The Cryo-EM Effect: Structural Biology of Neurodegenerative Disease Aggregates

Abstract

Neurogenerative diseases are characterized by diverse protein aggregates with a variety of microscopic morphologic features. Although ultrastructural studies of human neurodegenerative disease tissues have been conducted since the 1960s, only recently have near-atomic resolution structures of neurodegenerative disease aggregates been described. Solid-state nuclear magnetic resonance spectroscopy and X-ray crystallography have provided near-atomic resolution information about in vitro aggregates but pose logistical challenges to resolving the structure of aggregates derived from human tissues. Recent advances in cryo-electron microscopy (cryo-EM) have provided the means for near-atomic resolution structures of tau, amyloid-β (Aβ), α-synuclein (α-syn), and transactive response element DNA-binding protein of 43 kDa (TDP-43) aggregates from a variety of diseases. Importantly, in vitro aggregate structures do not recapitulate ex vivo aggregate structures. Ex vivo tau aggregate structures indicate individual tauopathies have a consistent aggregate structure unique from other tauopathies. α-syn structures show that even within a disease, aggregate heterogeneity may correlate to disease course. Ex vivo structures have also provided insight into how posttranslational modifications may relate to aggregate structure. Though there is less cryo-EM data for human tissue-derived TDP-43 and Aβ, initial structural studies provide a basis for future endeavors. This review highlights structural variations across neurodegenerative diseases and reveals fundamental differences between experimental systems and human tissue derived protein inclusions.

INTRODUCTION

Alzheimer disease (AD) was first described in 1906 (1), Parkinson disease (PD) in 1817 (2), and amyotrophic lateral sclerosis (ALS) in 1874 (3). Despite over a century of interest and investigation, these diseases do not have a US Food and Drug Administration (FDA)-approved progression-altering therapy. Protein aggregates are a hallmark feature of AD, PD, ALS, and many other neurodegenerative diseases that seem to play a role in driving the diseases. Specifically, aggregates of tau, amyloid-β (Aβ), α-synuclein (α-syn), and transactive response element DNA-binding protein of 43 kDa (TDP-43) are some of the most common. The general light-microscopic morphologies of these aggregates are known. Until recently though, the structures and specific conformations of these aggregates at molecular resolution have remained elusive.

Understanding the molecular structures of aggregates related to neurodegenerative disease may help discover therapies for neurodegenerative disease by enhancing our basic understanding of the molecular pathogenesis of these diseases. More and more evidence is emerging that the conformation and molecular organization of aggregates correlates with the disease type and progression. For example, even within the same disease, multiple system atrophy (MSA), the conformation of α-syn in aggregates correlated with aggressiveness of disease progression (4, 5).

In vitro aggregates have been used to model the structure and biochemistry of neurodegenerative aggregates. In vitro aggregates can be made using recombinant proteins similar to the proteins found in disease aggregates. Ex vivo fibrils, derived from autopsied human brain tissue, are a scarce resource, and so in vitro fibrils present an enticing alternative. In vitro fibrils are relatively easy to make and can be made from recombinant protein. Unfortunately, in vitro structures to date have not accurately represented the fibril structures found in human disease (6–14). Furthermore, inhibitors rationally designed against ex vivo structures have proved to be more effective than those designed against in vitro fibrils (15).

Technological Advances in Cryo-Electron Microscopy

The advent and improvement of cryo-electron microscopy (cryo-EM) has been paramount in the determination of aggregate structure. Traditionally, X-ray crystallography and solid-state nuclear magnetic resonance (ssNMR) spectroscopy were the preferred methods for determination of protein structures. However, crystallography requires the growth of crystals, which can be unpredictable and may not represent the native state of a protein in solution. ssNMR may preserve a natural conformation more than X-ray crystallography and can even be used to study protein dynamics but requires very stable proteins at very high concentration (mM scale), while yielding relatively low-resolution structures. In contrast, cryo-EM allows for the study of high molecular weight proteins or protein complexes that have been difficult or impossible by other methods, maintains the native state of a sample in solution, and can capture more than one conformation of a protein in the same experiment (16). Though room temperature EM was first used to determine a 3D structure in 1968 (17) and cryo-EM was proposed in 1974 (18), until recently cryo-EM technology has been inadequate in resolving structures to near-atomic resolution. However, since 2013, several advances in cryo-EM technology have allowed for even atomic resolution in ideal samples (∼1.2 Å) (19). New microscopes allow for highly coherent 300 kV illumination, automatic sample insertion that can store and handle several grids at one time, automatic liquid nitrogen refilling system, and increased imaging stability from an extensive lens system. These upgrades allow for more efficient data collection than was previously possible and allow for higher resolution information to be retained in the imaging process. Direct electron detectors have also been developed. Direct electron detectors are more sensitive and substantially faster than traditional cameras. These detectors can take many frames per second as opposed to requiring long exposure times for a single image. The stack of frames can be analyzed to correct for specimen movement and image blurring that reduces reconstruction resolution. These recent advances in cryo-EM have allowed it to overcome resolution barriers that previously prevented its widespread use in 3D structural determination.

Protein aggregates derived from human brains are not generally amenable to structural determination by X-ray crystallography or ssNMR (exceptions noted below). It is very difficult to obtain enough sample to reach concentrations required for ssNMR and they do not readily form crystals for X-ray crystallography. In contrast, cryo-EM allows direct 2D imaging of these proteins which can then be reconstructed using helical reconstruction to obtain near-atomic resolution 3D structures of the fibrils (20). Helical reconstruction has proved key in the ability to reconstruct a near-atomic resolution 3D structure for ex vivo neurodegenerative disease associated fibrils (discussed below).

Here, we review recent structures of ex vivo, in vivo, and brain-derived aggregates of tau, Aβ, α-syn, and TDP-43 solved using cryo-EM.

Tau: Heterogeneity and Structure

Tau has been associated with many diseases, accordingly dubbed “tauopathies.” The diseases within the tauopathy category differ from each other both clinically and pathologically, including formation of aggregates containing different isoforms of the tau protein.

Tau’s native function mostly involves assisting with the assembly of tubulin into microtubules and the stabilization of microtubules (21). In its native form, tau exists in humans in an equal ratio of 6 isoforms that can be grouped into 2 main categories called 3R and 4R, containing 3 or 4 repetitions respectively of a 31- or 32-residue motif (Fig. 1A). These repeats make up the microtubule binding repeat (MTBR) of tau. In 3R, the second repeat (R2) is missing. Tau also has a proline rich region located N-terminal to the MTBR with additional N-terminal and C-terminal extensions beyond the proline rich region and MTBR, respectively.

