Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease affecting 1 in 50,000 individuals per year and is characterized by the loss of upper and lower motor neurons in the brain and spinal cord [1]. The typical age of onset is late middle life and patients typically present with progressive muscle weakness and atrophy. The prognosis for ALS is dire, with most patients succumbing to neuromuscular respiratory failure within 2–3 years of symptom onset. At present, ALS treatment is focused primarily on symptom management, with three FDA-approved therapies each demonstrating modest benefits in the clinic [2] and one highly effective, recently approved therapy for SOD1-mediated ALS [3]. Approximately 10% of ALS patients experience a disease that is hereditary in nature, typically in an autosomal dominant manner. Over 30 genes have been identified as potential drivers of ALS and the most prominent mutations occur in chromosome nine open reading frame 72 (C9ORF72) and the enzyme superoxide dismutase 1 (SOD1). In addition, more than half of ALS mutations occur in genes encoding RNA binding proteins (RBPs) such as fused in sarcoma (FUS), TAR DNA-binding protein 43 (TARDBP), heterogeneous nuclear ribonucleoprotein A1 (HNRNPA1), T-cell restricted intracellular antigen-1 (TIA1), and Ataxin 2 (ATXN2), amongst others [4]. The remaining 90% of patients show no apparent family history of the disease and hence are said to have “sporadic” ALS. Despite a gap in our understanding of the missing heritability of sporadic ALS, it likely results from a combination of environmental and genetic risk factors, including described mutations in C9ORF72, TDP-43, FUS and SOD1 amongst others [5].

The majority (97%) of all ALS patients, including sporadic cases, have a post-mortem histopathological presentation of nuclear TDP-43 depletion and accompanying cytoplasmic accumulation in motor neurons of the brain and spinal cord. It should be noted, however, that familial ALS patients with mutations in SOD1 or FUS do not share this TDP-43 pathology [6], [5]. Conversely, approximately 45% of frontotemporal lobar dementia (FTLD) patients display TDP-43 pathology and share several common genetic mutations with ALS [7]. Cytoplasmic TDP-43 aggregates are ubiquitinated as well as hyperphosphorylated, and have been shown to form insoluble amyloid-like fibrils. These insoluble aggregates are suggested to promote neuronal cytotoxicity through a combination of TDP-43 gain- and loss- of function mechanisms [8]. Truncated forms of TDP-43 arising from alternative translation or proteolytic cleavage, mainly the C-terminal fragments (CTF) of 35 kDa and 25 kDa, can also be found in these aggregates. These cytoplasmic aggregates also sequester a plethora of RBPs including FUS, ATXN2, TIA-1, Staufen, RBM45 (RNA Binding Motif 45), PABP1 (Polyadenylate-binding protein 1), and numerous others [reviewed in [9]]. In healthy neurons, while TDP-43 is mostly nuclear, it also can shuttle between the nucleus and the cytoplasm. Exactly how TDP-43 is exported from the nucleus and forms cytoplasmic aggregates in ALS is not fully understood but is proposed to be via a combination of aberrant changes in nucleocytoplasmic transport and TDP-43 solubility [10]. Of relevance, it has been reported that there is an age-dependent decline in nuclear pore components [11].

