Endogenous Proteins Controlling Amyloid β-Peptide Polymerization

We report that certain plasma proteins, at physiological concentrations, are potent inhibitors of amyloid β-peptide (Aβ) polymerization. These proteins are also present in cerebrospinal fluid, but at low concentrations having little or no effect on Aβ. Thirteen proteins representing more than 90% of the protein content in plasma and cerebrospinal fluid were studied. Quantitatively, albumin was the most important protein, representing 60% of the total amyloid inhibitory activity, followed by α1-antitrypsin and immunoglobulins A and G. Albumin suppressed amyloid formation by binding to the oligomeric or polymeric Aβ, blocking a further addition of peptide. This effect was also observed when the incorporation of labeled Aβ into genuine β-amyloid in tissue section was studied. The Aβ and the anti-diabetic drug tolbutamide apparently bind to the same site on albumin. Tolbutamide displaces Aβ from albumin, increasing its free concentration and enhancing amyloid formation. The present results suggest that several endogenous proteins are negative regulators of amyloid formation. Plasma contains at least 300 times more amyloid inhibitory activity than cerebrospinal fluid. These findings may provide one explanation as to why β-amyloid deposits are not found in peripheral tissues but are only found in the central nervous system. Moreover, the data suggest that some drugs that display an affinity for albumin may enhance β-amyloid formation and promote the development of Alzheimer’s disease.

Alzheimer's disease (AD) 1 is associated with the accumulation of a specific form of amyloid in the brain parenchyma and in meningocerebral blood vessels (1)(2)(3). The primary components of the amyloid are polymers of a short peptide derived through proteolytic processing of a ubiquitous transmembrane protein (4,5) termed the ␤-amyloid precursor protein. The amyloid ␤-peptide is usually referred to as the A␤. It is present in two principal variants (2,6), one that contains 40 amino acid residues (A␤ , and one C-terminally extended variant that contains 42 amino acid residues (A␤  ). The longer variant has been suggested to be of major importance in the pathogenesis of AD because it has a greater tendency to form amyloid fibrils in vitro and possibly also in vivo (7)(8)(9). Certain mutations associated with familial AD lead to an increased secretion of the 42-amino acid form (10) and an enhanced accumulation of amyloid.
␤-Amyloid displays several important features that distinguish it from other types of amyloid. (i) The peptide forming the amyloid deposits is present at very low concentrations in the circulation. This is in contrast to peripheral amyloid disorders in which the amyloid proteins are present at high concentrations. Examples of such non-central nervous system amyloid proteins include serum amyloid A, myeloma protein, and transthyretin (11). (ii) The levels of A␤ are not higher and the peptide is not structurally different (except in extremely rare cases of familial AD) in individuals with the disease than in healthy controls (for a review, see Ref. 3). (iii) It is well known that most and possibly all nucleated cells in the body produce the A␤ (12,13); however, for unknown reasons, ␤-amyloid is only deposited in the central nervous system.
In the present study, we aimed at investigating why ␤-amyloid exclusively is formed in the central nervous system. Previous work has demonstrated that some plasma proteins and lipoproteins bind A␤ and serve as carrier proteins (14,15). Protein binding is a general mechanisms for the transport of endogenous substances such as hormones and lipids as well as clinically used drugs (16). Generally, it is only the non-proteinbound fraction of the substances that is biologically active. We therefore hypothesized that only the free fraction of A␤ can take part in the polymerization process generating amyloid fibrils. Hence, A␤-carrier proteins may have an important role in preventing amyloid formation by increasing the bound fraction.
The bulk of large proteins do not penetrate the blood-brain barrier efficiently. Thus, the levels of soluble proteins in the central nervous system are much lower than those in peripheral tissues. It has been estimated that the ratio between protein content in the CSF and plasma is approximately 0.004 (17). However, in contrast to large proteins, the A␤ levels are higher in CSF than in plasma (18,19), which probably reflects a higher rate of secretion from neuronal cells than from other cell types. Overall, this suggests that a smaller fraction of the A␤ is protein-bound in the central nervous system than in the periphery.
With this background, we decided to investigate whether plasma and CSF proteins can indeed inhibit ␤-amyloidogenesis. We established in vitro assays allowing quantitative and qualitative studies of amyloid formation in the presence of several different proteins. The proteins studied here represent more than 90% of the protein content in plasma and CSF (20 -25). Many drugs bind to plasma proteins, which can lead to interactions with severe consequences (16). If some drugs bind to the same site on plasma protein molecules as A␤, it may lead to increased levels of free A␤ and enhanced amyloid formation. Therefore, we decided to also address this possibility experimentally.
