Antiparallel Triple-strand Architecture for Prefibrillar Aβ42 Oligomers*

Background: Soluble Aβ42 oligomers, rather than insoluble amyloid fibrils, are toxic species in Alzheimer's disease. Results: We obtained structural restraints at all 42 residue positions in Aβ42 oligomers and performed structural modeling. Conclusion: In oligomers, each Aβ42 protein forms a single β-sheet with three antiparallel β-strands. Significance: Our novel structural model provides new structural framework for understanding oligomer-fibril interconversion and designing oligomer-targeted therapeutics. Aβ42 oligomers play key roles in the pathogenesis of Alzheimer disease, but their structures remain elusive partly due to their transient nature. Here, we show that Aβ42 in a fusion construct can be trapped in a stable oligomer state, which recapitulates characteristics of prefibrillar Aβ42 oligomers and enables us to establish their detailed structures. Site-directed spin labeling and electron paramagnetic resonance studies provide structural restraints in terms of side chain mobility and intermolecular distances at all 42 residue positions. Using these restraints and other biophysical data, we present a novel atomic-level oligomer model. In our model, each Aβ42 protein forms a single β-sheet with three β-strands in an antiparallel arrangement. Each β-sheet consists of four Aβ42 molecules in a head-to-tail arrangement. Four β-sheets are packed together in a face-to-back fashion. The stacking of identical segments between different β-sheets within an oligomer suggests that prefibrillar oligomers may interconvert with fibrils via strand rotation, wherein β-strands undergo an ∼90° rotation along the strand direction. This work provides insights into rational design of therapeutics targeting the process of interconversion between toxic oligomers and non-toxic fibrils.

A␤42 oligomers play key roles in the pathogenesis of Alzheimer disease, but their structures remain elusive partly due to their transient nature. Here, we show that A␤42 in a fusion construct can be trapped in a stable oligomer state, which recapitulates characteristics of prefibrillar A␤42 oligomers and enables us to establish their detailed structures. Site-directed spin labeling and electron paramagnetic resonance studies provide structural restraints in terms of side chain mobility and intermolecular distances at all 42 residue positions. Using these restraints and other biophysical data, we present a novel atomiclevel oligomer model. In our model, each A␤42 protein forms a single ␤-sheet with three ␤-strands in an antiparallel arrangement. Each ␤-sheet consists of four A␤42 molecules in a headto-tail arrangement. Four ␤-sheets are packed together in a face-to-back fashion. The stacking of identical segments between different ␤-sheets within an oligomer suggests that prefibrillar oligomers may interconvert with fibrils via strand rotation, wherein ␤-strands undergo an ϳ90°rotation along the strand direction. This work provides insights into rational design of therapeutics targeting the process of interconversion between toxic oligomers and non-toxic fibrils.
Alzheimer disease is a fatal neurodegenerative disorder and the most common form of dementia (1). Deposition of amyloid ␤ (A␤) 3 peptide in the form of amyloid plaques is a hallmark of Alzheimer pathology. Amyloids refer to fibrillar protein aggregates with common tinctorial and structural characteristics and are involved in a range of neurodegenerative, localized, and systemic disorders including Parkinson and Huntington diseases, and type II diabetes (2). Soluble A␤ oligomers have been increasingly recognized as primary neurotoxins in Alzheimer disease (3)(4)(5). Several A␤ oligomers have been identified in vivo (6,7), including dimers, trimers, and A␤*56. Different in vitro protocols have been used to prepare oligomers such as A␤-derived diffusible ligands (8), globulomers (9), prefibrillar oligomers (10), and amylospheroids (11). Because the molecular structures of these oligomers are unknown, it is impossible to know how many unique structures exist in these A␤ oligomers. Currently, structural classification of these oligomers is largely restricted to the use of conformation-specific antibodies (12). Based on immunoreactivity to the oligomer-specific polyclonal antibody A11, A␤ oligomers can be classified into A11-positive prefibrillar oligomers and A11-negative fibrillar oligomers (12).
One challenge in the structural studies of A␤ oligomers is related to their transient and heterogeneous nature. A␤ oligomers represent a series of intermediate assemblies on or off the pathway to fibril formation. Oligomers prepared using different protocols have been shown to be structurally diverse (13). Some A␤ oligomers have been shown to have similar parallel in-register ␤ structures as amyloid fibrils (14), and other oligomers adopt distinct structures (15)(16)(17)(18)(19). Heterogeneity can also occur within the same oligomer sample (20,21). Structural heterogeneity has been a major obstacle in obtaining high-resolution structural data.
Site-directed spin labeling (SDSL) in combination with electron paramagnetic resonance (EPR) spectroscopy has emerged as a powerful approach to characterize the structures of amyloid fibrils (22). The general strategy of SDSL includes substitution of a selected residue with cysteine and subsequent modification of the cysteine residue to produce a spin label side chain. The EPR sample can be in solutions, aggregates, or membrane environments, and of any size. As shown previously in the studies of A␤ and yeast prion protein Ure2p, EPR can resolve structural heterogeneity and separate different structural states (23)(24)(25)(26). Distance measurements with continuous-wave and pulsed EPR can cover a wide range of distances from 5 to 70 Å (27,28). These advantages make SDSL EPR a promising technique to obtain detailed structural information of the inherently heterogeneous A␤ oligomers.
