Expansion of Polyglutamine Induces the Formation of Quasi-aggregate in the Early Stage of Protein Fibrillization*

We examined the effects of the expansion of glutamine repeats on the early stage of protein fibrillization. Small-angle x-ray scattering (SAXS) and electron microscopic studies revealed that the elongation of polyglutamine from 35 to 50 repeats in protein induced a large assembly of the protein upon incubation at 37 °C and that its formation was completed in ∼3 h. A bead modeling procedure based on SAXS spectra indicated that the largely assembled species of the protein, quasi-aggregate, is composed of 80 to ∼90 monomers and a bowl-like structure with long and short axes of 400 and 190 Å, respectively. Contrary to fibril, the quasi-aggregate did not show a peak at S = 0.21 Å–1 corresponding to the 4.8-Å spacing of β-pleated sheets in SAXS spectra, and reacted with a monoclonal antibody specific to expanded polyglutamine. These results imply that β-sheets of expanded polyglutamines in the quasi-aggregate are not orderly aligned and are partially exposed, in contrast to regularly oriented and buried β-pleated sheets in fibril. The formation of non-fibrillary quasi-aggregate in the early phase of fibril formation would be one of the major characteristics of the protein containing an expanded polyglutamine.

Amyloid is a fibrillar deposit observed in various proteins associated with human neurodegenerative diseases including Alzheimer's diseases, Creutzfeldt-Jakob diseases, and polyglutamine diseases. Amyloid has a "cross-␤ structure," with ␤-strands perpendicular to, and backbone hydrogen bonds parallel to, the fibril axis (1,2). Recent in vitro studies have revealed that an intermediate termed protofibril is transiently formed prior to the formation of the amyloid fibrils of amyloid-␤ peptide and ␣-synuclein, the depositions of which in brain are hallmarks of Alzheimer's disease and Parkinson's disease, respectively (3)(4)(5)(6)(7). The protofibril as well as oligomeric states of amyloid-␤ peptide has been reported to induce cellular dysfunction and neuron death (8 -10). Recently, Bucciantini et al. (11) showed that non-disease-related proteins might generally form globular intermediates prior to the formation of amyloid fibrils and that the intermediates induce cellular toxicity. Thus, studies on the earliest phase of protein fibrillization, in particular the oligomerization and nucleation processes, are of great importance for identifying therapeutic targets as well as for understanding the molecular mechanism of fibril formation. In addition, structural characterization of such globular precursors of amyloid fibril would be important to reveal origins of the cellular toxicity of the amyloid-forming proteins.
Amyloid fibrils are also observed in the aggregates of proteins bearing an expanded polyglutamine (12,13). Expansion of polyglutamine to more than ϳ35 repeats in certain proteins induces the formation of intranuclear inclusions that are characteristic of polyglutamine diseases such as Huntington's disease and hereditary spinocerebellar ataxia (14). Although the protein aggregate including amyloid fibril might be closely involved in the cellular dysfunction that causes polyglutamine diseases (15), little is known about the effects of the expansion of glutamine repeats on the aggregation mechanism of the causative proteins. Although Poirier et al. (16) recently reported spheroids and protofibrils as precursors in the fibrillization of the N-terminal fragment of mutant huntingtin containing 44 glutamine repeats, an early phase of the fibrillization process of the protein has not been fully characterized. In particular, structural and physico-chemical properties of such precursors have remained unclear.
We have recently designed sperm whale myoglobin (Mb 1 ) mutants in which a varying length of glutamine repeats was inserted (17). Although the expansion of glutamine repeat in general renders the protein insoluble in water, we successfully established preparation of the mutant Mbs containing 50 (Mb-Gln 50 ) or 35 (Mb-Gln 35 ) glutamine repeats on a large scale, which is required for their structural analysis. 50 glutamine repeats inserted into protein are pathological, whereas 35 glutamine repeats lie between pathological and non-pathological status in polyglutamine diseases. In our previous study (17), Mb-Gln 50 and Mb-Gln 35 , but not a mutant Mb bearing shorter, 12 glutamine repeats (Mb-Gln 12 ), reacted with a monoclonal 1C2 antibody specific for an expanded polyglutamine (18), suggesting that the expanded polyglutamines in Mb-Gln 50 and Mb-Gln 35 form a characteristic structure similar to that of the native proteins that cause polyglutamine diseases. In addition, the tendency to form amyloid fibrils was highly dependent on the length of glutamine repeats inserted into Mb. Mb-Gln 50 and Mb-Gln 35 formed amyloid fibrils under physiological conditions rapidly and slowly, respectively, whereas Mb-Gln 12 and Mb wild-type (Mb-WT) did not form any amyloid fibrils under the same conditions (17). We suggest, by these observations, that Mb-Gln 50 and Mb-Gln 35 would be appropriate to investigate the effects of the expansion of glutamine repeats on the fibrillization process of polyglutamine-bearing proteins. To elucidate the formation mechanism of fibril would be crucial for better comprehension of the molecular basis for polyglutamine diseases.
