MS spectra in a data-dependent manner with up to five of the most intense ions that exceeded a pre-set threshold being subjected to fragmentation and analysis. The MS/MS data generated were analyzed, and matches to protein sequences in the NCBI non-redundant nr data base (human/trypsin subset) were determined using SEQUEST (75) program.
Sequence identification was based on the cross-correlation normalized for peptide length (XcorrЈ) and ␦ correlation (⌬Cn) scores. SEQUEST-derived peptide identifications and protein identifications were evaluated for statistical significance and filtered with the Peptide Prophet and Protein Prophet software tools (76). In each case the predicted M r and pI of the identification matched the M r and pI values on the gel within Ϯ5%.
Although an unambiguous identification cannot be made from a seven-amino acid sequence, in most cases a mammalian-

A␤-binding peptides identified by phage display
Only peptides found in two or more clones are shown. For peptides for which BLAST search yielded two or more results, the ambiguities were resolved by co-precipitation studies, and the correct protein is listed first. TD, tumor necrosis factor receptor-associated factor (TRAF) domain. constrained BLAST search identified only a single protein or in some cases two candidate proteins as having a close match (Table 1). These ambiguities were resolved by co-precipitation studies (below). The peptides found by phage display contained sequences found in phosphodiesterase 3A and 3B and a number of proteins involved in cholesterol transport and metabolism (sortilin and steroid 5␣-reductase type 2). The strongest bind-ing (17 clones) was to a peptide found in the P2X2 purinergic receptor and the protein Suppressor of Ty 3 homolog (SUPT3H) (77).

Amyloid fragment
To confirm the interactions between A␤ and the candidate proteins and to resolve the ambiguities in the phage display results, rabbit brain cytosol and brain N-lauroyl sarcosine extract or CHAPS extract were incubated with biotin-LC-␤-amyloid-(1-42). A␤-binding proteins were collected using avidin-agarose and analyzed by Western blotting. Because the protein interactions of some candidate proteins (such as 14-3-3) depend on Ca 2ϩ , the coprecipitation experiments were carried out in the presence and absence of added calcium. Because copper also binds to A␤ (78,79), Cu 2ϩ (10 M) was also added to some samples. These co-precipitation experiments confirmed that A␤ interacts with the following proteins in rabbit brain extracts: C-reactive protein precursor, 14-3-3⑀, 14-3-3␥, GDNF receptor, ERGIC-53, P2X2, ERGIC2, apolipoprotein B, and sortilin ( Fig. 1; see Table 3). No co-precipitation could be found for phosphodiesterase 3A, phosphodiesterase 3B, CNPase, tubulin, SorLa, KIDINS-220, SUPT3H, or CD34. The interactions of A␤ with 14-3-3 and ERGIC-53 were calcium-dependent. The co-precipitation signal with C-reactive protein precursor was very weak, possibly because of the low levels of expression of C-reactive protein precursor in brain. A␤ co-precipitated with the P2X2 purinergic receptor but not with CD34 or SUPT3H, indicating that peptide YQDSAKT represented P2X2. Similarly, co-precipitation experiments indicated that the peptide SVLDRQR corresponded to sortilin and not KIDINS-220 because A␤ co-precipitated only with sortilin ( Fig. 1); however, the sortilin band was relatively weak, suggesting low levels of sortilin in the rabbit brain. No co-precipitation was observed with a SorLA antibody. Interestingly, co-precipitation was observed for both ERGIC2 and ERGIC-53 (Fig. 2) despite the fact that a peptide matching LGSYKPS is only found in ERGIC2. Table 2 for details. AB, A␤; GDNFR, GDNF receptor. The inability to detect tubulin by co-precipitation in either cytosol or N-lauroyl sarcosine extract despite the strong signal observed by mass spectrometry is not surprising since binding of A␤ may be specific to particular oligomeric or fibrillar forms of A␤. Tubulin binding to A␤ (80) and carboxyl-terminal fragments of A␤ precursor protein (81) has been observed by previous researchers. Antibodies were not available for tumor differentially expressed protein or TMS membrane protein; thus, it was not possible to determine which of these two proteins bound A␤. The proteins that were confirmed by co-precipitation to interact with A␤ are summarized in Table 2.