In this work, we performed a comprehensive structural study on A␤42 oligomers prepared using a fusion protein, GroESubiquitin-A␤42 (GU-A␤42). This fusion protein construct forms highly ordered oligomers without further assembling into fibrils, and enables us to obtain detailed structural information of these A␤42 oligomers. The fusion protein system is similar to yeast prion proteins such as Sup35p and Ure2p, which contain both a prion domain and a globular domain, and the globular domain does not participate in the amyloid formation of these yeast prion proteins (29). The fusion protein approach also provides for some other unique applications. For example, a split luciferase-A␤ system allows high sensitivity detection of oligomer formation in mammalian cells (30). Fusion protein approaches also enable studies of mutational effects at specific residue positions in yeast (31) and Escherichia coli (32) cells and in vivo high throughput screening of small molecule inhibitor libraries (33). Fusion proteins also facilitate structural characterization of A␤ fragments using x-ray crystallography (34).
These GU-A␤42 oligomers recapitulate the characteristics of prefibrillar oligomers, such as immunoreactivity to oligomerspecific antibody A11 (12). For structural studies with EPR, spin labels are introduced, one at a time, at all 42 residue positions of A␤42 sequence. Residue-specific mobility analysis using EPR reveals three ordered segments at residues 1-10, 13-23, and 28 -42. Distance measurements show two major intermolecular distance distributions at each of the 42 residue positions: 9 -10 Å and 15-17 Å. These results allow us to suggest a triplestrand antiparallel model for the A␤42 prefibrillar oligomers. Our model for prefibrillar oligomers points to a mechanism of oligomer-fibril interconversion wherein rotation of ␤-strands reorganizes the ␤-sheets of the oligomers into new fibril ␤-sheets that run (i.e. have ␤-hydrogen bonding) approximately perpendicular to the original ␤-sheets of the oligomers. We term this mechanism of nucleated conformational conversion (35) (conversion of an oligomer from one conformation to another without adding or losing material) to be "strand rotation."

Preparation of A␤42 Fusion Proteins and Full-length A␤42-
The DNA construct of GroES-ubiquitin-A␤42 (36) and the deubiquitylating enzyme Usp2cc (37) were kindly provided by Dr. Rohan T. Baker at Australian National University (Australia) and Dr. Il-Seon Park at Chosun University (South Korea). Single cysteine mutations at various sites were introduced into A␤42 sequence using QuikChange kit (Agilent Technologies). Mutations were confirmed with DNA sequencing. Expression of GroES-ubiquitin-A␤42 in E. coli and their purification were performed as previously described in Ngo and Guo (38). Expression of Usp2cc and removal of fusion protein GU to prepare full-length A␤42 was performed as previously described in Gu et al. (39). The purity of A␤42 fusion proteins was checked with SDS-PAGE. The fusion protein GU without A␤42 was prepared using the same protocol as GU-A␤42.
Spin Labeling of A␤42 Fusion Proteins-Dithiothreitol was added to purified A␤42 fusion proteins to a final concentration of 10 mM and was allowed to incubate for 20 min at room temperature to break any disulfide bonds. Then A␤42 fusion proteins were buffer exchanged into the labeling buffer (7 M guanidine hydrochloride, 50 mM NaCl, 20 mM MOPS, pH 6.8) using a 5-ml HiTrap desalting column (GE Healthcare). The spin labeling reagent MTSSL (1-oxyl-2,2,5,5-tetramethylpyrroline-3-methyl methanethiosulfonate, Enzo Life Sciences) was at 10-times molar excess and then incubated for 1 h at room temperature. Spin-labeled A␤42 fusion proteins were then precipitated with methanol, air dried, and then stored at Ϫ80°C.
Preparation of A␤42 Fusion Protein Oligomers-Methanol precipitated A␤42 fusion proteins were resuspended in PU buffer (50 mM phosphate, 8 M urea, pH 10.0) to a final concentration of 1 mM. Then A␤42 fusion proteins were diluted 20-fold in PBS buffer (50 mM phosphate, 140 mM NaCl, pH 7.4) to 50 M. After incubation at room temperature for 24 h, oligomers form loosely associated precipitates. As a control, the fusion protein GU without A␤42 was also precipitated with methanol, resuspended in PU buffer, and followed by 20-fold dilution to PBS buffer and incubation at room temperature for 24 h.
Transmission Electron Microscopy-GU-A␤42 oligomer samples were diluted with water to 5 M and then applied onto glow discharged copper grids covered with 400 mesh formvar/carbon film (Ted Pella) and negatively stained with 1% uranyl acetate. Samples were examined under a JEOL JEM-1200EX transmission electron microscope with an accelerating voltage of 80 kV.
X-ray Powder Diffraction-GU-A␤42 oligomers were collected by centrifugation at 20,000 ϫ g for 20 min, and washed with water. The GU sample was buffer exchanged to water using a HiTrap desalting column (GE Healthcare). The lyophilized powders for both GU and GU-A␤42 were mounted on the tip of a broken glass rod. Then, the specimen was placed on the goniometer of an in-house x-ray machine and shot using a Rigaku-FR-D x-ray generator equipped with a Rigaku HTC imaging plate detector.
Thioflavin T Fluorescence Assay-Thioflavin T (Sigma) was dissolved in PBS and filtered with a 0.22-m filter. GU and GU-A␤42 samples were diluted to a final concentration of 5 M in PBS containing 50 M ThT. Fluorescence was measured immediately on a Jasco FP-6200 spectrofluorometer. Excitation was at 440 nm (5-nm slit width), and emission was at 485 nm (5-nm slit width).
