Water-soluble Abeta (N-40, N-42) oligomers in normal and Alzheimer disease brains.

Ultracentrifugation and graded molecular sieving, as well as a sensitive sandwich enzyme-linked immunosorbent assay were used to isolate and quantitate the amounts of water-soluble oligomers of beta amyloid (Abeta) peptides N-40 and N-42 in cerebral cortex of normal and Alzheimer disease (AD) brains. AD brains contained 6-fold more water-soluble Abeta (wsAbeta) than control brains. The majority of water-soluble peptides in most AD cases was A beta N-42, representing 12 times the amount found in control brains. The wsAbeta was present in the form of monomers and oligomers ranging from less than 10 kDa to greater than 100 kDa. The amount of wsAbeta N-42 in AD brains is about 50 times greater than the level of soluble Abeta N-42 found in the CSF of AD patients. This disparity may be due to the rapid association of wsAbeta N-42 into fibrillar deposits and/or to the integrity of the anatomical barriers which separate the two extracellular spaces. In this paper, we consider soluble any form of Abeta which has not yet polymerized into its insoluble, filamentous form. This includes both the newly synthesized forms of Abeta and those peptides which may be loosely attached to insoluble filaments but which can, nevertheless, still be considered soluble. It has been previously shown that, once it has aggregated into its filamentous form, the Abeta peptides are resistant to disaggregation and degradation by a number of denaturing agents and aqueous buffers containing proteolytic enzymes. Therefore, it is likely that the water-soluble Abeta peptides we quantified are precursors to its insoluble, filamentous form. Consequently, reducing the levels of soluble Abeta in AD brains could have profound effects on AD pathophysiology.

The pathology of Alzheimer disease (AD) 1 is characterized by the deposition of the ␤ amyloid (A␤) peptides in the extracellular space of the brain parenchyma and in the walls of the cerebral blood vessels. These peptides are derived by proteo-lytic degradation of a larger molecule, the A␤ precursor protein, whose gene is localized on chromosome 21 (reviewed in Ref. 1). From a chemical point of view, A␤ is represented by polypeptide chains 40 to 42 amino acid residues long with a M r of 4,500 (2)(3)(4). Another class of amyloid associated with diffuse deposits contains the amino acid sequence of residues 17-42 (3 kDa), corresponding to the C-terminal sequence of the 1-42 A␤ form (5). All of these A␤ peptides are very insoluble and resistant to proteolytic degradation. These physicochemical properties primarily result from the C-terminal amino acid sequence of 12 to 14 hydrophobic residues which play an important role in the initial aggregation and insolubility of A␤. The amphipathic N-terminal region of the A␤ molecule, consisting of 28 amino acids, appears to be necessary for the polymerization of this peptide into cytotoxic 10-nm filaments. In addition, the Nterminal portion of the A␤ molecule is probably responsible for the binding to ancillary molecules such as apolipoproteins (6), glycosaminoglycans (7), ␣1-antichymotrypsin (8), the complement protein C1q (9), and metal ions (10).
According to their morphology, the A␤ peptide deposits can be classified as either fibrillar or amorphous. The fibrillar forms are usually observed at the center of and around the neuritic plaques where they are surrounded by dystrophic neurites and reactive glial cells. They are also observed in the parenchymal and leptomeningeal blood vessels, where they cause destruction of the vascular walls and myocyte degeneration (11,12). The amorphous forms of A␤ are mainly localized to diffuse deposits scattered throughout the brain's gray matter and apparently do not cause any observable pathological alterations. Interestingly, numerous cell lines in culture produce small quantities of soluble A␤ 1-40, which suggests a physiological role for the shorter A␤ peptide (13)(14)(15). The A␤ 1-42, on the other hand, seems to be the major component of the parenchymal and, to a lesser extent, the vascular deposits of AD (2)(3)(4)16). The abundance of this peptide suggests that it plays a relevant role in the pathophysiology of AD.
