The Solvent Protection of Alzheimer Amyloid-β-(1–42) Fibrils as Determined by Solution NMR Spectroscopy*

Alzheimer disease is a neurodegenerative disorder that is tightly linked to the self-assembly and amyloid formation of the 39–43-residue-long amyloid-β (Aβ) peptide. Considerable evidence suggests a correlation between Alzheimer disease development and the longer variants of the peptide, Aβ-(1–42/43). Currently, a molecular understanding for this behavior is lacking. In the present study, we have investigated the hydrogen/deuterium exchange of Aβ-(1–42) fibrils under physiological conditions, using solution NMR spectroscopy. The obtained residue-specific and quantitative map of the solvent protection within the Aβ-(1–42) fibril shows that there are two protected core regions, Glu11-Gly25 and Lys28-Ala42, and that the residues in between, Ser26 and Asn27, as well as those in the N terminus, Asp1-Tyr10, are solvent-accessible. This result reveals considerable discrepancies when compared with a previous investigation on Aβ-(1–40) fibrils and suggests that the additional residues in Aβ-(1–42), Ile41 and Ala42, significantly increase the solvent protection and stability of the C-terminal region Lys28-Ala42. Consequently, our findings provide a molecular explanation for the increased amyloidogenicity and toxicity of Aβ-(1–42) compared with shorter Aβ variants found in vivo

. The mechanisms by which a cytotoxic effect is exerted in vivo and the reasons why a pathologic self-aggregation is induced in certain individuals are complex and at present not completely understood. Prevention of A␤ assembly therefore constitutes a considerable therapeutic challenge, where an increased understanding regarding the properties of amyloid, and the pathways leading to its formation, is of utmost importance.
Because of the generic quaternary structure and the large size of amyloid structures, elucidation of their architecture provides a complicated problem, where traditional methods, such as crystal diffraction and solution NMR, are not readily applicable. However, two recent crystallographic studies on fibrous micro-crystals, grown from peptides with either seven or twelve residues, have revealed many interesting structural details about the cross-␤ spine of fibrils (10,11). Solid-state NMR performed on dried fibrils provides an alternative to the above mentioned techniques and has been used extensively to investigate the structure of A␤ amyloid. Studies on A␤- (10 -35), A␤- , and A␤-  suggest an arrangement where fibrils are formed by extended parallel ␤-strands arranged into two sheets (12)(13)(14). The results suggest a fibrillar core involving residues Val 12 -Val 24 and Ala 30 -Ala 42 and a loop spanning between residues Val 24 -Ala 30 (15). The structural restraints obtained from these solid-state NMR studies suggest a model where parallel ␤-sheets are organized in exact register (16). Other models have been proposed (9), among them ␤-helical (17) and nanotube (18)

models.
To gain information about the structural and dynamical properties of fibrils in an aqueous environment, a novel method using solution NMR spectroscopy in combination with hydrogen/deuterium (H/D) exchange was developed (19,20). This technique relies on the partial solvent protection of amide protons, either as a result of their involvement in hydrogen bonds between ␤-strands within the fibril or as a result of solvent exclusion. In aqueous solutions, amide protons located at the exterior of the fibril are more accessible to solvent compared with amide protons buried within the fibril interior, and consequently these will experience a higher hydrogen exchange rate. Through exchange of the surrounding water by D 2 O, followed by a rapid conversion of the fibrils into a monomeric NMR-detectable state, the H/D exchange pattern of the amyloid can be measured indirectly using solution NMR. The method pinpoints the fibrillar core in a residue-specific and quantitative manner and gives a view of the fibrillar dynamics in solution. This approach has therefore turned out to be a powerful complement to solid state measurement and has offered a more complete picture of the fibrillar properties of the amyloid in several cases (19 -25).
This study utilizes H/D exchange experiments and solution NMR spectroscopy to probe, for the first time, the fibrillar core of amyloid fibrils derived from A␤- . The obtained solvent protection pattern identifies two well protected regions, a result that differs considerably from a similar study on A␤-(1-40) (21). The additional residues, Ile 41 and Ala 42 , appear to significantly stabilize the fibril assembly, in particular the C-terminal parts of the peptide. These results provide a molecular explanation to the increased amyloidogenicity of A␤-  compared with the shorter variants of A␤ found in vivo.
