Alzheimer (cid:1) -Amyloid Homodimers Facilitate A (cid:1) Fibrillization and the Generation of Conformational Antibodies*

We reported previously that stabilized (cid:1) -amyloid peptide dimers were derived from mutant amyloid precursor protein with a single cysteine in the ectodomain juxtamembrane position. In vivo studies revealed that two forms of SDS-stable A (cid:1) homodimers exist, species ending at A (cid:1) 40 and A (cid:1) 42. The phenomenon of the trans-formation of the initially deposited 42-residue (cid:1) -amyloid peptide into the amyloid fibrils of Alzheimer‘s disease plaques remains to be explained in physical

In the amyloidogenic processing of APP, 1 the A␤ peptide is produced, circulates extracellularly, and usually does not de-posit as plaques (1,2). In the secretory pathway, APP is cleaved C-terminal from A␤ residue 16, thus precluding the formation of full-length A␤1-40/42 but generating 3-kDa fragments, termed A␤17-40/42 or p3/40/42 (3,4). Familial AD-linked mutations in presenilins 1 and 2 and APP can cause Alzheimer's disease (AD) by increasing the cellular production of A␤42 (5), thereby accelerating the polymerization of A␤42 and promoting cerebral accumulation of A␤ as an essential early event in AD pathogenesis (6,7). Furthermore, peptides ending at A␤42, i.e. A␤1-42 and A␤17-42, appear to be a major constituent of the diffuse plaques seen initially in AD and Down syndrome, and they have been proposed to serve as a nidus for the aggregation of the more abundant A␤1-40 peptides (8,9). The presence of mixtures of A␤ assemblies, ranging from monomers to insoluble amyloid fibrils, has made it difficult to ascribe amyloidogenicity or toxicity principally to one or another A␤ species. Naturally secreted oligomers of A␤ are believed to have a direct role in AD pathology, both because of their short and long term neurotoxic effects and their ability to give rise to fibrillar assemblies characterized by the adopted ␤-sheet structure (10 -12). Fibrillar A␤ has been observed to bind to APP residues His-110, Val-112, and Ile-113 belonging to a common ␤-strand within the cysteine-rich domain of APP (13). The N terminus of A␤ was found to influence the kinetics of aggregation, and increased amyloidogenicity was suggested to correlate positively with N-terminal deletions up to A␤ residue 16 (10).
Garzon-Rodriguez and co-workers (14,15) reported the existence of a stable A␤1-40 dimer in the absence of SDS and suggested a ␤-structure for the dimer. At least three findings indicate the involvement of the carboxyl third, i.e. the A␤17-40/42 sequence, in the formation of SDS-stable dimers. First, A␤ residues 1-28 do not form SDS-stable dimers by prolonged incubation (16). Second, a monoclonal antibody recognizing A␤ residues 17-24 detected SDS-stable A␤ dimers from human brain (17). Third, mutant APP with a single cysteine in the ectodomain juxtamembrane position (i.e. A␤ residue 28 changed to cysteine) spontaneously formed APP homodimers leading to a ϳ7-fold higher level of A␤ dimers in the cell culture medium (18). Possibly, this substitution fitted with the predicted amphipathic interface of the APP ␣-helical juxtamembrane domain and then enabled the increased A␤ production.
To study the effect of this specific mutant on the assembly of A␤ into oligomers and fibrils, we have synthesized two series of covalently cross-linked A␤ peptides, one of which terminates at residue 40 and the other at residue 42. Apparent differences in the aggregation profiles have been examined at a fixed time point by circular dichroism (CD) and electron microscopy (EM).
The high content of ␤-sheet structure in the A␤42 dimer has been confirmed by Fourier-transformed infrared (FTIR) mass spectrometry, and a conformation-dependent epitope of A␤ dimers has been detected by surface plasmon resonance (SPR).