FIGURE 1.

Tau fibril structures, common fragments used of in vitro tau aggregation, and the cores that compose fibrils. (A) The domains of full-length tau showing the N-terminal extension (red) that can have an N1 (orange) or N2 (yellow) insert, proline rich region (green), microtubule binding domain (pink/purple) that can have 3 or 4 repeats, and the C-terminal extension (blue). The common fragments K18 and K11 are shown with the PHF6 peptide, thought to be a nucleus of aggregation, labeled. The amino acids that make up the core of ex vivo fibrils are shown with the different cores that compose in vitro tau fibril structures that have been confirmed by cryo-EM. (B) The paired helical filament (PHF) tau fibril core structure that is found in brains of Alzheimer disease (AD) patients (PDB: 5O3L, EMDB: 3741). (C) The straight filament (SF) tau fibril core structure that is found in the brain of AD patients. It is less prevalent and less twisted than the PHF (PDB: 5O3T, EMDB: 3743). (D) The core folds of different tau protofilaments that have been solved by cryo-EM (PDB: 5O3L, 6GX5, 6NWP, 6TJX, 6QJH, 6QJM, 6QJP, 6QJQ) with the PHF6 (black amino acids) peptide that is thought be a nucleus of aggregation shown in the core.

Tau is known to aggregate into filaments in >20 clinicopathologically diagnosed diseases with different isoforms predominating in aggregates of different diseases. Diseases that will be discussed further in this review, along with their predominant tau isoforms, include AD (3R and 4R), chronic traumatic encephalopathy ([CTE], 3R and 4R neurofibrillary tangles [NFT’s], 4R thorn-shaped astrocytes [TSAs]), Pick disease ([PiD], 3R), and corticobasal degeneration ([CBD], 4R) (22, 23).

Prior to the advent of cryo-EM for robust image analysis, it was difficult to elucidate the atomic structure of physiologically relevant tau. Paired helical filaments (PHFs), a type of tau fibril found in AD, were first described and shown by EM in the 1960s. Later these fibrils were shown to consist of a rigid core and a so-called “fuzzy coat” (24–26). It was not until 1991 that Crowther correctly showed the general cross-section morphology, the protofilament orientation, and the dimeric nature of PHFs and straight filaments (SFs), another type of tau fibril found in AD (27). Cryo-EM has allowed a greater understanding of the molecular interactions that make up the PHF and SF folds. The first atomic structure of tau fibrils derived from human brain was published in 2017 (28), 26 years after the protofilament morphology was elucidated. Notably, recombinant tau is often used in studies but had not been compared with human brain-derived tau on a molecular level until recently (6). Additionally, a crystal structure of the PHF6 peptide, a portion of tau long thought to be a nucleus of aggregation, came out in 2007, before a disease-derived tau fibril structure had been solved (29, 30). Though helpful for research into how tau functions in vitro as a protein, the continued relevance to human disease of these traditional methods and findings has been brought into doubt by recent advances in structural knowledge of tau fibrils using cryo-EM.

Tau: Near-Atomic Structure

The first atomic structure of tau fibrils showed that SFs and PHFs from AD have the same protofilament but differ in the interprotofilament packing (Fig. 1B, C) (28). Additionally, it was shown that these structures were retained in multiple different sporadic cases of AD as well as in an inherited case of AD (31). The finding that different AD cases, both sporadic and inherited, share the same tau fibril morphology at an atomic level strengthens the theory that fibril structure may indeed play a role in clinical and pathological findings of neurodegenerative disease and is not in fact patient specific.

Though tau fibrils demonstrate consistent morphology among different cases of AD, fibril morphology is not consistent among different tauopathy diseases. All available tau fibril structures show an ordered core of beta strands of loops and turns that have different organizations and molecular interactions. Most of these structures lack much of the disordered “fuzzy coat” that normally surrounds tau fibrils due both to experimental procedures which include protease treatment of the brain isolates and averaging out of this disordered area during image analysis (exception noted below) (28, 31–34). Nonetheless, ex vivo tau fibril structures from 4 different tauopathies (AD, PiD, CTE, CBD) show different protofilament structures. The 4 cores are made of approximately the same residues but take different shapes in protofilament packing (Fig. 1D, E). These ex vivo tau filaments derived from human brain show that the filament cores include R3, R4, and 10-12 amino acids C-terminal to R4 no matter the isoform (28, 32–35). The remainder of the cores varies by disease (Fig. 1A). The folds of these aggregates may dictate how the tauopathies expand and propagate in the brain. Similarities and differences in the folds could be an underlying factor in the similarities and differences of tauopathies clinically and pathologically.

Aggregates seen in both CBD and CTE contain unidentified nonproteinaceous densities within the core that are not covalently linked to tau and so could not be posttranslational modifications within the fold (32, 35). Sequestration of materials has been shown to add to neurotoxicity (36), thus the nonproteinaceous densities may be evidence of tau fibrils sequestering necessary compounds within a cell, leading to exacerbation of neural cell death and dysfunction. The nonproteinaceous densities may also be an area to target for aggregate disruption if they provide stability for the fibrils.

The posttranslational modifications of AD and CBD fibrils were analyzed in the context of fibril structure. Notably, these modifications were able to be analyzed because the fibrils were not treated with proteases that have been shown to cleave posttranslational modifications in addition to the C- and N-terminal regions of tau (35). Acetylation and ubiquitination are the predominant posttranslational modifications within the core region of AD and CBD fibrils. Acetylation occurs at lysine residues, neutralizing the positive charge and making tau fibrils less soluble, more prone to aggregation, and more resistant to aggregate degradation (35, 37–39). Ubiquitination also occurs at lysine residues and has been seen by cryo-EM on fibrils (35). Ubiquitination is generally thought of as a marker for removal or protein aggregates (40). However, ubiquitin chains may actually stabilize aggregates by stacking with adjacent ubiquitin chains on adjacent monomers of the protofilaments (35). Additionally, modeling with densities seen in SFs of AD show that ubiquitin may stabilize the asymmetric interaction of these fibrils (35). Ubiquitin and acetylation likely play an important role in understanding how fibrils in vivo are stabilized and resistant to degradation by proteins, such as the 26S proteasome which often requires unstructured segments to capture substrates prior to degradation. Use of protease treatments effectively removes this crucial structural information, so should be used cautiously with in vivo-derived fibrils going forward in order to preserve native structure and conformations. Further investigation of posttranslational modifications in fibrils will be necessary to further clarify their exact role in vivo.

