Clinical history and neuropathology

Deidentified human biospecimens from deceased individuals were obtained from academic biorepositories. All tissue donors provided written or verbal consent to donate autopsied brains for use in biomedical research in accordance with the standards of each institution. This study was exempt from institutional review board approval (i.e., this study is not considered human subject research) in accordance with the University of California San Francisco institutional review board policy. We determined the cryo-EM structure of tau filaments from 0.5 g fresh-frozen frontal cortex samples from four individuals with DS. Case 1 was a 63-year-old man with documented dementia. Case 2 was a 51-year-old woman with documented dementia. Case 3 was a 46-year-old woman with no documented dementia. Case 4 was a 36-year-old woman with no documented dementia. The demographic and cliniconeuropathological details of these four cases are in Supplementary Table S1. Formalin-fixed samples from additional cases of DS and sAD were obtained for histological studies. Fresh-frozen samples from additional cases of DS and sAD were obtained for biochemical studies. The demographic and cliniconeuropathological details of the additional cases are in Supplementary Tables S2 and S3.

Immunohistochemistry

Deparaffinized fixed sections were pretreated in 98% formic acid for 6 min to enhance immunoreactivity. After blocking with 10% normal goat serum in phosphate-buffered saline (PBS) with 0.2% Tween 20 (PBST), sections were incubated at room temperature in primary antibodies overnight followed by secondary antibodies for 2 h. Primary antibodies were prepared in 10% normal goat serum and applied as combinations of either (a) anti-Aβ1-40 rabbit polyclonal (1:200; AB5074P, MilliporeSigma) and anti-Aβ1-42 12F4 mouse monoclonal (1:200; 805502, BioLegend); (b) anti-Aβ17-24 4G8 mouse monoclonal (1:1000; 800709, BioLegend) and anti-tau (phospho-S262) rabbit polyclonal (1:200; ab131354, Abcam); or (c) anti-tau AD conformation-specific GT-38 mouse monoclonal (1:150; ab246808, Abcam) and anti-tau (phospho-S396) rabbit monoclonal (1:200; ab109390, Abcam) antibodies. The polyclonal IgG H + L secondary antibodies were conjugated with Alexa Fluor 488 or 647 (A11029, A21235, A11008, and A21244, Thermo Fisher Scientific) applied 1:500 in 10% normal goat serum in PBST. Stained slides were scanned on a ZEISS Axio Scan Z1 digital slide scanner at 20× magnification. Excitation at 493, 553, or 653 nm was followed by detection at 517 nm (Aβ40 or phospho-tau), 568 nm (autofluorescence), or 668 nm (Aβ42 or tau), respectively.

Dye staining of brain sections for EMBER (excitation multiplexed bright emission recordings) microscopy

Formalin-fixed, paraffin-embedded brains were sectioned (8-μm thickness) and glass mounted. To reduce the autofluorescence in the brain tissue by greater than 90% intensity (e.g., from lipofuscin or hemosiderin), the sections were photobleached in a cold room for up to 48 h using a multispectral LED array [19]. The sections were deparaffinized, washed in PBS, and stained with 5 µM dye 60 for 30 min. The sections were washed with PBS and coverslipped with PermaFluor (Thermo Scientific) as the self-curing mounting medium. See Yang et al. for additional details [20].

EMBER microscopy data collection

Dye 60-stained fixed brain slices were imaged with a Leica Microsystems SP8 confocal microscope using a 40× water-immersion lens (1.1 NA), a white light and 405 nm lasers, and a HyD detector at 512 × 512-pixel resolution at 2× zoom. The optical plane was manually focused for each field of view. To reduce the background noise from the bottom of the slide, the LightGate was set to 0.5–18 ns. First, 110 images were acquired using the Λ/λ-scan mode with wavelength excitations of 470, 490, 510, 530, 550, 570, 590, 610, 630, 650, and 670 nm. The emission detection range started at 10 nm greater than the given excitation wavelength and ended at 780 nm, with a 20-nm window. For example, for the 470-nm excitation, the images were collected at 480–500, 500–520, 520–540, 540–560, 560–580, 580–600, 600–620, 620–640, 640–660, 660–680, 680–700, 700–720, 720–740, 740–760, and 760–780 nm. Then, in the λ-scan mode, 18 additional images were collected at 405-nm excitation with emission detection intervals of 20 nm from 420 to 780 nm. See Yang et al. for additional details [20].

EMBER data postprocessing, particle segmentation, and dimension reduction

The EMBER analytical pipeline is described in detail in Yang et al. [20]. In brief, we developed a set of custom scripts in MATLAB to process the raw fluorescent images and segment the aggregated protein deposits. The signal-processing algorithm for the analysis of particle-resolution EMBER spectra was executed in MATLAB with pca. Each identified EMBER particle from the particle segmentation was normalized to [0, 1] and then concatenated in an array for principal component analysis. The principal component scores PC1 and PC2 were plotted. Uniform manifold approximation and projection (UMAP) was performed in MATLAB with run_umap with default settings.