Recently, small amounts of apparently soluble low molecular mass A␤ peptides (3.0, 3.7, and 4.0 kDa) were isolated by immunoprecipitation from AD brain homogenates, which were not detected in normal brains (17,18). Utilizing the more sensitive sandwich ELISA soluble A␤ peptides have been identified in normal and AD brains (19). In the present paper, we have further investigated the A␤ water-soluble fractions based on a more rigorous centrifugal separation of AD and control brain homogenates free of detergents or chaotropic agents. In addition, we quantified the water-soluble oligomeric A␤ by sandwich ELISA in fractions obtained by ultracentrifugation and graded membrane filtration.

EXPERIMENTAL PROCEDURES
Human Tissue-Brains were obtained from 8 demented patients who died of AD (post-mortem delay time 4 -13 h). The left hemispheres were utilized for histological and morphometric analysis. The right hemispheres were frozen at Ϫ75°C until the moment of utilization. The histopathological analyses indicated that the brains of all 8 patients contained a large number of neurofibrillary tangles and neuritic plaques, fulfilling the diagnostic criteria of AD as determined by the Consortium to Establish a Registry for Alzheimer Disease and the NIH Neuropathology Panel for AD (20,21). The amyloid burden, that is, the percentage of A␤ area within a total field of 6 mm 2 , was estimated by immunostaining sections of the second frontal gyrus with the mouse monoclonal 4G8 antibody (against A␤ [17][18][19][20][21][22][23][24], then quantitating the amount of A␤ in these sections using computer-assisted image analysis (NIH Image). In addition, the APO E genotype was determined. Genomic DNA, isolated from unfixed cerebellum, was used for APO E genotype determinations using a previously described polymerase chain reaction method (22). The material used in this study amounted to 20 g of cerebral cortex, free of leptomeninges, from each frontal lobe. The 4 control cases utilized in this investigation had no clinical history of dementia or any other neurological involvement (Table I).
Methodology-The protocol of investigation employed in the present paper is schematically presented in Fig. 1. Each cerebral cortex was finely minced and gently, but thoroughly, homogenized in a glass Ten Broeck tissue grinder in the presence of 130 ml of 20 mM Tris-HCl, pH 8.5, 3 mM EDTA, 500 g/liter leupeptin, 700 g/liter pepstatin, 350 mg/liter phenylmethylsulfonyl fluoride, 100 mg/liter 1,10-phenanthroline, 100 mg/liter benzamidine, 50 mg/liter gentamicin sulfate, and 250 g/liter amphotericin B. Brain homogenates were spun at 135,000 ϫ g (Beckman SW28 rotor) for 2 h at 4°C. The viscosity of the recovered supernatant hampered passage of the specimen through the 100-kDa cutoff Centricon filter. Therefore, the sample was further centrifuged at 220,000 ϫ g (Beckman 41 Ti rotor) for 2 h at 4°C. The resulting small pellet (P220) was set aside for A␤ analysis and the supernatant (S220) was divided into two parts. The first portion was submitted to a series of membrane ultrafiltration steps with M r cutoffs at 100, 30, and 10 kDa (Centriprep, Amicon Inc., Beverly, MA). These generated three retentates designated as: Ͼ100 kDa, 100-30 kDa, and 30-10 kDa, and a final filtrate Ͻ10 kDa. The presence of A␤ in all of the above fractions was subsequently investigated by ELISA as described below. The second portion was centrifuged at 435,000 ϫ g for 90 min at 4°C (Beckman TLA 100.2 rotor) and separated into a pellet (P435) and a supernatant (S435). Two marker proteins of known molecular size, hemoglobin (64 kDa) and myoglobin (17 kDa), were independently utilized to establish the proportion of soluble peptides that can be pelleted at 435,000 ϫ g. Spectrophotometric measurements indicated that, under these conditions, 92% of the hemoglobin and 73% of the myoglobin were pelleted.