Tau: In Vitro Structure

In vitro tau aggregates are used experimentally to model the aggregation process that happens in vivo. In many cases, polyanions, such as heparin, are used to induce aggregation in vitro. Low-resolution negative-stain EM images have been used to claim that heparin-induced tau looks like that of human brain-derived tau. Cryo-EM of heparin-induced tau showed that heparin-induced 2N4R (2N-terminal inserts with 4 MTBR) tau had at least 4 different conformations and 2N3R had only one conformation in the analyzed samples (Fig. 1D). Only 3 structures of 2N4R were able to be solved, and they were termed “snake,” “twister,” and “jagged” (the structure of another morphology, termed “hose,” was not able to be solved). Notably, the cores of all 4 in vitro structures are different than the ex vivo structures. The in vitro cores are smaller and do not include the R4 repeat or beyond the R4 repeat, where all the known ex vivo structures contain all of R4 and an additional 10–12 residues (6, 28, 31–34). Though there are some shared structural properties at low resolution, near-atomic resolution cryo-EM has definitively shown that heparin-induced in vitro tau aggregates are not similar to ex vivo tau aggregates. There are substantial differences in the core region of the in vitro and ex vivo aggregates that likely affect generalizability of experiments using in vitro tau aggregates to model human disease and neurodegenerative processes.

The PHF6 peptide (306-VQIVYK-311 in 3R repeat) was previously identified as a nucleus of aggregation for tau (29). Crystallographic structures showed that PHF6 peptides pack antiparallel to each other (30); however, the new cryo-EM structures of ex vivo tau aggregates show that antiparallel stacking is not present in any of the 4 physiologic tau aggregates seen thus far. Additionally, synthetic fragments have been used to model tau aggregation. However, shorter fragments, such as the K18 and K19 fragments, stop at Lys372 and thus do not contain the entire region that comprises the core of ex vivo tau aggregate structures (41, 42). The original PHF6 peptide was identified with a K19 fragment and heparin (29), which may indicate that the PHF6 nucleus of aggregation is not relevant for in vivo tau aggregation. Interestingly, the PHF6 peptide was recently showed to have acetylation and ubiquitination, which are not present in in vitro experiments but as previously mentioned may have substantial effect on affinity to aggregate (43). Redefining the nucleus of aggregation in a physiologically relevant way may lead to more effective treatment options for tauopathies in the form of tau aggregation inhibitors. Clinical trials have fallen short so far in showing benefit in humans with tau aggregation inhibitors that showed promise in mice (44), but redefining the nucleus of aggregation may open more options to be explored. Additionally, using a fragment for in vitro experimentation that encompasses the full span of the ex vivo tau cores, such as K11 or K12 (45, 46), may yield more clinically relevant results. However, going forward, the cores made by tau fragments should be compared structurally to ex vivo structures to ensure disease relevance.

The structures of aggregates found in tauopathies have opened the door for deeper understanding of pathogenesis and potentially development of new treatments for a large class of debilitating diseases. However, many forms of tau aggregation in tauopathies have yet to be described, such as other 4R tauopathies (PSP, GGT, AGD) and TSAs from CTE. Also, for example, SFs have been found in PSP, but the question remains if they are similar to those found in AD (47). The diversity of folds already seen in a small subset of tauopathies would suggest that the field has only a glimpse into the clinically relevant conformations that tau can take in aggregation (48).

Aβ: Heterogeneity and Structure

Though necessary for the diagnosis of pathologic AD, Aβ has not yet had the same breakthroughs as tau when it comes to ex vivo structures. There have been numerous structures of in vitro-derived Aβ solved through a variety of techniques in a variety of conditions. These structures have shown that Aβ creates widely heterogeneous fibrils whose specific structures depend on the Aβ peptide chosen and the condition used (8–14, 49–54) (Fig. 2D). Despite efforts to solve structures of in vitro-derived Aβ seeded from brain-derived aggregates (50), to date only one structure of ex vivo Aβ(1–40) has been solved (Fig. 2B) (7), which largely differs from the in vitro-derived structures (Fig. 2A, D).

FIGURE 2.

Ex vivo and in vitro Aβ fibril structures that have been solved by cryo-electron microscopy (cryo-EM) and solid-state nuclear magnetic resonance spectroscopy (ssNMR). (A) The 3-fold symmetry structure from “patient 1” ex vivo-seeded fibril. These fibrils were formed by seeded aggregation with recombinant amyloid-β (Aβ) and ex vivo fibrils (PDB: 2M4J). (B) The atomic structure of Morphology I that is currently the only Aβ structure solved of ex vivo fibrils. The density of Morphology II and III are shown with the atomic structure of Morphology I fit into the density, as Morphology II and III are thought to be multiples of Morphology I (PDB: 6SHS, EMDB: 4864, 4866). (C) The right-handedness of the only ex vivo fibril and an example of left-handedness of an in vitro fibril. The ex vivo fibril is the only structure that has been shown to be right-handed, as all of the in vitro structures are left-handed (PDB: 6SHS, 5OQV). (D) The variety of fibril structures of Aβ(1–40) and Aβ(1–42) in vitro fibrils and how their structures are different than that of Morphology I (PDB: 2LMN, 2LMP, 2MVX, 2MPZ, 5AEF, 5KK3, 2NAO, 5OQV).

