Cellular prion protein targets amyloid-β fibril ends via its C-terminal domain to prevent elongation

Oligomeric forms of the amyloid-β (Aβ) peptide are thought to represent the primary synaptotoxic species underlying the neurodegenerative changes seen in Alzheimer's disease. It has been proposed that the cellular prion protein (PrPC) functions as a cell-surface receptor, which binds to Aβ oligomers and transduces their toxic effects. However, the molecular details of the PrPC–Aβ interaction remain uncertain. Here, we investigated the effect of PrPC on polymerization of Aβ under rigorously controlled conditions in which Aβ converts from a monomeric to a fibrillar state via a series of kinetically defined steps. We demonstrated that PrPC specifically inhibited elongation of Aβ fibrils, most likely by binding to the ends of growing fibrils. Surprisingly, this inhibitory effect required the globular C-terminal domain of PrPC, which has not been previously implicated in interactions with Aβ. Our results suggest that PrPC recognizes structural features common to both Aβ oligomers and fibril ends and that this interaction could contribute to the neurotoxic effect of Aβ aggregates. Additionally, our results identify the C terminus of PrPC as a new and potentially more druggable molecular target for treating Alzheimer's disease.

mechanism by which it causes neuronal dysfunction has remained mysterious. It is presumed that the disease process starts by the binding of A␤ oligomers to receptor proteins or lipids on the surface of neurons. However, the molecular identity of the relevant binding sites is uncertain.
In 2009, Laurén et al. (5) identified the cellular prion protein (PrP C ) as a cell-surface receptor for soluble oligomers of the A␤ peptide, an observation subsequently confirmed by other groups (6 -8). Binding of A␤ oligomers to PrP C was shown to produce neurotoxic effects, including suppression of long-term potentiation and retraction of dendritic spines; it was reported that these effects depended upon interactions between PrP C and metabotropic glutamate 5 receptor, resulting in activation of intracellular Fyn kinase (9 -12). In addition, disruption of the endogenous gene encoding PrP C was shown to rescue behavioral deficits as well as early mortality in certain AD transgenic models (13). Taken together, these results led to the proposal that PrP C is one of the receptors mediating the synaptotoxic effects of A␤ oligomers and that pharmacologic targeting of PrP C could represent a novel therapeutic strategy for treatment of AD (14 -16). Although some studies have disputed the importance of PrP C in mediating A␤ toxicity (17)(18)(19)(20), these discrepancies may result from the use of different experimental paradigms, variability in preparations of synthetic A␤ oligomers, or the involvement of additional neurotoxic pathways.
Most previous studies of PrP C -A␤ interaction have focused on oligomeric preparations of A␤ referred to as amyloid-␤derived diffusible ligands (ADDLs), which are produced by resolubilization of synthetic A␤ peptide in aqueous medium (21). ADDL preparations have the virtue of being enriched in soluble, oligomeric species, but they are very heterogeneous in terms of size, and they do not obviously correspond to any of the intermediate states that have been described during the polymerization of A␤ from a monomeric state. Under suitably controlled conditions, A␤ polymerizes by a well-studied process involving distinct steps of primary nucleation, secondary nucleation, and elongation (22)(23)(24). These kinetic steps have been mathematically modeled, and their rate constants have been determined (25)(26)(27). To explore more fully the nature of the PrP C -A␤ interaction, we sought to determine how PrP C affected the A␤ polymerization process and to use the results of these experiments to provide further insight into the nature of the A␤ species targeted by PrP C . Our results have important implications for understanding the role of PrP C as a receptor transducing the neurotoxic effects of A␤, and they suggest a novel approach for disrupting the PrP C -A␤ interaction for therapeutic purposes.

A␤ preparation and polymerization
We used carefully defined conditions for polymerization of A␤  (hereafter referred to simply as A␤), which have been described previously (22,23,28,29) and which have been shown to result in reproducible kinetic curves as monitored by thioflavin T (ThT) fluorescence. To ensure that our experiments began with monomeric A␤, we solubilized lyophilized peptide in 15 mM NaOH and used a Superdex 75 column to isolate monomers (Fig. 1a). A␤ in the monomer fraction pro-duced a random coil signature by circular dichroism (Fig. 1d, 0 min curve), consistent with the absence of significant ␤-sheetcontaining aggregates.
Upon incubation in physiological buffer, monomeric A␤ polymerized in a highly reproducible fashion based on ThT fluorescence, following characteristic sigmoidal kinetics (Fig.  1b). Starting with a monomeric A␤ concentration of 5 M, polymerization proceeded with a typical lag time of 50 -60 min, reaching a plateau after about 90 -100 min, with a half-time of 70 -80 min. Using analytical size-exclusion chromatography (SEC), we observed continuous depletion of monomers in the A␤ sample over time, reflecting their incorporation into filaments that were removed by filtration and ultracentrifugation a, size-exclusion chromatograph for preparation of A␤ monomers on a Superdex 75 column. Rectangle, the A␤ monomer peak collected for use in kinetic experiments. Arrows, elution volume of molecular weight standards. b, polymerization of A␤ (5 M) monitored with ThT. Each set of colored dots represents one polymerization run with triplicate samples (for a total of seven polymerization runs). c, analytical SEC (Agilent Bio Sec-3 column) of samples taken at different times from the polymerization reaction shown in b. Arrows, molecular weight standards. Samples were centrifuged and filtered to remove insoluble aggregates before injection into the column. d, far-UV circular dichroism of A␤ at the indicated times during the course of polymerization. The CD signature shifts from random coil to ␤-sheet. e, EM images of negatively stained preparations of A␤ (10 M) taken at 0 min (monomers), 38 min (early exponential phase), and 16 h (plateau phase). Scale bars, for monomer, 200 nm; for 38 min and 16 h, 500 nm. Samples were concentrated by centrifugation before imaging. f, SEC of ADDLs on a Superdex 75 column, showing a peak in the void volume. g, EM of negatively stained ADDLs showing small globular aggregates. Scale bar, 500 nm. h, ThT curves for ADDLs incubated under the same conditions used for polymerization of A␤ monomers shown in b. The ThT signal is much lower than in b and does not change substantially over 16 h.

