Sulfated Polysaccharides Promote the Assembly of Amyloid β1–42 Peptide into Stable Fibrils of Reduced Cytotoxicity*

The histopathological hallmarks of Alzheimer disease are the self-aggregation of the amyloid β peptide (Aβ) in extracellular amyloid fibrils and the formation of intraneuronal Tau filaments, but a convincing mechanism connecting both processes has yet to be provided. Here we show that the endogenous polysaccharide chondroitin sulfate B (CSB) promotes the formation of fibrillar structures of the 42-residue fragment, Aβ1–42. Atomic force microscopy visualization, thioflavin T fluorescence, CD measurements, and cell viability assays indicate that CSB-induced fibrils are highly stable entities with abundant β-sheet structure that have little toxicity for neuroblastoma cells. We propose a wedged cylinder model for Aβ1–42 fibrils that is consistent with the majority of available data, it is an energetically favorable assembly that minimizes the exposure of hydrophobic areas, and it explains why fibrils do not grow in thickness. Fluorescence measurements of the effect of different Aβ1–42 species on Ca2+ homeostasis show that weakly structured nodular fibrils, but not CSB-induced smooth fibrils, trigger a rise in cytosolic Ca2+ that depends on the presence of both extracellular and intracellular stocks. In vitro assays indicate that such transient, local Ca2+ increases can have a direct effect in promoting the formation of Tau filaments similar to those isolated from Alzheimer disease brains.

Pathogenesis in Alzheimer disease (AD) 3 is linked to the accumulation of the highly amyloidogenic self-associating amyloid ␤ peptide (A␤). The amyloid cascade hypothesis postulates that AD pathology is initiated by an extracellular accu-mulation of A␤ that in turn triggers a transmembrane signal having as ultimate effect the formation of neurofibrillary tangles by the microtubule-associated protein Tau (1-3), followed by collapse of the microtubular cytoskeleton. Some of the mechanisms that have been proposed to explain how extracellular A␤ exerts its cytotoxic effects include the promotion of oxidative stress (4), disruption of Ca 2ϩ homeostasis (5,6), the targeting and functional disruption of particular synapses by A␤ oligomers (7), and the stimulation of synthesis and release of toxic molecules such as nitric oxide (8). It is generally accepted that the aggregative state of A␤ is a paramount issue in determining its degree of neurotoxicity. Existing data have shown that soluble A␤ oligomers may be the key effectors of cytotoxicity in AD (1, 7, 9 -11). Other results, however, indicate that the neurotoxic activity of A␤ requires its aggregation in fibrillar form (12,13), yet it has also been suggested that the aggregates may be a mechanism of defense acting by concealing and immobilizing neurotoxic soluble A␤ (14). Among the different forms of A␤, the 42-residue fragment (A␤ 1-42 ) readily self-associates and forms nucleation centers from where fibrils can quickly grow. In a previous work, we characterized the different species that appear during A␤  fibrillogenesis in vitro, from oligomers through protofibrillar forms to mature fibrils (15). Although active debate is centered on the issue of which is the pathogenic species of the peptide that ultimately causes the synaptic loss and dementia associated with AD (16), comparatively less effort is being devoted to the study of endogenous factors that can promote or inhibit the formation of the candidate neurotoxic forms.
There are evidences indicating that the appearance of insoluble fibrillar structures enriched in ␤-sheets is facilitated by diverse environmental factors (17). The amounts of A␤ in vivo are generally much smaller than the concentrations required to induce fibril formation in vitro. It is therefore likely that other molecules exist that play an important role in the formation and deposition of amyloid fibrils and plaques. Carbohydrates are ubiquitous components of plasma, like glucose, that is found in brain tissue in millimolar amounts (18), and are also present in long-lasting structures of the extracellular matrix (ECM) such as proteoglycans. Proteoglycans are highly glycosylated proteins often carrying multiple negatively charged polysaccharides termed glycosaminoglycans (GAGs), which are among the most abundant components in the ECM of many tissues, including the brain (19,20). From a chemical point of view, GAGs are unbranched polymers of repeated disaccharide units, usually containing an amino sugar and a uronic acid. Except in the case of hyaluronan, GAGs have varying degrees of sulfation.
Proteoglycans and GAGs are found associated with all types of amyloid deposits (21)(22)(23) and have been implicated in the nucleation of fibrils (24). They can also stabilize mature fibrils against dissociation (25) and proteolytic degradation (26), and they have been shown to facilitate the formation of fibrils of amylin (27), apo-serum amyloid A (28), ␣-synuclein (29), prion protein (30), Tau (31), and A␤ (32). It has been proposed that GAGs may have a scaffolding role, promoting fibrillogenesisprone conformations of the amyloid precursor proteins (21). A␤  and other aggregating peptides like prion protein peptide 106 -126 and Tau peptide 317-335 share cationic motifs that may be involved in binding to the negative charges of sulfated GAGs (33,34). Other linear anionic polymers such as nucleic acids have also been described to bind prions with high affinity (35,36), suggesting that polymeric molecules with repeating anionic and hydrophobic surfaces might be efficient inducers of amyloid structure. Chondroitin sulfate proteoglycans have been found to be associated with the lesions of AD (37), and heparan and chondroitin sulfate GAGs attenuate the neurotoxic effect of A␤ in primary neuronal cultures (38) and in neuron-like cell lines (39). Because GAGs can bind A␤, it is conceivable that GAG-mediated neuroprotection is due to the sequestering of A␤, in agreement with the hypothesis that the generation of senile plaques in AD would be a partially protective response aimed at reducing A␤ neurotoxicity. Supportive of this view is the finding that low susceptibility of neurons and cortical areas to neurofibrillar deposition corresponds with high proportions of aggregating chondroitin sulfate proteoglycans in the neuronal microenvironment (40). Despite this clearly established relationship between GAGs, A␤, and AD onset and progression described above, little is known about the type of A␤-containing structures induced by carbohydrates in general and by GAGs in particular.