Aβ is a proteolytic product of amyloid precursor protein (APP). The function of APP has yet to be fully clarified and remains elusive, but it is thought that APP plays a role in the regulation and plasticity of CNS synapses (55). APP itself has been implicated by various mutants and multiplications in this gene which lead to early onset AD (56, 57). After Aβ is formed from APP in a proteolytic pathway, Aβ forms parenchymal plaques and vascular amyloid deposits in the brain (57). The quantity of brain amyloid has been shown to correlate with progression from mild cognitive impairment to AD (58, 59). Aβ(1–40) and Aβ(1–42) are 2 fragments of APP that have been heavily characterized as disease-relevant (57). Aβ(1–42) has been described as more pathogenic than Aβ(1–40) in AD (60, 61). Aβ(1–40) is thought to be the primary cause of vessel wall damage specifically (62), while Aβ(1–42) is thought to form toxic fibrillization intermediates that may lead to its increased pathogenicity (60, 61). Like tau in the tauopathies, structural variation in Aβ may correlate with AD phenotype (63–67). Some of this structural variation may result from different Aβ fragments with N- and C-terminal variations, such as Aβ(1–40), Aβ(1–42), Aβ(pyruoglutamate-modified 3-40/42). However, some of the structural variation has also been shown to occur with the same fragment type, as Aβ(1–40) aggregates with different structures have shown to exhibit varying toxicity to primary neuronal cells (63). Thus, understanding the structure of Aβ aggregates in vivo will likely provide insight into the pathogenic mechanisms of AD and potentially allow for targeted therapies or improved models of human disease.

Aβ: ssNMR Structures

The first brain-derived high resolution structure of an Aβ aggregate was determined using ssNMR (50) (Fig. 2A). This structure relies on seeded aggregation from brain-derived aggregates rather than the direct structural determination of the brain-derived aggregates themselves. The authors used the brain-derived aggregates in combination with recombinant Aβ(1–40) to generate fibrils in large enough quantity to perform ssNMR (50). It is worth noting that a recent paper focused on α-syn used a similar approach for MSA that yielded a diverse array of fibrils that do not recapitulate the ex vivo seeds in fibril diversity or structure (discussed further below) (68). For the case of Aβ, the patient from whom the seeding aggregates were derived, “patient 1,” did not have a classic AD history. The seeded fibrils had a single predominant structure that had 3-fold symmetry and incorporated the entire Aβ(1–40) peptide (Fig. 2A). They attempted to analyze aggregates from a second patient, “patient 2,” who had a more typical AD clinical history than “patient 1,” however they were unable to create an atomic model. Fibrils from “patient 2” also had a single predominant fibril type with 3-fold symmetry, but this fibril had a distinct backbone conformation and residue interaction from the fibrils derived from “patient 1” (50).

Subsequently, samples were analyzed from 3 types of AD: rapidly progressive AD, posterior cortical atrophy variant AD (PCA-AD), and typical, prolonged duration AD (t-AD). Similar to previous data, PCA-AD and t-AD had a single abundant Aβ(1–40) fibril that accounted for ∼80% of all fibrils (69). Notably, though, the structures seen in the PCA-AD and t-AD cohort did not match the predominant fibril from “patient 1” reported in 2013, indicating that “patient 1” was likely not a representative case (50, 69). Additionally, rapidly progressive AD showed a heterogeneous mixture of Aβ(1–40) fibrils that did not have a predominant fibril type. This structural analysis shows that clinical presentation may in fact be influenced by Aβ fibril structures in some situations in AD. However, as evidenced by a single structure seen in PCA-AD and t-AD cases, not all clinical variants can be traced solely back to Aβ fibril structure (69). Tau fibril structure and other nonaggregate factors may also play a role in disease progression. Also, Aβ(1–42) was not found to have a single predominant fibril type in any group, with each group having at least 2 highly prevalent fibril types for Aβ(1–42) (69), indicating that Aβ(1–42) has more variability in fibril type than Aβ(1–40). Perhaps Aβ(1–42) also plays a role in disease progression in conjunction with Aβ(1–40). Further structural characterization will need to be done to elicit how these other polymorphs compare to the structures reported in Lu et al 2013. Cryo-EM may play an important role in future structural determination due to its capacity to overcome the issues that have troubled previous attempts at in vivo fibril structure determination. At this point, however, evidence would suggest that Aβ fibrils are more diverse in AD than tau fibrils, which may be an obstacle for solving structures by cryo-EM and for elucidating which fibrils are most important for clinical presentation. The effect of Aβ fibril structure and its relation to tau structure will be an area of future exploration.

Aβ: Cryo-EM Structures

Although ssNMR provided the original breakthroughs in determining structures closer to being physiologically relevant, cryo-EM has recently provided a structure of directly imaged brain-derived Aβ aggregates (7) (Fig. 2B). The samples were taken from the meninges of AD brains that shared pathologic features suggesting vascular deposits of Aβ. At least 77% of fibrils had a conserved peptide fold (7). “Morphology I” showed 2 Aβ(1–40) molecules per cross-section that have solvent exposed N and C termini that allow other APP fragment types to fit into the structure (Fig. 2B). “Morphology I” has a relative 2-fold symmetry, distinct from the 3-fold symmetry seen by ssNMR (7, 50) (Fig. 2A, B). “Morphology II” and “Morphology III” have 4 and 6 Aβ(1–40) molecules respectively in cross-section that at low resolution seem to coincide with multiples of “Morphology I” (Fig. 2B). The other morphologies that make up 23% of the fibrils were unable to be characterized (7), thus the structures may not be representative of all Aβ packing in the vasculature of all AD patients. One of the most notable structural elements of the morphologies that were able to be characterized is the handedness of the β-sheet twist. The ex vivo structure shows a right-handed twist, as opposed to the left-handed twist that has been consistently shown in in vitro structures (11, 14, 54, 70–73) (Fig. 2C). As was also seen with tau, none of the Aβ interfaces of in vitro aggregate structures (Fig. 2A, D) match that of the ex vivo structure (Fig. 2B). Thus the in vitro aggregates likely do not completely recapitulate structural aspects of Aβ aggregation in human disease.

Aβ deposits in the meninges have low Aβ(1–42) content (62) and are morphologically distinct from parenchymal amyloid plaques, thus may not be representative of the atomic structure of pathologic Aβ plaques. There have been no ex vivo derived atomic structures of Aβ(1–42) fibrils as the fibrils from Kollmer et al do not contain Aβ(1–42) even though they hypothesize the fold could incorporate Aβ(1–42). As a major component of amyloid plaques, the structure of Aβ(1–42) aggregates are important for a comprehensive understanding of Aβ aggregates mechanism in disease processes.

