Glycosylphosphatidylinositol Anchoring Directs the Assembly of Sup35NM Protein into Non-fibrillar, Membrane-bound Aggregates*

Background: Sup35NM is a soluble protein that when misfolded forms amyloid fibril aggregates. Results: When tethered to membranes via a lipid anchor, Sup35NM aggregates adopt a non-fibrillar, membrane-bound structure. Conclusion: Lipid anchor-directed membrane association prevents assembly into fibrils. Significance: This may explain differences between prion diseases compared with related protein aggregation diseases because only prion diseases involve aggregation of a lipid-anchored protein. In prion-infected hosts, PrPSc usually accumulates as non-fibrillar, membrane-bound aggregates. Glycosylphosphatidylinositol (GPI) anchor-directed membrane association appears to be an important factor controlling the biophysical properties of PrPSc aggregates. To determine whether GPI anchoring can similarly modulate the assembly of other amyloid-forming proteins, neuronal cell lines were generated that expressed a GPI-anchored form of a model amyloidogenic protein, the NM domain of the yeast prion protein Sup35 (Sup35GPI). We recently reported that GPI anchoring facilitated the induction of Sup35GPI prions in this system. Here, we report the ultrastructural characterization of self-propagating Sup35GPI aggregates of either spontaneous or induced origin. Like membrane-bound PrPSc, Sup35GPI aggregates resisted release from cells treated with phosphatidylinositol-specific phospholipase C. Sup35GPI aggregates of spontaneous origin were detergent-insoluble, protease-resistant, and self-propagating, in a manner similar to that reported for recombinant Sup35NM amyloid fibrils and induced Sup35GPI aggregates. However, GPI-anchored Sup35 aggregates were not stained with amyloid-binding dyes, such as Thioflavin T. This was consistent with ultrastructural analyses, which showed that the aggregates corresponded to dense cell surface accumulations of membrane vesicle-like structures and were not fibrillar. Together, these results showed that GPI anchoring directs the assembly of Sup35NM into non-fibrillar, membrane-bound aggregates that resemble PrPSc, raising the possibility that GPI anchor-dependent modulation of protein aggregation might occur with other amyloidogenic proteins. This may contribute to differences in pathogenesis and pathology between prion diseases, which uniquely involve aggregation of a GPI-anchored protein, versus other protein misfolding diseases.

Prion diseases, also known as transmissible spongiform encephalopathies (TSEs), 3 are a group of protein misfolding diseases characterized by the accumulation of abnormal, aggregated forms of the prion protein, PrP. In its native, cellular form (PrPC), the prion protein is mostly ␣-helical and monomeric. However, it is able to undergo conformational conversion and polymerize, either spontaneously as a result of mutations or following infection, to a ␤-sheet rich structure termed PrPSc (1)(2)(3). Evidence points toward PrPSc, alone or associated with some co-factor, as the infectious agent of TSEs (reviewed in Ref. 4). PrPSc is thought to propagate by incorporating host PrPC into PrPSc aggregates by a seeded polymerization mechanism (5)(6)(7). Some other neurodegenerative diseases associated with protein misfolding are thought to spread within tissues by a similar "prion-like" seeded polymerization mechanism (8 -11). At present, TSEs are distinguished from other protein misfolding diseases, such as Alzheimer disease and various systemic and localized amyloidoses, by their ability to transmit between individuals and sometimes even between species (12). However, the relative transmissibility of other protein misfolding disorders is currently under intense investigation.
Another unique facet of TSEs among protein misfolding diseases is that the misfolded protein contains a glycosylphosphatidylinositol (GPI) lipid anchor (13). In prion-infected tissues, PrPSc aggregates may be membrane-associated, either on the cell surface or within intracellular vesicles, or may be deposited in the form of amyloid plaques in the extracellular space (11,14,15). "Amyloid" refers to an ordered fibrillar structure in which hydrogen-bonded ␤-sheets are oriented perpendicular to the fibril axis, forming an architecture known as cross-␤ (16). The cross-␤ arrangement yields characteristic reflections by fiber diffraction and typically stains with the amyloid-specific dyes Thioflavin T or S and Congo Red (17)(18)(19)(20). In most instances, PrPSc is membrane-bound as granular, diffuse, nonfibrillar structures, with amyloid plaques only manifesting in certain TSE diseases, such as kuru, Gerstmann-Straussler-Scheinker disease, variant Creutzfeldt-Jakob disease, and a minority of sporadic Creutzfeldt-Jakob disease cases (14,(21)(22)(23)(24)(25). The effect of GPI anchoring of PrP on TSE pathogenesis and infectivity has been investigated using a transgenic mouse that expresses PrPC lacking a GPI anchor (26,27). In these mice, dense amyloid PrPSc deposits form following infection. This is accompanied by altered clinical disease, prolonged incubation period, and poor susceptibility to peripheral routes of infection, suggesting that the GPI anchor strongly influences TSE pathogenesis (27,28).
Although much is known of the generic structure of amyloid (29,30) and the structure of PrP amyloid fibrils (1,(31)(32)(33)(34)(35)(36), detailed structural knowledge of non-fibrillar, membranebound PrPSc aggregates is limited. Prion rods have been described with the tinctorial properties of amyloid that assembled during purification using detergent extraction and proteinase K digestion of brain-derived PrPSc (37)(38)(39). When the prion rods were solubilized with a mixture of phospholipid and detergent, liposomes formed that contained 10 -100-fold higher infectivity but no fibrillar material as analyzed by immunoelectron microscopy (40,41). These experiments, like those using the anchorless PrPC mice, show that removal of PrPSc from the cell membrane allows polymerization into amyloid, that there is a relationship between infectivity and membrane association, and that membrane-associated aggregates differ morphologically from amyloid fibrils.
Although extraction from membranes results in dramatic morphological changes to PrPSc aggregates, there is evidence to suggest there are structural similarities between membranebound and amyloid forms of PrPSc. For example, analysis of scrapie-infected tissue slices using synchrotron infrared radiation showed that areas where PrPSc was detected by immunohistochemistry corresponded to deposits rich in ␤-sheet structure (42). However, diffuse, membrane-bound forms of PrPSc do not bind Thioflavin S, providing evidence that they are not amyloid (26). These data are consistent with imaging of PrPSc in infected brain tissue and cells by immunogold electron microscopy, where fibrillar forms of PrPSc were only observed for deposits that were not membrane-associated (e.g. see Refs. 43 and 44;reviewed in Ref. 15). This technique has also revealed that membrane-bound PrPSc gives rise to unusual membrane lesions, in particular plasma membrane invaginations on neurons and astrocytes (15,45,46). No similar membrane lesions were observed in the GPI anchorless PrPC mouse model, suggesting that only GPI-anchored PrPSc is able to induce such pathology (26,27).
Given the influence of GPI anchoring of PrP on PrPSc aggregation and pathogenesis in TSE disease, we have asked whether GPI anchoring might similarly modify the aggregation and biology of other amyloidogenic proteins. We initiated these investigations using a model system consisting of a GPI-anchored form of the highly charged, glutamine-rich N-terminal and middle (NM) prion domain from the yeast prion protein Sup35p (referred to here as Sup35 GPI ), stably expressed in N2a cells (47). When expressed in Saccharomyces cerevisiae in its native, soluble form, the function of Sup35p is as a translation termination factor (48). However, in the prion state, [PSI ϩ ], Sup35p adopts an alternative, amyloid conformation that is able to self-propagate through a templating mechanism and is inherited by daughter cells (49,50). Sup35p is not bound to the membrane of yeast cells and has a high propensity to form amyloid in vivo and in vitro (51)(52)(53)(54)(55). There is evidence that other yeast prion proteins (e.g. Ure2p) form amyloid in the yeast cytosol (56). In previous studies, we and others reported that Sup35NM is able to propagate as a prion in mammalian cells (47,57,58) and that GPI anchoring facilitates aggregate propagation between N2a cells, resembling mammalian prion behavior (47). In the present work, we go on to characterize the ultrastructural and biochemical features of GPI-anchored Sup35NM aggregates. The results show that GPI anchoring to the cell membrane directs the formation of aggregated, non-fibrillar forms of Sup35NM. By placing a GPI anchor onto a highly amyloidogenic protein that would otherwise fibrillize into amyloid, we have altered its biophysical properties to resemble those of PrPSc aggregates associated with TSE, highlighting the critical role of membrane association in modulating the assembly and ultrastructure of aggregates.
