Distinct Early Folding and Aggregation Properties of Alzheimer Amyloid-β Peptides Aβ40 and Aβ42

The amyloid β peptide (Aβ), composed of 40 or 42 amino acids, is a critical component in the etiology of the neurodegenerative Alzheimer disease. Aβ is prone to aggregate and forms amyloid fibrils progressively both in vitro and in vivo. To understand the process of amyloidogenesis, it is pivotal to examine the initial stages of the folding process. We examined the equilibrium folding properties, assembly states, and stabilities of the early folding stages of Aβ40 and Aβ42 prior to fibril formation. We found that Aβ40 and Aβ42 have different conformations and assembly states upon refolding from their unfolded ensembles. Aβ40 is predominantly an unstable and collapsed monomeric species, whereas Aβ42 populates a stable structured trimeric or tetrameric species at concentrations above ∼12.5 μm. Thermodynamic analysis showed that the free energies of Aβ40 monomer and Aβ42 trimer/tetramer are ∼1.1 and ∼15/∼22 kcal/mol, respectively. The early aggregation stages of Aβ40 and Aβ42 contain different solvent-exposed hydrophobic surfaces that are located at the sequences flanking its protease-resistant segment. The amyloidogenic folded structure of Aβ is important for the formation of spherical β oligomeric species. However, β oligomers are not an obligatory intermediate in the process of fibril formation because oligomerization is inhibited at concentrations of urea that have no effect on fibril formation. The distinct initial folding properties of Aβ40 and Aβ42 may play an important role in the higher aggregation potential and pathological significance of Aβ42.

The amyloid ␤ peptide (A␤), composed of 40 or 42 amino acids, is a critical component in the etiology of the neurodegenerative Alzheimer disease. A␤ is prone to aggregate and forms amyloid fibrils progressively both in vitro and in vivo. To understand the process of amyloidogenesis, it is pivotal to examine the initial stages of the folding process. We examined the equilibrium folding properties, assembly states, and stabilities of the early folding stages of A␤40 and A␤42 prior to fibril formation. We found that A␤40 and A␤42 have different conformations and assembly states upon refolding from their unfolded ensembles. A␤40 is predominantly an unstable and collapsed monomeric species, whereas A␤42 populates a stable structured trimeric or tetrameric species at concentrations above ϳ12.5 M. Thermodynamic analysis showed that the free energies of A␤40 monomer and A␤42 trimer/tetramer are ϳ1.1 and ϳ15/ϳ22 kcal/mol, respectively. The early aggregation stages of A␤40 and A␤42 contain different solvent-exposed hydrophobic surfaces that are located at the sequences flanking its protease-resistant segment. The amyloidogenic folded structure of A␤ is important for the formation of spherical ␤ oligomeric species. However, ␤ oligomers are not an obligatory intermediate in the process of fibril formation because oligomerization is inhibited at concentrations of urea that have no effect on fibril formation. The distinct initial folding properties of A␤40 and A␤42 may play an important role in the higher aggregation potential and pathological significance of A␤42.
Alzheimer disease (AD) 2 is a progressive neurodegenerative disease occurring in the elderly resulting in the accumulation of misfolded proteins and cognitive dysfunction (1). The major pathological hallmarks of AD are the deposition of senile plaques and accumulation of neurofibrillary tangles in specific regions of the brain. The plaque deposits consist predominantly of insoluble amyloid fibrils formed by amyloid-␤ peptide (A␤). The mutations associated with inherited forms of AD provide strong evidence that the aggregation of A␤42 is a causative factor in etiology of AD because the mutations increase the relative amount of A␤42 (2)(3)(4). A growing body of evidence indicates that prefibrillar oligomeric forms of A␤ may represent the primary pathological species and not the mature amyloid fibrils that accumulate in plaque deposits (4,5).
The A␤ peptide is generated from limited proteolysis from a type I transmembrane protein, amyloid precursor protein (APP) (6). The two most common isoforms of A␤ are A␤40 and A␤42, which vary by the length of the C terminus. Although secreted A␤40 is much more abundant, A␤42 is the major component in senile plaques (7,8). Biochemical studies show that A␤42 aggregates and forms fibrils more rapidly than A␤40 (9,10). Recent studies have demonstrated that overexpression of high levels of A␤40 alone do not result in overt amyloid pathology transgenic mice; however, expression of low levels of A␤42 results in a broad range of amyloid pathology (11). Therefore, A␤40 and A␤42 possess different biochemical properties and A␤42 is believed to be the major etiologic agent in pathogenesis of Alzheimer disease due to its enhanced aggregation or oligomerization properties.