Dot Blot Assay with A11 Antibody-GU sample and GU-A␤42 oligomers were diluted with PBS to 25 M. Then 5 l of GU and GU-A␤42 samples were spotted on a nitrocellulose membrane (Bio-Rad). The membrane was blocked in 10% fat free milk in Tris-buffered saline (TBST) (50 mM Tris, 150 mM NaCl, 0.05% Tween20) at room temperature for 1 h, followed by incubation with the rabbit polyclonal A11 antibody at 2.4 g/ml in 5% fat free milk, TBST at 4°C overnight. TBST was used to wash the membranes for three times. Then, the membrane was incubated with anti-rabbit HRP-conjugated second-ary antibodies (1:5000 in 5% fat free milk, TBST) (Jackson ImmunoResearch) for 1 h at room temperature, followed by further washing in TBST buffer for three times. The blots were developed using the Super Signal West Pico Chemiluminescent Substrate kit (Pierce).
Circular Dichroism Spectroscopy-Secondary structures of GU sample were analyzed by CD spectroscopy. The GU sample (200 l), 24 h after dilution from PU to PBS, was placed in a 0.1-cm path length quartz cell (Starna). A Jasco J-715 CD spectrometer was employed. The measurement were carried out in a wavelength range of 190 -260 nm at a rate of 20 nm min-1 with a step resolution of 0.5 nm, a time constant of 4 s and a bandwidth of 1 nm. The CD spectra were obtained by averaging 6 scans. The temperature was set at 25°C. The spectra were corrected by subtracting the buffer background. The HT voltages were above 800 for the wavelength range of 190 -203.5 nm, so only the CD data in the wavelength range of 204 -260 nm were reported. Because of the formation of insoluble oligomers, the GU-A␤42 samples were not studied with CD spectroscopy. The CD results are reported as mean residue ellipticity in Fig.  1F.
Aggregation Kinetics of A␤42-Purified full-length A␤42, without fusion protein partner, was buffer exchanged to 30 mM ammonium acetate, pH 10.0, then lyophilized and stored at Ϫ80°C. For aggregation experiments, lyophilized A␤42 powder was dissolved in CG buffer (20 mM CAPS, 7 M guanidine hydrochloride, pH 11), filtered through 0.2-m syringe filter (Corning 431212), and then buffer exchanged to PBS using a 5-ml HiTrap desalting column (GE Healthcare). The sample was then filtered through a 0.2-m syringe filter (Corning 431212), and concentration was determined using an extinction coefficient of 1.28 mM Ϫ1 at 280 nm. The A␤42 was diluted to 50 M with PBS, supplemented with 20 M ThT, either in the presence or absence of 2.5 M GU or GU-A␤42 samples. After mixing all components, 50 l of each mixture was transferred to the 384 well black polystyrene plate with clear bottom and PEG coating (Corning 3655). The plate was then sealed with a plastic film (Corning 3095). All these steps were performed on ice if possible. The aggregation was initiated by placing the plate in a Victor 3V plate reader (Perkin Elmer). The plate is incubated at 37°C with orbital shaking (1 mm shaking diameter, normal shaking speed). The thioflavin T fluorescence was measured through the bottom of the plate at every 5 min (with excitation filter of 450 nm and emission filter of 490 nm). Each sample was prepared in duplicates.
EPR Spectroscopy-EPR measurements were performed at X-band frequency on a Bruker EMX spectrometer equipped with the ER4102ST cavity at room temperature using a microwave power of 20 milliwatt. A modulation frequency of 100 kHz was used. Modulation amplitude was optimized to each individual spectrum. Scan width is 200 G. For each sample, 20 l of oligomer sample was loaded into glass capillaries (VitroCom) sealed at one end. EPR spectra in each figure panel were normalized to the same number of spins.
EPR Distance Analysis-Distance analysis was performed using the program ShortDistances, developed by Dr. Christian Altenbach at UCLA. The detailed fitting procedure to obtain distances has been previously described (40). The 20% labeled spectra were used as the spectra without dipolar interactions. To avoid over fitting of the experimental data, we emphasized on using minimum number of variable parameters. The width of the distance distribution was fixed at 2 Å. The distance, percentage of the spin labels at the fitted distance, and the percentage of non-interacting spin labels were allowed to vary. The fitted spectra, distances, and their relative populations are plotted in Fig. 8.
Modeling of A␤42 Oligomers-EPR distance information was used to create harmonic distance constraints for backbone hydrogen bonding for A␤42 plus four residues of ubiquitin (Lys-Arg-Gly-Gly) at the N terminus of A␤42. Models were varied in number of oligomers in the directions of ␤-hydrogen bonding (direction of sheets), and in number of sheets. Models were also varied in ␤-twist, which changes the overall twist along the ␤-sheet direction. Models were energy-minimized by molecular dynamics with harmonic distance constraints using the CNS program (41). The best fit model satisfied the distance constraints from EPR and also fit well to the x-ray powder diffraction, using the R o goodness of fit metric as previously described (18).

A␤42 Fusion Protein Forms A11-positive Prefibrillar
Oligomers-We studied the oligomer formation of A␤42 using a fusion protein containing GroES-ubiquitin at the N terminus of A␤42 sequence (36). Here this construct is termed GU-A␤42. Oligomers are formed by a 20-fold dilution from a denaturing buffer containing 8 M urea to phosphate-buffered saline (PBS) and incubated at room temperature for 24 h without agitation. The final GU-A␤42 concentration is 50 M. Shortly after dilution from urea to PBS, GU-A␤42 forms visible precipitates. Twenty-four hours after aggregation began, most of the GU-A␤42 was in precipitate, and soluble GU-A␤42 was below the detection limit by absorbance at 280 nm. We attribute the ability to form visible aggregates to the presence of GroES as a fusion protein partner. GroES is known to form oligomers (42,43). When we performed the same aggregation assay using fulllength A␤42 without the fusion protein, we did not observe any visible aggregates. At the same time, A␤42 quickly aggregates into amyloid fibrils in the absence of fusion protein partners. We conclude that the GroES fusion partner promotes the trapping of GU-A␤42 into oligomers and prevents fibril formation, allowing us to further characterize its structure below. One disadvantage, however, is that formation of precipitates prohibited us from characterizing the properties of GU-A␤42 oligomers using solution-based methods such as size exclusion chromatography, light scattering, and sedimentation.