Generation of N2a Cell Clones Expressing Sup35 Constructs-The procedure for construction and culture of cell lines stably expressing GFP-and mCherry (mC)-tagged proteins is described elsewhere (47). Stably transfected cells were subjected to multiple rounds of FACS sorting to select for high expressing cell populations. During the course of Geneticin selection and FACS sorting, aggregates of Sup35-GFP GPI appeared in the culture, creating a mix of cells that were positive or negative for aggregates. FACS sorting enriched the population for aggregate-positive cells, although aggregate-negative cells were still present (data not shown). Single cell cloning of these mixed cultures led to the isolation of stable cell lines that remained aggregate-free (Sup35-GFP GPI -Sol, for soluble) or aggregate-positive (Sup35-GFP GPI -Agg) over extended passage. When treated with preformed Sup35 aggregates, Sup35-GFP GPI -Sol cells support persistent propagation of Sup35-GFP GPI aggregates as shown elsewhere (47). FACS-sorted Sup35-mC GPI cultures contained a very high percentage of aggregate-positive cells without single cell cloning.
Fluorescence Microscopy-Wide field fluorescence microscopy images were acquired as described elsewhere (47) using 10ϫ Plan Fluor numerical aperture 0.3 or 40ϫ S Plan Fluor numerical aperture 0.6 objectives. Confocal images were obtained on a Nikon LiveScan confocal microscope as described elsewhere (47). Confocal images were deconvolved using Huygens (Scientific Volume Imaging) or AutoQuant (Media Cybernetics) software. Images were analyzed using Imaris and NIS-Elements software.
Detergent Insolubility, Filter Trap, and Chymotrypsin Resistance Assays-Assays for detergent insolubility and resistance to chymotrypsin digestion were performed as described elsewhere (47) with the exception that varying mass ratios of chymotrypsin/total protein were also tested where indicated.
Phospholipase Release of Cell Surface Sup35-GFP GPI -Cells were plated at 10% confluence and incubated for 2 days. To remove cell surface GPI-anchored proteins, the cells were incubated for 60 min at 37°C in conditioned medium either without or with 0.22 units/ml phosphatidylinositol-specific phospholipase C (PIPLC; MP Biomedicals). After chilling for 30 min at 4°C, the cells were stained for 20 min at 4°C with 10 g/ml Alexa Fluor 594-labeled wheat germ agglutinin (WGA) in conditioned medium to label the cell surface. The cells were then fixed at room temperature with 4% PFA, 0.05% glutaraldehyde, 5% sucrose in PBS for 25 min. After two rinses with PBS, the cells were imaged in PBS.
In Situ Seeding of Sup35-mCherry Aggregation-For in situ seeding by fixed cells, cells were fixed for 10 min at room temperature with 4% paraformaldehyde, 0.05% glutaraldehyde in PBS. After washing with PBS, residual fixative was quenched by incubation at 37°C for 24 -48 h with Opti-MEM medium supplemented with 10% fetal bovine serum (FBS). Quenching medium was then replaced with either fresh culture medium or 3-4-day conditioned medium from cells stably expressing anchorless mC or anchorless Sup35-mCherry (Sup35-mC). The medium was precleared by centrifugation at 3000 ϫ g for 5 min prior to the addition to fixed cells. After incubation for 5 days in the dark at 4°C, the samples were washed with PBS and imaged by confocal microscopy.
For in situ seeding by live cells, the medium of cells cultured for 3 days in Opti-MEM was replaced with fresh conditioned medium from cells expressing Sup35-mC. The conditioned medium was filtered through a 0.2-m pore size filter prior to addition to live cells. After incubation for 8 -10 h at 37°C, the cultures were washed twice with Opti-MEM and then imaged by confocal microscopy. After imaging, the cells were washed twice with Hanks' balanced salt solution, lysed with 1% sarkosyl in Tris-buffered saline (pH 7.5), and processed for chymotrypsin digestion as described elsewhere (47). Filtered conditioned Sup35-mC medium alone was also processed in parallel for comparison.
Cell-free Assay for Aggregation of Sup35-mC in Culture Supernatants-Precleared culture supernatants from cells expressing mC or Sup35-mC were prepared as described above. Reactions were initiated by mixing culture supernatants with microsomes prepared as described elsewhere (59) and normalized for total protein content by bicinchoninic acid assay (Pierce) in 1% SDS. We refer to the membrane fraction as "microsomal" for convenience, but as explained elsewhere (59), this fraction represents a crude total membrane preparation.
Total reaction volume was 200 l. After incubation for at least 5 days at 4°C in the dark, the samples were adjusted to 4% SDS, incubated for 30 min at room temperature, and then analyzed by filter trap assay and immunoblot.
Thioflavin T Staining-Cells stably expressing Sup35-mC GPI or mC GPI were plated at 10% confluence and incubated for 3 days. Cultures were then chilled at 4°C for 15 min before the addition of recombinant Sup35NM amyloid fibrils and further incubation for 140 min at 4°C. Recombinant Sup35NM fibrils were prepared with gentle rotation at 4 or 37°C as described elsewhere (47). The cells were then fixed with 4% paraformaldehyde, 4% sucrose in PBS for 30 min at room temperature. After rinsing with PBS, the residual fixative was quenched by incubation for 20 min at room temperature with 50 mM glycine in PBS. Next, the cells were stained for 30 min with freshly prepared 10 M Thioflavin T (ThT) in 50 mM glycine (pH 8.5), washed once for 5 min in PBS, and mounted in Prolong Gold for imaging. ThT staining was visualized by confocal microscopy using 442-nm excitation and a 480BP40 emission filter.
Serial Passage of Cells with Congo Red-Cells were serially passaged at 10-or 20-fold dilution each pass in Opti-MEM ϩ 10% FBS as described elsewhere (47). For treatment with Congo Red, the cells were continuously cultured in medium supplemented with Congo Red at the indicated final concentrations from a 2 mM stock in 90% PBS, 10% ethanol. Similar results were obtained using Congo Red from two different suppliers.
Scanning and Transmission Electron Microscopy of N2a Cells-Cells were cultured on Thermanox coverslips (Thermo Scientific) for transmission electron microscopy (TEM) or on clean silicon chips for scanning electron microscopy (SEM). Cells were fixed for 30 min at room temperature with 4% paraformaldehyde, 0.05% glutaraldehyde, 5% sucrose in HEPES-buffered saline (pH 7.4). The samples were then rinsed, and free aldehydes were quenched for 5 min by incubation with 50 mM glycine in phosphate-buffered saline (PBS). After aldehyde quenching, the samples were processed for either immunoperoxidase or immunogold staining as described below.
For immunoperoxidase staining, samples were incubated with primary antibody (anti-GFP) for 1 h at room temperature, washed twice with PBS, and then incubated for 1 h at room temperature with goat anti-mouse IgG F(ab) 2 -peroxidase-conjugated secondary antibody. After three rinses with PBS, the peroxidase-labeled samples were subsequently processed for TEM as described previously (60).