In the present study, we applied small-angle x-ray scattering (SAXS) to examine and compare the early stage of fibril formation of the mutant Mbs, because SAXS is a powerful technique for quantitative detection of smaller particles with accurate time dependence, compared with light scattering. We revealed that the expansion of polyglutamine from 35 to 50 repeats in a mutant Mb induced a large assembly of the protein in the early phase of fibrillization. The formation of a largely assembled species of Mb-Gln 50 was confirmed by electron microscopy. We further examined structural properties of ␤-sheets of expanded polyglutamines in the Mb-Gln 50 assembled species and determined its low-resolution structure based on SAXS spectra.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Recombinant wild-type and mutant Mbs were expressed in TB1 Escherichia coli and purified as reported previously (17). The concentration of ferric Mbs was determined by the extinction coefficient of 157 mM Ϫ1 ⅐ cm Ϫ1 at the Soret band in the UV-visible spectra. The sample solutions of Mbs in 10 mM MOPS buffer at pH 7.0 were centrifuged at 25,000 ϫ g for 30 min, before incubation at 37°C, to remove any pre-existing precipitates.
Congo Red Binding Assay-We incubated 100 l of 1.86 mg/ml Mb-Gln 50 and Mb-Gln 35 and 4.49 mg/ml Mb-Gln 35 , Mb-Gln 12 and Mb-WT solutions at 37°C. We also incubated the supernatant fraction (1.86 mg/ml) of a Mb-Gln 50 solution, which was pre-incubated at 37°C for 6 h at 37°C. The pellet of the Mb-Gln 50 solution, which was pre-incubated for 6 h, was resuspended with 100 l of 10 mM MOPS buffer (pH 7.0) and seeded into fresh 1.86 mg/ml Mb-Gln 50 and 4.49 mg/ml Mb-Gln 35 solutions (10% (v/v)). At appropriate time points, 5 l of the protein solutions was added to 10 M Congo red in 10 mM MOPS buffer (pH 7.0) and incubated at room temperature for 30 min. UV-visible spectra of these samples were measured with a Shimadzu 2400-PC spectrophotometer. Congo red bound to fibril was determined using the following equation: Congo red (mol/l) ϭ A 540 /25,295 Ϫ A 480 /46,306 (8).
Small-angle X-ray Scattering (SAXS)-The measurement of SAXS was performed by synchrotron radiation of RIKEN structural biology beamline I (BL45XU) at Spring-8 in Harima, Japan (19). The sampleto-detector distance was 1 or 2.2 m, which was calibrated using meridional diffraction of dried chicken collagen. The temperature of the sample was maintained at 37°C by an incubator, and the sample cell and stage were also set to 37°C. Using an x-ray image intensifier and cooled CCD detector (XR-IIϩCCD) (20), each scattering profile was collected for 1 s during which no radiation damage was found. The data were normalized to the intensity of the incident beam, and the buffer was subtracted. The radius of gyration (R g ) was determined by the Guinier approximation: I(S) ϭ I(0) exp(Ϫ4 2 Rg 2 S 2 /3), where S and I(0) are the momentum transfer and intensity at the zero scattering angle, respectively, with fitting ranges of S 2 (Å Ϫ2 ) from 5 ϫ 10 Ϫ6 to 40 ϫ 10 Ϫ6 . S is defined as S ϭ 2sin/, where 2 and are the scattering angle and the x-ray wavelength, respectively (21). The S range used for R g determination satisfied the condition 2 SR g Ͻ 1.3. The distance distribution function, P(r), was calculated by GNOM, which uses an indirect Fourier transform method (22). The maximum dimension, D max , was determined from the first zero cross-point of the P(r) function (21). The cross-sectional radius of gyration, R c , was determined by fitting the data points, using the equation I(S)S ϭ I c (0)exp(Ϫ 2 R c 2 S 2 /2) over S 2 ranges (Å Ϫ2 ) from 2 ϫ 10 Ϫ6 to 10 ϫ 10 Ϫ6 and from 8 ϫ 10 Ϫ6 to 23 ϫ 10 Ϫ6 for oligomer and quasi-aggregates, respectively. The spectral analysis by singular value deconvolution (SVD) was performed using SPECFIT (version 3.0, Spectrum Software Associates, Marlborough, MA), using the S versus S 2 ϫI(S) (Kratky) plot of the data.