Transmission electron microscopy (TEM) shows that the GU-A␤42 sample consists of globular oligomers (Fig. 1A). Most of the GU-A␤42 oligomers have a diameter of 10 -12 nm. No fibrils or any elongated protofibrils were observed by electron microscopy, even after more than 2 weeks of incubation at room temperature.
To establish the relationship between the GU-A␤42 oligomers and other A␤42 oligomers, we performed dot blot analysis using the oligomer-specific A11 antibody (10). Fig. 1B shows that the GU-A␤42 oligomers stained strongly with A11 antibody. This result suggests that the GU-A␤42 oligomers should be classified as "prefibrillar" oligomers (12), which are distinct from fibrillar oligomers that share the same epitope as amyloid fibrils.
X-ray powder diffraction of GU-A␤42 oligomers shows two sharp reflections at 4.6 and 10.0 Å (Fig. 1C), which have been previously suggested as characteristics of cross-␤ structure (44). The 4.6-Å reflection corresponds to the interstrand spacing within the same ␤-sheet, and the 10.0-Å reflection corresponds to the sheet to sheet spacing in the oligomers. The sharp nature of reflections at both 4.6 and 10.0 Å suggests that GU-A␤42 oligomers contain highly ordered ␤-sheet structure.
Thioflavin T (ThT) binding assay shows that GU-A␤42 has weak ThT binding (ϳ2.5-fold change in fluorescence intensity compared with ThT alone), and GU alone has only a marginally greater ThT signal (Fig. 1D). This is consistent with previous studies on full-length A␤42 oligomers, which also show much weaker binding to ThT than mature fibrils (45).
On SDS gel, the GU-A␤42 sample contains two major species of SDS-resistant oligomers (Fig. 1E). The apparent sizes for these two oligomers are 86 and 128 kDa according to the calibration with molecular weight standards. The molecular mass for GU-A␤42 monomer is 25 kDa. The commercial molecular weight standards consist of denatured single polypeptide chains and thus adopt extended structures. SDS-resistant GU-A␤42 oligomers, however, must adopt some compact structures to stay oligomeric. Therefore, GU-A␤42 oligomers would migrate faster on the gel than the molecular weight standards of comparable size. For these reasons, we conclude that the SDS-resistant GU-A␤42 oligomers are tetramers and hexamers. This finding is similar to our previous study showing that GU-A␤42 oligomers also form tetramers and hexamers in the presence of 8 M urea (38). Here the oligomer sample contains only 0.4 M urea. Even with such low urea concentration, no other oligomeric species were observed within the detection limit of Coomassie staining on the gel. Tetramers and hexamers are also the major oligomer forms for A␤42 without fusion protein partners, as revealed by ion mobility mass spectrometry (46), suggesting GU-A␤42 oligomerizes similarly as A␤42.
We also checked if the fusion protein GU alone is refolded under the condition of our oligomer preparation. As shown in Fig. 1F, the CD spectrum of the GU sample is qualitatively similar to previously published CD spectra of GroES (47) and ubiquitin (48), suggesting that the GU protein is refolded to the native structure under our oligomerization condition.
These results show that GU-A␤42 forms globular oligomers. Lack of fibril formation suggests fusion protein partners effectively trap GU-A␤42 in a stable oligomeric state. These GU-A␤42 oligomers recapitulate the characteristics of fulllength A␤42 prefibrillar oligomers including A11 reactivity, weak ThT binding, and presence of SDS-resistant oligomers.
Residue-specific Side Chain Mobility in A␤42 Oligomers-To study the structure of GU-A␤42 oligomers with EPR, we introduced spin labels, one at a time, at all 42 residue positions of A␤42 sequence. A commonly used spin labeling reagent (see "Experimental Procedures") was used to generate the spin label side chain named R1 ( Fig. 2A).
TEM studies show that the spin-labeled GU-A␤42 proteins form globular oligomers (Fig. 2B). The morphology of these oligomers is indistinguishable from the oligomers of wild-type GU-A␤42, suggesting that spin labeling has little effect on the formation of oligomers. Previously, we found that cysteine substitutions of hydrophobic residues (Ile-31, Ile-32, Leu-34, Val-39, Val-40, Ile-41) at the C-terminal region disrupted the formation of tetramers and hexamers (38). To check if spin labeling at these residue positions also disrupts oligomer formation, we performed SDS-PAGE analysis for the GU-A␤42 oligomer samples labeled at C-terminal residues (Fig. 2C). We found that tetramers and hexamers are largely unaffected by spin labeling. We propose that the hydrophobic nature of the nitroxide ring makes it very tolerable in A␤ oligomers. These results also suggest that hydrophobicity, rather than size, of the amino acid side chain is critical for A␤42 oligomerization.
