Atomic-resolution map of the interactions between an amyloid inhibitor protein and amyloid β (Aβ) peptides in the monomer and protofibril states

Self-association of amyloid β (Aβ) peptides is a hallmark of Alzheimer's disease and serves as a general prototype for amyloid formation. A key endogenous inhibitor of Aβ self-association is human serum albumin (HSA), which binds ∼90% of plasma Aβ. However, the exact molecular mechanism by which HSA binds Aβ monomers and protofibrils is not fully understood. Here, using dark-state exchange saturation transfer NMR and relaxation experiments complemented by morphological characterization, we mapped the HSA-Aβ interactions at atomic resolution by examining the effects of HSA on Aβ monomers and soluble high-molecular weight oligomeric protofibrils. We found that HSA binds both monomeric and protofibrillar Aβ, but the affinity of HSA for Aβ monomers is lower than for Aβ protofibrils (Kd values are submillimolar rather than micromolar) yet physiologically relevant because of the ∼0.6–0.7 mm plasma HSA concentration. In both Aβ protofibrils and monomers, HSA targets key Aβ self-recognition sites spanning the β strands found in cross-β protofibril structures, leading to a net switch from direct to tethered contacts between the monomeric Aβ and the protofibril surface. These HSA-Aβ interactions are isoform-specific, because the HSA affinity of Aβ monomers is lower for Aβ(1–42) than for Aβ(1–40). In addition, the HSA-induced perturbations of the monomer/protofibrils pseudo-equilibrium extend to the C-terminal residues in the Aβ(1–42) isoform but not in Aβ(1–40). These results provide an unprecedented view of how albumin interacts with Aβ and illustrate the potential of dark-state exchange saturation transfer NMR in mapping the interactions between amyloid-inhibitory proteins and amyloidogenic peptides.

The exact etiology of Alzheimer's disease is not fully understood, but the amyloid cascade hypothesis rationalizes a critical subset of the molecular phenotypes linked to the pathology of Alzheimer's disease (1). According to the amyloid hypothesis, the aggregation of A␤ 2 peptides in the brain contributes to the neuronal death and brain damage typically observed in Alzheimer's patients. The A␤ peptides are present in both the cerebrospinal fluid (CSF) and in blood plasma, but in the latter the self-association of A␤ into amyloids is inhibited primarily by human serum albumin (HSA), which binds ϳ90% of plasma A␤ (2, 3). The inhibition of A␤ self-association by HSA has both pathological implications and therapeutic potential (4 -7). Low serum albumin concentrations have been reported to be associated with increased cognitive impairment in elderly patients (4), and plasmapheresis with therapeutic albumin is currently being assessed in clinical trials as a potential treatment for mild to moderate Alzheimer's disease (8,9). The pathological and therapeutic potential of HSA warrants a comprehensive understanding of the molecular mechanism underlying the HSA-A␤ interactions, not only to improve the HSA therapeutic efficiency but also to elucidate basic principles of amyloid inhibition that will facilitate the design of new amyloid inhibitors (10 -17).
Although it is known that albumin inhibits amyloid formation by binding A␤ protofibrils with higher affinity than A␤ monomers and interfering with A␤ protofibrils-monomer recognition (11-13, 18 -20), several questions remain open about the molecular mechanism through which HSA prevents A␤ aggregation. First, because of the transient and elusive nature of the protofibrils formed under physiological conditions by the two major physiological species of A␤, i.e. 1-40 and 1-42 (21)(22)(23)(24), it is currently not fully understood how protofibrils are perturbed by HSA. Addressing this question is critical to explain how HSA modulates A␤ self-association (11)(12)(13). Second, it is currently unclear to what extent and how monomeric A␤ peptides bind HSA in plasma. Although the affinity of HSA for A␤ monomers is expected to be weak (25), the HSA concentration in plasma is high (ϳ0.6 -0.7 mM), and therefore lowaffinity interactions with HSA (K d Յ mM) are of potential physiological relevance.
To address these questions, we have prepared solutions of A␤  and A␤  either diluted to a primarily monomeric form or in a dynamic pseudo-equilibrium between monomers and high-molecular weight oligomeric protofibrils, stabi-lized through the use of low temperatures and desalting, as previously described (26). Under these experimental conditions, the interactions between monomeric A␤  or A␤ , denoted here as A␤40 1 or A␤42 1 , respectively, and the surface of soluble A␤ protofibrils, respectively denoted here as A␤40 n or A␤42 n , are effectively probed at atomic resolution by a combination of 15 N T 2 relaxation experiments and selective 15 N saturation transfer, as implemented through the dark-state exchange saturation transfer (DEST) NMR pulse sequence (26 -31).
Here, we utilize DEST, relaxation, and saturation transfer difference (STD) NMR experiments to monitor the interactions between unlabeled HSA and 15 N-labeled A␤  and A␤ . Based on these data, we propose a dual mechanism for the inhibition of A␤ self-association by HSA, whereby at plasma concentrations the latter interacts with both A␤ monomers and protofibrils, targeting key A␤ self-recognition sites and leading to a net switch from direct to tethered contacts between monomeric and protofibrillar A␤. In this context, the term "tethered" refers to the lack of direct contacts between a given residue of monomeric A␤ and the surface of A␤ protofibrils. Hence, a tethered residue of A␤ monomers is anchored to the A␤ protofibrils through other residues of monomeric A␤ that contact directly the A␤ protofibrils. The relative tethered versus direct contact probabilities are residue-specific (26), and the DEST experiment provides a means to quantify these tethered versus direct contact probabilities on a residue-specific basis (26). We also show that these HSA-A␤ interactions are isoform-specific, because distinct A␤(1-40) versus A␤  differences are observed in the interactions with albumin.

