Age-dependent formation of TMEM106B amyloid filaments in human brains

Many age-dependent neurodegenerative diseases, such as Alzheimer’s and Parkinson’s, are characterized by abundant inclusions of amyloid filaments. Filamentous inclusions of the proteins tau, amyloid-β, α-synuclein and transactive response DNA-binding protein (TARDBP; also known as TDP-43) are the most common1,2. Here we used structure determination by cryogenic electron microscopy to show that residues 120–254 of the lysosomal type II transmembrane protein 106B (TMEM106B) also form amyloid filaments in human brains. We determined the structures of TMEM106B filaments from a number of brain regions of 22 individuals with abundant amyloid deposits, including those resulting from sporadic and inherited tauopathies, amyloid-β amyloidoses, synucleinopathies and TDP-43 proteinopathies, as well as from the frontal cortex of 3 individuals with normal neurology and no or only a few amyloid deposits. We observed three TMEM106B folds, with no clear relationships between folds and diseases. TMEM106B filaments correlated with the presence of a 29-kDa sarkosyl-insoluble fragment and globular cytoplasmic inclusions, as detected by an antibody specific to the carboxy-terminal region of TMEM106B. The identification of TMEM106B filaments in the brains of older, but not younger, individuals with normal neurology indicates that they form in an age-dependent manner.

cerebral cortex 11 . Furthermore, the variant rs1990622 has been reported to correlate with reduced neuronal degeneration during ageing, independently of disease 11,12 .
Cryo-EM structure determination can also be used to identify previously unknown filaments. Here we have used cryo-EM to show that residues 120-254 from the luminal domain of TMEM106B form amyloid filaments in human brains. We initially observed TMEM106B filaments in the brains of individuals with familial and sporadic tauopathies, Aβ amyloidoses, synucleinopathies and TDP-43 proteinopathies. However, the role of TMEM106B filaments in disease remains unclear. They were not observed in brains from young individuals, but their presence in brains from older individuals with normal neurology (controls) indicates that TMEM106B filaments may form in an age-dependent manner. It remains to be determined how these findings relate to those from genetic association studies.
Using sarkosyl extraction protocols that were originally developed for α-synuclein 18,22 , we observed a common type of filament that seemed to lack a fuzzy coat in the cryo-EM micrographs from cases of various conditions with abundant filamentous amyloid deposits. Structure determination to resolutions sufficient for de novo atomic modelling revealed that the ordered cores of these filaments consisted of residues 120-254 from the carboxy-terminal, luminal domain of TMEM106B and that the filaments were polymorphic (Fig. 1). We solved the structures of TMEM106B filaments from a number of brain regions of 22 individuals with abundant amyloid deposits, and from the frontal cortex of 3 individuals with normal neurology and no or only a few amyloid deposits (cases 1-25; Methods, Table 1 and Extended Data Table 1).
The neurodegenerative conditions for which we solved structures of associated TMEM106B filaments included sporadic and inherited Alzheimer's disease, pathological ageing, corticobasal degeneration, sporadic and inherited FTLD (FTLD-TDP types A and C, and familial frontotemporal dementia and parkinsonism linked to chromosome 17 caused by MAPT mutations), argyrophilic grain disease, limbic-predominant neuronal inclusion body four-repeat tauopathy, ageing-related tau astrogliopathy, sporadic and inherited Parkinson's disease, dementia with Lewy bodies, multiple system atrophy (MSA) and amyotrophic lateral sclerosis. We observed three different TMEM106B protofilament folds (I-III; Fig. 1 and Extended Data Figs. [1][2][3][4]. Filaments with fold I were more common than filaments with folds II or III. For all three folds, we determined the structures of filaments that were made of a single protofilament. We also determined the structures of filaments comprising two protofilaments of fold I, related by C 2 symmetry. In each individual, we observed only filaments with a single fold, without a clear relationship between folds and diseases. The TMEM106B folds shared a similar five-layered ordered core comprising residues S120-G254 and contained 17 β-strands, each ranging between 3 and 15 residues. Our best maps for filaments with folds I, II and III had resolutions of 2.6, 3.4 and 2.8 Å, and came from case 1 (sporadic Alzheimer's disease), case 19 (MSA) and case 17 (MSA), respectively ( Fig. 1). TMEM106B remained fully glycosylated in all folds, as reflected by large extra densities corresponding to glycan chains attached to the side chains of N145, N151, N164 and N183. The fifth glycosylation site at N256 is outside the ordered core, with the C-terminal 20 residues being probably disordered. We divide the sequence that forms the ordered cores of the folds into three regions according to their degree of structural conservation: the amino-terminal region (S120-T166) is conserved in all three folds; the C-terminal region (Y211-G254) is conserved only in folds I and II; and the middle region (A167-M210) varies between folds.