The pellet retrieved after the 135,000 ϫ g centrifugation of the initial homogenate, containing the insoluble A␤, was dissolved in 10% SDS, Tris-HCl, pH 8. After standing 6 h at room temperature, the lysate was centrifuged at 135,000 ϫ g for 2 h (Beckman SW28 rotor) at 20°C. The ensuing pellet (P135) was washed twice with 20 mM Tris-HCl, pH 8, dissolved in 5 ml of 80% glass distilled formic acid, and centrifuged at 275,000 ϫ g for 30 min at 4°C. This step permitted the separation of insoluble lipofuscin from the solubilized A␤. The supernatant (S275) was concentrated by vacuum centrifugation, dialyzed (1,000 Da cutoff) against Tris buffer, and analyzed for A␤ by ELISA. The supernatant derived from the SDS-135,000 ϫ g centrifugation (S135) was spun at 275,000 ϫ g (Beckman 41 Ti rotor) for 2 h at 4°C. The resulting pellet was suspended in 100 mM Tris-HCl, pH 8, containing 2 mM CaCl 2 and digested with 10 g/ml DNase I (Worthington). Following centrifugation at 275,000 ϫ g for 2 h, the pellet (P275) was dissolved in 80% formic acid and submitted to size exclusion HPLC in a TSK-3000 SW column (0.7 ϫ 60 cm, Altex). The chromatography was developed in 80% formic acid as described previously (5). In addition, the S275 formic acid supernatant derived from P135 was also investigated by HPLC as described above.
ELISA-A sandwich ELISA (23) was used to quantitate specifically either A␤ peptides 1-42 (N-42), or all A␤ peptides of 1-28 amino acids or longer. All incubations were performed at room temperature. A mouse monoclonal antibody, 266, that recognizes amino acids 13-28 of A␤, was bound to the wells of microtiter plates and used as the capture antibody for the sandwich ELISA. Samples or solutions of A␤ of known concentrations were applied to the wells in phosphate-buffered saline containing Triton X-100 detergent (dilution buffer) for 1 h. The unbound antigen was then aspirated and the plates washed three times with Tris-buffered saline containing Tween 20. For the detection of A␤ 1-28 or longer, a biotinylated, mouse monoclonal antibody that recognizes the first 16 amino acids of A␤ (6C6) in dilution buffer was added to the wells, the plates were incubated for 1 h, and the wells washed as before. Streptavidin alkaline phosphatase diluted 1:1000 in dilution buffer was added to the wells and it was incubated for another hour, followed by three washes. The fluorescent substrate methyl umbellipheryl phosphate was added to the wells and fluorescence read in a fluorometer equipped with an excitation and emission filter of 365 and 450 nm, respectively. Standard curves for this assay were derived from values obtained from known concentrations of A␤ 1-40 ranging from 0.125 to 5 ng/ml.
The quantitation of A␤ 1-42 in samples was done with the capture antibody 266 and the rabbit polyclonal antibody 277.2 developed to A␤ residues 39 -43 (23). An alkaline phosphatase labeled-affinity purified F(abЈ) 2 fragment of donkey anti-rabbit IgG (HϩL), absorbed against mouse IgG, was added to the wells and incubated for 1 h. After aspirating and washing the plates, a chemiluminescent substrate, AMPPD, and an enhancer, emerald green, were added and the chemiluminescence was detected using a chemiluminometer. Standard curves for the assay were generated from samples of known concentrations (0.125 to 2.0 ng/ml) of A␤ 1-42. This assay fails to detect A␤ 1-40 or shorter A␤ species and has less than 5% cross-reactivity with A␤ 1-41 and A␤ 1-43. The values obtained from the above assays were subtracted from each other to determine the amount of A␤ containing amino acids 1-28 through 1-40 (N-40).