The interaction between monomeric A␤(1-40) (A␤40 1 ) and HSA is weak (K d ‫؍‬ ϳ0.1-1.0 mM) but physiologically relevant
To probe the interactions between HSA and A␤(1-40) monomers, we analyzed dilute (50 -60 M) solutions of 15 N-labeled A␤  in both the absence and the presence of equimolar amounts of HSA. Under these dilute conditions, the A␤  peptide is primarily monomeric, as supported by the observation of only marginal off-resonance DEST effects and the absence of significant residue-dependent variations (supplemental Fig. S1). No appreciable chemical shift changes are detected for A␤(1-40) upon addition of HSA (supplemental Fig. S2), consistent with low populations of the high MW A␤40 1 -HSA complex relative to free A␤40 1 . However, minor fractions of high MW complexes are often sufficient to result in marked enhancements in 15 N R 2 rates. Hence, we measured the 15 N R 2 relaxation rates of monomeric 15 N-labeled A␤  with and without equimolar amounts of HSA ( Fig. 1a and supplemental Fig. S3b, gray bars). Fig. 1a and supplemental Fig. S3b (gray bars) reveal that the majority of the significant HSA-induced 15 N R 2 enhancements occur within the segments that span the two ␤-strands (i.e. ␤ 1 and ␤ 2 ) involved in the cross-␤ structure typical of A␤ fibrils. The largest 15 N R 2 increases cluster in the 31-40 region ( Fig. 1a and supplemental Fig. S3b, gray bars), suggesting that these C-terminal residues are a primary site for the HSA-A␤40 1 interaction. These results were independently confirmed through STD-HSQC experiments (Fig. 1, c and e). Fig. 1 (c and e) reveals that saturation is transferred to multiple A␤40 1 sites, including a continuous stretch spanning residues 30 -40, as well as additional residues in the 12-24 region, such as the 17-19 segment in the central hydrophobic core. No major STD-HSQC cross-peaks were observed in the absence of HSA (Fig. 1c), confirming that the STD signals observed in the presence of HSA do indeed reflect the transfer of saturation from HSA to A␤  and genuinely report on the HSA-A␤40 1 interaction. Overall, the HSA-dependent enhancements in STD and 15 N R 2 rates consistently point to the 31-40 C-terminal region in ␤ 2 as a primary consensus site for the binding of A␤40 1 to HSA, with additional interaction sites close to or within the central hydrophobic core of A␤40 1 spanned by ␤1.
To estimate the K d value for the HSA-A␤40 1 interaction, a dilute solution of 15 N-labeled A␤(1-40) monomers was titrated with increasing concentrations of HSA, and the titration was monitored through the HSQC intensity losses caused by the HSA-induced R 2 enhancements (Fig. 1f). As the HSA concentration increases the signal intensity of several HSQC cross-peaks decreases in a dose-dependent manner (Fig. 1f). When the data of Fig. 1f are fitted using a Scatchard-like model, K d values in the 0.1-1.0 mM range are obtained. This result was confirmed by monitoring the binding isotherm through STD/ STR ratios (supplemental Fig. S3c). However, the exact values of the fitted K d should be interpreted with caution, because the site-specific affinities (K d ) depend also on the stoichiometry of the A␤40 1 -HSA complex. For instance, if only a single molecule of monomeric A␤(1-40) binds each molecule of HSA (i.e. n ϭ 1 in Fig. 1f), K d ϭ ϳ0.2 mM, but if each of the three homologous domains of HSA binds monomeric A␤(1-40) (i.e. n ϭ 3 in Fig. 1f), then K d ϭ ϳ0.6 mM (Fig. 1f). Despite these uncertainties, the data of Fig. 1f point to HSA binding monomeric A␤(1-40) with effective K d values in the 0.1-1.0 mM range, which is comparable with the concentration of HSA in plasma, and suggest that the A␤40 1 :HSA interactions, although weak, are physiologically relevant.