Much attention has focused on A␤ as causative agent for AD, but accumulating evidence points to disruptions in neuronal Ca 2ϩ signaling as a consistent progenitor for the disease, occurring prior to the development of the histopathological markers and cognitive decline (41). The Ca 2ϩ hypothesis of AD proposes that sustained and accumulated alterations in Ca 2ϩ homeostasis are a proximal cause in neurodegenerative diseases (42). Thus, alterations in the permeability to Ca 2ϩ of the cell membrane could be one of the events linking A␤ and Tau aggregation, the two processes central to the amyloid cascade theory. It is likely that if A␤ has a role in permeabilizing the cell membrane to Ca 2ϩ , this effect will depend on its aggregated state, i.e. globular oligomers and fibrils might elicit different responses. In turn, because A␤ aggregation is influenced by ECM components such as GAGs, these might be able to modulate a putative A␤-induced rise in cytosolic Ca 2ϩ . To tackle these issues, we have exposed A␤  to carbohydrates, studying the structures formed, their cytotoxicity, their ability to alter the permeability to Ca 2ϩ of culture cells, and the effect of A␤-induced Ca 2ϩ dyshomeostasis on Tau aggregation.

EXPERIMENTAL PROCEDURES
Preparation of A␤  and Tau-Chemicals and reagents were purchased from Sigma, except where otherwise indicated. A␤ 1-42 synthesized by Peptide Institute, Inc. (Japan) was purchased lyophilized in glass vials and stored at Ϫ80°C immediately upon arrival. As a rule, lyophilized A␤ 1-42 was dissolved to 10 M in 10 mM phosphate buffer, pH 7.4 (PB). When complete disaggregation of A␤ 1-42 was required, we followed a variant of Zagorski's protocol (43) as described (15). Samples were incubated in the dark for the times indicated in 1.5-ml Eppendorf tubes at 37°C with gentle rocking. For the production of control oligomeric or fibrillar preparations of A␤ 1-42 we followed established protocols (9). Briefly, oligomers were prepared by diluting 5 mM A␤   Electrophoresis and Immunoblots-Polyacrylamide gel electrophoresis was performed as described previously (15), in Tricine gels containing 0.1% SDS. Immunoblots were transferred to (and dot blots spotted on) a polyvinylidene difluoride membrane (Immobilon, Millipore) with a Mini Trans-Blot Cell (Bio-Rad), blocked in 0.1 M Tris-HCl, pH 7.5, 0.5% Tween 20, 1% Triton X-100, 3% bovine serum albumin, and incubated in the presence of rabbit anti-A␤ 1-40 , mouse anti-chondroitin-4sulfate (Acris Antibodies), or mouse anti-Tau (Zymed Laboratories Inc.), diluted 1:2000 in blocking solution. The enhanced chemiluminescence Western blotting detection system (Amersham Biosciences Corp.) was used to visualize the decorated bands or dots.
Atomic Force Microscopy-Images were obtained with a commercial MultiMode atomic force microscope controlled by a Nanoscope IV electronics (Digital Instruments, Santa Barbara, CA) equipped with either a 12-m scanner (E-scanner) or a 120-m scanner (J-scanner), or with a commercial MFP-3D (Asylum Research, Santa Barbara, CA). Oxide-sharpened pyramidal Si 3 N 4 tips mounted on triangular 100-m long cantilevers (k ϭ 0.08 N/m) were purchased from Olympus (Tokyo, Japan). Except where otherwise indicated, images were taken in liquid using a tapping mode cell without the O-ring seal, following established protocols (15). After the indicated incubation times and immediately before imaging, 10 l of the sample were allowed to adsorb for about 5-10 min at room temperature on freshly cleaved muscovite mica (Asheville-Schoonmaker Mica Co.) or highly ordered pyrolytic graphite (Nt-MDT Co., Zelenograd, Moscow, Russia), and finally overlaid with ϳ100 l of the corresponding incubation buffer. Mica-supported dipalmitoylphosphatidylcholine lipid bilayers were prepared as described (44). For images taken in air, the surface was carefully rinsed with deionized water and gently dried under a N 2 stream. Except where otherwise indicated, z scale for amplitude images was 0.2 V.