The structure obtained by ssNMR was seeded by brain-derived aggregates and thus likely reflects both ex vivo and in vitro aggregation processes (50). Though they do not match the cryo-EM structure of ex vivo Aβ, the ssNMR structures may be more similar to an in vivo aggregate than the purely in vitro-derived structures. However, this was only partially the case for MSA α-syn fibrils (68). Additionally, the “patient 1” Aβ aggregates used for seeded aggregation came from a patient with an atypical clinical history and deviate from the predominant structural polymorph seen in further studies of Aβ from AD brains (50, 69). As was seen with tau and Aβ, clinical history appears to correlate with aggregate structure, thus the structure determined by ssNMR (50) may be representative of an in vivo aggregation distinct from that seen in cryo-EM (7). Although there is some evidence that in vitro seeds produce repeatedly homogeneous fibrils by gross structure (63, 74), there has been no structural data to support the fibrils retaining the exact structure of the in vivo aggregate after seeding. In fact, α-syn data would refute this (68). Future structural studies will need to be done to determine if in vitro seeded aggregates retain high fidelity to in vivo-derived seeds for different fibrils in different conditions. If in vitro aggregates seeded from brain-derived fibrils do retain in vivo aggregate structure, it would make it easier to conduct experiments relevant to neurodegenerative disease since aggregates are now a pseudo-renewable resource in labs. In contrast, using only fibrils purified from brains is expensive, time-consuming, and depends on scarce source material coming from donated tissue.

Recently the field has made substantial strides toward understanding the structure of Aβ aggregates in vivo, although less progress has been made here than with tau fibrils. Inhibitors have been made which demonstrate that affecting the C-terminus of Aβ can disrupt aggregation while N-terminus inhibitors are relatively ineffective (75, 76). These inhibitors, developed using in vitro Aβ aggregates, may not have the same effect on the in vivo aggregation process involved in AD, which is known to involve a different structure and opposite helical handedness than in vitro aggregates. Continued structural studies of in vivo-derived seeds and ex vivo aggregates will allow for drug development and improved disease models that can effectively target and recapitulate disease-relevant Aβ aggregates that are thus far distinct from in vitro aggregates.

α-Syn: Heterogeneity and Structure

Synucleinopathies are defined pathologically by the presence of α-syn aggregates. Lewy body diseases consist of synucleinopathies with pathologic Lewy bodies and Lewy neurites which clinically manifest as PD, Parkinson disease dementia (PDD), or dementia with Lewy bodies (DLB). α-Syn aggregates in the form of glial cytoplasmic inclusions (GCIs) are characteristic of MSA (77–80). It has been shown previously that fibrils of α-syn exist in PD (77, 78, 81) and are expected to exist in other synucleinopathies (82–84). α-Syn fibrils formed in vitro also induce α-syn inclusions when injected into model animals (85–87).

α-Syn is a 140 amino acid protein encoded by the gene synuclein alpha (SNCA, Fig. 3A) whose normal function is thought to be related to synaptic vesicle transmission (79). Residues 1–60 compose a lysine-rich N-terminal region with KTK lipid-binding repeats for vesicle binding (88–90). Familial SNCA gene mutations are found in this N-terminal region, including: A30P (91), A30G (92), E46K (93), G51D (94), A53E (95), A53V (96), A53T (97). Residues 61–95 comprise the non-amyloid β component (NAC) region that has been shown as essential for aggregation. Small peptides derived from this region were easily able to aggregate (98, 99). Furthermore, β-synuclein, which is similar to α-syn but lacks amino acids 71–82 in the NAC, has not been found to aggregate like α-syn (81, 100). Removing this region from α-syn also prevents aggregation in vitro (100) further supporting the importance of this region in aggregation.

FIGURE 3.

Ex vivo and in vitro α-synuclein (α-syn) fibril structures and representative α-syn histology. (A) The domains of full-length α-syn, showing the N-terminal domain (red), non-Aβ component (NAC) (blue), and C-terminal tail (pink). Posttranslational modifications, specifically ubiquitination and phosphorylation, that may play a role in α-syn aggregation and disaggregation are labeled (above the domains diagram). Disease-related familial mutations of α-syn are also indicated (below the domains diagram). (B) The image shows a pigmented substantia nigra neuron with a Lewy body (left) adjacent to a normal neuron. Extracellular pigment is also present (right) indicative of neurodegeneration. Lewy bodies are found in a variety of neurodegenerative diseases including dementia with Lewy bodies and Parkinson disease (PD). The right image shows glial cytoplasmic inclusions that are found in multiple system atrophy (MSA). Both Lewy bodies and glial cytoplasmic inclusions contain α-syn. (C) Type I MSA fibril that is made of protofilament IA (light blue) and protofilament IB (dark blue) with its electron density map overlaid. The positively charged cavity made-up of K43, K45, and H50 from both protofilaments (yellow) likely contains a negatively charged nonproteinaceous material. K80 and K60 are both sites of ubiquitination, with K60 only being solvent exposed in IA, not IB. T72 is a site of phosphorylation that is buried in both IA and IB (PDB: 6XYO, EMD: 10650). (D) Type II MSA fibrils with electron density map from Type II₁ fibrils (fibrils with protofilament IIB₁ as opposed to protofilament IIB₂). Type II fibrils are each made of an IIA (light purple) and IIB (dark purple) protofilament. Protofilament IIB₁ and IIB₂ are overlaid and labeled as protofilament IIB. The fibrils have a positively charged cavity made by K43, K45, and H50 of both protofilaments that likely contains a negatively charged nonproteinaceous material. K80 and K60 are both sites of ubiquitnations, with K60 only being solvent exposed in protofilament IIA, not IIB. T72 is a site of phosphorylation that is found in a cavity in protofilament IIA, as opposed to being buried in protofilament IA (PDB: 6XYP, 6XYQ, EMD: 10651). (E) The K80-A91 region of Type IIB₁ and Type IIB₂ protofilaments that has a slightly different conformation. T81 phosphorylation may be what drives protofilaments to take either the IIB₁ or IIB₂ conformation as it is only solvent exposed in IIB₂. (F) The variety of in vitro α-syn fibril structures that mostly differ from the ex vivo MSA fibrils.