EXPERIMENTAL PROCEDURES
CD Spectroscopy-Alterations in the secondary structure content of the A␤-derived peptides were monitored by CD spectroscopy. Spectra were taken using a Jasco J-710 instrument with a PFD-350S temperature control device set at a sensitivity of 10 mdeg, a time constant of 4 s, and a scan speed of 5 nm/min. Typically, peptides were diluted to 100 g/ml in 1 mM Tris-HCl buffer, pH 7.5, with or without 1 mM dithiothreitol and measured in a 1-mm quartz cuvette, scanning from 240 to 190 nm. Protein concentrations for the calculation of mean residue ellipticity (⌰ mrw ) were determined by amino acid analysis of the sample after measurement. Curves are presented as the signal average of four transients with a similarly signal-averaged base line subtracted. A Fourier transform operation was carried out to remove high frequency noise from the signal.
Secondary structure content of the peptides was estimated from the far UV CD spectra as mean residue ellipticity using the PEPFIT program (19).
EM-Each peptide was dissolved to 1 mg/ml in deionized water (Milli Q, Millipore) supplemented with 5 l of 25% ammonia solution (measured pH 9.8) and incubated for 24 h at 37°C.
Aliquots (15 l) of the aged peptide solutions (1 mg/ml) were applied to freshly glow-discharged carbon-coated copper grids (300 mesh size) and negatively stained with 2% aqueous uranyl acetate. The negative staining preparations for all peptide samples were done in parallel within a time interval of about 10 min after the incubation period. All staining preparations were performed in duplicate. Grids were examined in a Zeiss EM 10A electron microscope (Oberkochem, Germany) at an acceleration rate of 80 kV. The magnification indicator of the microscope was routinely controlled by using a grating replica.
FTIR Measurements-The peptide was dissolved in 10 mM phosphate buffer D 2 O. The pH was adjusted to 8.7 with DCl and NaOH dissolved in D 2 O. The solution was incubated at 56°C for 4 h to reach full proton-deuterium exchange. To remove residual protons, the sample was lyophilized twice and dissolved in D 2 O again. For measurements, a peptide solution of 30 g l Ϫ1 in a temperature-controlled CaF 2 cell with a path length of 15 m was used. Spectra were recorded using a Phillips PU 9600 FTIR spectrometer whereby instrument optics and sample compartment were flushed with dry nitrogen gas. Buffer spectra recorded under the same conditions were subtracted from the sample spectra. The position of individual IR transitions in the amide I region between 1,700 and 1,600 cm Ϫ1 were resolved by second derivative spectra and by self-deconvolution routines. For secondary structure analysis the spectra were fitted using a set of six Lorentzian line shaped transitions.
The peptides to be cross-linked were allowed to form S-S bonds in 20% dimethyl sulfoxide. After a 12-h incubation, the sample was freezedried, dissolved in diluted ammonia, and injected onto a Jupiter 5-m C4 300 Å reversed phase column (Phenomenex, Torrance, CA). The dimeric fraction was collected, and identity was verified with mass spectrometry again.
Concentrations of dissolved peptides were determined by amino acid analysis according to the manufacturer's protocol after hydrolysis with 6 N HCl, 0.1% phenol for 24 h at 110°C (420A Amino Acid Analysis System, Applied Biosystems).
Antibodies and Western Blot-The monoclonal antibody W0-2 recognizing the N-terminal region of A␤ has been used as a control (22). Polyclonal A␤17-40 K28C dimer antibodies were raised against the synthetic peptide. Unconjugated peptide was used for rabbit immunizations. Immediately prior to immunization peptide solutions were prepared as described for EM, and complete Freund's adjuvant was added to a final concentration of 500 g/ml. After 4 weeks, subsequent immunizations were given with incomplete Freund's adjuvant three times at 2-week intervals. Blood samples were collected 10 days after each injection and stored at 4°C until assayed by SPR with immobilized synthetic peptides.