Pathological TDP-43 aggregates in the neurons of ALS patients have been proposed to originate from specific mRNA-protein containing granules called stress granules (SGs) [reviewed in [12]]. SGs are dynamic, cytoplasmic membraneless-organelles (MLOs) that form as part of a survival response to stressors such as heat shock, proteasome inhibition, viral infection, oxidative and osmotic stress. SGs sequester polyadenylated [poly(A)] mRNAs stalled in translation initiation until the stress has been alleviated, allowing cells to conserve energy and resources [13]. This model for ALS pathogenesis was put forward in part because TDP-43 aggregates have been reported to colocalize with SG markers [15], [14], [16], suggesting that persistent stress could lead to the mislocalization of TDP-43 into these aggregates. In addition, TDP-43 can be recruited to SGs in cell culture in response to stress, although it was noted in HeLa cells that the majority of TDP-43 remains in the nucleus to form nuclear granules [17]. A reduction of the SG component, Ataxin-2, has been reported to reduce TDP-43 aggregation, improving motor function and survival in mice expressing high levels of wild type TDP-43 [18]. Also, several ALS mutations occur in genes encoding proteins involved in cellular SG formation - such as TDP-43, TIA1, ATXN2, hnRNPA1, TAF15 (TATA-box binding protein associated factor 15) and FUS - which could disrupt SG dynamics and over time lead to the formation of amyloid-like, insoluble TDP-43 aggregates [19]. Accordingly, ALS mutations have also been linked to proteins involved in SG clearance, such as the protein disaggregase VCP (Valosin Containing Protein) [20] and the ubiquitin binding protein SQSTM1 (Sequestosome 1) [21]. However, this model of SGs seeding TDP-43 pathological aggregates remains controversial as other groups have failed to report colocalization of TDP-43 with SG factors in the brain and spinal cord of ALS patients, despite observing their association in cell culture [22], [23], [24]. Whether these differences are due to discrepancies in methods of detection or quality of fixed post-mortem tissue is inconclusive. In addition, TDP-43 has been shown to be capable of aggregating independently of SG formation [25], [26]. An optogenetic approach inducing TDP-43 oligomerization in the presence of blue light formed RNA-free cytoplasmic inclusions which did not show colocalization with SG markers. Intriguingly, the formation of these cytoplasmic inclusions could be inhibited upon treatment with an RNA oligonucleotide based on the TDP-43 target Clip-34 sequence, called “bait oligo”, underlining the effect of RNA-binding by TDP-43 in preventing aggregation. It should be noted that this group also did not detect poly(A) mRNA, an obligatory component of SGs, as colocalized with TDP-43 inclusions in spinal cords of ALS patients and FTLD hippocampus [26]. Moreover, the addition of amyloid-like fibrils in the form of fragmented TDP-43 to neuronal cells was sufficient to form cytoplasmic granules that accumulated endogenous TDP-43 and nuclear import factors, as well as result in nuclear TDP-43 depletion, whilst in the absence of canonical SGs [25].

Despite extensive proteomics-led characterization of the components of TDP-43 aggregates in cell culture, how pathological TDP-43 fibrillization initiates in the neurons of ALS patients is still unclear. Herein, we consider whether TDP-43 pathological aggregates can arise from MLOs other than SGs given that their protein components can overlap significantly [27]. We will focus on MLOs that contain TDP-43 and FUS as they are mutated in ALS and are present within pathological aggregates in sporadic ALS [9], [28]. Importantly, we will also discuss the functional consequences of ALS-linked TDP-43 and FUS mutations on these MLOs, underscoring early pathological events that may occur in ALS prior to cytosolic aggregate formation.

In addition to SGs, there are numerous other mRNA-RBP containing MLOs that serve to compartmentalize highly specific cellular functions. These are often referred to as biomolecular condensates and can be defined as membraneless cellular compartments which can be visualized by fluorescence microscopy as foci ranging from 20 nm to 1 µm in diameter [29]. These include the nucleolus, nuclear pore, nuclear speckles, Cajal bodies, paraspeckles, PML (promyelocytic leukemia protein) bodies, germ granules, P- (processing) bodies and mRNA transport granules. The common denominator of these biomolecular condensates is that they are formed by a process called liquid-liquid phase separation (LLPS). LLPS describes the reversible de-mixing/condensation of biomolecules (nucleic acids and proteins) from a solution into liquid droplets/condensates. These droplets are referred to as “liquid” as they are dynamic and can exchange components with the surrounding solution or fuse with other droplets. The cytoplasm or nucleoplasm can thus be conceptualized as biomolecules dissolved in a solvent, whereby at a certain threshold concentration the interactions between the biomolecules are more favorable than its interactions with the solvent, resulting in their de-mixing/condensation from the solution. This threshold concentration can be modulated by several factors other than protein/nucleic acid concentration, such as temperature, salt and ion concentration, as well as pH [30], [31]. Importantly, LLPS is highly sensitive to changes in protein/nucleic acid stoichiometry that can occur in response to cellular signaling in physiological or pathological contexts, as well as artificially via protein overexpression in cell culture.