The N-terminal region, S120-T166, forms the first two layers of the five-layered ordered cores. It comprises one long and five short β-strands  Article that constitute a tightly packed core with hydrophobic and neutral polar residues on one side, and a large polar cavity that is filled by solvent on the other side. The three glycosylation sites in this region are located in the outer layer, adopting an extended conformation. The N-terminal residue S120 in the inner layer is buried inside the ordered core, where it packs closely against E161 from the N-terminal region and H239 and E241 from the C-terminal region (Extended Data Fig. 4). The C-terminal region, Y211-G254, forms the two central layers of the ordered cores. It adopts a compact hairpin-like structure, the ends of which are held together by a disulfide bond between C214 and C253. Segment F237-E246 that packs against the N-terminal region has the same conformation in all three folds, whereas in the rest of the hairpin-like structure, 15 residues have opposite 'inward/outward' orientations in fold III compared with those in folds I and II. Moreover, despite similar interfaces between N-and C-terminal parts in all three folds, these regions in fold III are separated along the filament axis by one more rung than in folds I and II (Extended Data Fig. 3e).
The middle region, A167-M210, forms the fifth layer of the ordered cores and contains the fourth glycosylation site at N183. In fold I, this region packs loosely against the other side of the C-terminal hairpin-like region with the formation of three large amphipathic cavities. In fold II, these internal cavities are smaller than in fold I. We observed two subtypes of fold II (IIa and IIb) that differed mainly by the conformation of segment A167-I187. The packing of the middle region against the C-terminal region is tightest in fold III, leaving only one sizable cavity with a salt bridge between E206 and K220. Only fold III shows cis isomerization of P189. In folds IIb and III, there is a large extra density at the end of the side chain of K178, suggesting that this residue may be post-translationally modified. Likewise, there is an extra density in front of the side chain of Y209 in fold I, but not in the other folds (Extended Data Fig. 1). It is possible that these residues determine the formation of the different folds. Genotyping of all individuals (Table 1) showed that the alleles encoding T185 or S185 were equally represented. Individuals with fold I were homozygous for T185 or S185, or heterozygous, indicating that fold I can accommodate a threonine or a serine at position 185. Owing to the compatibility of both residues with the glycosylation motif at N183, no differences in the associated glycan densities were observed. Fold II was found only in case 19, which was homozygous for T185. Seven out of eight individuals with fold III were homozygous for S185, with the remaining individual being heterozygous. It is possible that the packing of the side chain of residue 185 in the interior of fold III leaves insufficient space to accommodate a threonine.
In all three folds, residues G177-N183 adopt a conserved conformation, with the positively charged residues K178 and R180 pointing outwards. In filaments made of two protofilaments with fold I, two pairs of these residues are on opposite sides of a contiguous extra density that runs  along the helical symmetry axis. As the cofactor responsible for this density probably does not obey the imposed helical symmetry, the map in this region is of insufficient quality to allow its identification. Although we did not solve the structures of filaments comprising two protofilaments of folds II or III, the micrographs of case 19, the only individual for which we observed filaments with fold IIa/b, and the micrographs of case 21, with fold III, also contained wider filaments that probably comprised two TMEM106B protofilaments (Extended Data Fig. 5).
In the absence of an experimentally determined native structure, we examined the structure of TMEM106B as predicted by AlphaFold 23 (Extended Data Fig. 6). Whereas the formation of amyloid filaments is often associated with natively unfolded proteins or low-complexity protein domains, the sequence S120-G254, which spans the ordered core of TMEM106B filaments, is confidently predicted to be a globular domain of the immunoglobulin-like β-sandwich fold. Glycosylation sites at N145, N151, N164 and N183 are positioned on the outside of the fold, and the disulfide bond between C214 and C253 is also predicted to form in the native structure. The β-sandwich domain is connected to a single transmembrane helix, without a flexible linker sequence. Moreover, there is a hydrophobic surface patch at this end of the domain, suggesting that it is positioned close to the membrane. It thus seems unlikely that the cleavage site at S120, the buried N-terminal residue in all TMEM106B filaments, can be accessed by lysosomal proteases. Shedding of the luminal domain may happen in a noncanonical way.