For immunogold staining, the samples were first blocked with goat blocking agent (Electron Microscopy Sciences) for 30 min at room temperature or overnight at 4°C. Blocking solution was then replaced with primary antibody (anti-GFP or 3F4) diluted in 2% BSA, 0.1% Aurion BSA-c TM (Electron Microscopy Sciences) in HEPES-buffered saline (HBS-BSAc). After incubation for 1 h at room temperature, the samples were washed three times for 5 min each in HEPES-buffered saline ϩ 0.1% Aurion BSA-c. The samples were then treated with anti-mouse Alexa Fluor 594-FluoroNanogold conjugate diluted in HBS-BSAc for 1 h at room temperature followed by four 5-min washes in HEPES-buffered saline ϩ 0.1% Aurion BSAc. After the final wash, samples were fixed with Karnovsky's fixative. Samples were then subjected to gold enhancement per the manufacturer's recommendations (Goldenhance-EM, Nanoprobes). The SEM samples were processed as described elsewhere (61). The TEM samples were processed in a Lynx tissue processor (Electron Microscopy Supplies) as described elsewhere (27). The durations of TEM processing steps were modified as follows: postfixation for 30 min, washing and dehydration for 15 min each, and respective epoxy resin infiltration steps with 50, 75, and 100% resin for 1, 4, and 8 -20 h each. In TEM images, a small population of electron dense spots was observed at high magnification, often associated with membranes, comprising particles much smaller in size than large enhanced gold particles that were present only on the cell surface. The small particles correspond to background from the gold enhancement procedure to detect nanogold, which involves using a metal-based solution to increase the size of the gold particles for easier observation. This background is acknowledged by the manufacturer (Goldenhance-EM, Nanoprobes). Cell-associated metal ions are probably able to react with this solution and thus are visible using TEM. Our interpretations are further supported by results from immuno-EM experiments using immunoperoxidase-based labeling instead of FluoroNanogold, which showed specific labeling of Sup35 only and no intracellular labeling (Fig. 12), and the presence of the small background particles in control samples where the FluoroNanogold secondary antibody was omitted. In addition, all immunolabeling was performed on non-permeabilized cells to specifically label cell surface GFP-tagged proteins.
Preparation and TEM of Biotinylated Sup35NM Fibrils-Recombinant Sup35NM fibrils were prepared with gentle rotation at 37°C as described elsewhere (47). The Sup35NM expression construct contains a C-terminal cysteine residue and His tag for purification (62). The fibrils were collected by centrifugation at 100,000 ϫ g for 30 min and rinsed with PBS prior to labeling. Biotinylation of pelleted fibrils was performed using a 10-fold molar excess of EZ-Link BMCC-biotin and incubation overnight at room temperature. Remaining free thiols were blocked using iodoacetamide. The fibrils were then collected by centrifugation as above, washed with PBS, and finally resuspended in PBS. Biotin labeling was confirmed by immunoblot analysis of replicate blots with anti-Sup35 M rabbit polyclonal antibody and NeutrAvidin-HRP (to detect biotin) as described previously (47). Biotinylated fibrils were also examined by TEM after staining with methylamine tungstate.
Cells were plated on 13-mm diameter Thermanox coverslips for TEM visualization of cell-associated fibrils. Cells were first chilled to 4°C prior to the addition of ϳ800 ng of biotinylated Sup35NM fibrils or PBS buffer. After incubation at 4°C for 80 min, the cells were fixed with 4% paraformaldehyde, 0.02% glutaraldehyde, 5% sucrose in PBS. Free aldehydes were quenched with 50 mM glycine in PBS prior to blocking for 30 min with 2% BSA in PBS. The samples were then incubated with a 1:20 dilution of Alexa Fluor 594-streptavidin-FluoroNanogold in 2% BSA, 0.1% Aurion BSA-c in PBS. Additional negative control samples were incubated with streptavidin-FluoroNanogold that was pretreated for 30 min with 10 mM D-biotin to block biotin-binding sites. Following several washes in PBS ϩ 0.1% BSA-c, the samples were fixed with Karnovsky's fixative and subjected to gold enhancement per the manufacturer's recommendations, including a procedure developed by Wanzhong He and Pamela Bjorkman to inhibit the background enhancement by cellular metal ions (see the Nanoprobes Web site) that was discussed above. We used a modification of the He and Bjorkman procedure with one part solution A and 5 parts each of solution B, solution C, and 0.05 M sodium phosphate, 0.1 M sodium chloride (pH 5.5). The samples were further processed for TEM imaging as described above with the exception that microwave-assisted processing was used for all steps.

Generation of Cell Lines Stably Propagating Sup35-GFP GPI
Aggregates-In order to characterize the aggregates formed from GPI-anchored Sup35NM, mouse N2a neuroblastoma cells were transfected with the constructs shown in Fig. 1. These comprise fusions of GPI-anchored GFP or mCherry to the C terminus of the amyloidogenic Sup35NM domain, which we refer to as Sup35-GFP GPI and Sup35-mC GPI , respectively. GFP with a GPI anchor signal was used as a control construct (GFP GPI ). Cells stably expressing GPI anchorless mCherry (mC) or GPI anchorless Sup35NM-mCherry (Sup35-mC) were also developed (47) to provide a source of substrate for in vitro conversion reactions. In transiently transfected cells, the distribution of Sup35-GFP GPI was indistinguishable from GFP GPI , showing a uniformly distributed cell surface population and an intracellular, largely perinuclear population, both typical of GPI-anchored proteins (not shown). After 24 days of culture with Geneticin selection to select for stably transfected cells and multiple rounds of FACS sorting to select for cells expressing high levels of GFP-tagged proteins, we observed that several cells in the Sup35-GFP GPI cultures exhibited bright, punctate cell surface aggregates of fluorescent protein (data not shown). These aggregates had arisen without application of exogenous preformed aggregates of Sup35NM. The amyloidogenic NM domain of the yeast prion Sup35 was expressed as a fusion to a GPIanchored fluorescent protein (GFP or mCherry) in mouse N2a neuroblastoma cells. GPI anchorless constructs were generated by insertion of a stop codon immediately prior to the GPI anchor addition sequence. An N-terminal signal peptide (SP) sequence was included in all constructs, which directs the proteins to the secretory pathway in mammalian cells. Shaded segment, Myc (GFP constructs) or HA (mCherry constructs) antibody epitope tag.
Limited dilution cloning of the Sup35-GFP GPI culture led to the isolation of cloned lines that either stably propagated high levels of Sup35-GFP GPI aggregates of spontaneous origin (Sup35-GFP GPI -Agg) or were devoid of aggregates (Sup35-GFP GPI -Sol) (Fig. 2, B and C). Sup35-GFP GPI -Agg cultures were highly enriched for aggregate-positive cells, but cells without visible aggregates were also present. Sup35-GFP GPI -Sol cells were not resistant to aggregate formation because self-propagating Sup35-GFP GPI aggregates could be induced by treating Sup35-GFP GPI -Sol cells with preformed recombinant Sup35NM (rSup35NM) fibrils (47) (and the present study). Regardless of the origin (spontaneous or induced), the Sup35-GFP GPI aggregates exhibited a similar morphology by fluorescence microscopy. The aggregates were localized to the cell surface and often, but not always, accumulated at points of intimate intercellular contact. Two classes of aggregate morphology were observed: irregular and contoured. Irregular aggregates formed amorphous shapes as indicated by the arrowhead in Fig. 2. Contoured aggregates accumulated with a shape that followed the contours of the cell surface ( Fig. 2, arrow). By contrast, the distribution in the non-aggregated Sup35-GFP GPI -Sol cells (Fig. 2B) resembled that seen in the GFP GPI control ( Fig. 2A).