RESULTS AND DISCUSSION
Previous studies from our laboratory and others have focused on the chemical analysis of the fibrillar A␤ present in neuritic plaques and in the walls of the blood vessels. Due to the extreme insolubility of these peptides, their purification required the utilization of a variety of detergents and/or strong chaotropic and denaturing agents. While these methods have been effective in isolating the A␤ from the amyloid deposits, they have prevented the characterization of the soluble A␤ which is the precursor of the insoluble, filamentous form. This soluble pool of A␤ is operationally defined as A␤ extracted from brain tissue under mild condition (i.e. homogenated in 20 mM Tris-HCl, pH 8.5) and is hereafter referred to as water-soluble A␤ (wsA␤). The isolation and analysis of these water-soluble low molecular weight forms, in the absence of the aforemen- tioned agents, may help to determine the degree to which they participate in the pathophysiology of AD.
In the present study, we separated the pools of water soluble and insoluble A␤ from AD and control brains (Fig. 1). The detection and quantitation of A␤ in our study, using sensitive sandwich ELISA, revealed that wsA␤ N-40 and N-42 are present in the aqueous extractable fraction of AD and control brains ( Fig. 2A and Table II). The level of insoluble A␤ in AD brains is, on the average, 100 times that of control brains, whereas the amount of wsA␤ in AD brains is about 6 times that detected in control brains (Fig. 2, A and B, and Table II). An interesting observation is that there is about 50 times the quantity of wsA␤ N-42 found in AD brain cortex as that detected by Vigo-Pelfrey et al. 2 in the CSF of AD patients (A␤N-42 ϭ 0.4 ng/ml) using the same ELISA technique. A possible explanation for the discrepancy between these two pools may be that in AD brain parenchyma, this peptide is rapidly incorporated into the insoluble 10-nm filaments characteristic of this disease, drastically decreasing its opportunity of being translocated into the CSF. An alternative explanation for the huge difference found could be due to the integrity of the physical barriers which exist between the extracellular space of the brain parenchyma and the subarachnoidal space. These barriers are represented by the glia limitans at the surface of the cerebral cortex and by the outer wall of the perivascular spaces which is derived from the leptomeninges (24). In support of this possibility are the presence of wisps of amyloid frequently observed in the molecular layer of the cerebral cortex, just beneath the glia limitans and also in adjoining perivascular spaces.
Analysis of the relative amounts of wsA␤ N-40 and N-42 revealed that the predominant form of wsA␤ in control brains is N-40. On the average, the N-40:N-42 ratio of wsA␤ in control brains is 75:25, whereas in AD brains this ratio is 48:52 (Table  III). The amount of wsA␤ N-40 found in 5 of the 8 AD brains studied (cases 2, 5, 6, 7, and 8) fell within the same range as that detected in the control group. Pathology reports of the three remaining AD brains (cases 1, 3, and 4), which contained higher quantities of wsA␤ N-40, indicated a significantly higher level of cerebrovascular amyloidosis than the other five cases. This elevated level of A␤ N-40 in cerebrovascular amyloidosis is in agreement with data recently reported by Younkin's group (25). Significantly, since the average amount of wsA␤ N-42 in the AD brain (19.0 ng/g of cortex) is higher than in the control group (1.6 ng/g of cortex), and the amount of N-40 was comparable in the aforementioned 5 cases, it is reasonable to conclude that it is the water-soluble N-42 peptide which is the main contributor to the fibrillar deposits of A␤ in AD.

FIG. 1. Flow chart showing the purification procedures utilized to separate the water-soluble and insoluble fractions of A␤.