We previously showed that distinct amyloid folds of tau, α-synuclein, Aβ and TDP-43 characterize different neurodegenerative diseases [13][14][15][16][17][18]20,21 . We now describe the presence of TMEM106B filaments in many of these diseases, without a correlation between folds and diseases. Therefore, we also examined 16 brains from individuals with normal neurology that varied in age between 20 and 101 years. By immunoblotting with an antibody raised to a peptide corresponding to residues 239-250 of human TMEM106B (antibody TMEM239), the sarkosyl-insoluble fractions from disease cases showed a band of 29 kDa, which probably corresponded to the 17-kDa C-terminal fragment plus 12 kDa of glycosylation and other modifications ( Fig. 2 and Extended Data Fig. 7). This band was not present in the brains from individuals with normal neurology aged less than 46 years, excluding the possibility that TMEM106B assembly was an artefact caused by tissue extraction. However, we consistently observed the 29-kDa band in the brains from control individuals older than 69 years. Interestingly, the 29-kDa band was not present in the frontal cortex from a 15-year-old individual with early-onset dementia with Lewy bodies 24 (Fig. 2b). In agreement with these observations, immunohistochemistry of brain sections with the antibody TMEM239 showed staining of inclusions in disease cases and older control individuals, but not in younger controls ( Fig. 3 and Extended Data Fig. 8). It is not known how these inclusions relate to lysosomes. Cryo-EM structure determination showed the presence of TMEM106B filaments with one or two protofilaments of fold I in the frontal cortex from three controls, aged 75, 84 and 101 years.
Our results suggest that amyloid filaments of the lysosomal protein TMEM106B form in an age-dependent manner in human brains, without Article a clear mechanistic connection to disease. Until now, the presence of abundant intraneuronal amyloid filaments in human tissues has always been associated with disease. Dominantly inherited mutations in the genes encoding tau, α-synuclein and TDP-43 cause neurodegenerative diseases. In addition, cryo-EM structures of amyloid filaments made of these proteins exhibit distinct folds that are characteristic of different diseases [13][14][15][16][17][18]21 . Although TMEM106B has been associated with frontotemporal dementias and other diseases, the evidence for a causal relationship between TMEM106B aggregation and disease remains unclear, and distinct TMEM106B folds do not characterize different diseases. Instead, our observations suggest that TMEM106B filaments form in an age-dependent manner. Like lipofuscin, a lysosomal complex of oxidized proteins and lipids that develops in an age-dependent manner in many tissues 25 , TMEM106B filaments may also form in lysosomes, even though staining for TMEM106B inclusions was not always associated with the presence of lipofuscin autofluorescence. Lysosomal dysfunction has been implicated in the pathogenesis of neurodegenerative diseases 26 . Further studies are needed to determine whether TMEM106B filaments can be found in tissues other than the central nervous system and to assess the role of filament formation in relation to human ageing and pathologies.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04650-z.

Clinical history and neuropathology
We determined the cryo-EM structures of TMEM106B filaments from the brains of 25 individuals (Table 1 and Extended Data Table 1). Most individuals have been reported previously 14,[16][17][18]20 . Unpublished cases are described below. Early-onset Alzheimer's disease (EOAD; case 3) was in a 58-year-old woman who died with a neuropathologically confirmed diagnosis following a 7-year history of memory loss. FTDP-17T (case 7) was in a 55-year-old man who died with a neuropathologically confirmed diagnosis following a 2-year history of behavioural changes, aphasia and dementia caused by a P301L substitution in MAPT. His brother, sister and mother were also affected. Sporadic PD (case 12) was in an 87-year-old male who died with a neuropathologically confirmed diagnosis following an 8-year history of PD. Inherited PD (case 14) was in a 67-year-old woman who died with a neuropathologically confirmed diagnosis following a 10-year history of PD caused by a G51D substitution in SNCA. FTLD-TDP-C (case 21) was in a 65-year-old woman who died with a neuropathologically confirmed diagnosis following a 9-year history of semantic dementia. ALS (case 22) was in a 63-year-old woman who died with a neuropathologically confirmed diagnosis of ALS stage 4, type B TDP-43 pathology, following a history of 2 years and 5 months of motor symptoms, without dementia. Control 1 (case 23) was a 75-year-old man who died of coronary heart disease without neuropathological abnormalities. Control 2 (case 24) was a 84-year-old man with mild tau pathology (Braak stage 1) who died of sepsis. Control 3 (case 25) was a 101-year-old man with mild tau pathology (Braak stage 1) and mild cerebral amyloid angiopathy who died of pneumonia.

