Cross-β helical filaments of Tau and TMEM106B in gray and white matter of multiple system tauopathy with presenile dementia

In neurodegenerative diseases and aging, the microtubuleassociated protein tau (MAPT) and the transmembrane lysosomal protein 106B (TMEM106B) become misfolded in different cell types and give rise to intracellular inclusions [3, 4]. The latter are made of amyloid filaments whose structures are being studied at the molecular level by cryogenic electron microscopy (cryo-EM). The nature of intracellular tau aggregates is determined by the participating tau isoforms and the structure of the amyloid filament(s) [1, 4]. TMEM106B aggregates, as discovered using cryo-EM, are composed of amyloid filaments that originate from the carboxy terminus of TMEM106B [3]. In the brain, the gray matter differs from the white matter for the cell types and the quantity of myelin. The gray and white matter both contain astrocytes, oligodendrocytes, and microglia. The gray matter contains nerve cell bodies, dendrites, axons, and synaptic terminals whereas the white matter contains axons as the only nerve cell component. The structure of tau and TMEM106B amyloid filaments present in the gray matter has been unveiled in several neurodegenerative diseases using cryo-EM; however, whether tau and TMEM106B filaments from the gray and white matter have the same fold is unknown. Neuropathologic, biochemical, genetic, and cryo-EM methods were used to study the gray and white matter from the frontal lobes of two individuals affected by multiple system tauopathy with presenile dementia (MSTD) (Supplementary Figs. 1–6). MSTD is a neurodegenerative disease caused by the MAPT intron 10 mutation + 3, which disrupts a stem-loop structure in the mRNA and leads to the presence of mainly four repeat (4R) tau isoforms in neurons and glia [2, 5–7]. Tau inclusions labelled by antibodies to phosphorylated tau and to 4R have different shapes in neurons, astrocytes, and oligodendrocytes. In the gray matter, tau inclusions are present in neurons and glia including tufted astrocytes, astrocytic plaques, and oligodendroglia with coiled bodies. In the white matter, tau inclusions are seen in oligodendrocytes with numerous coiled bodies and astrocytes (Supplementary Fig. 1, 2). TMEM106B inclusions in the gray and white matter were labelled with anti-TMEM239 (residues 239–250). In the gray and white matter, numerous intracellular inclusions were present mostly in the cell bodies and processes of astrocytes (Supplementary Fig. 2). Using a multidisciplinary approach, we characterized tau and TMEM106B amyloid filaments from the gray and the white matter. Relative to tau, we have determined that both areas contain filaments that have the AGD type 2 fold with a four-layered ordered structure accommodating amino acids 279–381 of tau, packing two protofilaments with C2 Md. Rejaul Hoq, Sakshibeedu R. Bharath, and Grace I. Hallinan have contributed equally to this work.

Immunohistochemical sections were counterstained with hematoxylin to show nucleus and cytoplasmic structure. In addition, the Gallyas silver method was used to demonstrate cytoplasmic inclusions, specifically of 4R tau, in nerve cells, astrocytes and oligodendroglia.
Sanger DNA analysis. Genomic DNA was extracted from frozen brain tissue.
Whole-exome sequencing (WES). Target enrichment made use of the SureSelectTX human all-exon library (V6, 58 megabase pairs; Agilent) and high-throughput sequencing was carried out using a HiSeq 4,000 (sx75 base-pair paired-end configuration; Illumina). Bioinformatics analyses were carried out as described [10]. In addition to the TMEM106B gene, the coding regions of genes corresponding to proteins reported to interact with the TMEM106B protein were screened for variants.
Extraction of tau and TMEM106B filaments. Sarkosyl-insoluble fractions were prepared from the cerebral cortex and the deep white matter of freshly frozen frontal lobes as previously described [15]. Briefly, ~4 g of gray and white matter tissues were homogenized in A68 extraction buffer consisting of 10 mM Tris-HCl, pH 7.4, 0.8 M NaCl, 1 mM EGTA, 5 mM EDTA, and 10% sucrose with protease and phosphatase inhibitors. Samples were centrifuged at 20,000 × g and the supernatants brought to 1% sarkosyl. Supernatants were incubated at room temperature (RT) while shaking. After centrifugation at 100,000 × g/1 h/4 °C, the sarkosyl-insoluble pellets were resuspended in 10 μl/g tissue 50 mM Tris-HCl, pH 7.4. The pellet was diluted in A68 extraction buffer and centrifugated at 20,000 × g/30 min/4 °C. The supernatant was centrifuged at 100,000 × g/1 h/4 °C and the final pellet resuspended in 20 mM Tris-HCl, pH 7.4 with 100 mM NaCl, and stored at 4 °C.
