Disulfide Bonds Reduce the Toxicity of the Amyloid Fibrils Formed by an Extracellular Protein

The misfolding of proteins into amyloid fibrils constitutes the hallmark of many diseases.[1] Although relatively few physicochemical properties of protein sequences—charge, hydrophobicity, patterns of polar and nonpolar residues, and tendency to form secondary structures—are sufficient to rationalize in general terms their relative propensities to form amyloid fibrils,[2, 3] other properties can also be important. One example is intramolecular disulfide bonds, which limit the ways in which a polypeptide can be arranged in a fibril through the topological restraints that they impose. Although disulfide bonds are present in 15 % of the human proteome, in 65 % of secreted proteins, and in more than 50 % of those involved in amyloidosis, our understanding of how they influence the properties of amyloid fibrils is limited.[4–6] We have examined the formation of fibrils by human lysozyme[7, 8] in the presence and absence (Figure 1 a,b) of its native disulfide bonds, and found that they profoundly influence the fibrillar morphology and cytotoxicity. 
 
 
 
Figure 1 
 
a) Structure (pdb code 1Lz1) of wild-type lysozyme (Lys) with the disulfide bonds shown in red. These were reduced as shown in (b) to obtain LysRA. c, d) Amyloid formation by Lys (c) and LysRA (d) monitored by light scattering (LS) at 500 nm and different ... 
 
 
 
As disulfide bonds stabilize folded proteins, they determine the conditions under which wild-type (Lys) and reduced and alkylated lysozyme (LysRA) are amyloidogenic. In agreement with previous reports, we found that it is necessary to incubate Lys under destabilizing conditions, such as low pH (pH 2.0) and high temperature (≥50 °C), to form amyloid fibrils within 24 h (Figure 1 c and Figure S1 in the Supporting Information).[8–10] By contrast, LysRA is amyloidogenic under milder conditions; at pH 2.0, for example, it forms fibrils at 20 °C (Figure 1 d and Figure S1 in the Supporting Information). We analyzed the conformational properties of Lys and LysRA by NMR spectroscopy and far-UV circular dichroism (CD) as a function of temperature. We found that Lys is folded at 20 °C (Figure 1 e,g) and experiences a well-defined unfolding transition at about 55 °C (Figure 1 g and Figure S2 in the Supporting Information).[10] By contrast, LysRA is unfolded at all temperatures (Figure 1 f,h). Our results, therefore, indicate that the presence of intact disulfide bonds decreases the rate at which lysozyme forms fibrils (Figure 1 c,d) by stabilizing the cooperatively folded native protein.[11] 
 
Disulfide bonds also determine the morphology of the fibrils. After 24 h of incubation under the mildest conditions that lead to aggregation (Figure 1 c,d), both Lys and LysRA had converted into fibrils as shown by transmission electron microscopy (TEM; Figure 2 a,b insets) and by thioflavin T (ThT) and Congo red (CR) binding (Figure 2 c,d and e,f, respectively). We analyzed the samples by far-UV CD and found that in both cases the spectra evolved from those corresponding to largely disordered proteins (Figure 2 a,b, blue) to those of species rich in β-sheet structure, with a minimum in the ellipticity at approximately 217 nm typical of amyloid fibrils (Figure 2 a,b, red). We also analyzed the amide I region (1580–1720 cm−1) of the infrared spectra of the fibrils by using attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectroscopy (Figure 3 a,b and Table S1 in the Supporting Information),[9] and found that LysRA fibrils are less rich in β-sheet structure than those formed by Lys (51.5 versus 72.5 %). 
 
 
 
Figure 2 
 
Aggregation kinetics of a) Lys at pH 2.0 and 60 °C and b) LysRA at pH 2 and 20 °C monitored by far UV-CD. Samples of fibrils formed after 24 h and isolated by ultracentrifugation have a fibrillar morphology (insets in (a) and (b)) and ... 
 
