β-Amyloid Is Different in Normal Aging and in Alzheimer Disease*

The mechanism of neurodegeneration caused by β-amyloid in Alzheimer disease is controversial. Neuronal toxicity is exerted mostly by various species of soluble β-amyloid oligomers that differ in their N- and C-terminal domains. However, abundant accumulation of β-amyloid also occurs in the brains of cognitively normal elderly people, in the absence of obvious neuronal dysfunction. We postulated that neuronal toxicity depends on the molecular composition, rather than the amount, of the soluble β-amyloid oligomers. Here we show that soluble β-amyloid aggregates that accumulate in Alzheimer disease are different from those of normal aging in regard to the composition as well as the aggregation and toxicity properties.

The mechanism of neurodegeneration caused by ␤-amyloid in Alzheimer disease is controversial. Neuronal toxicity is exerted mostly by various species of soluble ␤-amyloid oligomers that differ in their N-and C-terminal domains. However, abundant accumulation of ␤-amyloid also occurs in the brains of cognitively normal elderly people, in the absence of obvious neuronal dysfunction. We postulated that neuronal toxicity depends on the molecular composition, rather than the amount, of the soluble ␤-amyloid oligomers. Here we show that soluble ␤-amyloid aggregates that accumulate in Alzheimer disease are different from those of normal aging in regard to the composition as well as the aggregation and toxicity properties.
A series of evidence indicates that progressive cerebral accumulation of ␤-amyloid (A␤), 2 a proteolytic product of transmembrane protein APP, is the primary pathogenic event of Alzheimer disease (AD) (1). Recent clues indicate that small, soluble A␤ aggregates produce more severe synaptic dysfunction and neuronal damage than do A␤ polymers (2)(3)(4)(5). This behavior is common to all known pathogenic and nonpathogenic amyloidogenic peptides (6,7). Soluble A␤ is detectable early in the cerebral cortex of subjects at risk for AD pathology, several years before the formation and deposition of amyloid fibrils (8). Hence, the analysis of soluble A␤ in brain tissue allows the characterization of the toxic form of the peptide.
A strong argument against the amyloid hypothesis is the abundant and constant deposition of A␤ in the brains of elderly subjects, in the absence of signs of neuronal degeneration and dementia (9 -11). The reasons for the absence of pathogenic effect exerted by A␤ in normal aging are unknown. The issue has important therapeutic implications, because the major strategies to prevent and cure AD are focused on halting A␤ accumulation (12).
In brains from Alzheimer disease (AD) and Down syndrome patients, three major species of soluble A␤ have been identified by mass spectrometry: the full-length form, A␤1-42, which has a relative molecular mass of 4.5 kDa, and two N-terminal peptides truncated at residue 3 (A␤3-42) and residue 11 (A␤11-42) with relative molecular masses of 4.2 and 3.5 kDa, respectively (13,14). The 4.2-and 3.5-kDa bands are more prominent in familial AD carrying presenilin 1 mutations than in sporadic AD, suggesting that the ratio of soluble A␤ species may dictate the toxicity of the aggregates (15).
We predicted that the composition of soluble A␤ underlies the different effect exerted by the molecule in AD and in normal aging. To investigate this hypothesis, we studied the composition and properties of aggregation and toxicity as well as the damage produced on artificial membranes of soluble A␤, comparing these areas in sporadic AD and cognitively normal elderly subjects with abundant amyloid plaques in cerebral cortex.

MATERIALS AND METHODS
Tissues-We used frozen blocks and formalin-fixed sections of frontal cortex from 14 cases with late onset sporadic AD (mean age at death 80 Ϯ 8 years) (clinical history of disease; pathological diagnosis according to the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) criteria; post-mortem interval 8 h Ϯ 3) provided by the brain bank of Case Western Reserve University, Cleveland, OH, and from 11 cognitively normally aging (NA) elderly subjects (mean age at death 83.3 Ϯ 10 years; post-mortem interval 9.5 h Ϯ 4). The latter subjects had be tested neuropsychologically annually and agreed to be autopsied for research purposes (provided by the Alzheimer's Disease Research Center, University of Kentucky). Their neuropsychological scores were within the range of normal. In cerebral cortex abundant A␤ plaques were present, with absent or scarce neurofibrillary pathology. The amount of A␤ plaques, as shown with monoclonal antibody 4G8, was semiquantitatively evaluated in three nonadjacent sections of frontal cortex and was comparable with that of AD cases.