H/D Exchange of A␤-  Fibrils Analyzed by NMR-The fibrillar solution was prepared by incubating a sample of ϳ1 mM 15 N-labeled A␤-  in double distilled H 2 O containing 50 mM NaCl at 37°C for 10 -12 days during slow agitation. The resulting gel-like solution was split into three samples, and the pellet was recovered by short centrifugations at 13000 ϫ g. For two of the samples, the pellet (30 l) was diluted 30 times using D 2 O, pD 7.0. To ensure an essentially complete removal of H 2 O, the washing procedure was repeated once. These two samples were subsequently incubated in D 2 O for 20 and 120 min, respectively, which includes the duration for the buffer exchange procedure. The third sample contained fully protonated fibrils and served as a control. The pelleted fibrils in the three samples were rapidly dissolved into monomers in 80% HFIP/20% D 2 O (v/v) and 150 mM NaCl, buffered to pD 2.6 using d 3 -acetate and d 6 -acetic acid. Each sample acquired a peptide concentration corresponding to ϳ1 mM. Hydrogen exchange of each individual amide group of the A␤-(1-42) peptide was monitored by recording a series of heteronuclear two-dimensional 15 N-HSQC experiments, starting 7 or 8 min after dissolution of the fibrils. Each experiment was recorded with four transients/increment and 128 (t 1 ) ϫ 1024 (t 2 ) complex data points, resulting in an acquisition time of 10 min. To quantitatively monitor dissolution of the fibrils into a monomeric species, the integrals of non-exchangeable methyl groups were determined in one-dimensional proton spectra, recorded prior to each 15 N-HSQC spectrum. The H/D exchange experimental series on the fully protonated A␤-(1-42) fibrils determined the solvent protection levels of amide protons within the monomeric species. With residual H 2 O present in the pellet, the exchange to deuterium reached ϳ80%, a fact that is taken into account in the analysis.
Data Analysis and Fitting-Processing of the recorded 15 N-HSQC experiment was carried out with NMRPipe (26), whereas XWINNMR (Bruker Biospin) was used for processing and analysis of one-dimensional spectra. To monitor the post-trap H/D exchange after dissolution of the partly exchanged A␤-(1-42) fibrils in D 2 O, peak volumes from base-line-corrected 15 N-HSQC spectra were obtained by integration using NMRView software routines (28) and analyzed by non-linear fitting routines in the software Grace (plasma-gate.weizmann.ac.il/ Grace). The monomer concentration of the solution was determined from integration of the non-exchangeable methyl region in one-dimensional proton NMR spectra, which were recorded directly prior to each individual 15 N-HSQC spectrum. The integral (I) of the amide resonance of each protected residue in the monomeric state was fitted to a single exponential function, I(t) ϭ (I 0 Ϫ I inf )⅐e [Ϫksolv(monomer)⅐t] ϩ I inf , where I 0 is the extrapolated proton integral at time zero, k solv is the post-trap H/D exchange rate, and I inf is the integral at infinity. I 0 describes the degree of protonation of amide groups that survived the deuterium exchange period with D 2 O in the fibrillar state prior to re-solubilization. The actual fraction of protection within the fibrillar state is calculated by a comparison of corresponding peak integrals in a similarly treated but fully protonated reference sample. By definition, the relative ratio between I 0 in the two spectra is 100% in the absence of any H/D exchange process and 0% under conditions of complete H/D exchange.
Atomic Force Microscopy-An aliquot of the A␤-(1-42) fibril solution used for H/D exchange experiments was withdrawn and diluted to 5 M with double distilled H 2 O. 10 l of this sample was applied onto freshly cleaved ruby red mica (Goodfellow, Cambridge, UK). The material was further allowed to adsorb for 30 s and then washed three times with distilled water and air dried. The surface was analyzed with a Nanoscope IIIa multimode atomic force microscope (Digital Instruments) using Tapping Mode TM in air. A silicon probe was oscillated at 258 kHz, and images were collected at an optimized scan rate corresponding to 1-4 Hz. The image was flattened and presented in height mode using Nanoscope software (Digital Instruments).
