Structure-specific amyloid precipitation in biofluids

The composition of soluble toxic protein aggregates formed in vivo is currently unknown in neurodegenerative diseases, due to their ultra-low concentration in human biofluids and their high degree of heterogeneity. Here we report a method to capture amyloid-containing aggregates in human biofluids in an unbiased way, a process we name amyloid precipitation. We use a structure-specific chemical dimer, a Y-shaped, bio-inspired small molecule with two capture groups, for amyloid precipitation to increase affinity. Our capture molecule for amyloid precipitation (CAP-1) consists of a derivative of Pittsburgh Compound B (dimer) to target the cross β-sheets of amyloids and a biotin moiety for surface immobilization. By coupling CAP-1 to magnetic beads, we demonstrate that we can target the amyloid structure of all protein aggregates present in human cerebrospinal fluid, isolate them for analysis and then characterize them using single-molecule fluorescence imaging and mass spectrometry. Amyloid precipitation enables unbiased determination of the molecular composition and structural features of the in vivo aggregates formed in neurodegenerative diseases. The composition of toxic protein aggregates associated with neurodegenerative diseases is difficult to determine. Now, a method has been developed that can capture amyloid-containing aggregates in human biofluids using a structure-specific chemical dimer. This method—known as amyloid precipitation—enables unbiased determination of the molecular composition and structural features of the in vivo aggregates.

α -Synuclein, amyloid-β (Aβ) and tau are examples of proteins that self-aggregate in cross β-sheet motifs, and they are present in Lewy bodies, amyloid plaques and tau tangles, respectively 1 . These cross β-sheets (or amyloid structures) are found in the brains of people with neurodegenerative diseases such as Parkinson's disease (PD) and Alzheimer's disease (AD) [2][3][4] . Importantly, brain extracts containing misfolded Aβ from patients with AD and preformed α-synuclein fibrils induced cerebral β-amyloidosis and α-synuclein propagation, respectively, and associated pathologies in mice [5][6][7][8] . However, depletion of aggregates from an AD brain suppressed in vivo seeding capability 9 , reinforcing the idea that the induction of pathology is likely governed by the structure and concentration of the aggregate seeds 6,10 ; this highlights the importance of studying the aggregated protein as opposed to its monomeric counterpart.
The exact mechanism by which protein aggregates lead to progressive loss of neuronal cells and result in subsequent pathophysiologic effects like dementia and movement disorders remains poorly understood. It is known that subtle differences in amino acid content result in major structural changes that have an impact on the pathophysiology of these diseases [11][12][13][14] . Also, in vitro studies have revealed that protein aggregation is a dynamic process where a wide range of aggregates with variable sizes 1 and hydrophilicities 15 are formed, and that the aggregates become more toxic when they acquire a cross β-sheet structure 1,16 . We have previously studied the aggregation of α-synuclein in detail using super-resolution imaging and single-molecule fluorescence [15][16][17][18] . Aggregation proceeds by the formation of small soluble aggregates, which undergo slow structural conversions to small oligomeric species with increased β-sheet structure over 24 hours; these oligomeric species are cytotoxic to cells 16,17 . Aggregates with a β-sheet structure were also shown to be more effective at membrane permeabilization, leading to increased cell death 19 . Other studies showed that thioflavin T (ThT)-active aggregates of α-synuclein interact with ATP synthase and increase the probability of the opening of the permeability transition pore of mitochondria, ultimately leading to cell death 20 . Thus, ThT-active aggregates of α-synuclein are toxic to cells by a number of mechanisms. ThT-active aggregates are formed early in the aggregation reaction (from 1 hour onward) together with non-ThT-active species 15 . Both species are spherically symmetric and smaller than ~50 nm in size, and they are clearly distinct from high-aspect-ratio and longer fibrils formed at later times (24 h), which also have a different surface hydrophobicity. The fibrils are also much more highly ordered than the ThT-active species that form initially, as measured using fluorescence anisotropy, providing further evidence for a structural conversion 18 . Overall these experiments show that oligomeric non-fibrillar ThT-active species form early in the aggregation process, with distinct properties from fibrils, and are toxic to cells by a variety of mechanisms.
Moreover, it was recently reported that small soluble Aβ aggregates induced extensive membrane permeability, while larger β-sheet-containing aggregates were most effective at causing an inflammatory response in microglia cells 25 . These findings were replicated in a recent study of the aggregates present in the cerebrospinal fluid (CSF) of patients at different stages of AD 27 . The aggregates in the CSF of mildly cognitively impaired patients induced more membrane permeabilization, while larger β-sheet aggregates present in the CSF of AD patients were more effective at inducing inflammation 27 . Together these studies reinforce the idea that aggregates of different size and structure trigger different toxic mechanisms and that the relative proportion of these different aggregates changes during the development of the disease.