Immunoblot Analysis-Soluble A␤ was extracted from the watersoluble fraction of frontal cortex with a well established method described in detail previously (8). Briefly, frozen tissues were homogenized in 4 volumes of saline buffer (50 mM Tris, pH 7.6, 5 mM EDTA, 150 mM NaCl) containing protease inhibitors and centrifuged at 100,000 ϫ g for 1 h. Homogenization was also carried out with EDTA-free buffer. Soluble A␤ was immunoprecipitated from the supernatants (1 ml corresponding to 250 mg of tissue) adjusted to 1ϫ radioimmune precipitation assay buffer (150 mM NaCl, 1% Nonidet P-40, 0.5% cholic acid, 0.1% SDS, 50 mM Tris, pH 8, with protease inhibitors) with an antiserum raised against A␤1-40 synthetic peptide (RGP9) that selectively recognizes A␤ region 1-3, as demonstrated with the selective reactivity with A␤ starting at position 1 only ( Fig. 2A). Immunoprecipitation was carried out similarly with monoclonal antibody 4G8 (Signet Laboratories). To ascertain the state of aggregation of soluble A␤, immunoprecipitation was also performed following filtration of supernatants using Amicon Ultra centrifugal filter unit devices with low-binding Ultracel membranes with a 10-kDa cut-off (Millipore). Immunoprecipitated proteins were separated on Tris-Tricine 10 -18% gels and recognized by immunoblotting with monoclonal antibody 4G8, as well as with antibodies specific for N-terminal residues 1 (␣-N1), 3 (␣-py3), and 11 (␣-py11) (the last two cyclized to pyroglutamate at their N termini) (14). Antibodies specific for A␤42 and A␤40 were also used (IBL, Gunma, Japan). Brain-soluble fractions were also analyzed directly by immunoblotting following precipitation with methanol. Mixtures of synthetic peptides prepared as indicated below were immunoprecipitated with RGP9 anti-A␤ antiserum and analyzed by immunoblotting (Fig. 2B).
MALDI-TOF Analysis-Immunoprecipitated synthetic peptides and tissue samples were all treated under the same conditions. Dried agarose beads were resuspended in 50 l of 10% formic acid and agitated for 3 h at room temperature. 1 l of the supernatant was loaded directly onto the MALDI target using the dried-droplet technique and ␣-cyano-4hydroxycinnamic acid as matrix. Alternatively, 20 l of the supernatant was subjected to a single desalting/concentration step before mass spectrometric analysis over a ZipTipC 18 (Millipore Corp., Bedford, MA) and eluted in 1 l of 50% CH 3 CN ϩ 50% trifluoroacetic acid 0.2%. MALDI-TOF mass measurements were performed on a Voyager-DE STR (Applied Biosystems, Framingham, MA) operated in the reflectron mode. Spectra were calibrated externally using a standard peptide mixture.
Immunocytochemistry-Immunocytochemisty was performed on formalin-fixed, paraffin-embedded sections of frontal cortex. Adjacent 6-m-thick sections were processed according to the biotin-avidin method using antibodies specific for N-terminal residues 1 and 3 pyroglutamate (14). Sections were pretreated with 98% formic acid for 10 min at room temperature. The reaction was developed with 3,3Ј-diaminobenzidine as co-substrate. The number of reactive A␤ plaques was determined in 12 fields of the cortex spanning the entire cortical thick-FIGURE 1. Characterization of A␤ species in water-soluble fractions of cerebral cortex from NA and AD subjects. A, on immunoblots of immunoprecipitated peptides, the antibody 4G8 detected three bands of 4.5 kDa (B1), 4.2 kDa (B2), and 3.5 kDa (B3) in four representative AD and NA cases of all 25 examined. B, B1, B2, and B3 revealed with 4G8 are detected, respectively, with antibodies specific for A␤ residue 1 (␣-N1), A␤ residue 3 with pyroglutamate (␣-py3), and A␤ residue 11 with pyroglutamate (␣-py11). C, B1, B2, and more weakly B3 are recognized by the antibody specific for A␤42 in AD and NA cases. Only B1, corresponding to the full-length peptide, is labeled by the antibody specific for A␤40. D, water-soluble fractions directly analyzed by immunoblotting (detection with the antibody 4G8) show a pattern of reactivity identical to that obtained by immunoprecipitation. E, the quantification of B1 (white), B2 (black), and B3 (gray) reveals that cerebral watersoluble A␤ is composed predominantly of A␤1-42 (50%) in NA and A␤py3-42 (48%) in AD. Data are expressed as mean values Ϯ S.D.; n ϭ 14 for AD and n ϭ 11 for NA. All data are from triplicate experiments.