Extraction of TMEM106B filaments
Sarkosyl-insoluble material was extracted from frontal cortex (EOAD, FTLD-TDP-C and control cases 1-16), cingulate cortex (sporadic PD), temporal cortex (inherited PD and FTDP-17T) and motor cortex (ALS), essentially as described previously 22 . Similar extraction methods were used for all other cases, which have been described in the references in Extended Data Table 1. The original sarkosyl extraction method, which we used in our work on the cryo-EM structures of tau filaments from Alzheimer's disease, chronic traumatic encephalopathy and Pick's disease [13][14][15] , uses sarkosyl only after the first, low-speed centrifugation step 27 . A previously published method 22 also uses sarkosyl at the beginning (before the first centrifugation step). This protocol change was essential for detecting abundant TMEM106B filaments, possibly because clumped filaments end up in the first pellet when sarkosyl is not yet present in the original method. In addition, the previously published method 22 uses a gentler clearing spin at the end, which results in an increase in the amount of filaments in the final sample. In brief, tissues were homogenized in 20 vol (w/v) extraction buffer consisting of 10 mM Tris-HCl, pH 7.4, 0.8 M NaCl, 10% sucrose and 1 mM EGTA.
Homogenates were brought to 2% sarkosyl and incubated for 30 min at 37 °C. Following a 10-min centrifugation at 10,000g, the supernatants were spun at 100,000g for 20 min. The pellets were resuspended in 700 µl g −1 extraction buffer and centrifuged at 5,000g for 5 min. The supernatants were diluted threefold in 50 mM Tris-HCl, pH 7.4, containing 0.15 M NaCl, 10% sucrose and 0.2% sarkosyl, and spun at 166,000g for 30 min. Sarkosyl-insoluble pellets were resuspended in 50 µl g −1 of 20 mM Tris-HCl, pH 7.4 containing 100 mM NaCl.

Immunoblotting and immunohistochemistry
Immunoblotting was carried out as described previously 28 . Sarkosyl-insoluble pellets were diluted 1:3 and sonicated in a water-bath for 10 min at 50% amplitude (QSonica). They were resolved on 12% Bis-Tris gels (Novex) and the antibody TMEM239 (a rabbit polyclonal antibody that was raised to a synthetic peptide corresponding to residues 239-250 of human TMEM106B) was used at 1:2,000.
To enhance the signal, membranes were boiled in PBS for 10 min at 95 °C. For immunohistochemistry, formalin-fixed, paraffin-embedded 8-µm-thick sections were incubated overnight in xylene. Following deparaffinization, the sections underwent heat-induced epitope retrieval in Tris-EDTA buffer (10 mM Tris base, 1 mM EDTA, 0.05% Tween 20, pH 9). Peroxidase was quenched by incubation in 3% hydrogen peroxide in PBS containing 20% methanol for 30 min, followed by a 15-min incubation in BLOXALL endogenous blocking solution (Vector Laboratories). After a brief wash in PBS + 0.3% Triton X-100 (PBST), the sections were incubated in blocking buffer (2.5% bovine serum albumin, 5% horse serum in PBST) for 1 h at room temperature. This was followed by an overnight incubation at 4 °C with primary antibody in blocking solution (TMEM239 was used at 1:500 and N-terminal rabbit polyclonal TMEM106B antibody A303-439A (Bethyl Laboratories) 29 , which was raised to a synthetic peptide corresponding to residues 1-50 of human TMEM106B, was used at 1:250). After three washes with PBST, the sections were incubated with ImmPRESS-HRP polymer anti-rabbit detection antibody (Vector Laboratories) for 2 h at room temperature. After another three washes with PBST, Vector SG substrate (peroxidase) was added to visualize the antigen. Sections were counterstained with nuclear fast red and covered with a coverslip using Entellan mounting medium (Merck). Images were acquired with a QImaging Retiga 2000R CCD camera using an Olympus BX50 microscope.