The prion protein prevents amyloid-␤ fibril elongation
before chromatography (Fig. 1c). We did not detect a significant population of oligomeric species that migrated with a molecular size of Յ100 kDa at any time during the course of polymerization. This observation is consistent with previous reports that oligomers, although obligate intermediates during the polymerization process, never constitute Ͼ1% of the total A␤ mass during the course of the reaction, with the major species being monomers and fibrils, the ratio of which changes continuously aspolymerizationproceeds (22).Duringthepolymerizationprocess, the CD signature of the A␤ gradually shifted from random coil to ␤-sheet, consistent with incorporation of unstructured monomers into amyloid fibrils (Fig. 1d).
Using electron microscopy with negative staining, we observed that fibrils formed early during the polymerization process, consistent with previous reports that fibrils are first detectable during the lag phase (30,31) (Fig. 1e, center image). During the early exponential phase, the fibrils had lengths of 0.5-2 m and displayed a twisted morphology with diameters ranging between 4.1 Ϯ 0.8 nm in the narrow regions and 12.1 Ϯ 1.5 nm in the thick regions. At later time points, fibrils tended to clump together, and their lengths and morphology were more difficult to discern (Fig. 1e, right image). These structural features have been described previously (32). Only scattered fibrils were observed in the isolated monomer fraction at zero time, and these were much more difficult to locate on the EM grid. For the most part, fibrils were not detected on the surface of the grid when freshly prepared A␤ was applied (Fig. 1e, left image). Most likely, the scattered fibrils that were observed represent a very small proportion of the total A␤ present at this time point and are a reflection of how rapidly aggregates begin to form after monomers of A␤ are isolated.
For the purposes of comparison, we utilized ADDL preparations in some of the experiments described below. ADDLs are typically prepared by solubilizing A␤ in HFIP, drying to a thin film, resuspending the film in DMSO, diluting into aqueous medium, and incubating for 16 h at room temperature. These preparations consisted primarily of aggregates that eluted in the void volume of the Superdex 75 column, indicating a molecular size Ͼ 70 kDa (Fig. 1f). As described previously (21), these aggregates appeared as small, globular assemblies of 5-10 nm in diameter by EM (Fig. 1g). The ThT-binding signal obtained from ADDL preparations, even at 20 or 100 M, was considerably lower than the maximum signal achieved by fully polymerized A␤ at 5-10 M, consistent with the absence of long fibrils in the ADDL samples (Fig. 1h). The ThT signal of the ADDL preparations did not increase further with continued incubation, indicating that the aggregates did not fibrillize during the 16-h period after formation.

PrP delays fibril formation at substoichiometric ratios
We sought to determine what effect PrP has on the polymerization process itself. We found that recombinant PrP profoundly delays A␤ polymerization, even when present in amounts that are highly substoichiometric to A␤ (Fig. 2a). The half-time for polymerization was nearly doubled at a PrP/A␤ ratio of 1:160 and tripled at a ratio of 1:20 (Fig. 2b). However, even in the presence of PrP, polymerization eventually reached the same plateau value of ThT binding, indicating that PrP at the concentrations examined slowed, but did not prevent, conversion of monomeric to fibrillar A␤.
The inhibitory effect of PrP on A␤ fibril formation could also be visualized using semidenaturing detergent agarose gel electrophoresis (SDD-AGE), which provides a means of separating large, SDS-resistant amyloid fibrils from monomers and smaller aggregates on agarose gels. After 160 min, samples polymerized without PrP, which had reached the plateau phase of ThT binding, contained substantial amounts of fibrillar material, which migrated as a broad smear (Fig. 2c, lane 6). At this same time point, samples polymerized in the presence of increasing amounts of PrP contained decreasing amounts of fibrillar material on SDD-AGE (Fig. 2c, lanes 1-5), corresponding to the lower levels ThT binding reached by these samples. When SDD-AGE analysis was performed at 16 h, when plateau values of ThT fluorescence had been attained in all samples, there was no apparent difference in the amount of fibrillar material between PrP-containing and control reactions (Fig.  2d), again indicating that PrP in substoichiometric amounts slows but does not completely prevent fibril formation.
We also assessed the effect of PrP on polymerization of fluorescently labeled A␤ using fluorescence polarization (FP). In both the presence and absence of PrP, polarization increased with time, reflecting incorporation of labeled A␤ monomers into fibrils, which have a lower rotational mobility (Fig. 2e). This change in polarization was slower in the presence of PrP, consistent with an inhibitory effect on fibril formation. The FP signal plateaued at a similar value with and without PrP, suggesting again that PrP delayed but did not prevent fibril formation, with all of the monomers eventually being converted to the fibrillar form.