To study the side chain mobility using SDSL EPR, we prepared GU-A␤42 oligomers using a mixture of spin-labeled protein and wild-type protein at 1:4 molar ratio. This sample is referred to as "20% labeled." The EPR spectra of 20% labeled samples are shown in Fig. 3. In the 20% labeled sample, intermolecular spin-spin interactions are minimized, so the EPR spectral lineshape is mainly determined by the mobility of the spin label, which reflects local structure at the labeling site. The side chain mobility is estimated using inverse center line width of the EPR spectra (49). Based on the plot of residue-specific side chain mobility (Fig. 4), A␤42 consists of four structural segments. Segment 2 (residues 13-23) and segment 4 (residues 28 -42) are the most ordered segments in the oligomers. The relatively more flexible segment 3 (residues 24 -27) separates segments 2 and 4. X-ray powder diffraction studies suggest the presence of ordered ␤-sheet structures in GU-A␤42 oligomers (Fig. 1C). Therefore, we conclude that both segments 2 and 4 adopt ␤ structures, and segment 3 forms a turn connecting the two ␤ strands. Segment 1 (residues 1-12) has higher mobility than other segments (Fig. 4). However, the presence of two peaks at lowfield resonance line (labeled M and N in Fig. 5A) suggests that the EPR spectra in this segment are composed of two components, which we termed M and N. The M component has similar mobility as the well ordered positions (Fig. 5B). Using spectral subtractions, we were able to reveal the lineshape of component N, whose relatively narrow lines indicate high mobility (Fig. 5). Therefore, even though center line width measurement, which has contributions from both components M and N, shows that residues 1-12 have higher mobility than other residues, spectral subtractions show that these residues consist of two structural states: a structured state and a locally disordered state. The structured state has mobility similar to segments 2 and 4 in Fig. 4.
GU-A␤42 Oligomers Adopt a Distinct Structure from Fibrils-Most amyloid fibrils studied to date adopt parallel in-register ␤-sheets (50), in which the side chains at the same residue position stack on top of each other. When spin label is introduced at a ␤-strand site in fibrils, the stacking of spin labels leads to strong spin exchange interactions (23,26). As demonstrated in Fig. 6A, strong spin exchange interactions (i.e. spin exchange frequency Ͼ100 MHz) manifest as single-line EPR spectra due to the collapse of the low-field and high-field resonance lines to the center-field line. The EPR spectra of 100% labeled GU-A␤42 oligomers show significant broadening as indicated by lower spectral amplitude compared with 20% labeled oligomers (Fig. 7). However, the EPR spectral lineshape of GU-A␤42 oligomers is distinct from the single-line spectrum of A␤42 fibrils (Fig.  6B). Therefore, the EPR data suggest that the structure of GU-A␤42 oligomers is distinct from the parallel in-register ␤-sheet structure of A␤42 fibrils.
Distance Analysis and Structural Modeling of GU-A␤42 Oligomers-To gain detailed structural information, we analyzed the intermolecular distances between spin labels in the GU-A␤42 oligomers by spectral fitting (Figs. 7 and 8). Distances were obtained for every labeling position. All labeling positions show two intermolecular distances at ϳ9 -10 and 15-17 Å (Fig.  8A). For the 9 -10 Å distance, there are two possible structural origins. First, in a ␤-sheet structure, the interstrand distance is 4.75 Å, thus the distance between alternate ␤-strands is 9.5 Å. Second, the distance between adjacent ␤-sheets is ϳ10 Å (Fig.  1C). A face-to-back packing between ␤-sheets would also give rise to inter-residue distance of ϳ10 Å for all residue positions. For the 15-17 Å distance, a likely origin is the spacing between every third ␤-strands within the same ␤-sheet. A structural model consistent with both distances is presented below.
Distance analysis also shows the populations for spin labels at each measured distance. The population of spin labels at 9 -10 Å increases from ϳ20% for N-terminal regions to ϳ40% for C-terminal regions (Fig. 8B), suggesting that the repeat giving rise to the 9 -10 Å distances has differing structural orders between N-and C-terminal regions. The population of spin labels at 15-17 Å account for ϳ30 -40% of total spins, and it remains relatively unchanged from N terminus to C terminus (Fig. 8B). There is also a significant population of spin labels that are further apart (Ͼ20 Å) and do not contribute to the broadening of the continuous-wave EPR spectra. The different populations of spin labels reveal structural heterogeneity in GU-A␤42 oligomers. The N-terminal region is least ordered, and ϳ50% of spin labels at each labeling site are Ͼ20 Å apart. For C-terminal regions, only 20 -30% of spin labels are Ͼ20 Å apart. A less ordered N-terminal region and more ordered FIGURE 3. EPR spectra of spin-labeled GU-A␤42 oligomers. The oligomers were prepared with spin-labeled and WT proteins at 1:4 molar ratio to minimize the effect of intermolecular spin-spin interactions on EPR lineshape. The sample preparation is referred to as "20% labeled" in the text. Scan width is 200 G. FIGURE 4. Residue-specific side chain mobility in GU-A␤42 oligomers. The inverse center line width (␦ Ϫ1 ) is determined using the EPR spectra of the 20% labeled GU-A␤42 oligomers, in which spin-labeled and wild-type GU-A␤42 are mixed at 1:4 molar ratio. The inset shows how the center line width was measured from the EPR spectrum. SEPTEMBER 26, 2014 • VOLUME 289 • NUMBER 39

JOURNAL OF BIOLOGICAL CHEMISTRY 27305
C-terminal region are consistent with the two-component analysis for the N-terminal residues in Fig. 5. In structural studies of A␤42 fibrils (51,52), the N-terminal region is also less ordered than C-terminal region, suggesting that some common mechanisms may underlie the assembly of both oligomers and fibrils.