Construction of Low-resolution Model-Low-resolution particle shapes were restored from SAXS intensity profiles using a bead modeling procedure of DAMMIN (23). In the dummy atom minimization (DAMMIN), a protein molecule was approximated by densely packed small spheres (dummy atoms). Minimization was performed by the simulated annealing method (22), starting from the dummy atoms placed at random coordinates within the search sphere, sphere of diameter D max . The detailed procedure of DAMMIN has been described previously (24). The stability of the model was checked by repeating the minimization 10 times in different runs. The three-dimensional models were displayed using the program VMD.
Electron Microscopy and Immunoblotting-The negative staining for the 1.86 mg/ml of sample of Mb-Gln 50 and the 4.49 mg/ml of samples of Mb-Gln 35 , Mb-Gln 12 , and Mb-WT was performed as reported previously (17), and the images were recorded on a LEO 912AB electron microscope (LEO, Cambridge, UK) normally at a magnification of 63,000 ϫ g. For immunoelectron microscopy of Mb, the copper grid was treated with a polyclonal Mb (1:250 -500, Chemicon) or 1C2 antibody (1:50 -100, Chemicon), followed by a 5-nm gold-conjugated rabbit or mouse secondary antibody, respectively (1:200, Amersham Biosciences). Each 1.86 mg/ml of sample of Mb-Gln 50 with various incubations (0, 3, 10, 24, and 72 h) at 37°C was also processed to immunoblotting by the Mb (1:5000) and 1C2 (1:2000) antibodies.
X-ray Diffraction Experiment-The sample-to-detector distance was 30 cm, which was calibrated using meridional diffraction of dried chicken collagen. Mb-Gln 50 and Mb-Gln 35 fibrils grown from ϳ5 mg/ml of solution in 10 mM MOPS buffer at pH 7.0 were collected by brief centrifugation and mounted onto the last guard slit with a width of 0.6 mm. The sample-to-detector distance was 30 cm. The sample holder was vacuumized to remove the scattering of air, and the fibrils were exposed to x-ray radiation operating at ϳ90 mA. Data were collected at room temperature for 30 s on a modified RIGAKU R-Axis IV 2ϩ imaging plate detector. 2 Size Exclusion Chromatography-A 1.86 mg/ml of solution of Mb-Gln 50 and 4.49 mg/ml of solutions of Mb-Gln 35 and Mb-Gln 12 were incubated at 37°C, and 40 l of each sample at 3, 10, and 24 h was centrifuged at 25,000 ϫ g for 20 min. The supernatant was analyzed by size exclusion chromatography using the Superdex-200 in the SMART system (Amersham Biosciences). The running buffer was 10 mM MOPS containing 150 mM NaCl at pH 7.0, and the flow rate was 40 l/min.

Formation of Quasi-aggregate by Expanded Polyglutamine-We examined the fibril formation of mutant Mbs by a
Congo red binding assay. We incubated separate solutions (1.86 mg/ml) of Mb-Gln 50 and Mb-Gln 35 at 37°C and determined the amount of Congo red bound to fibril, using UVvisible spectroscopy (8). The time-dependent profiles showed that the fibril formation of Mb-Gln 50 was initiated in ϳ50 h, whereas Mb-Gln 35 did not clearly exhibit fibrillization within 200 h (Fig. 1A). We then increased the concentration of a Mb-Gln 35 solution from 1.86 mg/ml to 4.49 mg/ml and examined the fibril formation by the Congo red binding assay. The 4.49 mg/ml of solution of Mb-Gln 35 formed fibrils slowly with a lag time of ϳ150 h (Fig. 1A). Contrary to Mb-Gln 50 and Mb-Gln 35 , 4.49 mg/ml of solutions of Mb-Gln 12 and Mb-WT did not show the binding to Congo red upon incubation Ͼ200 h (Fig.  1A), indicating no fibril formation as reported previously (17). We also verified that the lag time of fibrillization did not change when we used mutant Mbs of different sample preparations.