Tissue Preparation and Immunohistochemistry-APP23 mice (23) were sacrificed by cervical translocation. After removal of the brain, one hemisphere was formalin-fixed and paraffin-embedded for immunohistochemical analysis, and the other hemisphere was snap-frozen for biochemical analysis. Immersion fixation was carried out using 4% buffered formalin at 4°C. Immunohistochemistry was performed on 4-m paraffin sections according to standard protocols. In brief, sections were deparaffinized in xylene and rehydrated. After treatment with 1% H 2 O 2 in phosphate-buffered saline to block endogenous peroxidase activity, sections were heated in a microwave oven in 0.01 M citrate buffer, pH 6.0. Pretreatment with 88% formic acid enhanced immunoreactivity. Sections were treated with fetal calf serum prior to the addition of the primary antibodies to block nonspecific binding sites. This was followed by an overnight incubation with the primary antibody at room temperature. Monoclonal primary antibody G2-10 (1:500) against A␤40 and polyclonal antibody MX-02 (1:500) against A␤17-40 K28C dimers were incubated overnight in a humid chamber at room temperature. Staining was visualized using the ABC method, with a Vectastain kit (Vector Laboratories, Burlingame) and diaminobenzidine as chromogen.
Real Time SPR Analysis-Real time binding experiments were performed on a BIACORE system equipped with the upgrade kit (BIA-CORE) (24). All experiments were performed at 37°C.
The instrument was operated with phosphate-buffered saline as eluent (1ϫ phosphate-buffered saline, 0.005% Surfactant P-20, pH 7.4). SPR buffers and solutions were filtered and degassed before use. The sensor chip surface was derivatized with 300 -3,000 response units (RU) of A␤ peptides. For coupling, 1 volume of a 1 mg/ml peptide solution in diluted ammonia was mixed with 9 volumes of 10 mM sodium acetate, pH 3.2, and was injected at 5 l/min onto the sensor chip surface activated by N-hydroxysuccinimide and 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide. This was followed by the injection of 1 M ethanolamine, pH 9, for 5 min to quench residual N-hydroxysuccinimide groups.

RESULTS
We assessed uniform and standardized engineered A␤1-42, A␤1-40, A␤17-42, and A␤17-40 homodimers and compared their biochemical and biophysical characteristics with corresponding wild-type "monomers" (set into quotations because freshly dissolved synthetic A␤ always yields a certain ratio of monomers, dimers, and trimers (25)). The generation of homodimers was achieved by introducing a Cys at position 28. This approach is based on the following assumptions. First, dimerization of A␤ is not only the first but also an essential step for A␤ fibrillization. Second, although it is distinct from the A␤ dimer in vivo, the K28C substitution is nearly authentic A␤ because this mutation was selected for an enhanced production of A␤ dimers from APP695 K624C-transfected SH-SY5Y cells (18).
Secondary structures of A␤ peptides generated by substituting Lys-28 to Cys-28, oxidation, and subsequent purification by HPLC were analyzed together with wild-type peptides by CD spectroscopy and confirmed by FTIR spectroscopy for the A␤1-42 K28C mutant (Fig. 1). Most strikingly, wild-type and mutant peptides ending at residue 42 displayed an even more pronounced ␤-structural transition than forms ending at resi-  F), and wild-type A␤17-40 (blue curve in F) were processed as described for full-length forms of A␤. due 40 (Fig. 1, A versus B). Compared with A␤1-42 it is apparent that the absence of the C-terminal two amino acids in A␤1-40 leads to a marked increase in the negative maximum at ϳ198 nm (Fig. 1, A and B, blue curves). Deconvolution of CD spectra indicates a ␤-sheet content of 48%, a content of unstructured coil of 22%, and negligible amounts of ␣-helix for the A␤1-42 K28C dimer (Fig. 1A, red curve), for the A␤1-42 "monomer" (Fig. 1A, blue curve) 23% ␤-sheet, and 52% random and 18% ␤-sheet and 70% random for the A␤1-40 "monomer" (Fig.  1B, blue curve). Dimeric A␤1-40 K28C (35% ␤-sheet, 55% unstructured coil) was even superior to "monomeric" A␤1-42 (23% ␤-sheet, 52% unstructured coil; Fig. 1B, red curve, versus A, blue curve). To summarize, A␤ K28C dimerization significantly increased the ␤-sheet content by a factor of 2 for both the 40 and 42 series.