The x-ray powder diffraction and near atomic-scale EPR distance information provide several structural restraints for prefibrillar A␤42 oligomers: (i) ␤-rich structure; (ii) sheet repeats of three strands; (iii) head-to-tail packing of A␤42 subunits within the same ␤-sheet, (iv) intramolecular antiparallel ␤-sheets; and (v) face-to-back packing of sheets. From these restraints, we developed an atomic model of prefibrillar A␤42 oligomers (Fig. 9). This model fits the powder diffraction data well, with a goodness of fit of 0.220 using the previously described R o metric (18), which has a scale of 0 (perfect fit) to 1 (worst possible fit). The best-fit model of prefibrillar A␤42 oligomers is composed of 16 A␤42 molecules arranged into four face-to-back packed ␤-sheets (Fig. 9A). Although the architecture of these oligomers is fibril-like, their size (ϳ60 ϫ 60 ϫ 35 Å in the absence of fusion protein partners) suggests that they are expected to have a globular appearance by transmission electron microscopy (Fig. 1A). A schematic of the model of prefibrillar A␤42 oligomers is shown in Fig. 9B to illustrate the measured distances from EPR. The model predicts another spin-spin distance at ϳ20 Å, which was not observed from the distance measurements. This is likely due to the fact that the upper limit of distance measurements with continuous-wave EPR is ϳ20 Å. Our analysis already revealed two interspin dis- Because of the similarity of these ordered spectra, an average spectrum was used as the approximation for the component M. Experimentally, EPR spectra are measured as the first-derivative of the absorption spectra. With increasing spin exchange interactions, the low-field (LF) and high-field (HF) resonance lines collapse with the center-field (CF) resonance line, giving rise to the so-called single-line spectra (red). Single-line spectra are characteristic of the parallel in-register ␤-sheet structure in amyloid fibrils. B, spectrum of A␤42 fibrils, spin-labeled at Leu-34, is characterized by a single-line feature, suggesting strong spin exchange interactions in parallel in-register ␤-sheet structures. In contrast, the spectrum of GU-A␤42 oligomers shows three spectral lines, suggesting that GU-A␤42 oligomers adopt a distinct structure from fibrils. Scan width is 200 G. tances at 9 -10 Å and 15-17 Å. Spin-spin interactions at ϳ20 Å produce very weak broadening in the EPR spectral lineshape. We attempted to include a third distance, which did not significantly improve the fitting. Therefore, we limited our analysis to two interspin distances. This structural model is also capable of accommodating the fusion protein partners, as illustrated in Fig. 9C. The fibril-like ␤-sheets in our oligomer model suggest that oligomers may be able to convert to fibrils through a mechanism of ␤-stand rotation (Fig. 9D).
GU-A␤42 Oligomers Promote the Fibrillization of A␤42-To further explore the relationship between the GU-A␤42 oligomers and full-length A␤42 oligomers (without the fusion protein partner), we studied the aggregation kinetics of A␤42 in the presence and absence of GU-A␤42 oligomers. As shown in Fig.  10, GU-A␤42 oligomers shortened the lag phase of the A␤42 aggregation. This suggests that GU-A␤42 oligomers likely adopt a structure that is on pathway to amyloid formation and support the possibility of oligomer-to-fibril conversion as proposed in Fig. 9D. In contrast, GU alone dramatically lengthened the lag phase of A␤42 fibrillization. This is likely due to the chaperone activity of GroES, consistent with reports in the literature that molecular chaperones generally have anti-amyloid activities (53).

DISCUSSION
Oligomeric assemblies of A␤, particularly A␤42, have been widely hypothesized as the primary neurotoxins that cause the pathology in Alzheimer's disease. Structural knowledge of A␤ oligomers is critical for understanding the structural basis of toxicity and the mechanism of Alzheimer pathogenesis. An accumulating body of evidence supports the notion that A␤ oligomers adopt a different structure from the parallel in-register ␤-sheet structure of A␤ fibrils (15)(16)(17)(18)(19). Particularly, Fourier transform infrared spectroscopy (FTIR) studies on A␤ oligomers prepared using various protocols (15)(16)(17)(18)(19) show a common 1695-cm Ϫ1 peak, which is absent in fibril samples, suggesting that antiparallel ␤ structure may be a common feature for A␤ oligomers (54). However, concerns have been raised regarding whether the 1695-cm Ϫ1 peak can distinguish parallel from antiparallel ␤-sheets (55,56). For example, Khurana and Fink (57) show that ␤-helix proteins give rise to an FTIR peak at ϳ1690 cm Ϫ1 . Ahmed et al. (16) interpreted the high frequency FTIR peak at 1675-1695 cm Ϫ1 as side chain vibrations of arginine, asparagine, and glutamine in their A␤42 oligomers. Detailed structural information such as inter-residue distances is very limited (16,58). Currently there are no structural models of oligomers based on extensive experimental restraints. Computational modeling studies have provided a number of structural models (59,60), but these models still await experimental validation.
In this work, we employed site-directed spin labeling to probe the side chain mobility (Fig. 4) and measure intermolecular distances (Fig. 8) at all 42 residue positions of A␤42 sequence in a prefibrillar oligomer preparation. Together with x-ray powder diffraction (Fig. 1C), the spin label mobility analysis reveals a turn at residues 24 -27, which connects two ␤-strands at residues 13-23 and 28 -42. This ␤-turn-␤ motif has also been observed in other A␤ oligomer preparations (61,62) and has been proposed in A␤42 fibril structural models (51,52). Mobility analysis also suggests that N-terminal residues consist of both a disordered state and a structured state (Fig. 5), consistent with recent hydrogen exchange and solid-state NMR   (Fig. 4) is shown in magenta. Shown on the right is a fibril model of A␤40 proposed previously (76). Prefibrillar oligomers may interconvert with fibril seeds by an ϳ90°rotation around the long axes of the ␤-strands. We term this conversion mechanism as strand rotation.
studies (61,62), which reveal a structured N-terminal segment. Distance measurements show two intermolecular distances at 9 -10 Å and 15-17 Å for every residue position, providing unambiguous support for the lack of parallel in-register ␤ structure in A␤42 oligomers. This work echoes a recent finding that A␤42 globulomers adopt structures distinct from fibrils (39).