The aggregation or oligomerization of A␤ has been the subject of numerous studies employing a variety of experimental approaches. Early studies indicated that A␤42 forms non-covalent, SDS-resistant species with apparent molecular weights of dimer, trimer and/or tetramer, whereas under the same conditions A␤39 and A␤40 migrate as a monomeric species (9,12,13). However, gel filtration analysis under physiological conditions indicates that the smallest A␤40 and A␤42 species co-elute at a position corresponding to an apparent molecular weight expected for a dimer (12,13). Fluorescence resonance energy transfer experiments suggest that low molecular weight A␤40 exists as a dimer (14), whereas NMR diffusion measurements suggest that it is primarily monomeric (15). Photochemical oxidative cross-linking studies suggest that monomer, dimer, trimer, and tetramer species of A␤40 exist in a rapid equilibrium (16). In contrast, the same method applied to A␤42 suggests that pentameric or hexameric aggregates are preferentially formed that represent a fundamental building block for higher order assembly (17). Although some of these results appear inconsistent, different experimental conditions and analytical methods could conceivably account for the inconsistencies.
Although extensive research has been done on the structural and kinetic properties of A␤ fibrillogenesis, the thermodynamic and initial folding properties have not been studied in detail. Unlike some amyloidogenic proteins that aggregate rapidly, seed-free A␤ at low concentration exhibits "nucleationdependent" kinetics with a significant lag phase on the order of hours to days to form fibrils (10). Taking advantage of this property, we examined the equilibrium folding properties of early species of A␤40 and A␤42 and determined their experimental energy values. We found the early species of A␤40 and A␤42 are different not only in tertiary and secondary structures but also in thermodynamic stability.

EXPERIMENTAL PROCEDURES
Peptide Synthesis-A␤ peptides and the short peptides indicated in text were synthesized by using fluorenylmethoxycarbonyl solid phase chemistry using a continuous flow semiautomatic instrument described previously (9). Peptides were purified by reverse phase high performance liquid chromatography, and the purity was analyzed by matrix-assisted laser desorption ionization mass spectrometry. Purified peptides were dissolved in 50% acetonitrile/water, aliquotted, and relyophilized.
Sample Preparation-Folded A␤ peptides were prepared as follows except where indicated. A denatured stock solution was prepared by dissolving lyophilized peptides at a concentration of 10 mg/ml in buffer containing 10 mM sodium phosphate, pH 7.4, and 10 M urea freshly before use. The urea-containing buffer was prepared as described (18) by weight measurement and refractive index. The denatured peptide stock in 10 M ureacontaining buffer was centrifuged at 15,000 ϫ g for 10 min at room temperature. The desired amount of denatured A␤ in 10 M urea was added into buffer A (10 mM sodium phosphate, pH 7.4) for refolding. The sample was immediately vortexed and centrifuged at 15,000 ϫ g for 10 min at room temperature and filtered through a 0.22-m filter. The supernatant was collected, and the concentration of A␤ was determined by UV absorption at 280 nm (⑀ ϭ 1280 cm Ϫ1 M Ϫ1 ) based on the Edhoch equation (19). The refolding yield calculated by absorbance before and after centrifugation is generally 80 -100%.
Native and SDS Polyacrylamide Gel Electrophoresis (PAGE)-Tris-Tricine Ready Gels, 16.5%, (Bio-Rad) without SDS were used for monitoring the A␤ species by native electrophoresis. A␤40 and A␤42 at 50 M were prepared as described above. Native PAGE was electrophoresed at 100 V for 5-6 h at 4°C. The samples were not heated, and the sample buffer does not contain SDS or ␤-mercaptoethanol. The gel was transferred to a nitrocellulose membrane, and the membranes were stained by direct blue 71 (Sigma) (20). The same samples were also examined by SDS-PAGE followed by Western blotting. SDS-PAGE was performed at room temperature with the presence of SDS and ␤-mercaptoethanol in the running and sample buffers. The samples were preheated. The nitrocellulose membrane after transferring was stained with direct blue and followed by Western blotting with A␤-specific antibody 6E10.