It is now understood that AD develops before the manifestation of clinical symptoms, so it is important to develop new diagnostic methods in readily available biofluids such as blood, urine and CSF. In particular, CSF is one of the major clearance systems and provides an accessible biofluid that can be used to assess extracellular protein aggregates. However, the protein aggregates present in CSF are at very low (subpicomolar) concentrations and they are very heterogeneous in size 28,29 . These two factors have significantly hindered the development of suitable tools to isolate and study protein aggregates from human biofluids. New methods are needed to isolate and characterize the low levels of aggregates present in human biofluids in order to better understand how compositional and structural differences in these aggregates impact cellular toxicity and contribute to disease pathogenesis. This is a fundamental step towards the development of effective therapeutic strategies and for early diagnosis of disease.
Until recently, protein aggregates implicated in neurodegeneration have largely been characterized using capture techniques based on antibodies or aptamers [30][31][32][33] . However, both antibody and aptamer capture strategies have a fundamental limitation that they only target aggregates of a selected protein, as well as having other problems such as epitope accessibility on misfolded proteins, inefficient targeting if the aggregated proteins contain post-translational modifications and difficulties in recognizing aggregates composed of oligomers formed by more than one type of protein 34,35 . To address these issues, we have developed a structure-specific chemical dimer designed to selectively bind cross β-sheet motifs instead of a specific protein epitope, allowing detection of the range of protein aggregates associated with neurodegenerative diseases 29,36 in an unbiased fashion. We have named this molecule the capture molecule for amyloid precipitation (CAP-1). Protein aggregates can be precipitated from solution by attaching CAP-1 to magnetic beads, a process that we refer to as amyloid precipitation (AP). This AP method enables an array of molecular and cellular techniques, ranging from single-molecule imaging to cytotoxicity studies, to be performed to characterize the structural and functional properties of protein aggregates.

Results
Rational design and characterization of a bio-inspired amyloidspecific probe. The design of CAP-1 was inspired by the structure of antibodies due to their natural high affinity to target specific molecules, based on their Y-shaped structure with two binding sites. This chemical molecule is designed to recognize β-sheet structures with high affinity and is based on a dimer of benzothiazole derivatives. This derivative contains structural elements of ThT for its photophysical and optical properties 36,37 , for detection of aggregate binding. It also contains structural elements of Pittsburgh Compound B (PiB) for its increased affinity to β-sheet structures compared with ThT, namely the lack of two methyl groups and charge in the benzothiazole group of PiB 38,39 , for efficient capture of β-sheet-containing aggregates. A recent study has used a similar approach to develop a multivalent positron emission tomography (PET) ligand to image Aβ aggregates in the brain 40 .
The synthesis of CAP-1 was achieved using established methods ( Fig. 1a for structure and Supplementary Figs. 1-9 for synthesis details). The trimeric species has two β-sheet binding sites for increased avidity (as previously described for a dimeric version of ThT 41 ) and has a third site for immobilization in this specific implementation, via biotin-streptavidin binding.
After initial spectral characterization of CAP-1 (excitation (λ ex ) = 355 nm and emission (λ em ) = 440 nm wavelengths ( Supplementary Fig. 10) and solubility measurements ( Supplementary Fig. 11)), we evaluated the binding of CAP-1 to α-synuclein monomers, oligomers and fibrils ( Fig. 1b-d). CAP-1 binds to oligomers as well as fibrils but not monomers (Fig. 1b-d and Supplementary Fig. 12). CAP-1, like ThT 42 , is suitable for total internal reflection fluorescence microscopy (TIRFM) and can be used to monitor the aggregation reaction, from small, early stage aggregates (typically after 4 h) to long, mature fibrils (after 24 h; Fig. 1c,d). Kinetic studies of the α-synuclein aggregation in the presence of CAP-1 or ThT showed similar changes in fluorescence, as expected ( Supplementary Fig. 13).
We selected α-synuclein as the model amyloid protein throughout this work, but we also achieved similar results using other amyloid proteins such as Aβ peptide fragment 1-42 (Aβ 42 ) and tau aggregates ( Supplementary Fig. 14), with comparable ratios of signal to background. The binding of CAP-1 to these protein aggregates further supports the specificity towards a cross β-sheet regardless of protein sequence, highlighting the value of using CAP-1 to target a 'structural epitope' . Such non-biased approach is key since the composition of in vivo aggregates remains elusive.
Finally, we determined the binding affinity of CAP-1 to α-synuclein and compared this affinity with that of ThT (Fig. 1e) using bulk fluorescence and sonicated fibrils (average length, 200 nm) to avoid heterogeneity in the structure and size of the aggregates ( Supplementary Fig. 15). Using an initial concentration of 100 nM α-synuclein (monomer equivalent), we obtained a dissociation constant K d (CAP-1) = 14 ± 5 nM and K d (ThT) = 1,400 ± 140 nM ( Supplementary Fig. 16), representing a 280-fold increase in affinity of CAP-1 compared to ThT. Dissociation constants often depend on the approach used, and previous studies reported a K d of ThT for α-synuclein fibrils from 588 nM to 100 μM (refs. 43,44 ). The increase in CAP-1 affinity towards amyloids compared to ThT can be explained by the combination of two key factors: CAP-1 being a dimer (as previously described, avidity increases affinity 41 ) and the absence of the N-methylated benzothiazole moiety (Supplementary Fig. 1) as seen for PiB 38,39 . The K d for a monovalent version of CAP-1 was 35 ± 12 nM, confirming that dimerization increased the binding affinity ( Supplementary Fig. 16). The K d for CAP-1 for Aβ 40 was 17 ± 8 nM, supporting the specificity of CAP-1 binding to cross β-sheet aggregates ( Supplementary Fig. 16). The K d of ThT for Aβ 40 was reported to be 2.3 μM (ref. 41 ), again demonstrating the increase of binding affinity of CAP-1 over ThT.