Structure Modeling-A model of the fibrillar structure of A␤-(1-42) was generated from the coordinates of the structural model of the A␤-(1-40) fibril kindly provided by Dr. Robert Tycko (16). The monomeric structure was modified manually in MOLMOL (29) by adding the missing N-and C-terminal residues, followed by energy minimization in Swiss-PdbViewer, version 3.7 (30), using the GROMOS96 43B1 force field.

Selecting a Suitable Solvent for Dissolution and NMR Analysis of A␤-(1-42) Fibrils-
The ability to quickly convert the fibrils into a monomeric state is essential for NMR detection of protected amide protons within the fibril. During this process, the H/D exchange pattern trapped by the secondary structures of the fibrils is transferred to and must be preserved in the monomeric state. By performing the experiments at a pH value of ϳ3 and at low temperatures, amide proton exchange rates could be minimized. Furthermore, an induction of secondary structure in the monomer considerably reduced the amide proton exchange rates via the formation of new hydrogen bonds. We found that a mixture of 80% HFIP/20% water (v/v) at pH 3 had the capability of rapid dissolution of A␤-(1-42) fibrils. In a previous investigation, performed under very similar conditions, the peptide formed a well structured conformation with two ␣-helices, comprising residues Ser 8 -Gly 25 and Lys 28 -Gly 38 and a bend centered at position Ser 26 -Asn 27 (31). Circular dichroism measurements showed a further increase of ␣-helical structure with an increasing concentration of NaCl up to a nearly saturated solution of ϳ150 mM salt (data not shown). The observed 15 N-HSQC spectrum of the A␤-(1-42) peptide in this HFIP/water solution and 150 mM NaCl, pH 3.0, displayed a well dispersed spectrum with an optimal separation of peaks at 15°C, a temperature at which the hydrogen exchange rate is relatively slow.
NMR Resonance Assignment of A␤-  in Solution-Sequencespecific resonance assignment turned out to be straightforward with few overlapping resonances. Most resonances were assigned via a sequential walk between the backbone amide protons in the nuclear Overhauser effect spectroscopy spectrum or via sequential or medium range nuclear Overhauser effect contacts, characteristic for ␣-helices. All 41 backbone amide resonances (residue 2-42) were identified. Our data are generally in agreement with the previously reported assignment and structure of A␤-(1-42) performed in HFIP (31).
Fibril Formation and Atomic Force Microscopy Analysis of A␤-(1-42)-Fibril formation was induced by dissolving ϳ1 mM 15 Nlabeled A␤-  in double distilled H 2 O containing 50 mM NaCl, followed by incubation at 37°C for 10 -12 days during slow agitation. The initially non-viscous solution acquired a gel-like appearance, and the presence of fibrils was verified using atomic force microscopy (Fig. 1). The fibrils formed were of varying length usually exceeding 1 m in length and with an average height corresponding to 3.5 nm. A repetitive pattern of nodules could also be observed, indicating a twist of the fibril with an average pitch of ϳ110 nm.

H/D Experiments and Determination of Protection Factors of A␤-(1-42) Fibrils-
The fibrillar material was easily collected from the gel-like fibril solution by centrifugation. H/D exchange was carried out by resuspension and incubation of the fibrillar pellet in D 2 O, followed by re-collection of fibrils using centrifugation. This procedure was repeated once to efficiently remove water (residual water, ϳ0.1% (v/v)) and non-aggregated material. After a total exchange period of either 20 or 120 min, the fibrils were converted into NMR-detectable monomers by dissolution in 80% HFIP/20% D 2 O (v/v) and 150 mM NaCl at pD 2.6. H/D exchange in the monomeric state was monitored by recording a series of 15 N-HSQC spectra over time. Fig. 2A shows a spectrum of the fully protonated peptide, whereas the spectra in Fig. 2, B and C, were recorded 12 and 196 min, respectively, after dissolution of the H/Dexchanged fibril material. The observed signal intensity of each amide resonance is determined by the level of solvent protection in the fibrillar state, the concentration of monomers, the rate for which fibrils are converted into monomers, and the H/D exchange rate in the mono-meric state. By recording one-dimensional proton NMR spectra prior to each 15 N-HSQC spectrum and integrating the non-exchangeable methyl region of these spectra, the dissolution rates, as well as the relative concentration between samples, could be determined (data not  shown). Our results showed that Ͼ94% of the total material was dissolved prior to the first recorded spectrum, and a correction was made by fitting a single exponential with a rate constant of 0.0015 min Ϫ1 . As a consequence of the H/D exchange, the intensity of the amide resonances will decline with time (Fig. 2). The signal decay rates of each individual amide resonance were obtained by fitting the corrected signal intensities to the equation under "Data Analysis and Fitting." An extrapolation to time zero of fibril dissolution gives the signal intensity (I 0 ) of each residue in the fibrillar state. Fig. 3 illustrates the curve fit of the post-trap decay for residues Ile 31 , Val 39 , and Ile 40 . This treatment enables identification of solvent-protected residues within the fibril and facilitates quantification of the protection level. The protection factors of individual residues are determined by calculating the ratio of the initial signal intensities obtained from samples pre-incubated in D 2 O and a fully protonated reference sample. Fig. 4 shows the obtained protection factors, which will be discussed below.
