R. H. Baloh, Department of Neurology, 660 South Euclid Avenue, Box 8111, Saint Louis, MO 63110, USA Fax: +314 362 3752 Tel: +314 362 6981 E-mail: firstname.lastname@example.org
Accumulations of aggregated proteins are a key feature of the pathology of all of the major neurodegenerative diseases. Amyotrophic lateral sclerosis (ALS) was brought into this fold quite recently with the discovery of TDP-43 (TAR DNA binding protein, 43 kDa) inclusions in nearly all ALS cases. In part this discovery was fueled by the recognition of the clinical overlap between ALS and frontotemporal lobar degeneration, where ubiquitinated TDP-43 inclusions were first identified. Later the identification of TDP-43 mutations in rare familial forms of ALS confirmed that altered TDP-43 function can be a primary cause of the disease. However, the simple concept that TDP-43 is an aggregation-prone protein that forms toxic inclusions capable of promoting neurodegeneration has not been upheld by initial investigations. This review discusses observations from human pathology, cell culture and animal model systems, to highlight our somewhat murky understanding of the relationship between TDP-43 aggregation and neurodegeneration.
Aside from neuronal cell loss and reactive gliosis, aberrant protein aggregation is the key feature of the pathology of most neurodegenerative diseases, including Alzheimer’s disease, Parkinson’s disease, Huntington’s disease and many others [1,2]. The importance of protein aggregation in neurodegenerative diseases is indicated by the presence of frequent large extracellular or intracellular inclusions, believed to be abnormal accumulations of misfolded proteins that cannot be degraded by normal mechanisms. Although the ubiquitous presence of protein inclusions in neurodegenerative diseases suggests they are central to pathophysiology, there is no universal agreement as to what role they play. Inclusions themselves have variously been theorized to be (a) the primary toxic species leading to neuronal dysfunction and death; or (b) epiphenomena which correlate with another toxic event but are themselves unimportant; or (c) structures which protect cells from harm by sequestering abnormally misfolded toxic proteins. Their presence alone on pathology is consistent with any of these possibilities.
Amyotrophic lateral sclerosis (ALS) is an adult onset neurodegenerative disease predominantly of the motor system, i.e. spinal motor neurons and upper motor neurons of the primary motor cortex. This leads to muscle wasting (amyotrophy) and degeneration of cortical primary motor neurons and their descending axons in the lateral columns of the spinal cord (lateral sclerosis). A subset of patients with ALS also develop frontal lobe dysfunction or frontotemporal lobar degeneration (FTLD), indicating that other non-motor cortical structures may also be involved and suggesting that ALS and FTLD may share a common underlying pathophysiology . Although it was recognized in 1991 that ubiquitin-positive inclusions are frequent in motor neurons in ALS patients on autopsy , the importance of these structures and their component proteins remained unknown for the next 15 years. Then in 2006 TDP-43 (TAR DNA binding protein, 43 kDa) was identified as a key component of insoluble ubiquitinated inclusions in both ALS and FTLD, using a mass spectrometry based approach [5,6]. Shortly afterward dominantly inherited point mutations in TDP-43 were identified as a cause of familial ALS (and less commonly FTLD), solidifying that altered TDP-43 function can directly initiate neurodegeneration [7–11].
This breakthrough discovery brought protein aggregation to the forefront in ALS research, similar to its place in other fields of neurodegeneration, and with it the same incipient questions are raised. What is the importance of TDP-43 inclusions? Are they a primary neurotoxic species, innocent bystanders, or part of a normal cellular protective response? This review will discuss findings related to TDP-43 aggregation in human pathology and in cell culture and animal model systems, to begin to synthesize answers currently available to these questions.
TDP-43 aggregation in human disease
Initial reports of TDP-43 inclusions in ALS and FTLD tissue delineated several key characteristics, and these have largely driven the various hypotheses as to the role of TDP-43 in disease (Fig. 1). TDP-43 is a multifunctional DNA and RNA binding protein, which shuttles between the nucleus and cytoplasm to mediate multiple aspects of RNA metabolism [12,13]. Whereas in normal cells the majority of TDP-43 resides in the nucleus, TDP-43-positive inclusions were most commonly observed in the cytoplasm or dystrophic neurites, with nuclear inclusions occasionally seen. In neurons with cytoplasmic TDP-43 inclusions, nuclear TDP-43 was frequently diminished or absent, suggesting that there may be a loss of nuclear TDP-43 function in these cells. Lastly, biochemical fractionation indicated that TDP-43 was cleaved into C-terminal fragments in the disease state, and became phosphorylated . Ultrastructure of TDP-43 inclusions did not show fibrils and did not stain with congophilic dyes, indicating that TDP-43 inclusions do not have characteristics of amyloid . Some constellation of these histopathology findings and biochemical abnormalities are loosely referred to in the literature as TDP-43 ‘pathology’. Based on the regional distribution of TDP-43 abnormalities in the neocortex, and the appearance of dystrophic neurites, a subclassification scheme was also suggested for different types of FTLD–TDP (FTLD with TDP-43 inclusions) . Correlation of this scheme to different clinical syndromes in the ALS/FTLD spectrum has thus far shown mixed results [16–18].