LLPS is catalyzed by multiple weak, non-covalent interactions between proteins and nucleic acids. Specifically, proteins contained within these biomolecular condensates, such as RBPs, are enriched in intrinsically disordered regions (IDRs) which contain low-complexity domains (LCDs) that lack secondary structure. Notably, prion-like domains (PrLDs) are a type of IDR with similar amino acid composition to Yeast prion domain. IDRs facilitate multivalency by promoting binding to several copies of itself or other proteins [31]. RGG (arginine-glycine-glycine) motifs, capable of mediating RNA and protein interactions, also occur in LCDs and favor multivalent interactions [33], [32]. Other RNA binding motifs such as RRMs (RNA recognition motifs) also promote multivalency and regulate condensation in the presence of RNA [31]. Unsurprisingly, proteins with ALS-linked mutations occurring within their low complexity regions as well as mutations that alter cellular localization and thus subcellular concentration, such as in the NLS (nuclear localization signal), demonstrate altered LLPS [34]. The effect of these mutations on LLPS have been mainly studied and discussed in the context of SG dynamics [reviewed in [35]]. However, here we will highlight how they can also affect other nuclear and cytosolic biomolecular condensates. It is pivotal to consider how all biomolecular condensates may be disrupted as a network during different stages of ALS pathogenesis as they are highly interconnected, sharing numerous protein and nucleic acid components [27]. Importantly, dynamic liquid droplets can be converted or “matured” into irreversible, fibrillar solids reminiscent of pathological aggregates [36], [37]. In addition, amyloid-like fibrils of TDP-43 have been shown to be sufficient to trigger aggregation of endogenous TDP-43, underlining the potential role of LLPS in seeding aggregation [25]. However, how disruptions in LLPS lead to protein aggregation in ALS neurons is still unclear given that most of these studies are performed in in vitro systems.

Familial ALS mutations in TARDBP and FUS each constitute approximately 5% of cases, whereby patients present with cytoplasmic inclusions of TDP-43 and FUS, respectively. These inclusions can also be found in sporadic ALS cases, with approximately 97% of patients showing TDP-43 aggregates and only approximately 1% depicting FUS cytoplasmic inclusions [38]. It has been suggested that FUS cytoplasmic mislocalization is more common than previously appreciated in sporadic ALS spinal cords, occurring in an unaggregated state rather than cytoplasmic inclusions [39]. In addition, FUS-positive inclusions in motor neurons of the anterior horn in sporadic ALS cases were found to form mostly “skein-like” inclusions compared to the mostly “granular” type observed for TDP-43 [28]. Intriguingly, TDP-43 and FUS are not present in the same pathological inclusion in patients, and TDP-43 pathology is not observed in patients with familial FUS mutations [38], even though they have been reported to co-aggregate in yeast [40]. It has been reported that FUS inclusions colocalize with TDP-43 in spinal motor neurons of sporadic ALS patients independent of FUS mutations, however, it was suggested that this may be due to discrepancies in immunostaining techniques [41].

TDP-43 is a highly conserved, ubiquitously expressed RNA/DNA binding protein encoded by the TARDBP gene on chromosome 1p36.22. TDP-43 has been ascribed a plethora of diverse cellular functions including transcriptional regulation, pre-mRNA splicing, alternative polyadenylation, miRNA biogenesis, translation, mRNA stability and transport [reviewed in [42]]. TDP-43 is mainly localized in the nucleus although it is capable of nucleocytoplasmic shuttling, containing an NLS within its N-terminal domain (NTD) and a putative NES (nuclear export signal) in its RRM2 [43]. However, this NES was demonstrated to be non-functional as it was neither necessary nor sufficient for TDP-43 nuclear export [44], [45]. The globular NTD has been shown to mediate TDP-43 dimerization and oligomerization in vitro and in cell culture [46]. TDP-43 contains two RRMs and shows preferential binding to UG-rich RNA sequences [47]. At its C-terminus, TDP-43 comprises a glycine-rich, unstructured PrLD that is reported to be aggregation-prone in vitro and in cells [48] (Fig. 1). Importantly, the vast majority of ALS-associated TDP-43 mutations occur in its PrLD, several of which have been shown to either disrupt or promote phase separation [34], [49].

FUS is a ubiquitously expressed member of the FET/TET family of RNA/DNA binding proteins encoded by the FUS gene on chromosome 16 and is implicated in numerous cellular functions such as transcriptional regulation, DNA repair, pre-mRNA splicing, miRNA processing, translation, mRNA stability and transport [reviewed in [50]]. Intriguingly, FUS shows numerous structural and functional similarities with TDP-43. FUS is also predominantly nuclear and can shuttle between the nucleus and cytoplasm [51], containing a reported non-functional NES within its RRM as well as a C-terminal NLS [44]. While FUS contains only a single RRM, it also comprises other RNA binding domains such as a zinc finger (ZnF) and RGG domains. FUS also contains an N-terminal PrLD that is capable of phase separation in vitro, although this effect was enhanced in the presence of its RGG domain [52] (Fig. 1). Intriguingly, the reversibility of FUS phase separation was shown to be affected by ALS mutations in the PrLD, leading to fibrillization over time [37]. Although numerous ALS associated mutations occur in the PrLD, unlike TDP-43, the vast majority occur in its C-terminal low complexity region encompassing the second RGG domain and NLS [50].