Cell Cultures and Ca 2ϩ Imaging-Cell culture media and other reagents were obtained from Biological Industries, unless specified otherwise. SH-SY5Y human neuroblastoma cells (CRL-2266, ATCC) were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (PAA), glutamine (2 mM), penicillin (50 units/ml), and streptomycin (50 g/ml). Cells were maintained at 37°C in the presence of 5% CO 2 , and the medium was replaced every 2 days. Cells were plated at a density of ϳ5000 cells/dish in 35-mm glass-bottom culture dishes (MatTek Corp.) for at least 24 h before transfection. One g of plasmid encoding cameleon YC3.60 (45) (kindly provided by Dr. Atsushi Miyawaki, RIKEN Brain Science Institute, Saitama, Japan) was mixed with 3 l FuGENE-6 transfection reagent (Roche Diagnostics Corporation, Basel, Switzerland) in 1.5 ml Opti-MEM serum-free medium (Invitrogen), and added to the cells. After 6 h, the medium was removed and cells were incubated in Ham's F-12 containing 15% fetal calf serum for 4 -6 days. One day before the measurements, the culture medium was changed to Ham's F12 containing 0.5% fetal calf serum. In the experiments where extracellular Ca 2ϩ was depleted, cells were rinsed twice with Hanks' balanced salt solution without Ca 2ϩ /Mg 2ϩ , and medium was changed to Hanks' balanced salt solution without Ca 2ϩ /Mg 2ϩ containing 2 mM EGTA before application of the amyloids. Thapsigargin (Molecular Probes, Eugene, OR), a specific blocker of sarcoplasmic/endoplasmic reticulum calcium ATPase Ca 2ϩ pumps, was added at a final concentration of 2 M in Ham's F-12 containing 0.5% fetal calf serum. Amyloids were applied by pipetting a fixed aliquot (100 l) of the preincubated solution into the temperature controlled recording chamber containing the cells in 900 l of medium (for a 1-ml final volume). Controls included the incubation of cells in HCl-containing medium, which in the absence of A␤ 1-42 did not induce any Ca 2ϩ response. The Ca 2ϩ ionophore ionomycin (Calbiochem) was used as a control at the end of each experiment to saturate YC3.60. Cells were visualized with a confocal laser scanning microscope Leica TCS SP2 (Leica Lasertechnik GmbH, Mannheim, Germany) adapted to an inverted Leitz DMIRBE microscope with a CO 2 -and temperature-controlled atmosphere (Life Imaging Services, Reinsch, Switzerland). Cameleon YC3.60 was excited with a 458 nm line of an argon ion laser. Cyan fluorescent protein (CFP) and yellow fluorescent protein (YFP) signals were collected by using a 458/514 double dichroic mirror and 510/590 (YFP) and 465/530 (CFP) emission ranges. Images (512 ϫ 512 pixels) were acquired every 5 s at a 400-Hz scan rate with a 40ϫ Leitz Pla-Apochromat objective (numerical aperture 1.25, oil). To quantitatively analyze the changes in fluorescence resonance energy transfer, raw CFP and YFP fluorescence intensity were quantified in Leica LCS software and the data obtained were exported to Microsoft Excel for further analysis. Fluorescence intensity was calculated as the average within the regions drawn around individual cells, and a suitable threshold was set to avoid background signal. Signals are expressed as normalized fluorescence ratio (R/R o ), where R is the fluorescence ratio at any time calculated as F YFP /F CFP and R o is the basal fluorescence ratio, which is obtained by averaging 20 scans before amyloid stimulation.
Immunocytochemistry-SH-SY5Y cells were cultured on coverslips in Petri dishes as described above. After incubation for the specified times in the absence or presence of A␤  HCl fibrils, cells were rinsed with PBS and fixed for 15 min with a solution of 3% paraformaldehyde in PBS containing 60 mM saccharose. Fixed cells were washed with PBS containing 20 mM glycine and permeabilized for 10 min with a solution of 0.05% Triton X-100 and 20 mM glycine in PBS. To block nonspecific antibody binding, coverslips were incubated for 20 min at room temperature in a blocking PBS solution containing 1% bovine serum albumin and 20 mM glycine. Cells were then incubated at 37°C in a humidified chamber for 1 h with monoclonal antibody anti-Tau or anti-␣-tubulin diluted in blocking solution at 20 g/ml. After a washing step, cells were incubated for 1 h at room temperature with secondary antibody goat anti-mouse conjugated to Alexa Fluor 546 (Molecular Probes). Confocal images were obtained with an Olympus Fluoview 500 confocal scanning laser microscope adapted to an inverted Olympus IX-70 inverted microscope and a 60ϫ Plan Apochromatic objective (numerical aperture 1.4, oil). Alexa Fluor 546 was excited at 543 nm with a He-Ne laser, and image size was 1024 ϫ 1024 pixels. Three-dimensional maximum projection images were obtained from 12 serial optical sections (z-step ϭ 500 nm) at the confocal microscope normal rate.
Fluorescence Spectroscopy and CD Assays-A stock solution of 10 mM thioflavin T (ThT) prepared in PB was filtered and diluted in PB to a final concentration of 20 M. Except where otherwise indicated, samples containing 20 M A␤ 1-42 and/or 2 mg/ml carbohydrate were mixed with an equal volume of 20 M ThT. Fluorescence was measured in 348-well plates (Nunc) using a BioTek FL600 spectrofluorometer with excitation and emission wavelengths of 440 and 485 nm, respectively. The reported values have been corrected by subtracting the buffer fluorescence in the absence of amyloid and/or carbohydrate. Samples were prepared in triplicate for each experiment.
CD measurements were performed with a UV-visible Jasco 715 spectropolarimeter equipped with a Peltier temperature control system using a 1-mm path length silica quartz cuvette (Hellma). Samples were measured at wavelengths between 190 and 250 nm with a 1 nm step resolution and an integration time of 2 s. Spectra recorded at 25°C were an average of 20 scans baseline-corrected from where the buffer spectrum was subtracted. The contribution of CSB to the CD signal in the A␤/CSB mixture was removed by subtracting the corresponding spectrum.