PrP does not prevent secondary nucleation by preformed fibrils
When a small amount of preformed fibrils is added at the start of an A␤ aggregation reaction, the rate-limiting, primary nucleation step is bypassed, shortening the lag phase and resulting in rapid formation of new fibrils by secondary nucleation and elongation (22). We investigated how PrP affected this process. We seeded a solution of 5 M A␤ with a 1% molar equivalent of preformed fibrils in the presence and absence of PrP and compared the results with equivalent unseeded reactions. As expected, in the absence of PrP, seeding significantly accelerated the polymerization reaction (Fig. 3a, 0 nM curve). PrP showed an inhibitory effect in seeded reactions (Fig. 3a), gradually damping the acceleration produced by the seeds. Importantly, however, the strength of the effect was reduced when compared with non-seeded reactions with equivalent amounts of PrP (Fig. 3b). For each concentration of PrP, the half-time was significantly decreased by the addition of seeds (Fig. 3c). This result held true independent of whether PrP was preincubated with the fibrils before their addition to the reaction (not shown). In another variation of the experiment, we found that fibrils formed in the presence of PrP accelerated polymerization reactions to nearly the same extent as fibrils formed in the absence of PrP (Fig. 3, d and e). Taken together, these results suggest that PrP does not have a major effect on the secondary The prion protein prevents amyloid-␤ fibril elongation nucleation phenomena that occur when reactions are seeded by preformed fibrils.

PrP selectively inhibits filament elongation
Given the mechanistically well-characterized features of A␤ polymerization under the controlled experimental conditions we employed, we had an opportunity to pinpoint which microscopic step(s) were being affected by PrP using a mathematical modeling approach based on the macroscopic ThT curves. This approach has been used successfully to characterize interactions between A␤ and several molecular chaperones (23). We first determined the integrated rate law for A␤ aggregation in the absence of PrP, using as a guideline published values for the key rate constants (see supplemental material). We then fit the ThT curves in the presence of PrP by systematically varying the rate constant for only one of the three molecular steps in the polymerization process: k n for primary nucleation, k 2 for secondary nucleation, and k ϩ for elongation. The best global fit to the data was achieved when the elongation rate (k ϩ ) was varied in response to PrP addition (Fig. 4a). The sum of residual errors for this fit was 1.7, compared with 7.7 and 17.7 for the fits to variations in k n and k 2 , respectively ( Fig. 4, b and c). The calculated values for k ϩ exhibit a strong influence of PrP concentration on elongation rate, which dropped to 6% of the uninhibited value in the presence of 250 nM PrP, a 1:12 ratio of PrP to A␤ ( Fig. 4a (inset), and supplemental Tables S1 and S2).

The prion protein prevents amyloid-␤ fibril elongation
As A␤ fibrils are thought to grow by monomer addition to the fibril ends (22), a plausible mechanism for the inhibitory effect of PrP on fibril elongation is that PrP binds specifically to the growing ends of the fibrils, preventing monomer addition. If one assumes that PrP is present in excess and that it binds rapidly to fibril ends as soon as they are generated (i.e. that binding is at equilibrium), it is possible to derive a mathematical expression, in the form of a Langmuir binding isotherm, relating the normalized values for k ϩ to the concentration of PrP. This expression incorporates an equilibrium constant, K eq , for binding of PrP to the fibril ends. The experimentally determined k ϩ values in the presence of increasing concentrations of PrP provide an excellent fit to this model (Fig. 4a, inset), yielding a value for K eq of 2.1 ϫ 10 7 M Ϫ1 , corresponding to an affinity constant of 47.6 nM. This quantitative analysis provides strong evidence that PrP selectively inhibits elongation of A␤ fibrils and suggests that it does so by binding tightly and selectively to fibril ends.