Monomeric A␤(1-42) (A␤42 1 ) also binds HSA but more weakly than A␤40 1
To investigate whether the ability of plasma albumin to bind A␤40 1 extends to A␤42 1 , we repeated the 15 N R 2 , STD-HSQC, and titration experiments ( Fig. 1, a, c, and f) also for dilute solutions of the longer A␤ isoform ( Fig. 1, b, d, and f). Fig. 1b shows that upon addition of albumin pervasive 15 N R 2 relaxation enhancements are observed for most A␤42 residues, indicating that under our experimental conditions, HSA binds A␤42 1 as well. However, the STD-HSQC and the titration data consistently point to HSA binding A␤42 1 with lower affinity than A␤40 1 , despite the longer length of the former peptide ( Fig. 1, d and f). The STD enhancement detected upon addition of albumin to A␤42 1 is marginal (Fig. 1d), in stark contrast with the major STD increase observed upon albumin addition to the shorter A␤40 1 under similar experimental conditions (Fig. 1c). The reduced STD contribution arising from HSA is consistent with a reduced fraction of albumin-bound peptide in the A␤42 1 versus A␤40 1 solutions. Furthermore, the titration data (Fig. 1f)

A␤ peptides bind HSA via a dual mechanism
confirm that in going from A␤40 1 to A␤42 1 , the site-specific dissociation constant is subject to a ϳ2-fold increase, resulting in a K d Ј value of 450 Ϯ 100 M, i.e. a value that is still nonnegligible compared with physiological albumin concentra-tions in plasma. Based on these affinities and considering a plasma concentration of HSA of 644 M, significant fractions of albumin-bound monomeric peptide are expected in plasma for both A␤  and A␤ . However, unlike in plasma,

A␤ peptides bind HSA via a dual mechanism
HSA concentrations in the CSF are limited to ϳ3 M, i.e. too low for significant HSA-A␤ 1 interactions but more comparable with the affinity of albumin for A␤ protofibrils (18). Hence, we proceeded to investigate the effect of HSA on more concentrated solutions of A␤(1-40) and A␤ , in which A␤ monomers are in a dynamic pseudo-equilibrium with A␤ assemblies.

Preparation of A␤(1-40) and A␤(1-42) protofibrils (A␤40 n and A␤42 n )
To examine the effect of HSA on the A␤ monomerprotofibril exchange process, we prepared a concentrated (300 M) solution of 15 N-labeled A␤ . A␤ peptides are known to form large amyloid protofibrils spontaneously when dissolved at concentrations higher than 100 M (26). If incubated at 37°C in the absence of albumin, the A␤(1-40) protofibrils grow into large fibrils, as shown by EM images at 150,000ϫ magnification (supplemental Fig. S4, left panels), which form highly compact and dense tangles, as shown by EM images at 1,000 -4,000ϫ magnification (supplemental Fig. S4, left panels). However, when the concentrated A␤(1-40) sample is left to equilibrate at low temperature (4 -10°C) for 7-10 days, a pseudo-equilibrium is reached in which A␤ monomers are in dynamic exchange with protofibrils (A␤40 n ) (26). Indeed, after 10 days of incubation at 10°C, our 300 M A␤(1-40) solution contained large (Ͼ10 nm) worm-like aggregates, as shown by dynamic light scattering (supplemental Fig. S5a) and electron microscopy ( Fig. 2a, left panels). Unlike the results obtained at higher temperatures (supplemental Fig. S4, left panels), which serve as our positive control, the A␤(1-40) sample at 10°C did not contain any observable fibrillar tangles or mature fibers (Fig. 2a, left panels). Similar conclusions were reached also for the preparation of A␤(1-42) protofibrils (A␤42 n ) at 10°C, as supported by DLS (supplemental Fig. S6) and EM data (Fig. 2b,
The presence of protofibrils in dynamic exchange with NMR-visible A␤ monomers in concentrated A␤(1-40) samples was independently confirmed by the average difference in NMR 15 N transverse relaxation rates (⌬R 2 ) between the 300 M and the diluted 60 M A␤(1-40) reference solution, which is greater than 1.00 s Ϫ1 (i.e. ϳ1.45 s Ϫ1 , supplemental Fig. S8A), as expected for NMR-invisible ("dark") A␤(1-40) protofibrils (32). The maximum value of ⌬R 2 provides an estimate of the pseudo first-order rate constant k on, app for the conversion from NMR-visible to invisible species (26), which based on supplemental Fig. S8A is ϳ2 s Ϫ1 . We then investigated how this dynamic steady state A␤40 1 -A␤40 n exchange is affected by HSA.

