Methods

Amyloid fibril preparation
Recombinant wild-type β2m was expressed in Escherichia coli and purified to homogeneity.26 The recombinant wild-type protein contained all 99 residues plus the N-terminal methionine, and the single disulfide bond was oxidised. Fibrils were formed by incubation of the protein (0.3–0.5 mg/ml) for 2–8 weeks at pH 2.5 in 25 mM sodium phosphate and 25 mM sodium acetate buffer containing 0.03% (wt/vol) sodium azide at 37 °C. At pH 7.0, fibrils were grown by elongation of heparin-stabilised seeds from fibrils formed at pH 2.5.24

Transmission EM
β2m fibrils were used without dilution and negatively stained with 2% (wt/vol) uranyl acetate on thin carbon films supported on holey carbon coated EM grids. Microscopy was carried out on a Tecnai 10 (FEI, Eindhoven, NL) operating at 100 kV with a low electron dose and defocus values of 700–900 nm. For cryo-EM, the fibrils were vitrified on holey carbon support films. Cryo-EM was performed using a Tecnai F20 with a field emission gun operating at 200 kV. Low electron dose images were recorded on Kodak SO163 film at 29,000×, with defocus in the range 1–2.6 μm.

Single-particle image processing
Cryo-EM images were digitised on a Zeiss SCAI photoscanner (ZI Imaging, Swindon, UK) with a pixel size of 7 μm. Cross-over repeats were selected by manually marking the cross-over positions using Ximdisp54,55 and were boxed out in SPIDER.56 Images were averaged to 0.35 nm/pixel at the specimen level and extracted into 1080 × 1080-pixel boxes for data processing. Defocus was determined from carbon adjacent to holes using Ctffind2,54,57 and image phases were corrected for the effects of the contrast transfer function using IMAGIC.58 Images were bandpass-filtered with a low-frequency cutoff between 70 and 20 nm (adjusted according to defocus) and a high-frequency cutoff of 1.16 nm and normalised. Images were aligned in SPIDER,56 with subsequent multivariate statistical analysis and classification of images in IMAGIC. An initial separation into the three structural types prior to classification was based on an analysis of the eigen images. Classes were all set to the same polarity. An iterative procedure of alignment, classification, and polarity reversal was performed until no classes of opposite polarity were obtained. Averaged power spectra of negative stain segments were obtained by multivariate statistical analysis and classification. The subunit repeat is easier to recognise in stained fibrils because of the higher contrast and as a consequence of single sided staining which avoids the overlap of upper and lower regions of the fibrils. This overlap obscures the subunit repeat pattern when the whole structure is seen in projection by cryo-EM because the repeats are out of register in the two halves of the fibrils.

Three-dimensional reconstruction
Three-dimensional reconstructions using SPIDER were carried out on the best classes, those showing the clearest protofilament substructure, using helical symmetry.56 Two parameters had to be determined for the reconstruction: the cross-over repeat (Fig. 1a) and the subunit repeat (Fig. 1d). The cross-over repeat was determined by selecting a small region around one cross-over from the class average and by using cross correlation to determine the position of the other cross-over. The cross-over length was checked visually on the class average. Determination of the subunit repeat was more difficult, and several methods were used to verify the results. The starting point was the 5.88-nm repeat observed in diffraction patterns of negative stain images (Fig. 1e). Then, each map was reconstructed with a range of subunit repeats (4.5–7 nm), and the maps were examined to find the ones with the most distinct features, the highest contrast, and the best-matching reprojections (Supplementary Fig. 4). Based on these comparisons, a subunit repeat of 5.78 nm was used for the A-type maps, and a subunit repeat of 5.25 nm was used for the B-type maps. Short segments were reconstructed, and these were helically averaged to generate a full repeat so that reprojections could be compared to the class averages.

Scanning transmission electron microscopy
Stock β2m fibril solutions were either diluted 20× to 100× in buffer or water and used immediately, or diluted 80× in an aqueous glutaraldehyde solution (final aldehyde concentration, 0.1%) and used after 5 min of incubation at room temperature. For mass measurement, 7 μl aliquots were adsorbed for 1 min to glow-discharged thin carbon films supported on holey carbon films on 200-mesh gold-plated copper grids. The grids were blotted, washed in 7–12 drops of quartz double-distilled water, plunge-frozen in liquid nitrogen, and freeze-dried at − 80 °C and 5 × 10− 8 Torr overnight in the microscope. Tobacco mosaic virus (kindly supplied by Dr. R. Diaz-Avalos, Institute of Molecular Biophysics, Florida State University) adsorbed to a separate grid, was air-dried, and served as mass standard. A Vacuum Generators STEM HB-5 interfaced to a modular computer system (Tietz Video and Image Processing Systems GmbH, D-8035 Gauting, Germany) was used to record 512 × 512-pixel dark-field images. An accelerating voltage of 80 kV, a nominal magnification of 200,000×, and doses of approximately 400 e/nm2 were used for the mass measurements. Repeated low-dose scans were also recorded from some grid regions to assess beam-induced mass loss. The IMPSYS59 and MASDET60 program packages were used for evaluation. Fibril segments were selected in square boxes and tracked. The total scattering within an integration box following their length was then calculated, and the scattering contribution of the supporting carbon film was subtracted. Division by the segment length gave the MPL. MPL values were corrected for beam-induced mass loss, scaled to the MPL of tobacco mosaic virus, binned, displayed in histograms, and fitted by Gaussian curves. The measured MPL was not dependent on either the dilution method or the glutaraldehyde treatment, allowing all data sets to be merged and displayed in a single histogram for the final analysis.