Clinically, PD is the second most common neurodegenerative disorder affecting 2%–3% individuals 65 and older, while DLB is thought to be the underlying cause of 10%–15% of all cases of dementia. PD is characterized by neuronal loss in the substantia nigra that causes striatal dopamine deficiency and leads to bradykinesia and other motor symptoms (101, 102). PD can, however, progress to PDD, and DLB can progress to motor dysfunction similar to that seen in PD. PD, PDD, and DLB are generally considered to be on the same disease spectrum with similarities clinically and pathologically including α-syn aggregates in the form of Lewy bodies and Lewy neurites (Fig. 3B). MSA is distinct clinically and pathologically from PD and DLB, with a lower prevalence in the general population. MSA is of sporadic onset and is characterized by parkinsonism, cerebellar ataxia, and/or autonomic failure (103–107). Similar to PD and DLB there are neuronal α-syn inclusions and neuronal cell loss in MSA, however, α-syn inclusions are more prevalent in oligodendrocytes as GCIs (Fig. 3B) (82, 108–113).

As is seen with tau fibrils, the structure of α-syn fibrils varies among disease states (114–119). It was also shown that GCIs from MSA are ∼10³-times more potent seeds of aggregation than fibrils derived from PD Lewy bodies (120). Recent characterization of amplified patient-derived fibrils suggests that PD and MSA fibrils share many characteristics, but MSA fibrils are more potent inducers of motor deficits, neurodegeneration, and α-syn pathology. Although DLB and PD are thought to be on the same disease spectrum, DLB fibrils did not yield the significant neuropathology seen with PD fibrils. Though still recent, these results may suggest that PD fibrils and DLB fibrils contained in Lewy bodies may be less similar than previously thought (4).

α-Syn: Near-Atomic Structure

There have been several structures of in vitro formed α-syn fibrils by cryo-EM and ssNMR (121–126). However, until recently, there had been no structures of α-syn derived from human brains, due to limited ability to purify large enough quantities (127). Schweighauser et al were the first to report a structure of ex vivo α-syn using cryo-EM, for which they used samples from 5 cases of pathologically confirmed MSA (5). Two filaments that looked identical visually on micrographs were actually different structurally when investigated computationally (Fig. 3C, D). Within the 2 filament types, there were 4 distinct protofilaments (IA interacts with IB, IIA interacts with IIB) (5). An increased ratio of the “Type I” to “Type II” filaments in the putamen was correlated with disease duration of the 5 cases. However, this correlation will need further investigation as some cases showed different ratios in other parts of the brain (5).

Both Type I and Type II filaments have a cavity, bound by K43, K45, and H50 on both A and B protofilaments, that contains nonproteinaceous density. This density is likely negatively charged due to a + 4 charge per level of the cavity. Type I filaments have a larger cavity containing more densities than the cavity of Type II filaments. There are also unconnected densities next to external lysines that have not been identified. Similar densities were seen in tau structures and may be negatively charged molecules of the fuzzy coat (5, 28, 31–33). The nonproteinaceous densities may suggest that fibrils are sequestering necessary molecules from other parts of the cell, leading to cell toxicity and death. Additionally, the nonproteinaceous densities may be key in maintaining fibril stability and thus could be explored for purposes of disaggregation and learning how fibrils grow and maintain their structure in vivo.

In vitro fibrils show high variability in protofilament interaction, even within classes of polymorphs (Fig. 3F). Some of the in vitro structures also contain the 3-layered L-motif seen in the ex vivo fibrils. These fibrils were notably assembled using negative polyanions that may stabilize the positively charged cavities, allowing α-syn fibrils to maintain their structure and elongate (121–124, 128). Similar to what was seen with tau and Aβ, no in vitro protofilament is a perfect match for the ex vivo α-syn structure, but the PDB-6PEO structure is the closest to protofilament IIB₂ with only minor differences (PDB-6A6B and PDB-6CU7 are similar to PDB-6PEO). Using very similar in vitro fibrils would likely yield the most disease-relevant results when ex vivo fibrils are unrealistic or unavailable. However, the structural differences should be noted as a caveat for complete generalizability to disease.

To determine if using ex vivo fibrils for seeded aggregation could generate in vitro fibrils similar to the ex vivo seeds, a subset of the same cases from that were used to determine the ex vivo MSA fibril structure were used for seeded aggregation (68). Disappointingly, none of the daughter fibrils fully recapitulated the ex vivo seeds. Also, notably, no structure derived from seeded aggregation yielded a fibril containing nonproteinaceous density that is seen in both Type I and Type II filaments (Fig. 3C, D), likely indicating substances present in cells that were not present in vitro have a role in the way α-syn folds into fibrils. Two cases that had a mixture of Type I and Type II filaments yielded a mixture of fibrils that look very different from the parent seeds (68). However, one case that only had Type II filaments yielded a fibril that contained a single protofilament that is quite similar to Type IIB protofilaments and a protofilament formed in vitro using recombinant α-syn (5, 68, 124). Since this case only had one fibril type present, this may have led to a more faithful recapitulation of the parent seed. Future work will need to be done to investigate the factors that lead to faithful propagation of a parent seed both in vitro and in vivo. At this point though, it seems that in vitro fibrils seeded with ex vivo fibrils do not recapitulate the structure of the ex vivo fibrils.

α-Syn: Disease-Associated Mutations

Familial SNCA mutations disrupt the stabilizing interactions of the known wild-type α-syn in vitro fold conformations. It has been shown that under the same conditions A53T, A53E, G51D, and E46K mutants of α-syn form fibrils with distinct morphologies from wild-type fibrils (123). The E46K mutant α-syn is toxic to neuronal cells (129). These fibrils were less stable and easily fragmented, but were also a better seed than wild-type fibrils, potentially indicating that toxicity is related to seeding rather than fibrils stability (130). These in vitro data suggest that mutations may affect fibril structure, stability, and effect on disease progression.

G51D and A53E mutations can cause mixed MSA and PD pathology in patients (95, 131, 132). G51 lies at the protofilament interface of the ex vivo MSA fibrils near K43, K45, and H50. Structurally, there may be room to accommodate the switch from glycine to aspartic acid; however, this switch would reduce the positive charge of the central cavity (5). It is difficult to speculate how a change in charge or amino acid side chain would affect the central cavity and the in vivo fibrils as a whole. However, it could change the nonproteinaceous compounds within the central cavity or potentially disrupt the fibril stability or even enhance fibril stability to make fibrils more stable and more pathogenic than wild-type α-syn fibrils leading to mixed MSA and PD pathology.

Some of the in vitro fibrils imaged may be able to explain how some of the familial mutations pack in vivo and the ex vivo structures solved may be able to accommodate some of the familial SNCA mutations. However, additional ex vivo α-syn structures of mutant fibrils will need to be solved in order to confirm the actual structure. The structure of fibrils from patients with familial mutations will be important to understanding why those mutations specifically cause disease and potentially lead to breakthroughs in understanding development and progression of sporadic disease as well.

