Amyloid beta-peptide is transported on lipoproteins and albumin in human plasma.

The amyloid β-peptide (Aβ) is the major constituent of neuritic plaques in Alzheimer's disease and occurs as a soluble 40-42-residue peptide in cerebrospinal fluid and blood of both normal and AD subjects. It is unclear whether Aβ, once it is secreted by cells, remains free in biological fluids or is associated with other proteins and thus transported and metabolized with them. Such knowledge of the normal fate of Aβ is a prerequisite for understanding the changes that may lead to the pathological aggregation of soluble Aβ in vivo, the possible influence of certain extracellular proteins, particularly apolipoprotein E, on plaque formation, and the pharmacology of putative Aβ-lowering drugs. To address the question of Aβ distribution in human biological fluids, we incubated fresh human plasma from 38 subjects with physiological concentrations (0.5-0.7 nM) of radioiodinated Aβ1-40 and seven plasma samples with Aβ1-42. Lipoproteins and lipid-free proteins were separated and analyzed for bound iodinated Aβ1-40. We found that up to 5% of Aβ added to plasma is bound to selected lipoproteins: very low density, low density, and high density, but not lipoprotein(a). The large majority (≈89%), however, is bound to albumin, and very little Aβ is free. Aβ distribution in plasma was not significantly influenced by apolipoprotein E genotype. We conclude that Aβ is normally bound to and transported by albumin and specific lipoproteins in human plasma under physiological conditions.

Alzheimer's disease (AD), 1 a common, progressive neurodegenerative disorder, is characterized by the presence of extracellular amyloid deposits in both the cerebral neuropil and meningocerebral blood vessels. The major component of these deposits, the 40 -42 amino acid amyloid ␤-peptide (A␤) (1), also occurs in the cerebrospinal fluid (CSF) and plasma of normal subjects and AD patients (2)(3)(4)(5). A␤ is constitutively secreted into the extracellular fluid of a variety of primary cultured cells, including neurons, astrocytes, fibroblasts, endothelial cells, and smooth muscle cells (2, 6 -9). It shows N-and Cterminal heterogeneity in vitro and in vivo (10 -13). Intensive effort has been directed at understanding the cellular generation of A␤ (for review, see Ref. 14) as well as its extracellular aggregation into insoluble amyloid fibrils (15)(16)(17)(18)(19), but very little is known about the fate of soluble A␤ between its secretion and the process of amyloid formation. For example, it is largely unclear how A␤ is transported and metabolized once it is released from the cell and to what extent A␤ in human body fluids is free or associated with other proteins. It is particularly important to address these questions in view of the fact that only a small percentage of all AD cases (5,10,20,21) have been shown to have an actual increase in A␤ production. The basis for excessive ␤-amyloid build up in the majority of AD cases is currently unknown and could involve in part a change in the normal transport or catabolism of A␤.
The major genetic risk factor associated with AD is the ⑀4 polymorphism of the apolipoprotein E (ApoE) gene (22,23). Inheritance of one or two ApoE4 alleles is associated with a highly significant increase in the number and density of cerebral and cerebrovascular A␤ deposits (24 -29). However, we recently showed that expression of ApoE4 had no influence on A␤ production in cultured cells (30). Several groups reported ApoE-A␤ binding under in vitro conditions (31)(32)(33)(34), and one study suggested A␤ binding to HDL 3 particles (35). Since ApoE is a constituent of lipoproteins present in blood and CSF and is consequently involved in lipid transport, it is important to determine whether A␤ secreted into physiological fluids becomes associated with and transported by lipoproteins, and if so, by which lipoproteins and whether ApoE influences this in an allele-specific manner.
