Deficiency of Disulfide Bonds Facilitating Fibrillogenesis of Endostatin*

Endostatin is an endogenous inhibitor of tumor angiogenesis and tumor growth. It has two pairs of disulfide bonds in a unique nested pattern, which play a key role in its native conformation, stability, and activity. Here, we constructed a disulfide-deficient variant of endostatin, endo-all-Ala, to examine the effects of the two disulfide bonds on fibrillogenesis of endostatin under nondenaturing conditions. Based on thioflavin T fluorescence, atomic force microscopy, far-UV circular dichroism, and Fourier transform infrared spectroscopy, we found that endo-all-Ala, which has a higher α-helical content compared with wild type, is prone to forming fibrils in a pH-dependent manner. Subsequently, more hydrophobic patches with a lower stability of endo-all-Ala were observed when compared with wild type, which possibly contributes to the propensity of amyloid formation of endo-all-Ala. To our surprise, the significant increase of the α-helical content in endostatin induced by trifluoroethanol can also facilitate fibril formation. In addition, the cytotoxicity of fibrillar aggregates of endo-all-Ala, which were generated at different stages of the fibril formation process, was evaluated by cell viability assay. The results indicate that the cytotoxicity is not due to the fibrils but rather due to the granular aggregates of endo-all-Ala. Moreover, endostatin was interestingly found to be reduced by glutathione at physiological concentrations. Our present work not only elucidates the correlation between the existence of disulfide bonds and the fibril formation of endostatin but also may provide some insights into the structural and functional basis of endostatin in Alzheimer disease brains.

Formation of amyloid-like fibrils is frequently considered to be a generic property of many proteins (1, 2), including but not limited to disease-related proteins (3,4). Although having no obvious amino acid sequence similarity, all amyloid fibrils appear to share a common characteristic cross-␤-sheet structure that forms the core of the fibrils (5). These amyloid fibrils, straight and unbranched, viewed under transmission electron microscope (5), can be detected through the binding of dyes such as thioflavin T (ThT) 2 (6) or Congo Red (7). Furthermore, amyloid formation arises primarily from the main chain interaction (8), and disulfide bonds in proteins usually play an essential role in amyloid fibril formation (9,10).
Endostatin is a 20-kDa C-terminal fragment of collagen XVIII (11). It can specifically inhibit vascular endothelial cell proliferation and migration and thus potently prevent angiogenesis and tumor growth without induced toxicity or acquired drug resistance (11)(12)(13). Endostatin is a globular protein with two pairs of disulfide bonds (Cys 33 -Cys 173 and Cys 135 -Cys 165 ) in a unique nested pattern (14). These disulfide bonds, making the whole molecule tightly packed, are intimately related to the native conformation, stability, and activity of endostatin (15). It has been reported that endostatin is acid-resistant with slow kinetics upon acidinduced unfolding, which may be attributed to the two pairs of disulfide bridges (16). In addition, endostatin is highly resistant to trypsin digestion at 37°C when the two pairs of disulfide bonds are removed (15), which may be considered to be one of the characteristics of fibrillar proteins like PrP sc (17). However, the relationship between fibrillogenesis of endostatin and the two pairs of disulfide bonds under nondenaturing conditions has not been investigated so far.
To investigate the effects of the disulfide bonds on fibrillogenesis of endostatin, an endostatin variant (endo-all-Ala), which is a genetically engineered disulfide-deficient variant of endostatin wild type (endo-WT), was constructed. In this study, substitution of all four cysteine residues with alanine, which is a conservative change (18), was shown to lead to an enhanced pH-dependent fibrillogenic tendency when compared with endo-WT in aqueous buffer with no denaturants. Unlike disulfide-reduced endo-WT (19), however, endo-all-Ala is free of thiol residues and thus cannot cause intermolecular covalent cross-linking and aggregation via disulfide bonds. Therefore, endo-all-Ala is a good model for the study of fibril formation without the complications of added denaturants or reductants. Furthermore, the fibrillogenesis of endo-all-Ala can be investigated without consideration of the intermolecular covalent cross-linking caused by thiol residues.
In our work, we found that endo-all-Ala rapidly forms amyloid-like fibrils at a mildly acidic pH. This mildly acidic pH always occurs in aged brains with cerebral acidosis (20). In contrast to endo-WT, endo-all-Ala at this acidic pH was shown to have more hydrophobic patches with a lower stability, which possibly contributes to the amyloid formation propensity of endo-all-Ala. Surprisingly, endostatin with a higher ␣-helical content induced by either mutagenesis or 2,2,2-trifluoroethanol (TFE) is especially prone to generating fibrils. Moreover, during the fibrillogenic process at various time periods, endo-all-Ala produces distinct morphological types of aggregates. Among these aggregates, the granular aggregates can cause cytotoxicity toward rat pheochromocytoma PC12 cells in a dose-dependent manner.