Fluorescence Emission and Circular Dichroism Spectroscopy-Fluorescence emission spectra were obtained with a SPEX Fluorolog spectrofluorometer (Jobin-Yvon). The fluorescence of Bis-ANS (4,4Ј-dianilino-1,1Ј-binaphthyl-5,5Ј-disulfonic acid dipotassium salt; Sigma) was obtained by exciting the samples at 400 nm in buffer A containing 5 M Bis-ANS and different urea concentrations as indicated. The fluorescence emission from 450 to 600 nm was monitored. CD spectroscopy was done using a Jasco Model J-800 (Jasco Inc.) spectropolarimeter for the far-UV CD spectra. The samples were in buffer A containing different urea concentrations as indicated. The samples were placed in a cylindrical quartz cell (Hellma) with the path length of 1 cm. The spectral data were collected from 250 nm and below.
Urea Denaturation-Both fluorescence and CD were employed to monitor the structural changes of A␤ upon urea denaturation. Three concentrations of A␤, 50, 25, and 12.5 M, were examined. A titration method was performed by titrating the unfolded protein stock in 8 M urea-containing buffer A to the folded protein stock in Ͻ0.2 M urea with a constant protein concentration. The folded protein stock was prepared by refolding the unfolded peptide in buffer A through rapid dilution of urea. Reversed titration was performed with titration of the folded protein stock to the unfolded stock. The final urea concentration ranged from 0.08 to 8 M urea depending on the experimental design. The experimental time is less than 1 h for one set of denaturation experiments. For urea denaturation by Bis-ANS fluorescence, the emission wavelength from 450 to 600 nm was monitored. The emissions at 508 nm for A␤40 and at 500 nm for A␤42 were collected, normalized, and plotted against urea concentration. For urea denaturation by far-UV CD, the signals at 220 nm were collected every 0.5 s for 60 s. The data were averaged, normalized, and plotted against urea concentration. Both Bis-ANS and CD data were normalized using the native signals in the lowest urea concentration as unity. The experiments were performed at 25°C.
Data Fitting and Analysis-The denaturation data were plotted and fit by KaleidoGraph 3.0 (Synergy Software) or Igor Pro 5 (Wave Matrix Inc.). The data of the denaturation of A␤40 were fit to a two-state mechanism (N % U) described by Santoro and Bolen (21). Briefly, total signal is contributed from fractions of the folded species N, f N , and the unfolded species U, f U with their signal properties, P. Slopes for pre-and post-transition are also incorporated into Equation 1.
The sum of the folded and unfolded fraction is 1, For a two-state mechanism with a folded monomer and an unfolded monomer (N % U) the equilibrium constant between N and U, K UN , as shown in Equations 2 and 3, is and where R is the gas constant, T is the absolute temperature, and ⌬G UN is the free energy between N and U states in the presence of denaturant. Because ⌬G UN is dependent on denaturant concentration linearly, Equation 4 describes the relationship between the free energy and denaturant.
The fittings were done with Equation 1 after incorporating the Gibbs free energy of unfolding in the absence of denaturant, The solvent-exposed surface area (⌬ASA) and the estimated number of residues can be obtained from Equations 5 and 6 (22).
For two-state mechanism with a trimeric species to unfolded monomeric species, Tri % 3U, or tetrameric species to unfolded species, Tetra % 4U, the equilibrium constants are expressed as Equation 7 and Equation 8, respectively, where P t is the total subunit concentration. By solving the thirdorder equation of f U for the trimer model (23,24), we obtain the solution as follows in Equations 9 and 10.
For the tetramer model, we obtain the physically meaningful root as follows by solving the fourth-order equation of f U as shown in .
The fitting of the denaturation data of A␤42 for the trimer model was done by Equation , m-value, f u , f trimer , f tetramer , and the slope of pre-and post-transitions were obtained. Moreover, the fitting of this model was performed using the global fitting package in Igor Pro 5 to increase the accuracy where ⌬G UT H 2O and m-value were set as global parameters, meaning only one value is generated while fitting to various data sets. The other parameters can be varied.
To describe the A␤42 folding with the three-state model either Tri % 3M % 3U or Tetra % 4M % 4U, we plotted various free energies as a function of urea and employed linear extrapolation to obtain the free energy in the absence of denaturant (26). Briefly, the first transition describes multimer dissociation. The monomer fraction, f M , can be obtained by the experimental data using Equation 14.
obs , T , and M are the observed trimeric/tetrameric and monomer signals. The signal at 2 M urea was used as the monomeric signal. By using Equations 14, 7, and 8, we obtained various free energies of multimer dissociation, ⌬G dissociation , in different concentrations of urea. The second transition represents the monomeric species unfolding (M % U); the unfolded fraction, f U , can be obtained by the experimental data using Equation 15.