FIGURE 1. Co-precipitation of A␤-(1-42) and A␤-binding proteins in rabbit brain extracts. See
To identify A␤-interacting proteins by mass spectrometry, biotin-LC-␤-amyloid-(1-42) was incubated with CHAPS extract from rabbit or mouse brain, collected with avidin-agarose, washed, and eluted with increasing salt and buffer concentrations. In one experiment, fresh and oligomerized biotin-LC-␤-amyloid were also compared. Because Cu 2ϩ has been shown to bind to A␤ (78,79), 10 M CuCl 2 was also added to some samples. Rabbit brain was used because the sequence of rabbit A␤ is identical to that of humans. The proteins were analyzed by SDS-polyacrylamide gel electrophoresis, eluted from the gel, and identified by liquid chromatography/ion trap tandem mass spectrometry. The proteins are shown in Table 3.
The predominant A␤-binding proteins found by mass spectrometry were 14-3-3 and CNPase. Similar patterns were observed for rabbit and mouse, although more mouse proteins than rabbit proteins were identified, presumably because the rabbit nr data base is relatively incomplete. The addition of Cu 2ϩ had no effect on the binding of A␤ for any of the proteins (Fig. 2). No differences were observed between fresh and oligomerized biotin-A␤ (not shown). Oligomerization of A␤ is essential for toxicity (82); however, oligomerization of A␤-(1-42) can occur rapidly (within minutes) in solution (83). Thus, the fresh biotin-A␤ used in these experiments may also have been partially oligomerized.
Effects on Phosphodiesterase Activity-Because both mass spectrometry and phage display both implicated some form of phosphodiesterase, we measured the effects of A␤ on cGMPsensitive PDE 3A and 3B. A␤ had no effect on commercially available purified phosphodiesterase 3A but produced a slight inhibition of phosphodiesterase activity in rabbit brain homogenate. A␤ peptides also produced a slight (Ϸ20%) inhibition of purified PDE 3B (Fig. 2). A␤ had no effect on the inhibition of PDE 3A or PDE 3B by cGMP (not shown). A␤ also had no effect on CNPase activity (Fig. 3D).
Protection against ␤-Amyloid-To test whether the A␤-binding peptides could protect against the toxic effects of A␤, we applied eight synthetic peptides mixed with Pro-Ject, a cationic lipid protein transduction reagent, to cultured H19-7/ IGF-IR cells. These cells are hippocampal neurons from Rattus norvegicus that have been immortalized by retroviral transduction of temperature-sensitive tsA58 SV40 large T antigen (84). After 18 h, oligomerized A␤-(1-42) was applied to the cells. The cells were collected after 7 days in culture and analyzed for markers of apoptosis (caspase-3 and pBad) and senescence (␤-galactosidase). The peptide sequences are shown in Table . The oligomerized A␤, analyzed by size-exclusion high performance liquid chromatography, consisted predominantly consisted of oligomeric species between 10 and about 60 kDa (Fig. 4).