To obtain a reliable H/D protection pattern, one must discriminate between the H/D amide proton exchange occurring as a result of the preceding incubation in D 2 O and H/D amide proton exchange occurring as a result of an unprotected position within the monomeric state. Therefore, the recorded series of 15 N-HSQC spectra, which monitor the H/D exchange in fully protonated fibrils (no pre-incubation in D 2 O) serves as a highly important reference. The result showed that 35 of 41 possible amide protons can be used for probing the solvent protection pattern of the fibril. Residues Ala 2 , His 6 , Asp 7 , Ser 8 , His 14 , and Asp 23 experienced an H/D exchange rate within the monomeric state, which is too fast to enable detection and thereby makes their level of protection within the fibril unattainable. This lack of protection is because of the solvent-exposed positions of these residues in the monomeric fold, either in the flexible N-terminal tail or, as in the case of Asp 23 , its position close to a turn.
The solvent protection pattern of the fibrils was determined from fibrillar samples that were pre-incubated in D 2 O for 20 or 120 min (Fig.  4). By using two different pre-incubation times, valuable information about the solvent exchange rates and structural stability of the fibrils could be obtained. Our results showed that, after a pre-incubation of 20 min, a total of 28 amides were observed, showing a protection pattern essentially covering two regions of the fibril. Both regions showed a slightly bell-shaped protection curve, where the first covered residues Glu 11 -Gly 25 , except for residues His 14 and Asp 23 , and the second region covered residues Lys 28 -Ala 42 with a very high degree of protection. Residues Glu 3 , Phe 4 , Arg 5 , Gly 9 , Tyr 10 , Ser 26 , and Asn 27 were all confirmed to be unprotected within the fibrillar state (Fig. 4). After 2 h of preincubation in D 2 O, the protection pattern was essentially the same for a majority of residues. However a slight decrease of protection was observed in the flanking residues of the first region (residues Glu 11 -Gly 25 ), and one residue, Val 12 , was now completely exchanged. Fig. 5 shows the obtained protection factors mapped onto our model of the A␤-(1-42) fibril.

DISCUSSION
The A␤ peptide is the main protein component of plaques found in patients with the neurodegenerative disorder Alzheimer disease. Production of A␤ is the result of proteolytic digestion of the significantly larger amyloid precursor protein resulting in a length of the peptide corresponding to 39 -43 residues. Most of the amyloid precursor protein mutations that ultimately result in early onset of AD have now been linked to an increased concentration of A␤ peptides either via an enhanced expression (6) or through an increased proteolytic processing (32,33). The aggregation propensity among A␤ fragments of different length differs significantly, and AD has moreover been linked to an increased proportion of longer variants, in particular the highly amyloidogenic 42-residue form (7,8). Senile or neuritic plaques in AD brains are further enriched with the A␤-(X-42/43) variants, which also represent the initial deposits found in vivo (5). A recent study suggests a pivotal role for A␤-(1-42) concerning development of parenchymal and vascular amyloid deposition in mice (34). Currently, the increased aggregation propensity for the longer variants, in particular A␤-(1-42), has not been explained on a molecular level.