 
 
 
 
Figure 3 
 
a, b) ATR-FTIR spectra of LysRA (a) and Lys fibrils (b), shown in black, with the contributions obtained by curve fitting colored as follows: red: β sheet, green: random/α helix, blue: turns and loops, gray: side chains. c) SDS-PAGE of ... 
 
 
 
To analyze the nature of the fibrillar core of the fibrils, we studied both types of fibrils by limited proteolysis (Figure 3 c,d). It was found that the fibrils formed by Lys are inert to proteolysis under the conditions used here but that those formed by LysRA are readily cleaved (Figure 3 c,d and Figure S3 in the Supporting Information) and have their diameter reduced from (5.1±0.6) to (3.6±0.6) nm (Figure 2 b and Figure S4 in the Supporting Information). The protease-resistant segment of the molecule, attributable to the core, is composed of about 80 residues, from residue 29 to 108. As susceptibility to proteases requires 10 to 12 unfolded residues,[12] our result is consistent with the number of residues in the β-sheet secondary structure determined by FTIR analysis (51.5 %, that is, 67 residues). We also probed the nature of the non-core regions by an 8-anilinonaphthalene-1-sulfonate (ANS) binding assay, in which interactions of this dye with solvent-exposed hydrophobic patches cause a blue shift in the maximum emission wavelength and an increase in emission intensity.[13] We found that the fluorescence intensity of ANS is higher in the presence of LysRA fibrils than in that of fibrils formed from Lys (Figure S5 in the Supporting Information), hence indicating a greater number of solvent-exposed hydrophobic residues. 
 
Since hydrogen-bonding interactions in the cross-β core stabilize amyloid fibrils,[14] we investigated whether differences in core size are reflected in their resistance to disaggregation. We measured the concentration of protein in equilibrium with fibrils at increasing concentrations of guanidine hydrochloride (GdnHCl)[15] and found that the fibrils formed by LysRA disaggregate at lower concentrations of GdnHCl than those formed by Lys (Figure 3 e). Our results indicate that the fibrillar core formed in the presence of disulfide bonds is larger than in their absence, thereby reducing the susceptibility of the fibrils to proteolysis and increasing their stability. 
 
Current evidence suggests that the most toxic forms of amyloid aggregates are not the mature fibrils but their less organized precursors.[16] In addition, recent studies have shown that partially structured fibrils can also give rise to toxicity as a result of their larger accessible hydrophobic area or by their greater tendency to generate toxic oligomeric species by fragmentation.[9] To investigate whether or not disulfide bonds alter the cytotoxicity of the fibrils, samples corresponding to protein concentrations of 5 to 20 μm were added to cultures of SH-SY5Y human neuroblastoma cells and the resulting changes in cell viability were measured using a calcein acetoxymethyl (AM) assay (Figure 3 f). The results, supported by an MTT assay (MTT=3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide; Figure S6 in the Supporting Information), show that the fibrils formed by LysRA have a significantly higher cytotoxic effect than those formed by Lys (p<0.0001; Figure 3 f), in agreement with the finding that ANS binding in amyloid species (Figure S5 in the Supporting Information) correlates with cytotoxicity.[17] This result shows that disulfide bonds can decrease toxicity by favoring the formation of highly structured amyloid fibrils, which suggests that disulfide bonds in extracellular proteins could be the result of evolutionary pressures[18] to minimize toxic aggregation in an environment where the redox potential favors disulfide bond formation. 
 
To investigate this hypothesis further, we analyzed the aggregation propensity of the human proteome using the well-established Zyggregator predictor.[3] We found that the sequences of extracellular proteins have higher intrinsic aggregation propensity than intracellular ones, an observation that has been related to the dilution that occurs upon secretion.[19, 20] We also found that disulfide bonds are associated with sequences of high aggregation propensity (Figure 3 g), which suggests that disulfide bonds have co-evolved with protein sequences[21] to minimize their propensity to form potentially toxic amyloid aggregates. This analysis would explain the high prevalence of disulfide bonds in extracellular proteins, where additional protective mechanisms that reduce misfolding and its consequences are likely to play a less significant role than inside the cell.[20, 22] 
 