By using Equations 15 and 2, we were able to plot the free energies of unfolding, ⌬G UM , as a function of urea. The data were fit linearly by Equation 4 to obtain the free energy in the absence of urea and the m-value.
Fibril and Prefibrillar Oligomer Assay-Amyloid fibrillization was monitored by thioflavin T binding (27) using a fluorescence plate reader (XS; Molecular Devices Inc.). Folded A␤ peptides at 50 M in buffer A containing 0.02% sodium azide in different urea concentrations were prepared. The samples were stirred with Teflon-coated microstir bars at 300 rpm at room temperature. An aliquot of 10 l was taken at the indicated time and mixed with 90 l of thioflavin T solution (30 M thioflavin T in buffer A) to detect thioflavin T fluorescence. The samples were excited at 442 nm, and the emissions at 485 nm were recorded. The emission was plotted against time as indicated under "Results." The background fluorescence of the buffer is subtracted. Dot blotting was also employed to monitor the appearance of oligomeric species in the fibrillization process. Aliquots of 2 l from the samples were evenly pipetted and dotted on nitrocellulose membranes at the indicated time. Then, the membranes were stained by direct blue staining or blotted with A␤ antibody A11 to detect prefibrillar oligomers (28).

RESULTS
The experimental paradigm we used to analyze the early conformations and aggregation states of A␤ was to refold A␤ from 10 M urea denatured stock solutions. This strategy has been commonly used to study the folding properties of proteins (18). We refer to the refolded A␤ as described under "Experimental Procedures" as "folded" A␤ (Ͻ0.2 M urea) and the A␤ in 6 M urea as denatured A␤.
First, we performed analytical size-exclusion chromatography to examine the assembly and hydrodynamic properties of folded A␤40 and A␤42 at 50 M. We found that both A␤40 and A␤42 co-elute as single peaks with an apparent molecular mass of ϳ11 kDa as previously reported (12,17) (supplemental Fig. S1). This indicates that the hydrodynamic radius of A␤40 is indistinguishable from that of A␤42. We also confirmed that the aggregation kinetics of 50 M folded A␤40 and A␤42 displayed a significant lag time before fibril or prefibrillar oligomer formation was observed (10). This indicates that the diluted A␤ is quasi-stable for a period of several hours to days, which allows analysis of its folding properties in the absence of detectable fibril or prefibrillar oligomer formation immediately after dilution.
To further examine the assembly state of A␤, we performed native gel electrophoresis with the folded A␤40 and A␤42 at 50 M (Fig. 1A). We observed one protein band for A␤40 and two bands for A␤42, indicating that the A␤42 sample contains two species that are not in rapid equilibrium because they separate into distinct bands on the time scale of the electrophoretic separation. The faster migrating bands of A␤40 and A␤42 closely co-migrate, although the A␤42 band migrates slightly slower than that of A␤40. When the same samples were electrophoresed under denaturing conditions in the presence of SDS on Tris-Tricine gels, we found both samples ran predominantly as monomers (Fig. 1B) but A␤42 contains a higher molecular mass band of 14.5 kDa that is recognized by A␤-specific antibody. This is the approximate size expected of a trimer (13.5 kDa). We have previously reported that this higher molecular mass A␤42 band runs as a size consistent with a tetrameric species on Trisglycine gels (9,12). The results indicate that A␤40 and A␤42 have different assembly states after dilution from urea stock solutions. The fact that A␤42 runs as a single symmetrical peak on gel filtration, but runs as two distinct species on native gel electrophoresis, suggests that the two species have approximately the same hydrodynamic radius and do not separate by gel filtration.
To examine the possible differences in the conformations of A␤40 and A␤42, we employed fluorescence and circular dichroism spectroscopy. We used Bis-ANS, which is known to bind to hydrophobic surfaces of partially folded proteins, to monitor the conformational differences. Bis-ANS has been previously used to characterize soluble A␤ conformations that are distinct from fibrils (29). We found the maximum emission of Bis-ANS fluorescence of 50 M folded A␤40 was shifted to ϳ508 nm compared with that of the denatured A␤40, which was ϳ520 nm (Fig. 1C). The differences of Bis-ANS binding between folded and denatured A␤40 show that the folded A␤40 contains hydrophobic clusters on the protein surfaces, indicating it is able to form tertiary structures rather than completely random coils. In addition, the fluorescence intensity of Bis-ANS significantly increased in the presence of folded A␤40.