Cell Viability Assays-SH-SY5Y were plated at 20,000 cells/ well in 96-well plates in 100 l of Ham's F-12 medium with 15% fetal calf serum. After 24 h at 37°C in 5% CO 2 atmosphere the medium was substituted by 100 l of peptide-containing solution prepared by dissolving 10 l of concentrated A␤ 1-42 in 90 l of Ham's F-12 with 0.5% fetal calf serum, to reach the desired final A␤ concentration. Incubation was resumed for the times indicated. 10 l of 4-[3-(4-iodophenyl)-2-(4-nitrophenyl)-2H-5-tetrazolio]-1,3-benzene disulfonate labeling reagent (WST-1, Roche Diagnostics GmbH) was added to each well, and the plate was incubated in the same conditions for 4 h. After thor-oughly mixing for 1 min on a shaker, the absorbance of the samples was measured at 450 nm. WST-1 in the absence of cells was used as blank. Samples were prepared in triplicate for each experiment.
Transmission Electron Microscopy-A␤ fibrils were fixed with 2% paraformaldehyde and 2.5% glutaraldehyde in PB and kept in the fixative for 3 h at 4°C. Fixed fibrils were washed and postfixed at 4°C with 1% osmium tetroxide in the same buffer containing 0.8% potassium ferricyanide. Then, the sample was dehydrated in acetone, with addition of 5% uranyl acetate in the 70% acetone step, infiltrated with Lowicryl HM20 resin during 2 days, embedded in the same resin, and polymerized at 60°C for 48 h. Ultrathin sections were obtained with a Leica Ultracut UCT ultramicrotome and mounted on Formvar-coated copper grids. Sections were finally stained with 2% uranyl acetate in water and 0.8% lead citrate. For Tau samples, a 5-to 10-l aliquot was placed on a Formvar-coated copper grid, allowed to stand for 1 min, washed with distilled water, and finally stained with 2% uranyl acetate for 1 min. All samples were observed in a JEM-1010 electron microscope (Jeol, Japan) operated at 80 kV.

Protofibrillar and Oligomeric A␤ 1-42 Can Form Different
Types of Associations-Atomic force microscope (AFM) visualization on highly oriented pyrolytic graphite (HOPG) provides a good resolution of molecular species that have affinity for hydrophobic surfaces. During the first stages of aggregation, A␤ 1-42 can adopt either a globular or a fibrillar structure depending on the method used to prepare the sample (Fig. 1). A␤ 1-42 directly dissolved in phosphate buffer or PBS quickly forms protofibrils up to 100 nm long (Fig. 1, A-D). Protofibrils have a characteristic ribbon-like shape ϳ1.5 nm high and ϳ5.5 nm wide (15) and can associate in elongated fibril-like clusters (Fig. 1, A and B). AFM images taken minutes apart have revealed rearrangements of individual protofibrils (Fig. 1, C and D) that include shuffling and growth. Pretreatment of A␤  with trifluoroacetic acid reduces the formation of protofibril nucleating centers, and globular structures with a mean height of ϳ5 nm deposit on a hydrophilic mica substrate as the major initial species (15). When this sample is imaged on HOPG, globular structures can be observed forming patches, or "rafts" (Fig. 1E), that tend to be confined between HOPG steps.
The association of protofibrils and/or of globular oligomers generates ϳ11.5 nm-wide fibrils of the nodular type (Fig. 1F), formed by the stacking of ϳ100 nm-long protofibril bundles (15). Nodular fibrils may evolve further to yield mature smooth fibrils, the highest order of A␤ 1-42 fibrillogenesis (15). High resolution AFM images of the globular structures deposited on HOPG (Fig. 1G) show that they have a diameter similar to that of fibrils, but their height varies in apparently discrete steps of ϳ2 nm (Fig. 1H), suggesting that the higher structures are stacks of several individual units.
Sulfated Polysaccharides Promote the Formation of Amyloid Structure-ThT fluorescence analysis showed that sulfated polysaccharides induced a Ͼ100% increment in the fluorescence of A␤ 1-42 -containing solutions ( Fig. 2A), an indication of increasing ordered amyloid ␤-sheet structure. This effect was not exhibited by control carbohydrate samples including mon-osaccharides, the chondroitin sulfate disaccharide unit ␣-⌬UA-[133]-GalNAc-4S, and the positively charged polysaccharide chitosan. According to the available literature, the concentration of A␤ 1-42 in the brain ranges from ϳ10 ng/ml as found in brain extracts or cerebrospinal fluid (46) to Ͼ1 g/ml inside tightly packed amyloid fibrils, indicating that locally the peptide:carbohydrate ratios will span a wide range of values, including the ones used in this work. Although at the concentrations used monosaccharides like glucose do not promote the formation of amyloid structure as sulfated polysaccharides like CSB do, A␤ 1-42 incubated in the presence of either has significantly less cytotoxicity than the peptide alone 48 h after addition to SH-SY5Y cells (Fig. 2B). Thus, to further characterize the effect of carbohydrates on A␤ 1-42 fibrillogenesis and cell viability we selected CSB and glucose, two abundant endogenous components of the brain ECM.
For comparative AFM studies of fibril formation where high resolution was not required we have used a mica substrate, which is less likely than HOPG to affect fibril deposition (47). A␤ 1-42 incubated at 37°C in PB forms fibrils and globular aggregates (Fig. 2, C and E), but in the presence of 2 mM glucose the sample is abundantly populated by globular structures with a height of ϳ2-5 nm and a width similar to that of fibrils (Fig.  2D). In the presence of CSB there is a remarkable enrichment in the number and length of fibrillar structures (Fig. 2F). These CSB-induced fibrils are very persistent and can be stored at 4°C for up to 40 days without apparent major alterations (Fig. 2H). Incubation of CSB in the same conditions but in the absence of A␤ 1-42 did not yield any detectable structures by AFM imaging on graphite or mica.