FIGURE 2.
A, on immunoblots of A␤ synthetic peptides, the antiserum RGP9 selectively recognizes full-length A␤ 1-42 but not A␤py3-42 and A␤py11-42. Accordingly, on immunoblots of brain-extracted water-soluble A␤, the antiserum RGP9 recognizes only B1, which corresponds to the full-length A␤, but not B2 and B3, in which the N-terminal truncated species migrate, which are recognized by monoclonal antibody 4G8. B, immunoblots of full-length and N-terminal truncated A␤42 synthetic peptides mixed in equal amount and kept for 24 h at 37°C, allowing their aggregation. All three A␤ peptides are immunoprecipitated by the antiserum RGP9. C, soluble A␤ peptides extracted from AD brains are undetectable by immunoblotting with the antibody 4G8 if previously filtered through a 10-kDa cut-off membrane. OCTOBER 7, 2005 • VOLUME 280 • NUMBER 40 ness in three AD and three NA cases using an ocular grid of 0.135 mm 2 at a final magnification of ϫ100.
Atomic Force Microscopy-Aggregation was initiated by incubation of the A␤ mixtures corresponding to AD and NA patterns at a concentration of 10 M in distilled water at room temperature. For AFM analysis, 20 l-aliquots of the sample were withdrawn at various times dur-ing the aggregation reaction, deposited on freshly cleaved mica, and dried under mild vacuum for 30 min. AFM images (amplitude data) were acquired in tapping mode using a Dimension 3000 microscope (Digital Instruments) equipped with a "G" scanning head (maximum scan size 100 m). Single beam, uncoated, silicon cantilevers were used (type TESP, Nanosensors, and RTESP, Nanodevices). The aggregate sizes were obtained by measuring the aggregate height in cross-section and the corresponding height in tapping mode AFM images.
Thioflavin T Binding Assay-One hundred l of AD and NA mixtures, 10 M, were diluted to 1 M in water, and thioflavin T 6 M was added to the solution. Fluorescence was measured using a PerkinElmer LS-5 luminescence spectrometer with excitation and emission at 440 m and 480 nm, respectively, with 5 nm bandwidth. Each sample and standard was done in duplicate.
Membrane Permeability Assay-Vesicle leakage induced by A␤ peptides was evaluated by means of the release of calcein (Sigma) as described previously (16). A␤ aggregates, obtained at different incubation times, were added at a final concentration of 1.5 M to a phospholipid vesicles suspension (lipid concentration 0.05 mM). To help protein insertion in the membrane, the sample was bath-sonicated for 15 min. The protein aggregation conditions were the same as those used for FIGURE 3. Low molecular weight aggregates of A␤ species immunoprecipitated with the antiserum RGP9 from water-soluble fractions of cerebral cortex of NA and AD cases. In addition to the monomeric forms, indicated as B1, B2, and B3, several bands migrating from 10.6 to 6 kDa are differently labeled by monoclonal antibody 4G8 and by antiserum ␣-N1, specific for full-length A␤, and by antiserum ␣-py3, specific for A␤ starting with pyroglutamate at position 3. The reactivity above 10.6 kDa is nonspecific. Three times more plaques are recognized by the antibody against A␤ residues 1 (␣-N1) than by the antibody specific for residue 3 (␣-py3) in the frontal cortex of normal elderly cases (A and B). The opposite ratio of plaque numbers is detected in AD cases (C and D). Magnification, ϫ80.

␤-Amyloid in Aging and in Alzheimer Disease
AFM. The samples were excited at 490 nm, and fluorescence emission was measured at 520 nm using an Aminco-Bowman spectrofluorimeter.
