Skip to main content
eLife logoLink to eLife
. 2019 Sep 25;8:e46574. doi: 10.7554/eLife.46574

Structural mapping of oligomeric intermediates in an amyloid assembly pathway

Theodoros K Karamanos 1,2,, Matthew P Jackson 1,2, Antonio N Calabrese 1,2, Sophia C Goodchild 1,2,, Emma E Cawood 1,2, Gary S Thompson 1,2,§, Arnout P Kalverda 1,2, Eric W Hewitt 1,2, Sheena E Radford 1,2,
Editors: Tricia R Serio3, John Kuriyan4
PMCID: PMC6783270  PMID: 31552823

Abstract

Transient oligomers are commonly formed in the early stages of amyloid assembly. Determining the structure(s) of these species and defining their role(s) in assembly is key to devising new routes to control disease. Here, using a combination of chemical kinetics, NMR spectroscopy and other biophysical methods, we identify and structurally characterize the oligomers required for amyloid assembly of the protein ΔN6, a truncation variant of human β2-microglobulin (β2m) found in amyloid deposits in the joints of patients with dialysis-related amyloidosis. The results reveal an assembly pathway which is initiated by the formation of head-to-head non-toxic dimers and hexamers en route to amyloid fibrils. Comparison with inhibitory dimers shows that precise subunit organization determines amyloid assembly, while dynamics in the C-terminal strand hint to the initiation of cross-β structure formation. The results provide a detailed structural view of early amyloid assembly involving structured species that are not cytotoxic.

Research organism: E. coli

Introduction

Oligomers have been the focus of amyloid research over decades because of their pivotal role in assembly and their potential cytotoxicity (Chiti and Dobson, 2017). Numerous aggregation-prone proteins (or their fragments) form oligomers (Benilova et al., 2012; Wei et al., 2011; Laganowsky et al., 2012; Apostol et al., 2013), some of which are cytotoxic (Laganowsky et al., 2012; Ono et al., 2009; Fusco et al., 2017), while others are not (Mannini et al., 2014). Many groups have attempted to elucidate the structure(s) of amyloid oligomers with different biological properties (Chiti and Dobson, 2017). However, their ephemeral nature, dynamic signature, and heterogeneity in mass and conformation provide significant experimental challenges. Hence, our current understanding of the structure of oligomers is often limited to low-resolution models (Fusco et al., 2017; Vestergaard et al., 2007; Campioni et al., 2010; Cremades et al., 2012), or to oligomers assembled from non-natural amino acids, short peptides, or protein fragments (Laganowsky et al., 2012; Apostol et al., 2013; Sangwan et al., 2017). Establishing a relationship between the oligomers observed and the mechanism of amyloid formation is also an important, but challenging, task. In some cases, oligomers have been shown to be ‘off-pathway’ since they have to dissociate for amyloid formation to proceed (Wu et al., 2010; Baskakov et al., 2002; Souillac et al., 2003; Bieschke et al., 2010). Characterization of such species, however, does not provide insight into the structural mechanism by which initially unstructured (e.g. Aβ42, α-synuclein) or natively structured proteins (e.g. lysozyme, transthyretin, antibody light chains, β2-microglobulin (β2m)) undergo conformational conversion to form the parallel in-register cross-β structure of amyloid (Iadanza et al., 2018a). Other oligomers have been shown to be on-pathway (Fusco et al., 2017; Cremades et al., 2012), or to form via secondary nucleation processes that enhance the rate of fibril formation (Cohen et al., 2013). Proteins in an oligomeric or aggregated form have also been characterized kinetically, thermodynamically and biophysically (Cohen et al., 2013; Cohen et al., 2018; Lenton et al., 2017). However, a detailed understanding of both the structural properties of oligomers and their role in assembly is needed in order to understand the structural mechanism(s) of amyloid formation and the origins of cytotoxicity, as well as to design inhibitors of the assembly process.

Here, we describe an integrative approach which uses kinetic modeling to identify oligomers formed on-pathway to fibril formation, NMR spectroscopy and other biophysical methods to determine their structural properties, and cellular assays to determine their cytotoxicity. The strategy employed can be applied to other assembling protein systems and draws on the powers of NMR to provide detailed structural information about individual precursors in dynamic equilibrium within complex mixtures of assembling species, and kinetic modeling to ascribe their role in amyloid formation. To exemplify this combined structural and kinetic approach we focus on the naturally occurring variant of human β2m, known as ΔN6. This variant lacks the N-terminal six amino acids and is formed by natural proteolytic truncation of the wild-type (WT) protein (Esposito et al., 2000; Eichner et al., 2011). WT human β2m (named herein as hβ2m) forms amyloid deposits in the joints of patients undergoing long term hemodialysis (Gejyo et al., 1985). However, hβ2m does not aggregate into amyloid fibrils at physiologically relevant pH and temperature on an experimentally accessible timescale in vitro (the pH in normal and diseased joints ranges from 5.5 to 7.4; Floege and Ehlerding, 1996). Addition of Cu2+ ions, detergents, organic solvents, glycosaminoglycans or collagen can drive hβ2m amyloid formation at neutral pH (Platt and Radford, 2009; Stoppini and Bellotti, 2015; Yamamoto et al., 2005; Benseny-Cases et al., 2019). These reagents partially unfold the native protein, facilitating cis-trans isomerization of Pro32 that initiates assembly (Eichner et al., 2011; Platt and Radford, 2009; Yamamoto et al., 2005; Chiti et al., 2001). By contrast with the intransigence of hβ2m to form amyloid in vitro, ΔN6 is highly amyloidogenic, forming fibrils rapidly in vitro in the absence of additives at pH 6–7 (Karamanos et al., 2016). ΔN6 forms ~ 30% of β2m in amyloid plaques in patients with dialysis-related amyloidosis (Bellotti et al., 1998). Previous studies have shown that ΔN6 can induce amyloid formation of hβ2m at near-neutral pH in vitro (Eichner et al., 2011) and can co-assemble with the WT protein into amyloid fibrils (Sarell et al., 2013). The structure of hβ2m in amyloid fibrils formed in vitro at low pH (pH 2.0) has also been solved recently using solid-state NMR and cryo-EM, revealing a parallel in-register cross-β structure typical of amyloid, which differs dramatically from the all anti-parallel immunoglobulin fold of the native precursor (Iadanza et al., 2018b). The atomic structure(s) of hβ2m amyloid fibrils formed in vivo, and those of ΔN6 formed in vitro or ex vivo, however, are not yet known.

Several examples of oligomers (dimers, tetramers and hexamers) of WT hβ2m have been reported previously (Calabrese et al., 2008; Eakin et al., 2006; Mendoza et al., 2011; Mendoza et al., 2010; Halabelian et al., 2015; Colombo et al., 2012; Rennella et al., 2013; Liu et al., 2011; Karamanos et al., 2014), with one report of a domain swapped dimer of ΔN6 stabilized by addition of a nanobody (Domanska et al., 2011). Since hβ2m is inert to aggregation at physiological pH and temperature in vitro, the oligomerization of the protein was stimulated by mutation and/or the addition of Cu2+ ions (Calabrese et al., 2008; Eakin et al., 2006; Mendoza et al., 2011; Mendoza et al., 2010), or by linkage of monomers via non-native disulfide bonds (Halabelian et al., 2015; Colombo et al., 2012). Although some of these oligomers form under conditions in which WT hβ2m may eventually form fibrils, the role of individual oligomeric species in the aggregation mechanism remains unclear. The oligomers formed in the initiating stages of aggregation of ΔN6 into amyloid also remain obscure.

Here, we show that amyloid formation of ΔN6 occurs via a remarkably specific assembly mechanism involving the transient formation of dimers and hexamers. Exploiting NMR methods able to analyze dynamic and lowly populated states (Anthis and Clore, 2015), we characterize these assemblies, yielding a structural model of the initiating events in ΔN6 aggregation in atomic detail. The results reveal the formation of head-to-head dimers that pack into symmetric hexamers that retain a native-like immunoglobulin fold and are not cytotoxic. The hexamers appear to be primed for further conformational change into the cross-β structure of amyloid by dynamic unfurling of their C-terminal β-strands. The results portray a detailed atomic view of the early stages of ΔN6 assembly that may enable the development of routes to combat disease by targeting the specific protein-protein interactions that define the early stages of assembly.

Results

Fibril elongation occurs through an oligomeric state

Previous results have shown that ΔN6 assembles rapidly into amyloid fibrils in vitro at pH 6.2, but not at pH 8.2 (Eichner et al., 2011), suggesting that lowering the pH increases the population of aggregation-prone species. Such species may also be relevant in vivo given the acidic microenvironment of the joints of DRA patients (Eichner et al., 2011; Bellotti et al., 1998; Karamanos et al., 2014). At pH 6.2 (close to its pI of 5.8) ΔN6 is dynamic, but retains a native-like immunoglobulin fold (Eichner et al., 2011). To determine the kinetic mechanism by which ΔN6 aggregates into amyloid fibrils, experiments were performed in which ΔN6 fibril seeds (20 μM monomer equivalent concentration) were incubated with different concentrations of ΔN6 monomers (20 μM to 500 μM) and the rate of amyloid formation was monitored by the fluorescence of thioflavin T (ThT). All experiments were performed at pH 6.2 at a total ionic strength of 100 mM (see Materials and methods). The simplest kinetic mechanism in which monomers add to the fibril ends would result in a linear dependence of the initial rate of fibril elongation versus the monomer concentration, with saturation at high monomer concentrations (Buell et al., 2014; Buell et al., 2010; Xue et al., 2009). Such behavior is observed for seeded assembly of acid unfolded monomers of hβ2m, which initially lack persistent structure (Platt et al., 2008), into amyloid fibrils at pH 2.0 (Figure 1a,b). By contrast, ΔN6 showed more complex behavior, with a clear non-linearity in the initial rate of elongation versus monomer concentration, in which rapid seeded growth occurs only above ~200 μM ΔN6 (Figure 1c,d). This indicates that fibril elongation by ΔN6 must involve addition of one or more oligomeric species to the fibril ends under the conditions employed.

Figure 1. Dependence of the fibril elongation rate on the concentration of soluble protein.

Figure 1.

Seeded elongation assays for (a) hβ2m at pH 2.0 monitored by ThT fluorescence. 20 μM of preformed seeds of hβ2m (formed at pH 2.0) and varying amounts of soluble protein were added, as indicated in the key. Note that the protein does not aggregate under these conditions in the absence of seeds on this timescale (Xue et al., 2008). The dashed line shows the initial rate of each reaction. (b) The initial rate of fibril elongation (shown in units of ThT fluorescence (a.u.)/h) versus the concentration of hβ2m added. The dashed line represents a prediction using a monomer addition model (see Table 4). (c) Seeded elongation assays for ΔN6 using 20 μM preformed seeds formed from ΔN6 at pH 6.2 as a function of the concentration of soluble ΔN6 added. Open blue symbols denote the ThT fluorescence signal of 500 μM ΔN6 in the absence of seeds. The dashed line shows the initial rate of each reaction. (d) The initial rate of fibril elongation (shown in units of ThT fluorescence (a.u)/h) versus the concentration of soluble ΔN6 added. The dashed line shows the dependence of the elongation rate (in units of ThT fluorescence (a.u)/h) on the concentration of monomer assuming a monomer addition model (see Table 4). The elongation rate for monomer addition shows a hyperbolic behavior as a function of monomer concentration, with a linear dependence at lower monomer concentrations, followed by a saturation phase at higher monomer concentrations. The simulation in (b) (dashed line) uses a slower microscopic elongation rate (ke) (Table 4) than that used in panel (d) and therefore saturation is not achieved by 410 μM protein in (b), but is in (d). Five replicates are shown for each protein concentration. Error bars show the standard deviation between all replicates.

Native-like dimers and hexamers form during Δn6 assembly

The concentration-dependence of ΔN6 elongation could be explained by an oligomer(s) acting as the elongation unit. To explore whether oligomeric species of ΔN6 are formed under the conditions employed, sedimentation velocity analytical ultracentrifugation (AUC), size exclusion chromatography (SEC), cross-linking, and NMR experiments were performed. These approaches report on the conformational properties and molecular weight distribution of the assemblies formed at different ΔN6 concentrations. Sedimentation velocity AUC experiments showed that ΔN6 forms discrete oligomers at pH 6.2, with monomers, dimers and higher order species with a sedimentation coefficient (S value) consistent with 6–9-mers (although the rapid equilibration of the species present prevents accurate determination of their mass and population) (Figure 2a). To investigate the molecular mass of the species present, ΔN6 was cross-linked after different incubation times in the absence of fibril seeds using 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide hydrochloride (EDC) (see Materials and methods) and the resulting species examined using SDS-PAGE (Figure 2b). This revealed the presence of hexamers during assembly (Figure 2b). The population of the hexameric species is decreased at later time points, presumably because it is consumed into fibrils (Figure 2b). Analytical SEC of ΔN6 at different protein concentrations without cross-linking revealed only monomers and dimers (Figure 2—figure supplement 1a), consistent with the higher order assemblies dissociating upon dilution on the column. However, when cross-linking was performed prior to SEC, higher molecular weight oligomers were observed, with these species being more abundant when higher protein concentrations were used (Figure 2—figure supplement 1b,c). At the highest concentration of ΔN6 used (500 μM) cross-linking resulted in the rapid formation of high molecular weight aggregates that elute in the void volume (Figure 2—figure supplement 1b,c). The population of these aggregates increases with time, accompanied by depletion of the oligomers, consistent with these species being capable of assembly into amyloid (Figure 2—figure supplement 1d).

Figure 2. ΔN6 oligomer formation.

(a) Sedimentation velocity AUC of ΔN6 at different concentrations, as indicated by the key. Note that the higher order species decrease in intensity at high protein concentrations (>200 μM) consistent with the formation of large aggregates that sediment rapidly before detection (see also Figure 2—figure supplement 1d). (b) SDS–PAGE of cross-linked ΔN6 (80 μM) at different time-points during de novo fibril assembly in the absence of fibril seeds (see Materials and methods). Note that dimers are not observed, presumably as they are not resilient to the vigorous agitation conditions used to accelerate fibril formation in these unseeded reactions, or are not efficiently cross-linked by EDC under the conditions used (see Materials and methods). A negative stain electron micrograph of ΔN6 after 100 hr of incubation is shown below. Scale bar – 500 nm. (c) The methyl region of the 1H NMR spectrum of ΔN6 at 400 μM (left) or 10 μM ΔN6 (right). (d) Per residue combined 1H-15N chemical shift differences between the 1H-15N HSQC spectrum of ΔN6 at 10 μM and 400 μM. Blue dots represent residues for which assignments are missing in both spectra. The dashed line represents one standard deviation (σ) of chemical shifts across the entire dataset. Residues that show chemical shift differences > 1σ are shown in yellow,>2σ are colored red, and residues for which the chemical shift difference is not significant (<1σ) are colored gray. Residues that are broadened beyond detection in the spectrum obtained at 400 μM are colored in magenta (see also Figure 2—figure supplement 2a). Residues are numbered according to the sequence of the WT protein. Arg 97 is hydrogen bonded to residues in the N-terminus and presumably is indirectly affected by the interaction. (e) The structure of ΔN6 (2XKU; Eichner et al., 2011) colored in the same scheme as (d). Pro32 is shown in blue space-fill. The buffer used in all experiments was 10 mM sodium phosphate pH 6.2 containing 83.3 mM NaCl (to maintain a constant ionic strength of 100 mM for all experiments), 25°C.

Figure 2.

Figure 2—figure supplement 1. Analysis of ΔN6 oligomerization.

Figure 2—figure supplement 1.

(a) Analytical SEC traces of uncross-linked ΔN6 at different concentrations as indicated in the key. (b) Analytical SEC traces of cross-linked ΔN6. (c) Zoom-in of the SEC traces shown in (b). The elution profile of protein standards is shaded in the background. (d) Analytical SEC traces of 500 μM ΔN6 0 hr (black), 4 hr (green) or 24 hr (blue) after cross-linking was performed. (e) Protein correlation times (τc) measured using a 1H-TRACT experiment (see Materials and methods) as a function of ΔN6 concentration at pH 6.2, colored as in (b). The black line represents a linear fit to the data. The correlation time of 600 μM ΔN6 at pH 8.2 is shown in blue. (f) The exponential decay rate (d) of the 1H NMR signal in a diffusion measurement using stimulated echoes as a function of ΔN6 concentration. The black line represents a linear fit to the data. The linear scaling of τc and d is predicted from the linear dependence of the overall assembly rate koveron on ΔN6 concentration using the calculated Kds and a monomer-dimer-hexamer model (inset) (see Materials and methods). Data points are colored as in (b). Error bars in (e) and (f) are calculated from the noise level of the spectra and are smaller than the marker points.
Figure 2—figure supplement 2. Estimation of dimer and hexamer Kd values.

Figure 2—figure supplement 2.

(a) 1H-15N HSQC spectra of 10 μM (green) or 400 μM (pink) ΔN6. Resonances which are broadened >80% at 400 μM ΔN6 are indicated on the spectrum. (b) The combined 1H-15N chemical shift differences that report on hexamer formation as a function of ΔN6 concentration (the data at 50 μM are excluded since at this concentration the equilibrium is dominated by dimer formation). The solid lines represent fits to a monomer-dimer-hexamer model using a dimer Kd of 50 μM and a hexamer Kd of 10 × 10−9 M2 (see Materials and methods). Error bars represent the standard deviation of the resonances that do not show significant chemical shift changes between 10 and 410 μM ΔN6. Data were acquired at 25°C in 10 mM sodium phosphate pH 6.2 containing 83.3 mM NaCl (total ionic strength of 100 mM). (c) 1H-15N HSQC spectra of 10 μM (panel i), 20 μM (panel ii), 50 μM (panel iii), 100 μM (panel iv), 200 μM (panel v) or 410 μM (panel vi) ΔN6. Residues are labeled in panel (i) according to the color scheme of Figure 2d. (d) Reduced χ (Benilova et al., 2012) surface produced by fits to the monomer-dimer-hexamer model using the 10 residues (11, 12, 23, 26, 50, 51, 52, 67, 68, 97) that showed the largest chemical shift changes. (e) HN RDCs as a function of the ΔN6 concentration (a single example for A15 is shown for clarity). (f) Reduced χ (Benilova et al., 2012) values for the fitting of RDC data over the 41 residues measured to the structure of ΔN6 as a function of the Kd value used to extrapolate the RDCs to 100% dimer (see Materials and methods). Error bars were calculated from the noise level of the experiment.

1H-NMR and 1H-15N HSQC NMR spectra of ΔN6 were next acquired to examine the properties of the oligomers formed. Significant changes in chemical shift and linewidth of individual resonances at different concentrations of ΔN6 were observed, consistent with the finding that ΔN6 self-assembles into higher molecular weight species at pH 6.2 (Figure 2c and Figure 2—figure supplement 2a). The residues most affected lie in the A strand and the BC, DE and FG loops, suggesting that these regions form the intermolecular interfaces in the higher molecular weight species (Figure 2d,e). Consistent with these observations, measurement of the rotational correlation time (τc) and diffusion coefficient of the sample, which reflect the average size and shape of the molecules formed, showed a linear dependence on ΔN6 concentration, consistent with protein oligomerization in which the resulting species are in dynamic exchange (Figure 2—figure supplement 1e,f). Together these results show that ΔN6 assembles into dimers and hexamers that are assembly competent, in dynamic exchange, and assemble via interfaces which are located in the apical region of the protein that surrounds Pro32 (Figure 2e).