Capture of protein aggregates using CAP-1-method of AP.
Following the characterization of CAP-1 binding to α-synuclein, we designed a protocol for isolation of protein aggregates from solution, which we named amyloid precipitation (AP).
The schematic of AP is outlined in Fig. 2a,b. After the conjugation of CAP-1 with magnetic streptavidin-coated beads (Fig. 2a), the beads are added to a solution containing protein aggregates such as recombinant α-synuclein solution or a biofluid. After 2 h at 4 °C with gentle mixing, the beads are separated using a magnet, and both fractions ('beads' and 'depleted') are analysed by TIRFM (Fig. 2c) and bulk fluorescence (Fig. 2d). We found some clumping of the beads in the absence of protein, but the presence of proteins prevented clumping, allowing us to perform AP. Figure 2c shows conjugated beads with CAP-1 after AP using α-synuclein fibrils (right) or phosphate-buffered saline (PBS) (left).
The presence of fibrils (right panel) attached to the beads is visible by the 'hairy' appearance of the beads and highlighted in the magnified bead, and contrasts with the plain look of beads without protein (left panel). Despite the heterogeneous bead-to-fibril attachment-some beads contain many small fibrils, and other beads contain fewer but longer fibrils-there is a significant difference in the diameter (measured as the fluorescence intensity profile) between beads in the presence or absence of α-synuclein fibrils, 105 nm (P = 0.0002), confirming the successful binding of fibrils to beads ( Supplementary Fig. 17).
In Fig. 2d, we tested the efficacy of AP towards α-synuclein fibrils (purple) versus α-synuclein monomers (orange; Supplementary  Fig. 18 for representative fluorescent spectra of both fractions). The difference between beads incubated with α-synuclein fibrils (100%, purple) and beads with α-synuclein monomers (33%, orange) highlights the absence of a cross β-sheet in the monomeric solution and corresponds to the fluorescence of CAP-1 alone. The fluorescence intensity for the samples, beads + CAP-1 + monomers (orange) and beads + CAP-1 + PBS (grey), is the same for both beads and supernatant, confirming that CAP-1 does not bind to monomers. The low fluorescence intensity detected for the supernatant of both samples, 11% for fibrils and 3% for monomers, reflects the presence of residual CAP-1 molecules released from the beads during the incubation and as expected is higher for the sample containing fibrils. Overall, the fluorescence increase between the supernatant (11%) and beads (100%) for the α-synuclein fibril sample demonstrates the successful pulling down (and concentration) of aggregates by the beads.
In both TIRFM (Fig. 2e) and bulk ( Fig. 2d) measurements, detection of protein aggregates is based on CAP-1 intrinsic fluorescence, highlighting its ability to strongly bind (capture) aggregates In purple are the amyloid binding regions, and in blue, the biotin used for surface attachment via streptavidin binding. b, Illustrative diagram highlighting the selective affinity of CAP-1 to cross β-sheets present in early stage aggregates and fibrils but not in monomers. c, TIRFM images of α-synuclein aggregation at 0 h, 8 h (red circles highlighting oligomers) and 24 h using 5 µM CAP-1 and 2.8 µM α-synuclein; λ ex = 405 nm. Insets show the areas in the white boxes. Scale bar = 5 μm; inset scale bar = 2 μm. d, Maximum fluorescence intensity increase of CAP-1 (20 µM) upon binding to 10 µM α-synuclein at different time points of the aggregation reaction using λ ex = 355 nm. Data are presented as the mean ± s.d. of n = 2 independent experiments. One-way analysis of variance (ANOVA; P = 0.0006) and Tukey's post hoc comparisons (**P < 0.0017, ***P = 0.0007 and not significant (NS) P > 0.05). e, Binding affinity of CAP-1 and ThT to α-synuclein. Increasing amounts of CAP-1 or ThT were added to 100 nM total monomer concentration of sonicated α-synuclein fibrils. The inset shows a zoomed-in view of the shaded area. Data are presented as the mean ± s.d. of n = 3 or n = 2 independent experiments for CAP-1 or ThT, respectively. The K d for ThT was obtained by fitting the experimental points to a hyperbolic curve (specific binding); K d (ThT) = 1,400 ± 132 nM. For CAP-1, a model for binding fluorescent ligands that takes account of the change in fluorescence between bound and unbound molecule was used; K d (CAP-1) = 14 ± 5 nM. For more details, see the Supplementary Methods. and work as an optical read-out for the presence of β-sheets. We also used atomic force microscopy (AFM), an orthogonal non-optical technique, to confirm the successful binding of α-synuclein fibrils to CAP-1 beads ( Supplementary Fig. 19). As shown in the three-dimensional (3D; height) image, fibrils localize preferentially close to the beads ( Supplementary Fig. 19a), once more demonstrating the preference of protein aggregates to CAP-1-coated beads.
