Amyotrophic lateral sclerosis/frontotemporal lobar degeneration pathology and genetics

Cytoplasmic aggregates of the DNA/RNA-binding protein fused in sarcoma (FUS), also known as translocated in sarcoma (TLS), define a subgroup of both amyotrophic lateral sclerosis (ALS; also known as Lou Gehrig's disease) and frontotemporal lobar degeneration (FTLD) [1–3]. In ALS, motor neurons degenerate, and patients rapidly succumb to progressive muscular weakening and paralysis. ALS is the most common form of motor neuron disease in adults, with a lifetime risk of 1/600 to 1/1,000 [4]. Disease onset occurs on average at the age of 55 years, and the 3-year survival rate is close to 50 % [5]. In FTLD, patients suffer from region-specific neurodegeneration in the frontotemporal cortex that affects higher cognitive functions such as speech, language, and personality. FTLD is the second most common cause of dementia under 65 years [6]. Many patients show signs of both motor neuron loss and dementia, and thus, ALS and FTLD are now considered extreme ends of a disease spectrum [7, 8].

Both ALS and FTLD have a strong genetic component because 20–50 % of patients have a family history of neurodegeneration (familial cases) [5, 6, 9]. The discovery of dominant disease-causing mutations in several genes further linked these diseases and led to a pathological subdivision defined by the main aggregating protein (FUS, TAR DNA-binding protein 43 (TDP-43), tau, or SOD1) [8–11].

Aggregation of the microtubule-associated protein tau (MAPT) in the form of neurofibrillary tangles is often caused by pathogenic tau mutations that alter its affinity for microtubules [12]. Cytoplasmic redistribution and aggregation of the nuclear DNA/RNA-binding protein TDP-43 (or TARDBP) is the key pathological feature in most FTLD and ALS patients, hence called FTLD/ALS-TDP [13]. The nuclear DNA/RNA-binding protein FUS forms cytoplasmic inclusions in neurons and glia in a distinct group of patients, termed FTLD/ALS-FUS [1–3, 14]. Mutations in TDP-43 cause predominantly ALS and rarely FTLD, and mutations in FUS cause exclusively ALS [8]. While loss-of-function mutations in the growth factor progranulin (GRN) cause FTLD-TDP [15], no other mutations have been linked to FTLD-FUS so far. The most common genetic cause for ALS and FTLD is the recently discovered massive expansion of a GGGGCC repeat in the first intron of the uncharacterized gene C9orf72, which causes TDP-43 pathology by an unclear mechanism possibly involving RNA toxicity [16–18]. Aggregation of TDP-43 and FUS strongly suggests that dysregulated RNA processing is an important factor in the pathogenesis of ALS and FTLD. In this review, we focus on the role of aberrant splicing in this process.

FUS pathophysiology in ALS and FTLD

FUS is ubiquitously expressed and is predominantly localized in the nucleus. FUS mediates mRNA transport by shuttling in and out of the nucleus [19, 20]. Mislocalization of FUS to the cytoplasm is presumably the first step in the pathophysiological cascade that leads to neurodegeneration. ALS-FUS mutations cluster around the C-terminal nuclear localization signal (NLS) and disrupt the nuclear import of FUS [1, 2, 21]. The other members of the FET family of proteins, Ewing sarcoma protein (EWS) and TATA-binding protein-associated factor 15 (TAF15), are imported normally into the nucleus. FUS inclusions therefore lack EWS and TAF15 [14]. In contrast, FTLD-FUS cases show inclusions where the entire FET family coaggregates [14]. Thus, in FTLD-FUS, which typically lacks FUS mutations, nuclear import of the FET family may be more broadly impaired, although other unrelated transportin 1 cargos are not affected. This topic was discussed in the recent review by Dormann and Haass [11].

Regardless of the mechanism, mislocalization of FUS results in a reduction of nuclear function and an increase of cytoplasmic FUS prone to aggregation [22]. Cytoplasmic FUS may result in toxic gain of function by disrupting extranuclear RNA metabolism [23].