Analysis of A␤ Polymerization-96-well plates (Maxisorp; Nunc) were coated with peptide by incubating them with a solution of A␤  or A␤ 1-40 (2.5 M) in Tris-buffered saline (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, and NaN 3 ). Solution (100 l) was added to each well, and the plates were incubated at 37°C with shaking for 48 h. The peptide solution was then flicked off. Staining with a solution of Congo red (20 M) in Tris-buffered saline showed that the polymeric peptide had bound to the wells (data not shown). After removal of the peptide solution, the plates were placed upside down on absorbing paper and allowed to dry. Coated plates were stored at Ϫ20°C in a desiccator. On the day of experiment, the plates were blocked by the addition of 300 l of PBS containing 0.05% Tween 20 (PBS-T) and 1% bovine serum albumin/well for 2 h at room temperature. The plates were then washed with PBS-Tween (0.05% Tween 20), and the fluid was flicked off. Biotin-A␤ 1-40 or biotin-A␤ 1-42 was dissolved in Me 2 SO and diluted with Trisbuffered saline with NaN 3 (0.05%). Unless stated otherwise, the final concentration of the labeled peptide was 20 nM. The plates were incubated overnight at 37°C with agitation. Nonbound peptide was removed by washing the plates three times with PBS-T (300 l/well). Streptavidin-peroxidase was diluted with PBS-T and 1% BSA and added to the plates (150 l/well). After incubation (2 h at room temperature), the solution was flicked off, and the plates were washed four times with PBS-T. Tetramethyl-benzidin was used as chromogenic substrate for the peroxidase. After termination of the reaction with sulfuric acid (0.33 M, final concentration), absorbance was measured at 455 nm with a SpectraMAX 250 96-well plate reader. Nonspecific binding was defined as the binding of biotin-A␤ to wells that had not been coated with A␤. There was a linear relationship between peroxidase activity and the amount of peptide bound (data not shown). Nonspecific binding was, on average, approximately 15% of total binding (data not shown). We also studied the incorporation of 125 I-A␤ into tissue sections of human AD brain using the method of Maggio et al. (29). In experiments with the prion protein-derived peptide PrP 106 -126 (26), similar methodology was used, but with two exceptions. First, the Maxisorp plates were coated with a solution of 10 M peptide for 14 days. Second, incubation with N-terminally biotinylated PrP 106 -126 was performed for 4 h. The method used was validated by several means. (i) biotinylated-A␤ 1-42 or A␤ 1-40 was incubated at a high concentration (10 M) for 72 h at 37°C and then examined by electron microscopy. Both peptides were capable of forming fibrils that were indistinguishable from nonbiotinylated controls. (ii) Both biotinylated peptides required the Maxisorp plates to be coated with amyloid fibrils in order for them to bind. When the plates were coated with truncated variants of A␤ (A␤ 12-28 , A␤ [35][36][37][38][39][40][41][42] , A␤ 10 -20 , A ␤1-16 , or A␤ [25][26][27][28][29][30][31][32][33][34][35], no significant binding over that obtained with noncoated control wells was observed. (iii) We also studied the incorporation of biotin-A␤ 1-40 into preformed A␤ 1-42 fibrils. A low concentration of fibrils (corresponding to 20 nM monomeric peptide) was incubated with equimolar amounts of biotin A␤ 1-40 . After overnight incubation and centrifugation, the material was stained with antibiotin Ig labeled with colloidal gold and negatively stained with uranyl acetate. Using this protocol, gold-labeled amyloid fibrils were observed, demonstrating that biotin-A␤ 1-40 could bind to the preformed fibers. The controls used were biotin-A␤ 1-40 or preformed A␤ 1-42 fibrils alone. In these control experiments, biotin-A␤ 1-40 did not produce any detectable fibrils, whereas the A␤ 1-42 fibrils were not labeled by gold.
Electron Microscopy-Negative staining was performed by adsorption of a 5-l aliquot of the sample to a carbon-coated 200-mesh copper grid for 60 s. Staining was done by adding 10 l of 2% uranyl acetate directly to the adsorbed sample droplet for 2 min and air drying after the removal of excess liquid with filter paper. Specimens were examined in a JEOL 1210 electron microscope operated at 100 kV. Digitized micrographs were recorded with a slow scan charge-coupled device camera (Gatan; model 679). Data acquisition with the slow scan chargecoupled device camera and processing of the digitized images were controlled by a Macintosh PowerPC 8500 using DigitalMicrograph software from Gatan. Images were printed on a Thermoprinter Phaser 440 (Tektronix). Magnification calibration was performed as described previously (27) using negatively stained catalase crystals.