The disulfide bonds of lysozyme inhibit the aggregation of this protein into amyloid fibrils by stabilizing the folded state—a fact that can be attributed to the reduction in the entropy of the unfolded state.[11] A partially unfolded state can nevertheless be populated as a result of changes in conditions, as in this work, or by mutations, as in patients with nonneuropathic systemic amyloidosis.[8] We have shown that, when this situation occurs, disulfide bonds allow the formation of fibrils with a large proportion of their sequence in the cross-β conformation. This result is at first sight unexpected, as one might have anticipated that the conformational constraints resulting from cross-linking would reduce the ability of the chain to fold into the complex β-sheet amyloid structure. It is, however, clear that lysozyme and other disulfide-linked proteins are able to form fibrils that contain a high fraction of sequence in the cross-β structure.[23] 
 
We conclude that intramolecular disulfide bonds can stabilize amyloid fibrils, as they do for the folded state, by decreasing the entropic penalty associated with the formation of this ordered form of protein structure. This can be concomitant with significant decreases in the toxicity of the resulting fibrils, which suggests that disulfide bonds have co-evolved with protein sequences to reduce toxic aggregation.[21]


Methods
Reduction and alkylation of human lysozyme -Wild-type human lysozyme (Lys) was dissolved at 0.1mM in the reduction buffer (6M GdnHCl, 0.1M TrisHCl, pH 8.5), a 10-fold molar excess of tris(2-carboxyethyl) phosphine hydrochloride was then added and the pH adjusted to 8.5. After incubation at 25°C for 2 hours, a 10-fold molar excess (over the total number of sulfhydryl groups in the protein) of iodoacetamide was added. The mixture was incubated 1 h at 25°C in the dark and then dialyzed extensively against 1% formic acid (pH 2.0). The purity of the reduced and alkylated protein (Lys RA ) was checked by SDS-PAGE and RP-HPLC. Quantitative derivatization of sulfhydryl groups was verified by ESI mass spectrometry. An increase in mass of 464.2 Da, relative to the Lys, was observed, corresponding to the addition of a carbamidomethyl group to each one of the eight cysteine residues of the lysozyme molecule.
Optical spectroscopy -Protein concentrations were evaluated from absorption measurements at 280 nm on a single-beam Cary 400 Scan spectrophotometer (Varian, Palo Alto, CA, USA). The extinction coefficients at 280 nm, calculated with the method of Gill & von Hippel (1989) [1] , were 36940 cm -1 M -1 for Lys and 36440 cm -1 M -1 for Lys RA .
Fluorescence measurements were carried out on a Varian (Palo Alto, CA, USA) model Cary Eclipse spectrofluorimeter in a temperature-controlled cell holder, utilizing a 2 mm x 10 mm path length cuvette. ThT binding was monitored by exciting the sample at 440 nm and recording the emission fluorescence spectrum from 450 to 600 nm. For each measurement, 8 µl of a 2.5 mM ThT stock solution prepared in PBS were added to a volume of fibrils corresponding to 20 µg and a volume of 0.5 ml was reached with the phosphate buffer. For the Congo red spectroscopic assay, 50 µl of a 70 µM fibril sample was added to 1 ml of 5mM phosphate buffer, 150mM NaCl pH 7.4 containing 1mM Congo red. The UV spectra between 400 and 700 nm were immediately acquired at 20°C: the difference between the resulting spectra and the one measured with only Congo red in the absence of protein was used as the effective bound Congo red absorbance [2,3] . For 8-anilo-1-naphtalene-sulfonic acid (ANS) titration, aliquots of ANS from a stock solution in 20 mM glycine (pH 2.0) containing 0.1 M NaCl, were added to the isolated fibrils, to a final ANS concentration ranging from 0 to 200 µM. The final protein concentration was 5 µM in all cases. The spectra were immediately acquired at 20°C, using an excitation wavelength of 350 nm and an emission range from 380 to 700 nm. The difference between the resulting fluorescence intensity at 470 nm and that measured with only ANS in the absence of protein was used as the effective bound ANS fluorescence. The extinction coefficient at 350 nm of ANS was 4950 cm -1 M -1 [4] .
Far-UV CD spectra of Lys and Lys RA were acquired in 20 mM glycine·HCl, 100 mM NaCl pH 2.0. Proteins were diluted to a final concentration of 7 µM and the spectra were acquired at 20 or 60°C using a 1 mm pathlength cuvette and a J-810 Jasco spectropolarimeter (Tokyo, Japan), equipped with a thermostated cell holder. For secondary structure analysis of fibrils, samples were analyzed in a Bruker BioATRCell II using a Bruker Equinox 55 Fourier transform infrared spectroscopy (FTIR) spectrometer (Bruker Optics Limited, UK) equipped with a liquid nitrogen cooled mercury cadmium telluride (MCT) detector and a silicon internal reflection element (IRE). For each spectrum 256 interferograms were coadded at 2 cm -1 resolution, and the buffer background was independently measured and subtracted from each protein spectrum before curve fitting of the amide I region (1720-1580 cm -1 ). Calculation of the second derivatives was used to identify peak maxima. Using this information, the raw spectra were then fitted to a series of Gaussian peaks with the identified absorbance maxima using an iterative curve-fitting procedure performed in Origin8 (OriginLab Corporation, MA, USA).
NMR spectroscopy of Lys and Lys RA at pH 2.0 -Uniformly 15 N-labelled human lysozyme was expressed and purified as described previously [5] , by growing cells with 15 (NH 4 ) 2 SO 4 as the sole nitrogen source. NMR samples containing 70 µM 15 Nlabelled Lys or Lys RA lysozyme in 0.1 M NaCl, 20 mM glycine buffer pH 2.0, made with 95% (v/v) H 2 O/5% (v/v) 2 H 2 O were prepared. 2D 15 N-1 H HSQC spectra were measured at 293 K in a Bruker Digital Avance Spectrometer at 800 MHz.
SDS-PAGE analysis -Fibrillar samples, previously dissolved in DMSO, were analyzed by SDS-PAGE using 4%-12% Bis-Tris NuPAGE gels (Invitrogen UK) in MES buffer under reducing conditions. Gels were stained using coomassie brilliant blue.