FUS animal and cell culture models have been generated to dissect loss- and gain-of-function pathomechanisms. Two FUS knockout mouse lines show surprisingly different phenotypes. Inbred knockout mice fail to suckle and die within a few hours after birth [24]. In contrast, outbred knockout mice reach adulthood, but the males are sterile [25]. Despite the different phenotypes, both knockout mice show genomic instability. Furthermore, loss of FUS alters neuron morphology. Cultured neurons from knockout mice have fewer mature spines, but more filopodia-like dendritic protrusion than wild-type neurons [26]. Early FUS knockdown in hippocampal neurons results in enlarged axonal growth cones with disorganized cytoskeleton [27]. Gain-of-function mouse models that overexpress wild-type FUS succumb to progressive paralysis and die after 12 weeks. These mice show FUS-positive inclusions in spinal cord motor neurons and, therefore, replicate some aspects of human pathology [2, 28]. Together, these results suggest an important role of FUS in neurons during development. They also point to a combined loss-of-function and toxic gain-of-function pathomechanism in ALS/FTLD-FUS [29]. Thus, it is critical to understand the physiological function of FUS in the brain.

Alternative splicing in the brain

Alternative splicing drives and vastly extends the diversity of the transcriptome and proteome. One single gene may give rise to many different protein isoforms, often with distinct functions [30–32]. Tailored protein function is possible due to tissue- and development-dependent regulation of alternative splicing. Compared to other tissues, the human brain shows exceptionally high levels of alternative splicing, with more than 40 % of genes being alternatively spliced [33]. A complex interplay of cis- and trans-acting elements regulates alternative splicing. The cis-acting elements are splicing enhancer and inhibitory sequences within the pre-mRNA that recruit trans-acting RNA-binding proteins (RBP), which may themselves be further regulated by posttranscriptional modifications. The spliceosome, a RNA–protein complex consisting of small nuclear RNAs (U1, U2, U4, U5, and U6) and several RBPs, catalyzes splicing [34].

FUS is also part of the spliceosome and directly binds pre-mRNA [27, 35–38] and the splicing factors: splicing component 35 (SC35), polypyrimidine tract-binding protein (PTB), and the serine arginine (SR)-related proteins SRm160 and SRp75 [39–41]. Splicing of pre-mRNA transcripts starts during transcription, and both processes are tightly integrated [42]. FUS also regulates RNA polymerase II-mediated transcription by binding its C-terminal domain and regulating its phosphorylation [43]. Thus, FUS may integrate transcriptional and splicing regulation through RNA–protein and protein–protein interactions.

Until recently, the analysis of FUS-mediated splicing was limited to artificial exogenous splicing targets [39, 44]. The recent identification of endogenous neuronal splicing targets such as MAPT/tau [27] will allow detailed analysis of regulatory elements and will help to pinpoint the role of FUS in alternative splicing.

Pathogenic mutations highlight the importance of alternative splicing in neurodegeneration. In FTLD-tau patients, for example, MAPT mutations around exon 10 alter its splicing, thereby causing tau aggregation and impairing the axonal function of tau [45]. Moreover, mutations in trans-acting factors such as the RBP survival motor neuron protein 1 (SMN1) cause spinal muscular atrophy [46, 47]. Thus, identifying the splicing targets of FUS will help us to understand the pathogenesis of ALS/FTLD-FUS.

FUS-mediated alternative splicing in the brain

RNA-binding and alternative splicing targets of FUS have been studied previously in cell culture models (human embryonic kidney 293 cells [23, 43], motor neuron-like cells NSC-34 [48]) or in Xenopus laevis embryos [49]. Recently, four independent studies analyzed RNA bound to FUS in neuronal tissue [35–38] using different cross-linking and immunoprecipitation (CLIP) technologies and next-generation sequencing [50, 51]. The four groups then correlated CLIP results to the splicing changes detected in FUS knockout brains or cultured neurons with FUS knockdown, in order to identify FUS splicing targets in the nervous system. The experimental approach and results are compared in Table 1.