Sup35-GFP GPI Aggregates Are Detergent-insoluble and Protease-resistant-The stability of prion aggregates of Sup35NM produced in yeast and in vitro may be characterized by insolubility in detergents and partial protease resistance (50,63,64). To determine the stability of Sup35-GFP GPI aggregates, we measured their solubility after treatment with 1% sarkosyl. As observed previously for Sup35NM aggregates produced in [PSI ϩ ] yeast cells and induced Sup35-GFP GPI aggregates (47), Sup35-GFP GPI -Agg aggregates resisted solubilization in sarkosyl using a sedimentation assay, revealed as a band at around 80 kDa (Fig. 3A, arrow). The spontaneous aggregates were highly stable because they also resisted solubilization in a filter trap assay that employs a strong denaturing detergent (4% SDS) (Fig.  3B). Non-aggregated Sup35-GFP GPI was completely soluble in 4% SDS and able to pass through the membrane; the aggregated form was retained and detected on the membrane. Furthermore, chymotrypsin digestion of cell lysates expressing Sup35-GFP GPI -Sol resulted in the disappearance of an 80-kDa anti-N domain immunoreactive band with increasing enzyme concentrations (Fig. 4). There was a clear size shift after the addition of  very low enzyme concentrations as the protein was reduced in size due to cleavage; at higher concentrations, non-aggregated Sup35-GFP GPI was completely degraded. Conversely, the aggregated form produced by Sup35-GFP GPI -Agg cells acquired chymotrypsin resistance, with only very high concentrations able to fully eliminate N domain immunoreactivity, similar to recombinant Sup35NM amyloid fibrils, suggesting a conformational change in the N domain of Sup35-GFP GPI . The absence of any size shift for Sup35-GFP GPI from Sup35-GFP GPI -Agg cells after the addition of chymotrypsin is a further indication that the aggregated protein is in a protease-resistant conformation. These data show that spontaneous Sup35-GFP GPI aggregates possess detergent insolubility and protease resistance profiles similar to those of other Sup35NM aggregates.
Immunoblots in Figs. 3 and 4 suggested that Sup35-GFP GPI expression in soluble cells was slightly lower than in aggregatepositive cells (Fig. 3, lanes 2 and 4, and Fig. 4, lanes 1 and 7), despite both clones being isolated from the same parental culture that had undergone multiple rounds of FACS sorting. Quantitative analysis of these immunoblots and others (data not shown) indicated that Sup35-GFP GPI expression in Sup35-GFP GPI -Sol cells was about 56 Ϯ 13% (n ϭ 4) lower than in Sup35-GFP GPI -Agg cells. To investigate whether the higher Sup35-GFP GPI expression level was necessary for spontaneous formation of aggregates, we analyzed an independent Sup35-GFP GPI -Agg clone in which aggregates had also arisen spontaneously (clone D8). Aggregated Sup35-GFP GPI was abundant in Sup35-GFP GPI -Agg (clone D8) cells, as assayed by fluorescence microscopy (Fig. 5, compare A with B), detergent insolubility (data not shown), and chymotrypsin resistance (data not shown). Immunoblotting (Fig. 5C) with anti-Sup35 N domain antibody revealed that the expression level of GFP-tagged protein in Sup35-GFP GPI -Agg (clone D8) cells was 52 Ϯ 3.9% lower than in Sup35-GFP GPI -Sol cells. These data show that high expression of Sup35-GFP GPI was not required for spontaneous Sup35-GFP GPI aggregation in N2a cells.
Sup35-GFP GPI Aggregates Seed Aggregation of Sup35-mC-To establish that the aggregates in Sup35-GFP GPI -Agg cells have self-propagating activity like infectious recombinant Sup35NM fibrils, we used several different types of seeding assays. The substrate for the assays was anchorless Sup35-mC as released into conditioned medium of stably transfected cells ( Fig. 6A). Medium from cells expressing anchorless mC protein lacking the Sup35NM domain served as an additional control (Fig. 6A). We first conducted an in situ seeding assay analogous to one used previously (47) by incubating fixed Sup35-GFP GPI -Agg cells and control cell lines with conditioned medium from cells expressing Sup35-mC, followed by visualization of new Sup35-mC aggregate formation by confocal microscopy. As shown in Fig. 6B, bright accumulations of Sup35-mC (arrowheads) were detected on Sup35-GFP GPI aggregates (arrows) present on Sup35-GFP GPI -Agg cells (Fig. 6B, E-G). Faint diffuse accumulation of Sup35-mC was observed intracellularly, but this was also common to Sup35-GFP GPI -Sol and GFP GPI control cells, indicating that it was independent of Sup35 expression in the host cells and probably due to nonspecific binding of Sup35-mC. In an additional control experiment, no mCherry protein was detected on Sup35-GFP GPI aggregates when Sup35-GFP GPI -Agg cells were incubated with anchorless mC control culture supernatant (Fig. 6B, A-D), showing that seeded aggregation of mCherry-tagged protein required the Sup35NM domain.
To provide one form of biochemical evidence of seeded Sup35-mC aggregation, we conducted a cell-free aggregation assay using culture supernatant substrate mixed with crude total membranes (microsomes) from Sup35-GFP GPI -Agg cells as a source of seeds where new aggregate formation was detected by filter trap assay in the presence of 4% SDS using an mCherry-binding antibody, anti-RFP (Fig. 6C). Only microsomes from Sup35-GFP GPI -Agg cells induced the formation of SDS-insoluble Sup35-mC (Fig. 6C, anti-RFP panels). Consistent with the confocal microscopy data, no insoluble mCherry was induced when Sup35-GFP GPI -Agg microsomes were incubated with mC control culture sups. This provided a separate line of evidence establishing the Sup35NM domain-dependent specificity of interactions between Sup35-GFP GPI aggregates and Sup35-mC.
To prove that Sup35-mC assembly on Sup35-GFP GPI aggregates led to conformational conversion of Sup35-mC, we devel-  1-6) and Sup35-GFP GPI -Agg (lanes 7-12) cell lysates were digested with increasing concentrations of chymotrypsin (Ct) and analyzed by immunoblot, probing with anti-Sup35 N domain antibody. Arrow, Sup35-GFP GPI . Chymotrypsin-resistant material was only observed in Sup35-GFP GPI -Agg cell lysates, suggesting a conformational change in Sup35NM.  . Aggregates from Sup35-GFP GPI -Agg cells seed formation of GPI anchorless Sup35-mC aggregates. A, stably transfected cells expressing anchorless forms of mC or Sup35-mC were assayed by immunoblot for expression of the respective proteins. Cell lysates were normalized for total protein and probed with anti-HA or anti-Sup35 M domain antibodies. Culture supernatants were assayed for mCherry-tagged protein by immunoprecipitation. High levels of anchorless mC and Sup35-mC were released to culture supernatants. B, in situ seeding by fixed Sup35-GFP GPI aggregates. Cells were chemically fixed prior to incubation with culture supernatants from cells expressing the indicated forms of GPI-anchorless mCherry protein and confocal imaging. Arrows, Sup35-GFP GPI aggregates. Arrowheads, newly induced Sup35-mC aggregates. White in the merge panel (G) indicates areas of co-localization. Images correspond to a single 0.5-m optical z slice from near the middle of the cells. Scale bar, 10 m. C, cell-free seeding assay. Microsome fractions from Sup35-GFP GPI -Agg cells as a source of seeds or GFP GPI control cells (or PBS buffer control) were incubated with the indicated substrates. Reactions were analyzed by filter trap assay to detect either preexisting (anti-GFP panels) or newly induced mCherry-tagged (anti-RFP panels; anti-RFP binds mCherry) SDS-insoluble aggregates. D, in situ seeding by Sup35-GFP GPI aggregates on live cells. Live cells were incubated with culture supernatants from cells expressing Sup35-mC followed by confocal imaging. Arrows, Sup35-GFP GPI aggregates. Arrowhead, newly induced Sup35-mC aggregate. White in the merge panel (C) indicates the area of co-localization. Images correspond to a single 0.5-m optical z slice from near the middle of the cells. Scale bar, 10 m. E, chymotrypsin resistance assay. After imaging, cultures in D were lysed, digested with chymotrypsin, and immunoblotted with anti-HA tag antibody (lanes 5-10). Control samples of Sup35-mC culture supernatant treated in parallel were also immunoblotted with anti- Sup35  N (lanes 1 and 2) and M domain antibodies (lanes 3 and 4). Chymotrypsin-negative lanes contain one-quarter sample equivalents loaded in chymotrypsin-treated lanes. Arrow, full-length Sup35-mC. Open arrowhead, chymotrypsin-truncated Sup35-mC. oped a modified in situ seeding assay using live cells. In this assay, live cells were first incubated for 8 -10 h with conditioned medium containing Sup35-mC prior to confocal imaging. As in the experiments described above, we observed highly specific accumulation of Sup35-mC on cell surface Sup35-GFP GPI aggregates (Fig. 6D, A-D) that was absent from aggregate negative Sup35-GFP GPI -Sol control cells (Fig. 6D, E-H). This established that the specific binding of Sup35-mC to Sup35-GFP GPI aggregates in Fig. 6B was not attributed to the use of fixed cells.