RP-HPLC and Mass
Spectrometry -RP-HPLC analyses were performed on a Vydac C18 column (4.6 mm × 150 mm; The Separations Group, Hesperia, CA) eluted with a linear gradient of acetonitrile containing 0.05% (v/v) trifluoroacetic acid from 5% to 22% in 5 min and from 22% to 50% in 17 min, at a flow-rate of 0.6 ml/min. The effluent from the column was monitored by measurement of the absorbance at 226 nm. The identity of protein fragments was assessed by nanoelectrospray mass spectrometry (NanoESI-MS) with a LTQ-FT Ultra mass spectrometer (Thermo Scientific, USA). The monoisotopic mass values were analyzed using massXpert.
Transmission Electron Microscopy -Samples were applied to formvar-coated nickle grids, stained with 2% (w/v) uranyl acetate solution and imaged on a JEOL 1010 transmission electron microscope operating at 80kV. Images were taken with a Megaview III camera and digitized with the software AnalySIS (Soft Imaging System). TEM images were analysed using ImageJ.  Proteolysis of fibrils -Lysozyme fibrils, isolated by ultracentrifugation and resuspended at 70 µM, were subjected to proteolysis in the aggregation buffer at 20°C, using pepsin at an enzyme:substrate ratio of 1:50 (w/w). After proteolysis, fibrils isolated by ultracentrifugation were dissolved in DMSO for SDS-PAGE analysis and in 7.4 M GdnHCl under shaking for analysis by RP-HPLC.
Cell Culture -SH-SY5Y cells were cultured in Dulbecco's Modified Eagle Medium (Invitrogen) with the addition of 10% fetal bovine serum at 37°C in a humidified 5% CO 2 incubator. The cells were plated in Costar (3595) 96-well plates (Corning) using serum-free Neurobasal medium (GIBCO) and were then incubated with 1 µM aggregates for 48 h. The percentage of viable cells present after 48 h was assessed by adding Calcein AM and measuring fluorescence of calcein, the product of its hydrolization by intracellular esterases, and by adding MTT and measuring the absorbance of the product of its reduction by mitochondrial dehydrogenases.