Because Mb-Gln 50 and Mb-Gln 35 formed fibrils under the physiological condition, we examined the early phase of fibrillization of the mutant Mbs by SAXS. We incubated separate solutions (1.86 mg/ml) of the mutant Mbs at 37°C and measured their SAXS spectra with various incubation times. The Guinier plot of Mb-Gln 50 showed an increase in the intensity at S 2 ϭ ϳ1 ϫ 10 Ϫ3 Å Ϫ2 upon incubation at 37°C, suggesting formation of an assembled species of the mutant Mb (data not shown). We plotted the scattering profiles of Mb-Gln 50 as S versus S 2 ϫ I(S) (Kratky plot) to clarify the protein assembly. As shown in Fig. 1B, an intensity of the peak at S ϭ 1.4 ϫ 10 Ϫ3 Å Ϫ1 was clearly increased with incubation times. The peak showed a maximum scattering intensity in ϳ3 h of incubation, and the intensity was constant until 27 h (Fig. 1C). Because a peak at the extremely small-angle region (S ϭ 1.4 ϫ 10 Ϫ3 Å Ϫ1 ) corresponds to a large-sized particle, our result indicated that a largely assembled Mb-Gln 50 was formed in the early time of the lag phase (ϳ50 h) of fibrillization. Hereafter, we referred such a largely assembled species of protein, which was formed in the early stage of the Mb-Gln 50 fibrillization, as a quasiaggregate. In contrast to Mb-Gln 50 , SAXS spectra of a 1.86 mg/ml of solution of Mb-Gln 35 did not change upon incubation at 37°C for more than 20 h, suggesting an absence of such an assembled species of protein as observed for Mb-Gln 50 (data not shown). Thus, we measured SAXS spectra of a 4.49 mg/ml of solution of Mb-Gln 35 (Fig. 1D), which facilitated the fibril formation as shown in Fig. 1A. However, the peak at S ϭ 1.0 ϫ 10 Ϫ3 Å Ϫ1 did not increase with incubation times (Fig. 1E), which confirmed the absence of an assembled species of protein in the early stage of the fibrillization for Mb-Gln 35 . In addition, we did not detect any peak at the extremely small-angle region in SAXS spectra for Mb-Gln 12 and Mb-WT upon incubation at 37°C over 50 h, indicating absence of any largely assembled species of protein (data not shown). The reliability of time-dependent SAXS spectra was verified by repeating measurements with mutant Mbs of different sample preparations and with different incubation times.
To gain more physico-chemical insights into the formation of the quasi-aggregate of Mb-Gln 50 , we measured time-dependent SAXS spectra at various temperatures (25,29,33, and 37°C) for Mb-Gln 50 . The formation of quasi-aggregates was greatly affected by temperature, and representative scattering profiles at 25°C and 33°C are shown in Fig. 2, A and B. The scattering intensity at S ϭ 1.4 ϫ 10 Ϫ3 Å Ϫ1 was plotted against incubation times, and the formation rate of the quasi-aggregate was calculated. We found that these plots were best fitted by a single exponential (Fig. 2C), indicating that the formation of the qua-si-aggregate follows the first-order kinetics. We also show an Arrhenius plot of the formation rate of the Mb-Gln 50 quasiaggregate in Fig. 2D. The activation energy for the formation of the quasi-aggregate was calculated from the slope of the Arrhenius plot (22.8 Ϯ 1.8 kcal/mol). Furthermore, we examined the effects of the Mb-Gln 50 quasi-aggregate on fibrillization of Mb-Gln 50 or Mb-Gln 35 by the Congo red binding assay. Elimination of the quasi-aggregate, which accumulated upon incubation at 37°C for 6 h, in the Mb-Gln 50 sample prolonged a lag time of fibrillization from ϳ50 to ϳ70 h (Fig. 1A). On the other hand, addition of the Mb-Gln 50 quasi-aggregate into a fresh Mb-Gln 50 or Mb-Gln 35 solution as a seed eliminated or shortened the lag time, respectively (Fig. 1A).
We ascertained the formation of quasi-aggregate by electron microscopy. We carried out the negative staining for mutant Mbs, which had been incubated at 37°C for 3 h, and observed them by electron microscopy. A typical electron micrograph is shown in Fig. 3A. We clearly detected particles with a length of ϳ20 -45 nm in the Mb-Gln 50 sample, whereas such particles were not observed for Mb-Gln 35 , Mb-Gln 12 , and Mb-WT samples (data not shown). This electron micrograph suggested that the observed particles for Mb-Gln 50 correspond to the quasiaggregate, which was indicated in the SAXS experiment for Mb-Gln 50 (Fig. 1B), because the formation of quasi-aggregate was completed in 3 h (Fig. 1B). We further performed an immunoelectron microscopic study for the quasi-aggregate with a polyclonal Mb and a monoclonal 1C2 antibody (18). Although some of the quasi-aggregates initiate to assemble each other by further incubation during the immunostaining procedure, the quasi-aggregate reacted with both of the Mb and 1C2 antibodies (Fig. 3, B and C). We also investigated the reactivity of these antibodies to the Mb-Gln 50 fibrils that were formed by incubation at 37°C for 7 days. The Mb antibody reacted with the fibrils along the fibers (Fig. 3D), whereas the 1C2 antibody recognized only ends and branch sites of the fibrils (Fig. 3E). We further investigated this observation by the following immunoblotting experiment. Each sample of Mb-Gln 50 with various incubations (0, 3, 10, 24, and 72 h) were separated by SDS-PAGE and processed for immunoblotting by the Mb and 1C2 antibodies. We focused on the gel top in the immunoblot, because the quasi-aggregate and/or fibril can be detected at the gel top. We found a clear difference in the immunoreactivity to the gel top in 72 h between the Mb and 1C2 antibodies. Although the Mb antibody stained the gel top in 72 h, the reactivity of the 1C2 antibody to the gel top was obviously decreased in 72 h (Fig. 3F).