After imaging, the cells were lysed and assayed for chymotrypsin-resistant Sup35-mC as a measure of conformational conversion. In control experiments involving chymotrypsin digestion of conditioned medium containing Sup35-mC without exposure to Sup35-GFP GPI cells, no chymotrypsin-resistant bands were detected by immunoblotting with anti-Sup35 N domain antibody (Fig. 6E, lane 1). However, a prominent chymotrypsin-resistant band of about 61 kDa was detected by immunoblotting with anti-Sup35 M domain and anti-HA tag antibodies (Fig. 6E, lanes 3 and 11, open arrowhead), consistent with the expected size shift due to N domain removal by chymotrypsin (50). This allowed us to specifically track the conversion of Sup35-mC in lysates of Sup35-GFP GPI cells treated with Sup35-mC conditioned medium by immunoblotting with anti-HA antibody and detecting chymotryptic digestion of the N domain by size shift. As shown in Fig. 6E, chymotrypsinresistant bands were only observed in lysates of Sup35-GFP GPI -Agg cells treated with Sup35-mC conditioned medium (lane 9). The predominant band was of similar apparent molecular weight to full-length Sup35-mC present in both undigested control samples and Sup35-mC conditioned medium alone ( lanes 6 and 12, arrow), providing strong evidence of conformational conversion of Sup35-mC. Collectively, the four assays described above show that aggregates in Sup35-GFP GPI -Agg cells are able to induce the formation of new Sup35 aggregates and therefore have self-propagating activity.
Cell Surface Sup35 GPI Aggregates Share Properties with Membrane-bound, Non-fibrillar PrPSc Aggregates-GPI-anchored PrPSc aggregates are known to resist cleavage with PIPLC, which is able to release non-aggregated GPI-anchored proteins, including PrPC, from the membrane (65). To test whether Sup35-GFP GPI aggregates shared this property with PrPSc, we monitored live cells expressing both non-aggregated and aggregated forms of Sup35 before and after PIPLC treatment. GFP fluorescence corresponding to cell surface, non-aggregated Sup35-GFP GPI , as shown by co-localization with cell surface WGA staining, was present in untreated control cells (Fig. 7, PIPLC (Ϫ) arrowheads). Cells producing non-aggregated Sup35-GFP GPI lost all cell surface GFP fluorescence after the addition of PIPLC, indicating that the enzyme had successfully cleaved Sup35-GFP GPI from the cell surface, with only fluorescence from intracellular Sup35-GFP GPI remaining. This was true for both Sup35-GFP GPI -Sol cells and those cells in Sup35-GFP GPI -Agg cultures that lacked visible surface aggregates. However, the aggregated form resisted release (Fig. 7, arrows), similar to GPI-anchored PrPSc aggregates. Surprisingly, WGA co-localized extensively with Sup35-GFP GPI aggregates (Fig. 7,  arrows), suggesting an accumulation of glycosylated molecules at these sites. Sup35-GFP GPI aggregates therefore appear to have similar properties to PrPSc, and the results suggest that PIPLC resistance is a generic property of membrane-bound GPI-anchored protein aggregates.
To further characterize the properties of Sup35 GPI aggregates, we examined cells expressing high levels of aggregated mCherry-tagged GPI-anchored Sup35 (Sup35-mC GPI ) treated with the fluorescent dye ThT. ThT, and its analog ThS, are amyloid-specific dyes commonly used to identify amyloid aggregates. Contoured aggregates tended to predominate in Sup35-mC GPI cells (Fig. 8, G-I, arrows) and did not bind ThT. By contrast, positive control cultures of mC GPI cells spiked with rSup35NM amyloid fibrils had numerous very small ThT-positive aggregates (Fig. 8, D-F). mC GPI cells were used instead of Sup35-mC GPI cells in this case to eliminate the possibility of rSup35NM fibrils inducing new Sup35-mC GPI aggregates, which could confound interpretation of the data. Additional positive control experiments using Sup35-mC GPI cells treated with fibrils of recombinant Ure2p(1-80), a known amyloidforming protein, showed that ThT staining also had the sensitivity to detect even small Ure2p(1-80) amyloid aggregates (data not shown). Therefore, had the Sup35-mC GPI aggregates FIGURE 7. Cell surface Sup35-GFP GPI aggregates are resistant to release by PIPLC. Distribution of aggregated (Sup35-GFP GPI -Agg; rows 1 and 2) and non-aggregated (Sup35-GFP GPI -Sol; rows 3 and 4) Sup35-GFP GPI in cells with and without PIPLC treatment is shown. Plasma membrane staining at 4°C with WGA (magenta channel) was used as a marker for the cell surface. Areas of WGA and Sup35-GFP GPI co-localization appear white in the merge panels.
The arrowheads indicate non-aggregated Sup35-GFP GPI in both soluble and aggregated cell cultures. This staining was absent in samples treated with PIPLC, indicating removal from the cell surface. Aggregates (arrows) were resistant to PIPLC-induced release (row 2, arrows). Samples were imaged by confocal microscopy, and images represent a single 0.5-m optical z slice. Scale bar, 10 m.
had similarly low amyloid content as the small rSup35NM and Ure2p(1-80) aggregates, we would have observed them as ThT-positive. Because some amyloid fibrils stain poorly with ThT, we also attempted to stain Sup35-mC GPI and Sup35-GFP GPI aggregates each with at least one of six other fluorescent amyloid dyes, where the respective fluorescence emission spectra of GFP/mCherry and the dye did not overlap. Some of these dyes are reported to have a dramatically higher affinity for amyloid fibrils or achieve brighter in situ staining than classic amyloid dyes like ThT, ThS, and Congo Red (66 -68). The additional dyes tested included ThS, alkaline Congo Red (69), K114 (70), BSB (71), BTA-1 (66), and Amylo-Glo (68). Where possible, we attempted staining procedures on both live and fixed cells. In every case, there was no evidence of staining GPI-anchored Sup35 aggregates (data not shown). We also observed that all six additional dyes tested showed a higher level of diffuse background intracellular staining of cells than ThT. These data suggest that the membrane-bound GPI-anchored Sup35 GPI aggregates were not amyloid in nature. Furthermore, propagation of Sup35-GFP GPI aggregates was not affected by prolonged serial passage (Ͼ12 passes) in the presence of up to 20 M (the highest concentration possible without inhibiting cell growth) of the amyloid-binding dye Congo Red (data not shown), which has been reported to inhibit propagation of recombinant Sup35NM amyloid fibrils in vitro (72). There was also no evidence of Congo Red binding to Sup35-GFP GPI aggregates under these conditions when viewed by fluorescence microscopy, a sensitive technique for visualizing Congo Red (73). Together, these data show that GPI anchoring imparts proper-ties to Sup35-GFP aggregates such that they resemble nonfibrillar, membrane-bound PrPSc aggregates.