Although it was initially suspected that TDP-43 pathology may be specific for ALS and a subset of FTLD cases (now referred to as FTLD–TDP), it was quickly found that abnormal TDP-43 staining was present in many neurodegenerative diseases, including Alzheimer’s disease, Parkinson’s disease, diffuse Lewy body disease and many others (Table 1). Recently, abnormal TDP-43 staining was also observed in the brains of 29% of cognitively normal control subjects over age 65 . In all of these cases, the same histology characteristics are seen, i.e. neuronal and glial cytoplasmic inclusions of TDP-43 with nuclear clearing, dystrophic neurites and infrequent intranuclear inclusions. However, the distribution of TDP-43 pathology appears to be distinct in these different conditions. While ALS and FTLD–TDP brains have widespread abnormalities in most regions of the brain including neocortex, abnormal TDP-43 staining is usually limited to the limbic system in other neurodegenerative diseases, and in normal older subjects (references in Table 1). These findings indicate that although the presence or absence of TDP-43 pathology is not specific for a particular disease, the regional distribution of abnormal TDP-43 staining may correlate with particular disease entities.
Table 1. TDP-43 pathology in human disease. IBMPFD, inclusion body myopathy, Paget’s disease and frontotemporal dementia; IBM, inclusion body myositis; NCI, neuronal cytoplasmic inclusions; GCI, glial cytoplasmic inclusions; NII, neuronal intranuclear inclusions; DN, dystrophic neuritis; CI, cytoplasmic inclusions. The relevance of the location of the TDP-43 inclusions (cytoplasmic versus nuclear, glial versus neuronal) is unclear at present.
a Seen in all sporadic ALS cases, and non-SOD1 familial cases. b Distinct subtypes may exist in different forms of FTLD–TDP. c By definition. If all FTLD with ubiquitinated inclusions are included, TDP-43 pathology is observed in 80–94%.
NCI and GCI ≫ NII DN, nuclear clearing
Widespread: spinal motor neurons, frontal and temporal neocortex, basal ganglia, limbic structures
Similar abnormal TDP-43 staining is also observed in both inclusion body myopathies and myofibrillar myopathies, muscle diseases where cytoplasmic accumulation of misfolded proteins is the characteristic feature [20–22]. The presence of TDP-43 staining in myofibrillar myopathies is particularly interesting, as these are due to mutations in various components of the myofibrillar apparatus causing them to misfold and aggregate. The fact that TDP-43 pathology is observed in these conditions supports that translocation of TDP-43 from the nucleus and cytoplasmic aggregation can be induced simply as a response to misfolded protein stress in the cytosol.
Taken together the human pathology studies indicate that (a) cytoplasmic aggregation and nuclear clearing of TDP-43 are a consistent pattern across the spectrum of diseases where TDP-43 has been investigated, from brain to muscle; (b) TDP-43 pathology is not specific to ALS or FTLD and is observed in most neurodegenerative diseases and frequently in cognitively normal individuals over age 65; (c) the regional distribution of TDP-43 pathology varies across different neurodegenerative diseases, with a more widespread pattern seen ALS/FTLD, and a limbic predominant pattern in most other conditions. Although TDP-43 inclusion pathology is not specific for a particular disease entity, the fact remains that mutations in TDP-43 cause ALS and rarely FTLD, but not other neurodegenerative diseases. This supports that there is a unique connection between TDP-43 dysfunction and ALS/FTLD, but the nature of this connection remains to be clarified. Although abnormal TDP-43 staining is non-specific, this does not mean it is unimportant in diseases other than ALS/FTLD, and instead may support that altering TDP-43 aggregation or function could be a valid therapeutic target across many neurodegenerative disorders.