Furthermore, we followed the morphological changes of Mb-Gln 50 upon incubation at 37°C for 1 to ϳ7 days, using electron microscopy. The quasi-aggregates of Mb-Gln 50 were associated with each other to form largely assembled clusters, but fibrils were not observed at 1 day (data not shown), whereas fibrils were clearly detected at 3 days. We display typical electron micrographs of Mb-Gln 50 at 3 or 5 days in Fig. 3, G-I. In the electron micrographs, we frequently observed the fibrils that were connected to and associated with clusters of the quasiaggregates (arrow, Fig. 3, G-I).
Models of Mb-Gln 50 Monomer, Oligomer, and Quasi-aggregates-To gain more structural information on the quasi-aggregate, we attempted to perform SVD analysis for the timedependent SAXS spectra. Because the SVD analysis required more data points in the first h of protein fibrillization, we measured SAXS of Mb-Gln 50 (1.86 mg/ml), again in shorter incubation intervals from 1 to 60 min at 37°C (Fig. 4A). The resultant 11 scattering profiles were analyzed using SPECFIT. The SVD analysis showed two species, and the scattering profiles were best fitted by single exponentials. This analysis together with the presence of an isoscattering point at S ϭ 4.4 ϫ 10 Ϫ3 Å Ϫ1 in the SAXS spectra (Fig. 4A) indicated that this reaction follows the first-order kinetics. Therefore, we deconvoluted the time-dependent SAXS profiles of Mb-Gln 50 into two spectra, which displayed the scattering profiles of the initial (0 min) and final (60 min) states of Mb-Gln 50 . In the first-order kinetic scheme, the scattering profile of the final state (60 min) corresponded to that of the quasi-aggregate. We calculated Kratky plots of the two states of Mb-Gln 50 and show them in Fig. 4B. From the scattering profiles, we calculated R g values of the initial and final states of Mb-Gln 50 (68 and 146 Å, respectively). We realized that the initial state (0 min) was not a monomer but an oligomer, because the R g of 68 Å was larger than that of the Mb-Gln 50 monomer (27 Å) (25). We found, by SAXS and dynamic light scattering analyses, that the Mb-Gln 50 monomer is rarely isolated as a complete monomeric state and that some oligomerized species are formed during the purification. From the SAXS spectra, we also calculated crosssectional radii of gyration, R c , of the Mb-Gln 50 oligomer (30 Å) and quasi-aggregate (92 Å). In addition, D max values of the oligomer (213 Å) and quasi-aggregate (463 Å) were determined by the pair distance distribution functions using GNOM. We found that these values were not changed by different data sets, confirming the reliability of the scattering profiles of the Mb-Gln 50 oligomer and quasi-aggregate. We summarized the structural parameters of the Mb-Gln 50 monomer, oligomer, and quasi-aggregate in Table I.
On the basis of the SAXS spectra, bead models of the Mb-Gln 50 oligomer and quasi-aggregate were constructed by a fitting procedure, using DAMMIN. The calculated model structures are illustrated in Fig. 4C. The converged model of the oligomer showed an elongated structure that has long and short axes of 210 and 50 Å, respectively. The model structure of the quasi-aggregate showed a bowl-like shape with some torsion, having long and short axes of 400 and 190 Å, respectively, which approximately corresponded to those observed by electron microscopy in Fig. 3A. The model structure of the Mb-Gln 50 monomer was also constructed from the SAXS spectrum by the bead modeling shape-determination method. We calculated the particle volumes of the Mb-Gln 50 monomer, oligomer, and quasi-aggregate and listed them in Table I. We obtained the relative ratios between monomer, oligomer, and quasiaggregate and revealed that the oligomer and quasi-aggregate are composed of 4 to ϳ5 and 80 to ϳ90 monomers, respectively.
␤-sheet Structure of Polyglutamine in Quasi-aggregate and Fibril-We found, by the SAXS experiment, that the expansion of polyglutamine from 35 repeats to 50 caused the different mechanism in the early stage of protein fibrillization. Next, we explored the effects of the expansion of polyglutamine on the end-state fibril structure by x-ray diffraction analysis for the Mb-Gln 50 and Mb-Gln 35 fibril. The x-ray diffraction patterns of both dried fibrils showed a strong ring at a spacing of ϳ4.8 Å and a weak diffraction ring at the 8 to ϳ 9 Å region (Fig. 5, A  and B). To inspect the spacing values, we plotted the circular averaged intensity against S (Å Ϫ1 ) (Fig. 5C). This analysis clearly showed diffraction rings at spacings of 4.79 Å Ϫ1 and 8.33 Å, and 4.83 Å Ϫ1 and 8.42 Å for the Mb-Gln 50 and Mb-Gln 35 fibril, respectively. Although both spacing values of the diffrac-tion rings for the Mb-Gln 50 fibril were slightly smaller than those for the Mb-Gln 35 fibril, the spacing values did not change substantially by the expansion of glutamine repeats.