DISCUSSION
Identification of the functional motifs of a protein along with identification of its binding partners can give valuable clues about its biological role and, in the case of A␤, can provide insight on ways of blocking its toxicity. We have shown that A␤ peptide binds to 14-3-3, ERGIC-53, ERGIC2, sortilin, P2X2, apoB, steroid reductase, and the C-reactive protein precursor. The interactions with 14-3-3 and sortilin are of particular relevance to Alzheimer disease. Sortilin expression has been shown to correlate inversely with AD neuropathology (85). Levels of LR11/SorLA, which is also the receptor for apolipoprotein E, are decreased in sporadic AD (86). Its apparent role is to target proteins in the Golgi for transport to late endosomes. The 14-3-3 protein, which has many similarities to the Parkinson disease-associated protein ␣-synuclein (87), is found in neurofibrillary tangles (88), binds to tau, and is involved in phosphorylation of tau by glycogen synthase kinase 3␤ (89,90). Glycogen synthase kinase 3␤, which is the principal enzyme involved in phosphorylating tau, is regulated by 14-3-3 (91). Increases in 14-3-3 have been reported in patients with Alzheimer disease (92). Phosphorylation of Ser-9 in glycogen synthase kinase 3␤ (GSK-3␤) promotes binding of GSK-3␤ to 14-3-3. Although A␤ is produced in the trans-Golgi and ER, it has also been found in the cytosol (26 -29), where 14-3-3 is usually

Sequences of A␤-binding peptides tested in cultured cells for their ability to protect against A␤
The binding region on A␤ found by phage display is also shown. The probable identity of each peptide was determined by BLAST using co-precipitation studies to discriminate between high-scoring proteins. TD, tumor necrosis factor receptorassociated factor (TRAF) domain. located. However, 14-3-3 also has important functions in the Golgi and endoplasmic reticulum (93,94) and in ER-cell surface transport (95). 14-3-3 was also one of a small number of proteins found to be significantly oxidized after intracerebral injection of A␤-(1-42) (96). Thus, binding of A␤ to 14-3-3 could be a missing link that connects two important pathways in AD (A␤ oligomerization and neurofibrillary tangles). Based on the overlapping peptides used in the phage display experiment, the binding region for phosphodiesterase and steroid reductase can be narrowed down to amino acids 12-20 of A␤ (VHHQKLVFF). Likewise, the binding sites for sortilin and apoB are within the region of amino acids 25-25, and the binding region for P2X2 is within the region of amino acids 12-28 (Fig. 6). This region includes a protein kinase C pseudo-phosphorylation site (amino acids 26 -30, SNKGA). Although no phosphorylation of this site by protein kinase C has been reported, it binds protein kinase C and is critical for the inhibition of protein kinase C by low micromolar concentrations of A␤-(1-42) and A␤-(25-35) (11).

Peptide
Orner et al. (97) also recently used phage display to identify proteins that might mediate aggregation of A␤. The predominant peptide motif found by of Orner et al. (97) was a fragment identical to a region of A␤ itself (QKLVFF) containing the ␣-secretase cleavage site and the "hydrophobic patch" (Fig. 6), suggesting that this region is critical for A␤ aggregation. This supports earlier results indicating that the two phenylalanines are critical for A␤ self-aggregation (98). An N-methylated KLVFF peptide, in which the backbone NH group is replaced by an N-methyl group, was shown to reduce the cytotoxicity of A␤-  in cultured PC12 cells (99). Because of its low molecular weight, A␤ is difficult to detect in SDS gels, and interaction of A␤ with itself was not seen in our experiments. The fact that A␤ was not seen in our phage display may mean that the affinity of A␤ for itself is lower than its affinity for the other proteins detected. This is consistent with our experience and the experience of previous researchers with A␤, which requires a minimum of 48 h before oligomers form, even in a 250 M solution of pure A␤ (cf. Fig.  4). It is also possible that, because our A␤ was already aggregated, it becomes less likely to bind to additional A␤ molecules.
Therefore, it is likely that the hydrophobic region of A␤ and the nearby protein kinase C binding region are the most critical regions of A␤ for both its effector action and for aggregation of A␤. Peptides or other small molecules that bind to this region should protect against the toxic effects of A␤ by competitively blocking its ability to bind A␤ effector proteins such as sortilin. This was confirmed by our neuroprotection studies. Peptides #3 and #4, based on steroid reductase and sortilin, respectively, were the most effective at protecting cultured neurons from the apoptotic effects of A␤. Further experimentation is necessary to determine whether these or similar peptides, when administered intracerebroventricularly or delivered across the blood-brain barrier, can protect against A␤ in vivo. However, delivery to the brain may not be necessary for A␤-binding peptides to be effective. Experiments and clinical tests with A␤ antibodies have shown that chelation of A␤ in the circulation is useful in principle as a therapeutic strategy against AD.