The present study describes how hydrogen exchange in combination with solution NMR spectroscopy was used to determine the solvent protection of A␤-(1-42) fibrils formed under physiological pH, ion concentration, and temperature. The results suggest that the protected core of the fibril covers two regions of the A␤-(1-42) sequence, spanning residues Glu 11 -Gly 25 and Lys 28 -Ala 42 . In between these protected regions, two completely solvent-accessible residues, Ser 26 and Asn 27 , are found. Moreover, residues Glu 3 , Phe 4 , Arg 5 , Gly 9 , and Tyr 10 are exposed, which suggest that the N-terminal part of the peptide is solvent-accessible. As expected, the observed protection pattern of each protected region has a slightly bell-shaped appearance (Fig. 4), consistent with a decrease of solvent accessibility toward the core of the fibril. An extended pre-incubation time of the fibrils in D 2 O (from 20 to 120 min) caused no significant change in the protection of the core residues but affected the flanking regions to some extent, in particular Val 12 , which becomes completely exchanged (Fig. 4). These results agree well with those obtained in previous mass spectrometry studies, where the kinetics of the H/D exchange of backbone amide protons in A␤-  were investigated (35,36). The possibility of obtaining residue-specific information about quantitative solvent protection levels and dynamics of amyloid fibrils is unique for the method and was initially described in our study on a short A␤ fragment (19,20).
Overall, our findings are in good agreement with solid-state NMR studies on A␤-(10 -35), A␤-(1-40), and A␤- , where an in-register parallel arrangement of residues Val 12 -Ala 42 were proposed (12-14). Furthermore, solid-state NMR data on A␤-(1-40) suggest that the region comprising residues Val 24 -Ala 30 lacks ␤-strand conformation (15). This supports our results indicating that Ser 26 and Asn 27 are completely exposed within the fibril and only partial protections of residues Gly 25 and Lys 28 are observed. The outcome of our investigation also supports the model of Tycko and co-workers (16). In their model, the peptide forms a so-called "cross-␤ unit" defined as two extended ␤-strands brought together via a bend in position Gly 25 -Gly 29 . In addition, parallel in-register cross-␤ structure is formed via intermolecular interactions, and the appropriate cross-sectional dimensions are fulfilled by juxtaposition of two cross-␤ units (16). Based on this model and our results, we have generated a new model for A␤-(1-42) (Fig. 5). The overall high protection factor observed for the C-terminal part of the peptide (residues Lys 28 -Ala 42 ), as compared with the region comprising residues Glu 11 -Gly 25 , is readily explained by a structural arrangement where two cross-␤ units associate via their hydrophobic C-terminal part, forming a less solvent-exposed core region of the fibril (Fig. 5, A  and B). The width of the cross-section of the fibril model is ϳ4 nm, which is in accordance with our morphological data (Fig. 1). The atomic force microscopy image of A␤-(1-42) fibrils suggests a twisted architecture and an average height between nodules measured to 3.5 nm, in good agreement with the fibril model suggested (Fig. 5C).
A comparison of our results on A␤-(1-42) fibrils, with the results from a recent H/D exchange study on the A␤-(1-40) fibril, show both similarities as well as interesting discrepancies. The N-terminal part, residues Asp 1 -Lys 28 of both A␤-(1-42) and A␤-(1-40), displays an essentially identical solvent protection, where position Glu 11 is partly protected and positions Gln 15 -Gly 25 show a bell-shaped protection pattern. Both peptides also lack a significant solvent protection for Ser 26 and Asn 27 , further supporting previous investigations suggesting a loop within this region of the fibril (15). However, striking discrepancies are observed in the C-terminal part of the peptides, starting from Gly 29 to the C terminus. Our result suggests almost complete solvent protection at positions Gly 29 -Ala 42 (Fig. 4), whereas the study carried out on A␤-(1-40) is suggestive of a partly unprotected C-terminal end with only residues Lys 28 , Ala 30 , Ile 32 , Leu 34 , and Met 35 significantly protected (21). Although their study utilizes a longer incubation time in D 2 O (25 h), a direct comparison is justified, because previous H/D exchange mass spectrometry studies on A␤-(1-40) fibrils have shown that the bulk of solvent-accessible backbone protons exchange well before 2 h (35), and only a minor average difference is observed (0.9 protons) between 2 and 24 h (36). Similarly, the protected core of the A␤-(1-42) fibril in our study was formed already at an incubation time of 20 min with minor