The source of the A␤ that is deposited extracellularly in the brain and its microvasculature in AD is not known, so that the distribution of A␤ in blood, CSF, and brain interstitial fluid needs to be understood. The protein constituents of blood and CSF are well known (36,37), and a comparison of their compositions indicates that CSF is in part an ultrafiltrate of plasma. At least two considerations argue for initially evaluating A␤ distribution in blood, the number and volume of samples that can be obtained and the ready availability of plasma from many ApoE-genotyped subjects. Here, we have employed traditional methods for lipoprotein isolation (density gradients) and protein characterization (nondenaturing gels and Western blots) to show that exogenous A␤ 1-40 and A␤  , when incubated in fresh human plasma at physiological (nanomolar) concentrations, bind to certain plasma proteins. The large majority of the A␤ is bound to albumin and consequently found in fractions with a specific density Ͼ1.21 g/ml. A smaller but consistent portion of A␤ is bound to very low density lipoproteins (VLDL), low density lipoproteins (LDL), and high density lipoproteins (HDL). The amount of free exogenous A␤ is very small and variable. Our experiments define the distribution of A␤ in human plasma and analyze the effects of the different naturally occurring ApoE isoforms on this distribution.

EXPERIMENTAL PROCEDURES
Synthetic Peptides A␤ 1-40 and A␤ 1-42 , synthesized by fluoroenylmethoxycarbonyl chemistry, were purchased (QCB, Inc.) and stored either lyophilized or as stock solutions (10 Ϫ3 M) in 75% dimethyl sulfoxide, 25% hexafluoro-2propanol at Ϫ20°C to reduce A␤ aggregation during storage. Peptides were characterized by reverse phase HPLC (RP-HPLC), laser desorption mass spectrometry, and amino acid analysis (38) and gave satisfactory results in all cases.

Radioiodination
Peptides for radioiodination were purified to near homogeneity (98%) by RP-HPLC using a gradient of acetonitrile (MeCN) in 0.01 M trifluoroacetic acid and a C 18 column. Radioiodination was by the method of Maggio et al. (39). Briefly, A␤  was radiolabeled at the tyrosine residue at position 10 by oxidative iodination using Na 125 I and chloramine T. Peptide and unincorporated iodine were separated using reverse phase adsorption. The oxidized methionine residue at position 35 was reduced from the sulfoxide to the native thioether form using 2-mercaptoethanol. Radioiodinated A␤ peptides were purified to a specific activity of about 2000 Ci/mmol (one 125 I per molecule) by RP-HPLC. 125 I-A␤ was found by RP-HPLC to be essentially free of oxidation or degradation during these experiments and over months of storage, consistent with previous studies (39). The monoiodinated 125 I-A␤ (35.3% MeCN) was cleanly separated from the uniodinated (33.8% MeCN), diiodinated (38.8% MeCN), and oxidized (32.7% MeCN) forms of the peptide by RP-HPLC (39). Nonradioactive iodination of A␤ with 127 I and purification under identical conditions gave a product with m/z ϭ 4480 Ϯ 4 (M ϩ Na ϩ ; matrix-assisted laser desorption isonization time-of-flight mass spectrometry), consistent with the monoiodinated, reduced methionine form of the peptide. Acetonitrile was removed from stock-labeled peptide solutions by evaporation under a nitrogen stream. Before use, an aliquot of each peptide was electrophoresed on a 16% Tris-Tricine gel (Novex, Inc.) to search for degradative products or aggregates. All peptides employed showed neither gel-excluded high molecular weight material nor oligomers under these conditions. One batch A␤ 1-40 , however, showed a degradative product, while the other gave only a single, clean 4-kDa band, without any smaller products. Both peptides behaved essentially the same in our assays. The A␤ 1-42 also showed smaller products beside the 4-kDa band. Aliquots of Ϸ2.5⅐10 6 cpm were stored at Ϫ20°C so that one aliquot could be used per plasma sample and additional freeze/thaw cycles could be avoided. To exclude batch-specific effects, two different batches of iodinated A␤ 1-40 were employed in these experiments.

Plasma Samples
Blood samples from nonfasting subjects were obtained from two different hospitals (Brigham and Women's Hospital and Massachusetts General Hospital), and ApoE genotyping was performed as described previously (40,41). Blood from AD and control cases was collected into EDTA (final concentration, Ϸ0.15%). In general, on the day blood was drawn plasma was recovered by a Ϸ25-min centrifugation at 200 ϫ g (a 10-min centrifugation at 1700 ϫ g was shown to give the same results). One ml of fresh plasma was incubated with 2.2-2.5⅐10 6 cpm of iodinated A␤ (final concentration, Ϸ2.6 -3 ng/ml) for 60 -70 min at room temperature (25°C) and then separated on a NaBr gradient.