Table 1 Comparison of experimental conditions of FUS CLIP/exon array studies
Full size table

The four studies are largely consistent in their conclusions: firstly, in the brain, FUS regulates primarily alternative splicing events rather than transcription or constitutive splicing. Secondly, FUS binds several thousand transcripts and favors very long introns. FUS-binding sites often flank the regulated alternatively spliced exon. However, only 42 % [38] to 55 % [37] of transcripts differentially spliced after FUS knockdown were direct binding targets of FUS. Additionally, two studies that applied CLIP technology to human and mouse brain tissue [37, 38] found highly comparable RNA-binding profiles and a high correlation of binding targets between humans and mice.

Thirdly, no simple RNA sequence can explain the RNA-binding pattern of FUS. In a fraction of targets ranging from 10 % [35] to 60 % [37], different groups detected a significant preference for G/C [35], C/U [38], GGU [36], or GUGGU [37] motifs, although the enrichment was rather low. The GGU and GUGGU motifs are similar to the GGUG motif identified previously through in vitro affinity selection [44]. Two groups also evaluated the enrichment of RNA structure motifs, such as the short-stem loop motif proposed by Hoell et al. [23]. Ishigaki et al. [35] found a modest enrichment of short-stem loop in FUS RNA targets, but Rogelj et al. [36] did not. Further biochemical studies are necessary to fully understand the RNA-binding specificity of FUS in the brain [38].

Fourthly, gene ontology analysis revealed that FUS splice targets are predominantly involved in the following pathways: axonogenesis, axon guidance, cell adhesion, neuron projection, vesicle transport, and cytoskeletal organization [35–38]. Among other splicing events, loss of FUS leads to inclusion/exclusion of exon cassettes (e.g., MAPT/tau) [35–37], selection of alternative 3′ untranslated regions (UTRs) (e.g., ABLIM1) [36] and intron retention (e.g., small nuclear ribonucleoprotein 70 (snRNP70)) [38]. Intron retention typically leads to insertion of a premature stop codon and may be a mechanism to regulate protein abundance through nonsense-mediated mRNA decay. Interestingly, a previous study using FUS knockdown in X. laevis observed extensive intron retention with functional effects on the fibroblast growth factor pathway [49].

Finally, the studies comparing binding targets of TDP-43 and FUS detected only few RNAs bound by both proteins. Despite their structural homology, these two proteins seem to regulate a vastly different set of genes [36, 37]. Also, TDP-43 binds its targets with surgical precision, whereas FUS typically binds broadly along nascent transcripts with long introns, leading to a characteristic saw-tooth pattern of binding [36, 37]. This could indicate a role of FUS in stabilizing nascent RNA during transcriptional elongation.

Despite the consensus regarding pathways regulated by alternative splicing, only six genes were identified in at least three studies (Table 2), and 71 genes were identified in at least two studies (Table 3). Only netrin G1 (NTNG1), previously linked to Parkinson's disease (PD) and schizophrenia, was identified in all four studies [35–38]. MAPT/tau was identified in three studies and, additionally, also in our candidate-based approach [27, 35–37]. FUS also binds MAPT/tau mRNA in the human brain [38]. The overlapping targets NTNG1, MAPT, ABLIM1, NRCAM and BRAF are discussed below.

Table 2 Alternative splicing upon loss of FUS. Hits were identified in at least three studies
Full size table
Table 3 Putative splicing targets of FUS. Hits were identified in at least two studies
Full size table

The differences in experimental approach and statistical analysis are probably responsible for the limited overlap of splicing targets (Table 1). Differences in transcript abundance between whole brain tissue [36, 37], cultured neurons [35], or neurons differentiated from mouse embryonic stem cells [38] also limit the comparison. The use of FUS knockout brains [36, 37] in contrast to FUS knockdown in vitro [35, 38] or in vivo [37] could also account for the differences. Finally, Nakaya et al. [38] report several RBPs among the FUS targets, for example, EWS [36, 38] and snRNP70 [38], and suggest that FUS cross-regulates the RBP network. The lists of FUS-regulated genes inevitably include indirect splicing events.

Interestingly, independent studies of TDP-43 that applied CLIP technology [52–54] also showed limited overlap among targets [51]. A recent comparison of these studies also points to methodological differences as the underlying cause [51].