Preparation of sarkosyl-insoluble tau for biochemical characterization

Sarkosyl-insoluble Tau was prepared by diluting 250 µL 10% brain homogenate in equal volume of A68 buffer (10 mM Tris–HCl, pH 7.4, 0.8 M NaCl, 1 mM EGTA, 5 mM EDTA, 10% sucrose, protease and phosphatase inhibitors), followed by centrifugation at 13,000 × g at 4 °C for 20 min1. The supernatants were kept on ice, and the pellets were resuspended in 250 μl of A68 buffer and centrifuged for another 20 min. Both supernatants were pooled. Samples were then incubated with 1% sodium lauroyl sarcosinate (1614374, Sigma Aldrich) at room temperature for 1 h at 700 rpm followed by centrifugation at 100,000 × g at 4 °C for 1 h. Pellets were then washed and resuspended in 50 mM Tris–HCl, pH 7.4 at half the original volume of brain homogenate. Samples were used immediately or snap-frozen in liquid nitrogen and stored at − 80 °C.

Homogenous time-resolved fluorescence assay (HTRF) for tau aggregation

HTRF was performed on sarkosyl-insoluble extracts using Tau Aggregation kits (6FTAUPEG, Revvity) with 384-well microplates (Perkin Elmer) per manufacturer instructions. Briefly, anti-human TAU-d2 conjugate and anti-human TAU-Tb3+-cryptate conjugate antibody stocks were diluted at a concentration of 1:19 with detection buffer and premixed 1:1 immediately before plating. Samples were prepared by diluting sarkosyl-insoluble extracts at a concentration of 1:4 with 50 mM tris buffer. Samples (10 µL) were then added to each well followed by 10uL of premixed antibodies. Plates were sealed and incubated for 2 h at room temperature. Plates were read at 665 and 620 nm using the PHERAstar FSX Microplate Reader followed by analysis using MARS Data Analysis Software.

Western blot analysis

Samples were selected based upon normalized insoluble tau protein content as determined by the HTRF Tau Aggregation assay. Sarkosyl-insoluble tau samples were treated with or without trypsin (final trypsin concentration 2 mg/mL) at room temperature for 1 h at 1400 rpm. The samples were then resolved using 4–15% precast polyacrylamide gels and then transferred to a nitrocellulose membrane. Briefly, samples were mixed with 4X Laemlli sample buffer (1610747, Bio-Rad) and 2-mercaptoethanol followed by boiling for 5 min for protein denaturation. Following membrane transfer and blocking, tau protein was probed using the following primary antibody pairs: (1) T13 tau (1:1000, sc-21796, Santa Cruz) and 4R tau (1:1000, ab218314, Abcam); (2) GT-38 (1:1000, ab246808, Abcam) and TauC4 (1:1000, ABN2178, Sigma-Aldrich). Membranes were incubated with primary antibody pairs overnight at 4 °C with rocking; all primary antibody cocktails were prepared in Intercept blocking buffer (LI-COR). Following a TBST wash, membranes were incubated with goat anti-rabbit secondary antibody conjugated with IRDye 680RD (1:10,000, #926-68071, LI-COR) and goat anti-mouse secondary antibody conjugated with IRDye 800CW (1:10,000, #926-32210, LI-COR) in Intercept blocking buffer for 1 h at room temperature. Fluorescence was detected with the Odyssey Fc Imager (LI-COR) using channels 700 and 800. The PageRuler Plus pre-stained protein ladder (PI26619, Thermo Scientific) was used to provide a molecular weight marker. Images were analyzed with the LI-COR Acquisition Software.

Filament purification

For case 1, tau filaments were purified from the frontal cortex as previously described in Fitzpatrick et al. [13]. For cases 2–4, we achieved improvements in the tau filament purification, including reduced background ferritin contamination, by following methods previously described for purification of Aβ filaments [21]. Here, 0.5 g brain tissue was homogenized in 20 volumes (w/v) of extraction buffer and brought to 2% sarkosyl for incubation at room temperature for 1 h. The homogenates were then centrifuged at 10,000 × g for 10 min, and the resulting supernatants were additionally spun at 100,000 × g for 60 min at 4 °C. The pellet from the second spin was resuspended in 1 mL/g extraction buffer and centrifuged at 3000 × g for 5 min. The supernatants were diluted three-fold in buffer containing 50 mM Tris–HCl, pH 7.4, 150 mM NaCl, 10% sucrose, and 0.2% sarkosyl and centrifuged at 100,000 × g for 30 min. The resulting sarkosyl-insoluble pellets were resuspended in 100 µL/g 20 mM Tris–HCl, pH 7.4, containing 50 mM NaCl, prior to vitrification for cryo-EM.