PrP binds both monomeric and fibrillar forms of A␤ but with different affinities
To provide biochemical evidence for this model, we examined binding interactions between PrP and the two major spe-cies present during the polymerization reaction: monomers and fibrils. For comparison, we also analyzed PrP binding to ADDLs, an interaction that has been previously characterized. We employed three different techniques: surface plasmon resonance (SPR), dissociation-enhanced lanthanide fluorescence immunoassay (DELFIA), and FP.
For SPR experiments, we tagged recombinant PrP with a c-Myc epitope at its C terminus and captured it on the SPR chip using 9E10 antibody. This strategy was adopted to leave the N-terminal domain of PrP, which contains the two putative A␤-binding sites, free to interact with A␤ that was injected over the chip in the mobile phase. We compared the binding ability of A␤ samples taken at 0 min, representing mainly monomer, and at 16 h, representing fully polymerized fibrils. We found that the 0-min sample displayed detectable binding (ϳ180 resonance units (RU) after 240 s of injection) only at the highest concentration of A␤ (15 M) (Fig. 5a). The 16-h sample gave much larger responses, ranging from 180 to 700 RU over a concentration span of 0.9 -15 M (Fig. 5b), presumably reflecting the larger molecular weight of fibrils compared with monomers. ADDLs also bound to PrP, consistent with previous reports,withresponsemagnitudesintermediatebetweenmono- The prion protein prevents amyloid-␤ fibril elongation mers and fibrils (supplemental Fig. 1a). These results demonstrate that PrP binds efficiently to fully polymerized A␤ fibrils and to ADDLs. PrP may also bind weakly to A␤ monomers, although the much smaller mass of monomers compared with fibrils would make binding of this species more difficult to detect. It was not possible to calculate reliable affinity constants and stoichiometries for the PrP-A␤ binding reaction from these SPR data due to the fact that binding of the analyte (A␤) did not reach saturation during the injection phase. These anomalies of PrP-A␤ interactions in SPR experiments have been noted before (6), and may be related to rebinding or self-association of A␤.
To assess PrP-A␤ interactions under equilibrium conditions, we employed DELFIA. A␤ samples were incubated in plastic wells containing immobilized PrP, and the amount of bound A␤ was then measured using anti-A␤ antibody 6E10. We observed binding of monomers (0 min, both fresh and frozen), fibrils (16 h), and ADDLs, with apparent dissociation constants of 1.07 Ϯ 0.34 M for fibrils and 0.11 Ϯ 0.04 M for ADDLs (supplemental Fig. 1b). Monomer binding did not approach saturation; therefore, the dissociation constant was not calculated. The actual K d values for the aggregated species are likely to be much lower than the apparent values, because the molar concentration of these forms is only a fraction of the total A␤ concentration by a factor equivalent to the number of subunits in each aggregate. Thus, this assay confirms that PrP binds A␤ monomers, polymers, and ADDLs, with a higher affinity for the latter two species.
If PrP binds selectively to the ends of A␤ fibrils, then the amount of binding should be directly related to the effective concentration of fibril ends in the A␤ sample. To test this prediction, we used DELFIA to compare binding of PrP to fibrils before and after shearing of the fibrils via sonication. A sample of sonicated fibrils should contain a larger number of fibril ends compared with an unsonicated sample at the same concentration. As predicted, sonication of fibrils increased PrP binding (supplemental Fig. 2).
Both SPR and DELFIA are surface-based binding techniques, which may be subject to artifacts resulting from potential interference by the substrate with the binding interaction between PrP and A␤. We therefore tested PrP-A␤ interactions using FP, a solution-based technique. As the fluorescent probe, we utilized an N-terminal fragment of PrP (residues 23-109), which encompasses the two previously identified binding regions for A␤ aggregates (5,6). This fragment was labeled on a C-terminal cysteine residue with Alexa Fluor 488 C 5 maleimide (Fig. 5c). We compared the change in polarization for samples taken at different time points of an A␤ polymerization reaction (Fig. 5d). The zero time (monomer) sample did not produce a polarization shift, whereas the samples taken at 24 min (lag phase), 48 min (early exponential phase), and 90 min (plateau phase) produced progressively greater shifts at equivalent A␤ concentrations. These results demonstrate that PrP(23-109) binds to A␤ fibrils that form during the polymerization process. The data do not necessarily indicate a lack of binding to monomers (M r ϭ 4,500), which would be too small to produce a measurable shift in FP values, as demonstrated by the fact that an even larger ligand, an anti-PrP antibody (M r ϭ 150,000), did not cause a shift in polarization (Fig. 5e). However, ADDLs did produce a polarization shift consistent with a size intermediate between monomers and fibrils (Fig. 5f).

The structured, C-terminal domain of PrP is required for inhibition of A␤ fibril elongation and also influences binding to monomers
It has been shown previously that the unstructured N-terminal domain of PrP contains two polybasic regions (residues 23-27 and 95-105), which are required for binding to ADDLs, whereas the globular C-terminal domain is dispensable for this function (Fig. 6a, top schematic) (5,6). We tested the roles of the N-and C-terminal domains in the ability of PrP to inhibit the growth of A␤ fibrils. Fig. 6a shows schematic diagrams of the deletion constructs used for these experiments.
As expected, removing the entire N-terminal domain (yielding construct 110 -230) completely abolished the inhibitory effect of PrP on A␤ fibrillation (Fig. 6, compare b and c). To our surprise, however, we found that the isolated N-terminal domain (residues 23-119), which includes both of the putative ADDL-binding sites, had no effect on polymerization (Fig. 6d). A more C-terminally extended construct, 23-144, which ends just before the first ␣-helix, had a weak inhibitory effect at the highest concentrations, but much less than full-length PrP(23-230) (Fig. 6e). Taken together, these results imply that both the N-terminal and C-terminal domains of PrP are required for efficient inhibition of A␤ fibril elongation. We also tested a construct (⌬105-125), which is missing only a short hinge region connecting the N-and C-terminal domains, and found that it inhibited polymerization less effectively than the wildtype protein, implying that this region is also important for inhibitory activity (Fig. 6f). Fig. 6G summarizes the relative polymerization half-times for each of the PrP constructs.
To determine whether the observed differences in the ability of the PrP constructs to inhibit A␤ polymerization were due to alterations in their binding affinity for A␤ fibrils, we carried out DELFIA-binding assays. Surprisingly, the two C-terminally deleted constructs (23-119 and 23-144), as well as the internally deleted construct (⌬105-125), all of which showed greatly diminished ability to inhibit A␤ polymerization, displayed relatively unimpaired affinity for A␤ fibrils (Fig. 7a). In contrast, the 110 -230 construct, which is missing the N-terminal, ADDL-binding domains, exhibited significantly reduced binding to fibrils. These results imply that fibril binding, like ADDL binding, depends primarily on sites in the N-terminal domain. Importantly, although the C-terminal domain is not required for fibril binding, this domain (along with the hinge region) nevertheless plays a crucial role in the ability of PrP to inhibit fibril elongation.
We also used DELFIA to test the ability of the different PrP constructs to bind to A␤ monomers. Surprisingly, we found that the C-terminally deleted construct 23-119 displayed a greatly reduced ability to bind to A␤ monomers (Fig. 7b), although it showed an affinity for fibrils comparable with fulllength PrP (Fig. 7a). The 23-144 construct displayed slightly reduced monomer binding. As expected, the 110 -230 construct exhibited virtually no monomer binding. These results suggest that the C-terminal domain of PrP (particularly resi-