The sandwich ELISA was performed in all boxed fractions. P and S refer to pellets and supernatants, respectively. Brain homogenates were spun at 135,000 g for 2 h instead of the accepted criteria of 100,000 ϫ g for 1 h. This higher centrifugation yielded a better separation between insoluble A␤ filaments and the more soluble monomeric and oligomeric forms of this peptide. The water-soluble material was further spun at 220,000 ϫ g and the resulting supernatant subjected to ultrafiltration and ultracentrifugation. The water-insoluble pellet, obtained from the initial brain homogenate, was lysed in 10% SDS and centrifuged at 135,000 ϫ g. The resulting pellet and supernatant (P135 and S135, respectively) were further analyzed. The former contained large quantities of insoluble fibrillar amyloid A␤ 1-40, 1-42 (5). The latter fraction (see large shaded box) also carried the 3-kDa A␤ 17-42 peptide which was purified by HPLC as described previously (5). The A␤ 17-42 is the most insoluble of all A␤ species. This peptide only disaggregates into smaller micelle-like particles in SDS or into monomers in the presence of strong denaturing agents such as 80% formic acid (5)

FIG. 2. Histograms depicting the amounts of water-soluble and insoluble A␤ in AD and control brains as quantitated by sandwich ELISA.
A, the distribution of wsA␤ recovered in the S220 and P220 fractions. The amount of amyloid in the control cases (A-D) ranged from 5.3 to 10.2 ng/g and in the AD cases (1-8) ranged from 21.0 to 89.1 ng/g. A␤ in the P220 control group is not shown in the histograms since it amounts to only 0.1 ng/g on average. B, total amount of insoluble A␤ (P135) in the control and AD cases. The absence of value in AD case 4 is due to its total investment in the chromatographic study. The quantities of amyloid in the control cases (A-D) ranged from 5.1 to 20.8 ng/g and in the AD cases (1-8) ranged from 377.3 to 3000.0 ng/g. C, ultrafiltration values of A␤. Centricon membranes partitioned S220 into an apparent continuous distribution of wsA␤ which could range from monomeric and dimeric (Ͻ10 kDa) to polymeric (Ͼ100 kDa). The amounts of wsA␤ N-40 and N-42 in each of the filtrates and final retentate is shown in Table III. strated what appears to be a continuous distribution of monomeric and oligomeric wsA␤ ranging in size from less than 10 kDa to more than 100 kDa (Fig. 2C). On the average, the proportion of wsA␤ N-42 in the AD brains decreased from about 70% in the Ͻ10 kDa (monomeric and dimeric) fraction to about 40% in the 100-30 kDa and Ͼ100 kDa (octameric and larger) fraction (Table III). This may provide evidence that the incorporation of wsA␤ N-42 into insoluble filaments occurs only after it has oligomerized into octameric and larger molecules. However, it is unclear if the apparent decrease in the proportion of N-42 in the higher molecular weight fractions is actually due to a reduction in the level of the peptide or to the inability of the ELISA to measure accurately the level of N-42 due to steric hindrance caused by antibody binding to the aggregated peptides. The results of our ultracentrifugation studies also lend support to the presence of low M r wsA␤ in S220 fraction, since 50% of these peptides were sedimented at 435,000 ϫ g. Under this experimental condition, 73% of myoglobin (17 kDa) is pelleted. Hence, it could be assumed that a large quantity of wsA␤ in S220 is smaller than 17 kDa.
In a previous article, Tamaoka et al. (19) recovered soluble A␤ from successive Tris buffer extractions of brain homogenates in which they utilized a 1:2 ratio of cortical tissue to buffer. They suggested that the apparently "newly" solubilized A␤ was derived from a "reversible" depolymerization of this peptide from insoluble amyloid deposits. In our experience, the high density of the homogenate and the lower centrifugal forces (100,000 ϫ g, 15 min) do not permit extraction and separation of the soluble A␤ fraction. It is also possible that the magnitude of the mechanical forces necessary to disperse such a dense homogenate may account for the shearing of insoluble A␤ into their soluble fraction.