Recently, Meyermann and co-workers (21) manifested the production of endostatin by neuronal cells and the localization of endostatin to A␤ plaques in Alzheimer disease brains. In the cytosolic environment of the nervous system, proteins may be exposed to strong reducing influences, such as GSH (22,23), antioxidants (24), and thiol-disulfide oxi-doreductases (25). Furthermore, superoxide produced under pathological conditions is a potent reducing reagent, which can reduce disulfides in proteins (26). Our current study reveals that endostatin can be reduced by GSH in a simulated physiological environment in vitro. The structure of disulfide-reduced endostatin is similar to that of endo-all-Ala, which was verified by tryptophan emission fluorescence and far-UV CD spectra. Thus, the behavior of endo-all-Ala may be relevant to that of disulfide-reduced endostatin in vivo. These findings, combined with the in vitro cytotoxicity study, not only elucidate the effects of the unique nested disulfide bonds on fibrillogenesis of endostatin but also may provide some insights into the structural and functional mechanisms of endostatin in Alzheimer disease brains. Besides, protein aggregation can result in major economic and technical problems in biotechnology and pharmaceutical industries (27). Our study may consequently illuminate some basic mechanisms for inhibiting the aggregation of endostatin during its extensive preparation, storage, and delivery for therapy.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Endo-WT and endo-all-Ala were expressed and purified essentially as described elsewhere (15). Briefly, proteins were expressed in Escherichia coli strain BL21 (DE3) with MGGSHHHHH at their N termini and purified via affinity chromatograph on Ni 2ϩ -nitrilotriacetic acid column (Qiagen). Endo-WT and endo-all-Ala were refolded and purified by Protgen. The activity of refolded endostatin was analyzed and verified on the basis of the endothelial cell proliferation and migration assays as described (28). Protein concentrations were determined according to Edelhoch's method (29).
Fibril Formation-Endo-WT and endo-all-Ala were first dissolved in 30 mM acetate (pH 5.5) to make a stock solution, which was then diluted by aliquots into 10 mM sodium acetate buffer at different pH values. The stock proteins were filtered through a 0.2-m cut-off filter to remove large particles that may act as seeds for fibril growth prior to the addition of buffer. The final protein concentration was 37.5 M. ThT fluorescence was carried out after the samples were incubated at 37°C for 81 h.
ThT Binding Assay-Fluorescence was measured with a Hitachi FL-4500 fluorescence spectrophotometer using cuvettes that have an optical path length of 1 cm. The temperature was maintained constant at 20°C using an external bath circulator. Prepared fibril samples were diluted to 30 M ThT in 10 mM sodium acetate buffer, pH 6.0, with a final concentration of 5 M, and then the emission fluorescence intensity at 485 nm was monitored with the excitation at 442 nm. The slit widths were 5 and 10 nm for excitation and emission, respectively. For each sample, the fluorescence intensity at 485 nm was corrected by subtracting the emission intensity recorded before the protein was added to the ThT solution. In fibril growth kinetics experiments, at chosen time points, 70-l aliquots of protein were removed from the fibril solution prepared above for the ThT binding assay. The ThT stock was filtered three times before use.
Atomic Force Microscopy (AFM)-Fibril formation was monitored by AFM to identify and characterize intermediate species appearing on the aggregation pathway. At the time points of 0, 11, 41, and 81 h, a diluted aliquot (10 l; 1:70 dilution in water for endo-WT and endo-all-Ala) of protein was sampled from the protein solution and deposited on a piece of freshly cleaved mica for 5 min. The mica was then rinsed with deionized water and blow-dried with compressed nitrogen. The dried substrate was transferred to AFM (Nanoscope IIIa Multimode AFM equipped with an E-scanner; Digital Instruments). Tapping mode images were obtained. The scan size was 2 m. The image of endo-WT in 30% TFE was acquired with the same procedure. At least three regions of the mica surface were examined to confirm that similar structures existed through the sample. AFM image analysis was measured using the Nanoscope software (version 5.12r3).
CD Measurements-Far-UV CD was recorded on a Jasco J-715 spectropolarimeter equipped with a temperature-controlled liquid system. The stock of proteins was diluted in 2 mM sodium acetate by aliquots to a final protein concentration of 2 M. Cuvettes of 1-cm path length were used over the wavelength range between 190 and 250 nm. An average of four scans was obtained for all of the spectra. Photomultiplier absorbance did not exceed 600 V in the spectral region analyzed. Data were corrected for the buffer contributions. All of the measurements were performed under nitrogen flow. The results are expressed as mean residue ellipticity [] in units of degrees ϫ cm 2 ϫ dmol Ϫ1 ϫ 10 3 . When the ␣-helical propensity of endo-WT was examined, endo-WT was dissolved into the sodium acetate buffer at pH 6.0 containing different concentrations of TFE (0, 5, 10, 15, 20, and 30%), and then CD measurements were performed after the pH was checked.
Fourier Transform Infrared (FTIR) Spectroscopy-Labile hydrogens in endostatin were replaced by deuterium through repeated cycles of lyophilization and dissolution in D 2 O. The spectra of endo-WT and endo-all-Ala were recorded using freshly dissolved solutions of proteins (20 mg/ml). Amyloid fibrils were formed by incubation of endo-WT and endo-all-Ala in D 2 O, pD 6.0 (measurements not corrected for isotope effects) for 81 h at 37°C. The fibrils were concentrated by centrifugation before analysis. The FTIR spectra were collected on a PerkinElmer Life Sciences FTIR spectrophotometer equipped with a variable path length cell and purged with a continuous flow of dry nitrogen. Sample aliquots were placed between CaF 2 windows separated by a 0.2-mm Teflon spacer. The cell path length was kept constant during all experiments. For each spectrum, a 256-scan interferogram was collected in single beam mode at 20°C using a 4 cm Ϫ1 resolution. Fitting of the amide IЈ band was performed using Gaussian curves.