Affinity-grid preparation, vitrification, and data collection

GO deposition and assembly of the affinity grids was performed essentially as described [22, 23]. Briefly, to functionalize the GO surface, grids (Au QUANTIFOIL, R 1.2/1.3, 300 mesh, Quantifoil Micro Tools GmbH) were submerged in dimethyl sulfoxide with 10 mM dibenzocyclooctyne-PEG-amine (DBCO-PEG4-amine; CCT-A103P, Vector Laboratories) overnight at room temperature, washed, and then incubated with 20 µL of a 1 mM solution of azide polyethylene glycol (PEG) maleimide and methoxyl PEG azide (5 and 2 kDa molecular weight, respectively; Nanocs) at a 1:9 ratio for 6 h. Maleimide-functionalized grids were then washed with water and then ethanol, dried, and stored at − 20 °C. Prior to vitrification, GO grids were first incubated with 3 µL 250 nM recombinant protein G (Z02007, GenScript) in resuspension buffer (100 mM KCl and 40 mM HEPES, pH 7.4) for 15 min in a Vitrobot Mark IV System (Thermo Fisher Scientific) at 100% relative humidity and then washed and inactivated with 20 mM 2-mercaptoethanol in the resuspension buffer. Then, 3 µL anti-phospho-tau AT8 (S202 and T205) monoclonal antibody (1:100; MN1020, Invitrogen) was applied for 15 min, and then the grids were washed three times. Next, 3 µL purified tau filaments from case 2 or 4 were applied, and the grids were incubated for 15 min; then the grids were washed and mechanically blotted for 3–4 s and plunge frozen. For cases 1 and 3, conventional methods were used: 3 µL purified tau filaments were applied to glow-discharged holey carbon grids and mechanically blotted for 3–4 s at 100% relative humidity and plunge frozen. For all datasets, super-resolution movies were collected using a Titan Krios microscope (Thermo Fisher Scientific) operating at 300 kV and equipped with a K3 direct electron-detection camera (Gatan, Inc.) with a BioQuantum energy filter (Gatan, Inc.) with a slit width of 20 eV. Super-resolution movies were recorded at 105,000× magnification (pixel size: 0.417 Å/pixel) with a defocus range of 0.8–1.8 µm and a total exposure time of 2.024 s fractionated into 0.025-s subframes.

Helical reconstruction

All data-processing steps were done in RELION [24]. The movies were motion-corrected using MotionCor2 [25] and Fourier-cropped by a factor of two to give the final pixel size of 0.834 Å per pixel. The dose-weighted summed micrographs were directly imported to RELION [26, 27] and were used for further processing. Contrast transfer function (CTF) was estimated using CTFFIND 4.1 [28]. The filaments were picked manually with a box size of 1120 pixels downsampled to 280 pixels. 2D classification was used to separate homogeneous segments for further image processing. After the first round of 2D classification, image alignment was performed to correct for the variable in-plane rotation of the fibril projections, as previously described [29]. For each well-resolved 2D class, the angle α between the x axis of the image and fibril-growth direction and the displacement δ of the fibril along the y axis from the center of the box were measured in ImageJ [30]. The nonzero values for α and δ were corrected using two MATLAB codes, as follows:

$$} \to } + \,}\left( }} \right)$$

$$} \to } + \,}\left( }} \right)$$

The new values of these parameters were updated in the “particles.star” file. Following the manual alignment, further 2D classification was performed to make sure all the 2D classes were well centered and aligned in the boxes. 3D initial models were then created de novo from the 2D class averages using an estimated helical rise of 4.75 Å and the crossover distances of the filaments estimated from the 2D classes with 1120-pixel boxes using RELION’s relion_helix_inimodel2d feature. The 3D initial models were low-pass filtered to 10 Å for further 3D classification. The particles corresponding to each initial model were re-extracted with 280-pixel boxes for further 3D classification. The best particles were selected from the 3D classification for each dataset. The final rise and twist were optimized in 3D autorefinement. The final 3D densities were sharpened using postprocessing in RELION, and the final resolution was determined from Fourier shell correlation at 0.143 Å from two independently refined half-maps using a soft-edged solvent mask. The final resolutions of tau PHF were 2.7 Å for case 1, 3.1 Å for case 2, 2.9 Å for case 3, and 5 Å for case 4. The final resolutions of tau SF were 3.0 Å for case 1, 3.2 Å for case 2, and 3.1 Å for case 3. For case 4, the resolution of the minor population of chronic traumatic encephalopathy (CTE)–like filaments was 7.8 Å. Imaging and reconstruction statistics are summarized in Supplementary Table S4.

Model building and refinement

For PHF and SF model building and refinement, previous models [31] were initially docked into the final maps for cases 1 and 2. A single strand of previously solved PHF (Protein Data Bank ID: 7NRQ) or SF (Protein Data Bank ID: 7NRS) was used to rigid-body fit with the final postprocessed density map in ChimeraX [32]. An initial round of real-space refinement was performed using PHENIX [33], after which the model was used against a single half-map in ChimeraX with the ISOLDE plugin [34] to check for clashes and rotamers. A final round of PHENIX real-space refinement was performed, and MolProbity [35] statistics for each model are summarized in Supplementary Table S5.