The prion protein prevents amyloid-␤ fibril elongation
dues 120 -144) influences binding to A␤ monomers, although this region is not essential for binding to A␤ fibrils.

Discussion
There has been considerable interest in the unexpected ability of PrP C to bind A␤ aggregates (5)(6)(7)(8), both because of evidence that PrP C may transduce some of the neurotoxic effects of such aggregates in AD (9 -13) and because of the possibility that exogenous PrP or anti-PrP antibodies could be used as therapeutic agents to neutralize such toxic effects (14 -16, 33). Whereas many previous studies have focused on binding of PrP to heterogeneous preparations of soluble oligomers (ADDLs), we have characterized the effect of PrP on the polymerization of A␤ under highly reproducible conditions in which conversion of monomeric to fibrillar forms proceeds via a series of kinetically well-characterized steps, including primary nucleation,

The prion protein prevents amyloid-␤ fibril elongation
secondary nucleation, and elongation (22)(23)(24). Our results have allowed us to pinpoint which of these steps is inhibited by PrP, as well as which A␤ species and molecular sites PrP is likely to bind. Our study has implications for understanding the neurotoxicity of A␤ oligomers, and it suggests new approaches to targeting PrP for therapeutic purposes in AD.
Taken together, our results suggest a molecular model in which PrP binds tightly to the ends of growing fibrils, specifically inhibiting the elongation step of fibril growth (Fig. 8a). This mechanism is supported by several pieces of evidence. Most importantly, it provides an extremely close fit of the ThT polymerization curves to published differential equations (22,(25)(26)(27) describing the kinetics of A␤ polymerization. In this scheme, the data are best modeled by assuming that PrP specifically reduces k ϩ , the rate constant for fibril elongation, and that it does so by binding to fibril ends with an equilibrium dissociation constant, K d , in the nanomolar range. In contrast, models based on inhibition of primary or secondary nucleation result in very poor fits to the data. The model shown in Fig. 8a is also consistent with the substoichiometric nature of PrP inhibition, because only 1-2 PrP molecules would need to bind to each fibril to completely block elongation. This mechanism is also consistent with our observation that PrP has relatively little effect on seeded polymerization, which depends strongly on secondary nucleation. In contrast, chaperones that inhibit secondary nucleation of A␤ dramatically retard seeded polymerization reactions (28). Finally, we have demonstrated, using several different techniques (SPR, DELFIA, FP), that PrP binds to fibrillar A␤, as would be predicted by this model. Although the K d for this interaction is difficult to calculate from our data due to uncertainty in the actual concentration of fibrils being analyzed, it is likely to be in the submicromolar range, consistent with the value arrived at from the kinetic modeling.
We have made the unexpected observation that fragments of PrP encompassing only the N-terminal domain show a greatly reduced ability to inhibit A␤ polymerization, although their affinity for A␤ fibrils is relatively unaltered. Our data demonstrate that the N-terminal domain of PrP, encompassing the two previously described ADDL-binding sites, is essential for binding to fibrils but that this interaction alone is insufficient to block elongation. Rather, an additional involvement of the globular C-terminal domain is required. Because the globular domain itself lacks significant fibril-binding activity, the question arises as to how this domain contributes to elongation inhibition. One possibility is that binding of the N-terminal domain positions the C-terminal domain in proximity to the fibril end, sterically blocking access of additional A␤ subunits. Alternatively, binding of A␤ to the N-terminal domain of PrP may cause a conformational change in the C-terminal domain that unmasks additional A␤-binding sites in that region. We have found that deletion of a short hinge region (residues 105-125) linking the N-and C-terminal domains impairs the ability of PrP to inhibit polymerization, consistent with the idea that inhibitory activity depends on interactions between the two domains. NMR experiments support such an interdomain docking mechanism (34). We note that, in a previous study, the C-terminally truncated construct PrP(23-144) was reported to inhibit A␤ polymerization as effectively as full-length PrP (33). In contrast, we found that this construct had significantly reduced inhibitory potency, a discrepancy that could be due to the different polymerization conditions used in the two experiments.
Interestingly, we have found that PrP binds weakly to A␤ monomers, and this binding is diminished when the C-terminal domain of PrP (particularly residues 120 -144) is deleted. This observation suggests that the PrP C-terminal domain contributes to recognizing the unstructured conformation of the A␤ monomer. This conformation might be present transiently after a new monomer is added to the end of the growing fibril, before it is locked into the cross-␤-structure characteristic of the rest of the fibril (35), thus explaining why C-terminally deleted PrP constructs have reduced ability to inhibit elongation. Alternatively, it is possible that PrP binding to the zero time sample, which we interpret as monomer binding, really represents binding to a minor population of small oligomers or short fibrils that forms rapidly during the time required to carry out the DELFIA assay. In any case, the fact that PrP inhibits A␤ polymerization at substoichiometric ratios makes it unlikely that it acts primarily by binding to A␤ monomers. Previous studies have documented binding of PrP to A␤ fibrils as well as to ADDLs, although PrP has generally been said to lack affinity for A␤ monomers (5, 6). However, weak binding of PrP to mono-