To estimate the dissociation constants for dimer and hexamer formation, the chemical shifts and residual dipolar couplings (RDCs) of individual resonances were measured as a function of ΔN6 concentration from 10 to 410 μM (Figure 2—figure supplement 2a–e). Significant chemical shift differences were observed when the ΔN6 concentration was increased from 10 μM to 50 μM without significant line broadening (Figure 2—figure supplement 2c, panels i-iii). Increasing the protein concentration to 100 μM caused a decrease in the chemical shift differences (Figure 2—figure supplement 2c, panel iv), which then increase again in magnitude at 200 μM and 410 μM, accompanied by significant line broadening (Figure 2—figure supplement 2a and c, panels v,vi). This complex behavior is consistent with a monomer-dimer-hexamer equilibrium in which the monomers and dimers have different chemical shifts, while the chemical shifts of dimers and hexamers are similar (an assumption that is supported by our structural models, see below), and the exchange rate between monomers and hexamers is significantly faster than that between monomers and dimers. Therefore, the monomer-dimer equilibrium dominates the equilibrium (and the observed chemical shift) at low concentrations (50 μM). At higher concentrations the dimer is depleted relative to the hexamer and the chemical shift observed becomes a complex combination of the population of each species, the exchange rate between each species, and the difference in chemical shift of each residue in each assembly. Fitting the chemical shift data to a monomer – dimer – hexamer model yields a Kd for dimer formation of ≤50 μM, while that of hexamer formation is ~10 ± 5 x 10−9 M2 (Benilova et al., 2012) (see Materials and methods and Figure 2—figure supplement 2b,d), indicating that once dimers form they have a high affinity for one another. Importantly, the monomer – dimer – hexamer model with the estimated affinities adequately describes the observed increase in the τc and the observed diffusion coefficient versus protein concentration (Figure 2—figure supplement 1e,f), independently supporting the model derived. Increasing the Kd for dimer formation to >100 μM results in unrealistically low values for the hexamerization Kd (Figure 2—figure supplement 2d). Moreover, measurement of RDCs versus protein concentration results in a biphasic curve (Figure 2—figure supplement 2e), consistent with a multi-equilibrium process. Using these data, the RDCs of the dimer species can be calculated for a range of estimated Kd values (see Materials and methods). Fitting the dimer RDCs to the structure of ΔN6 (2XKU; Eichner et al., 2011), shows significantly poorer fits to the predicted RDC values assuming a dimer Kd higher than 50 μM (Figure 2—figure supplement 2f). To explain the chemical shift and RDC data, therefore, the dimer Kd must be ≤50 μM.

Specific interfaces determine aggregation

To map the interfaces involved in ΔN6 oligomer formation in more detail, intermolecular paramagnetic relaxation enhancement (PRE) experiments were performed. The PRE depends on the distance between a paramagnet and adjacent nuclei and can provide distance information about (transient) binding interfaces for nuclei that are within ~20 Å of the paramagnetic center (Clore and Iwahara, 2009), quantified by the effect of the spin label on the relaxation rates of each amide proton (the HN2 PRE rate). 14N-ΔN6 was spin-labeled with (1-oxyl-2,2,5,5-tetramethyl-D3-pyrroline-3-methyl) methanethiosulfonate (MTSL) by creating Cys variants at positions 20, 33, 54 or 61. Each protein (60 μM) was then mixed with 15N-ΔN6 (60 μM) at pH 6.2. At this total protein concentration, the PREs observed are dominated by the monomer-dimer equilibrium (35% of molecules are monomer, 51% of ΔN6 molecules are in dimers and 14% of ΔN6 molecules are in hexamers, determined from the Kd values measured above). These experiments (Figure 3) showed increased HN2 rates for residues in the A strand and the BC, DE and FG loops when the spin label is attached to residues 33, 54, or 61, suggestive of a head-to-head interaction involving the apical regions of the protein (Figure 3a–c and e). In accord with this conclusion, when MTSL is attached at position 20 at the distal side of the protein (Figure 3e), the HN2 rates are vastly decreased (Figure 3d).

Figure 3. Identification of interacting surfaces in ΔN6 dimers.

Intermolecular PRE data for the self-association of ΔN6. 15N-ΔN6 (60 μM) was mixed with an equal concentration of (a) 14N-(S33C)ΔN6-MTSL; (b) 14N-(L54C)ΔN6-MTSL; (c) 14N-(S61C)ΔN6-MTSL; or (d) 14N-(S20C)ΔN6-MTSL in 10 mM sodium phosphate buffer, pH 6.2 containing 83.3 mM NaCl (a total ionic strength of 100 mM). The resulting Γ2 rates are color-coded according to the amplitude of the PRE effect (see scale bar: gray-insignificant (<20 s−1), yellow->20 s−1, red->50 s−1, pH 6.2, 25°C). Blue dots in the plots are residues for which resonances are not assigned (na) at pH 6.2. Red crosses indicate high HN2 rates for which an accurate value could not be determined. Control experiments showed that the small PREs arising from14N-(S20C)ΔN6-MTSL arise from non-specific interactions with MTSL itself. Solid black lines depict fits to the PRE data for the dimer structure shown in Figure 4a. Note the poor fits for some residues which are sensitive to hexamer formation (14% of ΔN6 molecules) under the conditions used. Residues are numbered according to the WT sequence and the position of β-strands (2XKU; Eichner et al., 2011) is marked above each plot. (e) The structure of ΔN6 (2XKU; Eichner et al., 2011) with the BC loop shown in magenta, the DE loop in green and the FG loop in yellow. The MTSL attachment sites are highlighted as spheres.

Figure 3.

Figure 3—figure supplement 1. Lack of a hexamer population precludes aggregation of ΔN6 at pH 8.2.

Figure 3—figure supplement 1.

(a) Aggregation assays for 60 μM ΔN6 monitored by ThT fluorescence at pH 6.2 (red) or pH 8.2 (blue), 37°C with agitation (600 rpm). Five replicates are shown. Negative stain transmission electron micrographs of samples at 100 hr are shown alongside in the same color code. (b) Sedimentation velocity AUC traces for 120 μM ΔN6 at pH 6.2 (red) or pH 8.2 (blue). (c,d) Intermolecular PRE values for ΔN6 at pH 8.2. 60 μM 15N-ΔN6 was mixed with (c) 60 μM of 14N-(S61C)ΔN6-MTSL or (d) 60 μM of 14N-(S20C)ΔM6-MTSL in 10 mM sodium phosphate buffer, pH 8.2 containing 86.6 mM NaCl (total ionic strength 100 mM). Γ2 rates are color-coded according to their amplitude (blue-not assigned, gray-insignificant (<20 s−1), yellow->20 s-1, red->50 s−1 at pH 8.2, 25°C). Residues are numbered according to the WT sequence. The position of β-strands (from 2XKU; Eichner et al., 2011) is marked above each plot.
Figure 3—figure supplement 2. Mapping the interface of ΔN6 self-association at pH 8.2 using CPMG experiments.

Figure 3—figure supplement 2.

15N Relaxation dispersion CPMG data for residues (a) V49, (b) E74 and (c) Y78 at 1200 μM (blue) or 600 μM ΔN6 (red). Solid lines represent fits to a fast exchange model (see Materials and methods). (d) Plots of Rex (defined as R2,eff50Hz - R2,eff680Hz) per residue at different concentrations of ΔN6 at pH 8.2 as indicated in the key. The dashed line represents one standard deviation of the mean. (e) Correlation plot of HN RDCs measured at pH 6.2 versus those back-calculated from the structure of ΔN6 (2XKU; Eichner et al., 2011; Eisenberg et al., 1984) at pH 7.5. (f) The structure of ΔN6 monomers (2XKU; Eichner et al., 2011) colored according the amplitude of the Rex values at 1200 μM shown in (d). The results show that the interface between interacting monomers at pH 8.2 involves interaction between β-sheets mediated by residues in the B, D and E β-strands and adjacent residues in the DE loop. This interface is very different to the loop-loop interactions that create the dimer interface at pH 6.2 (see Figure 3e).

To determine whether the head-to-head dimers are critical for aggregation, the AUC, PRE and fibril growth experiments were also performed at pH 8.2 where ΔN6 does not assemble into amyloid fibrils even after extended incubation times (Figure 3—figure supplement 1a). The sedimentation velocity AUC experiments revealed that monomers and tetramers are formed at pH 8.2, but not hexamers, with the equilibrium in favor of the monomer (Figure 3—figure supplement 1b). Consistent with this, the τc of 600 μM ΔN6 at pH 8.2 is ~12 ns, in marked contrast with the τc of ~50 ns predicted for 600 μM ΔN6 at pH 6.2 (Figure 2—figure supplement 1e). Finally, intermolecular PRE experiments at pH 8.2 showed small Γ2 rates irrespective of the site of MTSL labeling (Figure 3—figure supplement 1c–d), suggesting that the monomers bind with different affinity and/or via different interfaces at this pH. To investigate these hypotheses, CPMG relaxation dispersion NMR experiments were performed. These experiments are able to detect excited states populated to as little as 1% of the total protein in solution (Hansen et al., 2008). Concentration-dependent CPMG profiles of residues in the B strand, D strand, DE loop, E strand and EF loop were observed at pH 8.2 (Figure 3—figure supplement 2a–d), indicating that the binding interface for tetramer formation differs substantially from the loop-loop interactions in the apical region of the protein that dominate assembly at pH 6.2, despite the fact that ΔN6 retains an immunoglobulin-like fold at both pH values (Figure 3—figure supplement 2e,f). As a consequence of the altered interface that forms at pH 8.2, hexamers and fibrils do not form. Together these results indicate that the head-to-head dimers formed at pH 6.2 are uniquely able to assemble into the hexamers that are crucial for fibril assembly.

Different dimer structures determine amyloid inhibition and propagation

To generate dimer structures consistent with the experimental data obtained, simulated annealing molecular dynamics calculations were performed. The calculations converged to two dimer structures (Figure 4a, Figure 4—figure supplement 1 and Table 1). In the lowest energy model (model A), the ΔN6 monomers are arranged in an extended conformation with the N-terminal residues M6 and I7 (WT numbering), along with the BC, DE and FG loops forming the interface (Figure 4a). The inhibitory dimer of ΔN6:murine β2m (mβ2m) was previously determined using a similar approach (Karamanos et al., 2014). This dimer also has a head-to-head subunit arrangement but is characterized by a more acute angle between ΔN6 subunits in which the monomers interact predominantly through the BC and DE loops (Karamanos et al., 2014) (Figure 4b, Video 1). Thus, distinct protein dimers formed from closely related sequences (mβ2m and hβ2m are 70% identical and 90% similar in sequence) give rise to fundamentally different outcomes of assembly.

Figure 4. Structural models of ΔN6 dimers.

Structural models of (a) the lowest energy ΔN6 homodimer (dimer A) and (b) the ΔN6-mβ2m heterodimer that inhibits ΔN6 fibril assembly (Karamanos et al., 2014). Interface residues (identified as those residues that have any pair of atoms closer than 5 Å) are shown in a ball and stick representation on one subunit and are colored in space fill in gold in (a) or red in (b) on the surface of the second subunit. ΔN6 is shown in the same pose (blue) in (a) and (b). The BC, DE and FG loops are shown in magenta, green and yellow, respectively, and the position of attachment of MTSL for the PRE experiments (residues 20, 33, 54 and 61) is highlighted in spheres. PDB files are publicly available from the University of Leeds depository (https://doi.org/10.5518/329). See also Video 1.

Figure 4.

Figure 4—figure supplement 1. Alternative ΔN6 dimer structures.

Figure 4—figure supplement 1.

Intermolecular PRE data for the self-association of 15N-ΔN6 (60 μM) mixed with 60 μM of (a) 14N-(S33C)ΔN6-MTSL, (b) 14N-(L54C)ΔN6-MTSL, (c) 14N-(S61C)ΔN6-MTSL, or (d) 14N-(S20C)ΔN6-MTSL) with the PRE effect color-coded according to its amplitude (blue dot- residues not assigned, gray-insignificant (<20 s−1), yellow->20 s−1, red->50 s−1, pH 6.2, 25°C). Red crosses indicate high HN2 rates for which an accurate value could not be determined. Solid black lines represent back-calculated PREs from the high energy dimer structure shown in (e) and (f). The small PREs arising from14N-(S20C)ΔN6-MTSL result from non-specific interactions with MTSL itself. (e and f) The structural model of dimer B shown in different orientations. In each diagram, one subunit is shown in cartoon representation (BC loop (magenta), DE loop (green) and FG loop (yellow)) and the second is shown as a surface. Interface residues are highlighted as balls and sticks on the first subunit and shown in gold space-fill on the second subunit. The MTSL attachment sites are highlighted as spheres and the positions of attachment (20, 33, 54 or 61) are labeled. The interface in dimer B involves a more extensive inter-subunit interface in the apical loops than observed in dimer A (Figure 4a). The resulting interface for dimer B does not describe the PRE data (solid black line) as well as the lower energy model of dimer A shown in Figure 4a (Materials and methods and Table 1).

Table 1. Agreement between experimental and back-calculated intermolecular PREs for the two dimer structures (dimer A and dimer B (see Figure 5—figure supplement 3).

RMS values are shown comparing the measured versus the predicted values from the structure PREs measured from S33, L54 and S61. Data from position S20 were not used as they arise from non-specific interactions with MTSL.

PRE term RMS dimer A RMS dimer B
S33C(ΔN6)-ΔN6 (s−1) 18.65 15.10
L54C(ΔN6)-ΔN6 (s−1) 29.02 27.44
S61C(ΔN6)-ΔN6 (s−1) 19.44 23.27
*High PREs (Å) 2.78 3.79

*High PREs refer to PREs in the BC loop (measured from S33, L54 and S61) that (due to their large value) could not be measured accurately and therefore are incorporated as loose distance restraints.

Video 1. Comparison of productive and inhibitory dimers.

Download video file (2.1MB, mp4)
DOI: 10.7554/eLife.46574.012

The ΔN6 subunit in each dimer (this study) is shown as a dark blue cartoon, while the second ΔN6 monomer in the productive dimer and the mβ2m subunit in the inhibitory dimer (Karamanos et al., 2014) are shown as light blue and red, respectively. The BC, DE, FG loops are colored in magenta, green and yellow, respectively, while the intra-dimer interface residues are shown as sticks on both subunits.

Structural models of on-pathway hexamers

Although the majority of the intermolecular PREs can be satisfied by the dimer A structure, the fits are not perfect (Figure 3a–d), presumably since ~14% of ΔN6 molecules form hexamers at the concentration of ΔN6 employed (120 μM). The PRE experiments were thus repeated at higher concentrations (320–400 μM) of ΔN6, wherein > 40% of ΔN6 molecules are predicted to be in hexamers. These experiments revealed a pattern of HN2 rates similar to those obtained at 120 μM ΔN6 (Figure 5—figure supplement 1a–d), with the highest HN2 rates involving the N-terminus, BC, DE and FG loops, suggesting that similar interfaces are formed in the dimer and hexamer. CPMG experiments at 180 μM ΔN6 (26% of ΔN6 molecules are monomer, 48% are dimer, and 26% are hexamer) and 480 μM ΔN6 (13% of ΔN6 molecules are monomer, 32% are dimer and 55% are hexamer) showed that residues in the apical regions of ΔN6, surrounding Pro 32, are also in concentration-dependent exchange at both ΔN6 concentrations at pH 6.2, in support of this conclusion (Figure 5—figure supplement 2).

The ordered nature of assembly (monomer, dimer, hexamer) and the identification of the interfaces involved, allowed us to generate models for the hexameric species (Figure 5—figure supplement 3). The measured PREs were converted into distances and simulated annealing molecular dynamics calculations were performed to create hexamer structures consistent with the experimental PRE and chemical shift data using the lowest energy dimer model (dimer A shown in Figures 4a and 5 - Figure 5—figure supplement 3a), as well as the less favorable dimer model B (Figure 4—figure supplement 1e and Figure 5—figure supplement 3b), as starting points. Note that the structure calculation strategy employed does not require knowledge of the dimer and hexamer populations (see Materials and methods). Starting from dimer A (Figure 4a) the structure calculation resulted in a hexamer in which the three dimers trimerize to form a compact daisy-like structure (Figure 5a–c). The PREs back-calculated from this model are consistent with the experimental data (Figure 5—figure supplement 4). Importantly, hexamer structures generated from dimer B (Figure 4—figure supplement 1e) resulted in poorer fits to the PRE profiles (Materials and methods and Table 2).

Figure 5. Structural model of ΔN6 hexamers.

(a–c) Sphere representations of the hexamer model formed from dimer A rotated by 90° in each view. Subunits belonging to the same dimer are colored in different tones of the same color. (d) The monomer-monomer (intra-dimer) interface is highlighted in green on the surface of the dimer formed from subunits 1a and 1b (within dimer A), with the other dimers shown as cartoons. (e) The inter-dimer interface is colored red on the surface of the dimer formed from subunits 1a and 1b, with the dimers shown as cartoons. (f) As in (e), but showing the dimer formed from subunits 1a and 1b, superposed with the mβ2m subunit in the inhibitory ΔN6-mβ2m dimer (Karamanos et al., 2014) (green cartoon). The ΔN6-ΔN6 and ΔN6-mβ2m dimers were aligned on the ΔN6 subunit 1b. Schematics of the assemblies are shown at the bottom colored as in (d–f). Note that the BC, DE and FG loops are highlighted as thicker chains in blue, green and cyan, respectively, in d-f. PDB files are publicly available from the University of Leeds depository (https://doi.org/10.5518/329). See also Video 2.

Figure 5.

Figure 5—figure supplement 1. Intermolecular PREs at high ΔN6 concentration.

Figure 5—figure supplement 1.

Intermolecular PRE data for the self-association of (a) 240 μM 15N-ΔN6 mixed with 80 μM 14N-(L54C)ΔN6-MTSL, (b) 200 μM 15N- ΔN6 mixed with 200 μM 14N-(L54C)ΔN6-MTSL, or (c) 80 μM 15N- ΔN6 mixed with 240 μM 14N-(S33C)ΔN6-MTSL. PRE data are color-coded according to their amplitude (blue dots-not assigned, gray-insignificant (<20 s−1), yellow->20 s−1, red->50 s−1, pH 6.2, 25°C). Red crosses indicate high HN2 rates for which an accurate value could not be determined. (d) Raw PRE data for residue 85V when 60 μM 14N-(L54C)ΔN6-MTSL was mixed with 60 μM 15N-ΔN6 (left) or when 200 μM 14N-(L54C)ΔN6-MTSL was mixed with 200 μM 15N-ΔN6 (right). Solid lines represent single exponential fits for the paramagnetic (black) or the diamagnetic samples (red).
Figure 5—figure supplement 2. Additional interfaces do not form in the ΔN6 hexamer.

Figure 5—figure supplement 2.

15N relaxation dispersion CPMG data for residues (a) 37, (b) 67, and (c) 83 at 180 μM ΔN6 (26% ΔN6 molecules are monomers, 48% are in dimers, 26% are in hexamers) (red) or 480 μM ΔN6 (13% ΔN6 molecules are monomers, 32% are in dimers, 55% are in hexamers) (black). Solid lines represent fits to the fast exchange model, yielding values of kexbind of 1790 ± 290 s−1 at 180 μM ΔN6 and kexbind of 1170 ± 196 s−1 at 480 μM ΔN6 (see Materials and methods). (d) Plots of Rex per residue defined as R2,eff50Hz - R2,eff680Hz. The dashed line represents one standard deviation of the mean calculated for all data points. Residues are numbered according to the WT sequence. Significant CPMG profiles are observed for residues in the N-terminus, A strand, BC, DE and FG loops, in excellent agreement with the intermolecular PRE data shown at 120 μM and 320 μM ΔN6 in Figure 3 and Figure 5—figure supplement 1. Residues which are severely broadened at 480 μM, thereby precluding accurate determination of their Rex values, are shown as black crosses. Crucially, when the protein concentration was increased the residues which show significant CPMG profiles are unchanged suggesting that the dimers and hexamers share a similar interface. (e) The structure of ΔN6 (2XKU; Eichner et al., 2011) colored according to the Rex amplitude as indicated in the scale bar. Trans Pro32 is shown in space-fill (pale blue).
Figure 5—figure supplement 3. Initial docking of dimer structures to create hexamer models.

Figure 5—figure supplement 3.

Plots of RMSD (to the lowest energy structure) versus total energy for hexamers generated by docking of (a) the lowest energy dimer structure (dimer A) or (b) the higher energy dimer (dimer B). The 50 lowest energy hexamer structures are marked as red circles. The hexamers that were selected for the next round of structure calculation for each dimer starting model are marked with green arrows. The structural model of dimer A and dimer B are shown alongside colored as in Figure 4—figure supplement 1.
Figure 5—figure supplement 4. Intermolecular PREs back-calculated from the hexamer structural model generated from dimer A.

Figure 5—figure supplement 4.