The globular and fibrillar structures formed by A␤ 1-42 during its aggregation do not enter the SDS gels used for Western blot analyses (Fig. 2I). However, in the presence of CSB, smaller species are formed that do enter the upper part of the gel. From their size, ranging from 200 to ϳ1000 kDa, these entities correspond to protofibrils. Protofibrils are highly hydrophobic and do not settle on mica efficiently (15), which explains why they are not observed in the CSB-containing samples shown in Fig. 2  (F and H). The less aggregative peptide A␤ 1-40 is not much affected by the presence of CSB (Fig. 2I). A␤ 1-42 fibrils formed in the presence of CSB were centrifuged down and washed with PB, and the resulting pellet was examined in dot blot assays for the presence of CSB. Whereas A␤  was abundant in the fibril precipitate, CSB was not detected in dots containing up to 100 g of A␤ 1-42 (data not shown). Anti-CSB antibody controls were able to detect down to 1 g of the GAG.
CSB and Glucose Reduce the Cytotoxicity of A␤  Oligomers and Fibrils-As a reference to study glucose-and CSBinduced oligomers and fibrils, we followed previously established protocols for the generation of both types of structures (9). Essentially, fibrils were obtained by incubation in 10 mM HCl (HCl fibrils), whereas oligomers were prepared by incubation in Ham's culture medium (Ham's-oligos). Both Ham's-oligos and HCl fibrils showed concentration-dependent cytotoxicity on the neuroblastoma cell line SH-SY5Y (Fig. 3A).
To study the effect of carbohydrates on the toxicity of oligomers and fibrils, A␤  was incubated in the same conditions described above for the formation of both types of structures, including in this case glucose or CSB at the start of the process. After addition of the A␤-containing samples to cells, aliquots were removed from the culture medium at 6 and 24 h of incubation and analyzed by Western blot (Fig. 3B). After 6 h the medium treated with Ham's-oligos showed a dominant band corresponding to a ϳ55-kDa species, whereas the medium treated with HCl fibrils contained aggregated structures that did not enter the gel, consistent with the presence of large fibril-like forms. The presence of glucose and CSB stimulated aggregation in oligomer-forming conditions and slowed down aggregation in fibril-forming conditions. Beyond 6 h of incubation with the cells, oligomeric species were significantly reduced, probably as a consequence of their progressive incorporation into fibrillar or aggregated forms. Based on these results, we explored the effect of the presence of glucose and CSB in oligomer-and in fibril-forming conditions on the viability of cells 6 h after addition of the samples to the cultures.
The results indicate that preincubation in the presence of both carbohydrates significantly reduces the toxicity of oligomers and fibrils, as it does for untreated A␤ 1-42 (Fig. 3C). This effect is particularly clear for CSB, which is able to restore 100% cell viability.
Oligomeric and Fibrillar A␤  Forms Have Different Effects on Cytosolic Ca 2ϩ Levels-We have used fluorescence resonance energy transfer technology to study if the several oligomeric and fibrillar forms of A␤ 1-42 had a differential effect on the levels of cytosolic Ca 2ϩ in culture cells. SH-SY5Y cells that had been transfected to express the Ca 2ϩ -sensitive indicator yellow chameleon YC3.60, composed of cyan and yellow fluorescent proteins (45), were treated with A␤ 1-42 incubated in a variety of conditions, and the corresponding variations in cytosolic Ca 2ϩ were analyzed by confocal fluorescence microscopy (Fig. 4A). Of all samples tested, only HCl fibrils had a dramatic effect in triggering a pulse of intracellular Ca 2ϩ increase followed by a drop in cytosolic fluorescence below resting conditions. Ionomycin applied after HCl fibril exposure did not result in the expected rise in cytosolic fluorescence. The presence of CSB in HCl fibril-forming conditions did not significantly affect the initial Ca 2ϩ pulse, but eliminated the subsequent fluorescence drop. We examined the possibility that the effect induced by HCl fibrils could result from the influx of extracellular Ca 2ϩ or from Ca 2ϩ liberated from intracellular stores. To do so, cells were either placed in Ca 2ϩ -free medium or treated with thapsigargin, respectively. The absence of extracellular Ca 2ϩ in the presence of intracellular Ca 2ϩ completely abolished the normal response to HCl fibrils. Following pretreatment with thapsigargin, which induces the leakage of Ca 2ϩ from the ER into the cytosol, but in the presence of extracellular Ca 2ϩ , HCl fibrils did not trigger the cytosolic Ca 2ϩ pulse, although the fluorescence drop phase was maintained. Controls done in the presence of 10 mM nickel, a voltage-gated calcium channel blocker, did not influence the HCl fibril effect (data not shown).
The different response of cytosolic Ca 2ϩ levels to stimulation with HCl fibrils and with CSB fibrils suggests that both types of fibrillar species may have significant differences in structure. Indeed, high resolution AFM images of HCl fibrils resemble immature nodular fibrils, short and bent (Fig. 4B), whereas CSB fibrils are clearly of the smooth type, long and straight (Fig. 2, F  and H). Consecutive AFM scans of HCl fibrils show increasing fibril disruption characteristic of weakly structured entities (Fig. 4, B and C), whereas fibrils formed in the presence of CSB are not altered after repeated scanning by the AFM tip (Fig. 4, D and E), suggesting that they have a much more stable architecture. ThT assays reveal that samples containing CSB fibrils have a fluorescence signal severalfold that of HCl fibril samples (Fig.  4F), indicating that the former have a much more ordered ␤-sheet structure. The presence of CSB in HCl fibril-forming conditions also induces a severalfold increase in ThT fluorescence. The high signal obtained reflects the concentration dependence of A␤ fibrillogenesis. Because only a small volume could be added to cell cultures, A␤ 1-42 concentration was very high during the preincubation step (200 M versus 20 M in the standard ThT assay of Fig. 2A). Repeated pipetting of HCl fibrils results in a sample with abundant protofibrils (Fig. 4G). Artificial membranes overlaid with such protofibril-enriched preparations show protuberances consistent with the insertion of protofibrils into the lipid bilayer (Fig. 4, H and I).