Images were taken on a Tecnai G2 Spirit Twin scope equipped with an AMT CCD Camera.
Tau seeding assay. Tau seeds were prepared as per filament extraction protocol. Prior to addition to cells, the resuspended pellets were sonicated with a probe sonicator at 75 W, for a total of 25 × 500 ms pulses. The probe was cleaned with isopropanol and water between samples. HEK 293 T cells stably expressing the repeat domain of Tau with a P301S mutation, fused at the C-terminus to either CFP or YFP, were obtained from ATCC [18]. Cells were cultured in DMEM supplemented with 10% FBS, 1% pen/strep, and 1% GlutaMax (Invitrogen). Cells were plated at a density of 50,000 cells/well in a 12-well plate, onto coverslips pretreated with 0.1 mg/ml poly-d-lysine for fixed cell imaging. Cells were incubated overnight at 37 °C and 5% CO2. The following day, 1 µl of sonicated seed material was combined with Lipofectamine 2000 (Thermo Fisher) and OptiMem medium, and incubated for 20 min at room temp. This transduction complex was then added onto the biosensor cells. Cells were incubated as before for 48 h. Cells were then washed with 1X PBS, fixed in 4% paraformaldehyde in PBS, stained with DAPI, and mounted onto microscope slides for imaging. Images were merged and cropped using ImageJ [25]. to search all de novo peptides above 15% score for over 300 potential PTMs and mutations. A 0.1 % peptide FDR cutoff (-10lgP≥21.8), PTM Ascore > 10, mutation ion intensity >1 and de novo only score > 80% were applied to the data followed by PEAKS LFQ analysis. The bio-informatic analysis Gene Ontology of identified proteins was done by DAVID Bioinformatics Resources 6.8 (https://david.ncifcrf.gov). The Venn diagram was generated using BioVenn (http://www.biovenn.nl/). Tau and TMEM106B high-resolution cryo-EM imaging. Cryo-EM grids of gray and white matter extracts of the two MSTD patients were prepared in a biosafety level 2 cabinet. 2-3 µl of the sample were applied on a graphene oxide coated EM grid, then washed with 10 mM Tris pH 7.8 before vitrifying using a semi-automated Gatan CP3 cryo-plunger. High resolution cryo-EM movies were collected on a FEI Titan Krios at 300 kV with a Gatan K3 detector mounted on a Quantum energy filter with 20 eV slit width (Table S1).
Helical reconstruction. CryoSPARC [23] was used to obtain motion corrected movies during data collection. The remaining steps of data processing were carried out in RELION 3.1 [24]. Filaments were manually picked and initially extracted with a box size of 1024 pixels and down-scaled to 256 pixels, to facilitate identification of filament types.
Then several rounds of 2D classification were carried out to find a homogeneous subset for different types of filaments. Helical rise was estimated from the first layer line centered on the meridian in the power spectrum of the 2D classes. Helical twist was estimated from the observed cross-over distance along the filaments. Iterative 3D classification and 3D refinement using a bare cylinder as the initial reference were carried to yield the final density map. The overall resolution was calculated from gold standard Fourier shell correlations at 0.143 between two independently refined halfmaps using the trueFSC.py program in JSPR [33].
Density map and atomic model analysis. The final map was sharpened using phenix.auto_sharpen [35] and a central region (~15 Å) of the filament was extracted for modeling using e2proc3d.py of the EMAN2 suite [34]. To distinguish polarity and handedness of the model, we manually built polyalanine models using COOT [6] in both directions, in two maps of opposite handedness resulting in four polyalanine models. To identify the protein sequence corresponding to the electron density map, phenix.sequence_from_map [36] was run with each of the four map/model pairs against all annotated sequences (8171 sequences) known to be expressed in the human brain (https://www.uniprot.org/uniprot/?query=tissue%3Abrain+reviewed%3Ayes+organism% 3A"Homo+sapiens+%28Human%29+%5B9606%5D"&sort=score). In addition to these sequences, the 18 isoforms of tau (nine splicing and nine hypothetical isoforms; https://www.uniprot.org/uniprot/P10636#sequences) were also included in the analysis.

Tau and TMEM106B in gray and white matter of individuals with MSTD.