Continuous NaBr Gradient
The protocol used here is a modification of that of Kelley and Kruski (42). After incubation with radioiodinated A␤, the plasma was adjusted to a density of 1.31 g/ml, transferred to a polyallomer centrifugation tube, and carefully layered with 3, 3.8, 3, and 1.2 ml of NaBr solutions having densities of 1.210, 1.063, 1.019, and 1.006 g/ml, respectively. Separation was accomplished using a Beckman SW41 swinging bucket rotor run for 24 h at 37,000 rpm and 20°C in a Beckman L8-55M ultracentrifuge. Fractions (0.5 ml) were collected by hand from the top of the tube, and all 24 fractions/gradient were analyzed for counts/ min/l and for protein content (using a Bio-Rad colorimetric assay). The fractions were then stored at 4°C for future use. Fractions of an iden-tical NaBr gradient, run in parallel without a plasma sample, were collected the same way, and their densities were determined using an Anton Paar DMA35 density meter.

Calculation of Percent Counts/Min
Counts of all 24 fractions in a gradient were summed and set equal to 100%. The fractions containing VLDL, LDL, HDL, and LFP were defined in each gradient by its protein profile (see Fig. 1), the counts for the respective protein groups summed and the percentage of iodinated A␤ (% counts/min) calculated for each group.
Correction for HDL-Small amounts of albumin could be found starting with fraction 16 (detected by Western blotting); these do not significantly contribute to the protein readings but do influence the counts/ min profile, because the amount of A␤ bound per mg of albumin is dramatically higher than that bound per mg of HDL protein (see Table  II). Therefore, the counts found in the HDL fractions 16 -18 needed to be corrected for the overlapping albumin-bound A␤ signal before the percent counts/min bound to HDL was calculated. For this correction, we assumed that the counts/min associated with the second half of the HDL peak (fractions 16 -18) would follow the same counts/min/protein ratios as the counts/min associated with the first half of the HDL peak (fractions 13-15). Thus we calculated:

Nondenaturing Agarose Gels
NaBr gradient fractions were dialyzed against 0.15 M NaCl, 0.01% EDTA, pH 7.0, and concentrated using Centricon 30 (Amicon, Inc.) for lipoproteins or Centricon 10 for lipid-free plasma proteins. Usually, the one or two peak fractions of each lipoprotein class were combined, resulting in a 100%, Ϸ75%, and Ϸ40% representation of all VLDL, LDL, and HDL, respectively. Also, the concentration factor was not uniform among samples, so that these gels give only qualitative information. Amounts of 3 or 10 l were electrophoresed on nondenaturing agarose gels (which separate by charge, not size) and stained for proteins with Amido Black or for lipids with Fat Red, respectively. Nondenaturing gels, stains, and plasma protein standards were purchased from Ciba Corning, Inc., and used according to the manufacturer's instructions.
For another experiment, unlabeled A␤ 1-40 was incubated for 1 h at room temperature with purified human serum albumin (HSA; Calbiochem, Inc.) or LDL (Perimmune, Inc.) at a molar ratio of 1:1. Ten-l aliquots of these mixtures (each containing 2 g of A␤) were electrophoresed alongside corresponding amounts of A␤ 1-40 , HSA, and LDL.

RESULTS
To begin to characterize the protein distribution of A␤ in biological fluids, we concentrated on blood because of its ready availability, volume of fluid per sample and the ability to obtain a reasonable number of ApoE-genotyped samples to address the question of an ApoE allele-specific effect on A␤ distribution among plasma proteins. However, the experimental paradigm we describe here can easily be modified to perform CSF analyses, as well.