Until now, we have used mature α-synuclein fibrils (sonicated, 200 nm; non-sonicated, >1 μm) as a model of protein aggregation to test AP. However, in biological fluids such as CSF, the aggregates present are smaller. These 'early stage' soluble aggregates, or oligomers, have been shown to be much smaller than the optical diffraction limit (~250 nm) 42 and differ in size, shape and structure from fibrils, as seen using higher resolution methods such as spectral point accumulation for imaging in nanoscale topography (sPAINT) 15 and AFM 25 . For this reason, we used α-synuclein aggregates collected at the 8 hour time point to maximize the number of these oligomers 15,16 and to validate the AP method for use in a biological context. We used electron microscopy (EM) to characterize aggregates present at 8 h ( Supplementary Fig. 22) and confirmed their subdiffraction limit size (~30 nm). The results in Fig. 2e show the number of fluorescent puncta before and after AP: 6.0 × 10 −2 μm -2 and 3.6 × 10 −4 μm -2 , respectively (Supplementary Fig. 21 for TIRFM images). In the presence of CAP-1, the number of protein aggregates in solution after pull-down is reduced to background levels (Fig. 2e, grey column; 4.2 × 10 −4 ± 2.3 × 10 −4 μm -2 ). In the absence of CAP-1, there was partial removal of aggregates ( Supplementary Fig. 21), suggesting unspecific binding to the beads, but this removal was negligible compared to the virtually complete depletion, 99.4%, in the presence of CAP-1. Overall, these experiments demonstrate that AP can be used to capture α-synuclein oligomers.
Next, we investigated the use of mass spectrometry (MS) to quantify the amount of α-synuclein enriched on the beads after pull-down, as MS will allow identification of molecular composition of the amyloids captured using AP. For this, we used high-resolution parallel reaction monitoring MS (PRM-MS). After AP, α-synuclein was eluted from the beads and digested using trypsin, converting full-length α-synuclein into small peptides, namely (α-synuclein peptide from residue 13 to residue 21) α-syn 13-21 , α-syn [35][36][37][38][39][40][41][42][43] , α-syn 46-58 , α-syn 61-80 and α-syn 81-96 . In order to confirm the specificity of CAP-1, we compared the presence and absence of CAP-1 during the AP. In the presence of CAP-1, the amount of individual tryptic α-synuclein peptides recovered was 5 to 13 times higher, depending on the peptide, than without CAP-1 ( Supplementary Fig. 23). This agrees with TIRFM results (Fig. 2e). The PRM-MS spectrum in Fig. 2f shows the relative abundance of the α-syn 13-21 peptide fragment ion (y n ) in the fragment ion spectrum in the presence (right) and absence (left) of CAP-1.

AP followed by PRM-MS of α-synuclein spiked in human CSF.
AP is an unbiased method to capture amyloid protein from solution, allowing subsequent MS identification of proteins present in such aggregates 45,46 . As CSF is a complex biofluid made of more than two thousand different proteins 47 , we first determined the sensitivity of CAP-1 beads to capturing known amounts of α-synuclein spiked in CSF.
Increasing amounts of either purified α-synuclein monomers (t = 0 hours) or an α-synuclein mixture of monomers (>95%) and oligomers (<5% (ref. 16 ); t = 8 hours) were spiked in control CSF (Fig. 3a for the outline of the experiment and Supplementary  Fig. 25 for TIRFM representative images). After AP, the beads were trypsin-digested and analysed by PRM-MS. In Fig. 3b the amount of α-syn 13-21 peptide recovered as a function of the initial α-synuclein concentration spiked is shown. Naturally occurring α-synuclein oligomers present in CSF were undetectable (Supplementary Tables  3-8 for list of proteins pulled down). For concentrations equal to and below 1 nM, monomers were not detected, while in 1 nM of mixed species, 28 femtomoles (28 pM) of α-syn 13-21 captured were detected (Supplementary Fig. 24 for other peptides). For α-synuclein concentrations higher than 1 nM, the increase in α-syn 13-21 detected is linear and about three times higher for the mixed species sample than for the monomers (Fig. 3b). CAP-1 beads captured 0.6% of total α-synuclein monomers (orange) spiked in CSF and 2.3% of the total α-synuclein mixture (monomers >95% and oligomers <5%, purple). This means that almost no monomers in solution are captured, while approximately 50% (2.3% out of <5%) of the oligomers added to CSF are captured. For this reason, the threefold change in the total amount of α-syn 13-21 recovered (Fig. 3b) represents a large difference in capture affinity between the monomer, which is present at high concentration, and the low concentration of aggregated α-synuclein. This result confirms the specificity of AP in capturing protein aggregates compared to the monomers in complex biofluids such as CSF.
