Human Serum Albumin Can Regulate Amyloid-β Peptide Fiber Growth in the Brain Interstitium

Background: 95% of amyloid-β in blood plasma binds to albumin with a Kd of 5 micromolar. Results: Physiological, micromolar levels of albumin inhibit amyloid fiber formation. Conclusion: Nearly half of Aβ in the CSF will be bound to albumin and inhibited from forming fibers. Significance: Levels of albumin in CSF must represent a risk factor and therapeutic target in Alzheimer disease. Alzheimer disease is a neurodegenerative disorder characterized by extracellular accumulation of amyloid-β peptide (Aβ) in the brain interstitium. Human serum albumin (HSA) binds 95% of Aβ in blood plasma and is thought to inhibit plaque formation in peripheral tissue. However, the role of albumin in binding Aβ in the cerebrospinal fluid has been largely overlooked. Here we investigate the effect of HSA on both Aβ(1–40) and Aβ(1–42) fibril growth. We show that at micromolar cerebrospinal fluid levels, HSA inhibits the kinetics of Aβ fibrillization, significantly increasing the lag time and decreasing the total amount of fibrils produced. Furthermore, we show that the amount of amyloid fibers generated directly correlates to the proportion of Aβ not competitively bound to albumin. Our observations suggest a significant role for HSA regulating Aβ fibril growth in the brain interstitium.

Fibrillar amyloid-␤ (A␤), 2 a 39 -43-residue peptide, is the major constituent of extracellular plaques in the brain interstitium of Alzheimer disease (AD) patients. The most prevalent variations are A␤(1-40) and A␤ , with A␤(1-42) being the more amyloidogenic form. Genetic alterations underlying familial AD indicate that A␤ plays a central role in the disease as inherited AD is associated with mutations in A␤ or its increased production (1).
Interestingly, unlike systemic amyloid-related diseases, although A␤ is found in the blood plasma at 0.1-0.5 nM concentrations (2), similar to A␤ levels found in the cerebrospinal fluid (CSF) (3), A␤ amyloid deposits are largely found in the interstitium within the brain. Human serum albumin (HSA) binds 90 -95% of the A␤ found in blood plasma (4,5). HSA is the most abundant protein in blood serum, at a concentration of ϳ640 M (6), but has a markedly reduced concentration in the CSF of typically 3 M (7). This may explain why A␤ plaques are only observed in the extracellular space of the brain and not the peripheral tissue.
The affinity of monomeric A␤ for HSA has been determined, and a dissociation constant (K d ) of 5-10 M for both A␤  and A␤(1-42), based upon a 1:1 stoichiometry, has consistently been reported by a number of groups (5,8,9). Indeed, with 640 M concentration of albumin in blood plasma, a K d of only 30 M would be sufficiently tight to bind more than 95% of physiological A␤.
The level of HSA in the CSF (3 M) and a micromolar affinity (K d ϭ 5-10 M) for A␤ suggests that although the majority of A␤ is bound to HSA in the blood, the capacity of albumin to bind A␤ in the brain interstitium will be quite sensitive to changes in HSA levels. Therefore, variations in the capacity of HSA to bind A␤, through decreased HSA concentration or competition for the A␤ binding site, could play a role in the buildup of toxic A␤ oligomers and fibers within the brain.
There is some debate over whether HSA binds to A␤ monomers or oligomers. Milojevic et al. (10 -12) have performed a series of experiments that have produced results consistent with HSA binding to oligomers but not monomers or fibrils. Similarly, it has been suggested that HSA traps A␤ in an oligomeric form (13). Others have suggested that HSA binds monomers of A␤ (5,8). A study using immobilized A␤ polymers and biotin labeling indicates that HSA can inhibit soluble A␤ addition to immobilized A␤ seeds, suggesting A␤ polymerization inhibition (9).
As HSA interacts with A␤, we were interested in how HSA might affect the kinetics of fibril formation, fiber nucleation, and elongation. Although it is often suggested that albumin inhibits fiber formation, we know of no direct study of the kinetics of amyloid formation in the presence of CSF levels of albumin. Here we study inhibition of A␤ fibril growth in detail. With an aim to understand fiber formation and kinetics in vivo, we monitored fibril growth rates using physiologically relevant (1-10 M) and substoichiometric concentrations of HSA to determine the concentration-dependent inhibition of A␤ fibril formation.