Cell Viability Assay-7-day SKNBE cells differentiated with retinoic acid were plated at a concentration of 5 ϫ 10 4 cells/well in 96-well plates in 100 l of media. The AD-and NA-A␤ mixtures were aged at a concentration of 10 M in phosphate-buffered saline at room temperature for 24 h, conditions that favor the formation of stable oligomeric aggregates (21). Peptides preparations were dissolved in culture media at a concentration of 1 M to obtain the highest toxic effect (4) and were added to the cells for 24 -48 h in serum-free conditions. Staurosporine (0.1 M) was used as positive control for apoptosis (data not shown).
MTT reagent was reconstituted in phosphate-buffered saline to 5 mg/ml as described by the manufacturer's protocol (Sigma). The solubilization solution was 10% SDS in 0.01 M HCl. 10 l of MTT was added to each well and incubated for 3 h at 37°C. 100 l of solubilization solution was added to each well and incubated overnight at 37°C. The absorbance of the samples was measured at 595 nm (Microplate reader, Bio-Rad).
Statistical Analysis-For statistical analysis, an unpaired Student's t test with Bonferroni's post-hoc test was used.

RESULTS
Composition of Soluble A␤ Immunoblot Analysis-We examined by immunoblotting the soluble A␤ species present in the cerebral cortex from subjects with sporadic AD (n ϭ 14) and from cognitively NAmatched controls with abundant amyloid plaques and scarce neurofibrillary pathology (n ϭ 11). As expected (8,13,17), following immunoprecipitation with RGP9 antiserum and detection with monoclonal antibody 4G8, soluble A␤ resolved into three bands of 4.5 kDa (B1), 4.2 kDa (B2), and 3.5 kDa (B3) in all cases (Fig. 1A). B1 was identified as the full-length A␤ (B1) using an antibody specific for A␤ starting at position 1. The other two bands reacted with antibodies specific for A␤ pyroglutamylated at position 3 (B2) and A␤ pyroglutamylated at position 11 (B3) (Fig. 1B), indicating that B2 and B3 are mostly (see mass spectrometry) composed of modified N-terminal truncated species, as reported previously (13,15). The antibody specific for A␤42 recognized all three bands in NA and AD cases, whereas only B1 was labeled by the antibody specific for A␤40 (Fig. 1C). The intensity of A␤40 reactivity was equal in NA and AD cases. This experiment indicated that most soluble A␤ belongs to the A␤42 form and that the A␤40 species is represented only by the full-length peptide (see mass spectrometry), which likely derives from the parenchymal amyloid angiopathy (18,19) present in both NA and AD cases.
An identical pattern of A␤ species was observed in immunoblots of brain-soluble fractions either immunoprecipitated with monoclonal antibody 4G8 or analyzed directly following protein precipitation with methanol (Fig. 1D). The results were not influenced by the presence or not of EDTA in the homogenizing buffer.
In its original state, soluble A␤ was present as small aggregates of all A␤ species, which partially disaggregated under reducing conditions of PAGE. This fact was demonstrated by two experiments: (a) the absence of A␤ in soluble fractions filtered through 10-kDa cut-off membranes (Fig. 2C); (b) the co-precipitation of N-terminal truncated A␤ species together with the full-length forms (Fig. 1A) upon immunoprecipitation with the antiserum RGP9, which recognized only the first three A␤ residues ( Fig. 2A). The state of aggregation of soluble A␤ was confirmed by the presence of several SDS-insoluble oligomers ranging from 6 to 10.6 kDa (Fig. 3). The variable reactivity of the A␤ oligomers with antibodies specific for different A␤ regions likely depends on changes of the epitopes exposure of A␤ species within the aggregates. All of the described bands did not appear when membranes were incubated with the preimmune serum (polyclonal antibodies ␣-N1 and ␣-py3). A nonspecific reactivity was instead observed above 11 kDa.