1-Anilinonaphthalene 8-Sulfonate (ANS) Binding Assay-ANS binding assay was carried out with an FL-4500 fluorescence spectrophotometer (Hitachi). ANS was added with a final concentration of 10 M in 10 mM sodium acetate at different pH values. Endo-WT and endo-all-Ala were added with a final concentration of 1 M. The ANS emission was scanned between 400 and 600 nm with an excitation wavelength of 350 nm. The titration curves of ANS binding were monitored with the emission wavelength at 478 nm. A circulating water bath was connected to the fluorescence spectrophotometer with the temperature adjusted to 37°C.
Light Scattering-Protein samples were prepared as described under "Fibril Formation." To monitor the protein polymerization, light scattering was performed immediately once the pH of the protein solutions was adjusted. Light scattering was measured in an FL-4500 fluorescence spectrophotometer (Hitachi). Both the excitation and emission wavelengths were set to 400 nm with a spectral bandwidth of 1 nm. All data were collected at 37°C.
Measurements of Tryptophan Emission Fluorescence-Fluorescence measurements were taken using a Hitachi FL-4500 fluorescence spectrophotometer with a 1-cm light path at 37°C. The excitation wavelength at 288 nm was chosen for the main contribution of tryptophan residues. The emission fluorescence spectra were collected between 300 and 400 nm with slit widths of 5 and 10 nm for excitation and emission, respectively. The scan speed was 240 nm/min. The protein concentration was 1 M in 2 mM sodium acetate (pH 6.0) or 2 mM Tris-HCl (pH 7.4). To avoid intermolecular cross-linking, the fluorescence of reduced endostatin was immediately tested after the pH of the solution was adjusted.
Urea-induced Unfolding and Data Analysis-Stock solutions of endo-WT and endo-all-Ala were diluted into urea of different concentrations with a final protein concentration of 1 M. The solution buffer is 2 mM sodium acetate (pH 6.0). The chosen temperature was 37°C for mimicking the physiological condition. After the solutions were incubated at this temperature for enough time to reach the kinetic equilibrium, fluorescence measurements were performed. The excitation wavelength was 288 nm, and the emission was monitored at 350 nm.
Urea-induced unfolding curves were analyzed by a two-state equation using the procedure of Santoro and Bolen, in which the native and unfolded base lines were determined by the data inside as well as outside the transition zone (30). Endo-all-Ala cannot be normalized to a twostate curve because of the marginal stability and the difficulties in finding its base lines.
Cell Culture-PC12 cells (rat pheochromocytoma) were routinely cultured in Dulbecco's modified Eagle's medium (Hyclone) containing 10% horse serum and 5% fetal bovine serum in a 5% CO 2 humidified atmosphere at 37°C. 100 g/ml streptomycin and 100 units/ml penicillin were added to the medium.

3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium Bromide (MTT)
Assay-Aliquots of solutions containing different morphological types of aggregates of endo-all-Ala prepared as described under "Atomic Force Microscopy (AFM)," with 10 M final concentration or granular aggregates of endo-all-Ala with 1-10 M final concentrations were added to the cell medium. After a 24-h incubation, 10 l of a stock MTT solution in phosphate-buffered saline was added to give a final concentration of 0.5 mg/ml and incubated for another 4 h. After removing the medium with phenol red, 100 l of cell lysis buffer (Me 2 SO) was added to each well, and the samples were mixed thoroughly by repeated pipetting with a multichannel pipettor. The cells were incubated at 37°C for 30 min, and the absorbance at 570 nm was measured on an enzyme-linked immunosorbent assay plate reader (Bio-Rad) with a reference wavelength of 630 nm. Each sample was assayed in hexad, and the assays were repeated twice.
Statistical Analysis-The results are expressed as mean values Ϯ S.E. Multiple comparisons were performed by one-way analysis of variance, and differences with p Ͻ 0.05 were considered significant.
Protein Reduction Assay-Endo-WT was incubated with GSH of 5-20 mM at 37°C, pH 7.4. The concentration of protein was 0.5 mg/ml. The reactions were terminated at different time intervals by adding Laemmli SDS-PAGE nonreducing loading buffer. The samples were not boiled and then were analyzed on 12% SDS gels followed by Coomassie Brilliant Blue staining.