We considered only the top splicing targets for the discussion on potential implications for ALS/FTLD.

FUS regulates alternative splicing of proteins related to axonal biology

MAPT/tau

Three FUS CLIP/exon array studies and our candidate-based study identified increased MAPT/tau exon 10 inclusion upon loss of FUS [27, 35–37]. Additionally, both our study [27] and that of Lagier-Tourenne et al. [37] reported an enhanced inclusion of exon 2 (ENSMUSE00000107966) and exon 3 (ENSMUSE00000107958) upon FUS knockdown (Table 2). Although formal proof that FUS regulates splicing of MAPT/tau in the human brain is still missing, there is strong evidence for functional conservation [37, 38].

The MAPT gene, encoding the protein tau, consists of 16 exons and is mainly expressed in the nervous system. Tau shows a complex alternative splicing regulation of an N-terminal cassette (exons 2 and 3) and exon 10 that leads to six different isoforms (0N3R, 1N3R, 2N3R, 0N4R, 1N4R, 2N4R) [55] (Fig. 1a). Inclusion of exon 2 or exons 2 and 3 adds one or two short acidic regions (termed 1N and 2N) in the so-called projection domain. Inclusion of exon 10 inserts a fourth microtubule binding region (4R), which increases affinity to microtubules compared to the shorter 3R isoforms [56]. During development, expression shifts toward longer isoforms.

Fig. 1
figure 1

Genomic structures and isoforms of MAPT/tau and NTNG1. a The human MAPT gene encodes for the tau protein and contains 16 exons. E0 and E14 (white) are noncoding, E6 and E8 (brown) are not expressed in the human brain, and E4a is only expressed in the peripheral nervous system. The N-terminal part of the protein shows a complex alternative splicing of the cassettes E2 and E3. Inclusion of E3 is coupled to the inclusion of E2. Inclusion of E2 or E2/E3 adds one or two acidic regions (1N, 2N) in the projection domain (N-terminal part). Alternative splicing of E10 is regulated independently. Exons 9–12 code for microtubule binding domains. Inclusion of E10 adds extra fourth microtubule binding region (4R). b The human NTNG1 gene contains 10 exons. E1 (white) is noncoding. Complex alternative splicing of E5, E6, E7, E8, and E9 results in at least nine different isoforms termed G1a, G1b, G1c, G1d, G1-e, G1-l, G1-m, G1-n, and G1-o. E exon

Full size image

Tauopathies are characterized by neurofibrillary tangles, which consist of aggregated hyperphosphorylated tau. Such aggregates are found in corticobasal degeneration, progressive supranuclear palsy, frontotemporal dementia and parkinsonism linked to chromosome 17 (FTDP-17), as well as Alzheimer's disease [57]. Tau mutations in FTDP-17 patients promote tau aggregation either by disturbing the tightly controlled 4R/3R ratio [58] or by affecting the interaction of tau with microtubules [56]. The changes in microtubule stability directly affect the transport along microtubules. Interestingly, increased 4R expression may cause neurodegeneration even in the absence of visible tau aggregation [59]. In mouse models of tauopathies, toxicity precedes tau aggregation [60]. In PD, 4R expression is correlated with progression to dementia without detectable tangles [61–63]. The H1 MAPT haplotype, which enhances 4R expression [64], has been genetically linked to PD. Finally, tau is also genetically [65] and pathologically [66] linked to the Guam variant of ALS. Thus, a shift toward 4R tau may contribute to neurodegeneration in ALS/FTLD-FUS, despite the lack of overt tau aggregation [67, 68].

Netrin G1

NTNG1, also known as laminet-1, belongs to the netrin family, with the distinction of being membrane bound via a glycosylphosphatidylinositol (GPI) anchor [69–71]. Netrins provide important guidance cues during brain development [72]. For example, NTNG1 and its ligand NGL-1 regulate axon guidance as well as synapse formation and maintenance [70]. Nine different isoforms resulting from alternative splicing of the exons 5–9 have been reported so far [71]. The protein contains several laminin- and non-laminin-type epidermal growth factor (EGF) domains.