Intermolecular PRE data for the self-association of ΔN6. 15N- ΔN6 (60 μM) was mixed with 60 μM of (a) 14N-(S33C)ΔN6-MTSL, (b) 14N-(L54C)ΔN6-MTSL, (c) 14N-(S61C)ΔN6-MTSL, or (d) 14N-(S20C)ΔN6-MTSL. The data are color-coded according to their amplitude (blue dots-not assigned, gray-insignificant (<20 s−1), yellow->20 s−1, red->50 s−1, pH 6.2, 25°C). Red crosses indicate high HN2 rates for which an accurate value could not be determined. Solid black lines represent back-calculated PREs from the lowest energy hexamer structure (arising from dimer A) shown in Figure 5. The RMS distances (Å) between the intermolecular distances that were used as restraints and those back-calculated from the hexamer structural model are shown (inset) for each dataset (see Materials and methods).
Figure 5—figure supplement 5. Conformational and biochemical properties of ΔN6 hexamers.

Figure 5—figure supplement 5.

(a) ESI-IMS-MS analysis. Collision cross section (CCS) distributions for each observed charge state of hexameric ΔN6. The charge state for each CCS distribution is indicated. Note that the CCS of the lowest (most native; Vahidi et al., 2013) charge state (15+) is consistent with the hexamer model generated from dimer A (labeled A (green)), but not the models generated from dimer B (labeled (B(i)), (B(ii)) and (B(iii)) for the three conformers labeled in Figure 5—figure supplement 3b). (b) Hydrophobicity of the hexamer interface. The surface of dimer one in the hexamer is colored according to the Eisenberg hydrophobicity scale (Arg = −2.53, Ile = 1.38) (Eisenberg et al., 1984) with the other dimers shown as cartoons. A key is show alongside. The view on the left-hand side shows the surface that is packed against dimers 2 and 3 in the hexamer (interior), with the view on the right-hand side showing the exterior surface of the assembly. (c, d) Fluorescence emission spectra of ANS (200 μM) incubated with (c) ΔN6 monomers (green), (d) dimers (open symbols) or hexamers (red) (eluting at 17 mL, 15 mL and 11 mL, respectively obtained with/without cross-linking, as indicated, using SEC; Figure 5—figure supplement 6). The fluorescence emission spectrum of ANS in buffer alone is shown in blue. ANS bound to the partially folded Im7 variant L53A I54A (Spence et al., 2004) (1 μM) is shown for comparison (black). This was used as a model for a compact native-like folding intermediate (Spence et al., 2004) (see text).
Figure 5—figure supplement 6. ΔN6 oligomers are not cytotoxic to SH-SY5Y cells.

Figure 5—figure supplement 6.

Toxicity of cross-linked (solid line/gray bars) or uncross-linked (dotted line/white bars) ΔN6 species following purification by analytical SEC. Cell toxicity was assessed using MTT reduction, cellular ATP level, generation of reactive oxygen species (ROS), and LDH release assays. For assays of MTT reduction, ATP levels and ROS production, the data are normalized to PBS (100%) and NaN3-treated controls (0%). LDH release is normalized to detergent lysed cells (100%) and PBS buffer treated controls (0%). The error bars represent mean S.E, * p 0.05. No evidence for cytotoxicity was observed for any protein species under the conditions employed.

Table 2. Agreement between experimental and back-calculated intermolecular distances for different hexamer structures.

RMS values are shown comparing the measured versus the predicted distances from each structural model for distances measured from S33, L54 and S61. Data from position S20 were not used as they arise from non-specific interactions with MTSL. See also Figure 5—figure supplement 3.

PRE term Hexamer 1
RMS (Å)
Hexamer 2(i)
RMS (Å)
Hexamer 2(ii)
RMS (Å)
Hexamer 2(iii)
RMS (Å)
S33C(ΔN6)-ΔN6 2.34 2.68 2.58 2.53
L54C(ΔN6)-ΔN6 1.25 2.33 2.26 1.87
 S61C(ΔN6)-ΔN6 2.22 2.7 2.68 3.11

In the hexamer models generated from dimer A the dimeric subunits are arranged in a helical manner twisted by ~120°, creating a hexamer that is ~60 Å in diameter and 75 Å in length. This hexamer model is consistent with the collision cross-section (CCS) of ΔN6 hexamers measured using the lowest charge state (15+) (the most native-like species; Vahidi et al., 2013; see Materials and methods) detected using Electrospray Ionization Ion Mobility Spectrometry – Mass Spectrometry (ESI-IMS-MS), but the measured CCS is inconsistent with hexamers derived using dimer B (Figure 5—figure supplement 5a). The monomer-monomer and dimer-dimer interfaces in the best fit hexamer structure (Figure 5a–e) involve similar, but not identical, regions, with the inter-dimer interface extending further into the β-sheet containing the A, B, E and D β-strands, while the intra-dimer interface is formed mostly through the BC and DE loops (Figure 5d–e and Video 2). The formation of dimers generates a hydrophobic surface which becomes buried in the hexamer (Figure 5e, Figure 5—figure supplement 5b and Table 3). Consistent with this, the cross-linked hexamers show a small (1.3-fold) increase in fluorescence in the presence of the hydrophobic dye 8-anilino-1-naphthalenesulfonic acid (ANS), that is much smaller than the ~100 fold increase in ANS fluorescence observed for a typical ‘molten globule’ state (Semisotnov et al., 1991), but similar in magnitude to ANS bound to the highly structured on-pathway folding intermediate of Im7 (monitored using the trapped equilibrium mimic of this species, Im7 L53AI54A (Spence et al., 2004) (Figure 5—figure supplement 5d). The ΔN6 dimers show a similar increase in ANS fluorescence as the hexamers despite having a larger exposed hydrophobic surface area, possibly because ANS binds more weakly or has a lower quantum yield when dimer-bound (Figure 5—figure supplement 5c–d). The interface formed in the inhibitory ΔN6-mβ2m dimer overlaps with the surface required for hexamerization, but not for ΔN6-ΔN6 dimerization (Figure 5f), rationalizing why mβ2m is able to inhibit amyloid formation (note that the Kd of the mβ2m:ΔN6 complex is 70 μM (Karamanos et al., 2014), similar to that (~50 μM) estimated here for ΔN6 homo-dimerization). The dimers and hexamers were incubated with SH-SY5Y cells, a cell line that is commonly used in studies of amyloid toxicity (Laganowsky et al., 2012; Fusco et al., 2017; Campioni et al., 2010; Jakhria et al., 2014; Leri et al., 2016; Giorgetti et al., 2008), and which has been shown previously to take up monomeric and fibrillar β2m (Jakhria et al., 2014). Interestingly, there was no evidence for cytotoxicity in assays for 3-(4,5-dimethylthiazol-2-yl)−2,5-diphenyltetrazolium bromide (MTT) reduction, lactate dehydrogenase release, reactive oxygen species formation and cellular ATP level (see Materials and methods) (Figure 5—figure supplement 6). However, rapid dissociation of the uncross-linked oligomers, prevention of conversion to a cytotoxic form by cross-linking, or cytotoxicity requiring different cell types or prolonged exposure (>24 hr) to the oligomers cannot be ruled out.

Table 3. Analysis of dimer and hexamer interfaces.

The buried surface area is calculated as the sum for the two subunits for each complex. Interface residues were identified as those residues that lose at least 10% of accessible surface area upon oligomer formation.

ΔN6 dimer A ΔN6 hexamer
Buried Surface Area (Å2) 1233 4201
% Charged residues in the interface 28 18
% Hydrophobic residues in the interface 44 54

Video 2. ΔN6 assembles into dimers and hexamers.

Download video file (2.9MB, mp4)
DOI: 10.7554/eLife.46574.022

The two ΔN6 subunits in the dimer (dimer A) are shown as blue cartoon and gray cartoon/transparent space-filling representations, respectively. The BC, DE and FG loops are colored magenta, green and yellow, respectively. The intra-dimer interface residues are shown as sticks on one subunit and as orange transparent spheres on the second subunit. The hexamer assembly is then shown as a space-filling model, with dimer one shown in dark blue/light blue, dimer two in dark yellow/light yellow and dimer three in magenta/pink. In the last part of the video only dimer one is shown as spheres while dimers 2 and 3 are shown as transparent cartoons. The intra-dimer interface is shown in green and the inter-dimer interface is shown in red.

Hexamer dynamics may prime further assembly into amyloid

The hexamer shown in Figure 5 retains a native-like immunoglobulin fold in which the β-strands are anti-parallel. Hence, a major conformational rearrangement has still to occur for ΔN6 to form amyloid fibrils in which the β-strands stack in a parallel in-register structure (Debelouchina et al., 2010) (R. Silvers, Y. Su, R.G. Griffin, and S.E. Radford, unpublished). Hints of how this conformational change may be initiated were obtained by quantitative analysis of the CPMG data shown in Figure 6, Figure 5—figure supplement 2 and Figure 6—figure supplement 1. Globally fitting these data for residues which lie in the dimer and/or hexamer interfaces (residues 26, 34, 35, 37, 51, 59, 65, 66, 67, 83, 84, 85, Figure 5d–e) to a two-state fast exchange model yields an exchange rate, kexbind, of 1790 ± 290 s−1 (Figure 6a,b). Distinct CPMG profiles were observed, however, for residues 87, 89, 91 and 92 which lie in the G strand of monomeric ΔN6 and which are not involved in the dimer-dimer interfaces (they show no significant concentration-dependent chemical shifts, nor PREs are observed for these residues at low or high protein concentration (Figure 6—figure supplement 1a). The CPMG data for these residues presumably report on conformational changes that result from hexamerization rather than the direct binding event itself. The CPMG data indicate that these residues exchange with a lowly populated (2%) excited state with an interconversion rate, kexG, of 205 ± 150 s−1, 10-fold slower than kexbind (Figure 6c–d and Figure 6—figure supplement 1a–e). Therefore, a distinct process, possibly local unfolding of the C-terminal β-strand, occurs when the hexamer is formed that is driven by the free energy of hexamer formation (ΔG°hexamer ~4 kJ/mol). At ΔN6 concentrations of 480 μM kexG is increased to 1170 ± 196 s−1 (Figure 6—figure supplement 1f–i), consistent with increased hexamer formation enhancing the observed rate of dynamics of the G strand. Hexamer formation thus potentially destabilizes the G-strand of ΔN6, causing local unfolding of this region of the polypeptide chain (although further experiments measuring the sign of the chemical shift change would substantiate this conclusion). This may then lead to more catastrophic structural reorganization of the hexamer into the parallel in-register structure of amyloid (note that the G-strand sequence forms a β-strand in the ΔN6 fibril core; Su et al., 2014). Whether structural conversion occurs within the hexamer, at the fibril end, or requires further, more elaborate molecular steps such as active participation of the fibril surface, or disassembly into smaller structural units prior to fibril assembly, remains to be seen.

Figure 6. G-strand unfurling may occur upon hexamer formation.

15N CPMG relaxation dispersion data at 750 MHz (magenta) and 950 MHz (red) (180 μM ΔN6, pH 6.2 (26% ΔN6 molecules are monomers, 48% are in dimers, 26% are in hexamers) for residues (a) 51, (b) 37, (c) 89, and (d) 92. Residues 37 and 51 report on intermolecular interactions that describe dimer and/or hexamer formation (schematic, top left), while residues 89 and 92 do not lie in an interface and report instead in the dynamics of the G strand in the different assemblies formed. The position of all five residues used in the cluster analysis of G strand dynamics is shown in spheres on the structure of ΔN6 (blue cartoon, top right). Pro32 is shown as a magenta sphere. Solid lines represent global fits to the Bloch-McConnell equations (Materials and methods) for each cluster of residues. The extracted parameters of the global fit for the two processes (kexbind and kexG) are indicated above the plots.

Figure 6.

Figure 6—figure supplement 1. Hexamer formation increases the dynamics of the G strand.

Figure 6—figure supplement 1.

(a) Location of the G strand in relation to the dimer and hexamer interfaces. Dimer one in the hexamer is shown in a cartoon representation while dimers 2 and 3 are shown as semi-transparent surfaces. The positions of the amide protons for residues 87, 89, 91, 92 are shown as gray spheres and the residues that take part in both the dimer and hexamer interfaces are shown as red spheres on the structure of dimer 1. A schematic of the assembly is shown alongside. 15N relaxation dispersion CPMG data for residues (b) 87, (c) 89, (d) 91 and (e) 92 at 950 MHz (red) and 750 MHz (magenta) of 180 μM ΔN6, pH 6.2. Solid lines represent the global fits to all residues in the cluster to the slow exchange model which yields a kexG of 205 ± 150 s−1. CPMG data for the same residues (f) 87, (g) 89, (h) 91 and (i) 92 at 750 MHz (blue) and 600 MHz (gray) using 480 μM ΔN6. Solid lines represent global fits to the fast exchange model which yields a kexG of 1170 ± 196 s−1.

A unified model of Δn6 polymerization

As a final test of the validity of the model of ΔN6 assembly proposed we assessed the ability of the structural, kinetic and thermodynamic parameters deduced above to describe the observed rates of fibril formation measured using ThT fluorescence, as well as the τc values versus ΔN6 concentration measured by NMR, and the fibril yield. Using the dimer and hexamer structural models shown in Figures 4 and 5 and the estimated Kd values for their formation, all of the derived experimental data could be recapitulated (Figure 7). Fitting the seeded fibril growth data to different kinetic models that assume (i) monomers to add to the fibril ends (Figure 7—figure supplement 1a); (ii) monomers are in exchange with a monomeric excited state that is responsible for growth (Figure 7—figure supplement 1b); or (iii) dimers are the elongation units (Figure 7—figure supplement 1c), fail to describe the seeding data (Materials and methods and Table 4). By contrast, a model assuming addition of hexamers describes the ThT kinetic profiles well (Figure 7a), while a more complicated monomer-dimer-tetramer-hexamer model does not improve the fit significantly (Figure 7—figure supplement 1d). The populations of monomer, dimer and hexamer, together with the derived structural models, are also consistent with the observed dependence of τc on protein concentration (Figure 7b). Finally, the amount of hexamer formed (in the absence of seeds) is also predictive of the fibril yield (Figure 7c,d) consistent with the hexamer being required for fibril formation. This conclusion is also supported by the appearance of hexamers early during assembly in the absence of seeds and their disappearance as fibrils form (Figure 2b).

Figure 7. The monomer-dimer-hexamer model describes the thermodynamics and kinetics of fibril elongation.

(a) Global fits (blue solid lines) to the fibril elongation kinetics monitored by ThT fluorescence assuming a hexamer addition model at different concentrations of soluble ΔN6 (dots) (Materials and methods and Table 4). The concentrations of ΔN6 are colored according to the key. The average of five replicates is shown. (b) Protein correlation times (τc) measured using NMR (red) and back-calculated values (green) using the populations of monomers, dimers and hexamers predicted from the monomer-dimer-hexamer model and the correlation times of the dimers and hexamer structural models shown in Figures 4 and 5. (c) The fibril yield (after 100 hr) of each elongation reaction. SDS-PAGE analysis of the whole reaction (shown in (a)) before centrifugation (W) or of the supernatant (S/N) after centrifugation at the different concentrations of ΔN6, as indicated. (d) Bar-charts showing the % of insoluble material (gray) measured using densitometry of the gel shown in (c). The % hexamer population in the absence of seeds (black) predicted by the monomer-dimer-hexamer model at each ΔN6 concentration correlates with the % insoluble material (gray). Note that the fibril yield is low since fibrils cannot form when the monomer concentration falls significantly below the Kd for dimer formation (50 μM).

Figure 7.

Figure 7—figure supplement 1. Alternative kinetic models do not describe the kinetics of seeded fibril growth.

Figure 7—figure supplement 1.

Global fits (blue solid lines) which assume that (a) a monomer, (b) a monomer excited state, or (c) a dimer, add to the fibril ends do not describe the observed fibril growth kinetics monitored using ThT fluorescence at different concentrations of soluble ΔN6 (dotted lines and key). A more complex monomer-dimer-tetramer-hexamer model (d) does not improve the quality of the fit compared with that shown in Figure 7a.

Table 4. Reaction schemes, rate equations and rate constants for the fibril elongation models tested.

X represents the species that add onto the fibril ends.

Module Variant Reaction scheme Rate equations Rate constants
Pre-polymerization No Pre-polymerization
(Monomer addition)

X=X1

d[X]dt=i=2N-keFi-1X+ke'Fi k1=ke
k1'=ke',
Monomer conformational exchange X1k1k1X1
X=X1
d[X1 ]dt= k1 X1 + k1X1
d[X]dt={k1X1k1X1+i=2NkeFi1X+keFi
k1,ke
k1',ke',
Dimer addition X1+X1k1k1X2
X=X2
d[X1]dt=-2k1X1X1+2k1'X2
d[X]dt={k1X1X12k1X+i=2NkeFi1X+keFi
k1,ke
k1',ke',
Hexamer addition X1+X1k1k1X2
X2+X2+X2k1k1X6
X=X6
d[X1]dt=-2k1X1X1+2k1'X2
d[X2 ]dt=k1 X1 X1  k1X2  3k2 X2 X2 X2 + 3k2X6 
d[X]dt={k2X2X2X2k2X+i=2NkeFi1X+keFi
k1,k2,ke,
k1',k2',ke',
Monomer-Dimer-Tetramer-Hexamer X1+X1k1k1X2
X2+X2k2k2X4
X2+X4k3k3X6
d[X1]dt=-2k1X1X1+2k1'X2
d[X2 ]dt=k1 X1 X1  k1X2  2k2 X2 X2 + 2k2X4  k3 X4 X2 + k3X6 
d[X4 ]dt=k2 X2 X2  k2X4  k3 X4 X2 + k3X6 
k1,k2,k3ke
k1',k2',k3',ke',

X=X6

d[X]dt={k3X4X2k3X+i=2NkeFi1X+keFi

Polymerization XXFkekeF1kekeF2...FNXX d[F]dt=-keXF1+ke'F2
d[Fi]dt=ke X Fi1 keFi ke X Fi+ keFi+1  2i<N
d[FN]dt=keXFi-1-ke'Fi

Discussion

Understanding the molecular details of oligomer formation is vital if we are to understand why proteins aggregate into amyloid and why different species have different toxicities (Iadanza et al., 2018a; Lu et al., 2013). Here, we present a general strategy, summarized in Figure 8—figure supplement 1, which allows the identification of oligomeric intermediates in amyloid assembly and enables their structural characterization. By combining the powers of NMR to detect lowly populated species in dynamic exchange, with complementary techniques such as AUC and cross-linking, oligomeric intermediates can be identified and structurally characterized in atomic detail. Importantly, to link these intermediates to the mechanism of aggregation, the derived affinities, stoichiometries and structural models can then be used to globally model the time course of fibril assembly. The strategy presented is not only applicable to protein aggregation, but to any weakly self-associating protein system. Given that the balance between monomers, dimers, higher molecular weight oligomers and fibrils could depend critically on the experimental conditions, including the pH, temperature, protein concentration, amount of seed added, buffer composition and ionic strength, the same protein, or a closely related protein variant, could assemble via different mechanism(s) under different conditions. Indeed, aggregation of many amyloidogenic proteins, including hβ2m (Iadanza et al., 2018b), is known to result in polymorphic fibrils (Close et al., 2018; Fitzpatrick et al., 2017; Colvin et al., 2016; Zhang et al., 2019) that could extend via different mechanisms. The approach described here can distinguish between such different assembly pathways and may be able to shed light on the role of individual oligomeric species in aggregation and the origins of amyloid polymorphism.

Using the workflow derived, we show that elongation of ΔN6 amyloid seeds proceeds via a specifically organized hexamer (Figure 8). This finding contrasts with the more common view of monomer addition to fibril ends that has been observed for Aβ40/42 (Cohen et al., 2018), α-synuclein (Buell et al., 2014), huntingtin exon 1 (Vitalis et al., 2009) and for unfolded hβ2m at pH 2.0 (Figure 1a) (Xue et al., 2008), while oligomers are thought to play critical roles in the primary/secondary nucleation phases of the assembly of these proteins (Cohen et al., 2018). By contrast with these initially disordered proteins, the monomeric precursor of ΔN6 assembly is structured, a scenario that accounts for more than 20 of the 70 human proteins known to cause amyloid disease (Sipe et al., 2016). Other amyloid precursors that are initially structured show an inability to self-seed (e.g. transthyretin; Hurshman et al., 2004), or display a non-classical dependence of the elongation rate on protein concentration (e.g. light chains; Blancas-Mejía et al., 2017). Whether these and other structured protein precursors assemble by a mechanism akin to that of ΔN6 could be answered by applying the integrated kinetic and structural approach described here to further examples of this set of proteins.