Research with a similar goal was carried out by Blanchard et al. (100,101), who used a combinatorial library of hexapeptides containing random combinations of Ala, Ile, Val, Ser, Thr, and Gly, selected on the basis of their ability to complex with A␤, to identify potential therapeutic peptides. Additional peptides were engineered by adding prolines and amino-or carboxylterminal modifications. However, none of the peptides in either Blanchard et al. or Orner et al. (97) matched any proteins found in the present study. This would not be unexpected, considering that Blanchard et al. used rational design rather than library screening to create their peptides. A similar approach was taken by Chalifour et al. (102), who used rational peptide design to create inhibitors of A␤ fibrillogenesis using D-amino acids.
Our work confirms the interactions between A␤ and tubulin, CNPase, and myelin basic protein, found by Verdier et al. (80), who studied synaptosomal proteins that co-precipitated with fibrillar A␤. Differences between their results and ours may be  Our evidence from mass spectrometry and phage display as well as that of Verdier et al. (80) indicate that some form of phosphodiesterase 3 or CNPase interacts with A␤. However, in our purified system A␤ had only a minor effect on PDE activity, and co-precipitation experiments from brain cytosol or detergent extracts showed little evidence of a direct interaction. Therefore, it is possible that the interaction between A␤ and PDE is indirect. Phorbol ester-induced phosphorylation of PDE 3A promotes binding to 14-3-3 (104). Phosphorylation of PDE 3B by protein kinase A also promotes binding to 14-3-3 (105). Thus, the interaction of A␤ with phosphodiesterase and CNPase may be mediated by 14-3-3. Further investigation of the interactions between A␤ and 14-3-3 may shed additional light on the possible involvement of PDE and CNPase.
We have also confirmed that A␤ interacts with proteins involved in inflammation. C-reactive protein, a member of the pentraxin family, is an acute-phase protein normally found in plasma. However, C-reactive protein immunoreactivity is also detectable in temporal cortex of AD patients (106), more specifically in neurofibrillary tangles (107,108). C-reactive protein mRNA is also detectable in pyramidal neurons, indicating that it is synthesized in the brain, and C-reactive protein is up-regulated in AD (109). Patients with the pathogenic apolipoprotein APOE4 allele have lower levels of C-reactive protein than normal patients (110).
A␤ also binds to the ERGIC marker protein ERGIC-53, which is involved in the calcium-dependent transport of glycoproteins such as APP from the ER to the Golgi intermediate compartment (111), where presenilin is located (112). Thus, ERGIC-53, like sortilin, may serve a transport function. Nicastrin, a glycoprotein component of the ␥-secretase complex, interacts with ERGIC-53 (113). However, a role of ERGIC-53 in AD has not been established. ERGIC2 (originally named PTX1) is a similar transport protein originally found in prostate (114) and may also serve to transport APP or A␤.
Apolipoprotein B-Apolipoprotein B, like ApoE, is a component of low density lipoproteins and is a suspected factor in ather-  osclerosis. ApoE4 is a genetic predisposing factor in AD. Serum apoB levels are increased in AD (115,116), and although predominantly a serum protein, apoB is also found in hippocampus, where it is associated with hippocampal amyloid deposits and neurofibrillary tangles (117). Overexpression of apoB in mice increases APP expression in mice fed a high cholesterol diet (118). P2X2-Purinergic receptors are up-regulated in AD. Recent studies have shown that caffeine and adenosine receptor antagonists prevent A␤-induced cognitive deficits in mice (119) and reduce A␤ production (120). P2X2 has been shown to interact with Fe65 (121), an adaptor protein for APP that associates with tau in vivo (122).