NTNG1 is the only target identified in all four FUS CLIP/exon array studies. Rogelj et al. [36] report an increased inclusion of exon 9 (ENSMUSE00000511732) upon FUS knockdown, while Ishigaki et al. and Lagier-Tourenne et al. [35, 37] report exon inclusion in a PCR amplicon spanning exon 7 (ENSMUSE00000670473) to exon 9 (ENSMUSE00000511732). Exons 8 and 9 insert laminin-type EGF domains. In contrast, Nakaya et al. [38] report an increased inclusion of exon 10 (ENSMUSE00000947279) that codes for an extracellular EGF domain. Little is known about the differential function of NTNG1 isoforms. Based on the number of EGF-like domains, the isoforms might have different affinities for the ligand NGL-1 [69]. Selective reduction of the G1c isoform has been observed in familiar cases of schizophrenia [73] and bipolar disorders [69]. Interestingly, NTNG1 is also linked to PD in a genome-wide association study [74].

ABLIM1, NRCAM, and BRAF

Altered splicing of ABLIM1, NRCAM, and BRAF upon loss of FUS was reported in at least three studies (Table 2). The actin-binding LIM protein ABLIM1 is expressed throughout the body and exists in three different isoforms ABLIM1-s, ABLIM1-m, and ABLIM1-l, which differ in the number of LIM domains [75]. In contrast to NTNG1 and MAPT/tau, the reported effects in splicing vary among the studies. Lagier-Tourenne et al. and Nakaya et al. detected preferential skipping of exons ENSMUSE00000292146 and ENSMUSE00000292118, respectively [37, 38]. The specific function of these exons is unknown. Rogelj et al. [36] detected an alternative 3′ splice site event at ENSMUSE00000640490, which is almost identical to exon ENSMUSE00000292146 except that it is annotated to be 5 bp longer. The protein product of this transcript lacks the C-terminal vinillin headpiece that mediates the binding of F-actin. Lastly, Lagier-Tourenne et al. [37] detected an alternative start site at ENSMUSE00000793956. The mouse genome database (NCBIM37) lists only one short transcript with this alternative start site, which also lacks the C-terminal vinillin headpiece. ABLIM1 binds F-actin, bridges the actin cytoskeleton, and is known to mediate axon guidance and outgrowth [76]. Interestingly, netrin signaling has been shown to activate ABLIM1 [77].

NRCAM is a transmembrane protein that belongs to the L1 family of cell adhesion molecules [78]. Alternative splicing of NRCAM results in more than a dozen isoforms that are differentially regulated during development [79]. The function of the different isoforms is unknown. Similar to ABLIM1, there is limited overlap in the reported affected exons upon loss of FUS. Rogelj et al. [36] reported skipping of exon ENSMUSE00000325244, which codes for an Ig-like beta sandwich domain. In contrast, Lagier-Tourenne et al. and Nakaya et al. [37, 38] report skipping of exon ENSMUSE00000325376 and inclusion of exon ENSMUSE00000325135, respectively (Table 2). Interestingly, these are neighboring exons, and the latter also encodes an Ig-like domain. NRCAM is crucial for axon growth and guidance, synapse formation, and neurite outgrowth [78] and has been linked to different forms of cancer and autism [78].

The BRAF gene codes for the B-raf protein member of the Raf family of kinases (including A-raf and C-raf). Raf kinases are part of the mitogen-activated protein kinase (MAPK) cascade, which activates gene expression upon growth factor stimulation. B-raf is highly expressed in the CNS [80] and is the major activator of the extracellular signal-regulated kinase (ERK)1/2 pathway in neurons [81, 82]. Both in vitro and in vivo studies have demonstrated the essential role of B-raf in neuronal survival and differentiation [82, 83]. Interestingly, conditional double knockout of B-raf and C-raf resulted in reduced axon growth [83]. Alternative splicing of BRAF results in 10 isoforms expressed in different tissues. The longest isoforms are abundant in the CNS [80]. Ishigaki et al. [35] report exon skipping in the region encompassing exons ENSMUSE00000618025, ENSMUSE00000951452, and ENSMUSE00000562746 after FUS knockdown. Lagier-Tourenne et al. [37], however, report preferential inclusion of exon ENSMUSE00000618025. This discrepancy could reflect differences in FUS regulation of alternative splicing in different cell populations in the brain, since Ishigaki et al. [35] analyzed cultured neurons, and Lagier-Tourenne et al. [37] analyzed whole mouse brain. Nakaya et al. [38] report skipping of exon ENSMUSE00000618032 upon loss of FUS. All known isoforms can activate ERK1/2 signaling [80], but alternative splicing might modulate kinase activity and substrate specificity, thus altering growth factor signaling [80].