An approach to probe how HSA perturbs the A␤ monomer-A␤ protofibril interactions: the ⍜ values
The ⌬R 2 versus residue profile measured for concentrated (Ͼ100 M) A␤ samples is typically combined with DEST data from 15 N-selective saturation experiments at multiple offsets to map the A␤40 1 -A␤40 n interactions (26). The combined analysis of the ⌬R 2 and DEST profiles provides a residue-specific constant, defined as K 3 , which quantifies the partitioning of a given A␤ residue i between direct and tethered contacts with the A␤ protofibril surface. However, this quantitative approach requires highly precise DEST measurements with 15 N-selective saturation at multiple offset frequencies and hence long acquisition times during which the pseudo-equilibrium of A␤ solutions may change, thus biasing the comparisons with the data acquired in the presence of HSA. As a first step toward minimizing this experimental bias and reducing the DEST acquisition time, while still probing how HSA perturbs the A␤ monomer-protofibril equilibrium, we analyzed the DEST data similarly to the analysis of the traditional 1 H STD experiments. Specifically, we calculated the DEST difference (⍜), or single point DEST, which is defined here according to the following equation, where I°refers to the DEST intensities measured at far offresonance 15 N frequency offsets (e.g. Ϯ 35 kHz), which serve as a reference, whereas Ist refers to the DEST intensities measured at 15 N frequency offsets sufficiently close to the carrier frequency ("near off-resonance offsets") to saturate A␤ protofibrils but still sufficiently far from the 15 N resonance frequencies of monomeric A␤ to minimize direct saturation of A␤ monomers (e.g. Ϯ 2-8 kHz). The ⌺Ϯ notation indicates that the intensities at the positive and negative offset values were averaged to remove to first order the effect of the 15 N chemical shift resonance offset on the on single point DEST (⍜). The single point DEST (⍜) reports primarily on the width of the DEST intensity versus 15 N-saturation offset profile and is expected to range between 0 and 1, with the former (latter) value approached by A␤ monomers residues subject to primarily tethered (direct) contacts with the surface of the A␤ protofibrils.
To test the effectiveness of the proposed single point DEST (⍜) method in probing the A␤40 1 -A␤40 n interactions, we computed the ⍜ values using previously acquired A␤(1-40) DEST data, for which the K 3 direct versus tethered contact partitioning coefficients were determined through the combined McConnell fitting of DEST and R 2 data (32 Fig. S9a. Supplemental Fig. S9a shows that the ⍜ versus residue profiles calculated using different near off-resonance values, i.e. Ϯ2 kHz, Ϯ4 kHz, and Ϯ8 kHz, exhibit similar patterns, with maxima located at similar residue numbers. Given these similarities, we opted for Ϯ4 kHz as the optimal near off-resonance saturation frequency, because it provides the advantage of a balanced compromise between adequate saturation of the A␤ protofibrils, resulting in significant crosspeak attenuation (up to ϳ40%, supplemental Fig. S9a) and optimal A␤ protofibril versus monomer selectivity (32). The minimal effect of the selective 15 N saturation at Ϯ4 kHz on the A␤ monomer is proven by the negligible magnitude of the ⍜ values measured for a dilute (60 M) solution of primarily monomeric A␤(1-40) (supplemental Fig. S1, black bars).
To test the suitability of ⍜ values at Ϯ4 kHz to estimate the K 3 values, we compared the residue profiles for both ⍜ and K 3 as shown in supplemental Fig. S9b. The ⍜ versus K 3 comparison (supplemental Fig. S9b) indicates that the two parameters exhibit similar trends, with both profiles featuring maxima at similar residue positions. Hence, the ⍜ residue profile provides an approximation to the relative K 3 values, as required to effectively probe the effect of HSA on A␤ protofibrils.
Next, we checked to what extent the ⍜ parameter is field-dependent. For this purpose, the single point DEST (⍜) was measured at Ϯ4 kHz and 700 MHz for a concentrated (300 M) A␤(1-40) solution equilibrated for 7 days at 4°C and compared to those computed using the published data acquired for a similar A␤ (1-40) sample at Ϯ4 kHz and 900 MHz (18, 23) (supplemental Fig. S9c). Despite the difference in field, and consequently in relative offsets as well, the two plots in supplemental Fig. S9c exhibit overall similar relative trends with two maxima centered at residues ϳ18 and ϳ33, consistently showing that the N-terminal residues prefer tethered states, whereas the central hydrophobic core and part of the C-terminal residues are more likely to be engaged in direct contacts with the protofibril surface.