Membrane-associated Sup35-GFP GPI Aggregates Exhibit a Non-fibrillar Morphology-To examine the morphology and ultrastructure of the Sup35-GFP GPI aggregates, non-permeabilized cells were immunogold-labeled on the cell surface using an anti-GFP primary antibody and Alexa Fluor 594-conjugated FluoroNanogold secondary and imaged by electron microscopy. Fig. 9 shows SEM images of untransfected negative control cells (A and B) and cells expressing either the GFP GPI control construct (C and D) or Sup35-GFP GPI (E-M). Fig. 9, L and M, shows an absence of gold labeling of Sup35-GFP GPI -Agg cells when either primary or secondary antibody was omitted, further demonstrating the specificity of immunolabeling. In all transfected cells labeled with both primary and secondary antibodies, gold particles (observed as white dots) decorated the cell surface, frequently on filopodia (Fig. 9, C-J). However, whereas the distribution of gold particles on Sup35-GFP GPI -Sol cells closely resembled that of cells expressing the GFP GPI control construct (Fig. 9, D and F, arrowheads), dense accumulations of gold particles were observed only in cells where aggregates formed (Fig. 9, G-J, arrows). Aggregates tended to cluster on protruding regions of the membrane in what resembled vesicular structures. These observations were supported in TEM images of sectioned cells (Fig. 10), where the distribution of gold particles (black dots) in Sup35-GFP GPI -Sol cells was similar to that in GFP GPI -expressing cells (Fig. 10, A and B, respectively, black arrowheads). Untransfected cells had rare cell surface labeling (D), and only in Sup35-GFP GPI -Agg cells were there dense accumulations of gold particles (Fig. 10C, black  arrow). TEM control images of Sup35-GFP GPI -Agg cells labeled without primary antibody (data not shown) were indistinguishable from images of untransfected N2a controls (Fig.  10D), with only rare gold particles and no areas of dense gold labeling corresponding to aggregates ever observed. The TEM images also showed that the vesicular structures associated with aggregates were composed of a single lipid bilayer (Fig.  10E). FluoroNanogold immunolabeling visualized by EM correlated well with visualization of aggregates in the same samples by wide field fluorescence microscopy (Fig. 11), providing additional evidence that the presence of high concentrations of gold particles corresponded to aggregates of Sup35-GFP GPI .
Differences in aggregate ultrastructure (i.e. contoured and irregular) were further illustrated in immuno-EM experiments using immunoperoxidase-based labeling instead of Fluoro-Nanogold (Fig. 12). Contoured aggregates resided in a linear fashion along the plasma membrane (Fig. 12D), and irregular aggregates were more localized and projected further from the membrane (Fig. 12, E and F), with both types of aggregates associated with vesicular structures (arrows), as suggested above. The arrowheads in Fig. 12 indicate staining of Sup35-GFP GPI -Sol cells (C), which resembled that of the GFP GPI control construct (A), whereas there was no staining in untransfected negative control cells (B).
In order to provide a positive control for TEM experiments, biotin-labeled rSup35NM fibrils were prepared and added to GFP GPI control cells (Fig. 13). The biotin conjugate enabled specific immunodetection of the rSup35NM fibrils. GFP GPI FIGURE 8. Sup35-mC GPI aggregates do not stain with Thioflavin T. Cultures were stained with the amyloid-specific dye Thioflavin T. mC GPI , negative control cells stably expressing GPI-anchored mCherry (A-C). mC GPI ϩ rSup35NM fibrils, mC GPI cells treated with rSup35NM amyloid fibrils as a positive control (D-F). Sup35-mC GPI , cells stably expressing high levels of aggregated Sup35-mC GPI (G-I). Sup35-mC GPI was used instead of GFP-tagged versions of the construct for these experiments due to spectral overlap between GFP and ThT. Sup35-mC GPI aggregates did not exhibit any fluorescence when treated with ThT (G-I, arrows), unlike rSup35NM fibrils (D, arrowheads). Scale bar, 10 m. cells were used because these cells do not contain aggregates and are not susceptible to aggregate induction following fibril treatment, although concurrent experiments were carried out on Sup35-GFP GPI -Sol cells with similar results. Immunoblotting with anti-Sup35M antibody and streptavidin established that the rSup35NM fibrils were biotinylated (Fig. 13A). TEM showed that the morphology of biotinylated fibrils resembled that of the unlabeled Sup35NM fibrils (Fig. 13B). Wide field fluorescence microscopy analysis of fibril-treated cells followed by immunolabeling with streptavidin-FluoroNanogold showed that fibrils were bound to the cell surface, as would be expected due to incubation at 4°C (SA-FNG; middle panel). The immunolabeling was specific because control samples either without fibril treatment or with fibril treatment followed by immunolabeling with streptavidin-FluoroNanogold preadsorbed with biotin were negative, showing that labeling was dependent on fibril treatment and mediated by biotin (Fig. 13C,  SA-FNG, left and right panels). Further examination by TEM revealed gold-labeled fibrils (Fig. 13D, B, arrowheads) and unlabeled (Fig. 13D, C, arrowheads) fibrils clearly visible on the cell FIGURE 10. Ultrastructure of membrane-bound Sup35-GFP GPI aggregates by transmission electron microscopy. Cells were immunogold-labeled without permeabilization with anti-GFP antibody and imaged by TEM. The enhanced FluoroNanogold particles were observed as large, electron-dense particles (black arrows and arrowheads). White arrows in E and F indicate very small electron-dense objects that corresponded to background from the gold enhancement procedure to detect nanogold, as shown by their appearance in negative control samples where the FluoroNanogold secondary was omitted (e.g. F). Scale bars, 2 m (A-D) and 100 nm (E and F). Dense accumulations of gold particles were only present in Sup35-GFP GPI -Agg cells (C and E, black arrowheads) and did not appear fibrillar.
surface. The diameter of rSup35NM fibrils (ϳ12 nm (54)) was expected to be about half the diameter of Sup35-GFP GPI fibrils based on the closest available estimates (52,53,74). To ensure clear visualization of the small rSup35NM fibrils, FluoroNanogold particles were enhanced using a modified gold enhancement procedure that blocks the background nucleation by cellular metals seen in Fig. 10F and produces smaller enhanced FluoroNanogold particles. To allow for direct comparison, we repeated anti-GFP/FluoroNanogold immunolabeling of Sup35-GFP GPI -Agg cells using the modified gold enhancement procedure. The results were completely consistent with those described for Fig. 10, with dense FluoroNanogold labeling corresponding to Sup35-GFP GPI aggregates only visualized on small vesicles that accumulated at the cell surface (Fig. 13D, E and F). The rSup35NM fibrils were readily distinguished from Sup35-GFP GPI aggregates, providing further evidence that GPI-anchored Sup35 aggregates are not fibrillar.
Similar patterns of staining were seen in SEM, TEM, fluorescence, and immunoperoxidase experiments, where cell surface aggregates specific to Sup35-GFP GPI -Agg cells were clearly observed. Furthermore, Sup35NM amyloid bound to cells was easily visualized by TEM and appeared fibrillar in morphology. Together, these data show that Sup35-GFP GPI aggregates differed in their ultrastructure and morphology from non-aggregated forms and that although aggregates may take on alternative morphological characteristics (irregular or contoured), no elongated, fibrillar, or rodlike structures resembling amyloid fibrils were ever observed.