In contrast, the fluorescence intensity of Bis-ANS in the presence of folded A␤42 is ϳ10-fold higher than that of folded A␤40 (Fig. 1C). The emission maximum of A␤42bound Bis-ANS is blue-shifted to ϳ495 nm, and the intensity of folded A␤42 is much higher than denatured A␤42. The differences in the Bis-ANS emission suggest that either there are different Bis-ANS binding sites between the folded A␤40 and A␤42 due to structural differences or it is simply the result of different partition of bound and free Bis-ANS. Previous studies of Bis-ANS fluorescence indicate that Bis-ANS does not bind significantly differently to A␤40 and A␤42 at pH 2.4, suggesting that A␤ structures and aggregation states are substantially similar at low pH (29).
We also employed far-UV CD to compare the secondary structural properties of the folded and denatured A␤40 and A␤42. Folded A␤40 is mostly random coil (Fig. 1D) as previously reported (30,31). However, it is not completely unfolded because there are significant intensity differences between the folded and denatured A␤40. In comparison with A␤40, folded A␤42 contains some ␤-structure that displays a minimum at 216 nm. Denatured A␤42 contains significantly less structure than folded A␤42. Because of the presence of high urea concen- tration that absorbs light, we are unable to monitor lower wavelength for the denatured samples.
To further analyze the assembly and equilibrium folding properties of A␤, we examined A␤40 and A␤42 as a function of urea and peptide concentration. First, we performed denaturation studies by titration as described under "Experimental Procedures." Bis-ANS binding to A␤ in different urea concentrations was monitored. We also performed the denaturation by reverse titration to examine the reversibility. If the data generated by titration and reverse titration are equivalent, the folding of the protein is considered to be reversible where no aggregation or irreversible reaction is involved in the folding process (18). The results demonstrate that A␤ folding is reversible because the Bis-ANS emission spectra from titration and reversed titration are identical (supplemental Fig. S2). The results also indicate that both A␤40 and A␤42 have reached a quasi-equilibrium within the time for each titration (Ͻ30 s) where no further folding, unfolding, or aggregation events are rapidly proceeding. The reversibility of A␤ also allows us to examine the thermodynamics of the peptide assembly.
For A␤40 (Fig. 2A), the Bis-ANS emission at 500 nm and CD signal at 220 nm were plotted against urea concentration. Similar denaturation curves were obtained with both Bis-ANS binding and far-UV CD. This indicates the changes of the exposed hydrophobic surface (tertiary structures) and secondary structures are concomitant. The consistency between Bis-ANS binding and CD data also supports the conclusion that Bis-ANS binding monitors the overall conformational changes rather than selective species under these conditions. We examined different concentrations of A␤40 to determine whether the structural changes are due to quaternary associations. If multimers are present at equilibrium, the denaturation curves will change accordingly due to dissociation of the multimeric species, whereas if the protein exists as a monomer the denaturation curves will not change as a function of peptide concentration.
A␤40 at 50, 25, and 12.5 M were examined for both Bis-ANS binding and far-UV CD in different urea concentrations. The data from three concentrations obtained by both methods overlay with no apparent concentration dependence ( Fig. 2A). A single transition between ϳ0.75 and ϳ4 M urea was obtained. The pretransition (Ͻ1 M urea) and the post-transition (Ͼ4 M) are also affected by urea concentration but in linear fashions. The cooperative change is due to the structural changes of A␤ at intermediate urea concentrations. The data were accurately fit by a two-state model (N % U), where a folded, or native, monomer and an unfolded or denatured monomer are present at equilibrium. The free energy (⌬G UN H 2O ) of A␤40 obtained from the fit is ϳ1.1 kcal/mol. The m-value, a cooperativity parameter, is ϳ0.7 kcal/mol/M. An unfolded fraction of 0.18 was obtained in the absence of urea. The midpoint of the transition, [urea]1 ⁄ 2 , calculated from the fit is at 1.6 M, which is close to the experimental data. This indicates that fitting the data to a two-state model is precise. The residual of the fit is shown in Fig. 2A. Many small globular proteins (Ͻ100 kDa) adopt two-state folding mechanisms at equilibrium. The statistical study of proteins adopting a two-state mechanism provides calculations of the exposed surface area upon protein unfolding and the estimated number of residues in the protein using the m-value (22). The exposed surface area of the folded monomeric A␤40 is predicted to be 2963 Å 2 , and the estimated number of residues of A␤40 is 41.6, which is close to the actual number.