Figure 8. Fibril formation in atomic detail.

Schematic representation of the mechanism of amyloid formation for ΔN6. During folding of hβ2m, a highly dynamic intermediate with a flexible A strand is populated prior to formation of the native-like intermediate termed ΙT, which has a native-like fold but contains a non-native trans X-Pro32 bond. The latter species is mimicked by ΔN6 and formed in vivo by proteolytic degradation of the WT protein (Bellotti et al., 1998). Only IT/ΔN6 is primed for aggregation, while the intermediate with the flexible A strand is not able to assembly directly into amyloid (Karamanos et al., 2016). As reported here, ΔN6 forms elongated head-to-head dimers (upper image, center) which assemble into hexamers. Alternative dimers involving interactions between the ABED β-sheets in adjacent molecules formed at pH 8.2 (lower image, center) do not associate further into fibrils. Murine β2m (mβ2m) also interacts with ΔN6 at pH 6.2 to form head-to-head heterodimers. The subunit orientation is different in this heterodimer (Karamanos et al., 2014), occluding the hexamer interface and inhibiting assembly (central image). ΔN6 hexamers can elongate fibrillar seeds and show enhanced dynamics in the G strand which could represent the first step towards the major structural reorganization required to form the parallel in-register amyloid fold. How this final step occurs, however, remains to be solved.

Figure 8.

Figure 8—figure supplement 1. A workflow to enable weakly self-assembling systems to be analyzed in structural, kinetic and thermodynamic detail.

Figure 8—figure supplement 1.

A schematic overview of the strategy employed to study the aggregation of ΔN6 which can be extended to other systems. Careful examination of kinetic rates of aggregation leads to the identification of possible aggregation pathways. Structural methods (AUC, SEC, cross-linking, ESI-IMS-MS) can then be used to identify the molecular weight and collision crosssection of the species involved. NMR chemical shift analysis and measurements of RDCs can be used to determine estimates of Kd which in turn can be used to determine conditions under which different species are populated. More detailed NMR studies lead to structural models of these species, while stabilization of the intermediates by chemical cross-linking aids the assessment of their cytotoxicity. The structural and kinetic information collected leads to the generation of kinetic models whose ability to describe the progress of aggregation monitored by ThT fluorescence is tested using numerical methods. Agreement is suggestive of the validity of the kinetic mechanism of assembly and the identity and structural properties of oligomeric intermediates formed.

Figure 8—figure supplement 2. Examples of some previously characterized oligomers of WT hβ2m and ΔN6.

Figure 8—figure supplement 2.

(a, b) Cu2+-stabilized H13F hβ2m hexamer (Calabrese et al., 2008; Eakin et al., 2004; Calabrese and Miranker, 2007; Antwi et al., 2008). (c) Domain swapped ΔN6 dimer (Domanska et al., 2011).

Here, we show that ΔN6 dimers and hexamers with well-defined interfaces involving the apical regions of the protein are required for fibrils to form under the conditions employed (Figure 8). By contrast, formation of other interfaces, such as that observed here for ΔN6 at pH 8.2 and the previously reported mβ2m:ΔN6 heterodimer (Karamanos et al., 2014) are not able to assemble into amyloid fibrils (Figure 8). The arrangement of subunits in the ΔN6 dimer and hexamer observed here is different to that in a previously reported structure of a domain swapped ΔN6 dimer (Figure 8—figure supplement 2c). However, the G strand that is responsible for the domain swap is dynamic in the hexamer structure presented here, consistent with this edge β-strand being able to dissociate from the β-sandwich to form both structures. A variant (H13F) of hβ2m has also been reported to form hexamers in the presence of Cu2+ ions (Calabrese et al., 2008) (Figure 8—figure supplement 2a,b). In the crystal structure of this species, the dimers and hexamers interact in a side-to-side or head-to-head manner to create a ring-like assembly, in marked contrast with the daisy-like organization of monomers in the ΔN6 hexamers shown in Figure 5. Real-time NMR studies of the folding of hβ2m have also revealed protein concentration-dependent exchange-broadening in the apical loops of its transient folding intermediate IT (Rennella et al., 2013), an observation that has been attributed to head-head oligomers, in agreement with the data presented here for ΔN6 which structurally mimics IT (Eichner et al., 2011). The interfaces observed in the ΔN6 dimer and hexamer also differ from the canonical inter-sheet stacking between immunoglobulin domains in antibodies, suggesting that the structural features described here are specific to the dimers and hexamers involved in amyloid assembly. Taken together, the results show that β2m can form different protein-protein interactions, only a specific set of which results in species capable of assembly into amyloid.

Although many studies have attributed the toxicity of amyloid to oligomeric species (Chiti and Dobson, 2017), we show here that the dimers and hexamers of ΔN6 are not cytotoxic, at least under the conditions employed, possibly because they are structured and bury substantial hydrophobic surface area. Interestingly, the oligomerization of ΔN6 has been linked to increased toxicity in Caenorhabditis elegans models (Diomede et al., 2012). Since amyloid formation can proceed via multiple pathways, it is possible that the cytotoxic species of ΔN6 formed in the C. elegans body wall muscle are different to those formed here in vitro. For several proteins, cytotoxicity has been ascribed to off-pathway oligomers that accumulate in the lag time of assembly, consistent with amyloid formation being protective for the cell (Bieschke et al., 2011). Interconversion between different forms of oligomers may also be required for cytotoxicity (Fusco et al., 2017; Cremades et al., 2012). Such a process could be compromised in the cross-linked species of ΔN6 used here.

In summary, by taking advantage of the power of NMR spectroscopy to visualize transient species, and combining these experiments with detailed analysis of the kinetic, thermodynamic and hydrodynamic properties of the aggregating ensemble of species, we have been able to determine an atomic structural model of two oligomeric species required for amyloid formation of ΔN6 at pH 6.2, and have generated a model that describes a potential mechanism of fibril elongation from these states. Our findings portray an assembly mechanism that is remarkably well-defined, involving the formation of specific protein-protein interfaces that are unique to the initiating stages of amyloid assembly. Substantial conformational changes have still to occur, however, for the hexameric intermediate to form the cross-β structure of amyloid. How this is achieved remains an open question, but could involve binding to the fibril ends and/or fibril surfaces. Most importantly, the results reveal a remarkable specificity to the early stages of ΔN6 amyloid assembly that involves the formation of well-defined oligomeric species via specific interfaces, the precise details of which determine the course of assembly. These findings suggest new avenues to combat disease by specific targeting of the early intermediates in the amyloid cascade which, at least for ΔN6, involve specific interactions between non-native, assembly-competent states.

Materials and methods

Protein expression and purification

The pINK plasmid containing the ΔN6 gene was transformed into BL21 DE3 plysS E. coli cells. 2 L flasks containing 1 L of LB or HDMI (1 g/L 15N-NH4Cl, 2 g/L 13C-glucose) media were inoculated with 10 mL of starter culture. Cells were incubated at 37°C, 200 rpm until they reached an OD600 of ~0.6 and then the expression of ΔN6 was induced by the addition of 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). Expression was allowed to continue overnight at 37°C and cells were harvested the next morning using a Heraeus continual action centrifuge at 15,000 rpm. The cell pellet containing ΔN6 as inclusion bodies was lysed by the addition of 50–100 mL buffer containing 100 μg/mL lysozyme, 50 μg/mL DNAse I, 50 μg/mL phenylmethanesulfonyl fluoride (PMSF), 10 mM Tris HCl pH 8.2. Further cell disruption was performed using a constant cell disrupter system (Constant Systems) at a pressure of 20.0 kpsi. Inclusion bodies were isolated using centrifugation (15,000 g) for 40 min, 4°C and the inclusion body pellet was washed with 10 mM Tris HCl pH 8.2 four times. ΔN6 was then solubilized in 10–20 mM Tris HCl pH 8.2 containing 8 M urea (MP Biomedicals) and refolded by dialysis (3000 MW cutoff membrane) against 2–5 L of the same buffer lacking urea. The refolded protein was centrifuged for 30 min at 15,000 g to pellet-insoluble material and the supernatant was loaded onto a Q-Sepharose (GE Healthcare) column equilibrated with 20 mM Tris HCl pH 8.2. Bound protein was eluted with a gradient of 0–400 mM NaCl in the same buffer over 800 mL and the protein was freeze-dried after dialysis in 18 MΩ H2O or concentrated using 3000 MW cutoff centrifugal concentrators (Sartorius). The freeze-dried protein was re-suspended in 10 mM sodium phosphate buffer pH 7.0, filtered through 0.2 μm filters (Fisher Scientific) and gel filtered using a HiLoad Superdex-75 Prep column (Amersham Biosciences), calibrated with a standard gel filtration kit (GE Healthcare). The monomer peak was collected, concentrated, aliquoted and stored at −80°C or dialyzed into 18 MΩ H2O and freeze-dried. Cys mutants of ΔN6 were created as described in reference (Karamanos et al., 2014) and purified as above, except that 2 mM dithiothreitol (DTT) was added before gel filtration.

Aggregation assays

ΔN6 seeds were formed by incubation of 800 μM protein in 10 mM sodium phosphate buffer, pH 6.2 containing 83.3 mM NaCl (to give a total ionic strength of 100 mM), 0.02% (w/v) NaN3 with 200 rpm shaking on a thriller shaker (Peqlab) at 37°C for 2 weeks. hβ2m seeds were formed by incubation of 800 μM protein (expressed and purified as described in Karamanos et al., 2014) in 10 mM sodium phosphate buffer pH 2.0, containing 50 mM NaCl, 0.02% (w/v) NaN3 with 200 rpm shaking at 37°C for 2 weeks. The resulting fibrils were sonicated for 15 s to create fibril seeds. For seeding reactions, samples containing 50–500 μM hβ2m or ΔN6 in pH 2.0 or pH 6.2 buffers, respectively, containing 10 μM thioflavin T (ThT) were incubated quiescently at 37 °C in sealed 96 low binding well plates (Thermo Scientific). De novo ΔN6 fibrils were formed by incubating 60 μM ΔN6 in 10 mM sodium phosphate buffer, pH 6.2, containing 83.3 mM NaCl, 0.02% (w/v) NaN3 with 600 rpm shaking in a 96-well plate at 37°C (lag time ~20 hr) or in an 1.5 mL Eppendorf tube (lag time ~100 hr). Control experiments monitoring seeded fibril growth of ΔN6 at pH 8.2 were performed in 10 mM sodium phosphate buffer, pH 8.2 containing 86.6 mM NaCl (total ionic strength 100 mM, identical to that used at pH 6.2) and 0.02% (w/v) NaN3. Fluorescence was monitored at 480 ± 10 nm after excitation at 440 ± 10 nm using a FLUOROstar Optima micro-plate reader (BMG Labtech).

Analytical ultracentrifugation

For sedimentation velocity experiments, a sample of 450 µL of protein was dialyzed overnight with 10 mM sodium phosphate buffer, pH 6.2 containing 83.3 mM NaCl or 10 mM sodium phosphate buffer, pH 8.2 containing 86.6 mM NaCl (each buffer has a total ionic strength of 100 mM). The sample was inserted in double-sector Epon centerpieces equipped with sapphire windows and inserted in an An60 Ti four-cell rotor. Absorbance data at the appropriate wavelength were acquired at a rotor speed of 48,000 rpm at 25°C. Data were analyzed using the c(s) continuous distribution of the Lamm equations with the software SEDFIT (Brown and Schuck, 2006),

D(s)=218ρkTs1/2(η(f/f0)w)3/2((1v¯r)/v¯)1/2,

where D(s) is the diffusion coefficient, k Boltzmann’s constant, T the temperature in K, s the sedimentation coefficient, f the frictional coefficient, f0 the frictional coefficient of a compact smooth sphere, η the solvent viscosity, ρ the solvent density and the partial specific volume.

At concentrations over 200 μM 20% of the material sedimented during the initial 3000 rpm run, consistent with the hexamers forming high-molecular-weight species that sediment before the c(S) data are acquired.

Chemical cross-linking and analytical SEC

ΔN6 (10 μM - 500 μM) was incubated at room temperature without shaking in 10 mM sodium phosphate buffer, pH 6.2 containing 83.3 mM NaCl (total ionic strength of 100 mM), 0.02% (w/v) NaN3 overnight. A 100-fold molar excess of 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide hydrochloride (EDC) (final concentration 1 mM - 50 mM) was added to the reaction, incubated for 10 min with gentle mixing, followed by the addition of 5 mM sulpho-N-hydroxysuccinimide (NHS) for 5 min at room temperature. Cross-linking was then quenched by the addition of 10-fold molar excess (over the concentration of EDC) of Tris HCl pH 8.0, or for cellular cytotoxicity assays, Dulbecco’s PBS, and samples were then analyzed immediately using an analytical Superdex S75 10/300 GL column (GE Healthcare) equilibrated with the same buffer. A similar protocol was used to cross-link ΔN6 during de novo fibril formation. A 500 μL volume of 80 μM ΔN6 in 10 mM sodium phosphate buffer, pH 6.2 containing 83.3 mM NaCl, 0.02% (w/v) NaN3 was incubated in a 1.5 mL micro-centrifuge tube at 37°C with 600 rpm vigorous shaking on a thriller shaker. Under these conditions, the lag time is ~100 hr instead of ~20 hr when the protein is incubated in a 96-well plate (Figure 2b and Figure 3—figure supplement 1a). Samples were cross-linked at various time-points during assembly by addition of 8 mM EDC, incubated for 15 min, followed by addition of 5 mM NHS, also incubated at room temperature for 15 min. The cross-linking reaction was quenched by addition of 200 mM ammonium acetate before samples were subjected to analysis by SDS-PAGE. Given the unavoidable dilution of the samples and their re-equilibration during the SEC run, quantitative analysis of the SEC traces of cross-linked and uncross-linked samples was not performed.

Measurement of ΔN6 correlation times

Rotational correlation times (τc) of ΔN6 at different concentrations were measured in 10 mM sodium phosphate buffer pH 6.2 containing 83.3 mM NaCl, or the same buffer at pH 8.2 containing 86.6 mM NaCl (total ionic strength for each sample of 100 mM), 25°C using a 1H-TRACT experiment (Lee et al., 2006) with delays of 0.002–0.064 s in a Varian Inova NMR spectrometer operating at 750 MHz. At each delay, the signal intensity of resonances in the amide region was integrated and the resulting curve fitted to a single exponential decay function in order to calculate the relaxation rates of the TROSY (Rα) and anti-TROSY (Rβ) spins. The difference Rβ - Rα was then converted to the correlation time (Lee et al., 2006). Errors were calculated using duplicate measurements.

Diffusion NMR measurements on ΔN6

Diffusion NMR experiments were performed on ΔN6 samples at different concentrations using pulsed field gradient (PFG) NMR spectroscopy using stimulated echoes with bipolar gradients performed on a Bruker Avance III 750 MHz spectrometer. A series of 1H spectra were collected as a function of gradient strength (g), the signal (S) was integrated and fitted to:

S/S0=exp(-dg2)

where S0 is the signal intensity at zero field gradient, d is the observed decay rate and g is the strength of the field gradient pulses. The decay rate (d) is directly proportional to the diffusion coefficient, D, of the protein (Stejskal and Tanner, 1965).

Chemical shift perturbation and calculation of Kd values

1H-15N TROSY spectra of ΔN6 at different concentrations were collected using a 750 MHz Bruker Avance III spectrometer. The combined 1H and 15N chemical shift difference was calculated using the function:

Δδ=(5*δ1Η)2+(δ15Ν)2

Chemical shift data at 10 μM, 20 μM, 100 μM, 200 μM and 410 μM ΔN6 were fitted to a monomer (X1) - dimer (X2) – hexamer (X6) model:

X1+X1k1k1X2+X2+X2k2k2X6

The equilibrium concentration of hexamer [X6] was calculated by numerical integration of the above model using scripts written in Python and converted to fractional saturation. The observed chemical shift (Δδ) is then given by:

Δδ=Bmax6*[Χ6][X1]

where Bmax is the maximum chemical shift difference. To obtain estimates for the monomer-monomer and dimer-hexamer Kds a grid search was performed by fixing the dimer Kd (k’1/k1) and the hexamer Kd (k’2/k2) to different values (Figure 2—figure supplement 2d). Excellent fits were produced using a dimer Kd <~50 μM, while the hexamer Kd shows a narrow distribution centered at ~10 ± 5 x 10−9 M (Benilova et al., 2012) (Figure 2—figure supplement 2d). To further validate the estimation of the dimer Kd, RDC experiments were performed as a function of ΔN6 concentration (Figure 2—figure supplement 2e). ΔN6 was aligned in 10 mg/mL of PF1 phage (Asla Scientific) and HN RDCs were measured using ARTSY (Fitzkee and Bax, 2010). The biphasic behavior of the RDCs suggests a three-state equilibrium in agreement with the monomer-dimer-hexamer model. The first/second transition at lower protein concentration (blue/pink dashed line in Figure 2—figure supplement 2e) presumably reports on the monomer-dimer/dimer-hexamer equilibrium, respectively. In order to extract RDCs of the dimer species the blue dashed line was extrapolated to 100% dimer using various Kd values. The resulting data were fitted to the structure of ΔN6 in order to calculate the alignment tensor of the dimer. Using a Kd greater than 50 µM results in a decrease in the goodness of the fit (Figure 2—figure supplement 2f), unless a large conformational change in the monomer is invoked upon dimer formation. However, based on the chemical shift data shown in Figure 2—figure supplement 2a,c ΔN6 remains native-like across all concentrations, thus placing an upper limit of the dimer Kd at ~50 μM in agreement with the grid search of the chemical shift data (Figure 2—figure supplement 2d). Note that the calculated tensor depends highly on the correct RDC values and therefore RDCs were not used in the structure calculations described below. Chemical shift perturbations for 10 residues that show significant chemical shift perturbations (11, 12, 23, 26, 50, 51, 52, 67, 68, 97) were fitted globally to this model, with representative examples shown in Figure 2—figure supplement 2b. Errors on the measured peak positions were calculated as the standard deviation of the mean for residues that show insignificant chemical shift changes. Errors on the fitted parameters were computed using Monte Carlo calculations with 100 steps.

To calculate populations of different species, a monomer-dimer-hexamer model was treated numerically, that is the kinetic equations that describe the time-evolution of the concentration of each species were integrated to τ=∞, after equilibrium was reached, yielding the equilibrium concentration (in molar units) of monomers, dimers and hexamers. Since the dimers consist of two monomers and hexamers of six monomers, these concentrations are then converted to populations (of monomers in the form of dimer or hexamer) using the relationship:

pn=n[An][Mtot]

where n is the oligomer order, [An] the equilibrium concentration of the oligomeric state and [Mtot] the total protein concentration. The overall rate of assembly, konover, for this three-state model is given by:

konover=k1appk2appk1k2app,

where

k1app=2k1[Meq]k2app=3k2[Deq]2

And therefore:

konover=k1appk2appk1k2app=6k1k2[Meq][Deq]2k1+3k2[Deq]2

The overall konover rate of assembly and therefore the total population of oligomers scales linearly as a function of the monomer concentration (see inset in Figure 2-Supplement 1f).