To determine the role of Cys-28 in maintaining the conformation, time course CD studies performed under reducing conditions with the A␤1-42 K28C dimer revealed that its predominant ␤-sheet structure was unchanged even in the presence of a 50-fold molar excess of dithiothreitol. CD spectra collected within 30 s of mixing were superimposable on spectra at 24 h and the control (Fig. 1C). The reduction of disulfide bonds was analyzed by matrix-assisted laser desorption/ionization timeof-flight (MALDI-TOF) mass spectrometry (data not shown). This indicates that treatment of A␤ K28C dimers with dithiothreitol per se was not sufficient for changing the conformation adopted after initial dimerization.
To analyze more accurately the high content of ␤-sheets in the A␤1-42 K28C dimer, FTIR spectroscopy was applied (Fig.  1D). The overall spectrum of the A␤1-42 K28C dimer with an absorption maximum for the amide I band at 1,625 cm Ϫ1 is typically observed for structures with mainly ␤-sheet content (26). Analysis of the spectra using a set of 6 Lorentzian line shaped IR transitions revealed a content of 64% ␤-sheet, 21% random coil, 9% turns, and 6% ␣-helix. This interpretation is slightly different from the CD spectrum (48% ␤-sheet, 22% random coil, 30% turns, and 0% ␣-helix). This inconsistency in regard to absent ␣-helical frequencies measured by CD spectroscopy is most likely because of the different methodology and has been observed before in other examples of nonhelical protein conformations analyzed by FTIR (26).
Amyloid ␤-peptides 17-40 and 17-42 derived by ␣and ␥-secretase cleavage of APP are major constituents of diffuse plaques with a fibrillar morphology observed by transmission electron microscopy (10,27) and were reported to induce neuronal apoptosis (28). Therefore, we also analyzed K28C substituted A␤17-40 and A␤17-42 by CD. The CD spectra revealed 40% ␤-sheet for the A␤17-42 K28C dimer (Fig. 1E) and 28% ␤-sheet for the A␤17-40 K28C dimer (Fig. 1F). Most interestingly, the lack of the N-terminal 16 amino acids reduced the ␤-sheet content by roughly a fifth for A␤17-42 and A␤17-40 "monomers" (Fig. 1, blue curves in E and F) compared with full-length peptides. Furthermore, for both the truncated and the full-length forms, the absence of the C-terminal amino acids Ile-Ala had little effect on the secondary structure of the "monomeric" forms but reduced the ␤-sheet content by roughly one-third for the corresponding K28C dimers (Fig. 1, A and B,  E and F, red curves).
To compare the amyloidogenic nature of K28C dimeric peptides relative to wild-type A␤, EM images of mutant and wildtype A␤1-42/40 and A␤17-40/42 peptides from freshly dissolved and 24-h aged solutions were generated under identical conditions. Analyses revealed a contrasting morphology for "monomeric" and K28C dimeric fractions. The regulated dimerization of A␤1-42 K28C improved the characteristic amyloid fibril morphology and underlined the importance of C-terminal amino acids Ile-Ala for fibril formation. A␤1-42 K28C generated ribbon-like filaments with a smooth contour measuring a 10 -40-nm and 5-10-m length ( Fig. 2A). This appearance strongly contrasted with mature fibrils of A␤1-42 "monomers" (Fig. 2B). Studies on A␤42 "monomers" and K28C dimers revealed that the A␤ fibril morphology was affected dramatically in the A␤1-42 K28C dimer. For the wild-type A␤1-42 the incubation time of 24 h led to the formation of beaded protofibrils with 5-13 nm average diameter and 0.15-1.5 m length (Fig. 2B), analogous to protofibril dimensions reported previously (29 -31). A second type of fibrils was of longer filaments of up to 3.5 m (Fig. 2B).
Thus, three significant differences of dimer fibrils were observed. First, the estimated diameter and the length of the A␤1-42 K28C was three times enlarged compared with the wild-type. Second, the fibrils did not appear beaded but straight. Third, the proportion of protofibrils was much lower, suggesting that the assembly/maturation was much faster or was an irreversible process. In each case, fibrillogenesis of A␤1-42 K28C must have started with dimers as initial building blocks and further oligomerized to tetramers and higher oligomers as described for the nucleation and growth of wildtype amyloid ␤-protein fibrils (32).