Surface Plasmon Resonance Spectroscopy-Interactions between albumin and monomeric/polymeric A␤ were measured using a BiaCore 2000 instrument (BiaCore AB, Uppsala, Sweden) essentially as described previously (28). Briefly, monomeric biotin-A␤ 1-40 at a concentration of 115 nM was attached to a sensor cell to which streptavidin had been coupled. The peptide was stored in Me 2 SO and diluted in running buffer (see below) immediately before it was coupled to the chip. Under these conditions, no evidence suggesting that the peptide polymerized could be obtained (data not shown). Polymeric peptide was attached via a monoclonal antibody against amino acid residues 2-8 of A␤ (BAP-1A). Unless otherwise stated, the running buffer used contained 10 mM Hepes, pH 7.5, 150 mM NaCl, and 0.05% P20 detergent. After each experiment, the cells were washed with ethanolamine (1 M, pH 8.5) until the sensor signal remained stable in contact with running buffer.

RESULTS
Albumin-mediated Inhibition of A␤ Polymerization-In the first set of experiments, we investigated whether albumin, which is quantitatively the most important A␤-binding protein (14) and also the most abundant protein in plasma and CSF, could interfere with the incorporation of biotin-A␤ 1-40 into amyloid fibrils. The test system used is based on the finding that A␤ monomers bind with high affinity to preformed polymers of A␤ (29). We immobilized A␤ 1-42 polymers as seeds (7) in 96-well plates and measured the incorporation of soluble biotin-A␤  in the presence of human serum albumin (HSA) or BSA. In Fig. 1A, the effects of different concentrations of HSA and BSA on biotin-A␤ 1-40 incorporation are shown. The highest concentrations of albumin used corresponded approximately to the plasma levels of a healthy human adult (21). Both HSA and BSA had the capacity to completely inhibit the incorporation of biotin-A␤ 1-40 into immobilized A␤ polymers with apparent IC 50 values of 10 and 12 M, respectively. The effects of HSA on the polymerization of A␤ 1-42 in this system were also studied. Here, nonlabeled A␤ 1-40 or A␤ 1-42 was immobilized in the Maxisorp plates as described under "Experimental Procedures." Biotinylated A␤ 1-40 or A␤ 1-42 was then allowed to bind the immobilized peptide in the presence of various concentrations of HSA. The IC 50 values of HSA on the inhibition of A␤ 1-40 or A␤ 1-42 binding were essentially identical (data not shown), suggesting that HSA is indeed capable of inhibiting polymerization of the two major forms of the A␤.
These findings were confirmed in a second set of experiments in which the incorporation of 125 I-A␤ 1-42 into brain tissue sections from an individual with AD was measured. As seen in Fig.  1B, the radiolabeled peptide bound to the amyloid deposits in the tissue, as demonstrated previously (29). In the presence of 227 M HSA, binding was heavily reduced (Fig. 1C). Measurement of the incorporated radioactivity using a phosphorimager showed that overall binding (binding to amyloid deposits in the tissue and background together) had been reduced with 55% by the addition of HSA.
A␤ incubated at high concentrations also rapidly polymerizes in the absence of preformed polymers, but through primary nucleation (30,31). Therefore, in other experiments, we studied the effects of HSA on soluble A␤ 1-40 and A␤  in the absence of seeds. As seen in Fig. 2, A and C, both peptides formed fibrils when incubated for 24 h at 37°C at a concentration of 20 M. This concentration is approximately 6,000 times higher than that in CSF (17,29,39). When incubated in the presence of 227 M HSA, the polymerization of A␤ 1-40 into amyloid fibrils was completely inhibited (Fig. 2B). Under the same conditions, A␤ 1-42 only formed occasional fibrils (Fig. 2D). Moreover, spherical structures of 10 -30 nm in diameter were also detected frequently.