The single point DEST (⍜) profiles reveal that HSA promotes a switch from direct to tethered-contacts for the A␤(1-40) residues involved in protofibril growth
The ⍜ profile was measured again after adding 70 M of unlabeled HSA to the equilibrated 300 M 15 N-labeled A␤  solution. With the exception of the worm-like aggregates becoming slightly more electron dense in the presence of HSA, the addition of HSA did not result in any major change in the protofibril integrity as monitored by DLS (supplemental Fig.  S5) and EM (Fig. 2a) under our experimental conditions. However, significant changes in ⍜ values were observed for most A␤ A␤ peptides bind HSA via a dual mechanism residues ( Fig. 3a and supplemental Fig. S11), consistent with an overall perturbation of the direct versus tethered A␤(1-40) monomer-protofibril contact distribution arising from interactions of HSA with A␤40 n and possibly with A␤40 1 .
The contribution of free HSA-A␤40 1 complexes to the HSAdependent ⍜ changes observed in Fig. 3a is expected to be negligible based on two main lines of evidence. First, under the experimental condition of Fig. 3, the fraction of A␤40 1 bound to HSA is minimal because of its weak affinity. Second, when unlabeled HSA is added to a dilute solution of 15 N-labeled A␤40 1 , no significant changes were observed in the DEST profile (supplemental Fig. S1). These observations suggest that the HSAdependent ⍜ changes observed in Fig. 3a arise mainly from direct interactions between HSA and the A␤40 n protofibrils, which bind albumin with higher affinity than A␤(1-40) monomers (10,18). Hence, the residue-specific ⍜ variations caused by HSA (Fig. 3, a and c) are a valuable indicator of how HSA binding to A␤40 n affects the direct versus tethered contact partitioning of A␤(1-40) monomers at the surface of the A␤40 n protofibrils provided that contributions from changes in k on, app or the populations of protofibrils are ruled out (as explained in the next section).
As shown in Fig. 3 (a and c), the ⍜ values for the majority of A␤(1-40) residues are decreased in the presence of HSA (Table  1), with the most significant ⍜ reductions observed for the central hydrophobic core and the region preceding the C terminus, at or near the in-register ␤-sheets of the protofibril structure. Only minimal or negligible differences were observed for the N terminus and the very C-terminal residues (Fig. 3, a and c). The decrease in ⍜ values observed upon HSA addition suggests that HSA promotes a switch from direct to tethered contacts between the A␤40 1 monomers and the A␤40 n protofibril surface. These results were independently confirmed at physiological temperature (37°C) by acquiring DEST data both in the absence and presence of albumin (supplemental Fig. S10). Supplemental Fig. S10 shows that incubation at 37°C for 24 h of a dilute A␤(1-40) sample results in a significant DEST effect for both ␤1 and ␤2 regions only in the absence of albumin, confirming that HSA leads to a switch from direct to tethered contacts between A␤(1-40) monomers and protofibrils. In addition, our EM data at 37°C (supplemental Fig. S4, 1,000ϫ and 4,000ϫ magnification images, right panels) suggest that albumin causes the fibril clumps to become more loosely packed than those observed in the absence of HSA. The EM images at   Fig. 3 (a and b). b Based on the data of Fig. 3 (c and d). The ⌬⌰ value depends on HSA concentration as shown in supplemental Fig. S11.

A␤ peptides bind HSA via a dual mechanism
150,000ϫ magnification also show that HSA coats the A␤ fibrils (supplemental Fig. S4, right panels), explaining the direct-to-tethered switch for the A␤40 1 -A␤40 n contacts observed by DEST (supplemental Fig. S10).
To probe to what extent the switch from direct to tethered contact states promoted by HSA resembles that caused by the spontaneous release of A␤40 1 from A␤40 n protofibrils, we compared the ⍜ change measured upon HSA addition to the concentrated 300 M A␤(1-40) sample (i.e. ⌬⍜ (Ϯ) has in Fig.  4) to the ⍜ change observed upon dilution of the concentrated (300 M) A␤(1-40) sample in the absence of HSA, i.e. ⍜ for concentrated A␤(1-40) assuming negligible ⍜ values at infinite dilution (Fig. 4). When all A␤(1-40) residues are included in the comparison, the ⌬⍜ (Ϯ) HSA and the ⍜ values for concentrated A␤(1-40) appear poorly correlated (Fig. 4a). However, when the ⌬⍜ analysis is confined to selected A␤(1-40) segments, such as residues toward the N terminus (i.e. 8 -12) and residues that span the C-terminal region, i.e. 37-40, a higher degree of correlation is observed (Fig. 4, b and d), suggesting that in these regions the effect of HSA addition is more comparable to that of A␤ dilution. Interestingly, the slopes of the correlations are different in the two segments (Fig. 4, b and d), showing that the extent of the dilution-like effect of HSA is region-specific.

Effect of HSA on the k on, app or the populations of protofibrils
It should also be considered that a decrease in the ⍜ value might result from a reduction not only in K 3 , but also in k on, app and/or the total population of protofibrils. To assess the contributions from the latter two parameters, it is important to evaluate the dark versus dilute R 2 change (⌬R 2 ) also in the pres-ence of HSA. In fact, k on, app typically equates the maximum value of ⌬R 2 (26). To obtain the dark versus dilute ⌬R 2 , we corrected for the R 2 contributions arising from the binding of HSA to A␤    Fig. 1a (i.e. ϳ18%). The R 2 rates of Fig.  1a (solid black bars) were then subtracted from the respective R 2 values measured for the dark sample in the presence of 70 M HSA to obtain the ⌬R 2 plot corrected for the R 2 contributions arising from the binding of HSA to A␤(1-40) monomers (supplemental Fig. S8a, filled black bars). The comparison of the corrected dark versus dilute ⌬R 2 profiles in the presence versus absence of HSA (supplemental Fig. S8a, filled versus open bars) reveals two important points.
First, supplemental Fig. S8a shows that, if decreases in k on, app and/or the total population of protofibrils occur upon addition of HSA, they are sufficiently limited to result in ⌬R 2 changes that fall within the experimental error margin of our R 2 data, i.e. HSA does not significantly decrease the ⌬R 2 values (supplemental Fig. S8a). Second, supplemental Fig. S8a reveals that, even after correction for the binding of free HSA to residual monomeric A␤  in the dark sample, a significant ⌬R 2 enhancement is observed upon HSA addition for several residues, especially in the C-terminal region (e.g. residues 37-40, supplemental Fig. S8a). This interesting effect is consistent with A␤(1-40) monomers interacting with protofibril-bound HSA and with the presence of multiple homologous domains in HSA. For example, although one domain of HSA binds the A␤(1-40) protofibrils, another domain of HSA may recruit A␤(1-40) monomers, which interact with albumin primarily through the C-terminal residues, as shown in Fig. 1 (a, c, and f). This example illustrates how A␤(1-40) monomers can interact with the A␤(1-40) protofibrils not only through direct contacts with the protofibril surface but also through indirect contacts mediated by protofibril-bound HSA.