Next, we asked whether conformational changes of dimerization at the N-and C-terminal ends of the K28C peptides could be monitored by immunoreactivity. To produce antisera specific for conformational epitopes of the A␤ K28C dimer, rabbits were immunized with unconjugated freshly dissolved A␤17-40 K28C dimers. The target antigen choice of the Nterminally truncated peptide was led by the following considerations. First, the majority of antibodies in mice immunized with A␤ fibrils are directed against the N-terminal 12 residues of A␤ and are capable of cross-reacting with the "monomeric" peptide (33). Second, the monoclonal antibody 4G8 recognizing A␤ residues 17-24, which was described to label naturally occurring SDS-stable A␤ dimers on the blot preferentially (17) indicated the minimum epitope required.
The presence of conformation-specific antibodies was detected by a direct binding assay with the polyclonal serum obtained (MX-02) and a monoclonal reference antibody (W0-2 to epitope A␤ residues 1-11 (22)). Antibody binding to A␤1-42 K28C dimer, A␤1-40 K28C dimer, and A␤1-42 "monomer" N-terminally coupled to the SPR biosensor CM5 chip surface (Fig. 4) was analyzed. The antibodies were made to flow across the immobilized forms of A␤, and binding events were evident from observed changes in the relative diffraction index (RU index), recorded as a function of time (Fig. 4). Primary data collected for the antibody binding assay in Fig. 4 show that W0-2 bound with comparable affinities to A␤1-42 K28C (Fig.  4A), A␤1-40 K28C (Fig. 4B), and A␤1-42 "monomers" (Fig.  4C), as indicated by the minor loss of signal during the dissociation phase (ϳ270 -420s). The absolute RU signal depends on individual amounts of peptide coupled to the specific chip sur- FIG. 4. Sensorgrams of the complete antibody/A␤ binding cycle. Polyclonal serum MX-02 raised against the A␤17-40 K28C dimer (diluted 1:50) and monoclonal W0-2 (5 g/ml) was injected onto an A␤1-42 K28C dimer surface (immobilized to 300 RU on the chip surface) (A), an A␤1-40 K28C dimer surface (immobilized to 1,500 RU) (B), and an A␤1-42 monomer surface (immobilized to 3,000 RU) (C). The injections were started at 100 s at a flow rate of 5 l/min. Fast association phases (100 -270 s) and varying dissociation rates (270 -420 s) were observed for W0-2 and the polyclonal serum depending on the specific antigen coated to the chip surface. The rapid drop in signal at 270 s of the diluted MX-02 serum is the result of changes in the refractive index caused by contaminating proteins within the crude serum. A detailed kinetic analysis of the monoclonal W0-2 will be published elsewhere. Immunoblots of APP23 mice brain homogenate (100 g of protein/lane), synthetic peptides A␤1-40 (100 ng/lane), and A␤1-42 (100 ng/lane) probed with monoclonal W0-2 (1 g/ml) and polyclonal MX-02 (anti-A␤17-40 K28C dimer) serum (diluted 1:500) are shown in D. The proteins were resolved under reducing conditions on a 16% Tricine gel. Cortical immunohistochemical staining of 18-month-old APP23 transgenic mice E-G. E, staining of vacular amyloid A␤ using anti-A␤17-40 K28C. F, staining of plaque amyloid A␤ using anti-A␤17-40 K28C. G, staining of plaque amyloid A␤40 using G2-10. Counterstaining was with hematoxylin. face and varied between 180 RU (Fig. 4A) and 1,700 RU for the monoclonal W0-2 (Fig. 4C). For each individual chip surface, the overlay of sensorgrams allows a rough estimate of the avidity of the polyclonal MX-02 serum raised against A␤17-40 K28C by comparing the relative resonance signal in the dissociation phase of A␤-specific W0-2 and MX-02 curves. The reactivity of MX-02 was significantly higher than for W0-2 when A␤1-42 K28C was immobilized (Fig. 4A) and similarly strong with the A␤1-40 K28C surface (Fig. 4B) but was significantly below the W0-2 signal with "monomeric" A␤1-42 on the chip surface (Fig. 4C). Taking into consideration that the conformation recognized by MX-02 is partially present in the "monomeric" A␤, a quantitative assessment of the cross-reactivity is impossible. Nevertheless, this suggests that a major portion of the polyclonal antibodies are directed against conformational epitopes that only exist in the K28C dimerized peptides.