changes at 2 h. Consequently, the differences observed provide strong evidence for two different fibrillar architectures in the C-terminal part of A␤-(1-42) and A␤- . A less structured C-terminal end of A␤-(1-40) is further supported from studies of proteolytic fragmentation in combination with mass spectroscopy (37) as well as by proline scanning of the A␤ sequence (38). Similar indications are reported in a recent solid-state NMR study, where no hydrogen bond distance information was obtained for residues Gly 29 , Gly 37 , and Gly 38 (39). The highly protected C-terminal region within A␤-(1-42) fibrils observed in the present study suggests that the addition of two C-terminal residues, Ile 41 and Ala 42 , in a significant manner increase the structure and stability in the whole C-terminal part of the peptide, possibly acting as a molecular zipper between the cross-␤ units along the fibril axis. The additional protection of at least 6 residues in the C-terminal region of A␤-(1-42) suggests that equally many intermolecular hydrogen bonds are formed, significantly shifting the equilibrium from a monomeric species toward an oligomeric assembly. Consequently, this finding provides a molecular explanation as to why A␤-(1-42) has a stronger propensity for fibril formation. Furthermore, it offers a likely explanation regarding the suggested pivotal role for A␤-  in the development of AD (5,34).
The observed discrepancies in solvent protection for the C-terminal parts of A␤-(1-42) and A␤-(1-40) may require a reconsideration of how the filaments within the fibril model of A␤-  assemble. An alternative model that better satisfies the H/D exchange results on A␤-(1-40) is created by juxtaposing two cross-␤ units in such a way that their N-terminal ␤-strands (residues Gln 15 -Gly 25 ) form the central core, whereas their C-terminal ␤-strands (residues Gly 29 -Val 40 ) face the solvent, thereby explaining the increased accessibility of the latter residues in A␤-(1-40) as compared with A␤- . In such an arrangement, the two cross-␤ units may be oriented either parallel or antiparallel with respect to each other. Of these, the second alternative seems preferable, as it creates favorable intermolecular electrostatic side-chain interactions between residues Lys 16 and Glu 22 in the core. Such an alternate lateral arrangement was previously presented together with the original model (16) but may play a much more important role than was previously assumed.
The basis for how A␤ species exert their toxic effect in vivo is a The protection is defined as the ratio of the observed intensity after a pre-incubation period in D 2 O over the intensity in a completely protonated sample (defined as 100%). Dark and light gray bars indicate the protection after 20 and 120 min of pre-incubation in D 2 O, respectively. # corresponds to 0% protection, and x represents residues that exchange too fast for detection. The error bars indicate the experimental uncertainty of the measurements. Residue names are indicated with their one-letter codes.
complex and not yet fully understood subject, although it appears that some form of peptide assembly is an inevitable step (39,40,41). Much attention has been focused on the recently discovered small globular oligomers of A␤, denoted ADDLs, which have been shown to exert a highly toxic effect (42). Interestingly, the propensity of ADDL formation varies significantly between different forms of A␤ peptides. Although the A␤-(1-40) variant has the ability to form oligomeric species (42), formation of the highly toxic ADDLs seems to be restricted to the A␤-(1-42) variant (43,44). Even though no structural information is available concerning ADDLs, it is likely that the peptide assembly, at least in part, shares similarities to the fibrillar state. The observation of a highly stabilized C-terminal region upon assembly of A␤-(1-42) compared with A␤-(1-40) may hence provide additional molecular clues to the formation of ADDLs. Interference with the most C-terminal part of A␤-(1-42) may thereby result in the inability to form ADDLs.
In conclusion, we have determined the solvent protection pattern of fibrils from the highly amyloidogenic A␤-(1-42) variant on a residuespecific basis, identified the fibrillar core residues, and attained information about the dynamical properties of the fibril. By comparing our result to previous studies on A␤-(1-40), our results revealed interesting discrepancies in the core structure and suggest that Ile 41 and Ala 42 play an important role for stabilization of the C-terminal residues from Gly 29 and forward. This finding may further provide a molecular explanation regarding their different propensity of aggregation and fibril formation and provide a novel target for drug design.