Because the Mb-Gln 50 fibril diffracted at the 4.8-Å spacing very strongly, we expected that a peak at S ϭ 0.21 Å Ϫ1 corresponding to the 4.8-Å spacing in ␤-pleated sheets would be observed in the high-S region of SAXS spectra. The peak at S ϭ 0.21 Å Ϫ1 was indeed unambiguously detected for the Mb-Gln 50 sample after incubation at 37°C for 120 h (bold arrow in Fig.  5D), when formation of the Mb-Gln 50 fibril was almost completed (Fig. 1A). We examined whether this peak is detected during formation of the quasi-aggregate. SAXS spectra of Mb-Gln 50 were measured after incubation at 37°C for 3 to ϳ12 h (Fig. 5D). As a result, we could not observe the peak at S ϭ 0.21 Å Ϫ1 in 3 h when formation of the quasi-aggregate was almost completed (Fig. 1B). Similarly, the incubation for 12 h was still insufficient for detection of the peak at S ϭ 0.21 Å Ϫ1 , although formation of the quasi-aggregate is definitely completed in 12 h (Fig. 1B).
Gel Filtration Experiment-We investigated whether we can isolate the quasi-aggregate of Mb-Gln 50 as a soluble state. A 1.86 mg/ml of solution of Mb-Gln 50 was incubated at 37°C for 3, 10, and 24 h, and the supernatant fraction was separated by size exclusion chromatography. Because the quasi-aggregate is composed of 80 to ϳ90 monomers of 26,000 (Table I), the molecular weight of the quasi-aggregate was estimated to be 2,000,000ϳ2,500,000, suggesting that the quasi-aggregate should be eluted at the void volume. However, we could not detect a peak at the void volume (data not shown), indicating that the quasi-aggregate does not exist in the supernatant but in the pellet fraction. We also verified absence of a peak at the void volume position for respective samples of Mb-Gln 35 and Mb-Gln 12  at 26,000 in the immunoblot (data not shown). Because we did not observe any fibrils at 1 day by electron microscopy, this result indicated that the material in the pellet fraction is the quasi-aggregate and that the quasi-aggregate was partially dissociated into monomers by the treatment with SDS.

Effect of Expansion of Glutamine Repeats on the Fibrillization
Process-The x-ray diffraction analysis revealed that both diffraction spacings in the Mb-Gln 50 fibrils (4.79 and 8.33 Å) are slightly smaller than but quite similar to those in the Mb-Gln 35 fibrils (4.83 and 8.42 Å) (Fig. 5, A and B). The diffraction rings at ϳ4.8 and ϳ8.4 Å correspond to the hydrogenbond distances between intra-and inter-␤-sheets, respectively (26). Because the inserted polyglutamine forms the ␤-sheet structure in the mutant Mbs (17), we suggest that the ␤-pleated sheets in fibril were constituted by the inserted glutamine repeats. Although intra-and inter-␤-sheets in the Mb-Gln 50 fibril were slightly more compact, presumably because of the longer glutamine repeats than those in the Mb-Gln 35 fibril, we can conclude that the expansion of glutamine repeats did not have substantial effects on the ␤-sheet structure in fibril. Interestingly, Perutz et al. (27) recently reported the same spacing values (ϳ4.8 and ϳ8.4 Å) in the diffraction ring of fibrils for N-terminal huntingtin, which contains similar glutamine repeats (Gln 51 ) to Mb-Gln 50 , suggesting that a similar length of polyglutamine, even in the different context of protein, forms a similar ␤-pleated sheet structure in fibril.
