Atomic-level differences between brain parenchymal- and cerebrovascular-seeded Aβ fibrils

Alzheimer’s disease is characterized by neuritic plaques, the main protein components of which are β-amyloid (Aβ) peptides deposited as β-sheet-rich amyloid fibrils. Cerebral Amyloid Angiopathy (CAA) consists of cerebrovascular deposits of Aβ peptides; it usually accompanies Alzheimer’s disease, though it sometimes occurs in the absence of neuritic plaques, as AD also occurs without accompanying CAA. Although neuritic plaques and vascular deposits have similar protein compositions, one of the characteristic features of amyloids is polymorphism, i.e., the ability of a single pure peptide to adopt multiple conformations in fibrils, depending on fibrillization conditions. For this reason, we asked whether the Aβ fibrils in neuritic plaques differed structurally from those in cerebral blood vessels. To address this question, we used seeding techniques, starting with amyloid-enriched material from either brain parenchyma or cerebral blood vessels (using meninges as the source). These amyloid-enriched preparations were then added to fresh, disaggregated solutions of Aβ to make replicate fibrils, as described elsewhere. Such fibrils were then studied by solid-state NMR, fiber X-ray diffraction, and other biophysical techniques. We observed chemical shift differences between parenchymal vs. vascular-seeded replicate fibrils in select sites (in particular, Ala2, Phe4, Val12, and Gln15 side chains) in two-dimensional 13C-13C correlation solid-state NMR spectra, strongly indicating structural differences at these sites. X-ray diffraction studies also indicated that vascular-seeded fibrils displayed greater order than parenchyma-seeded fibrils in the “side-chain dimension” (~ 10 Å reflection), though the “hydrogen-bond dimensions” (~ 5 Å reflection) were alike. These results indicate that the different nucleation conditions at two sites in the brain, parenchyma and blood vessels, affect the fibril products that get formed at each site, possibly leading to distinct pathophysiological outcomes.


Results
Seeding of Aβ40 fibril growth by parenchymal and vascular isolates. For two patients with AD + CAA (Patients 1, and 2), one with preclinical AD + CAA (Patient 3), and one patient with severe AD but only minimal CAA (Patient 4), brain parenchyma and cerebrovascular isolates were able to seed fibril growth from fresh solutions of Aβ40 by the criteria described in Methods, i.e., an increase in ThT fluorescence within 24 h, which is faster than unseeded control (Fig. 1), and rapid appearance of fibrils in TEM, Fig. 2)). This was the case even for meninges from Patients 3 and 4, where little or no Aβ deposition was observed histologically by immunostaining. In all instances, TEM appearance of the fibrils showed some degree of polymorphism. "Quiescent" conditions favor twisted ribbon morphology in TEM, and the use of "agitated" conditions favors striated (or "parallel") fibril morphology, in agreement with our earlier results 2,3,32 and those of others 1,47 , but the morphology is somewhat polymorphic nevertheless. This polymorphism may reflect the multiple weak interactions that occur during nucleation, of which solvent quiescence or agitation is only one factor.