Clinical information of cases #1 and #2 have been previously reported [29,30]. Case #1 was a 54-year-old female and case #2 was a 63-year-old female at the time of death. Both individuals had the intronic MAPT +3 splice-site mutation, a G-to-A transition in the intron following exon 10. In the gray matter of the frontal cortex of both cases, tauimmunoreactive neurons and neuropil threads were seen using antibodies against tau phosphorylated at Ser202/Thr205 (AT8) (Fig. 1) and 4R tau (RD4) (Fig. S2). In the subcortical white matter, numerous AT8 and RD4 immunoreactive intracytoplasmic inclusions in oligodendroglia (coiled bodies) were present. Coiled bodies were strongly positive for Gallyas silver staining (Fig. S2). Intracellular 3R immunoreactive deposits were not present in either gray or white matter (Fig. S2).
Immunohistochemistry of gray and white matter using anti-TMEM239 (residues 239-250) showed numerous cellular profiles containing immunopositive inclusions in the cytoplasm of the cell body and in elongated cytoplasmic processes in both cerebral cortex and subcortical white matter ( Fig. S1e-h). In the gray matter, the most numerous immunopositive cells appear to be astrocytes. Neurons appeared to have fewer immunopositive inclusions; however, considering the severe neuronal atrophy and loss that had occurred in the frontal cortex, it was difficult to clearly recognize the identity of some cells. Their identification was possible based on the morphology of the nucleus and the presence of immunopositive inclusions that were occasionally globular in shape and extended from the cytoplasm of the cell body to multiple cellular processes. Some immunopositive processes were seen near the vascular endothelium suggesting their astrocytic nature. In some instances, the immunopositive profiles appeared as small, dotted inclusions. In the white matter, immunopositive glial cells were less numerous than those observed in the gray matter. The lack of neurons in the white matter facilitated the recognition of the cellular identity. Both oligodendroglial cells and astrocytes appeared to be immunopositive with the TMEM239 antibody.
Biochemistry of tau from gray and white matter samples. Western blot analysis of the sarkosyl-insoluble fractions using antibodies to human tau (HT7) and phosphorylated tau (AT8) showed the presence of tau bands with a migration pattern corresponding to 4R tau, with identical electrophoretic mobility on either gray or white matter of both cases (Fig. S3a). Immunogold transmission electron microscopy preparations of dispersed tau filaments from gray and white matter showed twisted ribbon-like filaments that were decorated by the AT8 antibody, supporting the concept that the filaments were phosphorylated (Fig. S3b). Mass spectrometric analysis of the sarkosyl-insoluble fractions extracted from the gray and white matter of cases #1 and #2 revealed the presence of numerous post-translational modifications (PTMs) on tau, with no significant difference between gray and white matter or between cases. A number of proteins that co-purified with tau were identified by mass spectrometry and constitute the MSTD-tau "interactome" (Fig. S2m). A total of 2,006 proteins were found in association with tau deposits in the gray and white matter of both cases. Case #1 had 378 unique proteins in the gray matter, 339 proteins in the white matter, and 131 common proteins in the gray and white matter. Case #2 had 353 unique proteins in the gray matter, 377 proteins in the white matter, and 509 common proteins in the gray and white matter. The purified sarkosyl-insoluble fractions enriched for tau aggregates induced tau aggregation in a biosensor cell system. The tau isolated from the gray matter led to the formation of aggregates similar to those from tau isolated from the white matter of both cases. Tau isolated from the frontal cortex of a normal control or treatment with lipofectamine alone did not lead to the formation of aggregates (Fig.   S3c).
Cryo-EM of tau filaments from gray and white matter. We determined the structure of tau filaments at high resolution by cryo-EM imaging and 3D reconstruction. Visual inspection of cryo-EM micrographs showed the presence of different types of filaments in gray and white matter of both cases ( Figure S4a). 2D classification of filaments in RELION revealed that all filaments had a cross over distance of ~2,000 Å ( Figure S4b).
We identified a group of thinner filaments that had a width of 130-150 Å and a group of thicker filaments that had a larger variation in width (130-350 Å). The power spectrum of these 2D classes ( Figure S4c) displayed a layer line at ~4.8 Å with a peak on the meridian, suggesting that the filaments have around ~4.8 Å helical rise/subunit without 21 screw symmetry. Helical reconstruction in RELION revealed three distinct kinds of cross-β filaments in each of the four samples ( Figure S4b). Using RELION 3D classification, tau filaments were resolved using the following helical parameters, twist 0.41° and rise 4.8 Å. Filaments were made up of a doublet form of the filaments and were further refined to a resolution of ~4.5 Å. The doublets were composed of two protofilaments with C2 symmetry corresponding to the AGD type 2 filaments [29]. The AGD type 2 filaments have a four-layered ordered fold comprising residues 279-381 of tau. Type 1 and 3 AGD filaments were not observed. Filaments from gray matter were identical to filaments from white matter in both MSTD patients.