A␤  HCl Fibrils Induce Intracellular Tau Aggregation-After 24 h of continued exposure to A␤  HCl fibrils, the distribution of endogenous Tau in SH-SY5Y cells was examined by confocal fluorescence immunocytochemistry. Tubulin controls did not reveal significant changes in the microtubular network (Fig. 5, A and B). In untreated SH-SY5Y cells, Tau is  homogeneously distributed throughout the cytosol (Fig. 5C), whereas after 24 h in the continued presence of A␤ 1-42 HCl fibrils, Tau-containing fibril-like structures can be observed inside the cells (Fig. 5D). Western blot analyses of Tau in cell extracts did not reveal any SDS-resistant aggregated structures (data not shown).
Calcium and Pre-aggregated Tau Trigger Tau Aggregation-To explore if Tau could be aggregated by Ca 2ϩ in vitro we exposed recombinant Tau to a Ca 2ϩ concentration of 1.25 mM (Fig. 6A), that can be achieved locally in vivo in the vicinity of permeabilized membrane regions. In the absence of Ca 2ϩ , Tau shows a tendency to form ϳ200-kDa species that might correspond to a trimer or tetramer of the protein. As previously described (48,49), GAGs such as heparan sulfate and especially CSB promote the formation of larger Tau aggregates that do not enter SDS-polyacrylamide gels. In the presence of 1.25 mM Ca 2ϩ Tau becomes aggregated into large structures excluded from the gels. Mg 2ϩ has no effect on Tau even at the very high concentration of 10 mM. In addition, in the absence of Ca 2ϩ Tau can be aggregated by pre-aggregated Tau (Fig. 6B).
AFM imaging revealed that the large aggregated species formed by Tau incubated in the presence of 1.25 mM Ca 2ϩ were long fibrils (Fig. 7A), similar to those formed in the presence of CSB (50). In the absence of the aggregation triggers/enhancers Ca 2ϩ and CSB, Tau did not form any identifiable structures (data not shown), in agreement with existing data (51). In the vicinity of the fibrils, the surface was covered by a regular arrangement of parallel thin filaments with a mean width of 4.0 Ϯ 0.6 nm and a height of 0.9 Ϯ 0.2 nm (Fig. 7B). These dimensions are consistent with those of ␤-sheets existing in Tau fibrils that associate to form paired helical filaments (52,53). Transmission electron microscope (TEM) analysis permitted a better resolution of fibrils and filaments, whose growth and relative amounts are affected by incubation time and Ca 2ϩ and CSB concentration. Fig. 7C shows a group of small fibrils besides a grid-like expanse of short filaments. Higher resolution images indicate that the grid-forming structures (Fig. 7E) have a mean width of 5.1 Ϯ 0.7 nm. Fig. 7D shows characteristic larger fibrils with an internal structure made of long filaments that assemble into bundles. These fibrils can be several microns in length, and their constituent filaments have a mean thickness of 4.8 Ϯ 0.7 nm (Fig. 7F).
CSB Promotes the Formation of Highly Structured A␤ 1-42 Fibrils-CD analysis indicates that CSB promotes the formation of A␤ 1-42 species enriched in ␤ structure, according to the 9-fold increase in the minimum at 220 nm in the far-UV spectra  ( Fig. 8A). High resolution AFM images show that smooth A␤ 1-42 fibrils are straight rods containing internal parallel structures (Fig. 8B). Cross-section analysis of smooth fibrils (made 45°relative to the perpendicular for a better resolution) reveals 6 protofibril subunits and a fibril height of 11.5 nm (Fig.  8C). This image is consistent with a cylindrical structure composed of 12 subunits, of which 6 can be observed in an upper view such as that provided by AFM scans. AFM height measurements are very accurate, but lateral distances can be greatly distorted by the shape of the cantilever tip. In structures like the fibril shown in Fig. 8B the most reliable width dimensions are those obtained for the internal structures (15). The measured width for the topmost subunit is 3 nm, whereas the apparent width for the other subunits is increasingly magnified by their location closer to the vertical sides of the cylinder.
TEM images of ultrathin sections of CSB fibril pellets contain abundant smooth fibrils cut at different angles. Both transversal (Fig. 8D) and longitudinal sections (Fig. 8E) show that smooth fibrils have an electron dense outer shell surrounding a hollow inner core. The diameter of fibrils in TEM images is ϳ10 nm, slightly below the 11.5 nm measured in liquid medium with the AFM. To investigate the possible presence of CSB in the fibril structure, we prepared fibrils using CSB labeled with 2 nm gold beads bound to sulfate groups. Although some labeling could be detected by TEM in CSB fibrils, it was not significantly different from controls done with gold beads alone or conjugated to sulfated monosaccharides (data not shown).

DISCUSSION
A␤ is cleaved from its precursor protein in the membrane interface, and its cytotoxic effect is likely related to the amyloidlipid interaction (54). AFM imaging has revealed that certain oligomeric A␤ 1-42 species can form associations or rafts that have a high affinity for the amphipathic graphite/water interphase. Because A␤ oligomers are basically hydrophilic and do not settle well on graphite (15), we conclude that A␤ rafts are formed by a previously undescribed hydrophobic oligomeric species. TEM and AFM studies performed with artificial membranes have shown that, upon interaction with the membrane lipids, A␤ in fibrillar form reverts to globular peptide oligomers that associate into disordered domains (55). Thus, raft-forming oligomers could be part of a physiologically relevant pathway leading to the incorporation of soluble A␤ into cell membranes.