PRE experiments

The ΔN6 variants (14N-labeled) S20C, S33C, L54C and S61C (1–2 mg/mL) were incubated with 5 mM DTT for 20 min, excess DTT was removed using a PD10 gravity column (GE Healthcare) and the protein was then labeled immediately with MTSL by incubation with a 40-fold molar excess (over the total ΔN6 concentration) of the spin label for 4 hr in 25 mM sodium phosphate buffer, pH 7.0 containing 1 mM EDTA at room temperature. Excess spin label was removed by gel filtration (PD10 column) in the same buffer. Spin-labeled ΔN6 was used directly or stored at −80°C. In all cases, 100% labeling at a single site was confirmed using ESI-MS. For each PRE experiment, MTSL-labeled 14N-ΔN6 (10 μM −80 μM) was mixed with 15N-labeled ΔN6 (60 μM −240 μM) and the difference of the 1H R2 rates between oxidized and reduced (the latter created by addition of 1 mM ascorbic acid) MTSL-labelled 14N-ΔN6 was measured. Experiments were performed in 10 mM sodium phosphate buffer, pH 6.2 containing 83.3 mM NaCl or 10 mM sodium phosphate buffer, pH 8.2 containing 86.6 mM NaCl (a total ionic strength of 100 mM at each pH value). Data were recorded at 25°C using a 1H-15N correlation based pulse sequence with 5 or 6 time-points (0.0016–0.016 s) (Clore and Iwahara, 2009) and at least 32 scans per incremental delay, utilizing a Bruker Avance III 750MHz spectrometer equipped with a cryogenic probe. R2 rates were extracted by fitting the relaxation data to single exponentials using in-house scripts. The HN2 rate was then calculated as the difference between the R2 rate in the paramagnetic (R2, para) versus diamagnetic (R2, dia) sample:

Γ2=R2,paraR2,dia

Errors were calculated based on the noise of the experiment. The small PRE signal observed when ΔN6 is modified with MTSL at position 20 can be attributed to non-specific binding of the spin label itself to adjacent protein molecules, since addition of free MTSL resulted in a similar PRE profile (not shown). Thus, data arising from spin-labeled ΔN6 at position 20 were not included in quantitative analysis of the PRE experiments.

15N transverse relaxation dispersion CPMG experiments

15N transverse relaxation dispersion CPMG experiments were performed as described in reference (Loria et al., 1999) using samples dissolved in 10 mM sodium phosphate buffer containing 83.3 or 86.6 mM NaCl to maintain a constant ionic strength of 100 mM at pH 6.2 and 8.2, respectively. Spectra were acquired using a Varian Inova 500 MHz spectrometer using a fixed relaxation delay (τcpmg) of 48 ms or a Bruker Avance III 750 MHz or 950 MHz spectrometer using a delay of 40 ms. Spectra were processed using NMRPipe (Delaglio et al., 1995) and R2,eff rates were calculated using:

R2,eff=(IxI0)τCPMG

where Ix is the peak intensity in each experiment and I0 is the peak intensity in the reference spectrum (with CPMG train applied). CPMG data from two clusters of residues, one reporting on intermolecular interactions (12 residues) and the second reporting on the dynamics of the G strand (four residues) (see text) were fitted globally to the Bloch-McConnell equations (McConnell, 1958) describing a two-state exchanging system using the software package ‘relax’ (Morin et al., 2014). The fact that dimer and hexamer interfaces partly overlap, complicates the analysis of the CPMG data at pH 6.2. However, at the concentrations used, where either hexamerization is low (180 μM: 26% monomer, 48% ΔN6 molecules as dimer, 26% ΔN6 molecules as hexamer) or dimerization remains constant (480 μM: 13% monomer, 32% ΔN6 molecules as dimer, 55% ΔN6 molecules as hexamer) good quality fits to a simple two-state model were obtained. The calculated exchange parameters report on both dimer and hexamer formation. Due to this ambiguity, the apparent exchange rates obtained by fitting the CPMG data were not used in the kinetic modeling of the reaction, but used instead to report on the apparent differences in exchange dynamics of different residues as hexamer formation is enhanced.

Calculation of structural models

Structural models of dimers

Simulated annealing calculations were carried out in XPLOR-NIH (Schwieters et al., 2003). To account for the flexibility of the MTSL moiety, the paramagnetic group was represented as a five-membered ensemble. The computational strategy employed included three PRE potential terms (arising from S61C-ΔN6, L54C-ΔN6 and S33C-ΔN6) and classic geometry restraints to restrict deviation from bond lengths, angles and dihedral angles. Resonances for which an estimation of the R2 rate in the presence of the oxidized spin label was not possible were incorporated in the protocol as nOe-type restraints with an upper bound of 11.5 Å and a lower bound of 9 Å. Chemical shift perturbations observed upon binding were incorporated as sparse, highly ambiguous intermolecular distance restraints as described in reference (Clore and Schwieters, 2003). As chemical shifts can be influenced by numerous factors upon protein-protein interaction, the treatment of the derived data undertaken here results in a loose potential term that is unlikely to bias the structure calculation. Finally, the protocol included a weak radius of gyration restraint (Rgyr) calculated as 2.2*N0.38, where N is the number of residues in the complex. Rgyr is required in order to prevent bias towards more extended structures and tends to underestimate the true value of the radius of gyration (Kuszewski et al., 1999). C2 distance symmetry restraints alongside a non-crystallographic symmetry potential term were also implemented to ensure that the two monomers adopt the same conformation in the dimer. The aforementioned potential terms were then used in a rigid-body energy minimization/simulated annealing in torsion angle space protocol to minimize the difference between the observed and calculated Γ2 rates, starting from random orientations. The first step in the structure calculation consisted of 5000 steps of energy minimization against only the sparse chemical shift restraints, followed by simulated annealing dynamics with all the potential terms active, where the temperature is slowly decreased (3000–25 K) over four fs. During the hot phase (T = 3000 K) the PRE and nOe terms were underweighted to allow the proteins to sample a large conformational space and they were geometrically increased during the cooling phase. Proteins were treated as rigid bodies until the initiation of the cooling phase, where side chains were allowed to float (semi-rigid body calculation). The final step included torsion angle minimization using all potential terms. The calculations converged to two dimer structures shown in Figure 4a (lowest energy, termed dimer A) and Figure 4—figure supplement 1e (dimer B). Both dimers show a head–head configuration with dimer B showing a larger interface which extends from the BC and DE loops to the B and E strands. Fits to the PRE data are of lower quality for dimer B as judged by visual inspection of the fits and the restraints violation (RMS) (Table 1). However, both dimers were used as initial building blocks for calculation of the hexamer models.

Structural models of hexamers

Starting from dimer A or dimer B, an initial docking run was performed. Dimers were treated as rigid bodies and placed at random positions. Residues for which chemical shift differences were observed at high protein concentrations were used as sparse distance restraints alongside a geometry energy potential. Three-fold symmetry was imposed together with a non-crystallographic symmetry potential. The energy arising from the four potential energy terms was minimized in order to generate 1000 hexamer structures. The PRE potential energy was not used during the calculation but only in the scoring of the structures generated (together with the energy of the other four terms). Starting from dimer A, the plot of energy versus RMSD (to the lowest energy structure) (Figure 5—figure supplement 3a) shows the expected funnel shape with 44 of the 50 lowest energy structures sharing a backbone RMSD of 2–3 Å, indicating that these models are close to a structure that satisfies the PRE restraints. On the other hand, the 50 lowest energy hexamers built form dimer B show an RMSD of up to 35 Å with three clusters formed (Figure 5—figure supplement 3b). Therefore, these four hexamer structures (one arising from dimer A and three from dimer B) were taken forward for the next round of the protocol which consisted of an exhaustive simulated annealing calculation. Since it is difficult to define the extent to which the PREs arise from the dimer and hexamer, the PREs restraints were converted to distance restraints. Residues that show high PREs such that no peak was observed in the spectrum with oxidized MTSL were given no lower bound, while residues not affected by MTSL had no upper bound. This strategy removed some of the dimer – hexamer ambiguity and instead the protocol searched for hexamers that generally interact in the areas which show increased Γ2 rates at high ΔN6 concentrations, rather than quantitatively fitting the PRE data. The details of the simulated annealing run were similar to that performed to calculate the dimer structure, but included a three-fold (instead of two-fold) distance symmetry potential term (giving rise to hexamers with a D3 overall symmetry). The final stage of the protocol consisted of refinement in explicit water using XPLOR-NIH. Distances were converted back to PREs to allow comparison with the measured PRE data. Following this protocol, the hexamers produced from dimer A show increased PRE rates in the A strand and BC, DE loops in agreement with the PRE data (Figure 5—figure supplement 4). On the other hand, all hexamers assembled from dimer B show calculated PREs which describe the measured PREs less well (Table 2) (these fits are available on the University of Leeds publicly available library [https://doi.org/10.5518/329]). Note that dimer and hexamer models were generated and selected based only on the agreement with the NMR data. Cross-sections of the oligomers obtained from other experiments were used only as a check of consistency with the models determined. PDBs of the dimers and hexamers have been deposited in the University of Leeds publicly available library (https://doi.org/10.5518/329). The buried surface areas of dimers and hexamers were calculated using the program NACCESS (Hubbard and Thornton, 1993) which calculates the per residue accessible surface area (ASA) given a structural model. A cutoff of 10% loss in ASA between monomers and dimers/hexamers was used.

Native ESI-IMS-MS

ΔN6 samples were exchanged into a buffer consisting of 50 mM ammonium acetate, 50 mM ammonium bicarbonate pH 7.4 using Zeba spin desalting columns (Thermo Scientific) immediately before MS analysis. NanoESI–IMS–MS spectra were acquired using a Synapt HDMS mass spectrometer (Waters) with platinum/gold-plated borosilicate capillaries prepared in house. Typical instrument parameters were: capillary voltage, 1.2–1.6 kV; cone voltage, 40 V; trap collision voltage, 6 V; transfer collision voltage, 10 V; trap DC bias, 20 V; backing pressure, 4.5 mbar; IMS gas pressure, 0.5 mbar; traveling wave height, 7 V; and traveling wave velocity, 250 ms−1. Data were processed with MassLynx v4.1 and Driftscope 2.5 (Waters). Collison cross sections (CCSs) were estimated through a calibration approach using arrival-time data for ions with known CCS values (β-lactoglobulin A, avidin, concanavilin A and yeast alcohol dehydrogenase, all from Sigma Aldrich). Estimated modal CCSs are shown. Theoretical CCSs were calculated for hexameric model structures using the scaled projection approximation method implemented in IMPACT (Marklund et al., 2015) after performing in vacuo molecular dynamics simulations to account for structural alterations arising from transfer into the gas-phase, as previously described (Devine et al., 2017). Note that the best scoring model agrees with the CCS of the lowest charge state (15+) (which is considered to be most native; Vahidi et al., 2013) of the hexamer derived independently using the NMR data alone. The IMS-MS experiments thus serve as an independent validation of the structural model derived.

ANS binding

The ability of different ΔN6 species to bind 8-anilinonaphthalene-1-sulfonic acid (ANS) was measured by mixing 50 µL of each fraction obtained from analytical SEC of 1 μM ΔN6 (see above) with 200 µL of ANS to yield a final concentration of ANS of 200 µM. Fluorescence spectra were recorded using a ClarioStar plate reader (BMG Labtech) using an excitation wavelength of 370 nm and emission from 400 to 600 nm. The concentration of protein used was estimated to be ~240 μM (monomer), 3 μM (dimer) and 1 μM (hexamer). Experiments on Im7 L53A I54A were performed as described in reference (Spence et al., 2004).

Cytotoxicity assays

ΔN6 (240 μM) was cross-linked with EDC/NHS as described above. 500 µL of cross-linked material was resolved using a Superdex 75 analytical gel filtration column (GE Healthcare) using Dulbecco’s PBS as a mobile phase (Sigma #D8537). 1 mL fractions were collected. SH-SY5Y cells were obtained from an authenticated and mycoplasma free source (ATCC CRL-2266) and were passaged up to 10 times. The cells were mycoplasma tested and found to be negative. The cells were cultured as described previously (Xue et al., 2009) using 15,000 cells per well in 96-well plates (Corning #3595) for 24 hr in 100 µl of growth medium. This time point has been widely used in other studies of cytotoxicity and hence allows comparison of the results obtained with observations on β2m and other amyloid systems (Fusco et al., 2017; Campioni et al., 2010; Xue et al., 2009; Leri et al., 2016; Giorgetti et al., 2008).

Cells were then incubated with 50 µL of each fraction from SEC for 24 hr before analyzing cell viability. PBS alone was used as negative control and 0.02% (w/v) NaN3 was added as a positive control for cell death. Each experiment consisted of at least three repeats from two independent cross-linking reactions. The neuroblastoma cell line SH-SY5Y was chosen for our assays, as this cell line is a widely accepted model for the study of amyloid toxicity and has been used by other laboratories for β2m and other amyloid-forming sequences (Fusco et al., 2017; Campioni et al., 2010; Xue et al., 2009; Leri et al., 2016; Giorgetti et al., 2008).

For MTT assays, 10 µL of a 10 mg/mL solution of MTT (Sigma-Aldrich) was added to each well for 1.5 hr. Cell growth media and excess MTT were then removed and reduced MTT was solubilized using 50 µL DMSO per well. The absorbance of MTT was determined using a ClarioStar plate reader (BMG Labtech) at 570 nm with background subtraction at 650 nm. MTT reduction was calculated as a percentage of PBS buffer treated controls (100%) and cells treated with 0.02% (w/v) NaN3 (0%).

Cellular ATP was measured using the ATPlite Luminescence ATP detection assay (#6016963 Perkin Elmer) according to the manufacturer’s protocol. Luminescence was measured on a PolarStar OPTIMA plate reader (BMG Labtech). Cellular ATP was calculated as a percentage of PBS-buffer-treated controls (100%) and cells treated with 0.02% (w/v) NaN3 (0%).

Lactate dehydrogenase (LDH) release was measured using a Pierce LDH cytotoxicity assay kit (#88953 ThermoFisher Scientific) according to the manufacturer’s instructions. Absorbance was determined using a ClarioStar plate reader (BMG Labtech) at 490 nm with background subtraction at 680 nm. LDH release was calculated and normalized to detergent lysed cells (100%) and PBS-buffer-treated controls (0%).

Reactive oxygen species (ROS) production was determined using 10 µM 2’,7’-dichlorohydrofluorescein diacetate (H2DCFDA) (#D399 ThermoFisher Scientific). Cells were incubated with H2DCFDA for 10 min prior to the addition of ΔN6 samples from SEC. Fluorescence was recorded after further incubation for 45 min using a ClarioStar plate reader (BMG Labtech) at 540 nm. ROS production was calculated as a percentage of PBS buffer treated controls (100%) and cells treated with 0.02% (w/v) NaN3 (0%). 10 μM H2O2 was used as a positive control for the induction of ROS production and resulted in a 373 ± 21% ROS assay signal compared with incubation with PBS. Each experiment consisted typically of two-to-three independent experiments each containing five replicates per condition. The error bars represent mean S.E, * p 0.05. Raw data are available at (https://doi.org/10.5518/329).

Kinetic modeling of the rates of amyloid formation

The fibril growth kinetics for ΔN6 in the presence of ΔN6 fibril seeds shown in Figure 1c,d were fitted to five different kinetic models which consisted of two distinct modules (pre-polymerization and polymerization). In model (1) monomers are assumed to add to the fibril ends (this model contains two parameters, the elongation rate, ke, and a fibril depolymerization rate, ke). In model (2) the monomers are assumed to be in conformational exchange with a monomeric excited state that is responsible for elongation. Model (3) includes a monomer-dimer equilibrium followed by dimer addition to the fibril ends. Models (2) and (3) contain four parameters, monomer-monomer binding/unbinding rates (k1 and k1) and monomer conformational exchange rates (ke, ke). In the fourth model (4) a monomer-dimer-tetramer-hexamer equilibrium was considered. Finally, in model (5) a monomer-dimer-hexamer equilibrium was considered. This model contains six parameters, k1, k1 (monomer-monomer binding), k2, k2 (dimer-dimer-dimer binding) and ke, ke (exchange). The rate equations for all models are listed in Table 4 and were solved numerically using in house scripts written in Python. In the polymerization module, that describes the addition of the elongation unit (X) to the already formed fibrils, each assembly step was modeled explicitly (Table 4). The primary output of each model is the mass fraction of each species as a function of time. To convert the output of the program to ThT fluorescence curves, the mass of the elongated seeds was multiplied by a fluorescence factor (Ktht). Elongated seeds were assumed to be any species (Fi) that contain more monomers than the preformed seeds added in the assay (F0) (1 ≤ i ≤ N), where N represents the number of monomers in the fibril at the end of the reaction. The mass fraction of monomers present in a fibril was assumed to scale linearly with ThT fluorescence, giving the following expression for calculating the progress curves:

Fi(t)=i=1Ni[Fi]Ktht

Seeding data using all five starting ΔN6 monomer concentrations were fitted to each model globally sharing all rate constants using N = 200 (200 monomers in a fibril which would correspond to a fibril roughly 500 nm in length; White et al., 2009). The monomer-dimer Kd value (k1/k1) was fixed to 50 μM. Fitting the kinetic data to the hexamer addition model produces a hexamer Kd of ~1.9×10−9 M2 similar to the value of ~10 ± 5 x 10−9 M (Benilova et al., 2012) Kd obtained by fitting the chemical shift perturbation data on protein concentration, confirming the robustness of the model and the approach employed. Using the estimated Kd values to obtain the populations of dimer and hexamer (Pdim, Phex) and the structural models to calculate correlation times of the dimers and hexamers (τc,dim, τc,hex), the apparent correlation time at each ΔN6 concentration (τc,app) (computed as τc,app = Pmon τc,mon + Pdim τc,dim + Phex τc,hex, where Pmon/dim/hex is the population of dimer/hexamer, respectively and τc,mon/dim/hex is the correlation time of each species (9.8, 18.5 and 60.3 ns, respectively) calculated using the structural models by HYDROPRO (Ortega et al., 2011) matches the NMR measured τc versus ΔN6 concentration (Figure 7b).

Acknowledgements

We thank members of the Radford laboratory for helpful discussions, Nasir Khan for his excellent technical support and Amy Barker for her assistance with AUC. We also thank Alison Ashcroft for her long-term collaboration on native MS. TKK, SCG, EEC, EWH and SER acknowledge funding from the Wellcome Trust (089311/Z/09/Z, 204963 and 109154/Z/15/Z) and the European Research Council (ERC) under European Union’s Seventh Framework Programme (FP7/2007-2013) ERC grant agreement no. 322408. ANC is funded by the BBSRC (BB/K000659/1). We acknowledge the Wellcome Trust (094232) and the University of Leeds for funding the NMR instrumentation and the BBSRC (BB/E012558/1) for providing funds for the Synapt HDMS mass spectrometer.

Funding Statement

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Contributor Information

Sheena E Radford, Email: s.e.radford@leeds.ac.uk.

Tricia R Serio, The University of Massachusetts, Amherst, United States.

John Kuriyan, University of California, Berkeley, United States.

Funding Information

This paper was supported by the following grants:

  • Wellcome Trust 089311/Z/09/Z to Theodoros K Karamanos, Sheena E Radford.

  • Wellcome Trust 204963 to Sheena E Radford.

  • Wellcome Trust 109154/Z/15/Z to Emma E Cawood, Sheena E Radford.

  • European Research Council 322408 to Theodoros K Karamanos, Matthew P Jackson, Sheena E Radford.

  • Biotechnology and Biological Sciences Research Council BB/K000659/1 to Antonio N Calabrese, Sheena E Radford.

  • Wellcome Trust 094232 to Arnout P Kalverda, Sheena E Radford.

  • Biotechnology and Biological Sciences Research Council BB/E012558/1 to Sheena E Radford.

  • Wellcome Trust 092896MA to Theodoros K Karamanos, Sophia C Goodchild, Sheena E Radford, Eric W Hewitt.

Additional information

Competing interests

No competing interests declared.

Author contributions

Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing.

Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing.

Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing.

Conceptualization, Data curation, Formal analysis, Investigation, Writing—review and editing.

Conceptualization, Investigation, Writing—review and editing.

Conceptualization, Formal analysis, Investigation, Writing—review and editing.

Formal analysis, Methodology, Writing—review and editing.

Conceptualization, Supervision, Investigation, Methodology, Writing—original draft, Project administration, Writing—review and editing.

Conceptualization, Supervision, Funding acquisition, Writing—original draft, Project administration, Writing—review and editing.

Additional files

Transparent reporting form
DOI: 10.7554/eLife.46574.031

Data availability

Data are freely available at the University of Leeds Data Depository: https://doi.org/10.5518/329.

The following dataset was generated:

Karamanos TK, Jackson MP, Calabrese AN, Goodchild SC, Cawood EE, Thompson GS, Kalverda AP, Hewitt EW, Radford SE. 2019. Data from: Structural mapping of oligomeric intermediates in an amyloid assembly pathway. University of Leeds Data Depository.