When reactivity with the monoclonal antibody 4G8 was considered, B2 was significantly more prominent (accounting for 48% of total soluble A␤) in AD than in NA brains (p Ͻ 0.001), whereas B1 prevailed in NA cases where it represented 50% of the total (p Ͻ 0.001) (Fig. 1E). The overrepresentation of B2 or of B1 was observed in each AD and NA case, respectively. An identical ratio among the three bands was observed following detection with the antibodies specific for N termini 1, py3, and py11, indicating that B1, B2, and B3 correspond to the three above  OCTOBER 7, 2005 • VOLUME 280 • NUMBER 40 mentioned peptides. Moreover, when A␤1-42, A␤py3-42, and A␤py11-42 synthetic peptides were mixed in equal amounts and analyzed by immunoblotting with monoclonal antibody 4G8, we observed a reactivity that corresponded perfectly to the relative percentage of each peptide (Fig. 2B). The A␤1-x/A␤py3-x ratio was apparently maintained also in the amyloid plaques deposited in the frontal cortex of AD and NA subjects. In NA cases, the antibody to A␤ 1 recognized 3.3-fold more A␤ plaques than the antibody to A␤py3 (mean number of A␤ plaques in 12 fields of the frontal cortex. A␤ 1 185 Ϯ 42; A␤py3 56 Ϯ 16) (Fig. 4, A  and B), whereas the opposite ratio was observed in AD cases (A␤ 1 75 Ϯ 12; A␤py3 205 Ϯ 38) (Fig. 4, C and D).

␤-Amyloid in Aging and in Alzheimer Disease
Mass Spectrometry-MALDI-TOF mass spectrometry confirmed the presence of A␤1-42, A␤py3-42, and A␤py11-42 as major peptides, as well as the presence of A␤1-40 and other N-terminal truncated peptides ending at residue 42 (A␤2-42; A␤3-42; A␤4 -42) (Fig. 5, A  and B). We speculate that the latter truncated A␤ species migrate with B2, the 4.2-kDa band, even if we are not able to confirm it without antibodies specific for the various N termini. However, neither A␤1-40 nor the non-pyroglutamylated N-terminal truncated species were consistently detectable in all of the AD and NA cases examined (Fig. 5, A  and B). The ratio among different A␤ species observed by Western blotting was not evident by MALDI-TOF analysis, where the signals of A␤py3-42 and A␤py11-42 are significantly lower than that expected. This is due to a specific suppression of the signal derived from the N-truncated A␤ species, as shown in Fig. 5C and as demonstrated previously with other peptides (20).
Aggregation of Soluble A␤-We examined whether the different representations of soluble A␤ species resulted in different A␤ aggregates.
We applied the aggregation protocol that favors the formation of A␤ stable oligomers with a low concentration of short protofibrils (21), because oligomerization is required for A␤ neurotoxicity (2,5). A mix of the three synthetic A␤ peptides in the ratios corresponding to those detected in NA and in AD were aged at room temperature for 24 -48 h at a concentration of 10 M, and the state of aggregation was observed by AFM. At 24 h, A␤ peptides formed globular structures of 6.4 Ϯ 0.3 nm in diameter, the earliest detectable form of A␤ aggregation (22,23). Although these aggregates were present in both mixtures, they were clearly more numerous in the AD mixture (Fig. 6A). After 48 h of incubation the AD mixture displayed 12-nm-thick protofibrils, resulting from the assembly of thin (3-4 nm) subunits, which were absent in the NA mixture (Fig. 6B).
To measure the aggregation properties, we analyzed the two pools of synthetic peptides using a thioflavin T fluorimetric assay. The AD mixture revealed a higher rate of aggregation than the NA mixture, and the difference became statistically significant after 48 h (Fig. 6C, *, p Ͻ 0.05). This result was further confirmed by sedimentation assay (data not shown).
Toxicity of Soluble A␤-The effect of soluble A␤ pools on neurons viability was assessed by incubating neuroblastoma cells with the two different mixtures of synthetic peptides at 1 M in culture media. We used as the parameter of cell viability the MTT assay, which has been shown to be a sensitive indicator of A␤-mediated toxicity (22). After 24 h of treatment the AD mixture produced a 30% decrease of cell survival statistically different from that of untreated cells (**, p Ͻ 0.01) in comparison with a 20% decrease induced by the NA mixture (nonsignificative) (Fig. 6D). After 48 h of incubation, the AD and NA mix-

␤-Amyloid in Aging and in Alzheimer Disease
tures, respectively, determined a 50 and 30% decrease of cell survival, and the difference of the toxic effect of the two types of A␤ aggregates was significant (Fig. 6D, *, p Ͻ 0.05). The scrambled peptide A␤42-1 used as a negative control had no effect on cell survival (Fig. 6D).