Generally, we observed that Thioflavin T fluorescence reached a maximum, and then slowly declined. This pattern has been attributed to matting of fibrils, which can interfere with binding of the dye to fibrils, to selfquenching of ThT and related fluors, and to complexities of multiple aggregation products [42][43][44][45] . Fibrils obtained through successful seeding were examined by SSNMR (patients 1, 2, and 3) and XRD (the same three patients, and patient 4). In most cases, occipital lobe isolates were more effective at seeding than frontal lobe isolates. In addition to the above four patients, we attempted seeding using material from two control subjects (young age, no pathological or clinical evidence of AD or CAA). As expected, this material did not seed fibril formation by Aβ40 solutions and was not further examined.  In all cases, brain parenchyma refers to occipital lobe ("o"). Red circles = parenchyma seeding; blue circles = meninges (vascular) seeding; black squares = unseeded control performed at the same time as the seeding. Error bars are standard error of the mean of determinations from three samples, except for 1o and 4o, which were determinations from single samples. www.nature.com/scientificreports/ vascular-seeded Aβ fibrils. The major differences were found mainly in the position of peaks associated with Ala2 ( Fig. 5A), Phe4 (Fig. 5B) and Gln15 (Fig. 5C). Val12 and Leu34 peaks showed smaller chemical shift differences. The distinction between parenchyma-and vascular-seeded fibrils for Patients 1 and 2 was more obvious than for those of Patient 3, a point that is further discussed in Supplementary Information. In most cases, two broad peaks were observed at A2 for the Cβ atoms of brain parenchyma seeded samples, but only a single peak (Patients 2 and 3) or two peaks, one much stronger than the other (Patient 1), was observed for vascular-seeded samples. Cα and Cβ peaks for F4 and Q15 were also broad and of lower intensity than other peaks. Figure 6 and Supplementary Fig. 4 compare 13 C chemical shifts from CC2D spectra for parenchyma-and vascular-seeded samples. For six atoms (Fig. 6) mean chemical shifts for the vascular-and parenchyma-seeded samples differed significantly (from t-test, p < 0.01 for 3 atoms, p < 0.05 for 3 atoms). Although other atoms showed some differences, these were not considered statistically significant (from the t-test, p ≥ 0.05, Supplementary Fig. 4). In addition, we calculated secondary shifts (Δδ = δ fibril -δ coil ) as these strongly correlated with backbone conformation 46 , typically negative for 13 CO and 13 Cα, positive for 13 Cβ (Fig. 7). None of the six residues studied showed all three chemical shifts consistent with β-sheet structure, though for all of them, one or more atoms showed a secondary shift suggestive of β-sheet structure. (Note that for Gly residues in general 13 Cα secondary shifts are not correlated with secondary structure 47,48 ). In general, with a few exceptions (indicated in the figure), this pattern was more similar to that observed for brain-seeded fibrils by Lu et al. 3 than those observed for all-synthetic fibrils by Petkova et al. 1,48 or Paravastu et al. 47 .
Comparison of X-ray diffraction of brain parenchyma-and vascular-seeded Aβ40. X-ray diffraction of purely synthetic Aβ fibrils yielded peaks at 4.6-4.7 and 9.4-9.8 Å as expected for amyloid fibrils with a β-sheet structure. The calculated D-spacing values were similar for Aβ40 and Aβ42 fibrils, though the latter values may have been slightly lower (4.7 and 9.9 Å for Aβ40 and 4.6 and 9.6 Å for Aβ42). The peak at ~ 9.5-12 Å was slightly broader for Aβ40 than for Aβ42.
The main distinction between brain parenchyma-and vascular-seeded samples was the intensity of the broad peak at 9.5-12 Å. Some of the vascular-seeded samples (especially those from Patients 2 and 4) showed distinctly higher signal intensity at the ~ 10 Å peak than the sample seeded from the same patient's brain parenchyma. This is interpretable as greater order in packing between the β-sheets (i.e., the "side-chain dimension" of the cross-β structure).
Aβ42 fibrils are inefficient seeds for Aβ40 aggregation. A hypothetical explanation for the differences between fibrils seeded by brain parenchyma or blood vessels is that the Aβ42/Aβ40 ratio differs in Figure 2. Transmission electron microscopy images of brain parenchyma-and vascular-seeded Aβ40 fibrils. o = brain parenchyma (occipital) -seeded; m = vascular-seeded. Numbers are patient numbers (same as for Fig. 1). In all cases, magnification was 49,000 × magnification (with additional 1.4 × magnification from CCD camera). Scale bar indicates 100 nm. www.nature.com/scientificreports/ parenchyma and vascular Aβ deposits. It is possible, for example, that seeding from parenchyma represented a predominance of seeding by Aβ42 fibrils, while seeding with cerebrovascular amyloid could reflect a dominance of Aβ40 in the fibrils. (This point is discussed in more detail, below.) For this reason we tested the relative abilities of pre-formed Aβ40 and Aβ42 to seed fibril formation in solutions of Aβ40 ( Supplementary Fig. 5A).