References

  1. Anthis NJ, Clore GM. Visualizing transient dark states by NMR spectroscopy. Quarterly Reviews of Biophysics. 2015;48:35–116. doi: 10.1017/S0033583514000122. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Antwi K, Mahar M, Srikanth R, Olbris MR, Tyson JF, Vachet RW. Cu(II) organizes beta-2-microglobulin oligomers but is released upon amyloid formation. Protein Science. 2008;17:748–759. doi: 10.1110/ps.073249008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Apostol MI, Perry K, Surewicz WK. Crystal structure of a human prion protein fragment reveals a motif for oligomer formation. Journal of the American Chemical Society. 2013;135:10202–10205. doi: 10.1021/ja403001q. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Baskakov IV, Legname G, Baldwin MA, Prusiner SB, Cohen FE. Pathway complexity of prion protein assembly into amyloid. Journal of Biological Chemistry. 2002;277:21140–21148. doi: 10.1074/jbc.M111402200. [DOI] [PubMed] [Google Scholar]
  5. Bellotti V, Stoppini M, Mangione P, Sunde M, Robinson C, Asti L, Brancaccio D, Ferri G. Beta 2-microglobulin can be refolded into a native state from ex vivo amyloid fibrils. European Journal of Biochemistry. 1998;258:61–67. doi: 10.1046/j.1432-1327.1998.2580061.x. [DOI] [PubMed] [Google Scholar]
  6. Benilova I, Karran E, De Strooper B. The toxic aβ oligomer and Alzheimer's disease: an emperor in need of clothes. Nature Neuroscience. 2012;15:349–357. doi: 10.1038/nn.3028. [DOI] [PubMed] [Google Scholar]
  7. Benseny-Cases N, Karamanos TK, Hoop CL, Baum J, Radford SE. Extracellular matrix components modulate different stages in β2-microglobulin amyloid formation. Journal of Biological Chemistry. 2019;294:9392–9401. doi: 10.1074/jbc.RA119.008300. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bieschke J, Russ J, Friedrich RP, Ehrnhoefer DE, Wobst H, Neugebauer K, Wanker EE. EGCG remodels mature alpha-synuclein and amyloid-beta fibrils and reduces cellular toxicity. PNAS. 2010;107:7710–7715. doi: 10.1073/pnas.0910723107. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Bieschke J, Herbst M, Wiglenda T, Friedrich RP, Boeddrich A, Schiele F, Kleckers D, Lopez del Amo JM, Grüning BA, Wang Q, Schmidt MR, Lurz R, Anwyl R, Schnoegl S, Fändrich M, Frank RF, Reif B, Günther S, Walsh DM, Wanker EE. Small-molecule conversion of toxic oligomers to nontoxic β-sheet-rich amyloid fibrils. Nature Chemical Biology. 2011;8:93–101. doi: 10.1038/nchembio.719. [DOI] [PubMed] [Google Scholar]
  10. Blancas-Mejía LM, Misra P, Ramirez-Alvarado M. Differences in protein concentration dependence for nucleation and elongation in light chain amyloid formation. Biochemistry. 2017;56:757–766. doi: 10.1021/acs.biochem.6b01043. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Brown PH, Schuck P. Macromolecular size-and-shape distributions by sedimentation velocity analytical ultracentrifugation. Biophysical Journal. 2006;90:4651–4661. doi: 10.1529/biophysj.106.081372. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Buell AK, Blundell JR, Dobson CM, Welland ME, Terentjev EM, Knowles TP. Frequency factors in a landscape model of filamentous protein aggregation. Physical Review Letters. 2010;104:228101. doi: 10.1103/PhysRevLett.104.228101. [DOI] [PubMed] [Google Scholar]
  13. Buell AK, Galvagnion C, Gaspar R, Sparr E, Vendruscolo M, Knowles TP, Linse S, Dobson CM. Solution conditions determine the relative importance of nucleation and growth processes in α-synuclein aggregation. PNAS. 2014;111:7671–7676. doi: 10.1073/pnas.1315346111. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Calabrese MF, Eakin CM, Wang JM, Miranker AD. A regulatable switch mediates self-association in an immunoglobulin fold. Nature Structural & Molecular Biology. 2008;15:965–971. doi: 10.1038/nsmb.1483. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Calabrese MF, Miranker AD. Formation of a stable oligomer of beta-2 microglobulin requires only transient encounter with cu(II) Journal of Molecular Biology. 2007;367:1–7. doi: 10.1016/j.jmb.2006.12.034. [DOI] [PubMed] [Google Scholar]
  16. Campioni S, Mannini B, Zampagni M, Pensalfini A, Parrini C, Evangelisti E, Relini A, Stefani M, Dobson CM, Cecchi C, Chiti F. A causative link between the structure of aberrant protein oligomers and their toxicity. Nature Chemical Biology. 2010;6:140–147. doi: 10.1038/nchembio.283. [DOI] [PubMed] [Google Scholar]
  17. Chiti F, De Lorenzi E, Grossi S, Mangione P, Giorgetti S, Caccialanza G, Dobson CM, Merlini G, Ramponi G, Bellotti V. A partially structured species of beta 2-microglobulin is significantly populated under physiological conditions and involved in fibrillogenesis. Journal of Biological Chemistry. 2001;276:46714–46721. doi: 10.1074/jbc.M107040200. [DOI] [PubMed] [Google Scholar]
  18. Chiti F, Dobson CM. Protein misfolding, amyloid formation, and human disease: a summary of progress over the last decade. Annual Review of Biochemistry. 2017;86:27–68. doi: 10.1146/annurev-biochem-061516-045115. [DOI] [PubMed] [Google Scholar]
  19. Clore GM, Iwahara J. Theory, practice, and applications of paramagnetic relaxation enhancement for the characterization of transient low-population states of biological macromolecules and their complexes. Chemical Reviews. 2009;109:4108–4139. doi: 10.1021/cr900033p. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Clore GM, Schwieters CD. Docking of protein-protein complexes on the basis of highly ambiguous intermolecular distance restraints derived from 1H/15N chemical shift mapping and backbone 15N-1H residual dipolar couplings using conjoined rigid body/torsion angle dynamics. Journal of the American Chemical Society. 2003;125:2902–2912. doi: 10.1021/ja028893d. [DOI] [PubMed] [Google Scholar]
  21. Close W, Neumann M, Schmidt A, Hora M, Annamalai K, Schmidt M, Reif B, Schmidt V, Grigorieff N, Fändrich M. Physical basis of amyloid fibril polymorphism. Nature Communications. 2018;9:699. doi: 10.1038/s41467-018-03164-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Cohen SI, Linse S, Luheshi LM, Hellstrand E, White DA, Rajah L, Otzen DE, Vendruscolo M, Dobson CM, Knowles TP. Proliferation of amyloid-β42 aggregates occurs through a secondary nucleation mechanism. PNAS. 2013;110:9758–9763. doi: 10.1073/pnas.1218402110. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Cohen SIA, Cukalevski R, Michaels TCT, Šarić A, Törnquist M, Vendruscolo M, Dobson CM, Buell AK, Knowles TPJ, Linse S. Distinct thermodynamic signatures of oligomer generation in the aggregation of the amyloid-β peptide. Nature Chemistry. 2018;10:523–531. doi: 10.1038/s41557-018-0023-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Colombo M, de Rosa M, Bellotti V, Ricagno S, Bolognesi M. A recurrent D-strand association interface is observed in β-2 microglobulin oligomers. FEBS Journal. 2012;279:1131–1143. doi: 10.1111/j.1742-4658.2012.08510.x. [DOI] [PubMed] [Google Scholar]
  25. Colvin MT, Silvers R, Ni QZ, Can TV, Sergeyev I, Rosay M, Donovan KJ, Michael B, Wall J, Linse S, Griffin RG. Atomic resolution structure of monomorphic aβ42 amyloid fibrils. Journal of the American Chemical Society. 2016;138:9663–9674. doi: 10.1021/jacs.6b05129. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Cremades N, Cohen SI, Deas E, Abramov AY, Chen AY, Orte A, Sandal M, Clarke RW, Dunne P, Aprile FA, Bertoncini CW, Wood NW, Knowles TP, Dobson CM, Klenerman D. Direct observation of the interconversion of normal and toxic forms of α-synuclein. Cell. 2012;149:1048–1059. doi: 10.1016/j.cell.2012.03.037. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Debelouchina GT, Platt GW, Bayro MJ, Radford SE, Griffin RG. Intermolecular alignment in β2-microglobulin amyloid fibrils. Journal of the American Chemical Society. 2010;132:17077–17079. doi: 10.1021/ja107987f. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Delaglio F, Grzesiek S, Vuister GW, Zhu G, Pfeifer J, Bax A. NMRPipe: a multidimensional spectral processing system based on UNIX pipes. Journal of Biomolecular NMR. 1995;6:277–293. doi: 10.1007/BF00197809. [DOI] [PubMed] [Google Scholar]
  29. Devine PWA, Fisher HC, Calabrese AN, Whelan F, Higazi DR, Potts JR, Lowe DC, Radford SE, Ashcroft AE. Investigating the structural compaction of biomolecules upon transition to the Gas-Phase using ESI-TWIMS-MS. Journal of the American Society for Mass Spectrometry. 2017;28:1855–1862. doi: 10.1007/s13361-017-1689-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Diomede L, Soria C, Romeo M, Giorgetti S, Marchese L, Mangione PP, Porcari R, Zorzoli I, Salmona M, Bellotti V, Stoppini M. C. elegans expressing human β2-microglobulin: a novel model for studying the relationship between the molecular assembly and the toxic phenotype. PLOS ONE. 2012;7:e52314. doi: 10.1371/journal.pone.0052314. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Domanska K, Vanderhaegen S, Srinivasan V, Pardon E, Dupeux F, Marquez JA, Giorgetti S, Stoppini M, Wyns L, Bellotti V, Steyaert J. Atomic structure of a nanobody-trapped domain-swapped dimer of an amyloidogenic beta2-microglobulin variant. PNAS. 2011;108:1314–1319. doi: 10.1073/pnas.1008560108. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Eakin CM, Attenello FJ, Morgan CJ, Miranker AD. Oligomeric assembly of native-like precursors precedes amyloid formation by beta-2 microglobulin. Biochemistry. 2004;43:7808–7815. doi: 10.1021/bi049792q. [DOI] [PubMed] [Google Scholar]
  33. Eakin CM, Berman AJ, Miranker AD. A native to amyloidogenic transition regulated by a backbone trigger. Nature Structural & Molecular Biology. 2006;13:202–208. doi: 10.1038/nsmb1068. [DOI] [PubMed] [Google Scholar]
  34. Eichner T, Kalverda AP, Thompson GS, Homans SW, Radford SE. Conformational conversion during amyloid formation at atomic resolution. Molecular Cell. 2011;41:161–172. doi: 10.1016/j.molcel.2010.11.028. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Eisenberg D, Schwarz E, Komaromy M, Wall R. Analysis of membrane and surface protein sequences with the hydrophobic moment plot. Journal of Molecular Biology. 1984;179:125–142. doi: 10.1016/0022-2836(84)90309-7. [DOI] [PubMed] [Google Scholar]
  36. Esposito G, Michelutti R, Verdone G, Viglino P, Hernández H, Robinson CV, Amoresano A, Dal Piaz F, Monti M, Pucci P, Mangione P, Stoppini M, Merlini G, Ferri G, Bellotti V. Removal of the N-terminal hexapeptide from human beta2-microglobulin facilitates protein aggregation and fibril formation. Protein Science. 2000;9:831–845. doi: 10.1110/ps.9.5.831. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Fitzkee NC, Bax A. Facile measurement of ¹H-¹5N residual dipolar couplings in larger perdeuterated proteins. Journal of Biomolecular NMR. 2010;48:65–70. doi: 10.1007/s10858-010-9441-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Fitzpatrick AWP, Falcon B, He S, Murzin AG, Murshudov G, Garringer HJ, Crowther RA, Ghetti B, Goedert M, Scheres SHW. Cryo-EM structures of tau filaments from Alzheimer's disease. Nature. 2017;547:185–190. doi: 10.1038/nature23002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Floege J, Ehlerding G. β2-microglobulin-associated amyloidosis. Nephron. 1996;72:9–26. doi: 10.1159/000188801. [DOI] [PubMed] [Google Scholar]
  40. Fusco G, Chen SW, Williamson PTF, Cascella R, Perni M, Jarvis JA, Cecchi C, Vendruscolo M, Chiti F, Cremades N, Ying L, Dobson CM, De Simone A. Structural basis of membrane disruption and cellular toxicity by α-synuclein oligomers. Science. 2017;358:1440–1443. doi: 10.1126/science.aan6160. [DOI] [PubMed] [Google Scholar]
  41. Gejyo F, Yamada T, Odani S, Nakagawa Y, Arakawa M, Kunitomo T, Kataoka H, Suzuki M, Hirasawa Y, Shirahama T. A new form of amyloid protein associated with chronic hemodialysis was identified as β2-microglobulin. Biochemical and Biophysical Research Communications. 1985;129:701–706. doi: 10.1016/0006-291X(85)91948-5. [DOI] [PubMed] [Google Scholar]
  42. Giorgetti S, Raimondi S, Cassinelli S, Bucciantini M, Stefani M, Gregorini G, Albonico G, Moratti R, Montagna G, Stoppini M, Bellotti V. β 2-Microglobulin is potentially neurotoxic, but the blood brain barrier is likely to protect the brain from its toxicity. Nephrology Dialysis Transplantation. 2008;24:1176–1181. doi: 10.1093/ndt/gfn623. [DOI] [PubMed] [Google Scholar]
  43. Halabelian L, Relini A, Barbiroli A, Penco A, Bolognesi M, Ricagno S. A covalent homodimer probing early oligomers along amyloid aggregation. Scientific Reports. 2015;5:14651. doi: 10.1038/srep14651. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Hansen DF, Vallurupalli P, Kay LE. Using relaxation dispersion NMR spectroscopy to determine structures of excited, invisible protein states. Journal of Biomolecular NMR. 2008;41:113–120. doi: 10.1007/s10858-008-9251-5. [DOI] [PubMed] [Google Scholar]
  45. Hubbard SJ, Thornton JM. Department of Biochemistry and Molecular Biology, University College London, London, United Kingdom: 1993. [Google Scholar]
  46. Hurshman AR, White JT, Powers ET, Kelly JW. Transthyretin aggregation under partially denaturing conditions is a downhill polymerization. Biochemistry. 2004;43:7365–7381. doi: 10.1021/bi049621l. [DOI] [PubMed] [Google Scholar]
  47. Iadanza MG, Jackson MP, Hewitt EW, Ranson NA, Radford SE. A new era for understanding amyloid structures and disease. Nature Reviews Molecular Cell Biology. 2018a;19:755–773. doi: 10.1038/s41580-018-0060-8. [DOI] [PubMed] [Google Scholar]
  48. Iadanza MG, Silvers R, Boardman J, Smith HI, Karamanos TK, Debelouchina GT, Su Y, Griffin RG, Ranson NA, Radford SE. The structure of a β2-microglobulin fibril suggests a molecular basis for its amyloid polymorphism. Nature Communications. 2018b;9:4517. doi: 10.1038/s41467-018-06761-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Jakhria T, Hellewell AL, Porter MY, Jackson MP, Tipping KW, Xue WF, Radford SE, Hewitt EW. β2-microglobulin amyloid fibrils are nanoparticles that disrupt lysosomal membrane protein trafficking and inhibit protein degradation by lysosomes. Journal of Biological Chemistry. 2014;289:35781–35794. doi: 10.1074/jbc.M114.586222. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Karamanos TK, Kalverda AP, Thompson GS, Radford SE. Visualization of transient protein-protein interactions that promote or inhibit amyloid assembly. Molecular Cell. 2014;55:214–226. doi: 10.1016/j.molcel.2014.05.026. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Karamanos TK, Pashley CL, Kalverda AP, Thompson GS, Mayzel M, Orekhov VY, Radford SE. A population shift between sparsely populated folding intermediates determines amyloidogenicity. Journal of the American Chemical Society. 2016;138:6271–6280. doi: 10.1021/jacs.6b02464. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Kuszewski J, Gronenborn AM, Clore GM. Improving the packing and accuracy of NMR structures with a pseudopotential for the radius of gyration. Journal of the American Chemical Society. 1999;121:2337–2338. doi: 10.1021/ja9843730. [DOI] [Google Scholar]
  53. Laganowsky A, Liu C, Sawaya MR, Whitelegge JP, Park J, Zhao M, Pensalfini A, Soriaga AB, Landau M, Teng PK, Cascio D, Glabe C, Eisenberg D. Atomic view of a toxic amyloid small oligomer. Science. 2012;335:1228–1231. doi: 10.1126/science.1213151. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Lee D, Hilty C, Wider G, Wüthrich K. Effective rotational correlation times of proteins from NMR relaxation interference. Journal of Magnetic Resonance. 2006;178:72–76. doi: 10.1016/j.jmr.2005.08.014. [DOI] [PubMed] [Google Scholar]
  55. Lenton S, Grimaldo M, Roosen-Runge F, Schreiber F, Nylander T, Clegg R, Holt C, Härtlein M, García Sakai V, Seydel T, Marujo Teixeira SC. Effect of phosphorylation on a Human-like osteopontin peptide. Biophysical Journal. 2017;112:1586–1596. doi: 10.1016/j.bpj.2017.03.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Leri M, Bemporad F, Oropesa-Nuñez R, Canale C, Calamai M, Nosi D, Ramazzotti M, Giorgetti S, Pavone FS, Bellotti V, Stefani M, Bucciantini M. Molecular insights into cell toxicity of a novel familial amyloidogenic variant of β2-microglobulin. Journal of Cellular and Molecular Medicine. 2016;20:1443–1456. doi: 10.1111/jcmm.12833. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Liu C, Sawaya MR, Eisenberg D. β₂-microglobulin forms three-dimensional domain-swapped amyloid fibrils with disulfide linkages. Nature Structural & Molecular Biology. 2011;18:49–55. doi: 10.1038/nsmb.1948. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Loria JP, Rance M, Palmer AG. A TROSY CPMG sequence for characterizing chemical exchange in large proteins. Journal of Biomolecular NMR. 1999;15:151–155. doi: 10.1023/a:1008355631073. [DOI] [PubMed] [Google Scholar]
  59. Lu JX, Qiang W, Yau WM, Schwieters CD, Meredith SC, Tycko R. Molecular structure of β-amyloid fibrils in Alzheimer’s disease brain tissue. Cell. 2013;154:1257–1268. doi: 10.1016/j.cell.2013.08.035. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Mannini B, Mulvihill E, Sgromo C, Cascella R, Khodarahmi R, Ramazzotti M, Dobson CM, Cecchi C, Chiti F. Toxicity of protein oligomers is rationalized by a function combining size and surface hydrophobicity. ACS Chemical Biology. 2014;9:2309–2317. doi: 10.1021/cb500505m. [DOI] [PubMed] [Google Scholar]
  61. Marklund EG, Degiacomi MT, Robinson CV, Baldwin AJ, Benesch JL. Collision cross sections for structural proteomics. Structure. 2015;23:791–799. doi: 10.1016/j.str.2015.02.010. [DOI] [PubMed] [Google Scholar]
  62. McConnell HM. Reaction rates by nuclear magnetic resonance. The Journal of Chemical Physics. 1958;28:430–431. doi: 10.1063/1.1744152. [DOI] [Google Scholar]
  63. Mendoza VL, Antwi K, Barón-Rodríguez MA, Blanco C, Vachet RW. Structure of the preamyloid dimer of β2-microglobulin from covalent labeling and mass spectrometry. Biochemistry. 2010;49:1522–1532. doi: 10.1021/bi901748h. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Mendoza VL, Barón-Rodríguez MA, Blanco C, Vachet RW. Structural insights into the pre-amyloid tetramer of β-2-microglobulin from covalent labeling and mass spectrometry. Biochemistry. 2011;50:6711–6722. doi: 10.1021/bi2004894. [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Morin S, Linnet TE, Lescanne M, Schanda P, Thompson GS, Tollinger M, Teilum K, Gagné S, Marion D, Griesinger C, Blackledge M, d’Auvergne EJ. Relax: the analysis of biomolecular kinetics and thermodynamics using NMR relaxation dispersion data. Bioinformatics. 2014;30:2219–2220. doi: 10.1093/bioinformatics/btu166. [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. Ono K, Condron MM, Teplow DB. Structure-neurotoxicity relationships of amyloid β -protein oligomers. PNAS. 2009;106:14745–14750. doi: 10.1073/pnas.0905127106. [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Ortega A, Amorós D, García de la Torre J. Prediction of hydrodynamic and other solution properties of rigid proteins from atomic- and residue-level models. Biophysical Journal. 2011;101:892–898. doi: 10.1016/j.bpj.2011.06.046. [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Platt GW, Routledge KE, Homans SW, Radford SE. Fibril growth kinetics reveal a region of β2-microglobulin important for nucleation and elongation of aggregation. Journal of Molecular Biology. 2008;378:251–263. doi: 10.1016/j.jmb.2008.01.092. [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Platt GW, Radford SE. Glimpses of the molecular mechanisms of β2-microglobulin fibril formation in vitro: aggregation on a complex energy landscape. FEBS Letters. 2009;583:2623–2629. doi: 10.1016/j.febslet.2009.05.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Rennella E, Cutuil T, Schanda P, Ayala I, Gabel F, Forge V, Corazza A, Esposito G, Brutscher B. Oligomeric states along the folding pathways of β2-microglobulin: kinetics, thermodynamics, and structure. Journal of Molecular Biology. 2013;425:2722–2736. doi: 10.1016/j.jmb.2013.04.028. [DOI] [PubMed] [Google Scholar]
  71. Sangwan S, Zhao A, Adams KL, Jayson CK, Sawaya MR, Guenther EL, Pan AC, Ngo J, Moore DM, Soriaga AB, Do TD, Goldschmidt L, Nelson R, Bowers MT, Koehler CM, Shaw DE, Novitch BG, Eisenberg DS. Atomic structure of a toxic, oligomeric segment of SOD1 linked to amyotrophic lateral sclerosis (ALS) PNAS. 2017;114:8770–8775. doi: 10.1073/pnas.1705091114. [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. Sarell CJ, Woods LA, Su Y, Debelouchina GT, Ashcroft AE, Griffin RG, Stockley PG, Radford SE. Expanding the repertoire of amyloid polymorphs by Co-polymerization of related protein precursors. Journal of Biological Chemistry. 2013;288:7327–7337. doi: 10.1074/jbc.M112.447524. [DOI] [PMC free article] [PubMed] [Google Scholar]
  73. Schwieters CD, Kuszewski JJ, Tjandra N, Clore GM. The Xplor-NIH NMR molecular structure determination package. Journal of Magnetic Resonance. 2003;160:65–73. doi: 10.1016/S1090-7807(02)00014-9. [DOI] [PubMed] [Google Scholar]
  74. Semisotnov GV, Rodionova NA, Razgulyaev OI, Uversky VN, Gripas' AF, Gilmanshin RI. Study of the "molten globule" intermediate state in protein folding by a hydrophobic fluorescent probe. Biopolymers. 1991;31:119–128. doi: 10.1002/bip.360310111. [DOI] [PubMed] [Google Scholar]
  75. Sipe JD, Benson MD, Buxbaum JN, Ikeda S-ichi, Merlini G, Saraiva MJM, Westermark P. Amyloid fibril proteins and amyloidosis: chemical identification and clinical classification international society of amyloidosis 2016 nomenclature guidelines. Amyloid. 2016;23:209–213. doi: 10.1080/13506129.2016.1257986. [DOI] [PubMed] [Google Scholar]
  76. Souillac PO, Uversky VN, Fink AL. Structural transformations of oligomeric intermediates in the fibrillation of the immunoglobulin light chain LEN. Biochemistry. 2003;42:8094–8104. doi: 10.1021/bi034652m. [DOI] [PubMed] [Google Scholar]
  77. Spence GR, Capaldi AP, Radford SE. Trapping the on-pathway folding intermediate of Im7 at equilibrium. Journal of Molecular Biology. 2004;341:215–226. doi: 10.1016/j.jmb.2004.05.049. [DOI] [PubMed] [Google Scholar]
  78. Stejskal EO, Tanner JE. Spin diffusion measurements: spin echoes in the presence of a time‐dependent field gradient. The Journal of Chemical Physics. 1965;42:288–292. doi: 10.1063/1.1695690. [DOI] [Google Scholar]
  79. Stoppini M, Bellotti V. Systemic amyloidosis: lessons from β2-microglobulin. The Journal of Biological Chemistry. 2015;290:9951–9958. doi: 10.1074/jbc.R115.639799. [DOI] [PMC free article] [PubMed] [Google Scholar]
  80. Su Y, Sarell CJ, Eddy MT, Debelouchina GT, Andreas LB, Pashley CL, Radford SE, Griffin RG. Secondary structure in the core of amyloid fibrils formed from human β₂m and its truncated variant δn6. Journal of the American Chemical Society. 2014;136:6313–6325. doi: 10.1021/ja4126092. [DOI] [PMC free article] [PubMed] [Google Scholar]
  81. Vahidi S, Stocks BB, Konermann L. Partially disordered proteins studied by ion mobility-mass spectrometry: implications for the preservation of solution phase structure in the gas phase. Analytical Chemistry. 2013;85:10471–10478. doi: 10.1021/ac402490r. [DOI] [PubMed] [Google Scholar]
  82. Vestergaard B, Groenning M, Roessle M, Kastrup JS, van de Weert M, Flink JM, Frokjaer S, Gajhede M, Svergun DI. A helical structural nucleus is the primary elongating unit of insulin amyloid fibrils. PLOS Biology. 2007;5:e134. doi: 10.1371/journal.pbio.0050134. [DOI] [PMC free article] [PubMed] [Google Scholar]
  83. Vitalis A, Lyle N, Pappu RV. Thermodynamics of β-Sheet formation in polyglutamine. Biophysical Journal. 2009;97:303–311. doi: 10.1016/j.bpj.2009.05.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  84. Wei L, Jiang P, Xu W, Li H, Zhang H, Yan L, Chan-Park MB, Liu XW, Tang K, Mu Y, Pervushin K. The molecular basis of distinct aggregation pathways of islet amyloid polypeptide. Journal of Biological Chemistry. 2011;286:6291–6300. doi: 10.1074/jbc.M110.166678. [DOI] [PMC free article] [PubMed] [Google Scholar]
  85. White HE, Hodgkinson JL, Jahn TR, Cohen-Krausz S, Gosal WS, Müller S, Orlova EV, Radford SE, Saibil HR. Globular tetramers of β2-microglobulin assemble into elaborate amyloid fibrils. Journal of Molecular Biology. 2009;389:48–57. doi: 10.1016/j.jmb.2009.03.066. [DOI] [PMC free article] [PubMed] [Google Scholar]
  86. Wu JW, Breydo L, Isas JM, Lee J, Kuznetsov YG, Langen R, Glabe C. Fibrillar oligomers nucleate the oligomerization of monomeric amyloid β but do not seed fibril formation. The Journal of Biological Chemistry. 2010;285:6071–6079. doi: 10.1074/jbc.M109.069542. [DOI] [PMC free article] [PubMed] [Google Scholar]
  87. Xue WF, Homans SW, Radford SE. Systematic analysis of nucleation-dependent polymerization reveals new insights into the mechanism of amyloid self-assembly. PNAS. 2008;105:8926–8931. doi: 10.1073/pnas.0711664105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  88. Xue WF, Hellewell AL, Gosal WS, Homans SW, Hewitt EW, Radford SE. Fibril fragmentation enhances amyloid cytotoxicity. Journal of Biological Chemistry. 2009;284:34272–34282. doi: 10.1074/jbc.M109.049809. [DOI] [PMC free article] [PubMed] [Google Scholar]
  89. Yamamoto S, Hasegawa K, Yamaguchi I, Goto Y, Gejyo F, Naiki H. Kinetic analysis of the polymerization and depolymerization of beta(2)-microglobulin-related amyloid fibrils in vitro. Biochimica Et Biophysica Acta (BBA) - Proteins and Proteomics. 2005;1753:34–43. doi: 10.1016/j.bbapap.2005.07.007. [DOI] [PubMed] [Google Scholar]
  90. Zhang W, Falcon B, Murzin AG, Fan J, Crowther RA, Goedert M, Scheres SH. Heparin-induced tau filaments are polymorphic and differ from those in Alzheimer's and Pick's diseases. eLife. 2019;8:e43584. doi: 10.7554/eLife.43584. [DOI] [PMC free article] [PubMed] [Google Scholar]