Cryo-EM of TMEM106B filaments from gray and white matter. Helical reconstruction in RELION of the cross-β filaments (Fig. S4) revealed two additional types of filaments, a singlet and a doublet form of the same fold with helical parameters twist of 0.41° and rise of 4.8 Å. The singlet and doublet forms were resolved to 2.9 and 3.5 Å resolution, respectively. The doublet form has C2 symmetry, and it comprises two identical protofilaments of the singlet type. Ab initio model building efforts with phenix.map_to_model, buccaneer, deepTracer or manual model building in COOT revealed that the density map could not be explained with any of the isoforms of tau.
Manually built polyalanine models in two different directions were used to identify and score the density at each Cα position using phenix.sequence_from_map and to unambiguously assign the TMEM106B sequence to the electron density (Fig. S5). We observed extra densities which corresponded to glycosylation at residues N145, N151, N164 and N183 of TMEM106B. The singlet and doublet filaments of TMEM106B displayed the same fold, consisting of 17 β strands formed by residues Ser120 to Gly254, as previously described [26]. Each chain is stacked along the helical axis and these successive layers interact through main chain hydrogen bonds reminiscent of parallel β-strand packing. Within a chain, strands β1, β14, β15, and β16 make up the central core. The packing of the β-strands is further strengthened by the presence of a disulfide bond between Cys214 and Cys253. Interestingly, the N-terminal residue Ser120 was hidden in the interior of the fibril and away from the exposed solvent (Fig.  S6). The remaining residues of the C-terminus Arg255 to Gln274 were likely to be disordered and hence not resolved in the electron density. In the doublet filaments of TMEM106B, there appears to be no direct interaction between the two protofilaments.
Instead, the protofilament interactions appear to be mediated through a nonproteinaceous density at the two-fold axis. This density was surrounded by a pair of arginine and lysine residues (Lys178 and Arg180) from one chain and its dyad axis related monomer suggesting a negatively charged factor mediating the interprotofilament packing (Fig. S7). Mass spectrometric analysis of the sarkosylinsoluble fraction extracted from the gray and white matter of case #2 determined the presence of three tryptic peptides corresponding to residues 130-SAYVSYDVQK-139, 130-SAYVSYDVQKR-140, and 248-YQYVDCGR-255. Residue Arg255 is not included in the fibril core and may be part of the C-terminal fuzzy coat of TMEM106B.

TMEM106B and interacting partners' genetic analyses. Genotyping of TMEM106B
in both cases showed that MSTD individuals were homozygous for Thr185.
Homozygosity at codon 185 was confirmed by Sanger sequencing. WES analysis revealed a point mutation, c.401g>a, resulting in an asparagine replacing serine at residue 134 (S134N) in one allele of the TMEM106B gene in Case #1 (Table S2). The change (rs 147889591) was confirmed by direct sequencing. We also analyzed a number of TMEM106B interacting partners [22]. A total of 20 genes were analyzed.
Coding variants were also present in APOE, TMEM106C, VPS11, and KCNMA1 in

Discussion
In recent years, extensive cryo-EM studies have unveiled the structure of amyloid filaments involved in neurodegeneration. This work led to the discovery of the complexity of the structure of filaments formed by proteins such as tau, α-synuclein, amyloid-β, prion protein amyloid, TDP-43, and more recently, TMEM106B. It has been shown that filaments of these proteins extracted from the gray matter differ according to the specific disease process in which they are involved [3,4,7,8,12,16,17,29,38].
TMEM106B is a lysosomal transmembrane protein vital for lysosomal health and has been implicated as a risk factor and/or disease modulator in many neurodegenerative diseases [21,22], however, the relationship between TMEM106B and tau is not well understood [22,37]. At the cellular level, there is evidence that this protein is present in neurons, oligodendrocytes, astrocytes and microglia [21,22].
Recently, while investigating ex-vivo filaments from individuals affected by neurodegenerative diseases, it was serendipitously discovered that TMEM106B filaments co-exist with other aggregated protein filaments in sporadic and inherited tauopathies, amyloid-β amyloidoses, synucleinopathies and TDP-43 proteinopathies [5,9,20,26]. Insoluble TMEM106B accumulation has recently also been reported to be associated with multiple sclerosis and has been emphasized for its role in oligodendroglia [27]. It is important to note that TMEM106B filaments have been found in the brain of elderly individuals being particularly severe in individuals 75 years and older, but not present in the brain of individuals younger than 46 years of age [9,26].
Furthermore, TMEM106B folds seem to be similar across the different diseases unlike tau filaments. These data strongly suggest that TMEM106B filament formation is an aging-related process [26]. The mechanism of TMEM106B filament formation is unknown. Filaments may be formed within the lysosome and released or the C-terminal domain may aggregate in the cytosol once it is processed at Arg119/Ser120 and released.