The denaturation of A␤42 at 50, 25, or 12.5 M was also examined (Fig. 2B). As observed for A␤40, the Bis-ANS spectra titration curves were fully reversible (supplemental Fig. 2B), indicating that the sample reaches equilibrium. The Bis-ANS and far-UV CD curves give similar results, indicating that the tertiary and secondary structure changes coordinately. The denaturation of A␤42 shows a concentration dependence with a midpoint of transition that varies from ϳ1.4, ϳ1.1 to ϳ0.6 M urea for 50, 25, and 12.5 M, respectively.
Comparing the denaturation data of A␤42 at 50 M with that of A␤40 at the same concentration, A␤42 shows a flatter pretransition, indicating the structure in the presence of 0 to 0.75 M urea is not significantly affected by urea. The data show one cooperative transition suggesting A␤42 adopts an apparent two-state mechanism. The midpoint of the transition, ϳ1.4 M urea, is slightly lower than that of A␤40, 1.6 M urea.
The concentration dependence and the apparent single transition of A␤42 denaturation suggest a model of multimer and unfolded monomer equilibrium. Because we observed a higher molecular weight band on SDS gel electrophoresis that has an apparent molecular weight of a trimer, (Fig. 1B), we fit the data globally to a model of a trimer to three unfolded monomers (Tri % 3U). Both the Bis-ANS fluorescence and far-UV CD data of A␤42 at 50 and 25 M fit well to the trimer-unfolded monomer model (Fig. 2B). The residual of the fit is shown in Fig.  2B. We obtained a ⌬G UN H 2O of ϳ14.9 kcal/mol and an m-value of ϳ2.8 kcal/M/mol. The data for A␤42 at 12.5 M cannot be fit to the model because it lacks a pretransition state. This indicates that at 12.5 M A␤42 multimer is not significantly populated or it is not stable in the presence of low concentrations of urea. Because the higher molecular weight A␤42 band has also been reported as a tetramer (9,12), we fit the data to a tetramer to four unfolded monomer model (Tetra % 4U) (25) and found they fit equally well with little difference on the square value, 0.145 and 0.069 for trimer and tetramer model, respectively. With the tetramer model, we obtained ⌬G UN H 2O of ϳ21.6 kcal/mol and an m-value of ϳ3.4 kcal/M/mol (Fig.  2C). However, both fits reveal a large amount of unfolded species populated in the absence of urea by linear extrapolation of the post-transition.
An alternative three-state model is also possible with the presence of a folded monomeric intermediate. It is possible that a transition at higher urea concentration is present despite a large population of unfolded species. The three-state model either Trimer % 3M % 3U or Tetramer % 4M % 4U includes a transition of multimer dissociation from 0 to ϳ2 M urea and an unfolding of a monomeric intermediate from ϳ2 to 4 M urea. Due to the complexity of the fitting, we used the plotting method and extrapolation described under "Experimental Procedures" to obtain the energy values (Fig. 2D) 50 and 25 M). The monomer unfolding ⌬G UN H 2O of 5.8 kcal/mol was obtained from the second transition. Therefore, we suggest that trimer/tetramer dissociation is the major energetic cost rather than the unfolding of the monomeric species. Overall, the distinct equilibrium folding properties of A␤40 and A␤42 indicate that the early species differ by assembly state, structure, and stability.
Because we found Bis-ANS binds much better to the folded than the denatured A␤, we explored which regions contain the Bis-ANS binding site. We compared the binding of Bis-ANS to synthetic segments of A␤ including A␤1-12, 7-18, 13-24, 19 -30, and 25-36 (Fig. 3). The short peptides were denatured and refolded as described for the full-length protein. Interestingly, two of the peptides, A␤13-24 and A␤25-36, show dramatic Bis-ANS emissions, whereas the others do not interact with Bis-ANS. Bis-ANS does not interact with A␤19 -30, which possesses a protease-resistant segment (32) suggesting that the hydrophobic side chains within the region are buried. Therefore, it is likely that the flanking regions of the protease-resistant segment compose the solvent-exposed hydrophobic surfaces.