α-Syn: Posttranslational Modifications

Postmortem human tissue and transgenic mouse models of PD and other synucleinopathies show a correlation between posttranslational modifications and Lewy body formation. It has been suggested that these modifications may inhibit fibril formation and so may be occurring after the fibrils have already been formed (133). MSA filaments analyzed for the ex vivo MSA structures showed ubiquitination at K23, K60, and K80 (5). Based on the protofilament structures, K80 ubiquitination is only compatible with protofilament IIA, as it would clash with surrounding densities in the other protofilaments. Additionally, K60 is only solvent exposed in protofilament IA and IIA.

Similar to ubiquitination, phosphorylation may affect fibril type. T72 may be favored for phosphorylation in protofilament IIA over IA in the MSA fibrils because IA has T72 buried while IIA has T72 exposed to a cavity. T81 phosphorylation may be the driving difference between protofilament IIB₁ and IIB₂ as T81 is only solvent exposed in IIB₂. The only difference between the 2 protofilaments is at the T81-A90 region, which may be due to this phosphorylation (5).

C-terminal truncations have also been found in Lewy bodies and have previously been shown to have a better propensity to seed aggregation than full-length α-syn (134–138). Three in vitro structures comparing Lewy body-derived C-terminal truncations were analyzed by cryo-EM (128, 139, 140), demonstrating that the more α-syn was truncated, the more twisted the fibrils became. The structures also showed that the C-terminal truncations allowed for the protofilament cores to pack tighter than longer α-syn (128). The C-terminal truncated α-syn showed greater stability and faster seeding than longer α-syn, indicating that increased stability from tighter packing and increased seeding may account for aggregation in vivo.

Posttranslational modifications may precede aggregation and push aggregates to one type of fibril by limiting stable protofilament conformations. Modifications may also follow aggregation as a way to mark fibrils for break down, reduce their stability, or potentially even increase their stability. The ability or inability to tag fibrils may have ramifications for fibril clearance as some members of the hypothesized fibril clearance pathway require posttranslational modifications for target recognition. Understanding fibril structures can help with understanding how posttranslational modifications affect and interact with disease aggregates. Furthermore, structures can help elucidate if modifications push aggregates to one conformation or another.

Ex vivo structures for α-syn are important to facilitate the creation of rationally designed aggregation inhibitors. Indeed, rationally designed inhibitors based on recombinant α-syn fibrils were less effective against MSA seeded aggregates than recombinant aggregates (15). The recent ex vivo MSA fibrils, though, have opened the door for analysis of disease-relevant α-syn structures. However, MSA is a rare disease compared with other synucleinopathies, PD, PDD and DLB. Schweighauser et al note that they attempted to analyze DLB fibrils but were unable to due to the lack of helical structure compared with MSA fibrils. It has also been suggested that PD and DLB fibrils may be more heterogeneous than MSA fibrils and do not have the same structure (141). Pinning down the in vivo structure of α-syn may prove to be more difficult than tau as thus far diseases within the umbrella of tauopathies have concurred with the “one disease-one strain” hypothesis, while evidence for synucleinopathies points to a more heterogeneous population of fibrils to create a variety of disease phenotypes (5, 31, 32, 141).

TDP-43: Heterogeneity and Structure

TDP-43 was originally found from a screen of proteins that bind to the TAR element of HIV type I, but was later found to be the key aggregating protein in ALS and frontotemporal lobar degeneration with TDP-43 inclusions (FTLD-TDP) (142). As an RNA-binding protein, TDP-43 is involved in RNA regulation including splicing, and mRNA transport and stability (143, 144). The C-terminus of TDP-43 has a low-complexity domain ([LCD], Fig. 4A) that can form filamentous aggregates in vivo (145–147) and in vitro (148, 149). Toxicity of TDP-43 is thought to be, at least in part, related to a toxic loss of function due to TDP-43’s role in RNA metabolism and regulation (143, 144). Though mostly associated with ALS and FTLD-TDP, TDP-43 aggregates have also been found in AD, PD, CTE, Huntington disease, limbic predominant age-related TDP-43 encephalopathy (LATE), and inclusion body myopathies (150–154). Almost all familial disease-causing mutations of TDP-43 are found in the LCD (144). C-terminal fragments, ubiquitinated aggregates, and phosphorylated aggregates of TDP-43 have been found in neuronal inclusions of ALS and FTLD (142). Neuronal inclusions of TDP-43 can even act as seeds for new pathology in vivo (155, 156).

FIGURE 4.

In vitro transactive response element DNA-binding protein of 43 kDa (TDP-43) structures of fragments of the low complexity domain (LCD) and the full TDP-43 protein. (A) The domains of full-length TDP-43 including the N-terminal domain (NTD) (red), nuclear localization signal (NLS) (green), RNA recognition motif 1 (RRM1) (purple), RRM2 (pink), and LCD (blue) with SegA and SegB from the LCD indicated. (B) The core structures of in vitro fibrils that have been solved for SegA and SegB. SegA had 3 distinct fibrils, while SegB had one fibril conformation (PDB: 6N37, 6N3B, 6N3A, 6N3C).

TDP-43 aggregates have a wide array of clinical and pathological presentations, in addition to morphological and biochemical differences. There are at least 5 different histologic subtypes of FTLD-TDP, which have some correlation with genetic mutation status (157). When aggregates derived from human brains are used for seeds, the subtype of FTLD has a profound effect on seeding ability and cell toxicity. Seeding ability and toxicity notably correlate to disease course and likely indicate structural differences in TDP-43 aggregates (155, 158). Different truncations of TDP-43 are found in different regions of the brain and spinal cord in the same disease (159). Furthermore, antibodies to hyperphosphorylated TDP-43 have indicated differences in aggregate conformation with similarly modified TDP-43 proteins (160). The diversity of TDP-43 aggregates is likely one of the bases of the diversity of presentations of TDP-43 disease. It also poses a challenge for ultrastructural studies to elucidate how the TDP-43 aggregates differ, as many aggregates of the same conformation are needed to obtain a high-resolution reconstruction.