EXPERIMENTAL PROCEDURES
A␤ Production and Solubilization-A␤  and A␤(1-42) were purchased commercially from Zinsser Analytic and Cambridge Research Biochemicals, having been synthesized using solid phase Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistry. The peptides were characterized by 1 H NMR and HPLC, which indicated a single peak with the expected molecular mass. Met 35 was confirmed to be unoxidized. Lyophilized A␤(1-40) and A␤(1-42) were solubilized by dissolving 0.7 mg/ml A␤ in water at pH 10.5 and then rocked gently at 5°C for 72 h. The concentration of A␤ was determined using the tyrosine absorbance at 280 nm, ⑀ 280 ϭ 1280 M Ϫ1 cm Ϫ1 .
Human Serum Albumin-Essentially, fatty acid-free HSA was purchased from Sigma-Aldrich and solubilized in ultrahigh quality water (10 Ϫ18 ohms Ϫ1 cm Ϫ1 resistivity). The concentration was determined using the absorbance at 280 nm, Fiber Growth Assay-The kinetics of amyloid formation was monitored by the binding of thioflavin T (ThT) to amyloid fibers. ThT fluorescence occurs at 487 nm upon binding to amyloids; thus the ThT signal is related to the amount of amyloid produced. The ThT measurements were conducted on a BMG Galaxy FLUOstar fluorescence 96-well plate reader, using only the central 60 wells to minimize the effect of evaporation. Readings were obtained using an excitation wavelength of 440 nm and emission wavelength of 490 nm. Readings were taken every 30 min for ϳ400 h with the well plates being subjected to 30 s of orbital shaking prior to each measurement. 10 M A␤ was incubated with 20 M ThT and the required HSA concentration at 30°C in 160 mM NaCl. 30 mM HEPES buffer was used throughout to maintain pH at 7.4, with variations of 0.05 pH units or less. Ultra-high quality water (10 Ϫ18 ohms Ϫ1 cm Ϫ1 resistivity) was used at all times.
The data obtained can be fitted to the growth curve using the following equation (14) Y ϭ ͑ y i ϩ m i x͒ ϩ ͑v f ϩ m f x͒ where Y is the fluorescence intensity, x is the time, x 0 is the time at half-height of fluorescence (t 50 ), and is a time constant. This equation can be used to extract a number of empirical parameters including the lag time (t lag ϭ x 0 Ϫ 2), the apparent fiber growth rate (k app ϭ 1/), and the t 50 . The t 50 is influenced by both the nucleation and the elongation phases. Statistics-Fibril growth curves were presented using KaleidaGraph. Typically, kinetic parameters have been extracted from six raw traces, and the mean values and standard errors were calculated. Analysis of variance (ANOVA) was used to compare the kinetic parameters extracted from curve fitting at different concentrations of HSA. One-way ANOVA with Tukey's HSD post hoc tests was used to reveal significant differences at ␣ ϭ 0.05.
Transmission Electron Microscopy (TEM)-Glow-discharged carbon-coated 300-mesh copper grids were prepared using the droplet method, where 5-l aliquots of samples from the fiber growth assay were adsorbed for 1 min followed by 5 l of 2% (w/v) of phosphotungstic acid adjusted to pH 7.4, to neg-atively stain the sample. Images were recorded on a JEOL JEM-1230 electron microscope operated at 180 kV.

RESULTS
The kinetics of A␤ fibril formation was monitored using ThT, a dye that fluoresces upon binding to amyloid fibrils. The rate of fiber formation of A␤(1-40) was monitored over time in the presence and absence of increasing concentrations of HSA, using a 96-well plate fluorescence reader. A characteristic fibrillization growth curve was obtained, which has a lag phase and growth phase, due to nucleation and fibril elongation, respectively.
In Fig. 1, we show the fibrillization growth curves obtained from 10 M A␤(1-40) in the presence of 0, 3, 5, and 10 M HSA. Typically, six individual traces were obtained at each albumin concentration, and key kinetic parameters, including the t 50 , t lag , and k app , were extracted through fitting the characteristic sigmoidal growth curves (14). For comparison, the average t 50 of A␤ fibrillization in the absence of albumin is depicted as a dashed line on each trace. It is clear from Fig. 1 that fibril growth is delayed by all three substoichiometric concentrations of HSA as the elongation phase of the growth curves are all past the average t 50 for A␤(1-40) alone.