The prion protein prevents amyloid-␤ fibril elongation
mers has been observed in some cases (15), consistent with the results presented here.
Previous studies have identified several other proteins that act as chaperones affecting particular steps in the polymerization of A␤ or other amyloidogenic proteins (23). For example, DNAJB6 has been shown to block primary nucleation of A␤ by specifically targeting small oligomers (36), and proSP-C BRI-CHOS inhibits secondary nucleation by binding stoichiometrically along the surface of A␤ fibrils (28). A particularly relevant example is Ssa1, an Hsp70-type chaperone that blocks elongation of fibrils formed by the yeast prion protein, Ure2p, most likely by binding to fibril ends (37). The kinetic features of Ssa1 inhibition of Ure2p polymerization, based on ThT curves, are strikingly similar to those of PrP inhibition of A␤ polymerization described here, consistent with the conclusion that both proteins inhibit fibril elongation by binding to fibril ends. In contrast, the ThT curves for A␤ polymerization in the presence of PrP and DNAJB6 are quite distinct from each other, arguing that PrP does not target the small population of oligomers that forms transiently during the reaction.
Many previous studies have reported that PrP C binds tightly and specifically to ADDLs (5)(6)(7)(8). This observation has received considerable attention because ADDLs have been shown to have neurotoxic effects, such as suppression of long-term potentiation and retraction of dendritic spines (4,5). Thus, it has been proposed that PrP C may serve as a cell-surface receptor that binds A␤ oligomers and mediates their neurotoxic effects in the context of AD. What is the relationship between the ability of PrP C to bind ADDLs and its ability, demonstrated here, to bind to the ends of growing A␤ fibrils and inhibit elongation? ADDLs are clearly distinct from the A␤ fibrils that we generate in our experiments, in terms of their smaller size, their globular morphology, their much lower ThT-binding capacity, and their inability to seed polymerization when added to monomeric A␤ (not shown). Although often described as being oligomeric, ADDLs also appear to be different from the smaller oligomers (Ͻ100 kDa) that have been shown to accumulate transiently and at low levels (Ͻ1% of total A␤) when polymerization is performed under the conditions we have used here (22). Despite these differences, however, PrP may bind strongly to ADDLs because they display some of the same structural features as fibril ends (Fig. 8b). Perhaps ADDLs represent A␤ assemblies that are trapped in an elongation-ready mode capable of binding PrP. Although ADDLs do not seem to evolve to a fibrillar state, there is evidence that their morphology can change over time, with the appearance of nanotubular structures that have potent neurotoxicity and bind avidly to PrP C (8).
Our results have important implications for theories of how A␤ causes neurotoxicity in AD. Considerable evidence indicates that A␤ oligomers, rather than fibrils or monomers, have the greatest neurotoxic potency in vivo as well as in cell-based assays (2). The data presented here suggest that this may reflect structural features common to both oligomers and fibril ends and the ability of PrP C and perhaps other cell-surface receptors to recognize these features. According to this hypothesis, small, soluble assemblies of A␤, including oligomers, protofibrils, or nanotubes, may present a high molar concentration of protein surfaces that are structurally equivalent to fibril ends and that can therefore bind to and activate PrP C or other toxicity-transducing receptors on the neuronal surface. Consistent with this theory, structural studies suggest that A␤ fibril ends and toxic oligomers both display ␤-strands with dangling hydrogen bonds, which are not connected to sites on adjacent strands (35,38). Oligomer-specific antibodies are capable of recognizing such structures (39,40), and perhaps PrP C may do the same. It is also interesting to speculate that PrP C , which is a relatively abundant and widely distributed cell-surface protein on both neurons and glia, might influence the polymerization of A␤ within the brain, inhibiting elongation and contributing to the accumulation of soluble, neurotoxic assemblies.
It has been proposed that small-molecule ligands that bind to PrP C and prevent interaction with A␤ oligomers could represent useful therapeutic agents for treatment of AD (41). However, pharmacologically targeting the N-terminal domain of PrP C , which contains the major A␤-binding sites, is problematic, because this region is flexibly disordered, and does not present well-defined pockets for binding small molecules. In addition, ligands that interact with the N-terminal domain could produce adverse side effects, because this region has been shown to play a role in certain physiological activities of PrP C , such as neuronal development and cell adhesion (42,43). Our data raise the possibility that small-molecule ligands for the globular C-terminal domain of PrP, which is in principle more druggable, may specifically antagonize secondary interactions

The prion protein prevents amyloid-␤ fibril elongation
with oligomeric or fibrillar A␤. If so, these ligands might provide superior tools for preventing A␤ neurotoxicity.