Plasma was collected from fresh whole blood by centrifugation and incubated with physiological amounts of radioiodi-nated A␤  for ϳ1 h at room temperature. Typically, 2.2-2.5⅐10 6 counts/min were used, resulting in a final concentration of 2.6 -3 ng/ml A␤  . Samples were then subjected to density gradient centrifugation (see "Experimental Procedures"). The continuous NaBr gradient protocol followed here allowed the separation of all plasma constituents in one step, avoiding the disadvantages of sequential density centrifugation (see "Discussion"). Twenty-four fractions were collected per gradient, and three pieces of information extracted from each single fraction: the density, the protein amount, and the amount of iodinated A␤ (detected as counts/min). Fig. 1 shows an example of the protein, density, and counts/min profiles through an entire gradient. VLDL as well as chylomicrons float on the top and are found in the first fraction, which has a density of Յ1.010 g/ml, as indicated at the top axis of Fig. 1. LDL, HDL, and LFP are found in this gradient in the density ranges of 1.021-1.046 g/ml, 1.064 -1.172 g/ml, and Ն1.208 g/ml, respectively. These values are consistent with values found in the literature for chylomicrons (Յ0.94 g/ml), VLDL (0.94 -1.006 g/ml), LDL (1.019 -1.063 g/ml), HDL (1.063-1.21 g/ml), and LFP (Ն1.21 g/ml) using sequential ultracentrifugation (43). Since the amount of the dietary chylomicrons is negligible and was not sufficient to be detected in our samples, we will refer in the following to the first fraction(s) as "VLDL" fraction, although small amounts of chylomicrons might be present.
In addition to establishing the protein and density profile throughout the gradient, we examined the distribution of the iodinated A␤ 1-40 among the different lipoproteins and lipid-free proteins by counting all fractions. In a total of 38 human plasma samples, we found 0.06 Ϯ 0.04% of all A␤ counts in the VLDL fractions, 0.6 Ϯ 0.3% in the LDL, 4.3 Ϯ 1.9% in the HDL, and 89 Ϯ 5% in the LFP fractions. (For details about % counts/ min calculation and HDL counts/min correction, see "Experimental Procedures".) To confirm that the counts found in the lipoprotein gradient fractions represent A␤ 1-40 actually bound to the lipoprotein particles and not just material not completely sedimented into the LFP fractions, we characterized these fractions with a second, independent method: nondenaturing agarose gel electrophoresis, which separates free A␤ and lipoproteins. This commercially available gel system enables one to detect the large, intact lipoproteins directly (43,44) and can be stained for either protein or lipid (see "Experimental Procedures"). The protein and lipid stains of the VLDL, LDL, and HDL fractions of a representative gradient are shown in Fig. 2, A and B. The lipoproteins separated on our gradient had the same electrophoretic mobility as purified lipoproteins run in the same gel sytem (data not shown) and as described in the literature (43). The pre-␤ lipoprotein shown in the HDL lane is Lp(a), a lipoprotein which was detectable only in some plasma samples and which was very recently established as a new, major risk factor for coronary heart disease (45). Fig. 2C shows the autoradiogram of Fig. 2B and demonstrates that A␤ 1-40 found in the lipoprotein fractions colocalized with and bound to VLDL, LDL, and HDL, with none detectable on Lp(a), all of them having electrophoretic mobilities different from free, lipoprotein-unassociated A␤ (see Fig. 3). There was no difference of A␤ 1-40 binding to HDL 2 versus HDL 3 when corrected for milligrams of protein (data not shown). The signals in Fig. 2 cannot be compared quantitatively (see "Experimental Procedures").
The large majority of A␤ counts in human plasma was found in the LFP fractions of the gradient. To address whether this non-lipid-associated A␤ is free or bound to specific proteins, we ran the LFP fractions on nondenaturing agarose gels alongside plasma protein standards. Fig. 3A shows the last four fractions of a typical NaBr gradient. All major plasma protein classes (albumin, ␣ 1 -, ␣ 2 -, ␤-, and ␥-globulins) (Fig. 3A, std. lane) were recovered from the gradient in the LFP fractions, and they showed the same mobility as the commercial standards. When autoradiography of the gel shown in Fig. 3A was performed, the vast majority of A␤ 1-40 clearly colocalized with albumin (Fig.  3B). To confirm this association, the same fractions were immunoblotted with an anti-albumin antibody, and the albumin signal was found to parallel the A␤ signal (Fig. 3C). No detectable A␤ signal was associated with other protein classes, even after much longer autoradiographic exposure. The representa- tive gel shown in Fig. 3 has nearly no free A␤. We electrophoretically examined the LFP fractions of a total of 11 gradients (i.e. 11 plasma samples); in only two of the 11 plasmas did we find more than 1% of the iodinated A␤ 1-40 to be free, i.e. unbound to albumin (one having 5-10% and one 25-30% of the total A␤). By examining different batches of synthetic A␤ 1-40 , both iodinated and not iodinated, we observed two possible electrophoretic positions for free A␤ in these gels that were not mutually exclusive, migrating below albumin and/or between the ␣ 1 -and ␣ 2 -globulins (as indicated by asterisks in Fig. 3B).