The HSA-induced switch from direct to tethered contacts is A␤ isoform-specific
To test to what extent the albumin-induced remodeling of the direct versus tethered contacts observed for A␤(1-40) is isoform-specific, we extended the comparative DEST ⍜ analyses to A␤(1-42) (Fig. 3b). Fig. 3b reveals that, similarly to A␤ , also for A␤(1-42) albumin induces a pervasive reduction in ⍜ values (Table 1), without major perturbations of the morphology of the A␤ protofibrils in solution as supported by DLS (supplemental Fig. S6b) and EM data (Fig. 2b), except for the worm-like aggregates becoming more electron dense in the samples with HSA. However, the ⌬R 2 values for A␤(1-42) (supplemental Fig. S8b) reveal that the interpretations of the albumin-induced ⌬⍜ profiles of A␤(1-40) and A␤(1-42) (Fig.  3, c and d) are markedly different. Unlike for A␤ , for A␤(1-42) a reduction of ⌬R 2 is observed upon addition of HSA, suggesting that in this case HSA may act simply by reducing the effective k on, app and/or the total population of protofibrils. Furthermore, in the case of A␤ , no evidence is observed for HSA-mediated contacts between A␤(1-42) monomers and protofibrils, consistent with the weaker interactions between A␤(1-42) monomers and albumin, as per Fig. 1. Another clear A␤(1-42) versus A␤(1-40) difference is that in the case of A␤(1-42), the HSA-induced ⍜ reduction extends to the C-terminal residues (Fig. 3d), whereas for A␤ , the effect of HSA on the last five residues progressively decreases to negligible values for the last two residues (Fig. 3c).

Discussion
Our main findings are summarized in the scheme shown in Fig. 5. Fig. 5a illustrates that A␤(1-40) interacts with HSA through a dual binding mechanism involving both A␤(1-40) monomers and protofibrils. The former A␤(1-40) species are bound by HSA with K d values in the 0.1-1.0 mM range, whereas the latter bind albumin with affinities higher by ϳ2-3 orders of magnitude, typically in the ϳ M range (9,11). Hence, in plasma, where the physiological concentration of HSA is ϳ0.6 -0.7 mM (3), albumin is expected to interact with both A␤(1-40)  red line), as well as a secondary site spanning residues 17-24 (dashed red line). HSA also binds A␤(1-40) protofibrils with higher affinity than A␤(1-40) monomers (A␤40 1 ) and competes with the direct contacts between A␤40 1 and the A␤(1-40) protofibrils, thus inhibiting further growth into mature fibers that would otherwise occur under non-pseudo-equilibrium conditions in the absence of HSA, as explained in the text. However, HSA may also mediate indirect contacts between the A␤40 1 and the A␤(1-40) protofibrils. The A␤(1-40) residues involved in fibril cross-␤ structures are represented by thick arrows (23,47,48). One edge of the protofibril is displayed in gray to denote that HSA may not interact with both edges (12). b, model of A␤(1-42)-HSA interactions. The last two residues of A␤(1-42) stabilize a C-terminal turn in the monomeric peptide, possibly shielding the C-terminal region from HSA and reducing the affinity of A␤(1-42) monomers for HSA (dashed thin line rather than solid thick red line). However, the last two residues of A␤(1-42) also promote extensive HSA-induced protofibril perturbations (thicker red line), which now reach the C-terminal region of the second ␤-strand (full red rather than red/black C-terminal arrow). The black dashed circles/ovals highlight the sites of the A␤(1-40) versus A␤(1-42) differences.