Although pre-formed Aβ42 fibrils possess some ability to accelerate fibril formation by Aβ40, they are inefficient compared with pre-formed Aβ40 fibrils, as others have observed 49,50 . Furthermore, transmission electron Collagen in solution or as fibrils has only minor effects on Aβ40 aggregation. One important difference between the environment of blood vessels and that of brain parenchyma is that the former is rich in types I, IV and VI collagen, among others, while the latter is notably lacking in them. (This point is discussed in more detail in Supplementary Information.) For this reason, we explored the previously untested possibility that type I collagen, the most abundant form, could play a role in the differences between vascular-and parenchymaseeded Aβ40 fibrils. Type I collagen (rat tail tendon) was added to solutions of Aβ40, either as soluble protein or as pre-formed fibrils. Thioflavin T fluorescence assays ( Supplementary Fig. 5E) showed that the addition of collagen had at most only minor effects on the rate of fibril formation. Transmission electron microscopy images of brain parenchymal isolates used for seeding often contain a few endogenous collagen fibrils. For this reason, we also assessed the effects of collagenase treatment on the kinetics of Aβ40 fibril formation. Treatment of these isolates by collagenase had at most only minor effects on fibril formation kinetics, as shown by Thioflavin T fluorescence assays ( Supplementary Fig. 5F), strengthening the possibility that the structural differences of vascular-and parenchyma-seeded Aβ40 fibrils reflect those of the Aβ amyloids in the seeds.

Discussion
One of the defining features of amyloids is structural polymorphism  . Polymorphic forms of Aβ fibrils, furthermore, can be propagated as "replicate fibrils" by adding fibril seeds to fresh solutions of synthetic, and for NMR, isotopically labeled Aβ peptide. It has been shown that unseeded Aβ40 fibrils differ from brain-seeded Aβ40 fibrils in TEM appearance and SSNMR spectra 2,3,31,32 . There is a correlation between the SSNMR spectra of fibrils seeded from brain amyloid and the clinical subtype of Alzheimer's disease from which the patient suffered 31 . In particular, SSNMR spectra of fibrils seeded from amyloid patients with the rapidly progressive form of Alzheimer's disease differed from those from other types (the typical prolonged-duration form and posterior cortical atrophy variant) of AD. That is, there appears to be a "structure-malfunction" relationship between the SSNMR spectra, reflecting seeded Aβ40 fibril structure, and clinical phenotype of AD. It is likely that the seeded fibril structure represents a surrogate for the main neurotoxic species present in the brains of these patients. www.nature.com/scientificreports/ In this paper, we presented data suggesting that structural differences exist between fibrils seeded from brain parenchyma of patients with AD and those seeded from a cerebrovascular source of Aβ fibrils, the meninges. AD and cerebrovascular angiopathy (CAA) are conditions that affect partially overlapping sets of patients. Although the majority of patients with AD also have some level of CAA, and vice-versa, some patients have only, or predominantly one of these conditions. Of our patients, some (Patients 1, 2, and 4) had both conditions, of which two had severe CAA 2,4 ; another had milder disease. We obtained brain parenchymal and vascular amyloid-seeded Aβ fibrils for SSNMR from three patients, and for X-ray diffraction studies from a fourth. For the two control patients, having no histological evidence of AD or CAA, and for other such patients reported previously 2,3 , we saw no seeding.
The most notable finding from SSNMR was distinct patterns of chemical shifts in 2D-13 C-13 C SSNMR spectra of brain parenchyma-and cerebrovascular-seeded Aβ40 fibrils. This polymorphism was most apparent in Ala2, Phe4 and Gln15 Peaks in seeded material from Patients 1, 2, and 3. Of these, patients 1 and 2 had pathological diagnoses of AD and CAA. Patient 3 had only AD (no Aβ immunostaining was observed in blood vessels); nevertheless, sufficient seeding was possible for SSNMR and X-ray diffraction to be carried out. In these three patients, the same pattern of chemical shift differences was observed. The N-terminal domain has been shown previously to be an important site of polymorphism in Aβ fibrils. Future studies will include a comparison of this portion of the molecule with others: the two hydrophobic domains (residues 17-21, residues 30-40/42), the "bend region" (residues 22-29). www.nature.com/scientificreports/ X-ray diffraction studies of seeded Aβ fibrils from these same three patients also suggested differences between parenchymal and cerebrovascular seeds. Vascular-seeded samples tended to show greater signal intensity, which is interpretable as greater order in the D ~ 10 Å peak compared with brain parenchyma-seeded samples. The same samples also showed small differences in the position of this peak maximum.
For some patients 3 , a single predominant set of peaks is observed in SSNMR spectra. This is not always the case, however, as we showed in earlier results 2 , and has been shown by others more recently 51 . In the present study, at least three distinct sets of peaks exist: at least two polymorphs for vascular-seeded samples (as shown, for example, by two peaks for Cα and Cβ of Ala2), and one or two peaks for parenchyma-seeded samples, at different chemical shifts than those observed for vascular-seeded samples.