The proteins captured using CAP-1 beads should be enriched in amyloid-prone proteins or contain proteins in the CSF that bind amyloid proteins 35 . Using PASTA 2.0 (ref. 48 ) and RFAmyloid 49 , two highly cited web servers for the prediction of protein aggregation from sequence, we observed an increase in the total number of amyloid-prone proteins when using CAP-1 compared to unmodified beads (Supplementary Table 9). As expected, in the presence of CAP-1, there is an increase in the total β-strand content of captured proteins (25-26%) compared to not using the capture molecule (19%), providing further computational After incubation, beads bound to protein aggregates are isolated using a magnet. Both fractions, the depleted fraction (SN, supernatant) and enriched fraction (beads), can be analysed by bulk fluorescence and TIRFM. c, TIRFM of Dynabeads MyOne Streptavidin C1 beads conjugated with CAP-1 and in the presence (right panel) or absence (left panel) of 10 μM α-synuclein (αS) fibrils, using λ ex = 405 nm. α-Synuclein fibrils can be seen attached to the beads, creating a 'hairy' bead look (right panel inset), or, in other cases, having a single, long, thick spike. In the absence of protein aggregates, the beads have a plain look (left panel inset). Scale bars = 3 μm. d, Bulk fluorescence intensity (normalized) of beads (B) and supernatant (SN) after AP using 10 μM α-synuclein, with sonicated fibrils (5 days incubation), monomers and PBS only; λ ex = 355 nm and maximum λ em = 440 nm. SN represents the supernatant or 'depleted' fraction, and B the 'beads' fraction. The horizontal green area highlights that the fluorescence of beads + CAP-1 + monomers is the same as beads + CAP-1 without protein, supporting the idea that CAP-1 does not bind to monomers; that is, the fluorescence measured is due to CAP-1 alone. Data are presented as the mean ± s.d. of n = 2 independent experiments (each value corresponds to mean of n = 3 replicates), and differences between groups were analysed using the unpaired two-tailed Student's t-test; *P = 0.0345, **P = 0.0062 and ***P = 0.0004. e, Depletion of α-synuclein oligomers (time point 8 h of α-synuclein aggregation reaction) by AP and quantification of aggregates left in the supernatant (depleted fraction). Plotted is the fluorescent puncta counts (×10 2 μm -2 ) for the sample before and after AP using TIRFM. AP captures approximately 100% of oligomers in solution. Data are presented as the mean ± s.d. of n = 27 fields of view per sample for one representative experiment ( Supplementary Fig. 21 for TIRFM images). f, Schematic depicting AP followed by on-bead digestion. The lower panels show the α-syn 13-21 peptide fragment ion (y 2 to y 8 ) PRM spectrum in the presence (right) and absence (left) of CAP-1, recovered after AP from a solution containing 1 nM total α-synuclein (<50 pM oligomers). evidence of the ability of CAP-1 to select β-sheet-containing proteins (Supplementary Tables 1 and 2).
To evaluate the efficiency of AP in removing toxic amyloid species from CSF, we used a sensitive membrane permeability assay, developed previously 22 (Fig. 3c for outline of the experiment). CSF is diluted in a solution containing Ca 2+ ions and then added to liposomes containing a Ca 2+ -dependent dye. If CSF contains amyloids/ oligomers that cause membrane permeability, Ca 2+ ions enters the liposome, resulting in increased fluorescence. The increase in signal when the concentration of Ca 2+ ions equals the bath concentration is determined at the end of the experiment by adding ionomycin. This corresponds to 100% Ca 2+ ion entry and means that the measurement is quantitative with a scale from 0-100%. In Fig. 3d is plotted the average Ca 2+ influxes for CSF before AP (purple), CSF after AP ('depleted' fraction; white) and CSF after AP without CAP-1 (white with purple dots) using the same CSF as in Fig. 3b. AP removed most of the CSF proteins responsible for Ca 2+ influx, reducing membrane permeability from 27% to 6%. There is some depletion in the absence of CAP-1 due to non-specific binding to the beads [50][51][52] . Having established that AP is able to remove amyloid proteins from control CSF (Fig. 3b-d), we then decided to use CSF from PD patients in a separate set of experiments (Fig. 3e,f). TIRFM images showed a substantial decrease in the number of ThT-active species after AP (Fig. 3f), and we found a reduction of ~50% in Ca 2+ influx (Fig. 3e). This demonstrates that AP can capture amyloid aggregates from PD CSF. A non-specific binding to the beads without CAP-1 leads to some aggregate capture and a small reduction in membrane permeability (Fig. 3d,e, middle column).

Conclusions
Protein aggregates have been known to be implicated in neurodegenerative diseases for more than three decades 53 . Yet, despite much progress, there are still important technological limitations in isolating and characterizing the intermediate small species that are formed during the development of disease. This is due to the low abundance, small size and heterogeneity in conformation and composition of the aggregates [54][55][56][57] . Traditional immunocapture or immune recognition approaches have improved in being able to target misfolded or aggregated proteins but are not capable of distinguishing between aggregates of different structures that may have very different properties and toxicities 58 . The goal of this work is to design an unbiased method to capture and characterize all the aggregates with a cross β-sheet structure that are present in human biofluids to determine their composition.