HSA Inhibits the Polymerization of a Prion-derived Peptide-PrP 106 -126 represents the central core of the prion protein (for a recent review, see Ref. 26) and spontaneously forms amyloidlike fibrils. Similar to the A␤, prion protein can form amyloid deposits in the CNS and cause neurodegeneration. We there-fore decided to study whether albumin can also prevent polymerization of this peptide. As seen in Fig. 3, HSA dose-dependently inhibited the binding of biotinylated PrP 106 -126 to immobilized homologous peptide. The IC 50 value for HSA in this system was approximately 100 M, 10 times higher than that for A␤. This concentration represents less then one-sixth of the albumin concentration in blood but is more than 30 times higher than the albumin concentration in CSF. Hence, these After an overnight incubation at 37°C with agitation, the reaction was stopped and processed as described under "Experimental Procedures." The biotin-A␤ 1-40 incorporation in the absence of albumin is equal to 100%. The experiments were performed in quadruplicate. The experiment was repeated three times with essentially identical results. Data are indicated as the mean Ϯ S.E. The effect HSA on the incorporation of 125 I-A␤ 1-42 into genuine amyloid deposits in human AD brain tissue was also studied. In B and C, the effect of buffer alone or buffer containing HSA (227 M), respectively, is demonstrated. findings demonstrate that HSA displays a certain degree of specificity for ␤-amyloid.
Characterization of the Mode of Action for Albumin on A␤ Polymerization-Surface plasmon resonance spectroscopy allows protein-protein interaction studies in real time (28), and this methodology was therefore used to study how BSA and HSA interact with monomeric and polymeric A␤. Preformed A␤ 1-42 fibrils were immobilized to the sensor chip as described under "Experimental Procedures." A solution (25 M) of BSA (Fig. 4A) or HSA (Fig. 4B) was then allowed to flow through the cell. The protein bound avidly to the polymers, indicating that albumin indeed has an affinity for the polymeric peptide. In parallel flow cells, monomeric biotin-A␤ 1-40 was immobilized using streptavidin. In this case, no binding was observed with either BSA (Fig. 4A) or HSA (Fig. 4B). When the experiment was repeated using nonbiotinylated A␤ 1-40 that was immobilized with a monoclonal antibody, essentially identical results were obtained (data not shown).
Drug-enhanced A␤ Polymerization-It is well known that several clinically important drugs bind to albumin with various affinities. We speculated that some of these substances may bind to the same site(s) on the albumin molecule as A␤ and may therefore be able to displace the peptide from its binding site(s). This may lead to increased levels of free A␤ and the enhancement of amyloid formation. We therefore screened a number of albumin ligands with regard to their effects on biotin-A␤ 1-40 incorporation into preformed amyloid polymers in the presence or absence of 100 M HSA. It was found that tolbutamide, at concentrations corresponding to therapeutic levels (17), enhanced biotin-A␤ 1-40 incorporation in the presence but not in the absence of HSA (Fig. 5). This strongly suggests that tolbutamide is capable of interfering with A␤-albumin binding and indirectly stimulating amyloid fibril formation.
Endogenous Regulators of ␤-Amyloidogenesis-In these experiments, we investigated the effects of various plasma/CSF proteins on A␤ polymerization (Table I). The proteins listed in Table I represent more than 90% of the protein content in plasma and CSF. The concentrations tested covered the levels in plasma and CSF for all but two proteins, IgM and ␣ 1antichymotrypsin. The highest concentrations used were 0.55 and 1.0 M, respectively, which were lower than their plasma concentrations but higher than their CSF concentrations (see Table I). Seven of the 13 tested proteins had very little or no effect (i.e. the IC 50 was higher than the plasma concentrations). Of the remaining six proteins, three had IC 50 values in the range of 10 -30 M, and three had IC 50 values below 10 M. Albumin, ␣ 1 -antitrypsin, IgG, and IgA had IC 50 values that were substantially below their plasma concentrations, which strongly suggests that these proteins may be potent inhibitors of ␤-amyloidogenesis in vivo.
1 unit of inhibitory activity was defined as the number of mol/liter of protein required to inhibit polymerization by 50%  under the conditions specified under "Experimental Procedures." When taking the plasma concentration of the studied proteins into consideration, it is possible to estimate how much inhibitory activity each protein contributes (Fig. 6). Albumin is probably the most important regulator of ␤-amyloidogenesis in plasma. Although ␣ 1 -antitrypsin has an IC 50 value eight times lower than that of albumin (1.25 and 10 M, respectively), the concentration of the former is substantially lower (25.3 and 644 M, respectively). Therefore, despite its higher efficacy, it probably plays a less important role in the regulation of A␤ polymerization.
Cerebrospinal fluid contains essentially the same proteins as plasma, but the concentrations are considerably lower (see Table I). None of the tested proteins are present in the CSF in a concentration equal to or higher than its IC 50 value, which was obtained in the amyloid formation assay (see Table I). When comparing the total amount of inhibitory activity in plasma and CSF, we found that CSF contains only about 0.3% of that seen in plasma (Fig. 6).