A␤ peptides bind HSA via a dual mechanism
monomers and protofibrils, whereas in the CSF, where the concentration of HSA is ϳ3 M (3), only the HSA-A␤(1-40) protofibril interactions are anticipated to be physiologically relevant.
An unprecedented picture of how HSA perturbs the A␤(1-40) protofibrils is provided by the residue-resolution singlepoint DEST (⍜) profiles (Fig. 3, a and c). Our results show that the HSA-A␤40 n interactions affect preferentially A␤  residues involved in protofibril cross-␤ strand growth and promote a switch of these A␤(1-40) monomer-protofibril contacts from a direct to a tethered state (Fig. 5a and 3b and supplemental Fig. S10). For residues in the 8 -12 and 37-40 regions, the ⌬⍜ values arising from the HSA-induced direct-totethered switch correlate with the corresponding ⌬⍜ values observed upon dissociation of A␤(1-40) protofibrils into monomers (Fig. 4, b and d). These observations point to HSA shielding the protofibril at the sites of cross-␤ strand growth and competing with further A␤(1-40) monomer addition to the protofibrils (Fig. 5a), thus inhibiting protofibril growth into mature A␤(1-40) fibrils, as confirmed by our EM data (supplemental Fig. S4). However, the lack of global correlations between the effects of dilution and HSA addition on ⍜ (Fig. 4a) suggests that HSA does not simply act by completely shielding A␤(1-40) protofibrils from A␤(1-40) monomers. Additional effects are likely present, such as binding of HSA to a subtype of protofibrils or protofibril-binding sites or HSAmediated contacts between A␤(1-40) protofibrils and A␤(1-40) monomers.
In plasma, a further contribution to the inhibition of A␤(1-40) protofibril growth is provided by the albumin-A␤(1-40) monomer (A␤40 1 ) interactions, which affect multiple residues in the two ␤-strands involved in A␤(1-40) self-recognition, similarly to the albumin-A␤(1-40) protofibril (A␤40 n ) interactions (Fig. 5a). However, our data also reveal distinct differences between the interactions of HSA with A␤40 n and A␤40 1 . In the case of the latter species, a key site for binding with HSA is the C-terminal region of A␤(1-40) (i.e. [31][32][33][34][35][36][37][38][39][40], as confirmed by HSA-induced 15 N R 2 and 1 H STD enhancements that progressively increase in going from residue 37 to 40 (Fig. 1c). On the contrary, in the case of A␤40 n , the albumin-induced ⍜ variations observed at the C-terminal region of A␤40 n progressively decrease in going from residues 37 to 40 (Fig. 3c). In this respect, the interactions of HSA with A␤40 1 appear to complement those with A␤40 n (Fig. 5a), and together they provide an exhaustive and efficient coverage of the A␤(1-40) residues in both ␤-strands of the cross-␤ fibrils, which fully span both the central hydrophobic core and the C-terminal region (Fig. 5a), rationalizing the high anti-amyloidogenic potency of albumin in plasma (2,3).
The balance between A␤ monomer versus protofibril interactions with HSA is subject to marked changes in going from the A␤  to the A␤(1-42) isoform (Fig. 5b). The A␤(1-42) monomers bind HSA more weakly than A␤(1-40) monomers primarily because of a loss of interactions in the C-terminal region (Figs. 1 and 5b). A possible explanation for this observation is the stabilization by residues Ile 41 and Ala 42 of a C-terminal turn centered at Gly 37 and Gly 38 in monomeric A␤(1-42) (33-45) (Fig. 5b). To the extent that these intramolecular A␤(1-42) interactions compete with HSA binding, they account for the reduced HSA affinity for A␤42 1 versus A␤40 1 . However, the loss of C-terminal interactions with HSA at the level of A␤(1-42) monomers is compensated by an enhanced effect of HSA on the C-terminal contacts in the A␤(1-42) protofibrils (Fig. 3). Unlike A␤ , in the case of A␤ , the HSA-induced shift from direct to tethered contacts extends to the very C-terminal residues (Figs. 3d and 5b).
In summary, for both A␤(1-40) and A␤ , the HSA interactions of the C-terminal A␤ region relies on a monomer versus protofibril compensation strategy; however, the relative monomer versus protofibril balance is A␤ isoform-specific. In the case of A␤(1-40), the C-terminal A␤ segment mediates primarily the HSA-A␤ monomer interactions, whereas in the case of A␤(1-42), the C-terminal A␤ segment is perturbed primarily by the HSA-A␤ protofibril interactions (Fig. 5). Overall, the model emerging from our data (Fig. 5) provides a framework to understand the physiological role of HSA as an endogenous anti-A␤ amyloid agent in the CSF and plasma. In addition, the mechanism of Fig. 5 also addresses and clarifies previous discrepancies about the affinities of albumin for A␤ monomers versus protofibrils.
The approximately submillimolar affinity range proposed here based on our NMR data (Fig. 1f) is in agreement with previous surface plasmon resonance results indicating that monomeric A␤(1-40) did not lead to detectable binding to HSA at 25 M concentrations, irrespective of how monomeric A␤(1-40) was immobilized on the sensor chip, i.e. through monoclonal antibodies or streptavidin (3). However, K d values in the submillimolar range are markedly higher than previously reported M affinities between A␤(1-40) monomers and HSA based on low-resolution techniques, such as CD and immunoassays (13, 46 -48). This apparent discrepancy is explained considering that previous determinations of A␤(1-40)-HSA affinities (46,47) may reflect also contributions from A␤(1-40) oligomers, which are known to bind HSA more tightly than A␤(1-40) monomers (9,11). This interpretation is corroborated by a recent surface plasmon resonance investigation showing that early A␤ oligomers in equilibrium with monomers are sufficient to lower the effective measured K d value to the M range (18). In addition, the presence of A␤(1-40) oligomers is supported by the observation that the binding isotherm built using CD data leads to Hill coefficients significantly greater than 1 (1.4 -1.5) (46), which cannot be explained simply by the 1:1 binding of HSA and A␤(1-40) monomers. Furthermore, contributions to the CD spectra from both A␤(1-40) and HSA are not easily deconvoluted, whereas this problem is solved here by selectively 15 N-labeling A␤(1-40), but not HSA (supplemental Fig. S2), and by using NMR approaches that directly report on monomer interactions. Overall, our data on both monomeric A␤(1-40) and A␤(1-42) rule out that HSA binds the monomeric forms of these A␤ peptides with approximately micromolar K d values and are consistent with affinities lower by ϳ2 orders of magnitude, in the submillimolar range (Fig. 1f).