To characterize further the polyclonal MX-02 serum raised against A␤17-40 K28C, synthetic peptides A␤1-40, A␤1-42, and naturally occurring SDS-stable A␤ dimers were analyzed by Western blot. Crude homogenates of APP23 mouse brain and synthetic peptides yielded SDS-stable dimers of ϳ8 kDa plus "monomeric" A␤ with the conformation-specific MX-02 serum (Fig. 4D). The monoclonal antibody W0-2 recognizing the N-terminal unstructured region of A␤ demonstrated a significantly weaker immunoreactivity for SDS-stable dimers than MX-02. The apparent enhancement of immunoreactivity of polyclonal MX-02 with A␤1-40 is probably because of the presence of end-specific antibodies against the C terminus of A␤1-40. This result shows that antibodies within the MX-02 serum preferentially recognize conformation-specific epitopes that do not only exist in A␤ dimers with the K28C substitution (as indicated by the SPR data) but also in naturally occurring SDS-stable dimers.
To validate further the experiments using the artificially cross-linked peptide K28C as an antigen, immunohistological stainings were performed with the polyclonal serum MX-02. The serum recognized human A␤ in formalin-fixed and paraffin-embedded tissue from APP23 transgenic mouse brains (Fig.  4, E and F). An extracellular staining was observed in the amyloid component of plaques but not in dystrophic neurites (Fig. 4E). The polyclonal MX-02 antibody directed against A␤17-40 K28C dimers yielded a much stronger staining of vascular and plaque amyloid (Fig. 4, E and F) than the monoclonal G2-10, which specifically recognizes A␤40 (Fig. 4G). This suggests that A␤ dimers could be the species initially deposited in the vascular amyloid.

DISCUSSION
Characterization of the first step of A␤ fibril formation after the release of A␤ from its precursor is of paramount importance to understanding AD pathogenesis and will offer new therapeutic approaches. Based on our earlier study, where processing of K624C disulfide bonded APP homodimer mutants yielded homodimeric A␤ (18), stabilized A␤ homodimers have been generated by substituting Lys-28 to Cys-28 (K624C in APP695 numbering). This approach was employed to obtain a model for the fundamental A␤-A␤ interactions into fibrils by using A␤ K28C homodimers to scale down the assembly start to a two-step nucleation event, allowing conclusions to be made about the conformation of nascent A␤ from APP homodimers and about its further associative interactions. The relevance of A␤ K28C homodimers to the biophysical behavior of wild-type A␤ is strongly supported by the following published observations. A reductive methylation of Lys-28 slowed the rate of A␤1-40 aggregation but had little effect on protofibril structure (34) and was partially inaccessible to radiomethylation in A␤1-40 fibrils (35). A␤ homodimers play a direct role in A␤ fibril formation: the A␤ fragment 16 -35 was found to form a double-layered hairpin-like structure with the parallel ␤-sheet stabilized by an intrastrand salt bridge (D23-K28) (36,37). In accordance with our EM data and in conjunction with existing NMR and other structural data, the A␤ fibril is a hydrogenbonded, parallel ␤-sheet defining the long axis of the A␤ fibril propagation. Under the pH conditions used, the aggregation of A␤ K28C dimer peptides might have occurred according to the model suggested by Li et al. (38), in which the side chain of Lys-28 from one unit layer forms interlayer hydrogen bonds with the backbone atoms of Lys-28 in the adjacent unit layer. Thus, it can be assumed that in the A␤ K28C peptides the dimeric state was frozen by covalent cross-linking of the side chains of Cys-28 similarly to the interlayer hydrogen bonds observed for Lys-28 in wild-type A␤ and without altering any important inter-or intramolecular interactions.