There was some variation in the fibril growth kinetics, and so essentially, the identical experiment was repeated on three other separate occasions, an example of which is shown in supplemental Fig. 1. The fibril growth kinetics from these four separate experiments, for A␤(1-40) fibrillization in the presence of increasing concentrations of HSA, was averaged together to generate large n numbers (n ϭ 24), and the standard errors were calculated, as shown in Fig. 2. Variation in the kinetic parameters for fibril growth recorded on separate days was comparable with variation for individual fiber growth experiments. One-way ANOVA with Tukey's HSD post hoc tests was performed on these data; HSA increased the t 50 of A␤(1-40) fibril growth, but this was only significant (p Ͻ 0.05) at half a molar equivalent of HSA (5 M) and higher. Analysis of the t lag and k app indicated that HSA prolongs the lag phase of A␤ fibril formation with as little as 1 M HSA but has no significant effect on the rate of elongation of fibers (k app ) once nucleation has taken place. This means that the effect on the t 50 values is exclusively due to the inhibition of fiber nucleation, measured as the lag time (t lag ). It is clear that substoichiometric amounts of HSA at concentrations found in the CSF increase the lag time of A␤(1-40) fiber formation.
There was also a reduction in the total amount of fibers generated as indicated by the maximal fluorescence signal produced after more than 400 h. From Fig. 1, it is clear that the ThT fluorescence signal for 10 M HSA present is approximately half the intensity relative to A␤(1-40) in the absence of HSA. The mean maximum fluorescence intensities and the associated standard errors from the four repeat experiments of A␤(1-40) fibrillization were also calculated and are shown in Fig. 3a. The trend that can be observed from this is that the maximum fluorescence intensity, or total amount of fibers generated, decreases as the concentration of HSA present increases. This closely agreed with what is known about the K d of HSA for A␤. With a K d of 5 M and an equimolar mixture of A␤ and HSA (at 10 M), then 50% of the A␤ molecules will bind to HSA (see supplemental Affinity calculations). Interestingly, at this level of HSA, there was exactly a 50% reduction in total fibers generated for A␤(1-40) and A␤(1-42). Using the known dissocia-tion constant between A␤ and HSA (K d ϭ 5 M), it is possible to calculate the amount of free and bound A␤ for any concentration of A␤ and HSA by solving simple quadratic equations derived from the equation for the dissociation constant shown below equation 3; see also the supplemental material.  The strong correlation between the total amount of fiber generated and the predicted amount of available A␤ (not bound to HSA) is striking, as shown in Fig. 3b. In particular, the linear regression gives an R value of 0.998 and a gradient of 1.0. To confirm that albumin has indeed inhibited fiber formation, a second detection method was used. A␤(1-40) was incubated with 10 M HSA for 400 h as before (Fig. 1), and TEM images were obtained. In the absence of albumin, typical fiber  morphology was observed with numerous fibers generated (Fig.  4a). In the presence of albumin, almost no fibers or oligomers could be observed after searching many squares per grid (Fig.  4b).
Next we wanted to investigate the effect of albumin on A␤(1-42), the more amyloidogenic A␤ peptide found in plaques. A␤  showed similar fibril growth kinetics to A␤(1-40), as shown in Fig. 5, with all traces in the presence of 5 and 10 M HSA showing delays in fiber formation relative to A␤ alone. One-way ANOVA with Tukey's HSD post hoc tests showed that HSA significantly (p Ͻ 0.05) increased the lag phase and t 50 of A␤(1-42) fibril growth at concentrations over 5 M (supplemental Fig. 2). Once again, it can be observed that the maximal ThT fluorescence intensity for the fibril growth is reduced in the presence of albumin, indicating that as well as increasing the time required for fibrils to form, the total concentration of fibers formed is also reduced.

DISCUSSION
Perhaps because levels of albumin are so much lower in the CSF than the blood plasma, a role for HSA interacting with extracellular A␤ in the brain interstitium has largely been overlooked. However, 3 M concentration of albumin still represents the major protein component of the CSF. Furthermore, although the affinity of A␤ for HSA is quite modest (K d ϭ 5 M), it is clear that ϳ40% of A␤ within the CSF will be bound to HSA. Our results show that at physiological levels of albumin, the rate at which fiber formation is nucleated for both A␤(1-40) and A␤(1-42) is significantly inhibited. Moreover the total concentration of fiber generated is reduced by HSA; this suggests that HSA binds to A␤ molecules and traps them in a nonfibrillar form so that they are not available to form fibers.