Contrary to the similar fibril structures of the mutant Mbs, we revealed a major difference in the early stage of the fibrillization process between Mb-Gln 50 and Mb-Gln 35 . We demonstrated, by SAXS spectra and electron micrographs, that a largely assembled, non-fibrillar species of protein, quasi-aggregate, was formed in the early phase of fibrillization of Mb-Gln 50 . The formation of quasi-aggregate was almost completed within 3 h of incubation at 37°C, and the amount of quasiaggregate was constant until 27 h (Fig. 1B). Because the lag time in the fibril formation of Mb-Gln 50 was ϳ50 h (Fig. 1A), the quasi-aggregate was formed in the early stage of the lag phase, namely the nucleation process of fibrillization, and the amount of quasi-aggregate was constant until at least half-time of the lag phase. On the basis of SAXS spectra of the quasiaggregate, we constructed the bead model and determined the shape and dimensions of the quasi-aggregate (Fig. 4C). The SVD analysis revealed that only two species (oligomer and quasi-aggregate) exist during the formation of the quasi-aggregate, and that the formation of the quasi-aggregate follows the first-order kinetics (Fig. 4, A and B), which was consistent with the presence of an isoscattering point in the Kratky profiles (Fig. 4A). These results indicated that any stable intermediates are not formed in the formation process of quasi-aggregate from oligomers. Because the formation rate of quasi-aggregate was unambiguously affected by temperature (Fig. 2), we suggest that the formation of quasi-aggregate would be induced by hydrophobic interactions (28). The partial solubility of quasiaggregate in SDS (data not shown) would be in agreement with the major contribution of the hydrophobic interactions to the formation of quasi-aggregate, because the interactions between expanded polyglutamines are so strong that a quasi-aggregate formed by such interactions could not be readily dissociated into monomers by the treatment with SDS (29). We reported previously that the Mb-Gln 50 mutant, particularly its protein surface, is partially unfolded by the insertion of 50 glutamine repeats (17). The partially unfolding is one of the common structural features for amyloid-forming proteins (7). Hydrophobic residues exposed on the protein surface by the partial unfolding would promote the assembly of the Mb-Gln 50 protein and induce the formation of quasi-aggregate (Fig. 6). We found that the activation energy (22.8 Ϯ 1.8 kcal/mol) for the quasiaggregate formation was comparable with the value for the nucleation process of ␣-synuclein (17.9 kcal/mol) (28).
Protofibrils of amyloid-␤ peptide and ␣-synuclein are widely known to associate in tandem with each other and to form a mature fibril upon further incubation (3,4,6,7). In the present study, the removal of the quasi-aggregate in a Mb-Gln 50 sample prolonged the lag time of fibrillization, whereas the addition of the quasi-aggregate into a fresh Mb-Gln 50 solution eliminated the lag phase (Fig. 1A). These results suggested that the quasi-aggregate formed in the early stage of fibrillization would be involved in and facilitate the fibril formation, as previous seeding studies indicated (30,31). Then, we further examined the relation of the quasi-aggregate to a mature fibril in Mb-Gln 50 at various incubation times from 1 to 7 days, when the fibril formation was almost completed (Fig. 1A), by electron microscopy. Although tandem assembly of the quasi-aggregate was not clearly detected, we frequently observed the fibrils that  were connected to and associated with clusters of quasi-aggregates at 3 or 5 days (arrow, Fig. 3, G-I), which might support some involvement of the quasi-aggregate in fibril formation. It is noted here that the dimension of the quasi-aggregate (20 -40 nm) was somewhat larger than the diameter of fibril (10 -20 nm), implying that a fibril is not formed by the simple tandem association of the quasi-aggregate. If fibrils were formed by the assembly of quasi-aggregates, conformational rearrangement in the quasi-aggregate would be essential, as discussed in fibrillization of Ure2p (32), or some protein molecules would dissociate from the quasi-aggregates during the fibril maturation (27). Alternatively, the quasi-aggregate might play a role on the scaffold where monomers or oligomers interact and finally form a mature fibril, because we could not observe the tandem assembly of quasi-aggregates during the fibril formation by electron microscopy. Contrary to the formation of quasi-aggregate for Mb-Gln 50 , we could not observe such a largely assembled species of protein during the fibrillization of Mb-Gln 35 by SAXS and electron microscopy. A major difference in the physical property between Mb-Gln 35 and Mb-Gln 50 is that Mb-Gln 35 is not unfolded as greatly as Mb-Gln 50 (17). The lesser extent of unfolding of Mb-Gln 35 would prevent the exposure of hydrophobic residues on protein surface and the formation of quasi-aggregate in the early stage of fibrillization. Because the addition of the Mb-Gln 50 quasi-aggregate into a fresh Mb-Gln 35 solution accelerated the fibril formation of Mb-Gln 35 (Fig. 1A), we suggest that the absence of quasi-aggregate in the early phase is responsible for the slower formation rate of the Mb-Gln 35 fibril. Because Mb-Gln 35 did not form a quasi-aggregate, the mutant Mb would form a mature fibril in a different way from the quasi-aggregate-based fibrillization suggested for Mb-Gln 50 .