Preparation of A␤ monomers and ADDLs
Lyophilized A␤ 1-42 was purchased from the ERI Amyloid Laboratory, LLC (Oxford, CT). The peptide was solubilized in water, and one volumetric equivalent acetonitrile was added as a cryoprotectant before the solubilized peptide was separated into 1-mg aliquots, lyophilized, and stored at Ϫ80°C until use. For monomer preparation, the peptide was solubilized in 15 mM NaOH as described previously (44) without the addition of TCEP. Monomers were isolated by size-exclusion chromatography on a Superdex 75 10/300 GL (GE Healthcare) column using PBS as the running buffer. Fractions were collected and kept on ice for immediate use in ThT assays or were flashfrozen in liquid nitrogen and stored at Ϫ80°C until needed for EM or binding studies. The concentration of A␤ was estimated with a NanoDrop UV-visible spectrometer (Thermo Scientific) by reading the sample absorbance at 214 nm and applying Beer's law with an extinction coefficient of 76,848 M Ϫ1 cm Ϫ1 . ADDLs were prepared using a standard protocol (45,46) in which lyophilized A␤ peptide was solubilized in HFIP and dried to a film. The film was then solubilized in DMSO before dilution to a concentration of 100 M in Ham's F-12 phenol redfree medium (total DMSO 2% (v/v)), followed by incubation at room temperature for 16 h. For seeding assays, A␤ monomers were incubated for 16 h in PBS at 37°C, with or without 500 nM PrP. Fibril seeds were sonicated on ice for 10 min on 50% duty cycle before use.

ThT assay for A␤ polymerization
Kinetic assays for A␤ polymerization were conducted as described previously (22,29). Briefly, monomers of A␤ were diluted to a concentration of 3-10 M in PBS, and 10 M ThT was added from a stock of 1 mM. Recombinant PrP was added from a 1 mg/ml stock in water at the indicated concentrations. To follow ThT binding, 100-l samples were added to 96-well, half-volume, low-binding plates (Corning 3881), and florescence was read in a Synergy H1 multimode microplate reader (BioTek) every 2 or 6 min at 37°C (excitation 440 nm, emission 480 nm). Samples used for binding studies were removed directly from the wells and transferred to low-binding, 1.5-ml tubes, flash-frozen, and stored at Ϫ80°C until use.

Recombinant PrP
Full-length mouse PrP  and PrP(23-109)-cys sequences were synthesized by ATUM/DNA 2.0 (Newark, CA) in the vectors pJ414 and pJ411, respectively, using Escherichia coli-optimized codons, and were then subcloned into the pET101 vector using the Champion pET101 Directional TOPO expression kit (Invitrogen). The deletion variants were generated by site-directed mutagenesis using appropriate primers, and PrP(23-230) as the template. All constructs were verified by DNA sequencing. The pET101 vector was then transformed into BL21 Star chemically competent E. coli and expressed for 16 h via autoinduction (47). All constructs were expressed and purified as described previously (48), with minor modifications as follows. Cells were lysed, and inclusion bodies containing PrP(23-230), 23-230 -c-Myc, ⌬105-125, or His 6 -TEV-110 -230 were purified from the lysate. In the case of the N-terminal constructs 23-119, 23-109-cys, and 23-144, no inclusion body purification was required, and lysis buffer did not contain chaotropic agents. All constructs were then purified with an ÄKTA purification system (GE Healthcare) using a Ni 2ϩ -immobilized metal ion affinity column. For the construct 23-109-cys, 1 mM TCEP was added to all Ni 2ϩ buffers to prevent oxidation of the C-terminal cysteine. Protein was eluted from the Ni 2ϩ immobilized metal ion affinity column with 5 M guanidine HCl, 0.1 M Tris acetate, 0.1 M potassium phosphate (pH 4.5) while monitoring A 280 . Fractions spanning the elution peak were combined, and the pH was raised to 8 by titration with potassium acetate. For full-length PrP(23-230) as well as PrP⌬105-125, this was followed by storage at 4°C overnight to facilitate proper folding. For PrP(110 -230), which was expressed with a 5Ј-His 6 tag and TEV cleavage site (sequence MRGSHHHHH-HGENLYFQG), the eluent from the Ni 2ϩ column was desalted into 20 mM Tris, 20 mM KOAc, pH 8.0, and 0.1 mg of TEV protease was added. Enzymatic cleavage was allowed to proceed overnight at 4°C. The TEV protease was removed the following day with Ni 2ϩ resin, and the remaining protein was concentrated and purified using the Superdex 75 10/300 GL column in 20 mM Tris, 20 mM KOAc, pH 8.0. The protein was flash-frozen in this buffer and stored at Ϫ80°C until use. All other constructs were desalted using a HiPrep 26/10 desalting column (GE Healthcare) into 20 mM potassium acetate, pH 5.5, and then purified by reverse-phase HPLC using a C4 column (Grace/Vydac). Fractions containing the purified protein were pooled, lyophilized, and stored at Ϫ80°C for future use. Protein stocks were reconstituted in 0.2-m filtered water and quantified with a NanoDrop UV-visible spectrometer (Thermo Scientific) before use. The sequence for the 3Ј-c-Myc tag was EQKLISEEDL.