Material from meninges of Patients 3 and 4 supported seeding, despite only focal (Patient 3) or minimal (Patient 4) histologically clear, immunostainable Aβ in the blood vessels. Patient 4, however, had severe AD. We Figure 6. Analysis of chemical shifts of brain parenchyma-vs, vascular-seeded Aβ40 fibrils. A) Mean values for chemical shifts for three brain parenchyma-seeded (red) or vascular-seeded (blue) samples, using tissue from Patients 1, 2, and 3. Points represent means of these values, with standard deviations. For three of the peaks (*), parenchyma-and meninges-seeded samples were different by t-test to p < .01; for the other three peaks (**), parenchyma-and vascular-seeded samples were different by t-test to p < .05. Figure 7. Secondary chemical shifts (Δδ = δ fibril -δ coil ) for brain parenchyma and cerebrovascular-seeded fibrils. Also shown in the figure are values for secondary chemical shifts in brain-seeded Aβ40 samples reported by Lu et al. 3 . Values represent the mean chemical shifts for the spectra of Aβ40 fibrils seeded by material from Patients 1, 2, and 3. * indicates that all three samples gave chemical shifts typical of β-sheets; ** indicates one or two of the samples gave a value typical of β-sheets, i.e., negative for 13 CO and 13 Cα, positive for 13 Cβ. (Note that for Gly residues in general 13 Cα secondary shifts are not correlated with secondary structure 46 www.nature.com/scientificreports/ considered the possibility that meninges from these patients were "contaminated" by parenchyma. This possibility is unlikely at the outset, since care is taken to free meninges from any adherent parenchyma, but in addition, it seem unlikely since the SSNMR spectra (Patient 3) and X-ray diffraction pattern (Patient 4) of the seeded fibrils were similar to that of meninges-seeded fibrils obtained for the other patients. Furthermore, while parenchyma of course contains blood vessels, this component is minor, as shown by the fact that there is mutual exclusion of chemical shifts between parenchymal and meningeal samples. Thus, the most likely conclusion is that these shifts are tissue-specific. A possible concern in these studies is whether differences between parenchymal and vascular seeding are due to differences in Aβ fibril structure per se, or another component in the homogenate, which either accelerates Aβ aggregation or increases ThT fluorescences. One important control in the present study, as in previous ones 2, 3 , is the comparison of seeding by brain extracts of patients (with AD or CAA) with those without these diseases -i.e., brains from individuals with no neurological abnormalities ante-mortem or at autopsy. In addition, given the striking inefficiency of even Aβ42 to cross-seed Aβ40 solutions, the possibility of seeding from something other than Aβ fibrils seems somewhat unlikely. Nevertheless, an important caveat in these studies remains that we are studying the ability of amyloid-enriched brain extracts, not pure amyloids, to catalyze Aβ40 fibril formation.
An additional concern in these studies is the fidelity with which "daughter" or "granddaughter" fibrils replicate the parental fibrils. In the current studies, whenever possible, first generation fibrils were used, but this was not always possible. It is important to bear in mind that in the initial studies done by Petkova et al. 1 , differences in NMR spectra between different types of purely synthetic fibrils (i.e., not brain-seeded) could be passed at least three generations. Where there are mixtures of fibril types, however, as is surely the case in our studies, it is possible that some relatively non-abundant fibril type will out-compete the others and become dominant. Nonetheless, Figure 8. Fiber X-ray diffraction patterns for brain parenchyma-(red) and meninges-seeded (blue) Aβ40 fibrils. These show also showed reflections at 4.6-4.7 and 9.2-11.6 Å common to amyloids. www.nature.com/scientificreports/ even if that did occur, the fact would remain that difference exist beween parenchymal-and cerebrovascular amyloid preparations, and these differences are reflected in NMR spectra and other features of the seeded fibrils.
In conclusion, our data suggest a fundamental difference between Aβ aggregates (in this case, fibrils) of blood vessels and brain parenchyma in the three individuals studied in detail. As can be stated generally for seeding experiments of the types performed here, polymorphic differences in NMR spectra likely arise mainly during the nucleation phase of fibrillogenesis, i.e., because seeding bypasses the nucleation phase. Thus, our data suggest that Aβ fibrils formed in the presence of parenchymal amyloid differ from those formed in the presence of vascular amyloid, and these reflect, in turn, the different nucleation environments in parenchyma and blood vessels.