The structure of TDP-43 and its relationship to protein aggregation
TDP-43 is structurally similar to heterogeneous ribonucleoprotein (hnRNP) A/B family members in that it contains two RNA recognition motifs and a C-terminal domain rich in glycine residues  (Fig. 1). It was first identified as a protein capable of binding the TAR DNA (but not RNA) sequence of HIV1 to suppress transcription of the long terminal repeat . Later it was identified as a factor that binds to an intronic (UG) repeat element in the cystic fibrosis transmembrane receptor to modulate alternative splicing . The C-terminal region of TDP-43 is necessary for protein–protein interactions with other hnRNP proteins, and for mediating alternative splicing and transcriptional suppression [26–28]. Within the C-terminal region there is a particularly glycine-rich domain, and the neighboring region is also rich in glutamine (Q) and asparagine (N) residues [24,29]. Recent evidence supports that the C-terminal domain, in addition to mediating protein–protein interactions with other splicing factors, has properties of a Q/N-rich prion domain [29–31]. Despite the name, prion domains are not structurally related to the mammalian prion protein, but rather to yeast prions. Yeast prions contain Q/N-rich domains which can adopt an alternative conformation that recruits the native form of the protein into an inactive aggregate [32,33]. Although the degree to which TDP-43 has properties similar to other prion domain containing proteins remains to be explored, it is clear from numerous studies that the C-terminal region is critical for TDP-43 aggregation (see below). Furthermore, it is important to note that this region is the location of all but one of the ALS/FTLD associated mutations in TDP-43 identified to date, emphasizing that altering the function of this domain in some way probably contributes to neurodegeneration (Fig. 1).
TDP-43 aggregation and toxicity in model systems
Purified protein in vitro
Purified TDP-43 synthesized in bacteria is highly prone to aggregation and rapidly falls out of solution over time, supporting that the protein is intrinsically aggregation prone . The C-terminal region was required for this aggregation tendency, as a fragment of the N-terminus containing only the RNA recognition motif (RRM) domains remained soluble. Electron microscopy showed that purified full-length TDP-43 or C-terminal fragments formed amorphous aggregates and did not have properties of amyloid such as thioflavin-S binding. This is akin to the properties of inclusions in ALS/FTLD patient tissues, which likewise appear amorphous and non-amyloid. A second study of purified protein supported that the C-terminus of TDP-43 is critical for aggregation, and found that one particularly aggregation-prone subregion could also form amyloid fibrils . Finally, TDP-43 point mutations linked to ALS increased the propensity of purified TDP-43 to aggregate, supporting that increased tendency to aggregate may be an important property of ALS associated TDP-43 mutations .
Yeast models of TDP-43 aggregation and toxicity
Yeast provides an easily manipulated system which can be used to model properties of neurodegenerative disease proteins. Similar to findings with purified protein, TDP-43 is highly aggregation prone when overexpressed in yeast, and this property is dependent on the presence of the C-terminal domain [34,36]. Additionally, overexpression of TDP-43 is toxic to yeast, similar to what is seen in mammalian cells and animal models (see below). The yeast system was used nicely to show that the protein Pbp1 (an orthologue of ataxin-2), a polyglutamine containing RNA binding protein, modulates TDP-43 overexpression toxicity . This was recapitulated in a fly model, and furthermore screening of human patients supported that mid-range length expansions of the polyglutamine stretch in ataxin-2 were a risk factor for developing ALS . It remains to be determined whether overexpression of Pbp1 enhanced TDP-43 toxicity by promoting its aggregation, or alternatively by altering other aspects of RNA metabolism which are disrupted by TDP-43 overexpression.
Mammalian tissue culture models of TDP-43 aggregation and toxicity
In contrast to the strong tendency of exogenously expressed TDP-43 to form cytoplasmic aggregates in yeast, TDP-43 in cultured mammalian cells (endogenous or overexpressed) remains predominantly soluble, and in the nucleus [26,38,39]. The reason for this difference is unclear, but perhaps because yeast does not contain a TDP-43 orthologue the appropriate binding partners (either RNA or protein) are not present to help maintain TDP-43 in a soluble state. Several different manipulations are capable of promoting the formation of detergent-insoluble TDP-43 aggregates in cultured mammalian cells. These include mutating the nuclear localization signal , mutating the RNA binding domain , or expressing truncation mutants containing only the C-terminal domain (which are missing both the RNA binding motif and the nuclear localization signal) [40–42]. This suggests that altered TDP-43 RNA binding, nuclear localization or proteolytic cleavage can promote TDP-43 aggregation. While these are important insights, it has been hard to reconcile them with the fact that none of these properties is altered by ALS/FTLD associated mutations in TDP-43, making their relevance to TDP-43 inclusion formation in human disease tissue unclear.