Conclusions
Our data indicate that the inhibition of A␤(1-40) self-association by HSA relies on a dual mechanism, whereby at plasma A␤ peptides bind HSA via a dual mechanism concentrations albumin binds both A␤(1-40) monomers and protofibrils, targeting key A␤(1-40) self-recognition sites and triggering a net switch from direct to tethered contacts between monomeric and protofibrillar A␤ (Fig. 5). We also show that in going from the A␤(1-40) to A␤  isoform, the engagement of the C-terminal A␤ residues increases at the level of the A␤ protofibril-HSA contacts but decreases at the level of A␤ monomer-HSA interactions, which become weaker. These results provide an unprecedented view of the mechanism underlying the A␤ self-association by albumin. In addition, they also demonstrate the potential of a general NMR approach to probe at residue resolution how amyloid inhibitory proteins perturb A␤ self-association. Inhibitor-induced enhancements in 15 N R 2 rates and in 1 H STD-HSQC ratios unveil the sites of A␤ monomer interactions, whereas inhibitor-dependent changes in DEST under dark conditions, quantified as variations in the ⍜ observable, identify the A␤ protofibril residues affected by the inhibitor and, when combined with the analysis of dark versus dilute R 2 changes corrected for monomer interactions, provide a quantitative measure of how the inhibitor perturbs the direct versus tethered partitioning of the A␤ monomer-protofibril contacts. Hence, the DEST method (26) was essential to elucidate the inhibitory mechanism proposed here (Fig. 5).

Sample preparation
Peptide solutions were prepared as previously described (42). In brief, the A␤(1-40) or A␤(1-42) peptides (rPeptides Inc.) were dissolved in 3 mM NaOH at a concentration of 1 mg/ml and a pH adjusted to 11 using a 50 mM NaOH solution. Peptide solutions were then divided into different aliquots containing enough A␤ peptides to obtain 500-l final solutions at the desired A␤ peptide concentrations, i.e. 40 -60 and 300 M for dilute and dark A␤(1-40) samples, respectively, and 40 -50 and 150 M for dilute and dark A␤(1-42) samples, respectively. The aliquots were freeze-dried, and the lyophilized powder of each aliquot was resolubilized using 250 l of 3 mM Tris-HCl, pH 8.0, and buffer-exchanged with a Zeba desalting column pre-equilibrated with Tris buffer. Desalted solutions were adjusted to 50 mM HEPES buffer, 10% D 2 O, pH 6.8, by the addition of 250 l of 100 mM HEPES buffer, 20% D 2 O, pH 6.3, and up to additional 30 l of 100 mM HEPES buffer, pH 6.8, as necessary to fine tune the pH. A 1 mM HSA stock solution was prepared by dissolving lyophilized HSA, essentially fatty acid, and globulin free (A3782; Sigma), in 50 mM HEPES buffer, 10% D 2 O, pH 6.8. All solutions were treated with a chelating agent (Chelex 100; Sigma-Aldrich) to remove any residual metals. In all experiments involving concentrated A␤ samples (i.e. 300 M A␤  or 150 M of A␤(1-42)), the 15 N-R 2 relaxation and DEST experiments were acquired after 6 days of incubation at 10°C, and then HSA was added to an aliquot of the concentrated sample (ϳ7.5 days after sample preparation). Directly after HSA addition, DEST and 15 N-R 2 relaxation experiments were acquired again. The incubation at 10°C for 6 days was essential to reach a pseudo-equilibrium state between monomer and protofibrils in the concentrated samples. In addition, a 3 mM HSA stock solution was prepared specifically for the HSQC monitored titration. The HSA concentration was confirmed by measuring the absorbance at 280 nm using an extinction coefficient of 34,445 M Ϫ1 cm Ϫ1 (42,43). The A␤ peptide concentration was determined by UV absorption at 280 nm with an extinction coefficient of 1,490 M Ϫ1 cm Ϫ1 (42,43). Details about NMR, EM, and DLS data acquisition and processing are available in the supplemental text.