TDP-43: Near-Atomic Structure

Several high resolution structures of fibrils made from fragments of in vitro-derived TDP-43 have been investigated using cryo-EM, crystallography, and ssNMR (161–164). While these studies have provided insights into potential interactions between these fragments in in vivo aggregation, there has been limited data on how full-length TDP-43 or large C-terminal fragments fold, and these are the most relevant for human disease. This lack of structures has not been from a lack of effort (164), but rather from challenges posed by the TDP-43 protein and availability of in vivo aggregates.

Recently, atomic resolution structures have been obtained of 4 fibril polymorphs (Fig. 4B) that were derived from segments of the TDP-43 LCD (164). Cao et al note efforts to create aggregations from full-length TDP-43, a C-terminal fragment (amino acids 208–414), and the LCD (amino acids 274–414). These aggregates all failed to create ordered fibrils in vitro, but rather only made fibril clumps or disordered aggregates that could not be reconstructed to high resolution with current techniques (164). Thus, Cao et al chose to use known aggregate cores they refer to as SegA (amino acids 311–360) and SegB A315E (amino acids 286–331). SegA and SegB A315E were chosen because of their importance for aggregation in experiments and their ability to template full-length TDP-43 (148, 156, 164–168).

SegA fibrils showed 3 polymorphs termed SegA-sym, SegA-asym, and SegA-slow (Fig. 4B). The SegA-slow fibril is not compatible with full-length TDP-43 due to C- and N-termini residing inside the core of the fibril. However, SegA-slow may be compatible with TDP-43 fragments that may be more relevant to TDP-43-related disease aggregates than full-length TDP-43 (159). SegA-sym and SegA-asym alternatively can theoretically incorporate full-length TDP-43 in addition to truncated TDP-43, as the C- and N-termini are not confined inside the fibril core. SegB fibrils have a distinct fold from SegA fibrils (Fig. 4B).

Sarkosyl-insoluble fragments from the brain of an individual with FTLD-TDP could seed SegA, but not SegB A315E (164). SegB A315T is found in familial ALS, thus the inability to seed SegB A315E fragments could indicate that sporadic FTLD-TDP has different fibril structure that is not compatible with familial mutations (165). This discrepancy, though, will need to be investigated further to make clear conclusions.

As SegB A315E was not able to be seeded by fragments from human brain derived sarkosyl-insoluble pathologic TDP-43, aggregation processes may be different between TDP-43-related diseases, similar to what has been seen with ex vivo tau fibrils (28, 32–34). It is unknown, however, if individuals with the same TDP-43-related disease have the same structured aggregates or whether they have a mix of different aggregates. Additionally, based on Aβ and tau studies, seeding reactions do not necessarily result in the same fibril structure as in vivo aggregates (6, 7, 50, 169). Therefore, while SegA and SegB have been shown to be nuclei of aggregation, these in vitro structures may be different from the structures found in vivo.

TDP-43 poses unique challenges to both in vitro aggregation of full-length TDP-43 and extraction of ex vivo aggregates for imaging. TDP-43 is found in large protein complexes and can participate in multivalent interactions to assemble large connections of proteins. Cao et al speculate that this may allow large segments of TDP-43 to form unstructured aggregates or clumps (164). Distinct from the other proteins presented in this review, TDP-43 aggregates do not exhibit features of amyloid, especially by thioflavin staining, thus may not form the same structured aggregates (170). Additionally, TDP-43 has been shown to participate in liquid-liquid phase separation that is mediated by its multivalent interactions (171). TDP-43 may have a propensity to form phase separated liquids in the in vitro environment, but may fail to progress to an ordered aggregate, leading to amorphous aggregates. Amorphous aggregates are unable to be averaged to high-resolution by cryo-EM, as cryo-EM relies on many particles in the same conformation that are imaged from different views. Even though tau, Aβ, and α-syn fibrils formed in vitro thus far have not proved to truly reflect in vivo structures found in the human disease, it still remains formally possible that in vitro fibrils of full-length TDP-43 could provide insights into disease-relevant fibrils. TDP-43 also poses the challenge of only being present in small quantities in autopsied brains. The paucity of in vivo fibrils makes high resolution reconstructions by cryo-EM difficult on ex vivo aggregates. For tau, Aβ, and α-syn, in vivo fibrils are a rare resource, but are much more abundant in diseased brains compared with TDP-43 fibrils. With low quantity of fibrils, purification and extraction of the fibrils to high enough purity and quantity to allow for high resolution reconstructions by cryo-EM will be a persistent problem going forward in the struggle to determine TDP-43 fibrils’ structure. The structural determination of TDP-43 aggregates would assist in understanding not only disease initiation and progression but also in developing faithful disease models.

CONCLUSION

The fields of structural biology and neurodegeneration have seen rapid advances in their understanding of disease-associated aggregate structures in the past 4 years, aided by recent enhancements in cryo-EM. Tau fibrils have had the most notable breakthroughs, with structures from 4 different tauopathies, in vitro tau, and multiple cases of sporadic and familial AD resolved to near atomic resolution (6, 28, 31–34). The improvement of cryo-EM has allowed for higher resolution structural determinations of different particles with smaller quantities than was previously possible. These improvements have had notable impacts on neurodegenerative disease aggregates because of their relative scarcity and unique morphologies.

In vitro fibrils have been heavily used as the default in vitro model for studying neurodegenerative aggregates. However, as presented here, ex vivo fibrils structures question the fidelity of previous in vitro fibrils as a disease model and highlight the need to investigate better methods for creating in vitro fibrils. Additionally, animal models and cell models, though useful, have thus far fallen short of faithfully recapitulating human neurodegenerative disease for a number of reasons, likely including that they may not accurately recapitulate in vivo aggregate structure (172).

Finally, studying differences between aggregate structures could provide insight into the clinicopathologic heterogeneity that is observed across all neurodegenerative disease subtypes, shedding light on the initiation and propagation of neurodegenerative diseases’ neuropathologic inclusions to further guide therapeutic interventions to halt, slow, or even reverse disease progression.

 

Translational Neuropathology Research Laboratory, Department of Pathology and Laboratory Medicine, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, Pennsylvania, USA (BCC, EBL).

E.B.L. is supported by NIH R01NS095792, R56AG063344, P01AG066597, P01AG010124, P30AG072979, U54NS115322, and U19AG062418; Y.-W.C. is supported by a David and Lucile Packard Fellowship for Science and Engineering (2019-69645).

The authors have no duality or conflicts of interest to declare.