Based on the EPR distance data, we developed a novel atomic model for A␤42 prefibrillar oligomers (Fig. 9A). This model is distinct from all existing A␤ oligomer models in terms of both the A␤42 tertiary structure and quaternary subunit packing. In our model, each A␤42 subunit forms a single ␤-sheet with three ␤-strands in an antiparallel arrangement. Each ␤-sheet consists of four A␤42 molecules in a head-to-tail arrangement. Four ␤-sheets are packed together in a face-to-back fashion. The ␤-sheets have internal helical axes correlated in phase with a central axis, and this feature resembles that of toxic A␤42 fibrillar oligomers (named TABFOs) (18).
The size of our oligomer model is in general agreement with the reported size of in vitro A␤42 oligomers. For example, size exclusion chromatography studies show that A␤-derived diffusible ligands are ϳ65-80 kDa, corresponding to 14 -18 subunits (63). Globulomers are shown to consist of 12 subunits (9). Prefibrillar oligomers are eluted as a 90-kDa peak using size exclusion chromatography, corresponding to 20 subunits (64). The essence of our structural model is the antiparallel triplestrand architecture. Because of the head-to-tail arrangement between A␤ subunits within the same ␤-sheet, and the face-toback packing between ␤-sheets, our oligomer model is openended in both the backbone hydrogen bonding direction and the side chain direction, allowing the growth of the oligomer in size.
The fibril-like appearance of ␤-sheets in prefibrillar oligomers suggests that prefibrillar oligomers may interconvert with fibrils via a transition from a ␤-hairpin to a heterosteric zipper (Fig. 9D). The EPR distance restraints determined herein (Fig. 8) dictate that intersheet interactions are mediated by identical segments in oligomers (Fig. 9). Oligomers may convert to fibril seeds by strand rotation, wherein these identical segments rotate about 90°around their ␤-strand axes to make ␤-sheets of identical segments. The strand rotation model is illustrated in Fig. 9D. Such strand rotations would be facilitated by flexibility in the turn connecting the two ␤-strands. Indeed, our EPR data unambiguously reveal that this turn is flexible (Fig. 4). Furthermore, the possibility of oligomer-to-fibril conversion is supported by the result that GU-A␤42 oligomers reduced the lag phase of A␤42 aggregation (Fig. 10).
Nussinov et al. previously modeled various triple ␤-sheet structures for A␤42 amyloid (65). In their model, each A␤42 molecule consists of three ␤-strands with two turn regions: one at residues 25-29, and the other one at residues 9 -14. Each ␤-strand participates in a different ␤-sheet in the triple ␤-sheet structure, which has a better correlation with the hydrogen exchange data than the double ␤-sheet structure (65). The triple ␤-sheet structure has been previously reported for A␤40 fibrils based on solid-state NMR data (66). Hydrogen exchange studies of A␤42 fibrils also found that N-terminal region is protected in approximately half of the population (52). These studies highlighted the structural importance of N-terminal region, which is likely structured in the amyloid fibrils of both A␤40 and A␤42. In the structural model of A␤42 oligomers, we also show that each A␤42 molecule consists of three ␤-strands at similar residue positions as previously found in A␤42 fibrils (52,65). At the tertiary structure level, however, three ␤-strands from the same A␤42 molecule participate in the same ␤-sheet in an antiparallel fashion. Therefore, the A␤42 structure in the oligomers can be characterized as triple-strand, single ␤-sheet, which distinguishes it from the A␤42 structure in fibrils.
Structural heterogeneity or polymorphism has been observed for A␤ fibrils (67,68). For A␤ oligomers, different oligomer preparation protocols have been reported, and these oligomers are classified largely by morphology, size, and immunoreactivity to mono-and polyclonal antibodies (12). The findings in this work allow us to assess the heterogeneity of the underlying molecular structure in A␤42 oligomers. The EPR spectral lineshape (Fig. 3) and spin label mobility profile (Fig. 4) suggest that residues 13-42 adopt a single ␤-turn-␤ conformation at secondary structure level. This ␤-turn-␤ secondary structure is very similar to those observed in A␤42 fibrils (51,52). This similarity can be rationalized in the general framework of hierarchical protein folding (69). Assuming the same force is driving both fibril and oligomer formation, it is not surprising to see structural similarity between oligomers and fibrils at secondary structure level. In contrast to the rest of A␤42 molecule, the N-terminal residues 1-12 show significant amount of heterogeneity. In the N-terminal region, a locally disordered conformation co-exists with a well-ordered conformation. The population of the disordered conformation is ϳ10 -20% of the total population (Fig. 5). In A␤42 fibrils, a previous hydrogen exchange study shows that ϳ50% of the total population for the N-terminal region is structured (52). The higher percentage of the ordered population in oligomers is likely due to the presence of the fusion protein partner, which traps and stabilizes the oligomeric state. Intermolecular distance measurements reveal another layer of structural heterogeneity at quaternary structure level. We observed two major distance distributions at 9 -10 Å and 15-17 Å, but the spin label population for these two distances combined is ϳ50 -70%. The rest of the spin labels, at ϳ30 -50% of total population, give rise to distances over the detection limit of continuous-wave EPR, which is ϳ20 Å (Fig. 8B). The spin label population with Ͼ20 Å distances may represent other structures that are not currently modeled. The spin label mobility studies (Figs. 3 and 4) shows that these other structures are also highly ordered, at least at secondary structure level.