In this study, we presented the synthesis and characterization of a structure-specific chemical dimer designed to capture from solution the protein aggregates associated with neurodegeneration. This molecule has been specifically developed to bind and isolate a target molecule based on secondary structure (the presence of β-sheets) using chemical head groups that form the basis of PET ligands 59,60 . Previous studies have made use of dimerized ligands (protein/peptide 61,62 or ThT 41 ) as a way to improve binding affinity to a particular target molecule. To the best of our knowledge, this is the first study to exploit the increased affinity of dimerized ligands in order to enable isolation/precipitation of the target species based on its structure rather than its protein composition.
The CAP-1 structure was successfully designed and then demonstrated to bind and isolate aggregates with amyloid structure (fibrils of Aβ 42 , tau and α-synuclein), but crucially not monomers, using synthetic aggregates. Furthermore, the K d values for CAP-1 binding Aβ 40 and α-synuclein fibrils were comparable, which suggests that there will be no selective bias in capturing aggregates of different proteins. Importantly, we have also demonstrated successful AP of early aggregates of α-synuclein formed after 8 hours of aggregation, which are predominantly oligomers (Fig. 2e). Our data show that most of the oligomers are selectively removed by CAP-1 but not by the beads alone ( Supplementary Fig. 21). This is strong evidence that CAP-1 captures oligomers of α-synuclein as well as fibrils.
However, it is also critical to demonstrate translational relevance. In vivo, the complexity of the biofluids that surround the central nervous system and the brain tissue itself makes the detection of small amyloids a major challenge 63 . We demonstrated that the sensitive detection of amyloid-containing aggregates can be performed in CSF by amyloid pull-down using CAP-1, followed by bead digestion and detection by MS. We observed an increased number of amyloid-prone proteins when using CAP-1 compared to plain beads and an increase in the total content of the β-strand, highlighting the strength of our AP method in enriching β-sheet-containing proteins (Supplementary Tables 1 and 2).
An important question to consider is whether CAP-1 will capture toxic aggregates. Our previous work showed that the oligomers formed after a structural conversion to a more proteinase-K-resistant structure are cytotoxic to neurons 16 , and the formation of β-sheet-active species has been shown to lead to increased membrane permeabilization 19 . These species were also shown to be ThT active 15 and hence should be captured by CAP-1. We also previously showed that monomers of Aβ or α-synuclein cannot cause membrane permeabilization but aggregates can 22 , and hence it is likely that only protein aggregates (and not monomers) present in CSF cause membrane permeabilization. Since our experiments using CSF showed that CAP-1 captures the species responsible for membrane permeabilization and calcium ion entry, then, included in the species captured by CAP-1 should be the toxic aggregates present in CSF responsible for disrupting calcium ion homoeostasis in vivo. This data therefore show that CAP-1 captures toxic aggregates from CSF. However, we cannot rule out that there will be non-ThT-active species that are also toxic and will not be captured by CAP-1, but these species appear not to be the dominant toxic species present in CSF as measured by the membrane permeabilization assay.
This new capability for unbiased detection and capture of amyloids, coupled with a MS approach to provide the molecular composition, is a powerful combination because it has the potential to define the amyloids present in the brain, CSF and other biofluids. The ability of disease-causing aggregates (containing β-sheets) to cause membrane permeabilization has been previously correlated with their cytotoxic potential 22,25 . Importantly, in AP, the beads contain negligible monomers and, since it is possible to remove the captured aggregates from the beads, this approach allows further characterization of the human-derived aggregates and cytotoxicity experiments to be performed (Supplementary Fig. 20). Further improvements to the design of the capture molecule are possible by optimizing the linker length and the head groups and by synthesizing multimeric molecules to further improve the sensitivity and selectivity of AP 64,65 . This may further increase the affinity of the dimer of CAP-1 over the monomeric version. There are also important advantages to using the AP method in terms of stability and resistance to degradation compared to conventional antibodies [66][67][68] .
One particular advantage of this approach is the unbiased selection of amyloid conformations of proteins, regardless of their molecular identity. The appearance of misfolded and aggregated proteins is likely important in the early stages of disease. Developing complementary methods that do not require protein-specific approaches, but rather structural approaches, may identify the important biomarker in any disease. This is particularly true as the structural conformation is the biggest determinant of toxicity in disease, and therefore this approach may select for the pathogenic biomarker of disease.