The A␤ binding to lipoproteins and albumin is not due to iodination of A␤, because noniodinated, synthetic A␤ 1-40 binds to HSA and LDL as well (Fig. 4). The A␤-albumin complex was also detectable on SDS-polyacrylamide gel electrophoresis (data not shown).
We also investigated the A␤ 1-42 distribution in plasma and found it similar if not identical to A␤ 1-40 . Fig. 5 shows the protein stain (Fig. 5A) and the lipid stain (Fig. 5B) of the LFP and lipoprotein fractions and the autoradiogram of both (Fig.  5C). Again, nearly all A␤ 1-42 is bound to VLDL, LDL, HDL, and albumin, none to Lp(a). In a total of seven human plasma samples, we found 0.4 Ϯ 0.2% of all A␤ 1-42 counts in the VLDL fraction, 1.4 Ϯ 0.4% in the LDL, 3.9 Ϯ 1.2% in the HDL, and 92 Ϯ 2% in the LFP fractions, which is in the same range as found for A␤  .
The analyses summarized above demonstrate that iodinated A␤ 1-40 incubated at physiological doses with fresh human plasma is bound to VLDL, LDL, HDL, and albumin. ApoE is known to be a constituent of chylomicrons, VLDL particles, and a subclass of HDL particles. To test the possibility of an alleledependent effect of ApoE on A␤ distribution among lipoproteins, we compared a total of 17 ApoE 3/3, nine ApoE 3/4, nine ApoE 4/4, and three ApoE 2/3 plasma samples. These samples were obtained from two different hospital-based clinics, included AD and control cases, and were examined using two different batches of iodinated A␤ 1-40 peptide to exclude any batch-specific effects. The percentage distribution of total iodinated A␤ 1-40 found in the indicated fractions of the 38 plasma samples is listed in Table I. We saw no difference in A␤ distribution among the four ApoE genotypes examined. To confirm this impression, we performed a specific analysis of matched pairs of plasma samples that were run under the closest experimental conditions possible (Table II). Each of two such comparisons contained four plasma samples, two with the ApoE 3/3 genotypes and two with either the 3/4 or the 4/4 genotypes. The plasmas examined in each comparison were obtained on the same day from one clinic, prepared simultaneously, incubated with the same batch of iodinated A␤  , and run on the same density gradient centrifugation. The two sets shown in Table II represent the two different peptide batches used in this study. This closer comparison again revealed no significant ApoE allele-specific differences in the A␤ 1-40 distribution among the major plasma protein classes (Table II). We corrected all %counts/min values for the total protein amount in each fraction and still saw no allele-specific differences (Table II). Because the protein amounts were, in general, closely similar in FIG. 3. Iodinated A␤ 1-40 is associated with albumin. A, the lipid free protein fractions, 21-24, of a typical plasma NaBr gradient (see Fig. 1) were separated on a nondenaturing agarose gel together with a plasma protein standard (std.) and stained for total protein with Amido Black. B, autoradiogram of the gel shown in A. The nomenclature for the conventional human plasma protein groups that can be separated by nondenaturing gel electrophoresis is given on the left. Asterisks indicate the positions found for free A␤. Inset C shows a Western blot of the same four density gradient fractions using an anti-albumin antibody.
FIG. 4. Unlabeled A␤ 1-40 is associated with albumin and lipoproteins. A␤ 1-40 was incubated with purified HSA or LDL and separated on a nondenaturing agarose gel. A␤ was detected by Western blot using the antibody 6E10. This antibody does not cross-react with HSA and LDL (lanes 4 and 5). The electrophoretic mobilities of conventional human plasma protein groups are given on the left.

FIG. 5. Iodinated A␤ 1-42 is associated with plasma lipoproteins and albumin.