As in yeast, overexpression of TDP-43 is toxic and promotes cell death in cultured mammalian cells or primary neurons [29,41,43–45]. In most of these studies full-length TDP-43 remained soluble and in the nucleus, which indicates that aggregation and inclusion formation are not the only mechanism by which overexpression of TDP-43 can disrupt cellular function. Indeed, recent evidence supports that overexpression of exogenous TDP-43 promotes the degradation of endogenous TDP-43 mRNA through a 3′UTR binding site in an autoregulatory loop [46,47]. Given that TDP-43 binding sites are present in many mRNA species , it is likely that overexpression of TDP-43 directly alters the splicing or stability of a large number of RNA targets, in addition to its own message. Therefore it appears that the toxicity of TDP-43 overexpression in mammalian cells can be induced either by the formation of insoluble TDP-43 in the cytosol and proteotoxic stress [41,43], or by direct alteration in splicing and stability of mRNA targets . Which of these is more prominent, or relevant to human disease, is unknown.
Initial studies of cytoplasmic TDP-43 aggregates focused on the assumption that they represent inclusion bodies, i.e. abnormal accumulations of misfolded ubiquitinated TDP-43 that could not be properly degraded by the cell. However, those familiar with RNA binding proteins recognized that translocation from the nucleus and association with cytoplasmic RNA granules (stress granules or processing bodies) is common for these proteins . In several cell lines, TDP-43 was sometimes incorporated into cytoplasmic stress granules, either from oxidative stress, proteasome inhibition, heat shock or osmotic stress [50–53]. Stress granules are dynamic cytoplasmic bodies containing numerous mRNAs and RNA binding proteins that are involved in translational suppression during a broad range of cellular stressors mentioned above . It is important to note that, unlike the core stress granule proteins (TIA-1, TIAR), TDP-43 itself does not induce stress granules, and knockdown of TDP-43 does not alter their formation, suggesting stress granule organization is not a primary function of TDP-43 . Despite these issues, the concept that cytoplasmic TDP-43 aggregates are actually functional ribonucleoprotein complexes has important implications for TDP-43 pathology in human disease . This would provide a potential explanation for why cytoplasmic TDP-43 aggregates are observed in a wide variety of neurodegenerative diseases (and myodegenerative diseases), as cells in these cases are under proteotoxic stress due to the presence of other misfolded proteins and proteosome dysfunction. Therefore TDP-43 aggregation in these diseases may simply reflect a normal tendency of TDP-43 to incorporate into cytoplasmic RNA granules. As two studies examining whether TDP-43 aggregates in ALS/FTLD patient tissue stain with other markers of stress granules are currently at odds, further work is needed to clarify this possibility [50,53].
It is also important to consider that cytoplasmic RNA granules in neurons, both in axons and dendrites, may have unique properties compared with those in non-neuronal cells . Indeed, TDP-43 was found to be partially localized to dendrites of cultured hippocampal neurons, and upon activity stimulation this localization was enhanced . In contrast to non-neuronal cells, TDP-43 colocalized with markers of P-bodies rather than stress granules. P-bodies partially overlap with stress granules in their component proteins and RNAs, and are thought to be involved in constitutive RNA degradation . Therefore the particular cytoplasmic RNA granules that TDP-43 associates with may be both cell type and context dependent.
TDP-43 aggregation and toxicity in animal models
Numerous animal models overexpressing either wild-type or disease mutant TDP-43 have been generated. One aspect common to all of these models is that overexpression of TDP-43 (wild-type or disease mutant) in worms, flies, zebrafish, mice or rats is toxic in a dose-dependent manner and able to produce neurodegeneration. Unfortunately beyond that point of universal agreement, many of these models differ from one another with respect to whether TDP-43 forms insoluble aggregates, is mislocalized to the cytoplasm, or shares other core features of human TDP-43 pathology.
Invertebrate and fish models
Although a subset of these models developed evidence of TDP-43 inclusions or insolubility [59,60], most did not [44,61–63], despite producing similar toxicity and neurodegeneration (Table 2). Therefore, a recurrent theme in these studies is that TDP-43 inclusions are not necessary to promote neurotoxicity. Furthermore, in at least one case, the tendency to form insoluble inclusions (observed in a nuclear export deficient mutant) correlated with decreased rather than increased toxicity . A second consistent finding of these studies is that the presence of the functional RNA binding motif (RRM1) was required for toxicity of TDP-43 overexpression. This was determined by expressing only the C-terminal domain [44,59,64], or by introducing point mutations into the RNA binding domain . This is somewhat at odds with cell line models, which found that expressing the C-terminal domain alone readily promotes cytoplasmic aggregates and is more toxic than full-length protein [29,41]. This suggests that in these animal models TDP-43 toxicity is not through formation of inclusions but instead is via direct modulation of TDP-43 RNA binding partners.