DISCUSSION
Plasma and CSF proteins with affinity to A␤ serve as carriers for the peptide (14,15). This study and previous studies (32,33) have demonstrated that A␤-binding proteins in plasma and CSF may also have a function in regulating ␤-amyloid formation. The most abundant plasma protein, albumin, is present in a concentration more than 60-fold higher than its IC 50 in the A␤ polymerization assay used here (see Table I). Albumin is the most abundant protein in CSF, but it is present at a concentration below its IC 50 value at which only a partial inhibition of A␤ polymerization is obtained (see Fig. 1). Plasma also contains significant levels of other proteins, such as ␣ 1 -antitrypsin, IgG, and IgA, that are capable of inhibiting A␤ polymerization. The other studied proteins capable of inhibiting polymerization are also present in CSF in concentrations substantially below their IC 50 values (see Table I). These results point to a dramatic difference between plasma and CSF: the former contains large quantities of inhibitory proteins, whereas the latter contains small quantities of inhibitory proteins. For unknown reasons, ␤-amyloid deposits are not formed outside the central nervous system (34). The present results suggest that the high concentrations of inhibitory proteins in plasma prevent the formation of ␤-amyloid in peripheral tissues, but the low levels in CSF do not block ␤-amyloid formation in the central nervous system. This conclusion is also supported by previous experimental data ( 35), showing that CSF only partially inhibits the formation of thioflavin-binding amyloid from synthetic A␤ 1-40 .
Pathologically reduced levels of albumin might promote ␤-amyloidosis and possibly also AD. In clinical studies, it was observed that anti-inflammatory drugs may have beneficial effects on AD (36). Levels of albumin are often reduced in association with inflammation (25) and, hence, the antiamyloidogenic activity in plasma and CSF is also reduced. However, even heavily reduced plasma levels of albumin are probably still sufficiently high to prevent amyloid formation in peripheral tissues. It may be different in the central nervous system. Because albumin (and other inhibitory proteins) is present in low concentrations having limited effects on amyloid formation FIG. 6. A␤ polymerization inhibitory activity of individual proteins and the total content of inhibitory activity in plasma and CSF of these proteins. A, 1 inhibitory unit is defined as the number of mol/liter of the indicated protein required for a 50% inhibition of A␤ polymerization under the conditions defined under "Experimental Procedures" and the Fig. 1 legend. B, the total inhibitory activity in plasma and CSF contributed by the studied proteins specified in Table I.   TABLE I IC 50 values for the inhibition of biotin-A␤ 1-40 incorporation into immobilized A␤

polymers
The two columns to the right indicate plasma and CSF concentrations of the studied proteins in healthy adults (21)(22)(23)(24)41 (35), even small reductions in albumin levels in association with inflammation may lead to increased amyloid formation. The structural background as to why A␤ binds albumin and other proteins is not known. However, it is reasonable to assume that hydrophobic interactions are involved. It was surprising that monomeric A␤ did not display binding to albumin when studied by surface plasmon resonance spectroscopy, considering the findings of Biere et al. (14) showing that soluble A␤ binds albumin and lipoproteins. One explanation may be that A␤ molecules rapidly form small, soluble, oligomers with an affinity to albumin (37,38).
Tolbutamide is a drug used to regulate blood glucose levels in diabetes mellitus. It also displays a high affinity for albumin. As a result, its clinical use is often associated with interactions with other drugs when the compounds compete for the same binding site on the albumin molecule (17). Here, we found that tolbutamide, at concentrations corresponding to therapeutic levels, enhanced amyloid formation in the presence but not in the absence of HSA. A reasonable explanation is that tolbutamide and A␤ bind to the same site on albumin. Tolbutamide may therefore displace A␤ from albumin and generate higher free A␤ fractions that can participate in amyloid formation. Drugs that can penetrate into the central nervous system, bind to the A␤ site(s) on albumin, and increase the free fraction of the peptide may thus be capable of enhancing amyloid formation in vivo.
Mutations affecting proteins capable of binding A␤ may promote the development of AD (39,40). It is therefore possible that mutations affecting the proteins studied here may also have an impact on the development of AD through a similar mechanism.
In conclusion, the present data suggest a novel and possibly important physiological role for albumin and other plasma/CSF proteins in controlling amyloidogenesis in the central nervous system and possibly also in peripheral tissues. The data also suggest that drugs with certain pharmacokinetic properties may be capable of enhancing amyloidogenesis. Moreover, the reduced levels of albumin seen in association with inflammatory reactions may provide an opportunity for the A␤ to polymerize and thereby more easily form amyloid in the central nervous system.