It is unknown whether tau filaments in gray matter and in white matter have identical structures in the same disease; however, it has been proposed that oligodendrocytes and astrocytes play an important role in tau propagation [2,11]. A similar question can be asked about the structure of TMEM106B in the same disease.
Therefore, the central aim of this study was to determine the near atomic structure of tau and TMEM106B in the white matter of individuals with MSTD and compare with the filament structure of both tau and TMEM106B isolated from gray matter.
Depending on the isoforms involved in the specific disorder [13], tau aggregates are found in neurons or in neurons, astrocytes and oligodendrocytes in the gray matter; whereas, tau aggregates are found in oligodendrocytes and astrocytes in the white matter. We isolated sarkosyl-insoluble material from gray and white matter from two individuals affected by MSTD [30][31][32] and determined the structures of tau to be identical within each brain region. We observed the same tau structure, corresponding to the AGD type 2 fold in both the cerebral cortex and subcortical white matter of the frontal lobe of the two individuals. The AGD type 2 fold was originally described in the gray matter from the MSTD cases #1 and #2, and the gray matter from two cases of argyrophilic grain disease (AGD), one of aging-related tau astrogliopathy (ARTAG), and three of frontotemporal dementia (FTD) with an intronic MAPT mutation (IVS10+16) [29]. Our data from gray versus white matter in MSTD suggest a common mechanism of tau folding that is indistinguishable between brain regions. Moreover, the sarkosylinsoluble fractions derived from gray and white matter in MSTD induced similar seeded aggregation on a FRET-based biosensor cell line. Tauopathies caused by intronic MAPT splice-site mutations (+3 and +16) disrupt a stem-loop structure in the mRNA, which leads to elevated levels of 4R tau isoforms in the brain [19,32], therefore it remains to be determined the exact role that tau propagation plays in tau folding in different brain areas and in the potential involvement of neurons and glia in the sequence of events leading to tau misfolding and propagation in vivo.
TMEM106B filaments were obtained from the same isolated sarkosyl-insoluble material prepared to extract tau filaments [14]. Other studies reported TMEM106B filaments isolated using variations of the sarkosyl extraction protocol [5,9,20,26]. It remains to be seen whether small variations in the extraction protocol could facilitate or hinder the extraction of TMEM106B. We observed the presence of TMEM106B filaments with an ordered core comprising residues S120 to G254. We identified singlet and doublet filaments of TMEM106B that display the same fold, containing a fivelayered core corresponding to fold I of Schweighauser et al. [26]. TMEM106B fold I has been seen to be able to accommodate threonine at codon 185 (T185), with both MSTD individuals being homozygous for T185. As previously described, we observed the presence of densities corresponding to four glycosylation sites (N145, N151, N164, and N183). Case #1 was heterozygous for a polymorphism at codon 134 (S134N) that is inside the core; however, due to the similarities in the electrodensity of both residues, it was not possible to establish by cryo-EM whether filaments contained Ser or Asn at 134. Asn134 may not be glycosylated due to the lack of a consensus sequence for N-glycosylation at 134 (Asn-Xaa-Ser/Thr (where Xaa is not Pro) in the TMEM106B sequence. By mass spectrometry analysis, we identified TMEM106B peptides in gray and white matter in MSTD. Interestingly, the tryptic peptides were identical to the peptides previously reported in FTLD-TDP [5]. The trypsin proteolysis/mass spectrometry approach provides additional insights into the structure of TMEM106B within the fibrillar polymer. Of the eleven potential trypsin-cleavage sites in the core (residues S120 to G254), only five were accessible to trypsin. Lack of cleavage at Lys124 in the β1 strand supports the atomic model generated by cryo-EM, in which the N-terminal residue Ser120 is hidden in the interior of the fibril and away from the exposed solvent. Interestingly, the Lys178 and Arg180 residues that are involved in the interactions between two protofilaments of fold I and a non-proteinaceous density at the two-fold axis were also fully protected.
This study has identified for the first time that tau filaments in gray and white matter have identical structures in individuals with MSTD, corresponding to the AGD type 2 fold. Although additional cases and diseases may need to be analyzed, our data support the notion that there is not region/cell type-specific folding of tau, and that the same tau fold may be found in neurons and glia in the same disease. Our finding of TMEM106B in gray and white matter regions of the brain imply a similar mechanism of TMEM106B fibrillization in both neuronal and glial cell types. However, considering that the frontal cortex of the two MSTD patients had undergone severe atrophy and neuronal loss, we need to be cautious in assigning the presence of the TMEM106B extracted