Table 2. Features of invertebrate and zebrafish TDP-43 overexpression models. CTF, C-terminal fragments; dNLS, deleted nuclear localization signal; dNES, deleted nuclear entry signal; HMW, high molecular weight; FFLL, point mutation construct in RRM1 which blocks RNA binding; NR, not reported; S409/410 AA, phosphorylation-deficient mutant.
TDP-43 inclusions or insolubility
Exogenous TDP-43 localization
Differential toxicity of mutant constructs
snb-1 (pan neuronal)
RRM1, C-terminal deletion, and dNLS not toxicC. elegans tbp-1 orthologue also toxic if overexpressed
Some differences between these studies are hard to reconcile at present. First is the issue of differential toxicity of ALS disease mutants. Some found that ALS mutants were more toxic than wild-type TDP-43, whereas others (using the same promoters and outcome measures) found that the mutants were less toxic (Table 2). These studies also do not agree on whether mutating the nuclear localization signal of TDP-43 to mislocalize it to the cytosol made it more toxic  or less toxic [44,63,64] than wild-type TDP-43. One potential origin for these differences is that transgene expression level across studies is hard to compare, and all studies found that level of overexpression is a key factor in determining toxicity.
Numerous rodent models overexpressing TDP-43 have now been produced, and are reviewed in detail elsewhere . With regard to protein aggregation, what is again surprising is that insoluble TDP-43 inclusions are a minor component of the pathology relative to the degree of neurodegeneration produced as a consequence of TDP-43 overexpression. This is consistent with observations from TDP-43 overexpression toxicity in other in vivo model systems discussed above. While TDP-43 inclusions are a minor component of the pathology in rodent models, they recapitulate most other core features of human TDP-43 pathology, including nuclear clearing, C-terminal fragmentation and phosphorylation [67–72]. While large inclusions are not seen, there are still ways that aberrant aggregation of TDP-43 could play a role in promoting neurodegeneration in these models. For example, neurons in which TDP-43 inclusions form may undergo cell death rapidly, and therefore the inclusions are seen infrequently. Alternatively, overexpression of TDP-43 could promote the formation of toxic oligomers (particularly of the C-terminal fragment), which remain soluble and are not incorporated into large inclusions. Finally it is possible that TDP-43 aggregation is not involved in TDP-43 toxicity in these models, and instead altered mRNA processing or other underlying pathways are at play.
A promising aspect of the rodent models is that, across several promoter systems, TDP-43 toxicity is relatively selective to layer V cortical neurons and spinal motor neurons, overlapping with the vulnerable populations in ALS and FTLD [67,68]. This suggests that whatever the mechanism of toxicity of TDP-43 overexpression it appears to tap into the same underlying molecular and cellular pathways disrupted in human disease, making it worthy of continued investigation.
Within a few years since the discovery of TDP-43 inclusions in ALS and FTLD, our knowledge has expanded rapidly. We have learned that TDP-43 aggregation is observed in a diverse set of neurodegenerative and myodegenerative diseases, not just ALS and FTLD. Furthermore the importance of the C-terminal domain, the location of the majority of the ALS-associated mutations, for promoting self-aggregation in addition to mediating protein–protein interactions has come to light. Recent studies have pointed to the possibility that TDP-43 inclusion pathology may reflect an exaggeration of normal accumulation of TDP-43 into various cytoplasmic RNA granules. Finally, while detergent-insoluble inclusions were the key feature leading to the connection between TDP-43 and neurodegeneration, existing animal models clearly demonstrate that TDP-43 toxicity (at least in the context of overexpression) can occur without the formation of TDP-43 inclusions. While somewhat disconcerting, the same ambiguous relationship between inclusion formation and neurodegeneration is familiar to many other fields. In Alzheimer’s disease models, increased amyloid-β production promotes amyloid plaque formation and synaptic dysfunction, but not neurodegeneration . By contrast, in polyglutamine diseases (including spinocerebellar ataxias and Huntington’s disease) toxicity and neurodegeneration do not require inclusion formation, analogous to what is observed in TDP-43 model systems [74,75]. Therefore, it appears that we are just at the beginning of our understanding of the relationship between TDP-43 aggregation, inclusion formation and neurodegeneration. Though a challenging topic, an improved understanding of this relationship will be necessary as we consider ways to modulate TDP-43 aggregation for potential therapeutic benefit in ALS, FTLD and other neurodegenerative diseases.
R.H.B. is supported by the NIH/NINDS (K08 NS055980 and R01 NS069669), the Muscular Dystrophy Association (135428) and the Children’s Discovery Institute, and holds a Career Award for Medical Scientists from the Burroughs Wellcome Fund.