Our modeling explains both the distances at 9 -10 Å and 15-17 Å with a unifying structure (Fig. 9). It is likely that other structures satisfying only one set of distances also exist. In Fig.  11, we modeled two such alternative structures in which only FIGURE 11. Alternative models for prefibrillar A␤42 oligomers. In alternative model A, the A␤42 oligomer adopts a triple ␤-sheet structure. Alternative model B is very similar to model A except for the disordered N-terminal region. Balls represent spin labels to demonstrate the 10 Å spacing in model A. Only 6 A␤42 molecules are modeled in each structure to show basic architecture, without consideration of the size of the oligomer. Note that the 15 Å distance is not accounted for in these alternative models. FIGURE 12. Schematic summary of oligomer-fibril interconversion mechanisms. An amyloid protein (e.g. A␤42) is depicted as a U-turn of two ␤-strands. The green ␤-strand is N-terminal to the yellow ␤-strand. Oligomers are known to take many forms, two of which are represented here: ␤-cylindrin (77) (Oligomer 1, represented as a cylinder) and cross-␤ oligomers (Oligomer 2, represented as a stack of U-turns) (Stroud et al. (18) and this work). Fibril seeds have an amyloid spine made of a pair of ␤-sheets, represented as laminae upon which ␤-strands are superimposed). Unlike oligomers, which are limited in size, fibrils have either one or two growing ends onto which monomers may add indefinitely. A, in conversion by homogeneous nucleation, oligomers (shown here as a cylinder), must first dissociate before nucleating to form fibril seeds. B, in conversion by secondary nucleation, oligomers with fibril-like structure (as described in the present work) may add new monomers that form a fibril repeat different from the oligomer repeat. In the context of amyloid, such differences in repeats are known as polymorphisms. The example of polymorphism conversion by packing polymorphism is shown here. C, oligomers may convert to fibrils by nucleated conformational conversion (35). Two types of nucleated conformational conversion have been proposed: barrel unrolling and strand rotation. In conversion by barrel unrolling, an amyloid oligomer ␤-barrel (i.e. ␤-cylindrin) breaks open by dissolution of ␤-hydrogen bonding between one pair of ␤-strands to produce a linear ␤-sheet (77). To form a fibril seed, two such linear sheets must interact through lateral association. Strand rotation conversion, described in the present work and elsewhere (74,75), entails the rotation of ␤-strands by ϳ90°around their long axes. This rotation turns the ␤-sheets of the oligomer (running horizontally in the bottom panel) into fibril ␤-sheets (running vertically). the 9 -10 Å distance is accounted for. In alternative model A (Fig. 11), each A␤42 molecule adopts the same three ␤-strands, but each ␤-strand participates in a different ␤-sheet in an antiparallel fashion. Except for the N-terminal ␤-strand, the rest of the residues adopt a structure that is very similar to the antiparallel structure of A␤40 D23N fibrils (70). Although conceptually similar to the triple ␤-sheet models of Nussinov et al. (65), this alternative model distinguishes itself in the N-terminal region, which folds back to the double ␤-sheet of the C-terminal region. In alternative model B (Fig. 11), the N-terminal region adopts a locally disordered conformation. This may partly explain the structural heterogeneity observed for this region in our spin label mobility analysis (Figs. 4 and 5) and distance measurements (Fig. 8).
Our structural models have implications about the toxicity of oligomers. Several different oligomer preparations have been shown to have cytotoxicity (8 -11). If different oligomers exert their toxicity through similar mechanisms, these oligomers may have similar structural features underlying their toxicity. This work and previous studies (15)(16)(17)(18)(19)39) suggest that one common structural feature observed in different oligomers is the antiparallel ␤-sheet. Oligomer-specific polyclonal antibody A11 recognizes pore-forming bacterial toxin ␣-hemolysin and block its toxicity (71). The membrane-spanning core of ␣-hemolysin is a ␤-barrel consisting of antiparallel ␤ structures (72). The antiparallel structure may explain the reactivity of our GU-A␤42 oligomer with the A11 antibody. An alternative explanation to toxicity is that toxicity is associated with properties that are related to the aggregation process and are not specific to particular oligomer species (73). One candidate for such a property is hydrophobicity. In A␤ fibrils, due to symmetric packing of protofilament, the C-terminal hydrophobic region is packed inside the fibril core (50). In the oligomer model of this work (Fig. 9), the packing between adjacent ␤-sheets is face to back, so more hydrophobic residues are exposed to solvent compared with A␤ fibrils.
Structural conversion from oligomers to fibrils is a critical step in A␤ aggregation. In Fig. 12 we summarize several potential mechanisms for oligomer-fibril interconversion. The mechanism of strand rotation for oligomer-fibril interconversion has been proposed for A␤ based on biochemical studies of A␤40 protein (74,75). Here we provide direct structural evidence for strand rotation conversion in that (i) exhaustive EPR distance restraints dictate the model for prefibrillar oligomers and (ii) EPR mobility measurements demonstrate the essential flexibility of the turn connecting the two ␤-strands. Conversion between toxic oligomers and non-toxic fibrils may represent an attractive point of therapeutic intervention to treat Alzheimer disease. Understanding the conversion mechanisms is essential for rational design of potential therapeutics targeting this process.