Finally, we examined the effect of different urea concentrations on fibril and prefibrillar oligomer formation, using thioflavin T to monitor fibril formation and A11 anti-oligomer antibody (28) to monitor oligomerization. We analyzed the aggregation kinetics of folded A␤40 and A␤42 at 50 M with continuous stirring at room temperature (Fig. 4). By stirring the folded A␤40 and A␤42 individually at 0.2 M urea, a lag phase from 0 to ϳ 6 h followed by an increase in fibril formation was observed (Fig. 4, A and B). The fibrillization of A␤40 is significantly more sensitive to urea disruption than A␤42. This result suggests the folded structure of A␤42 promotes fibrillization. The results are consistent with the reported data on the fibrillization of A␤40 measuring by light scattering in the presence of urea (33). Similar to A␤40, A␤42 fibrillization is slower at 2 and 4 M urea. However, the maximal extent of A␤42 fibrillization in 2 M urea is slightly higher than that of the folded A␤42 in 0.2 M urea. The difference between final thioflavin T levels of A␤42 in 0.2 and 2 M urea is statistically significant with a p value of 0.034. Interestingly, we found that A␤42 in 4 M urea is still capable of fibrillization although the unfolding of A␤42 is complete. This suggests that the unfolded monomer is also amyloidogenic.
Prefibrillar oligomer formation was determined by immunoblotting with the conformation-dependent, oligomer-specific antibody (A11) (Fig. 4C). This antibody does not recognize oligomers smaller than approximately hexamer (28). Equal amounts of protein were dotted as determined by total protein staining (data not shown). In 0.2 M urea, oligomers begin to appear around 16 -52 h of incubation for A␤40 and A␤42 and increased with longer incubations, well after the time where fibril formation is maximal as determined by thioflavin fluorescence. The results are not due to limited sensitivity of the dot blot assay because it is at least 10-fold more sensitive for oligomers than the thioflavin T assay is for fibrils. This indicates that the normal kinetic relationship where oligomers are observed prior to fibril formation is reversed in the presence of urea.
At urea concentrations Ͼ0.2 M, oligomer formation was not observed for either A␤40 or A␤42. The fact that A11-positive oligomers are not observed at early times indicates that the structure of the folded multimer of A␤42 is immunologically distinct from the oligomeric state. The results also indicate that oligomer formation is significantly more sensitive to disruption by urea than fibril formation. The fact that fibril formation of A␤42 proceeds efficiently at 2 and 4 M urea, concentrations where no oligomers are detected, also indicates that oligomers are not an obligate intermediate for fibril formation and that oligomers and fibrils represent distinct alternative pathways of aggregation.

DISCUSSION
A␤ aggregation is a critical aspect of Alzheimer disease pathogenesis, and increasing evidence points to the role of relatively small aggregates or soluble oligomers as the primary pathogenic species (4). Unlike fibril formation, relatively little is known about the assembly states and energetics of initial folding and early aggregation. We found that the initial folded structures and properties of A␤40 and A␤42 are different and these differences are important for the ability to form higher order structures, such as amyloid fibrils. A␤40 exists as an unstable monomer population containing a large fraction of random coil but is not completely unfolded. It likely adopts a collapsed structure as previously reported (34). A␤40 displays a simple two-state monomeric to unfolded monomer model similar to most small proteins. Its free energy of folding is marginal with the value of 1.1 kcal/mol and a low m-value, 0.7 kcal/ mol/M, indicating the structure is not strongly dependent on the denaturant concentration. The estimated residue number of A␤40, ϳ41.6, based on two-state globular proteins is close to the actual number 40, but an over-estimation of the m-value could result from the large slope of the pretransition (35). Therefore, because A␤40 is unlikely to adopt a globular structure, the actual solvent-exposed surface area exposed upon unfolding can be smaller than the estimation as described under "Results." Unlike A␤40, A␤42 displays a denaturation profile that depends on the peptide concentration. One apparent transition was observed, and the data fit well with either a two-state trimer to unfolded monomer model (Tri % 3U) or tetramer to unfolded monomer model (Tetra % 4U), whereas the data do not fit to a dimer to unfolded monomer (D % 2U) model. Furthermore, the large post-transition could represent an unfolding transition of the monomeric species. Thus, the trimer/tetramer dissociation is the major step compared with the monomer unfolding in the three-state model (Tri % 3M % 3U or Tetra % 4M % 4U). The free energy of A␤42 dissociation is much larger than that of A␤40. Nevertheless, A␤42 at 12.5 M no longer displays a pretransition and is unable to be fit to the same model, suggesting that very little trimer/tetramer exists at this concentration.