Cloning TMEM106B C-terminal fragment (120-274) incorporated in pET3A was purchased from Genscript. The construct lacking residues 239-250 (Δ239-250) was made using in vivo assembly 30  concentrator, and used for immunoblotting to establish the specificity of the antibody TMEM239 (Extended Data Fig. 7).

Cryo-EM
For all cases, except EOAD, FTDP-17T, LNT, sporadic PD, inherited PD, FTLD-TDP-C, ALS and control cases 1-3, the cryo-EM datasets have been described in the references in Extended Data Table 1. For the remaining cases, resuspended sarkosyl-insoluble pellets were applied to glow-discharged holey carbon gold grids (Quantifoil R1.2/1.3, 300 mesh) and plunge frozen in liquid ethane using an FEI Vitrobot Mark IV. FTLD-TDP-A, FTLD-TDP-C and ALS samples were treated with 0.4 mg ml −1 pronase for 50-60 min before glow discharging, which further improved the TMEM106B filament yield. Images for cases of EOAD, FTDP-17T, LNT, FPD, FTLD-TDP-C and ALS were acquired using EPU software on Thermo Fisher Titan Krios microscopes, operated at 300 kV, with a Gatan K2 or K3 detector in counting mode, using a Quantum energy filter (Gatan) with a slit width of 20 eV to remove inelastically scattered electrons. Images for EOAD, sporadic PD and control cases 1-3 were acquired on a Thermo Fisher Titan Krios, operated at 300 kV, using a Falcon-4 detector and no energy filter.

Helical reconstruction
Movie frames were gain corrected, aligned, dose weighted and then summed into a single micrograph using RELION's own motion correction program 31 . The micrographs were used to estimate the contrast transfer function (CTF) using CTFFIND-4.1 (ref. 32 ). All subsequent image-processing steps were performed using helical reconstruction methods in RELION (refs. 33,34 ). TMEM106B filaments were picked manually, as they could be distinguished from filaments made of tau, Aβ, α-synuclein and TDP-43 by their general appearance and the apparent lack of a fuzzy coat. TMEM106B filaments comprising one or two protofilaments were picked separately. For all datasets, reference-free 2D classification was performed to select suitable segments for further processing. Initial 3D reference models were generated de novo from the 2D class averages using an estimated rise of 4.75 Å and helical twists according to the observed crossover distances of the filaments in the micrographs 31 for datasets of cases 10 (LNT; folds I-s and I-d), 18 (MSA; fold I-d), 19 (MSA; folds IIa and IIb) and 17 (MSA; fold III). Refined models from these cases, low-pass filtered to 10-20 Å, were used as initial models for the remaining cases. Combinations of 3D auto-refinements and 3D classifications were used to select the best segments for each structure. For all datasets, Bayesian polishing 35 and CTF refinement 36 were performed to further increase the resolution of the reconstructions. Final reconstructions were sharpened using the standard post-processing procedures in RELION, and overall final resolutions were estimated from Fourier shell correlations at 0.143 between the two independently refined half-maps, using phase randomization to correct for convolution effects of a generous, soft-edged solvent mask 37 . Further details of data acquisition and processing for the datasets that resulted in the best maps for five different TMEM106B filaments (filaments made of one or two protofilaments with fold I, as well as filaments made of one protofilament with fold IIa, fold IIb or fold III) are given in Extended Data Table 2.
Model building TMEM106B was identified by scanning the human proteome with different sequence motifs 38 , deduced from initial maps of folds I and III. , was the most effective, resulting in a hit for only TMEM106B, the sequence of which corresponded well to the entire maps. Atomic models comprising three β-sheet rungs were built de novo in Coot 39 in the best available map for each of the five different structures. Coordinate refinement was performed in ISOLDE (ref. 40 ). Dihedral angles from the middle rung, which was set as a template in ISOLDE, were also applied to the rungs below and above. For each refined structure, separate model refinements were performed for the first half-map, after increasing the temperature to 300 K for 1 min, and the resulting model was then compared to that same half-map (FSC work ) as well as the other half-map (FSC test ) to confirm the absence of overfitting. Final statistics for the refined models are given in Extended Data Table 2.