The types of fibril formed in brain parenchyma and cerebral blood vessels could depend to some extent on the heterogeneous nucleators to be found at each site. Among the known heterogeneous nucleators of Aβ aggregation are lipids (especially gangliosides and cholesterol), divalent metal ions, proteins such as collagen, and glycosaminoglycans (reviewed in 52). As we have pointed out 52 , these nucleators are not pure catalysts in the strict sense, because they can influence not only the rate of the reaction, but also the final products of the reaction. In the case of Aβ, our data suggest that this could include the structure of fibrils formed at these two sites.
It is likely that some or all of these would differ between brain parenchyma and blood vessels. We considered two possible differences between the environments in which Aβ aggregates in parenchyma and cerebral blood vessels: the relative contents of different Aβ isoforms, and the possible effects of extracellular matrix, in particular, type I collagen, on Aβ aggregation.
In previous papers, we presented data indicating that most of the seeding from "brain amyloid extract" is actually due to Aβ40, not Aβ42. For example, we previously noted 3 that polymorphic Aβ42 fibrils prepared in vitro do not seed the growth of Aβ40 fibrils, and concluded that brain-seeded Aβ40 fibrils most likely arose from Aβ40 fibrils (not Aβ42 fibrils) in the brain tissue. This was a point that we considered important to test formally herein. Thus, we presented data (Supporting Fig. 5A-D) that although Aβ42 was able to seed fibril formation from Aβ40 solutions, it did so very inefficiently, i.e., at high seed concentrations. Such high concentrations of Aβ42 might pertain to parenchymal neuritic plaques, and less likely to cerebrovascular Aβ deposits.
As to the effects of extracellular matrix, we considered the possibility, previously untested, that addition of Type I collagen, either in solution or as pre-formed fibrils, could affect Aβ40 aggregation rate (ThT fluorescence), but we saw no effect of these additions. Similarly, addition of collagenase to brain or meninges preparations did not affect these assays. Although these are assays of rate, not fibril structure, they suggest that differences between vascular-and parenchyma-seeded Aβ40 fibrils might be unrelated to differences in type I collagen contents of these two environments.
Our results add to the growing evidence that Aβ fibrils are not structurally homogeneous. One of the important goals in understanding and helping patients with dementia is to distinguish between Alzheimer's disease and other conditions, including cerebral amyloid angiopathy. Accordingly, diagnostic and therapeutic agents that target amyloid fibrils and their precursors will need to take fibril polymorphism in general, and the distinction between vascular and parenchymal Aβ aggregates in particular, into account.

Materials and methods
Peptide synthesis. Wildtype, unlabeled Aβ40 and Aβ42 peptides (DAEFR 5 HDSGY 10 EVHHQ 15 KLVFF 20 AEDVG 25 SNKGA 30 IIGLM 35 VGGVV 40 IA) were synthesized using standard FMOC synthesis procedures on an Applied Biosystems 433A synthesizer, essentially as described previously 53 . (Aβ(1-40) and Aβ  are henceforth referred to as Aβ40 and Aβ42, respectively.) Acetic anhydride capping was employed after addition of each residue; residues R5, S8, V12, V18, V24, and N27 of full-length Aβ were double-coupled. In some cases, the iso-acyl strategy of 54 was used (see below). Peptide was cleaved from resin with 9.45 mL TFA, 0.25 mL H 2 O, 0.25 mL EDT, and 0.1 mL TIPS for 0.25 mmol peptide. This mixture was added to resin in an ice bath; after 5 min, cleavage was continued for 2 h, room temperature. After filtration to remove resin, peptide was triturated by addition of cold diethyl ether. Peptide was dissolved in a small volume of 50/50/0.1 = Water/acetonitrile/TFA (v/v/v) and lyophilized.