Structural Features of Quasi-aggregate Formed in the Early Stage of Fibrillization-We examined structural features of the ␤-sheet formed by the inserted polyglutamine in the quasiaggregate of Mb-Gln 50 . Although the Mb-Gln 50 fibril, which was prepared by the incubation for 120 h at 37°C, clearly showed the peak at S ϭ 0.21 Å Ϫ1 in the Kratky plot, corresponding to the 4.8-Å spacing of orderly aligned ␤-pleated sheets, such a peak was not observed in the Mb-Gln 50 sample in 3 to ϳ12 h incubation when formation of the quasi-aggregates is completed (Fig. 5D). This result suggested the absence of orderly aligned ␤-pleated sheets in the quasi-aggregate. We also addressed the structural property of ␤-sheet in the quasiaggregate by the Congo red binding assay. The formation of quasi-aggregate was completed in 3 h, and the amount of quasi-aggregate was constant until 27 h. The level of bound Congo red to Mb-Gln 50 in 3-27 h was quite lower than that in ϳ120 h (Fig. 1A), when fibrils were clearly observed by electron microscopy. This result suggested that the quasi-aggregate has a low affinity for Congo red, contrary to the high affinity of fibril for Congo red. Because Congo red binds to regularly oriented ␤-pleated sheets in fibril (33), we interpreted that the low affinity of the quasi-aggregate for Congo red might correspond to unordered ␤-sheets in the quasi-aggregate. This interpretation seemed to be reconciled with the observation that the 1C2 antibody was not reactive to the fibril but to the quasi-aggregate of Mb-Gln 50 in the immunoelectron microscopic and immunoblotting experiments (Fig. 3, B-F). The Mb-Gln 50 fibril was not labeled by the 1C2 antibody along with the fibers, and the reaction of the 1C2 antibody with the gel top in the immunoblot was decreased in 72 h. This decreased reactivity was correlated with the increase in the amount of fibril at the gel top, because fibrils were not observed at 1 day but clearly detected at 3 days by electron microscopy (Fig. 3G). The low reactivity of the 1C2 antibody to fibril suggested that ␤-sheets of expanded polyglutamines in fibril are regularly oriented, interact each other extensively, and form a core structure in the fibril such that the 1C2 antibody is not accessible to the buried, expanded polyglutamines. On the other hand, the 1C2 antibody was reactive to the quasi-aggregate as evidenced by the electron micrograph (Fig. 3C) and the immunoblot (3, 10 and 24 h in Fig. 3F). These results implied that ␤-sheets of expanded polyglutamines in the quasi-aggregate do not yet form regularly aligned ␤-pleated sheets and do not interact each other so extensively as to inhibit the accessibility of the 1C2 antibody. On the basis of these results, we propose that the ␤-sheets of expanded polyglutamines in the quasi-aggregate of Mb-Gln 50 would not be orderly aligned and are partially exposed, which are quite different from the regularly oriented and buried ␤-pleated sheets in fibril.
Previous studies reported that some fibril precursors such as protofibril or an oligomerized form of amyloid-␤ exhibited cellular toxicity in vitro and in vivo (8 -11). In our study, the presence of the Mb-Gln 50 quasi-aggregates in the pellet fraction unfortunately prevented us from isolating the quasi-aggregate as a soluble state and evaluating its cellular toxicity. This observation was in contrast to isolation of the amyloid-␤ protofibril as a soluble form by gel filtration chromatography (5). Because the protofibril of amyloid-␤ constitutes 30 to ϳ50 monomers (4,500) (34), its molecular weight is estimated to be 135,000 to ϳ225,000, which is much smaller than that of the Mb-Gln 50 quasi-aggregate (2,000,000 to ϳ2,500,000). The higher molecular weight together with the extreme insolubility of expanded polyglutamines would render the Mb-Gln 50 quasiaggregate insoluble in water. Although we could not investigate cellular toxicity of the quasi-aggregate, the formation of quasi-aggregate only for the expanded polyglutamine-bearing protein suggested that the quasi-aggregate might be implicated with the cellular dysfunction that is caused by the expansion of glutamine repeats.
In summary, we revealed that the expansion of polyglutamine to 50 repeats in protein induced the formation of a largely assembled non-fibrillar species of the protein, a quasiaggregate, in the early stage of protein fibrillization. A bead modeling procedure based on SAXS spectra showed that the quasi-aggregate is a bowl-like structure with long and short axes of 400 and 190 Å, respectively. We suggested, by spectroscopic and electron microscopic studies, that properties of the ␤-sheets formed by expanded polyglutamines in the quasiaggregate are quite different from those of regularly oriented and buried ␤-pleated sheets in fibril. FIG. 6. Proposed formation mechanism of quasi-aggregate for the protein bearing an expanded polyglutamine. Expansion of glutamine repeats in protein induces partial unfolding of the protein, which promotes formation of the quasi-aggregate by hydrophobic interactions. The quasi-aggregate is suggested to facilitate the formation of fibril.