Seeding with Amyloid Fibrils Induces Non-fibrillar Aggregate Formation-To determine whether Sup35-GFP GPI aggregates would adopt fibrillar amyloid structure when seeded by amyloid fibrils, Sup35-GFP GPI -Sol cells were treated with rSup35NM amyloid fibrils and then serially passaged many times to dilute out the input fibrils as described previously (47). This treatment led to induction of self-propagating Sup35-GFP GPI aggregates, shown by SEM after immunogold staining in Fig. 14. Cells stained with anti-GFP had abundant gold particles with two classes of distribution. Some cells exhibited dispersed gold particles, largely on filopodia, in a distribution typical of non-aggregated Sup35-GFP GPI observed in Sup35-GFP GPI -Sol cells (arrowhead). Areas with very high concentrations of gold particles corresponding to Sup35-GFP GPI aggregates were also detected, often near points of intercellular contact (arrows). Only rare gold particles were present in negative control samples labeled using an irrelevant isotypematched primary antibody (Fig. 14A). The ultrastructure of the amyloid fibril-induced Sup35-GFP GPI aggregates was very similar to the aggregates of spontaneous origin in the Sup35-GFP GPI -Agg cells (Fig. 9, G-J), consisting of membrane-associated clusters of vesicle-like material with no evidence of amyloid fibrils.

DISCUSSION
The experiments outlined here investigated ultrastructural and biophysical properties of aggregates formed by a GPI-anchored form of Sup35NM in a neuronal cell line. This protein is known to have a high propensity to form amyloid both in vitro and in vivo (51)(52)(53)(54)(55). However, by directing Sup35NM to the plasma membrane via the addition of a GPI anchor sequence at the C terminus, we altered its aggregation properties. We have shown using biochemical and imaging methods that when Sup35NM is tethered to the cell membrane by a GPI anchor, it maintains the ability to aggregate, but the structures formed are not fibrillar. Our data suggest that GPI anchoring can influence protein aggregation by directing the formation of smaller, membrane-bound protein assemblies and blocking formation of higher order aggregates.
When stably expressed in N2a cells, the Sup35-GFP GPI fusion protein existed either as non-aggregated protein or as self-propagating aggregates that arose spontaneously with no exogenous seeding. We do not believe that GFP had any influence on aggregation in this system because Sup35-GFP GPI -Sol cells could be stably propagated long term (Ͼ50 passes) without forming aggregates, and the GPI-anchored GFP control construct never aggregated spontaneously. Furthermore, GFPtagged Sup35NM does not spontaneously aggregate when overexpressed in the cytosol of either prion-negative ([pin Ϫ ] [psi Ϫ ]) yeast (75,76) or N2a cells (58), but Sup35-GFP can convert to amyloid in [PSI ϩ ] yeast (53), (52) and in vitro (74). Why spontaneous aggregates arose in some cells but not others is unclear, although it does not appear to require the highest Sup35-GFP GPI expression possible, because the Sup35-GFP GPI -Sol clone used in these experiments expressed higher overall levels of Sup35-GFP GPI than another Sup35-GFP GPI -Agg clone called D8. Our system is the first to examine Sup35NM aggregation in association with membranes, which may influence the efficiency of the assembly process compared with Sup35NM in solution. By virtue of its GPI anchor, Sup35-GFP GPI should be localized to lipid raft regions of the membrane. Tethering to membranes and clustering in lipid raft domains may increase the effective local concentration of GPI-anchored Sup35NM such that a critical concentration for spontaneous aggregation can be reached, as has been reported previously (77). Such spontaneous aggregates have not been observed in systems where Sup35 is expressed in the cytosol (57,58), possibly because in this environment, where proteins are less restricted, the local critical concentration required for aggregation is not  A and B) and cells producing non-aggregated (C) or aggregated (D-F) Sup35-GFP GPI were stained without permeabilization with anti-GFP primary antibody and secondary antibody conjugated to HRP, which catalyzes the conversion of DAB substrate into a product detectable by TEM. Staining was present all along the surface membranes of transfected cells, with occasional small regions of more intense staining in both GFP GPI and Sup35-GFP GPI -Sol cells (arrowheads). The small intensely stained regions probably correspond to concentrations of GPI-anchored protein in individual raft membrane microdomains. Sup35-GFP GPI -Agg cells showed a similar staining pattern but were distinguished by the presence of very large, intensely stained regions that appeared to consist of cell surface accumulations of vesicles (arrows) resembling aggregate structures observed by immunogold labeling. The location, morphology, relative staining intensity, and specificity to Sup35-GFP GPI -Agg cells all strongly indicate that these large, intensely stained regions correspond to Sup35-GFP GPI aggregates. No staining was observed on N2a control cells (B) or on additional control samples for all cells where incubation with primary antibody was omitted (data not shown). All scale bars, 500 nm. easily reached. In addition, cytosolic chaperones may detect early misfolded isoforms and assist with their refolding or degradation; such mechanisms are not known to exist in the extra-cellular environment. Because GPI-anchored Sup35 aggregates can spread intercellularly (47), the spontaneous development of an aggregate in even a single cell in the culture could ultimately  , 100 nm). C, wide field fluorescence microscopy images of GFP GPI cells treated with biotinylated fibrils and immunolabeled using FluoroNanogold. For cells treated with fibrils, the immunostaining (SA-FNG; middle panel) appeared as both punctate spots (arrows) and more uniform cell surface labeling (arrowhead). Negative control samples using either GFP GPI cells not treated with fibrils and labeled (left column) or GFP GPI cells treated with fibrils and labeled with streptavidin-conjugated FluoroNanogold pre-adsorbed with biotin (right column) demonstrate labeling specificity. SA-FNG, FluoroNanogold channel. GFP, GFP channel. Scale bars, 10 m. D, TEM visualization of biotinylated rSup35NM fibrils added to GFP GPI cells. GFP GPI cells were treated with fibrils and immunolabeled as in C. Fibrils (black arrowheads) were clearly visible in both FluoroNanogold-labeled (subpanel B) and unlabeled (subpanel C) samples. No fibrils or immunolabeling were observed in untreated cells (subpanel A). Sup35-GFP GPI aggregates (arrows) on Sup35-GFP GPI -Agg cells immunolabeled with anti-GFP (subpanels D and E) or without primary antibody (subpanel F) and processed using the same modified gold enhancement procedure as in subpanels A-C are shown for comparison. Note that subpanels E and F were acquired at twice the magnification of subpanels A-D and that immunolabeled structures are not fibrillar. Scale bars, 100 nm.
lead to induction of aggregation in other cells to amplify the "infection." We confirmed that the aggregates in Sup35-GFP GPI -Agg cells were detergent-insoluble and protease-resistant and had self-propagating activity. Hence, they were biochemically similar to Sup35NM amyloid fibrils infectious for yeast cells (78) and Sup35-GFP GPI aggregates induced by such fibrils (Figs. 3, 4, and 6). The Sup35-GFP GPI aggregates could be characterized on the basis of their appearance as either irregular or contoured. By fluorescence microscopy, irregular aggregates were amorphous and spherical, whereas contoured forms possessed a more elongated shape. It is not known whether these two forms reflect alternative packing arrangements at a molecular level or if they only differ in their higher order ultrastructure nor if one may morph into the other. The morphology of aggregates at points of intercellular contact can change as cells move, as shown by time lapse live cell imaging (47). Alternatively, it is possible that the different morphologies represent different cellular responses to accumulation of membrane-bound protein aggregates, with the amorphous aggregates representing more aggressive attempts by cells to shed aggregates by budding off small vesicles from the cell membrane. Different Sup35NM aggregate morphologies have also been reported for Sup35NM expressed in the cytoplasm of N2a cells, but this seems related to clonal differences between N2a cell lines (57). Whatever the explanation, cultures of two independent single cell clones of Sup35-GFP GPI -Agg cells contained both types of aggregate morphologies. Likewise, both irregular and contoured aggregates appear in cultures of Sup35-GFP GPI -Sol cells where aggregates have been induced by treatment with rSup35NM fibrils (47). Thus, further investigation is required to determine the basis for different aggregate morphologies in our system.