Although glucose and CSB restore the viability of cells treated with otherwise toxic A␤ 1-42 concentrations, the clear difference between both carbohydrates in their ability to promote fibrillogenesis suggests distinct mechanisms. In the case of glucose the cytoprotective effect might be partially related to its role as an energy source that the cell metabolism can use to overcome the amyloid insult. 4 Electrophoretic analysis indicates that the promotion of fibril formation mediated by CSB involves protofibrils rather than oligomers. This observation is in agreement with data indicating that oligomers are not an obligatory intermediate in the process of fibril formation, as suggested by the finding that oligomerization is inhibited at concentrations of urea that have no effect in fibril formation (56). Both A␤ oligomers and fibrils have been described to be key agents in the Alzheimer-related cytotoxicity. However, different oligomeric and fibrillar types exist, and carbohydrates seem to promote more innocuous forms of both. A possible mechanism to explain functional differences between structurally similar species could be related to the formation of crosslinked structures by advanced glycation end-products (57), which might lock oligomers in fixed conformations preventing dynamic events otherwise leading to cytotoxicity. CSB, on the other hand, could act by speeding up the pathway toward highly stable smooth fibrils, which likely consist of densely packed ␤-sheet stacks.
Side-by-side association of long protofibrillar rods has been proposed in models derived from high resolution AFM images of soluble A␤ 1-42 oligomers (58). However, a stacked protofilament model for A␤ fibril structure has an important flaw: in such an assembly, A␤ fibrils could experience an unlimited growth in thickness by lateral addition of more protofibrils or protofilaments. Instead, A␤ fibrils have a well-defined cross- section of ϳ11.5 nm (15), and any higher order structures result from the association of fibrils, not from individual fibril enlargement. A model that can explain this characteristic of A␤ fibrils is the wedged cylinder structure (Fig. 9A). In A␤ 1-42 , residues 1-17 are disordered, whereas residues 18 -42 form a ␤-strand-turn-␤-strand motif formed by residues 18 -26 (␤ 1 ) and 31-42 (␤ 2 ) (59). A␤  stacking ultimately leads to the formation of hydrophobic ␤-sheets 5.5 nm wide and 1.5 nm high termed protofilaments, which can intertwine to form helical structures found at the core of the nodular type of fibrils (15). Nodular fibrils likely undergo internal rearrangements to yield smooth fibrils, the final stage in A␤  fibrillogenesis. This internal remodeling seems to consist of switching from a helical to a completely parallel protofilament alignment. We propose that this change results in the transition from a rectangular to a wedged conformation, where the loose 1-17 residues form the narrow end. In this model, wedged protofilaments ϳ3 ϫ 5.5 nm fill the same volume as rectangular-section protofilaments 1.5 ϫ 5.5 nm. Finally, wedged protofilament association could assemble a cylindrical rod containing 12 protofilament subunits. Tightly packed ␤-sheets would form the electron-dense outer shell observed in TEM images, whereas the unstructured 1-17 region would occupy the relatively hollow core. A cylindrical structure ϳ11.5 nm in diameter has a circumference of ϳ36 nm that can accommodate precisely 12 wedged subunits with an external arc of ϳ3 nm each. In this model, the hydrophobic interactions driving A␤ self-association will promote the formation of a structure energetically more stable than a stacked protofilament assembly, where a much larger hydrophobic surface remains exposed to the aqueous external milieu. Once the cylindrical structure is closed it cannot grow further in width, thus precluding the formation of thicker fibrils. A multimeric array of protofilaments ϳ3 nm across organized in a tubular configuration was already described for A␤ 1-40 fibrils with an average diameter of ϳ7 nm (60). A hollow core has been also deduced from electron micrograph image reconstruction of transthyretin fibrils (61), indicating that this can be a general feature of amyloids.
A␤ has a cluster of basic amino acids at the N terminus that are considered critical for GAG binding (34) (Fig. 9B). In agreement with this view we have shown that sulfated polymeric GAGs are efficient promoters of A␤ fibril formation. Other studies (62) have shown that CSB-derived monosaccharides represent the minimal GAG subunit required for A␤ binding and that lateral aggregation between A␤ fibrils or the transition of protofilaments into mature amyloid fibrils requires at least a sulfated GAG disaccharide, with the disulfated derivative being the most effective. These results suggest that subtle changes in the GAG backbone and distribution of sulfation have significant effects on the ability of chondroitin sulfate to organize A␤ into amyloid fibrils. However, the observation that, unlike A␤ 1-42 , A␤ 1-40 does not increase significantly its tendency to aggregate in the presence of CSB (24) indicates that the main factor behind amyloid fibril formation is the intrinsic peptide propensity toward fibrillogenesis. Sulfated polysaccharides could act as a template, interacting with the 1-17 tail of A␤  in nascent protofibrils (Fig. 9B), stabilizing them and facilitating their growth. Our failure to detect CSB in the final structure suggests that the GAG is shed from the fibrils during their growth, but we cannot rule out a small relative CSB content below the detection limit of current methods. Other techniques such as solid-state NMR or x-ray diffraction analysis might provide answers, although the insolubility and polydispersity of A␤ fibril preparations are serious obstacles.