Alexa Fluor 488-labeled mouse PrP(23-109)
Alexa Fluor 488 C 5 maleimide (Thermo Fisher Scientific) was dissolved in water at a stock concentration of 1 mM. Lyophilized 23-109-cys (N1-cys) was dissolved in 20 mM MOPS, pH 7.4, and 0.5 mM TCEP to a concentration of 100 M, and Alexa Fluor 488 was added dropwise with stirring to a final ratio of 1:1, giving a final concentration of 50 M N1-cys, 500 M Alexa Fluor 488. This solution was protected from light and allowed to incubate at room temperature for 2 h on a benchtop rotator. After 2 h, 1 ml of the solution was injected into an analytical C3 column (Zorbax 300SB C3, Agilent) on an Agilent 1200 Infinity HPLC system, and the peptide peak was collected and lyophilized. Confirmation of successful linkage was made by MALDI-TOF mass spectrometry.

Analytical SEC
Analysis of monomer depletion during A␤ polymerization was performed using an Agilent Bio Sec-3 300 Å column running in PBS on an Agilent 1200 Infinity HPLC system. A␤ samples were spun down at 16,000 rpm in a benchtop centrifuge The prion protein prevents amyloid-␤ fibril elongation and filtered with a 0.2-m filter to remove insoluble aggregates before injecting 500 l onto the column.

Circular dichroism
Far-UV spectra (193-250 nm, 1-nm bandwidth) were collected at 37°C from samples of 20 M A␤ that had been polymerized for different times using a Jasco J-815 spectropolarimeter (Jasco, Inc.) with a 1-mm path length quartz cell. Raw data (in millidegrees) were converted to mean residue ellipticity using a molecular mass for A␤(1-42) of 4514.1 Da.

Electron microscopy
To prepare fibrils for imaging, samples of A␤ were spun down at 100,000 ϫ g in a TLA 55 fixed-angle rotor (Beckman Coulter) for 30 min. The initial sample volume was 900 l. 890 l of the supernatant was removed before diluting the remaining sample 1:2 in ultrapure water. The sample was applied as a 4.5-l droplet to a glow-discharged, 300-mesh copper grid and allowed to incubate for 4 min before washing 12 times with filtered, ultrapure water. The grid surface was then stained for 1 min in 2% uranyl acetate and dried for 3 min. Images were taken using a Philips CM12 120KV transmission microscope. Scale bars were added to images, and measurements of fibrils were made using ImageJ, with each reported size representing an average of 30 independent measurements.

Semidenaturing detergent-agarose gel electrophoresis
SDD-AGE of A␤ fibrils was performed as described previously (49). The gel was prepared with 1.5% agarose in Tris acetate buffer (25 mM Tris acetate, pH 8.5, 250 mM glycine, 0.1% SDS). Samples were not boiled after Laemmli loading buffer was added to avoid further aggregation. The gel was allowed to run for 4 h at 100 mV at 4°C to achieve best separation of samples. The samples were capillary-transferred overnight to a 0.2-m PVDF membrane in Tris acetate buffer with 20% MeOH. Membranes were blocked overnight in 3% (w/v) BSA in TBST before immunoprobing with 6E10 anti-A␤ mouse IgG (BioLegend) and anti-mouse IgG (Sigma).

Surface plasmon resonance (SPR)
SPR was performed using the ProteOn XPR36 protein interaction array system (Bio-Rad) as described previously (7). 9E10 anti-c-Myc antibody (Invitrogen) was immobilized on a Pro-teOn GLM sensor chip using the standard protocol for amine coupling. PrP(23-230)-c-Myc (ligand) was captured on the surface, followed by an analyte flow step with A␤ monomers, fibrils, or ADDLs (injection period 240 s, 50 l/min flow rate). Nonspecific binding interactions between A␤ and the 9E10 antibody were subtracted from the sensorgram using the Pro-teOn analysis software.

Dissociation-enhanced lanthanide fluorescence immunoassay (DELFIA)
To coat 96-well DELFIA yellow plates (PerkinElmer Life Sciences), PrP was diluted to 100 nM in 20 mM MOPS, pH 7.4, and 50 l was added to each well. The plate was incubated for 1 h while shaking at 400 rpm on a Thermomixer R incubator (Eppendorf), and then the wells were washed five times with 300 l of TBS with 0.05% Tween 20 (TBST). The plates were blocked for 1 h with 0.1% BSA in TBST, followed by another wash step. ADDLs were diluted in PBS, 50 l was added to each well, and the plate was incubated at 400 rpm for 1 h and then washed. For A␤ time points, samples at each time point were diluted in PBS, and 50 l was added to each well before incubating the plate for 10 min. Plates were washed, and anti-A␤ antibody 6E10, diluted in DELFIA assay buffer (PerkinElmer Life Sciences), was added at a concentration of 1 g/ml and incubated for 1 h, followed by another wash. Secondary antibody (DELFIA Eu-N1 anti-mouse IgG, PerkinElmer Life Sciences) was added at a concentration of 0.3 g/ml and incubated for 1 h. Plates were washed, and DELFIA enhancement solution (PerkinElmer Life Sciences) was added at 100 l/well and incubated for 15 min before time-resolved fluorescence was measured in a Synergy H1 multi-mode microplate reader (BioTek) (excitation 320 nm; emission 615 nm, 100-ns delay).

Fluorescence polarization
Fluorescently labeled PrP(23-109) was mixed with A␤ samples in PBS. Fluorescent polarization was measured on a Synergy H1 multimode microplate reader fitted with a Green FP (485/528) filter cube (BioTek). Polarization values were calculated using the Synergy Gen 5 software (BioTek).