RESULTS
pH Dependence of Fibril Formation of Endo-all-Ala-It has been well established that fibril formation requires appropriate physicochemical conditions, among which pH exerts a significant influence on the whole process of fibrillogenesis (31). Hereby, the pH dependence of fibril formation of endo-WT and endo-all-Ala was evaluated initially. Samples were incubated at 37°C for 81 h at pH ranging from 3.6 to 6.6, and fibrils were quantified by ThT fluorescence. The enhancement in the fluorescence intensity of ThT upon binding to ordered protein aggregates is a rapid and sensitive method to show the presence of fibrils (6). ThT fluorescence as a function of pH for endo-WT and endo-all-Ala was shown in Fig. 1A. As to endo-WT, no increased fluorescence was observed regardless of various pH values. The fluorescence for endo-all-Ala, however, increased rapidly from pH 4.3 to 6.0 and reached a maximum at pH 6.0. Above pH 6.0, the ThT fluorescence dropped markedly, and this process was accompanied by the spontaneous precipitation of endo-all-Ala (data not shown). These results therefore reveal that fibrils from endo-all-Ala form most readily under mildly acidic conditions from pH 4.3 to 6.0, whereas fibrils from endo-WT are not present, and only within this range of pH does endo-all-Ala exist in a well defined aggregated conformation after incubation for 81 h.
The ThT fluorescence of endo-all-Ala increased nearly 200-fold from pH 4.0 to 6.0 in Fig. 1A. Then we determined the time course of the fibril formation of endo-all-Ala at pH 4.0 and 6.0 and 37°C using ThT fluorescence (Fig. 1B). Amyloid fibril formation is a nucleation-dependent process, in which protein molecules gradually cluster together to form a nucleus, and the kinetics of fibrillogenesis can be described as a characteristic sigmoidal time course curve (32). Unexpectedly, the fibril kinetic curve of endo-all-Ala at pH 6.0 was not sigmoidal and exhibited no observable lag phase, which suggests that endo-all-Ala forms fibrils rapidly without nucleation phases. After 81 h of incubation, the ThT fluorescence of endo-all-Ala at pH 6.0 increased to a maximal platform, indicating the end of fibril formation, whereas it kept unchanged at pH 4.0, which coincided with the results shown in Fig. 1A.
Fibril Formation of Endo-all-Ala at pH 6.0-AFM images of time-dependent morphological types of endo-all-Ala aggregates generated by incubation at pH 6.0 and 37°C in vitro were shown in Fig. 2. Consistent with our previous fibril kinetics data in Fig. 1B, amorphous granular aggregates composed of spherical particles predominated immediately

Fibrillogenesis of Endostatin
after the pH was adjusted ( Fig. 2A). Eleven hours later, short fibrils emanated from the globular granules, which indicated that fibrillar protein aggregates were beginning to form (Fig. 2B). In a similar sample recovered at 41 h, the fibrillar material was clearly evident, and the fibril extension continued (Fig. 2C). This fibrillar type resembles the protofilaments observed in other fibrillogenic systems, according to the size and characteristics of the aggregates (33). After prolonged incubation for 81 h, the AFM image showed that the fibrils produced were long, smooth, and unbranched and possessed a rope-like structure with a width of 30 -40 nm (Fig. 2D). In some instances, the image may represent overlapping fibrils. Based on the different morphological types of the aggregates, we classified them as granule, short fibril, protofilament, and fibril, respectively. The dimensions, including heights, widths, and lateral sizes, of the four types of the aggregates on mica were quantified by section analysis, and the results are shown in Table 1. Along with the increased height and width, the aggregates of endo-all-Ala continued to elongate as the growth of fibrils progressed. In contrast, when endo-WT was incubated under the same conditions for 81 h, there were only small spherical particles, which were probably the protein monomer molecules in appearance, with no evidence for extended fibrils characteristic of fibril formation (Fig. 2E).
To further probe the properties of fibrils, we explored the secondary structural changes of endo-all-Ala due to the fibril formation at pH 6.0 by far-UV CD (Fig. 3). The difference of the spectra between endo-all-Ala before and after being incubated for 81 h showed a negative peak at 215 nm, which is a diagnostic mark for the presence of ␤-structure (34). Cross-␤-structure has been represented as the fundamental polypep-tide arrangement of fibrils (5), whereas the secondary structure of endo-WT did not alter after the incubation.
Finally, we used FTIR to test the presence of fibrils. FTIR spectroscopy can be used in the study of protein secondary structure in the dissolved, aggregated, and solid state (35). It is therefore a powerful tool for the study of conformational changes occurring during endostatin aggregation. As expected, the FTIR spectrum of endo-WT after the incubation did not make any change compared with that of the protein incubated before (Fig. 4A). As shown in Fig. 4B, the FTIR spectrum in the amide IЈ region of endo-all-Ala before incubation was dominated by a strong band at 1640 cm Ϫ1 , indicating a high proportion of random coil structure. Moreover, endo-all-Ala had a little amount of intermolecular ␤-structure based on the bands of 1618 and 1681 cm Ϫ1 , which is accordant with the previous results in Figs. 1B and 2A. When the pD of the solution was adjusted to 6.0, endo-all-Ala could form granular aggregates immediately. After the incubation of endo-all-Ala at 37°C for 81 h, the decomposition analysis of the spectrum in Fig. 4C showed that the relative intensity of the 1640 cm Ϫ1 band decreased, and those of the bands appearing at 1615 and 1683 cm Ϫ1 increased, suggesting a strong  were freshly diluted to 2 mM NaAc (pH 6.0). Fibrils formed from endo-WT (filled circles) and endo-all-Ala (filled triangles) were prepared in 10 mM NaAc (pH 6.0) and then were diluted to 2 mM NaAc. Protein concentration was 2 M. Data were collected at 20°C. [] represents mean residue ellipticity with the unit of degrees ϫ cm 2 ϫ dmol Ϫ1 ϫ 10 3 .