Decision letter

Editor: Tricia R Serio1

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Structural mapping of oligomeric intermediates in an amyloid assembly pathway" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and John Kuriyan as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This manuscript from Radford and colleagues addresses the pathway of amyloid assembly by β2-microglobulin. The studies focus on characterizing both the structure and physiological impact of the assembly-competent species, using changes in pH and a truncated form of the protein to gain these insights. In particular, extensive experimental and modelling analyses are combined to gain unique insights into the early stages of amyloid formation for this protein, which has been found in amyloid deposits in patients with dialysis-related amyloidosis. The results provide an atomic view of ΔN6 retaining a native-like structure upon forming dimers and hexamers, which do not exhibit toxicity. This finding is noteworthy since oligomers have often been reported to be particularly cytotoxic, and the structure is distinct from other previously studied forms of the protein. The new results here help explain previous studies and provide clues as to how the major structural conversion to amyloid may occur.

Essential revisions:

The reviewers have identified a number of key questions, which should be addressed in a revised manuscript. The reviewers raised multiple questions regarding the conclusions and have requested additional information on the identity, quantitation, and thermodynamics to support them:

Hexamer abundance and structure:

1) The experimental determination of the hexamer Kd at ~10 ± 5 nM2 (ranging to 1.9 nM2 for kinetic modelling) is not consistent with the data. For example, using the values reported in the paper for 180 µM monomer of 48% dimer and 24% hexamer (subsection “Structural models of on-pathway hexamers”, first paragraph), assuming the remainder, 28%, is monomer, Kd for dimer to monomer is 59 µM and Kd for hexamer to dimer is 11x10-9 M2 = 11 x 109 nM2 = 11 x 103 µM2. This also will impact the derived ΔGº value. Use of the incorrect value for the hexamer Kd might have major consequences throughout the paper, though probably the conclusions will for the most part be unaltered. Elsewhere in the paper, for 180 µM monomer the% dimer is 36% and% hexamer is 15, 14 or 17%. There are also variations in% dimer and hexamer for 480 µM, etc. The authors should address the reasons for the variations.

2) The AUC seems to show that at low protein concentrations (where there presumably should be very few of these), there is equal amount of monomer and hexamer protein if not more hexamer. It's unclear how these qualify as rarely populated states, so further or a more quantitative explanation would help clarify this.

3) Subsection “Kinetic modelling of the rates of amyloid formation”, and correlation time calculations for Figure 7. Monomer will be present (particularly at the lower concentrations), why is this not included in the calculation? The omission of monomer may explain why the calculated values are too high at low protein concentration. Please also provide more specifics on how the correlation time was calculated for the dimer and hexamer from the models of the structures. It may be of interest to include tetramer in Figure 7—figure supplement 1.

4) Figure 2—figure supplement 1 – why are the signal intensities so different in panels A and B when nominally the same amount of protein was loaded onto the same column? It seems likely that the very small higher order oligomer signal that is seen in panel B is also present in panel A if the signal magnitudes were comparable. It seems important to address this since establishing the nature (i.e. dissociating upon dilution in the absence of cross-linking) and size of the excited species is important to the conclusions of the manuscript

5) It would be of interest to further define the hexamer and distinguish between the possible structural models. For Figure 6, please provide additional specifics for the analysis of protein concentration dependence of relaxation e.g. define the kex values in terms of constituent processes/terms and the observed protein concentration dependences. Can different protein concentration dependence for different residues be used to distinguish the dimer and hexamer interfaces? For the process involving strand G, is the chemical shift change towards random coil? Could strand G be involved in an interface? Also, for data in Figure 5 and its figure supplements, can more information be determined e.g. Figure 5—figure supplement 1 could include additional PRE data; Figure 5—figure supplement 2, consideration of residues 63, 64; Figure 5—figure supplement 4, why is the back calculation for hexamer model A done for PRE data at 120 µM where hexamer will be minimally formed? The agreement is stated to be good, but this is not clear.

6) Figure 5—figure supplement 5: Based on all 3 charge states, multiple hexamer models appear possible, however, the 15+ state is the main one considered. Please clarify the rationale for this analysis. The Materials and methods describe ANS measurements for dimer, but no dimer data are shown. While not essential, it would be of interest to include such data for the dimer, and this may provide support for the modelled hydrophobic surface in the dimer. Please clarify the extent of fluorescence change when ANS binds to the Im7. The presentation is a bit confusing because hexamers show a larger increase ANS than monomer, but hexamers are stated to show only a small increase.

Dynamics:

1) Please address when ΔN6 samples may or may not be at equilibrium in the various experiments, and how this may impact the data analysis.

2) Subsection “A unified model of ∆N6 polymerization”: how does Figure 2B shows disappearance of hexamers during the elongation phase?

3) It appears that relatively little monomer is consumed in the aggregation reactions (for example, consider the gel in Figure 2B). This is not addressed directly in the paper and it seems like having the majority of the material not aggregation prone must have a large effect on the interpretation of the data, i.e. the aggregation assays are reporting not on the major process that is occurring in the solution, but a relatively minor side reaction. There is not a gel for the pH 2 aggregation experiments, but it seems likely that must more of the monomer material is consumed?

4) Hexamer population is reported to correlate with amount of aggregated protein; please elaborate on the possible basis for this correlation.

Cell Toxicity:

The cell toxicity studies are weak and should be removed; a single concentration at a single timepoint in a neuroblastoma cell line is that is of questionable physiological relevance is considered.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Structural mapping of oligomeric intermediates in an amyloid assembly pathway" for further consideration at eLife. Your revised article has been favorably evaluated by John Kuriyan as the Senior Editor, and Tricia Serio as the Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below. Note that no further experimentation is required to address these issues.

1) Questions regarding quantitative analysis remain:

a) The proportions of monomer, dimer and hexamer for 120 μM total monomer are not consistent with those at higher concentrations (for 120 μM, the Kd dimer is 57 μM versus 50 μM, this should be addressed). The additional information on how the relative populations and Kds were calculated is pertinent, but not clear. Please define konover in the paper, give information on the origin of this equation (including reference), and include equations for how kon overall is related to τc and d.

b) Some additional support is needed for the authors' conclusions regarding larger species not really being quantitatively observable for AUC, but also SEC experiments, where the absorbance for larger species appears lower than expected. Can the authors use absorbance values more quantitatively?

c) Additional information is needed to show that independently prepared samples may indeed be compared i.e. they do not differ in uncontrolled ways due to unaccounted for variations in the timecourses or levels of assembly (oligomerization/aggregation), i.e. can the samples be considered to be consistently comparable (prior to and after completion of the time course of assembly/aggregation)?

d) Figure 2B: hexamer appears at 48 hours and disappears, while monomer appears unchanged. Presumably this is because at 80 μM total monomer, relatively little hexamer forms and gives rise to aggregates. So, one may reach an equilibrium at ~100 hours, where only monomer and dimer are significantly populated. It may be helpful to add some text or a figure to make this point clear, earlier in the paper. As for the lack of observation of dimer, could this be because the dimer is not effectively crosslinkable?

e) Given the experimental considerations the normalization of the chromatography signals may be reasonable; however, the relative signals within each panel are not internally consistent. Why is this?

2) Questions remain regarding the toxicity assay:

a) The toxicity assays are at best only a weakly supporting data: it is unclear why this is at all relevant for β2-microglobulin which does not specifically impact neuronal cells. It seems likely the cells simply don't take up β2-microglobulin, and good positive controls for these experiments are lacking.

eLife. 2019 Sep 25;8:e46574. doi: 10.7554/eLife.46574.040

Author response


Essential revisions:

The reviewers have identified a number of key questions, which should be addressed in a revised manuscript. The reviewers raised multiple questions regarding the conclusions and have requested additional information on the identity, quantitation, and thermodynamics to support them:

Hexamer abundance and structure:

1) The experimental determination of the hexamer Kd at ~10 ± 5 nM2 (ranging to 1.9 nM2 for kinetic modelling) is not consistent with the data. For example, using the values reported in the paper for 180 µM monomer of 48% dimer and 24% hexamer (subsection “Structural models of on-pathway hexamers”, first paragraph), assuming the remainder, 28%, is monomer, Kd for dimer to monomer is 59 µM and Kd for hexamer to dimer is 11x10-9 M2 = 11 x 109 nM2 = 11 x 103 µM2. This also will impact the derived ΔGº value. Use of the incorrect value for the hexamer Kd might have major consequences throughout the paper, though probably the conclusions will for the most part be unaltered. Elsewhere in the paper, for 180 µM monomer the% dimer is 36% and% hexamer is 15, 14 or 17%. There are also variations in% dimer and hexamer for 480 µM, etc. The authors should address the reasons for the variations.

We thank the reviewers for pointing out the discrepancies in the quoted populations in our manuscript. These have been corrected throughout. We apologise that we did not spot this in our submitted manuscript.

We also thank the reviewer for pointing out that we made an error in converting the hexamer Kd from M2 to nM2. We agree with the reviewer that the hexamer Kd is 11 x 10-9 M2 = 11 x 109 nM2 = 11 x 103 µM2, as can be calculated from the% monomer, dimer and hexamer at 180 µM of 26%, 48% and 26%, respectively. It is important that we point out here that all calculations in our submitted manuscript were performed using the correct populations and Kd. of 11 x 10-9 M2. Our error was in converting the units of M2 to nM2, which was only done editorially when writing the manuscript. Hence the error has no effect on any of the results presented. We apologise for making this error and the concerns it resulted in. The ΔGº value for hexamer formation accordingly is ~ 4 kJ/mol. We have corrected these values in the revised version of our manuscript.

In re-reading our manuscript, we also realised that we were not clear in describing how the relative populations and Kds were calculated. In our revision (revised Materials and methods subsection “Chemical shift perturbation and calculation of Kd values”), we now explain in detail how this was performed.

2) The AUC seems to show that at low protein concentrations (where there presumably should be very few of these), there is equal amount of monomer and hexamer protein if not more hexamer. It's unclear how these qualify as rarely populated states, so further or a more quantitative explanation would help clarify this.

In a system of interacting species (such as ΔΝ6) the measured S value obtained by sedimentation velocity AUC is a population-weight average. Thus peak areas cannot be interpreted directly as populations. We have clear evidence for hexamer formation from cross-linking experiments both during assembly (SDS PAGE) (Figure 2B) and at equilibrium (via SEC) (Figure 2—figure supplement 1). While the sedimentation velocity data are consistent with hexamers we cannot rule out the possibility that the peak observed corresponds to a weight average of species, some of which may be larger than hexamer. We have amended the subsection “Native-like dimers and hexamers form during ΔN6 assembly” to make this point clear.

3) Subsection “Kinetic modelling of the rates of amyloid formation”, and correlation time calculations for Figure 7. Monomer will be present (particularly at the lower concentrations), why is this not included in the calculation? The omission of monomer may explain why the calculated values are too high at low protein concentration.

We thank the reviewers for pointing this out. Monomers were indeed taken into account in the calculation, and we erroneously omitted to state this in the Materials and methods. We apologise. We have now corrected this in the revised version (subsection “Kinetic modelling of the rates of amyloid formation”).

Please also provide more specifics on how the correlation time was calculated for the dimer and hexamer from the models of the structures.

The correlation times were calculated using the structural models of the monomers, dimers and hexamers using HYDROPRO. We have added these details into the Materials and methods subsection “Kinetic modelling of the rates of amyloid formation”.

It may be of interest to include tetramer in Figure 7—figure supplement 1.

We have now included this figure in our revised manuscript. The addition of two additional rate constants (for dimer-tetramer formation) does not produce a significant improvement in the quality of the fits (new Figure 7—figure supplement 1D) compared with the monomer-dimer-hexamer model shown in Figure 7A. We therefore, chose to focus on the simpler model of monomer-dimer-hexamer.