Synthesis of Aβ containing Site-specific 13 C and/or 15 N Labels. An Aβ40 peptide with multiple 13 C and 15 N labels was synthesized, having uniform 15 N and 13 C labels at A2, F4, G9, V12, Q15, and L34. Labeled amino acids were from Cambridge Isotope Laboratories. Amino acids were protected using BOC-ON (Sigma) as follows: amino acids were dissolved in 1/1 = dioxane/water (v/v, 6 mL/mmol of amino acid), and stirred for 3 h at 22 °C. After the reaction, H 2 O and ethyl acetate (EtOAc) were added; the aqueous layer was retained and washed with EtOAc. After acidifying the aqueous layer to pH ~ 2.5 (ice-cold 1 M HCl), the product was extracted into EtOAc and dried, first under N 2 , followed by vacuum. Reaction products were verified by ESI-MS.
Synthesis of labeled peptides on either 0.15 or 0.25 mmol scale proceeded using a mixture of tBOC and FMOC strategies. The first amino acid was pre-loaded on the resin (Boc-Val-PAM resin); synthesis then continued using FMOC chemistry up to the point of adding a labeled residue. For tBoc synthesis, a coupling solution (made fresh daily) of 0.5 M HBTU (stored at 4 °C) in DMF and a capping solution of Ac 2 O/DIPEA/DMF (2:1:17) were used. Peptide was deprotected with 2 × 1 min washes with TFA, followed by rinsing with DMF. HBTU in DMF (0.975 eq) was added to the protected amino acid (1 eq) and dissolved by vortexing. DIPEA (1.30 eq) was added to the resin until a color change occurred, after which the amino acid was added and allowed to react. After 10 min, additional DIPEA (1.30 eq) was added. After an additional 10 min, the resin was washed with DMF. Capping solution was added and allowed to react for 10 min, and the resin was washed with DMF again. After addition of the tBOC-protected residues, the peptide was extended using FMOC chemistry. After the synthesis, the N-terminus was deprotected, the resin was washed with DCM and dried in a vacuum dessicator. For cleavage, peptide was mixed with p-cresol and anhydrous HF (1/1/10 = w/v/v), with stirring at 0 °C for 1 h, after which HF www.nature.com/scientificreports/ was evaporated. The reaction vessel was rinsed with TFA to collect the peptide, and filtration and precipitation with Et 2 O were as described above.

Synthesis and conversion of iso-acyl bond in Aβ40 and Aβ42.
In some cases, Aβ40 and Aβ42 were synthesized using the dipeptide Boc-Ser(Fmoc-Gly)-OH (Novabiochem) for G25-S26 54 . Standard Fmoc synthesis was carried out from residue 40 (or 42) to N27. The dipeptide and HOBt (both 1 mmol) were dissolved in 2.5 mL of 3/1 = DCM/DMF (v/v), after which DIC (1.1 mmol) was added. After 10 min, the activated dipeptide formed; since the solution had formed a slurry, it was filtered through a syringe filter. The activated dipeptide was then added to the reaction vessel containing the resin and allowed to react with the deprotected peptide for 60 min. The reaction vessel was vortexed for 60 min and vented every 10 min. Standard FMOC synthesis was resumed, beginning with V24 and continuing to the N-terminus. The usual cleavage conditions were used. Following RP-HPLC purification and lyophilization, the iso-acyl bond was converted to a peptide bond by dissolving it in 1 M NH 4 HCO 3 (to ~ pH 7) for 2 h, after which the Aβ peptides were re-purified by RP-HPLC.
Thioflavin T, transmission electron microscopy. Fibril formation was monitored using standard Thioflavin T fluorescence assays 55 using a Hitachi F-2000 fluorescence spectrophotometer (λ ex 446 nm, λ em = 490 nm, averaged over 10 s, three measurements per sample. To estimate lag periods, measured values were fitted to the equation of a stretched exponential equation, were T 0 = fluorescence at time 0, T eq = fluorescence at infinite time, t = time, k is a rate constant (s −1 ) and n is a dimensionless parameter. Fibrils were examined by transmission electron microscopy by previously described procedures 53 .
Isolation of brain parenchymal or vascular amyloid. Frozen, unfixed brain samples were obtained from The University of Chicago Medical Center at the time of autopsy. Brain parenchymal amyloid was isolated essentially as described previously 2,3,32 , using an isolation procedure modified from 56 .