A␤1-40 and A␤1-42 K28C homodimers demonstrated why A␤1-42 is more amyloidogenic than A␤1-40. To the best of our knowledge unaggregated and homogeneous starting material indicated that Ile-41 and Ala-42 of A␤1-42 forces the fibril length and that the K28C mutant displayed an even further pronounced ␤-sheet transition than forms lacking of these residues. This conclusion is supported by K28C mutants of A␤17-40 and A␤17-42, which also showed a significantly increased ␤-sheet compared with wild-type peptides. The ribbonlike uniform morphology of A␤1-42 dimers is best explained by the tendency of preformed fibrils to aggregate laterally. This supports an in-register parallel ␤-sheet organization as reported previously (39,40) for protofibrils. Recently, the first NMR data on A␤1-42 fibrils (41) found an in-register parallel ␤-sheet organization together with ϳ15% of fibrillized A␤1-42 occurring in an antiparallel ␤-sheet structure. Also, IR spectroscopy data suggested that A␤ may form an antiparallel ␤-sheet with a turn located around residues 26 -29 (Ser-Asn-Lys-Gly) (42).
Most interestingly, when we investigated the influence of the N-terminal ␤-strand segment (A␤ residues 10 -22), a growth arresting effect was found for Ile-41 and Ala-42 of A␤17-42 in sharp contrast to A␤ full-length peptides (see above). These findings, based on a contemporaneous analysis of both forms, are inconsistent with previous studies where a superior amyloidogenicity for A␤17-42 over A␤17-40 was suggested from sedimentation data (10). In this study, only A␤17-42 was analyzed and exhibited a fibrillar morphology by transmission electron microscopy (10). A␤17-40 was analyzed independently by others (27). As shown by our EM data the replacement of Lys-28 with Cys enhanced the inhibitory effect of the Nterminal deletions of A␤17-40 and A␤17-42 on fibril formation further indicating the distinct inverse relationship between the ␤-sheet structure and fibril formation. From the observations that the K28C mutant of A␤17-40/42 is rather inhibitory to fibril growth and dimerization further lowers the mean fibril diameter can be concluded that A␤17-40 aggregation normally occurs in an antiparallel manner.
Taken together, our findings support the emerging model that two ␤-strand segments in each peptide molecule span residues 10 -22 and 30 -35 with a loop or hinge between the two ␤-sheets located at residues 23-29 including Cys-28 of the mutant peptide, which enabled the regulated dimerization of A␤1-42 (Fig. 5). In this model, the C terminus of A␤1-42 would be in close contact to the residues 25-29 containing the bend of the peptide backbone that links the two ␤-sheets in mature fibrils. This is supported by SPR data obtained with C-terminal specific monoclonal antibodies revealing that the C-terminal region of the peptide is sensitive to conformation-induced epitope masking in fibrils and implying that A␤1-42 protofibrils associate by an end-to-end coalescence.
The physiological relevance of our findings is emphasized by a number of reports on A␤ dimers occurring in vivo. The presence of A␤ dimers in the cortex has been suggested to initiate the accumulation of A␤ in the human brain (17). Nonfibrillar SDS-stable dimers have been characterized as neurotoxic derivatives (43) and selectively blocked hippocampal long term potentiation in the absence of monomers, protofibrils, or fibrils (12). Rather than being initiated soon after the generation of A␤ in discrete intracellular vesicles as suggested by Walsh et al. (12), our findings demonstrate for the first time that a biochemically defined assembly of A␤ into A␤ dimers probably represents the initial step in amyloidogenesis. It does not only offer an explanation as to why fibrils are formed despite the apparent low concentrations of A␤ in the nervous system but also answers the question of the origin of a unusually high activation energy for the transition of a monomeric ␣-helical intermediate into a ␤-sheet conformation. A prevention of fibril formation by reducing dimer concentrations could have significant relevance to the treatment of amyloidoses where oligomers have already been implicated in disease processes (44).
The peptides used in this study should be useful for screens of such compounds and a monoclonal antibody recognizing A␤1-42 homodimers (which are potentially the earliest forms of synaptotoxic A␤ oligomers) might be useful for A␤ amyloid related therapeutic approaches by impeding its precipitation into existing plaques. The biological or pathological relevance of A␤ homodimers remains to be elucidated further, and a monoclonal antibody generated against A␤ K28C dimers should be most useful to achieve this goal.