In the presence of physiological micromolar levels of albumin, some but not all A␤ is trapped in a nonfibrillar form of A␤. Indeed, the total concentration of fibers generated with increasing levels of HSA (Fig. 3) shows a remarkably close agreement with the amount of free and albumin-bound A␤ that would be expected for a K d of 5 M. The predicted percentage of free A␤ based on the affinity of HSA for A␤ is shown in Fig. 3b. HSA can bind to A␤ in a nonfibrillar form as the final amount of fibers produced is reduced by the presence of HSA. Our data strongly support HSA interacting with monomeric A␤ to form a 1:1 complex with a K d of 5 M. Although binding oligomers of A␤ cannot be ruled out, they must be small, less than five A␤ monomers, as none are observed by TEM. It is notable that a    Fig. 3b (R ϭ 0.998) implies a K d very close to 5 M. Albumin binding to the monomeric A␤ effectively reduces the concentration of free A␤ available to form fibers. The lag phase in fibril growth is strongly affected by the A␤ concentration, which would explain the reduction in lag times observed. It seems unlikely that the fiber growth inhibition is due to HSA interacting with A␤ fibers as the fiber elongation growth rates are not reduced once a nucleating seed has formed. This suggests that HSA has a role in preventing the formation of nucleating seeds, but once they are formed, it has less effect in monomer addition to the ends of growing fibers.
There have been a number of protein binding partners of A␤ indicated, in particular, the cellular prion protein (15) but also serum amyloid P (SAP) (16), islet amyloid polypeptide (IAPP) (17), and transthyretin (18). The composition of plaques from the brain supports the idea that HSA interacts with a nonfibrillar/monomeric form of A␤ as HSA is not found within plaques, in contrast to serum amyloid P (19).
HSA is known for its ability to bind many hydrophobic molecules, in particular a number of pharmaceuticals, including warfarin and diazepam, as well as endogenous fatty acids (6). A crystal structure of HSA with fatty acids bound has indicated the location of hydrophobic pockets formed within HSA (20), which may be where part of the hydrophobic C-terminal residues of A␤ could bind to HSA. It has been suggested that there are pharmaceutical and endogenous hydrophobic molecules that could compete with A␤ binding to HSA (9). Albumin is also responsible for the transport of the labile pool of Cu 2ϩ ions in blood plasma (21). Of note, the A␤-Cu 2ϩ interaction and its role in AD have received significant interest (22)(23)(24), and Cu 2ϩ may be transferred from A␤ to HSA (25), meaning that albumin could have two mechanisms by which it inhibits fiber formation.
The concentration of HSA that will delay the fibrillization of A␤ is of interest, in particular, the level where HSA has a significant effect on A␤ fiber formation in vitro, as it is strikingly similar to the extracellular albumin levels found in vivo. HSA is found at a concentration of ϳ3 M in the cerebrospinal fluid and brain interstitium (26). Although HSA is at substantially lower levels than in blood plasma, 3 M still represents a major constituent of the CSF and the brain interstitium. With 0.1-1 nM A␤ and a 5 M K d for albumin, we can assume that 37% of A␤ will be bound to albumin in the CSF. Considerable variations in CSF levels of albumin (1-5 M) have been reported (27). A reduction to 1 M levels of albumin will lead to 83% of the A␤ remaining free in solution (see supplemental Affinity Calculations). Thus a decrease in the concentration of HSA could cause an increase in the amount of amyloid fibrils produced and as such lead to the buildup of amyloid plaques. It may therefore be possible that a small change in HSA in the CSF and interstitium may represent a significant risk factor in AD. However, both elevated and reduced levels of albumin have been reported in AD patients (28,29). It has been noted that A␤ plaques are only ever observed in the brain and not the peripheral tissue, probably because HSA concentrations are of a much higher magnitude (640 M) in the blood plasma (26). Interestingly, serum albumin levels, in vivo, decrease with age, which is also a known risk factor in AD (30). Furthermore, levels of albu-min are often reduced in association with inflammation, also linked with AD.
In conclusion, although the affinity of A␤ for HSA is not thought to be remarkable (K d ϭ 5 M), it follows that with levels of albumin in the CSF and brain interstitium of 3 M, 40% of A␤ in the CSF will be bound to albumin. We show that this A␤ bound to albumin is trapped in a nonfibrillar form, thus reducing the amount of A␤ available to form fibers. Furthermore, based on our fiber growth studies, small variations in albumin in CSF are therefore likely to regulate the amount of extracellular fibril formation in the CSF and brain interstitium. Currently, HSA is showing promise in a phase II clinical trial, using an albumin plasma exchange schedule to reduce levels of A␤ in blood plasma (31). A reduction of the A␤ pool in plasma will, in turn, reduce A␤ levels in the CSF as A␤ is in a dynamic equilibrium and is able to cross the blood-brain barrier (32).