For isolation of vascular amyloid, the procedure was modified as follows. Tissue was dissected to separate meninges and other separable vessels from brain parenchyma. Tissue was homogenized as described 2,3,32 in 20 volumes of cold DAB-S buffer (10 mM Tris pH 7.4, 0.25 M sucrose, 3 mM EDTA, 50 µg/ml gentamicin, 0.25 µg/ ml amphotericin B) in a Tenbroek homogenizer, stirred overnight at 4° C, and then subjected to sequential sucrose density gradient centrifugation (at 1.2 M, then 1.9 M sucrose) as described. In initial experiments, it was observed that for parenchymal samples, the upper layer obtained from the last sucrose density gradient step had the bulk of the seeding activity, but for meningeal samples, the pellet contained greater seeding ability. Presumably, this reflects the greater lipid content of the parenchymal sample. Thus, for meningeal seeding, the pellet rather than the supernatant was retained for further processing. The samples were otherwise treated similarly, i.e., amyloid material was further purified by collagenase treatment, brief extraction using SDS-containing buffers, and centrifugation steps as described 32 . The final pellet isolate was then briefly sonicated (~ 15 s at output 3 and duty 50%, in an ice bath, with longer times and increased power for some meningeal samples) to disperse the brain material prior to quantification. Total protein was measured using the BCA assay (Pierce) and Aβ content was assessed by CNBr digestion and RP-HPLC, as described 32 .
Formation of "Replicate" fibrils from amyloid-enriched brain extracts. Since the percentage of the protein consisting of Aβ peptides was not known, as in previous studies 2,3,32 seeding was demonstrated empirically through elimination or major foreshortening of the lag period in ThT fluorescence curves. In general, this required ~ 5 mg of protein (Bradford assay) for 1 mg of synthetic peptide; in some cases, this led to sufficient seeding for SSNMR in the first generation, though in some cases, a second generation of seeding was necessary. Aβ peptides constitute < 1% of the total protein, as we have stated in previous publications 2, 3,47 . Whenever possible, first generation fibrils were used. This was not always possible. Previous studies have shown, however, that there is fidelity in TEM images and SS-NMR spectra, and that the fibrils are indeed replicates of the previous generation. The initial studies done by Petkova et al. 1 showed that differences in NMR spectra between different types of purely synthetic fibrils (i.e., not brain-seeded) could be passed at least three generations. Unless otherwise stated, transmission electron micrographs, X-ray diffraction patterns and SSNMR spectra refer to second generation fibrils.
To assess the presence of seeding, an "unseeded control" (Fig. 1 and elsewhere) was produced on solutions of Aβ40 that were allowed to fibrillize under "quiescent" conditions. This may be a more accurate reflection of conditions that might occur in vivo, though the rate of fibrillization is slower than under "agitated" conditions. An additional caveat is that this process is exquisitely sensitive to minor variations in reaction temperature over the course of many days, and other variables such as pH, ionic strength, DMSO concentration -in spite of attempts to control these variables. This concern is mitigated, however, because in all cases, the same conditions (buffer, temperature, etc.) were used for both seeded and unseeded reactions, so that they could legitimately be compared. Furthermore, our emphasis was to demonstrate that we had achieved seeding, rather than to make a detailed study of the kinetics of unseeded fibrillization. Nevertheless, we narrowed down the main source of variability to the DMSO concentration. This is discussed further in Supporting Information (Supporting Fig. 6). Briefly, while variations in DMSO concentration affect the kinetics of the reaction, even much large variations in DMSO concentration have very minor effect on 1 H chemical shifts (≤ 0.008 ppm) and no observed effects on 13 C chemical shifts. We conclude, therefore, that the small variations DMSO concentration in our control www.nature.com/scientificreports/ experiments, while affecting kenetics of fibrillization of unseeded samples, did not affect the chemical shifts reported for seeded fibrils elsewhere in this paper.
Collagen and collagenase. As stated and described previously 32 , tissue routinely was treated with collagenase (0.2 mg/mL collagenase CLS3 (Worthington), in 50 mM Tris pH 8.0, 2 mM CaCl 2 ) at the same time as it was treated with DNase. In some experiments described below, the effects of collagenase on Aβ40 fibrillization kinetics were assayed by Thioflavin T fluorescence measurements. To study the effects of soluble collagen on Aβ40 fibrillization kinetics, rat tail tendon collagen was isolated by extraction using dilute acetic acid as described in 57 .