A wedged cylinder is consistent with the majority of data available for protofibril and protofilament structure and explains why protofibrils 5.5 nm wide give rise to fibrils precisely twice that width (15,63). This model is compatible with the existence of certain oligomeric A␤ species like the so-called doughnuts (11,64) and A␤*56, a recently described A␤ assembly in vivo (65). A␤*56 is an apparent dodecamer detected in the brains of a transgenic mouse line expressing a human A␤ precursor protein variant linked to AD. Whether A␤*56 represents a stable assembly in vivo is currently unknown. The cross-section of the wedged cylinder model contains an A␤ dodecamer, which might exist as an independent soluble association of 12 A␤ subunits arranged in a circle (Fig. 9A). Doughnuts and A␤*56 could actually be one and the same entity: a dodecamer whose center would contain the dangling 1-17 regions that confer it a less structured conformation (the hole in the doughnut). Our high resolution AFM images on HOPG of raft-forming globular structures have dimensions that closely fit those predicted for such disc-shaped dodecamers that can stack their hydrophobic faces to form cylinders. A ϳ55-kDa species consistently appears in Western blots transferred from gels run in the presence of SDS, indicating the existence of a very stable species in solution that could correspond to A␤*56.
Resistance of different cell types to A␤-induced toxicity appeared in part related to the ability of cells to counteract alterations of intracellular free Ca 2ϩ , suggesting that membrane destabilization and the subsequent early derangement of ion balance is a key event leading to amyloid-induced cell death (66). Several mechanisms have been proposed to account for the Ca 2ϩ -mobilizing actions of amyloids in their oligomeric and fibrillar aggregation states. These include a direct interaction with membrane components to destabilize the membrane structure (67), insertion into the membrane to form a cationconducting pore (10,68,69), activation of cell surface receptors coupled to Ca 2ϩ influx (70,71), and oxidative stress leading to dysregulation of mitochondrial Ca 2ϩ homeostasis (72).
Exposure of neuroblastoma cells to various aggregative states of A␤ 1-42 elicits different Ca 2ϩ responses. Unstructured fibrilinduced cytosolic Ca 2ϩ elevation requires the presence of both extracellular and intracellular Ca 2ϩ . A second effect of unstructured fibrils is a rapid leakage of fluorescence from the cells that suggests a massive plasma membrane permeabilization. In agreement with this observation, A␤  has been described to induce a strong membrane destabilization in giant unilamellar vesicles (73). Our observation that protofibrils derived from weakly structured fibrils can insert into lipid bilayers indicates a role for protofibrils in the membrane perturbations induced by A␤. The presence of CSB in unstructured fibril forming conditions might accelerate the incorporation of protofibrils into stable fibrils, an effect that can contribute to the reduction of cytosolic calcium variations and the recovery of cell viability. Resting cytosolic free Ca 2ϩ is normally maintained at very low levels despite enormous transmembrane concentration gradients so that even a brief increase might disrupt important cell processes. Indeed, Ca 2ϩ dyshomeostasis is implicated in both necrotic and apoptotic cell death (74), and amyloid-induced Ca 2ϩ changes are directly correlated with cell viability (75). An intriguing possibility in neurological diseases is that an enhanced plasma membrane ion conductance may cause a sustained depolarization and increased electrical excitability, leading to excitotoxic injury (5). These considerations are in agreement with the hypothesis that the toxic effects of A␤ on neurons are various, both in their mechanism and possibly also in their temporal component (76). For instance, A␤ monomers, oligomers, and fibrils are cytotoxic after a few hours according to cell viability assays (9), whereas soluble oligomers have been also associated to synaptic dysfunction in AD brain (77), which suggests a longer time of action.
AD pathophysiology has been related to sustained increased intracellular Ca 2ϩ levels (78). Here, we propose that pernicious, long term cumulative effects might result from brief local Ca 2ϩ influxes that over time perturb intracellular mechanisms, thus leading to cell collapse. In vitro, Ca 2ϩ promotes the formation of Tau filaments with a structure very similar to that observed in brain-derived paired helical filaments (79), suggesting that increments in the ion concentration may trigger the deposition of nucleating Tau aggregates in vivo (80). If, as we have shown, aggregated Tau can capture soluble Tau in the absence of Ca 2ϩ , such small Tau nuclei can act as seeds for the growth of paired helical filaments (52), even if Ca 2ϩ levels have been restored back to normal conditions. The results presented here are consistent with data suggesting that A␤ 1-42 induces a rise in resting Ca 2ϩ levels that is associated with an ER Ca 2ϩ leak ultimately leading to the pathological accumulation of Tau (81). Ca 2ϩactivated proteases such as calpain have been related to Tau neurofibrillary pathology (82).
Our data suggest that sulfated GAG-mediated neuroprotection is due to the sequestering of A␤. Thus, GAGs of the extracellular matrix and cell surface could act as key modulators of amyloid homeostasis influencing which tissues or cell types are harmfully affected by the process of amyloidogenesis. Interestingly, chondroitin sulfate content has been shown to be inversely correlated with the amount of hyperphosphorylated Tau in cortical areas of patients with AD (40), a good marker for A␤-induced neuronal dysfunction (1,4). Weakly structured A␤ fibrils could be leaking neurotoxic soluble peptide into the surrounding tissue, thus functioning as reservoirs of the bioactive oligomers (Fig. 9C). This observation conciliates apparently contradictory data that point at both fibrils and soluble oligomers as the culprits in AD pathogenesis. On the other hand, our results support the hypothesis that fibrillar deposits formed in the presence of sulfated GAGs are very stable and constitute thermodynamic sinks from where the incorporated peptides cannot easily escape.