Fibrillogenesis of Endostatin
increase of intermolecular hydrogen-bonded ␤-sheet structure. The secondary structure of cross-␤-sheet is considered as one of the hallmarks of fibrils (5). The secondary structure assignments in amide IЈ of endo-WT and endo-all-Ala were shown in Table 2, and the structure revealed by FTIR was consistent with that from the far-UV CD. Taken together, all of these data provide evidence to confirm that deficiency of both disulfide bonds endows endostatin with a high tendency to rapidly form fibrils under nearly neutral conditions in which endo-WT is in the native state.
Hydrophobic Surface and Aggregation Propensity of Endo-all-Ala-Endo-all-Ala forms fibrils in a pH-dependent manner. The self-assembly of proteins to fibrils is guided by noncovalent interactions such as hydrophobic interaction between hydrophobic residues exposed on the surface of individual protein molecules (33). This prompted us to detect the exposure of the hydrophobic patches on endo-WT and endo-all-Ala with pH titration by the fluorescent compound ANS, which specifically binds to hydrophobic clusters of aminoacyl residues (36). Upon binding, the quantum yield of the dye became increased, and its fluorescence maximum shifted from 520 nm to shorter wavelengths (Fig. 5, A and B). The titration curves of endo-WT and endo-all-Ala by ANS are shown in Fig. 5, C and D. As illustrated in Fig. 5B, the addition of 10 M of ANS to 1 M endo-all-Ala resulted in a large increase in fluorescence intensity and a shift of maximum of emission wavelength to 470 nm from pH 2.3 to pH 6.0. ANS also bound to the hydrophobic patches contained by endo-WT when the pH was lowered from 5.7 to 1.9 (Fig. 5A). At pH 1.9, the ANS fluorescence intensity of endo-WT at 478 nm was greatly enhanced, and the maximal emission wavelength was blue-shifted because of the formation of an intermediate induced by acid, which agrees with our previous report (37). These results indicate that there are numerous hydrophobic patches on the surface of endo-all-Ala at pH 6.0 that can bind to ANS and change the fluorescence emission spectra. Endo-WT at pH 6.0, however, does not bind to ANS, which implies that it has far fewer exposed hydrophobic residues than endo-all-Ala.
Since endo-all-Ala has more hydrophobic patches on its surface than endo-WT, we conducted a light-scattering assay to quantify the polymerization of endo-WT and endo-all-Ala at different pH values. The light-scattering assay was performed immediately after the pH of the protein solution was adjusted. As shown in Fig. 6, unlike endo-WT, which did not self-assemble at any pH with the concentration of 37.5 M, endo-all-Ala started to polymerize beyond pH 6.0 and reached a maximum around pH 10. When the ThT fluorescence and light scattering were superimposed as a function of pH, it is of interest that the corresponding pH of the maximal ThT fluorescence of endo-all-Ala after 81 h of incubation was near the turning point of polymerization, which arose around pH 6.0. This result shows that fibril formation of endo-all-Ala is most prominent at the pH where polymerization begins to occur.
Conformational Stability of Endo-all-Ala-It has been reported by Dobson and co-workers (38) that the tendency of fibrillogenesis correlates inversely with the conformational stability of the native state of the protein. We thus tested the conformational stability of endo-WT and endo-all-Ala at pH 6.0 via urea-induced unfolding monitored by tryptophan emission fluorescence. The tryptophan fluorescence emission spectra showed that the maximal emission wavelength of endo-WT was at 318 nm when excited at 288 nm, and that of endo-all-Ala was at about 340 nm at pH 6.0 without urea (Fig. 7A). These results demonstrate that endo-WT has a very tight tertiary structure, whereas endo-all-Ala is close to a completely denatured state at pH 6.0 without denaturant. However, the structure of endo-all-Ala at pH 6.0 is not similar to that of

TABLE 2 IR band position and secondary structure assignments for the amide l' band of endo-WT and endo-all-Ala
Endo-WT and endo-all-Ala before incubation were lyophilized and then freshly dissolved in D 2 O through repeated cycles. The final concentration is 20 mg/ml. Amyloid fibrils formed by incubation of endo-all-Ala in D 2 O, pD 6.0, for 81 h at 37°C, were concentrated by centrifugation before analysis.

Fibrillogenesis of Endostatin
endo-WT at pH 2.0, where endo-WT also exists in a partially folded state, as measured by tryptophan emission fluorescence and far-UV CD (data not shown). With the increased concentration of urea, endo-WT and endo-all-Ala unfolded accompanied by red-shifting of the maximal emission wavelength (Fig. 7A). The unfolding process of endo-WT revealed apparently a two-state process (Fig. 7A). Based on Santoro and Bolen's two-state model (30), the normalized curve of endo-WT was shown in Fig. 7B. Endo-WT has a urea molarity at the midpoint of the unfolding transition curve (C m ) of 3.9 M and was fully unfolded at 5.5 M urea. Endo-all-Ala, of which the unfolding was not a two-state process, unfolded completely at 3.5 M urea (Fig. 7B). These data imply that endo-WT is much more stable than endo-all-Ala at pH 6.0, 37°C, which may be caused mainly by the nested disulfide bonds in endostatin.