Proteome analysis and aggregation propensity -
The set of all entries belonging to Human Complete Proteome was downloaded from Uniprot (2010_12, [6] , www.uniprot.org) and redudancy was eliminated at a sequence identity threshold of 100%. All entries with a localization associated with the membrane, the ones marked as fragments and the proteins with an existence level termed "uncertain" were fuiltered out from the dataset. The set of proteins was divided in 4 classes: intracellular with disulfide bonds, intracellular without disulfide bonds, extracellular with disulfide bonds, extracellular without disulfide bonds. Classification of every entry was done on the basis of the Uniprot annotation (presence or absence of disulfide bonds and annotated cellular location). For the calculation of the intrinsic propensity profiles for aggregation of polypeptide sequences, the intrinsic aggregation propensities of individual amino acids are defined as: values are combined to provide a profile, A p , which describes the intrinsic propensity for aggregation as a function of the complete amino acid sequence (1). At each position i along the sequence we define the profile A p as an average over a window of seven residues: where € I i pat is the term that takes into account the presence of specific patterns of alternating hydrophobic and hydrophilic residues and € I i gk is the term that takes into account the gatekeeping effect of individual charges c i : In order to compare the intrinsic propensity profiles of different proteins, we normalized A p by considering the average (m A ) and the standard deviation (s A ) of € A i p at each position i. We thus obtain the normalized intrinsic aggregation propensity profile: We calculated the average µ and the standard deviation σ over a set of random sequences: The parameters α are determined by minimizing the error between positive values of € Z i p (i.e., the Zagg score) and experimental aggregation rates.
In the present study, the Zagg score is normalized using Eq. (4) and 20400 human sequences available from Uniprot (http://www.uniprot.org/).
In the analysis of intracellular and extracellular sequences, proteins with sequence similarity higher than 50% were filtered out. For this purpose, the CD-HIT algorithm (http://weizhong-lab.ucsd.edu/cd-hit/) was applied to the four classes (i.e., intracellular with disulfide bonds, intracellular without disulfide bonds, extracellular with disulfide bonds, extracellular without disulfide bonds) that were employed for the analysis. The number of proteins employed in this work is reported below: Secreted proteins without disulfide bonds: 574 Secreted proteins with disulfide bonds: 1072 Non-secreted proteins without disulfide bonds: 9395 Non-secreted proteins with disulfide bonds: 164

S5
With respect to the statistical significance of Zagg the un-equal variance t-test was used: i) Extracellular proteins with and without disulfide bonds are discriminated by Zagg with p<=0.001, ii) Intracellular proteins with and without disulfide bonds are discriminated by Zagg with p<=0.001. Fig. S1 -Thioflavin T fluorescence in the presence of Lys RA (a) and Lys (b) samples after 24 h of aggregation at different temperatures (see Figure 1 c-d). SDS-PAGE of Lys RA (a) and Lys (b) fibrils isolated by ultracentrifugation after 24h of aggregation at different temperatures. As the aggregation temperature increases Lys RA undergoes more acid-catalyzed hydrolysis of Asp-X peptide bonds [7] and the fibrils are formed by a larger amount of protein fragments. However, for the Lys protein, the presence of fragments in the fibrils is negligible. The different sensitivity of the two proteins to acid-catalyzed hydrolysis of Asp-X peptide bonds is likely due to the fact that the Lys RA protein is more unfolded and therefore more exposed and flexible than the Lys protein.