In this study we considered the aqueous-extracted A␤ to be soluble if it failed to form a pellet after centrifugation at 220,000 ϫ g for 2 h. A pathway must exist by which newly synthesized water-soluble A␤ polymerizes into its fibrillar, insoluble form. It is yet to be determined at what stage in this polymerization process insolubility is conferred upon the A␤ peptide, but there is ample evidence to conclude that once it achieves its fibrillar form it is extremely insoluble in a variety of denaturing agents and is non-degradable by proteolytic enzymes. Both of these phenomena have been independently documented by both Selkoe's group (26) and by this laboratory (27), and very recently communicated by Harigaya et al. (18) as the aqueous insoluble A␤ fraction or SDS-formic acid soluble A␤ component. It is likely that during the polymerization process there exists an intermediate, oligomeric A␤ which is still soluble but may be loosely associated to the insoluble filaments prior to its complete incorporation. Both soluble forms of this peptide could potentially be precursors to the insoluble, filamentous A␤ prominent in AD.
We attempted in this study to determine the extent to which the wsA␤ quantitated by ELISA could actually be A␤ peptides which had dissociated from insoluble filaments during purification procedures. For this reason, we conducted correlation analyses between the observed amyloid burden and wsA␤, amyloid burden and insoluble A␤, and between water-soluble and insoluble A␤, to estimate the extent to which this may have occurred. If the wsA␤ quantitated is that which has dissociated from plaques, a strong correlation between ELISA-measured wsA␤ and A␤ deposition (as measured by either A␤ burden or total insoluble A␤) would be expected. Although there is a weak correlation between wsA␤ and A␤ burden (Fig. 3A), the absence of correlation between insoluble A␤ and A␤ burden (Fig. 3B) would suggest that the former correlation may lack relevance. The most likely explanation for this is that amyloid is unevenly distributed from region to region in the cerebral cortex. It is also possible that a weakly positive correlation does, in fact, exist between wsA␤ and amyloid burden. However, since amyloid formation is a dynamic process, the size of this watersoluble pool will likely fluctuate and the existence of a weakly positive correlation may or may not have pathophysiological significance. Fig. 3C shows that no correlation exists between wsA␤ and insoluble A␤. These data suggest that the majority of the ELISA-measured wsA␤ may be newly synthesized A␤ and a smaller percentage consists of loosely attached peptides which have not yet been incorporated into the insoluble A␤ filaments. Interestingly, when cases 3 and 8, possibly perceived as outliers, are removed from the analysis, the correlation between insoluble A␤ and A␤ burden becomes significant ( ϭ 0.923; p Ͻ 0.05). However, removal of these two cases results in no significant correlation between wsA␤ and A␤ burden ( ϭ 0.528, p ϭ 0.361). This lends support that most of the A␤ present in this soluble pool is that which has been newly synthesized. It is not yet known if all of the A␤ in this watersoluble pool, whether newly synthesized or dissociated from insoluble filaments, is ultimately incorporated into insoluble filaments or if some percentage of this pool is subject to degradation by proteolytic enzymes. The identification of a water-soluble pool of A␤ N-42 allows us to contemplate its possible role in AD pathology, independent of its potential contribution to the formation of neuritic plaques and vascular deposits. There is a discordance between the amount of fibrillar amyloid deposits and the extent of neuronal pathology in AD brains. The location of the neuritic plaques appears separate and distinct from the sites of greatest neuronal loss and synaptic pathology. These observations have caused many to question the relevance of the fibrillar A␤ deposits to the dementing process of AD. In our opinion, a possible element which could reconcile this seeming discordance is the existence of a potentially toxic, water-soluble pool of oligo-meric A␤ 1-42. Our results show that, on the average, the oligomeric wsA␤ N-42 pool is uniquely elevated (12-fold) in AD compared to that of normal brains. The M r of these oligomers would favor their diffusion throughout the brain parenchyma, however, this soluble pool would likely go undetected using conventional immunochemical techniques. Therefore, given our observation that a pool of water-soluble A␤ is present in the cerebral cortex of control and AD brains, the potential toxic effects of oligomeric A␤ upon nerve cells calls for further investigation.