Fractions of a NaBr density gradient containing VLDL, LDL, HDL, and LFP, respectively, in addition to a plasma protein standard (std.) and A␤ 1-42 (A␤) were separated on a nondenaturing agarose gel and stained for either protein (A) or lipid (B) (see "Experimental Procedures"). C shows the autoradiogram of A and B; the signals cannot be quantitatively compared directly (see "Experimental Procedures"). all plasmas examined, the summary data for all samples in the study listed in Table I were not significantly changed by the latter standardization (data not shown). However, small differences in the relative distribution of A␤ among individual samples (see Table II) may be the result of different amounts of lipoproteins or nonlipidated proteins found among human plasmas; these differences did not reach statistical significance. We also found no significant difference in A␤ 1-40 distribution among the plasma protein classes in the 24 AD and 14 control cases examined in this study. DISCUSSION The Alzheimer's disease A␤ is secreted constitutively by ␤-amyloid precursor protein-expressing cells into the extracellular fluid and is present in CSF and plasma, but how soluble A␤ is transported and metabolized in these extracellular fluids remains unclear. An important question in this regard is to what extent soluble A␤ in brain extracellular fluid and in bodily fluids in general is free or associated with proteins and lipoproteins. Under in vitro conditions A␤ 1-40 binds to a variety of proteins, among them the apolipoproteins E and J (31-34, 46, 47).
Here, we report the analysis of A␤ distribution in human biological fluids. To obtain reliable data about such proteinprotein interactions, one should examine fresh samples to obviate the vulnerability of the lipoproteins. We have found, for example, that lipoproteins in previously frozen plasma samples show alterations of A␤ binding. 2 In order to assess any effects of the three human ApoE isoforms, we focused our studies on ApoE-genotyped fresh human plasma. Although CSF would clearly be of interest as well, it is currently not possible to obtain a sufficient number of fresh, ApoE-genotyped CSF samples. CSF is in part an ultrafiltrate of plasma, so that many of the differences in the protein compositions of the two fluids are mainly quantitative. However, a major qualitative difference is in the lipoprotein patterns. CSF carries only high density li-poproteins, which seem to contain either ApoAI or ApoE (48,49).
Based on the results presented here, we conclude that iodinated A␤ 1-40 and A␤ 1-42 are bound to selected lipoproteins (VLDL, LDL, HDL, but not Lp(a)) and to albumin in fresh human plasma. Only a very small fraction of A␤ is free, i.e. not protein-associated. An earlier study (35) showed binding of biotinylated A␤ 1-40 to HDL 3 and very high density lipoproteins, but failed to show binding to VLDL and LDL. This is probably due to the much lower sensitivity of their method (biotinylated A␤ 1-40 and immunoprecipitation versus iodinated A␤ and counting of the whole fraction without further manipulation). They also could not detect any A␤ (protein-associated or free) in the LFP fractions, but again, the methods used are not directly comparable. The authors employed sequential ultracentrifugation for plasma fractionation requiring a total centrifugation time of 5.5 d, which means a roughly total processing time of at least 7 d versus 24 h for the continuous gradient used by us. To be able to detect A␤ in the LFP fractions, the authors needed to immunoprecipitate it, which might be problematic regarding the high amounts of human, native IgGs in this fraction, although the authors tried to remove them with protein G-Sepharose.
It is known that synthetic A␤ can show substantial variability in various in vitro assays, probably due to conformational changes in the peptide and therefore enhanced propensity to self-aggregate (50 -52). To account for A␤ 1-40 peptide batch differences, we included several quality controls, including checks for degradation, aggregation, and the electrophoretic migration of A␤ on nondenaturing gels. Two different peptide batches that proved to be reliable were employed in our experiments, enhancing confidence in the binding data obtained.