Solid-state NMR. NMR measurements were performed at the UIC Center for Structural Biology (at UIC) for all the specimens except for Specimen 1 m, and at the NIH, Laboratory of Chemical Physics for Specimen 1 m. For Specimen 1 m, NMR measurements were performed at 14.1 T (599.2 MHz proton NMR frequency), using a 1.8 mm MAS NMR probe from the research group of Dr. Ago Samoson (Tallinn University of Technology). The 2D 13 C-13 C spectrum was obtained with MAS at 13.6 kHz, using a 50 ms 13 C-13 C spin diffusion mixing period and 100 kHz proton decoupling with TPPM. The recycle delay was 1.0 s. The maximum t1 period was 4.47 ms. The total measurement time was 85 h.
For all other specimens, SSNMR experiments were conducted at a Bruker-Avance III 750 MHz spectrometer using a 1.3 mm MAS triple-resonance probe. The spinning speed was set to 50,000 ± 3 Hz and the cooling air temperature was set at -15 C, resulting in a sample temperature of ~ 21° C. The fpRFDR pulse sequence was employed to obtain 2D 13 C-13 C spectra 58,59 . During the cross-polarization period, the 13 C RF field amplitude was linearly swept from 25 to 41 kHz during a contact time of 2.0 ms while the 1 H RF amplitude was kept constant at 15 kHz. During the mixing period, fpRFDR 13 C-13 C dipolar recoupling sequence with a mixing time of 1.92 ms and a 13 C π-pulse width of 8 µs was used. Although this is shorter than many such pulses appearing in the literature where slower spinning speeds were used, the rationale for using it is the following (see also 59 ). The optimal fpRFDR recoupling condition is achieved when the π-pulse width is 1/3-1/2 of the rotor period. In our case, the MAS spinning speed was 50 kHz which corresponds to 20 µs rotor period; thus, the π-pulse width should be between 6-10 µs, of which 8 µs is the median. 1 H SPINAL-64 decoupling of 10 kHz was employed during the t 1 and t 2 periods. The total measurement time was ranged from 26 to 98 h.
Secondary chemical shifts were calculated by subtracting the random coil values for amino acids found in 60 . (An additional 1.7 ppm was subtracted from the random coil values to account for the use of DSS rather than TMS as a reference.) Chemical shifts were compared with those for unseeded Aβ fibrils 1,3,47 .
X-ray diffraction. X-ray diffraction studies on Aβ fibrils were conducted essentially as described previously [61][62][63][64] . Briefly, Aβ fibrils in 0.05 M sodium phosphate, pH 7.40, were transferred to SpectraPor 3,500 MWCO dialysis tubing and dialyzed against MilliQ H 2 O (typically, 4 × 2 L H 2 O over 6 h for 8 mL of sample). Samples were then transferred to microfuge tubes and centrifuged 3 min at 9,300 rpm and 1 min at 13,000 rpm. The majority of the supernatant was removed and the fibrils resuspended in the remaining fluid (≤ 50 mL). The fibril slurry was drawn into Charles Supper 1 mm quartz capillary tubes, which were then capped at one end with beeswax. To orient fibrils, the capillaries were placed in an NMR tube covered with tissue paper and left for 6-7 days in a 600 MHz instrument at 37° C. Capillary tubes were removed from the NMR tube and allowed to dry, using paper wicks and elevated temperature as necessary. The tubes then were capped at the other end with beeswax. For samples containing brain isolate, both ends were dotted with epoxy glue and the capillary tubes were enclosed between sheets 2.5 mm mylar film, secured with epoxy glue.
Synchrotron data were collected at Argonne National Laboratory, using the Advanced Photon Source beamline BioCAT with a CCD-based X-ray detector (PCCD 168,080/Aviex LLC). The beam was focused to ~ 10 × 20 mm using compound refractive lenses. Exposure times from 1-60 s were tested; exposures between 5 and 30 s were usually appropriate. Data were analyzed using FIT2D (http://www.esrf.eu/compu ting/scien tific /FIT2D /). Background images of empty regions of the same sample-containing capillary or of an empty capillary, collected for the same exposure times as sample data, were subtracted from sample images. 1-D and 2-D azimuthal scans were prepared using the ' cake' function. An angle of > 300 was analyzed except when an imperfect background subtraction occurred, in which case as large an angle as possible (> 180°) was used. When multiple images were available to be analyzed for a given capillary tube, only the results of the single image with lowest background were used in statistical analysis. Position of peaks was determined from the azimuthal scans, which had a 0.13-0.14 Å step between points.