Increase of ␣-Helical Content Facilitating Fibril Formation of Endostatin-
Earlier studies in our group demonstrated that the two disulfide bonds restrict the helical propensity of endostatin (15). When  they are absent, endo-all-Ala exhibits a higher ␣-helical content based on the negative peak at 222 nm of far-UV CD spectra (15). This view is also confirmed by FTIR in Fig. 4B. Surprisingly, we found that endo-all-Ala with more ␣-helical content was prone to yielding fibrils, which contrasts with the proposition that the presence of ␣-helix represents one way to suppress fibril formation (8). In order to further assess whether or not endostatin with a high ␣-helical proportion prefers fibril formation, we examined the fibrillogenesis of endo-WT in the presence of TFE of different concentrations. TFE has a "double edged" effect on proteins and peptides; on one hand, it can destroy the tertiary structure of proteins (39), and on the other hand, it can increase the ␣-helical structure of proteins and peptides via strengthening of the hydrogen bonds (40). Far-UV CD spectra acquired for endo-WT at six representative TFE concentrations were shown in Fig. 8A. Indeed, the ␣-helical structure of endo-WT increased with TFE concentrations from 0 to 30% at pH 6.0, according to the two negative peaks at 208 and 222 nm in the spectra. We are especially interested in finding that within this range of TFE concentration, endo-WT produced the characteristic increase of ThT fluorescence, which implies that the fibril formation occurs (Fig.  8B). Moreover, the AFM image of endo-WT incubated in 30% TFE displayed long, twisted, and rope-like fibrils in Fig. 8C as well as those of endo-all-Ala incubated for 81 h at pH 6.0 in Fig. 2D, which further confirms that endo-WT produced fibrils in 30% TFE. All of these data therefore lead to the conclusion that the increased content of ␣-helical structure facilitates fibril generation of endostatin.
Cytotoxicity of Endo-all-Ala Aggregates-Incubation of endo-all-Ala in 10 mM sodium acetate, at pH 6.0, 37°C, led to the formation of granule, short fibril, protofilament, or fibril at different time intervals, respectively (see Fig. 2). The cytotoxicity of the four various morphological types of aggregates formed in such experiments was detected by adding the aggregates of a final protein concentration of 10 M to PC12 cell culture medium. Prior to the test of cytotoxicity, the morphologies of the distinct aggregates were detected unchanged following the dilu-tion into the cell culture medium as confirmed by AFM (data not shown). The cytotoxicity of the aggregates was evaluated by MTT reduction inhibition assay, a standard indicator of cell physiological stress thought to be related to changes in intracellular trafficking, particularly in the pathway of exocytosis (41). The experiments revealed that early endo-all-Ala granular aggregates significantly decreased the levels of reduced MTT in PC12 cells, whereas highly structured fibrils formed by prolonged incubation at pH 6.0 had small effect (Fig. 9A). In a second series of experiments, the toxicity of granular aggregates at a range of final protein concentrations from 1 to 10 M formed by endoall-Ala was examined toward PC12 cells. The result showed that the granular aggregates resulted in an inhibition of MTT reduction ranging from 13% at the lowest protein concentration (1 M) to 28% at the highest protein concentration (10 M) tested with respect to the control experiments performed by incubating the same cells with the buffer solutions only (Fig. 9B). In sum, these experiments demonstrate that the cytotoxicity of endo-all-Ala aggregates depends on not only the morphologies of the aggregates but also the concentrations of the aggregates.
Endostatin Reduced by GSH-Meyermann and co-workers (21) recently revealed that endostatin can be produced by neuronal cells and co-accumulate with A␤ plaques in Alzheimer disease brains. The cytosolic environment of the nervous system, however, may expose proteins to strong reducing influences. For example, GSH is an abundant intracellular thiol-containing molecule (present at ϳ1-10 mM in neurons and glia) that can reduce disulfide bonds in proteins (42). To determine whether endostatin is sensitive to this reagent, we incubated endostatin with different concentrations of GSH at 37°C, pH 7.4, and then analyzed the samples by nonreducing SDS-PAGE. The band of slower migration suggests the disulfide-reduced protein. Fig. 10A demonstrated that endostatin can be partially reduced by 10 -20 mM GSH for 1 h of incubation. After the reaction for 8 h, the amount of corresponding reduced endostatin increased slightly. Surprisingly, the band for reduced endostatin disappeared when we detected the samples after 24 h, which may result from the oligomers formed by the active thiols in the reduced endostatin at pH 7.4. Moreover, endostatin can be completely reduced, and it simultaneously formed oligmers at 50 h of incubation (data not shown). In addition, the tertiary and secondary structures of disulfide-reduced endostatin were not altered compared with those of endo-all-Ala according to the tryptophan emission fluorescence and far-UV CD spectra (Fig. 10, B and C).