Effect of A␤ Species on Membrane Permeability-Because cytotoxicity of A␤ oligomeric species has been ascribed to the ability of A␤ to generate membrane pores (24), we analyzed the permeabilization of liposome membranes induced by early aggregates of A␤1-42 and A␤py3-42, the predominant peptides of soluble A␤ in NA and AD. We measured calcein release from unilamellar vesicles made of neutral and negatively charged phospholipids as an index of membrane permeability, as described previously (16). After 12 h of aggregation, A␤py3-42 caused a 23% increase of membrane permeability, which instead was only slightly altered by A␤1-42 (Fig. 7). The opposite conditions were observed with a longer time of incubation (24 h) of the peptides (Fig. 7), suggesting that A␤py3-42 more quickly reaches the state of aggregation that produces the highest membrane damage.

DISCUSSION
Our findings show that the soluble A␤ aggregates present in AD consistently and significantly differ in composition from the aggregates associated with NA and exhibit a higher neurotoxicity, which can be correlated with the predominance of the N-terminal truncated species over the full-length form. Among all N-terminal truncated peptides, A␤py3-42 is the prominent form, as we showed by comparing the reactivity of its specific antibody with that obtained with the monoclonal antibody 4G8 that recognizes all A␤ species (Fig. 1B). The prevalence of A␤py3-42 over the other N-terminal truncated species that probably migrate with the 4.2-kDa band (B2) (A␤2-42, A␤4 -42) is confirmed by mass spectrometry analysis, which consistently revealed only A␤py3-42 in all cases examined. The high relative amount of A␤py3-42 on the total A␤ load was reported previously by Harigaya et al. (25) using a specific sandwich ELISA. The highly pathogenic effect of A␤py3-42 is supported by the finding that the A␤py3-42 early aggregates alter the membrane permeability, suggesting that they form pores in the membrane as it has been proposed for other amyloidogenic peptides (26). Thus, depending on the relative representation of N-terminal truncated species, A␤ aggregates may be associated with very severe degeneration, as in the case of mutant presenilin 1 (15), or may exert a lower toxic effect, as in NA.
The A␤py3-42 species might be generated from APP through an alternative ␤-secretase cleavage or they can be produced from the fulllength 1-42 form by extracellular aminopeptidases and modified by glutaminyl cyclase to generate pyroglutamate (27). ␤-Amyloid-cleaving enzyme 1 (BACE1) is responsible of cleavages at position 1 and 11 of A␤ (28), and A␤ fragments starting with residue 3 were never reported when analysis of APP processing was carried out in vitro (29). However, in a double presenilin 1/APP mutant mice the A␤py3-42 form was detected (30), suggesting that in vivo BACE1 or another still unknown endoprotease might produce the cleavage at position 3. Following cyclization of N-terminal glutamate, A␤ may acquire partial resistance to most of the extracytoplasmic aminopeptidases (31)(32)(33), with ensuing accumulation of A␤py3-42 in AD brains. Moreover, the proteolytic cleavage of A␤ N-terminal cyclized fragments requires neprilysin (34 -36), a specific metallopeptidase that is reduced in AD (37). Therefore, proteolysis of A␤ at its N terminus, by limiting the rate of A␤ catabolism and enhancing its seeding capacity and toxicity, may play a critical role in the pathogenesis of AD. Therapeutic strategies in AD should particularly target A␤ species truncated at the N terminus. FIGURE 7. Calcein release induced in 1:1 molar ratio phosphatidylcholine/phosphatidylserine liposome membranes by A␤ aggregates monitored as a function of protein aggregation time. Early A␤py3-42 (OE) and A␤1-42 (Ⅺ) aggregates produced after 12 h of incubation cause a 23 and 5% increase, respectively, in membrane permeability. The difference in calcein release induced by the two peptides is less after 24 h of incubation. The data are expressed as a percentage of the maximum fluorescence, determined from the ratio (F Ϫ F 0 )/(F max Ϫ F 0 ), where F 0 is the fluorescence intensity before protein addition, F is the fluorescence measured during the release experiment, and F max is the maximum fluorescence, determined by adding 0.5% (w/v) sodium cholate at the end of the experiment to disrupt the vesicles and obtain complete content release.