Although we observed variations in the appearance of aggregates, they were relatively amorphous and did not resemble fibrillar material typically described as amyloid, as supported by several lines of evidence, including the absence of staining by seven different amyloid-binding fluorescent compounds ( Fig.  8) (data not shown) and ultrastructural characterization of Sup35-GFP GPI aggregates by immuno-EM (Figs. 9, 10, 12, and 13). Furthermore, aggregates but not fibrils were induced in Sup35-GFP GPI -Sol cells following the addition of rSup35NM amyloid fibrils, which propagate in vitro and in vivo by templating their conformation onto non-GPI-anchored Sup35 (78 -80). Given that the cells required several passages after fibril treatment in order to ensure removal of the input fibrils by dilution, we cannot rule out the possibility that the ultrastructure of Sup35-GFP GPI aggregates induced by acute treatment with rSup35NM fibrils changed during cell passage. We attempted to visualize the ultrastructure of acutely induced Sup35-GFP GPI aggregates, but very poor labeling of input fibrils and induced aggregates by colloidal gold secondary conjugates of different sizes has thus far prevented attempts to unambiguously differentiate input rSup35NM fibrils from newly induced Sup35-GFP GPI aggregates. Nevertheless, the combined data suggest that although the Sup35-GFP GPI aggregates share some of the biochemical characteristics (detergent insolubility, protease resistance, and self-propagation) of amyloid fibrils formed by the same protein, they lack other important defining features of amyloid. That said, amyloid is the only current structural model for protein aggregates that exhibit conformational templating activity.
It is possible that the GPI-anchored Sup35NM aggregates are intermediates on the pathway to amyloid fibrils, such as oligomers, protofibrils, or very short protofilaments, and that membrane tethering by the GPI anchor restricts maturation of GPI-anchored Sup35NM aggregates into long amyloid fibrils. Recent data indi- cate that even Sup35 oligomers are sufficient for prion transmission and infectivity (81), although whether these oligomers have any relation to GPI-anchored Sup35 aggregates is unknown. Distinguishing between these different possibilities will require further investigation of Sup35-GFP GPI aggregates.
TSE-associated GPI-anchored aggregates composed of misfolded prion protein usually appear as small, diffuse assemblies that do not appear to be amyloid fibrils (14,15,22). However, PrPSc can also form amyloid, demonstrated by the presence of extracellular plaques in some cases of TSE and in mice genetically modified to express anchorless prion protein, following extraction or release from cell membranes and in vitro (26,37,(82)(83)(84). Thus, similar to the observations made in this study, GPI anchor-mediated membrane association of the prion protein appears to prevent amyloid fibril formation.
It is difficult to rule out the possibility that membrane-bound GPI-anchored aggregates of PrPSc and/or Sup35NM form exceptionally short amyloid polymers that stain poorly with amyloid-binding dyes. However, prefibrillar aggregates of rSup35NM can bind Congo Red and ThT, indicating that fibrillar ultrastructure is not required for Sup35NM to bind these amyloid dyes (50). In the present study, we were able to detect short (ϳ100 -200-nm) rSup35NM amyloid fibrils associated with the cell surface by both ThT staining and TEM (Figs. 9 and 13D), providing a measure of the sensitivity of our assays. Regardless of the nature of the aggregates formed (amyloid or not), it is clear that GPI-anchored membrane association significantly affects aggregate ultrastructure.
Several of the assays described here have been used to characterize PrPSc aggregates, enabling a comparison between the properties of Sup35-GFP GPI aggregates with PrPSc. Interestingly, like membrane-bound PrPSc aggregates (65,85) and misfolded PrP associated with familial prion diseases (86,87), Sup35-GFP GPI aggregates resisted release from cells by PIPLC digestion (Fig. 7). Furthermore, like GPI-anchored PrPSc aggregates, they were protease-resistant, self-propagating, did not stain with amyloid-specific dyes, and did not appear fibrillar by EM. Thus, by placing a GPI anchor on Sup35NM, we have changed its properties and aggregation pathway, directing away from a typical amyloid fibril to an apparent non-fibrillar aggregate that has similar properties to membrane-bound PrPSc.
The presence of PrPSc on cell membranes is known to induce distinctive membrane pathologies (15,46), and we also noticed some membrane disturbances in cells where Sup35 GPI aggregates were present. Clustered, vesicle-like protrusions appeared to emanate from the plasma membrane, coincident with aggregates. The origin of these vesicles is not clear, although the association with Sup35 aggregates and absence from Sup35-GFP GPI -Sol cells suggests that they may form via membrane blebbing as a cellular response to aggregation. In addition, a significant proportion of Sup35-GFP GPI appeared to be localized on filopodia in cells regardless of whether they were expressing the non-aggregated or aggregated form, consistent with the observation that cholesterol-rich membranes (where GPI-anchored proteins reside) are enriched in filopodia (88,89). The filopodia extend from cell to cell and tend to be enriched at points of intercellular contact like GPI-anchored Sup35 aggregates. Thus, it is also possible that aggregation of GPI-anchored Sup35 on filopodia somehow leads to their degeneration into the vesicle-like structures that we observed. Taken together with observations from TSE-infected animals, our data suggest a common theme that assembly of membrane-tethered ordered protein aggregates can induce membrane perturbation. Further work is required to determine the earliest events associated with formation of GPIanchored Sup35 aggregates.
Because it is known that extraction of GPI-anchored PrPSc aggregates from the cell membrane results in amyloid formation (37,40,41), in situ studies such as those outlined here are important in understanding the properties of aggregates formed in vivo. Although a high resolution structure of membrane-bound PrPSc aggregates will be difficult to achieve using current techniques, which require a highly purified sample, ultrastructural and biochemical analyses can be fruitful. Such studies, including ours, have shown that the GPI anchor is intrinsic to the prion protein, that it has a role in determining the type of aggregate formed (fibrillar versus non-fibrillar), and that it consequently can affect prion pathogenesis (11, 13, 26 -28). Precisely how GPI anchoring prevents assembly into amyloid fibrils is unclear. The GPI anchor may inhibit assembly into amyloid fibrils due to steric interference (26). This could provide an explanation for why we did not see the formation of amyloid after seeding with fibrils but did observe clustering and aggregation induced by the fibrillar seed. Another suggestion is that aggregates may be sheared during attempts by the cell to endocytose them (11). Furthermore, the natural curvature of the cell membrane might impose a limit on how large an aggregate can become and prevent the assembly of membrane-bound proteins into long amyloid fibrils. Because Sup35 is normally expressed in the cytoplasm of yeast, we cannot completely rule out the possibility that environmental conditions encountered by GPI-anchored Sup35NM during trafficking inhibit its ability to form amyloid fibrils. However, numerous studies in the yeast prion literature rely on recombinant Sup35NM amyloid fibrils formed in the presence of dilute denaturants and oxidative conditions, conditions that are likewise far removed from the yeast cytoplasm. Hence, our observations are unlikely to be due simply to cell surface expression of Sup35NM.
We have shown that GPI anchoring to the cell membrane modifies the assembly of a protein that would otherwise form amyloid fibrils to form a non-fibrillar aggregate that shares biochemical and structural properties of PrPSc. This work gives insight into the mechanism by which ordered aggregates of misfolded proteins, such as PrPSc, might assemble when they are in association with membranes, along with the consequences of such processes for cell physiology. It remains to be seen whether other amyloidogenic proteins would mimic this behavior, but understanding how GPI anchoring directs aggregation and what conformations these aggregates might contain will be crucial in understanding its part in the generation of pathogenic and/or infectious species observed in TSE.