Roles of Disulfide Bonds in Fibril Formation of Endostatin-Disulfide
bonds usually play a key role in fibrillogenesis (9) as well as in the stability and activity of proteins (15,43). Endostatin is a specific inhibitor of endothelial cell migration, proliferation, and angiogenesis (11). It contains a large fraction of irregular loop structures and ␤-sheets as well as two pairs of disulfide bonds (14). Based on the crystal structure, the disulfide bonds are in a nested pattern, which links the central core and the peripheral structures (14). In the present study, we found that deficiency of the two pairs of disulfide bonds can facilitate the fibril formation of endostatin in a pH-dependent manner. This finding implies that the disulfide bonds are of significant importance in stabilizing the conformation and preventing the fibril formation of endostatin.
Amyloid fibril formation is a nucleation-dependent process in which protein molecules gradually cluster together to form a nucleus, and then to fibrils (44). To our interest, endo-all-Ala generates fibrils rapidly at a nearly neutral pH without an observable nucleation phase, as shown in Fig. 1B, which generally occurs in a seeded fibrillogenic process. The reason presumably lies in the fact that endo-all-Ala produces nuclear fibrils when the pH is adjusted to 6.0, which were verified by AFM and FTIR (Figs. 2 and 4B). This nearly neutral pH, at which endo-all-Ala forms fibrils rapidly, occurs commonly in Alzheimer disease with cerebral acidosis (20). In general, fibrillar proteins have an increased resistance to proteolysis (17). Our data presented here may give a reasonable explanation for the trypsin resistance of endo-all-Ala. The distinct morphological types of endo-all-Ala aggregates were identified by AFM in Fig. 2. The absorption of large aggregates (like fibrils) on mica in AFM measurements depends on multiple factors, including the surface chemical properties of the substrate, the pH and ionic strength of the adsorption buffer, and the incubation time (6). Nevertheless, the ability of a certain species to absorb will not change with time. Thus, the relative morphological changes of a species observed at each time point should provide reliable qualitative information during the course of aggregation. In addition, the time-dependent morphologies of endo-all-Ala were also obtained by transmission electron microscope. Moreover, Fink and co-workers (45) reported that surfaces can catalyze the formation of amyloid fibrils. In the process of sample preparation for our ex situ AFM experiments, however, the fibrils were incubated in solution, not on a mica surface; therefore, the possibility that the fibrils of endoall-Ala are induced by the surface can be excluded. The results of our MTT assay proved that the granular aggregates of endo-all-Ala are highly cytotoxic toward PC12 cells with dose dependence, whereas fibrils are less harmful (Fig. 9). These results of fibrillar cytotoxicity not only strongly support the previous proposition that early prefibrillar soluble oligmers from proteins are inherently cytotoxic (41) but also may give us the opportunity to investigate in detail the mechanism of the activity of endostatin in neuronal cells (21).
Contributions of Hydrophobic Surface and Destabilization to Fibril Formation of Endo-all-Ala-The ANS binding assay has been widely applied to probe the exposure of hydrophobic protein surfaces (36). The comparison between endo-WT and endo-all-Ala at pH 6.0 upon binding to ANS in Fig. 5 indicates that, in the case of endo-all-Ala, there are larger hydrophobic patches exposed on its surface than endo-WT, which may be attributed to the deficiency of disulfide bonds. For endostatin, the two disulfide bonds hold the whole molecule so tight that a symmetrical sphere conformation is formed (14). Analogous to other native proteins (46 -48), endo-WT folds with a hydrophobic core buried inside and a large number of charged residues on the surface. When the two disulfide bonds are deficient, the tightly packed conformation of endo-WT will possibly be disrupted; thus, the hydrophobic residues will be released and exposed on the surface. Furthermore, the trend in the titration curve of endo-all-Ala by ANS in Fig. 5D is almost consistent with that of the ThT titration curve in Fig. 1A as a function of pH. Therefore, the exposed hydrophobic clusters may be responsible for the predisposition of endo-all-Ala polymerization for intermolecular hydrophobic interactions. Additionally, charge effects cannot be excluded in the polymerization of endo-all-Ala as a function of pH. Notably, endo-all-Ala incubated at 37°C for 81 h formed the largest amounts of fibrils at the critical pH, where the polymerization began to occur (Fig. 6). This result will be helpful in the rational design of a suitable pH for amyloid fibril formation, which has an implication for further investigation of the mechanism of fibrillogenesis and protein aggregates responsible for cell damage. Above this critical pH, the yield

Fibrillogenesis of Endostatin
of fibrils from endo-all-Ala decreased, accompanied by spontaneous precipitation of this protein. The rapid precipitation reaction of endoall-Ala represents another pathway that competes with and even blocks fibril formation. These data not only suggest that the hydrophobic surface is crucial for fibrillogenesis of endostatin but also sequentially guide us to disturb the formation of ordered aggregates through accelerating precipitation.
Our group has reported that the unique nested pattern of the two pairs of disulfide bonds contributes to the stability of endostatin at neutral pH (15). Endo-all-Ala, lacking the two disulfide bonds, is nearly in a denatured state, given the maximal wavelength of the tryptophan emission fluorescence spectra (Fig. 7A). In addition, the stability of endo-all-Ala decreased dramatically compared with that of endo-WT under moderate acidic conditions in Fig. 7B, which probably also accounts for the propensity of fibril formation of endo-all-Ala. Indeed, after being incubated for 81 h at pH 6.0, 37°C, endo-all-Ala produced long and unbranched fibrils, whereas endo-WT maintained the monomer structure (Fig. 2). The statement that destabilizing proteins facilitates fibrillogenesis even accommodates many other proteins, which include, for example, immunoglobulin light chain (49), ␤2m (50), and superoxide dismutase (51).