AP may ultimately lead to early diagnostic tools. The capability to detect β-sheet aggregates present in an unbiased way will allow us to determine which protein aggregates change during the development of AD or PD and hence develop new diagnostic methods for early disease detection. Several studies show that aggregates of Aβ, α-synuclein and tau are present in human biological fluids such as the CSF and serum of patients affected with AD (Aβ and tau) and PD (α-synuclein) [69][70][71][72] . Since the aggregation and deposition of cross-β-sheet-rich protein aggregates start in the central nervous system from 5 to 15 years before clinical manifestations of disease, detection of these β-sheet-rich aggregates holds the promise of developing a long-awaited diagnosis of AD and PD at the clinically asymptomatic stage, as well as predicting disease progression and monitoring the effects of potential drugs. This method will also yield insights into disease mechanisms: it provides fundamental information about the role of protein aggregation in human disease, allowing one to study how protein homoeostasis is disrupted in humans and whether this is due to the aggregation of a specific protein such as α-synuclein in PD, or the more general aggregation of several aggregation-prone proteins such as Aβ, tau, α-synuclein and TAR DNA binding protein 43 (TDP43). It is also not known how the relative amounts of these different aggregates will change during the development of different neurodegenerative diseases. In the latter case, differences in the relative amounts of these different proteins in biofluids would define the different diseases. AP may also reveal if protein aggregates play a fundamental role in human physiology, since they are present from a young age, or form only as a result of ageing. Addressing all these questions is only possible using an unbiased method that detects all β-sheet aggregates.
In conclusion, we have successfully developed a molecule inspired by the trimeric shape of an antibody to target all aggregates containing cross β-sheet motifs present in complex biofluids. CAP-1 has two binding sites to improve avidity and a third moiety to enable surface immobilization and therefore capture of aggregates based on their structure, but not on their protein composition. This simple and versatile method allows the identification of molecular components of aggregates, using MS. Overall, this structure-based approach will pave the way to understanding the exact molecular species responsible for neurodegeneration in humans and consequently hasten development of simple and robust early diagnosis methods.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41557-022-00976-3.

Methods
This research complies with all relevant ethical regulations and was approved by the Ethics Committee at University of Gothenburg (EPN 140811).
NH bis-benzothiazole. Boc-protected bis-benzothiazole B (64 mg 0.08 mmol, 1 equiv.) was added to a solution of HCl (4 M in MeOH, 3 ml) at room temperature. After 1 h, aqueous NaOH (3 M, 10 ml) and EtOAc (10 ml) were added and the layers separated. The aqueous layer was further extracted with EtOAc (2 × 10 ml), and the combined organic layers dried over Na 2 SO 4 and concentrated to give the NH bis-benzothiazole C (55 mg, 80 mmol, 99%) as a light yellow solid. The melting point was 168-170 °C.
Prior to use, α-synuclein aliquots were ultracentrifuged at 350,000g for 1 h at 4 °C using a TL120.2 rotor (Beckman) in an Optima TLX Ultracentrifuge (Beckman) to remove possible seed contaminants. Two-thirds of the total volume in the tube was used as the supernatant fraction (monomers only) and removed with minimal perturbation to avoid remixing of unwanted seeds. Afterward, the protein concentration was determined using a nanodrop (molar extinction coefficent at 275 nm for tyrosine ε 275 nm (Tyr) = 5,600 M −1 cm −1 ), and then the α-synuclein was diluted in cold Tris buffer (25 mM) supplemented with NaCl (100 mM) at pH 7.4 and 0.01% NaN 3 (to prevent bacterial growth) to a final concentration of 70 μM. This solution was incubated in the dark at 37 °C with constant agitation at 200 r.p.m. (New Brunswick Scientific Innova 43), and aliquots were taken at desired times (0 h for monomers, 6-8 h for oligomers and 1-5 days for mature fibrils). All time points were imaged on the TIRFM set-up before any experiments to confirm the presence/absence of the desired α-synuclein intermediate species, that is, the absence of aggregates at t = 0 h and of fibrils at 6 to 8 h, and the presence of diffraction-limited size aggregates in the 6 to 8 h time point aliquots. All steps were carried out using LoBind microcentrifuge tubes (Eppendorf) to limit surface adsorption. For binding affinity experiments and pull-down (Fig. 2d), mature α-synuclein fibrils (5 days incubation) were sonicated with a probe sonicator (Bandelin, Sonopuls HD 20170) four times at 15 s each and at 10% power; the tube was placed on a beaker containing ice to minimize overheating effects on the tube walls, and afterward, the protein was aliquoted and stored at −80 °C until use.
Preparation and photophysical characterization of CAP-1. A CAP-1 1 mM stock solution was prepared in DMSO solvent, divided into 20 μl aliquots and stored at −20 °C. Aliquots were used once to avoid freeze and thaw cycles. The photophysical properties of CAP-1 were determined using a Varian Cary Eclipse fluorescence spectrophotometer. Experimental settings used were λ ex = 355 nm (5-10 nm bandwidth) and λ em = 370-600 nm (5-10 nm bandwidth). Ultravioletvisible absorption and fluorescence (both excitation and emission) spectral characterization of CAP-1 (20 μM) were carried out in both PBS and Tris (25 mM) supplemented with NaCl (100 mM, pH 7.4). To test the CAP-1 solubility in PBS ([CAP-1] PBS ), different dilutions were prepared between 0 nM and 200 nM, and the emission spectrum was recorded using λ ex = 355 nm. Data points were plotted as [CAP-1] PBS versus maximum fluorescence intensity. The linear relationship between concentration and fluorescence intensity (R 2 = 0.98; R, coefficient of multiple correlation) strongly suggests that CAP-1 obeys the Beer-Lambert law and is therefore completely soluble in the range used.