We found that the amount of iodinated A␤ bound to plasma lipoproteins is small but highly consistent. We did not detect an ApoE allele-dependent effect on A␤ distribution among the lipoprotein classes, suggesting that A␤ 1-40 is either not directly bound to in vivo lipidated ApoE or that there are no detectable differences in A␤ 1-40 binding among the different ApoE alleles   (32,33). In vitro binding studies using high doses of synthetic A␤ and ApoE proteins purified from plasma or from ApoE-transfected cell lines have led to varying conclusions about allele-specific ApoE-A␤ interaction (32,33,53). However, two recent studies could not show an allele-specific effect on ApoE-A␤ binding (34,47). Direct A␤ binding to ApoE is not necessarily a prerequisite for an ApoE allele-mediated effect on A␤ distribution and transport among lipoproteins. Lipoproteins are heterogeneous particles, the exact composition of which varies depending on the subject's genetic and dietary background. ApoE is principally found in chylomicrons, VLDL, ␤-VLDL, and a subclass of HDL particles (for review, see Ref. 22). However, the different ApoE alleles show preferential incorporation into different lipoproteins (54 -56) and have different receptor affinities, so that the compositions and, more importantly, the amounts and binding functions of lipoproteins are altered with a change in the ApoE alleles. ApoE4 carriers, for example, display elevated plasma cholesterol and LDL, whereas ApoE2 is associated with type III hyperlipoproteinemia characterized by decreased LDL-cholesterol concentrations and accumulation of ApoE-enriched ␤-VLDL (22,57). Thus, our finding that A␤ binds to plasma lipoproteins has the consequence that any change in the lipoprotein profile of a subject may have a subtle effect on A␤ distribution in plasma. In view of the fact that the ApoE4 allele is a dose-dependent risk factor for AD, not a dominant mutation, and that the average age of onset for AD is after age 70, one would not expect to find a dramatic effect of different ApoE genotypes on the binding and transport of A␤, but rather a more subtle one that might alter A␤ distribution in extracellular fluids (plasma, CSF, and brain interstitual fluid) and result in its gradual accumulation over decades in ApoE4 carriers to reach disease-precipitating status. Our finding that only a small percentage of A␤ (Ϸ5%) in plasma binds to lipoproteins is consistent with such a subtle, long term effect.
The majority of iodinated A␤ added to fresh plasma became bound to albumin. Albumin is the most abundant protein in both plasma and CSF (37). Beside its function in stabilizing the osmotic pressure of fluids, it is one of the most important physiological transport proteins. It binds and transports a variety of small molecules, including peptides and drugs, and plays a vital role in lipid metabolism by transporting fatty acids. Only 40% of body albumin circulates in plasma; the remaining 60% is found in extravascular tissue fluids, and albumin has a high turnover (for review, see Ref. 36). Only a small percentage of endogenous plasma albumin crosses the intact blood-brain barrier (BBB). However, this situation is altered during pathological opening of the BBB (58 -60). Head trauma, even mild head injury, has been reported to transiently open the BBB (61,62). Head trauma is known to play a role as a risk factor that can contribute to the development of AD (e.g. see Ref. 63); for example, massive cerebral A␤ accumulation has been reported in professional boxers experiencing multiple episodes of head trauma (dementia pugilistica) (64). However, the BBB seems to be focally disrupted in AD brains (e.g. see Ref. 65), and A␤ itself is capable of damaging the structure and function of cerebral endothelial cells that constitute the barrier (66). One can therefore speculate that an increased influx of albumin-bound A␤ across a transiently or permanently altered BBB could play a role in enhanced deposition of A␤ in the extracellular space of the brain. Indeed, the ability of circulating synthetic A␤ to accrue onto existing plaque and vascular amyloid deposits has been recently shown in living monkeys (67), and there is evidence that iodinated A␤ 1-40 injected intravenously into rodents can pass the bloodbrain barrier and accumulate within the brain parenchyma and CSF in vivo (68,69).
Albumin is universally distributed in plasma, CSF, and all tissues, so that A␤, once secreted by neuronal and non-neuronal cells, could be bound to it. Consequently, it is likely from our studies that A␤ in various tissues, including brain, is transported by and metabolized with albumin. In this regard, it would be of great interest to determine whether one or more of the Ն20 known human albumin alleles (36) are associated with a higher risk of chronically depositing A␤ and developing AD.
Finally, another consideration that arises from the data we report here relates to the design and development of antiamyloid drugs. To identify and characterize a compound that is supposed to interact directly with A␤ in vivo (e.g. a drug that inhibits A␤ aggregation), it is necessary to know whether A␤ is free or bound to plasma and CSF proteins, and precisely which proteins these are. The finding that virtually all A␤ added to fresh plasma at physiological concentrations becomes bound to albumin and certain lipoproteins will need to be taken into consideration when evaluating compounds that are intended to bind to, sequester, or clear A␤ in vivo.