Taken together, hydrophobic surface and stability are two important factors to control protein aggregation. When the two disulfide bonds are lacking, endostatin, whose tertiary structure is disturbed, is predisposed to form fibrils with an enhanced hydrophobic surface and decreased stability.
The Presence of Nonnative ␣-Helix Facilitating Fibril Formation of Endostatin-Our group reported previously that the polypeptide chain of endostatin has an intrinsic ␣-helix propensity and that this property is restricted by the two disulfide bonds (15,16). Endo-all-Ala, which lacks disulfide bridges and contains a higher content of ␣-helix in its secondary structure than endo-WT, prefers to form fibrils. TFE can destroy the tertiary structure (39), whereas it can also increase the ␣-helical structure of proteins and peptides (40). Surprisingly, when endo-WT was mixed with TFE of different concentrations, fibrillogenesis of endo-WT was favored along with the increased amount of ␣-helical structure (Fig. 8). These observations are seemingly inconsistent with previous understanding that a high ␣-helical proportion of pro-teins can suppress fibril formation (8). The interpretation is possibly that the formation of ␣-helix is due to a kinetic rather than a thermodynamic preference, whereas amyloid fibril formation may be determined by the stability of the formed ␣-helix. In an environment that favors the conformational stability, ␣-helix would prevent the polypeptide chain from converting into fibril structure. Contrarily, under other conditions where mutagenesis or organic solvents reduce the stability of the ␣-helical structure formed in proteins, the polypeptide is therefore ready to form fibrils. As for endostatin, the ␣-helical structures revealed by the far-UV CD spectra in Figs. 3 and 8A are not in the fully structured native conformation but are induced by either mutagenesis or TFE. This nonnative ␣-helical structure of endostatin is unstable and can easily transform to ␤-structure after incubation (Figs. 3 and 4), which is one of the hallmarks of fibrils. Nevertheless, the amount of fibril formed by endostatin as observed is not linearly dependent on the increased ␣-helical content induced by TFE. This may be due to other factors, such as the extent of denaturation involved in the fibrillogenic process. In summary, our data support the view that ␣-helical structure in native proteins represents a way to suppress fibril formation (8) as well as a supplementary view that the formation of unstable ␣-helix induced in proteins can also facilitate fibril formation.
Biological Significance and Implications-Our systematic study, which focuses on the fibrillogenesis of endo-all-Ala, may elucidate the roles of disulfide bonds in the fibrillogenesis of endostatin. On the other hand, endostatin was recently found to be produced by neuronal cells and co-localize with A␤ in Alzheimer disease (21). However, proteins in the cytosolic environment of the nervous system may be exposed to strong reducing conditions, which include 1-10 mM GSH (42). Additionally, antioxidants as defenses against oxidative stress in neurodegenerative diseases by their free radical scavenging activity accumulate in neurons to ϳ3 mM (24). It is also possible that thiol-disulfide oxidoreductases of the thioredoxin and glutaredoxin pathways (25) could impair disulfide formation. Moreover, superoxide produced under pathological conditions is a potent reducing agent, which can reduce disulfides in proteins (26). Our data showed that endostatin can be reduced by physiological concentrations of GSH, the main cytosolic thiol-disulfide redox buffer (Fig. 10A). The complete reduction of endostatin with GSH requires at least 50 h, which may be due to the compact structure of endostatin with two nested disulfide bonds. This susceptibility of the disulfide bonds to reduction is also possessed by amyotrophic lateral sclerosis-related SOD in vivo (52). Furthermore, the structural analog of disulfide-reduced endostatin, endo-all-Ala, has been proved to form fibrils more easily than endo-WT. The structural similarity between reduced endostatin and endo-all-Ala implies that the behavior of endo-all-Ala may be relevant to that of reduced endostatin in vivo. Hence, we speculate that reduced endostatin in neuronal cells possibly contributes to the accumulation of endostatin in amyloid plaques, and our work may give some biophysical and functional evidence of endostatin in Alzheimer disease. Additionally, protein aggregation is not merely a nuisance factor in many in vitro studies but also causes major economic and technical problems in the biotechnology and pharmaceutical industries (27). Recently, the anti-tumor efficacy of endostatin has been validated in a phase III clinical trial on 493 human non-small cell lung cancer patients (53), and endostatin has just been approved as a new drug by the State Food and Drug Administration of China. Nevertheless, the refolding of endostatin from inclusion bodies of E. coli has been proved to be very difficult, with a recovery of less than 1% at physiological conditions (11). In most instances, there is a kinetic competition between aggregation and folding (54). Our study concentrates on the fibrillogenesis of a disulfide-deficient variant of endostatin in the absence of denaturants and reducing reagents, which may imply that the correct pairing of the two nested disulfide bonds is essential in the refolding of endostatin. On the other hand, the findings of our study may consequently provide some clues for preventing the aggregation of endostatin in extensive preparation, storage, and delivery for therapy via approaches that include reducing hydrophobic patches on the surface and increasing the stability.