Measurement of protein aggregation in plate reader. Fluorescence kinetics measurements of α-synuclein with either CAP-1 or ThT were monitored using a FLUOstar Omega fluorescence plate reader (BMG Labtech) in bottom reading mode under quiescent conditions. Corning 96-well plates with half-area (3881, polystyrene, black with clear bottom) and non-binding surfaces sealed with aluminium sealing tape were used for each experiment. Monomeric α-synuclein (40 µM; +2.4 µM preformed fibrils) was incubated in the presence of 50 μM CAP-1 or ThT at 37 °C under quiescent conditions for 4 days with data point collection every 6 min (ref. 74 ).
α-Synuclein preparation for plate reader measurements. Monomeric α-synuclein was prepared from purified α-synuclein subjected to gel filtration using a Superdex 75 10/300 GL column (Cytiva Life Sciences) equilibrated in MES buffer (10 mM 2-(N-morpholino) ethanesulfonic acid, 1 mM ethylenediaminetetraacetic acid (EDTA), pH 5.5), and the peak corresponding to monomeric α-synuclein peptide was collected in a low-binding test tube (Corning) on ice. Seed fibrils were produced as described previously 74 . Briefly, concentrated stock α-synuclein monomer (100-200 µM) was incubated at 40 °C for 72 hours with a Teflon bar on an RCT Basic Heat Plate (IKA) in PBS. To estimate the fibril concentration (monomeric equivalent), the solution was centrifuged at 21,130g in a benchtop centrifuge (Eppendorf). The concentration of the remaining α-synuclein monomer in the supernatant was estimated by absorbance using a NanoDrop 2000 (Thermo Fisher Scientific) and was subtracted from the concentration at the start of the aggregation. The volume of the supernatant was subsequently replaced by MES buffer and the stock was aliquoted and stored at −80 °C. To prepare α-synuclein seeds, the fibril stock was diluted to 5 µM final concentration in low-protein-binding tubes and sonicated for 15 s (1 s on, 1 s off) using a probe sonicator.
Binding affinity. The binding affinity measurements were conducted on a Duetta spectrofluorometer (HORIBA). The experimental settings used for CAP-1 and S1 (a monovalent version of CAP-1) were λ ex = 355 nm (10 nm bandwidth) and λ em = 370-600 nm (10 nm bandwidth), while those for ThT were λ ex = 440 nm (10 nm bandwidth) and λ em = 455-600 nm (10 nm bandwidth). Sonicated α-synuclein (200 nm; Supplementary Fig. 14) was prepared as described above, and Aβ 40 fibrils were obtained by incubating 4 μM of monomeric Aβ 40 in PBS for 4 h at 37 °C with constant agitation followed by the same sonication procedure. Different concentrations of dye (CAP-1 and S1 up to 200 nM and ThT up to 10 μM) were incubated with 100 nM of either α-synuclein or Aβ 40 for 20 min before measurement in PBS. The K d for ThT was obtained by fitting the experimental points to a hyperbolic curve (specific binding). Since CAP-1 and S1 are fluorescent in the absence of binding to aggregates, an equivalent set of data points was also collected in the absence of protein (the dye only curve). A model that takes into account the change in fluorescence between a bound and unbound molecule was then used (refs. [75][76][77] and Supplementary Methods). This is important for CAP-1 and S1 since there is only a small increase in fluorescence on the binding of fibrils.
AP assay. The AP assay consists of the pull-down of protein aggregates (for example, α-synuclein) using streptavidin Dynabeads (MyOne Streptavidin C1, Invitrogen) conjugated with CAP-1. Briefly, 30 μl of beads/sample was removed from the vial, resuspended in 1 ml PBS and placed on a magnet for 2-3 min for separation, and the supernatant was discarded. This step was repeated three times. Afterward, the beads were resuspended in 1 ml of CAP-1 (30 μM), and the tube was placed in a revolver mixer for incubation at room temperature for 1 h. Following this, the tube was placed on the magnet for 2-3 min and the supernatant discarded. The beads were washed three times with PBS, as before. Finally, the beads were resuspended in 500 μl solution containing α-synuclein (10 μM; monomers alone, or a mixture of aggregates) and left in the revolver mixer for 2 h or overnight at 4 °C. In the end, the tube was placed on the magnet for 2 to 3 min, and 450 μl supernatant was removed to a clean tube and labelled as the 'depleted' fraction; both the depleted fraction and the 'beads' were kept at 4 °C until use. All steps were carried out in LoBind microcentrifuge tubes (Eppendorf) to limit surface adsorption.
Detection of bead-bound α-synuclein. After AP, both 'beads' (diluted 1:32 in PBS) and the 'depleted' fraction were added to a 96-well half-area plate with a clear bottom (Corning 3881) for the bulk fluorescence measurement. The plate was placed in a plate reader (CLARIOstar, BMG Labtech) and the fluorescence intensity (bottom reading) was measured straight away at room temperature using the following settings: end-point mode, excitation filter 440 nm and bandwidth