4) Figure 2—figure supplement 1 – why are the signal intensities so different in panels A and B when nominally the same amount of protein was loaded onto the same column? It seems likely that the very small higher order oligomer signal that is seen in panel B is also present in panel A if the signal magnitudes were comparable. It seems important to address this since establishing the nature (i.e. dissociating upon dilution in the absence of cross-linking) and size of the excited species is important to the conclusions of the manuscript

The SEC experiments on crosslinked and uncrosslinked material was performed on different days and on different AKTA instruments using different sample loops and volumes loaded. Therefore, the absolute signal intensity between panels a and b cannot be compared directly. However, the curves within each panel can be compared, and show that without crosslinking (panel a) only monomer and dimers are present in contrast to after cross-linking (panel b). To avoid this confusion, we have now normalised the signal in each plot to the maximum intensity in each panel.

5) It would be of interest to further define the hexamer and distinguish between the possible structural models.

In the course of our work we also evaluated hexamer models assembled from dimer B in Figure 5—figure supplement 3 (see Author response images 1-3). Importantly, hexamer models assembled from dimer B predict large PREs in regions of the protein other than the BC, DE and FG loops which is inconsistent with the observed data. For example, hexamer i, (Author response image 1) shows large PREs in the F strand an succeeding FG loop, hexamer ii (Author response image 2) shows very large PREs from residue 33 to the N-terminus and the FG, while hexamer iii (Author response image 3) shows PREs from S61C to the C strand. We chose not to show all of these structure calculations in the supplement for the sake of brevity and clarity. We have made these data available on the University of Leeds publicly available library (https://doi.org/10.5518/329) and noted this in our revised manuscript (subsection “Structural models of hexamers”).

Author response image 1. Evaluation of hexamer i, assembled for dimer B.

Author response image 1.

Back-calculated PRE rates (black lines) are shown in panels A-D. Top (E) and side (F) view of the structural model.

Author response image 2. Evaluation of hexamer ii, assembled for dimer B.

Author response image 2.

Back-calculated PRE rates (black lines) are shown in panels A-D. Top (E) and side (F) view of the structural model.

Author response image 3. Evaluation of hexamer iii, assembled for dimer B.

Author response image 3.

Back-calculated PRE rates (black lines) are shown in panels A-D. Top (E) and side (F) view of the structural model.

For Figure 6, please provide additional specifics for the analysis of protein concentration dependence of relaxation e.g. define the kex values in terms of constituent processes/terms and the observed protein concentration dependences. Can different protein concentration dependence for different residues be used to distinguish the dimer and hexamer interfaces?

Unfortunately, the complexity of the monomer-dimer-hexamer model, combined with the limited range of protein concentrations that could be used to perform the experiments (as the protein aggregates non-specifically at higher concentrations than those described in the manuscript), mean that we do not have sufficient data to enable us to interpret the CPMG experiments in the detail suggested. To do so would be a substantial undertaking, beyond the scope of the current paper.

For the process involving strand G, is the chemical shift change towards random coil? Could strand G be involved in an interface?

CPMG experiments do not provide the sign of the Δδ (but just its absolute per-residue value), and so we cannot say whether the peaks move towards their random coil values. We think it is unlikely that the G strand is in an interface, since just four residues, all localised in the same part of the protein, show this behaviour. This point is now made in the subsection “Hexamer dynamics may prime further assembly into amyloid”. Most importantly, no significant concentration-dependent chemical shifts are observed for these four residues, nor are any PREs are observed for these residues at low or high protein concentration. A kex that is close to an order of magnitude slower for an interface involving the G strand compared with the interface at the apical part of the protein is also unlikely. Hence, we conclude that it is unlikely that the G strand is involved in an interface (see the aforementioned subsection).

Also, for data in Figure 5 and its supplements, can more information be determined e.g. Figure 5—figure supplement 1 could include additional PRE data; Figure 5—figure supplement 2, consideration of residues 63, 64; Figure 5—figure supplement 4, why is the back calculation for hexamer model A done for PRE data at 120 µM where hexamer will be minimally formed? The agreement is stated to be good, but this is not clear.

As requested, we have added additional raw PRE data in Figure 5—figure supplement 1 (new panel D).

Regarding residues 63 and 64 – we could determine their Rex value at 180 μΜ, but not at 480 μΜ, as these resonances are too broad to measure at the higher concentration (hence the absence of a black bar). We have now amended Figure 5—figure supplement 2 (black crosses) to highlight residues that behave in this manner.

Figure 5—figure supplement 4 – the PREs measured at high protein concentrations suffer from line broadening due to the higher oligomer population and therefore several resonances disappear from the spectrum (red crosses in Figure 5—figure supplement 1). Thus, we chose to compare the models generated to the PRE data measured at the lower concentration where all three species contribute to the PRE data and the maximum number of resonances is observed.

6) Figure 5—figure supplement 5: Based on all 3 charge states, multiple hexamer models appear possible, however, the 15+ state is the main one considered. Please clarify the rationale for this analysis.

We discuss data only for the 15+ charge state, as this is the lowest charge state observed and hence the most compact and native-like. For the higher charge states (16+ and 17+) coulombic repulsion can result in partial or more complete protein unfolding. The consensus in the field (Vahidi et al., 2013) is that it is best to focus analysis on the lowest charge state observed, as this likely better reflects the solution phase structure of globular proteins using ESI-IMS-MS. We have edited the subsections “Structural models of on-pathway hexamers” and “Native ESI-IMS-MS”.

The Materials and methods describe ANS measurements for dimer, but no dimer data are shown. While not essential, it would be of interest to include such data for the dimer, and this may provide support for the modelled hydrophobic surface in the dimer. Please clarify the extent of fluorescence change when ANS binds to the Im7. The presentation is a bit confusing because hexamers show a larger increase ANS than monomer, but hexamers are stated to show only a small increase.

We have now added the ANS data for the dimer to Figure 5—figure supplement 5D. The dimers and hexamers each show a small increase in ANS fluorescence compared with monomers. However, given that the ANS intensity depends both on the amount of dye bound and the quantum yield of the dye in the bound state, we cannot quantitatively compare the differences in intensity with the structural models.

We chose to compare the ANS fluorescence in the presence of the ΔN6 oligomers and the folding intermediate of Im7, since the on-pathway Im7 folding intermediate is highly structured, reminiscent perhaps of the dimers and hexamers of ΔN6. Notably, the ANS fluorescence of the Im7 folding intermediate is much smaller than that observed when ANS binds to a typical ‘molten globule’ state, such the classical molten globule of apo α-lactalbumin (for which an ~100-fold increase in ANS fluorescence is observed (Semisotnov et al., 1991)). We have clarified the text in the subsection “Structural models of on-pathway hexamers” to hopefully make this point clear.

Dynamics:

1) Please address when ΔN6 samples may or may not be at equilibrium in the various experiments, and how this may impact the data analysis.

Based on the CPMG data and the calculated exchange rates of 1790 s-1 and 205 s-1, equilibrium should be reached within 5 ms. Hence, all NMR experiments were performed under equilibrium conditions.

2) Subsection “A unified model of ∆N6 polymerization”: how does Figure 2B shows disappearance of hexamers during the elongation phase?

Figure 2B shows the appearance of a hexamer band at early stages of fibril formation that decreases in intensity as fibril formation proceeds. Since we did not carry out a simultaneous measurement of ThT fluorescence in this assay we cannot ascribe the% hexamer to a precise stage in the elongation kinetics. We have changed the text in the subsection “Native-like dimers and hexamers form during ΔN6 assembly” to reflect this fact more clearly, and we have added an electron micrograph to show that fibrils indeed form at the end of the reaction under the conditions employed (see revised Figure 2B).

3) It appears that relatively little monomer is consumed in the aggregation reactions (for example, consider the gel in Figure 2B). This is not addressed directly in the paper and it seems like having the majority of the material not aggregation prone must have a large effect on the interpretation of the data, i.e. the aggregation assays are reporting not on the major process that is occurring in the solution, but a relatively minor side reaction. There is not a gel for the pH 2 aggregation experiments, but it seems likely that must more of the monomer material is consumed?

The reviewers are correct in their observation that in seeded reactions at pH 6.2 not all of the initial monomer is consumed into fibrils, as shown in Figure 7C. Indeed, this is unusual behaviour compared with ‘typical’ seeding reactions where the elongating unit is a monomeric unfolded protein and the fibril yield is typically >90%. This is indeed observed for unfolded β2m at pH 2.0, where the critical concentration is <8 μΜ and reactions measured at concentrations of e.g. 100 μΜ have fibril yields of > 92%. However, ΔΝ6 at pH 6.2 elongates through a very different mechanism involving structured hexamers, whose population is highly dependent on the monomer concentration. After the hexamers have added onto the fibril ends and, due to the relatively high Kd for dimerization (~50 μΜ), the remaining monomer concentration rapidly becomes too low to enable dimer formation (and therefore hexamer formation), stalling the seeding reaction. As an illustrative example, at 500 μΜ ΔΝ6 where the population of ΔN6 molecules in hexamers is 56%, ~85 μΜ monomers (plus some dimers) remain in solution after fibril elongation – a concentration similar to the dimer Kd. As seen in Figure 7C when the initial monomer concentration approaches the dimer Kd we effectively see no elongation. Therefore, the low fibril yield in the seeding reactions is a predictable outcome of the monomer-dimer-hexamer model and provides an independent validation of the notion that hexamer formation is the rate limiting step of fibril elongation.

4) Hexamer population is reported to correlate with amount of aggregated protein; please elaborate on the possible basis for this correlation.

This question has been answered in point 3 above.

Cell Toxicity:

The cell toxicity studies are weak and should be removed; a single concentration at a single timepoint in a neuroblastoma cell line is that is of questionable physiological relevance is considered

One of the interesting findings of our manuscript is that the on pathway β2m dimers and hexamers do not exhibit cytotoxicity, at least under the conditions explored, and this was highlighted as a ‘noteworthy’ discovery in the editor’s summary of our paper. We prefer, therefore, to retain this information in our manuscript, as any reader would be bound to wonder why cytotoxicity was not assayed when amyloid oligomers are being discussed. However, we agree that our experimental observations are not exhaustive, and that we focus on a single cell type at a single protein concentration and incubation time, which we now discuss specifically in our revised manuscript (subsection “Structural models of on-pathway hexamers”).

The neuroblastoma cell line SH-SY5Y was chosen for our assays, as this cell line is a widely accepted model for the study of amyloid toxicity and has been used by other labs for β2m and other amyloid forming sequences. The choice of SH-SY5Y cells was also made since it enables a direct comparison with other published studies investigating oligomer-associated toxicity. A statement to this affect has been added to the manuscript and references are cited to justify this statement (see the aforementioned subsection).

Whilst only a single protein concentration was used for our assays, no effect on any of the assays for cell viability was observed, despite incubating the cells with a monomer equivalent concentration of ~3 μM cross-linked oligomers (calculated from the A280 of the hexamer fraction from the SEC column and the extinction co-efficient of monomeric β2m). Notably, the monomer equivalent concentration of the β2m hexamers is ~10 fold higher than that used by Fusco et al. (2017) to demonstrate that α-synuclein B* oligomers are cytotoxic.

A single time point was used, but again this corresponds to the timescales typically used in the references cited to determine whether oligomers are toxic. Whilst this allows comparison with other published studies, we accept that we cannot rule out toxicity taking longer to manifest. This caveat is now clearly stated in the manuscript.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below. Note that no further experimentation is required to address these issues.

1) Questions regarding quantitative analysis remain:

a) The proportions of monomer, dimer and hexamer for 120 μM total monomer are not consistent with those at higher concentrations (for 120 μM, the Kd dimer is 57 μM versus 50 μM, this should be addressed). The additional information on how the relative populations and Kds were calculated is pertinent, but not clear. Please define konover in the paper, give information on the origin of this equation (including reference), and include equations for how kon overall is related to τc and d.

We are not sure what the reviewer is referring to regarding the dimer Kd at 120 μΜ. The values used throughout the paper are Kd, dim = 50 μΜ and Kd,hex = 10 x 10-9 M2. We checked and could find no mention of 57 µM in our manuscript.

About konover (the overall rate of protein assembly, now defined in the subsection “Chemical shift perturbation and calculation of Kd values”) let’s consider a simple 3-state monomer (M), dimer (D), hexamer (H) model:

M→k1app←k-1D→k2app←k-2H

where k1app, k2app are the apparent forward rate constants for dimer and hexamer formation respectively, k-1, k-2 are the reverse rate constants for dimerization and hexamerization, respectively. Based on elementary chemical kinetics the time evolution of the concentration of H and D is given by:

dH/dt=k2appD-k-2H (Equation 1)

dD/dt=k1appM-k1D-k2appD+k-2[H] (Equation 2)

In order to derive an expression for (konover) we simply need to eliminate [D] from Equation 1 and replace it with [M]. If we apply the equilibrium condition for Equation 2 (d[D]/dt = 0) we obtain:

D=k1appM+k-2[H]k-1+k-2app (Equation 3)

By substituting Equation 3 into Equation 1 and rearranging:

dH/dt=k1appk2appk-1+k2appM-k-1k-2k-1+k2app[H] (Equation 4)

with the term in front of [M] representing koveron. However, the reaction considered is not first order and we need to convert k1app, k2app to rate constants using the relationships:

k1app=2k1Meq

k2app=3k2[Deq]2

Where, [Meq], [Deq] are the equilibrium concentrations of monomers and dimers. Therefore,

konover=k1appk2appk-1+k2app=6k1k2Meq[Deq]2k-1+3k2[Deq]2 (Equation 5)

The overall observed τc and d depend on the shape of monomers, dimers and hexamers but most importantly on their populations. In turn, the oligomer populations depend on the ratio of k1, k2 with k-1, k-2 and the concentrations of monomers and dimers and hexamers at equilibrium. Since the off rates are not concentration-dependent, for simulation purposes we can fix their values, while restraining k1, k2 accordingly to satisfy the measured monomer and hexamer Kds. By doing so, the konovergiven by Equation 5 scales in the same way as the overall population of dimers and hexamers at any given monomer concentration and therefore τc and d are expected to scale in the same way. [Meq], [Deq] were calculated by numerical integration of the monomer-dimer-hexamer model at τ=∞ using various initial monomer concentrations (Mtot) and Kd, dim = 50 μΜ and Kd,hex = 10 x 10-9 M2. This discussion has now been added into the Materials and methods subsection “Chemical shift perturbation and calculation of Kd values”.

b) Some additional support is needed for the authors' conclusions regarding larger species not really being quantitatively observable for AUC, but also SEC experiments, where the absorbance for larger species appears lower than expected. Can the authors use absorbance values more quantitatively?

The fact that AUC at concentrations over 200 μΜ does not show hexamers is presumably because at these concentrations 20% of the material has already sedimented during the initial 3000 rpm run, consistent with the hexamers rapidly forming high molecular weight species that sediment before the c(S) data are acquired (Author response image 4).

Author response image 4. Raw AUC data.

Author response image 4.

Plot of absorbance at 280 nm across the AUC cell radius for the initial run (3000 rpm, blue) versus the first scan of the actual run at 48000 rpm (red).

One therefore would expect to be able to see high order species in the SEC of uncross-linked material. However, with the columns and sample loops used in the SEC experiment and using our estimated Kd values, a sample of 400 μΜ (hexamer population 48%) would be diluted roughly 5-times, resulting in a maximum hexamer population of only 6%. Thus, it is not surprising that we only observe monomers and dimers in the SEC of uncross-linked ΔΝ6. On the other hand, when crosslinked material is added on the column at high concentrations (400-500 μΜ) a peak is observed at the void volume even at 0h (black curve Figure 2—figure supplement 1D) consistent with rapid formation of larger species, consistent with the AUC data. These complications do not allow detailed quantification of AUC/SEC absorbance values, other than identification of the MW of the species present. This discussion has now been added to the subsection “Chemical cross-linking and analytical SEC”.

c) Additional information is needed to show that independently prepared samples may indeed be compared i.e. they do not differ in uncontrolled ways due to unaccounted for variations in the timecourses or levels of assembly (oligomerization/aggregation), i.e. can the samples be considered to be consistently comparable (prior to and after completion of the time course of assembly/aggregation)?

More than 20 NMR samples from at least 10 different protein preparations over 4 years were used in the present study. All of those gave consistent results, while controls (i.e. HSQCs) were routinely carried out in order to check sample quality before and after each NMR experiment.

d) Figure 2B: hexamer appears at 48 hours and disappears, while monomer appears unchanged. Presumably this is because at 80 μM total monomer, relatively little hexamer forms and gives rise to aggregates. So, one may reach an equilibrium at ~100 hours, where only monomer and dimer are significantly populated. It may be helpful to add some text (similar to response to Dynamics point 3) or a figure to make this point clear, earlier in the paper. As for the lack of observation of dimer, could this be because the dimer is not effectively crosslinkable?

The reviewer is right in pointing out that at 80 μΜ the population of hexamers is only 6%. The apparent appearance and disappearance of the hexamer band over time in Figure 2B shows that the hexamers go on to form aggregates that do not enter the gel. As for the absence of a dimer band, based on our dimer model only one lysine pair between residues 58 in monomer 1 and 91 in monomer 2 can potentially be crosslinked, making the dimer less amenable to crosslinking in comparison to the hexamer where many different crosslinks could be formed. We now note this in the figure legend.

e) Given the experimental considerations the normalization of the chromatography signals may be reasonable; however, the relative signals within each panel are not internally consistent. Why is this?

The main point of the SEC experiments is to show that in the absence of cross-linking we only observe monomers and dimers while after cross-linking an array of oligomers are observed, with dimers and hexamers being the most prevalent ones. Data in Figure 2—figure supplement 1B show a decreased monomer population at higher protein concentrations fully consistent with hexamer species becoming more prevalent. For the uncrosslinked samples in Figure 2—figure supplement 1A, the dimer peak increases as the protein concentration is increased as expected. The unavoidable dilution of the samples and their re-equilibration during the SEC run, and also the complications discussed in point b above, make it impossible to quantitate the relative peak intensity in SEC traces of the uncrosslinked samples. However, this does not affect our conclusion from these data, which was simply to show that crosslinking increases the number of oligomers observed. This discussion has now been added to the subsection “Chemical cross-linking and analytical SEC”.

2) Questions remain regarding the toxicity assay:

a) The toxicity assays are at best only a weakly supporting data: it is unclear why this is at all relevant for β2-microglobulin which does not specifically impact neuronal cells. It seems likely the cells simply don't take up β2-microglobulin, and good positive controls for these experiments are lacking.

SH-SY5Y cells are a commonly used model for amyloid toxicity and were chosen since they allow comparison of the toxicity of the β2m hexamers with studies of the toxicity of other oligomers in amyloid formation. The observed lack of toxicity of the dimers and hexamers is unlikely to be due to a failure of these species to be taken up by the cells. Indeed, we have shown previously that SH-SY5Y cells not only take up the β2m monomers, but also β2m amyloid fibrils (Jakhria et al., 2014). This is now stated in the text (subsection “Structural models of on-pathway hexamers”).

The legend for Figure 5—figure supplement 6 has been amended to include additional information on the assay controls and how the data are normalised. For MTT reduction, ATP levels and ROS production, the data are normalized to the PBS buffer control (100%) and NaN3 treated controls (a positive control for cell death (0% )). LDH release is normalized to detergent lysed cells (a positive control for cell lysis, 100% ) and PBS buffer treated controls (0% ). Additional text has also been added to the Materials and methods (subsection “Cytotoxicity assays) describing H2O2 as a positive control for the induction of ROS. We have placed the raw data for these in the University of Leeds data repository so they are accessible to interested readers.

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Data Citations

    1. Karamanos TK, Jackson MP, Calabrese AN, Goodchild SC, Cawood EE, Thompson GS, Kalverda AP, Hewitt EW, Radford SE. 2019. Data from: Structural mapping of oligomeric intermediates in an amyloid assembly pathway. University of Leeds Data Depository. [DOI] [PMC free article] [PubMed]

    Supplementary Materials

    Transparent reporting form
    DOI: 10.7554/eLife.46574.031

    Data Availability Statement

    Data are freely available at the University of Leeds Data Depository: https://doi.org/10.5518/329.

    The following dataset was generated:

    Karamanos TK, Jackson MP, Calabrese AN, Goodchild SC, Cawood EE, Thompson GS, Kalverda AP, Hewitt EW, Radford SE. 2019. Data from: Structural mapping of oligomeric intermediates in an amyloid assembly pathway. University of Leeds Data Depository.


    Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd

    RESOURCES