The concentration dependence of A␤42 aggregation agrees well with the reported critical concentration for micelle formation in vitro, ϳ20 M (12). However, A␤40 also displays the same critical concentration as A␤42 (12), and the stability of A␤40 ranging from 12.5 to 50 M is not affected over this concentration range, indicating that the initial folding and aggregation of A␤42 is not related to micelle formation. These results indicate the early folding and aggregation states of A␤40 and A␤42 are distinct, which may account for their different fibrillization properties and pathological significance.
Differences in the early aggregation states of A␤40 and A␤42 have also been observed by photochemical cross-linking (17). Whereas the predominant cross-linked products of 30 M A␤40 are dimer, trimer, and tetramer, cross-linking of 30 M A␤42 yields predominantly tetramer, pentamer, hexamer, and heptamer. These data have been interpreted to suggest that A␤42 contains higher order aggregates ranging from tetramer to octamer termed "paranuclei" that are in rapid equilibrium and are not observed in A␤40. Our data are consistent with this observation and suggest that the trimeric or tetrameric species we observe represent the paranuclei observed by cross-linking. This is also consistent with the observation that paranuclei are not observed by cross-linking at low concentrations of A␤ (1 M) (36), because the trimeric/tetrameric state is poorly populated at concentrations below 12.5 M. Our data further indicate that trimers or tetramers are relatively stable because they do not dissociate detectably during native gel electrophoresis and they have a free energy of formation of ϳ15 or ϳ21.6 kcal/ mol for trimer and tetramer, respectively.
There are some apparent differences in the size and stability of the paranuclei previously described and the trimer/tetramer reported here. This apparent discrepancy can be simply explained by the possibility that the photochemical cross-linking products are a mixture of products that result from the cross-linking of a trimer/tetramer and the random collisional cross-linking of additional monomeric subunits that are in rapid equilibrium and present in the mixture. Trimer/tetramer formation correlates well with the faster nucleation kinetics of A␤42 (10), suggesting that these small oligomers may be important for nucleation. However, their formation is not required for fibril formation because A␤40, which does not form such an oligomer at the same peptide concentration, forms fibrils and fibril formation by A␤42 occurs at concentrations of urea that disrupt the small oligomers.
We were also able to probe the solvent-exposed hydrophobic area of A␤ by Bis-ANS binding. LeVine (29) has shown that Bis-ANS at pH 3 binds to the early species of aggregation and the kinetics is opposite to the thioflavin T kinetics. The possible reason for the disappearance of Bis-ANS signal is that the hydrophobic surface is buried in the higher aggregative assembly in which thioflavin T binds to its ␤-structure. This assumption is consistent with the finding that high concentrations of naphthalene sulfonates including Bis-ANS are able to inhibit A␤ oligomerization (37).
We found that the regions A␤13-21 13 HHQKLVFFA 21 and A␤30 -36 30 AIIGLMV 36 are the potential Bis-ANS binding sites. The hydrophobic residues in these regions are most likely exposed, whereas the protease-resistant segment containing a structured loop, A␤21-30, is rather buried (32,38,39). Our results are consistent with the NMR studies showing that a hydrophobic cluster composed of A␤17-19 is involved in a hydrophobic patch on the protein surface (34).
A␤ aggregation is critical for the etiology of AD. We found that the degree of "foldness" is important for the aggregation properties of A␤. Both unfolded A␤40 and A␤42 in 6 M urea cannot fibrillize, although A␤42 seems to be more capable of fibrillization under these denaturing conditions. Nevertheless, oligomerization of A␤40 and A␤42 is strongly affected by denaturant. The appearance of prefibrillar oligomers requires the native, folded structure in the early stages of aggregation because oligomerization is prevented by low concentrations of urea. If the oligomeric pathway is required for toxicity, the folded structure of A␤ is therefore essential for the etiology of AD.
The differences in assembly, stability, and structure between A␤40 and A␤42 (Fig. 5) are illustrated in the energy landscape. If the fibrillization of amyloid is a downhill energy diagram (40,41), the stability of folded A␤42 is closer to that of the fibrillar state which contains the most stable structure. These results provide a rational basis for explaining the enhanced aggregation, deposition, and pathological significance of A␤42.