Consequences for ALS and FTLD

Tau, NTNG1, ABLIM1, NRCAM, and B-raf exemplify a common theme among FUS splice targets, because they all affect cytoskeletal organization and, in particular, axon growth and maintenance. The recent identification of ALS-causing mutations in profilin 1 [84], a protein that regulates actin polymerization, as well as mutations in the neurofilament subunit H [85], further highlights the importance of the cytoskeleton in the pathogenesis of neurodegenerative diseases. Altogether, FUS mislocalization to the cytoplasm may impair maintenance and repair of long axons. Interestingly, the earliest signs of ALS are axon retraction and denervation, which clearly occur before the loss of neuronal cell bodies [86, 87].

How could impaired FUS-mediated alternative splicing render neurons more vulnerable in ALS/FTLD-FUS? We propose the following model (Fig. 2): (1) in ALS or FTLD, pathogenic mutations or impaired nuclear import causes cytoplasmic mislocalization of FUS [88]. Reduction of nuclear FUS results in aberrant alternative splicing of axonal and cytoskeleton-related transcripts, such as MAPT/tau. (2) The network responsible for axonal growth, maintenance, and repair deteriorates, and neuronal connections are weakened. Secondary effects, such as misregulation of other RBPs [38] or their sequestration into aggregates, may enhance neurodegeneration. (3) Aging and other stressors [11, 88] can trigger denervation and early clinical symptoms. Lastly, (4) due to disrupted repair mechanisms, the damaged tissue cannot be repaired resulting in progressive neurodegeneration.

Fig. 2
figure 2

Hypothesis of splicing centered pathomechanism in ALS/FTLD-FUS. 1 Impaired nuclear import of FUS, due to mutations in the NLS or transport defects, causes mislocalization of FUS to the cytoplasm. This leads to loss of nuclear function and thus changes in alternative splicing of axonal and cytoskeleton related genes. 2 Altered splicing disturbs axonal growth and maintenance and results in axonal atrophy and loss of connectivity. 3 Aging and stress accelerate the process of neuronal denervation and lead to the first clinical symptoms. Connections to dendrites or neuromuscular junction (NMJ) are affected. 4 Due to disturbed repair mechanisms, the affected neurons are not able to cope with the stress and to repair damaged connections. The result is progressive degeneration observed in disease

Full size image

Conclusions and key open questions

We have reviewed here the alternative splicing targets of FUS reported by four independent groups. The most robustly identified targets are linked to cytoskeleton, axon growth, and maintenance [35–38]. However, FUS binds many more RNAs apparently without changing splicing or expression. How are these RNAs affected by FUS? Furthermore, it is unclear how FUS affects the bound long noncoding RNAs, for example, maternally expressed 3 (Meg3) and nuclear enriched abundant transcript 1 (NEAT1) [37]. Meg3 and NEAT1 have been linked to neurodegeneration, because their expression is significantly dysregulated in Huntington's disease [89]. Moreover, it will be important to understand how FUS aggregates might impair cytoplasmic RNA metabolism [23]. Finally, to fully understand the pathomechanism of ALS and FTLD with FUS pathology, we should also consider other potential roles of FUS in the nervous system, particularly genomic stability. Even postmitotic neurons seem to require constant genome repair, because cellular stress and normal synaptic activity can cause double-strand breaks [90].

In conclusion, we now know that loss of FUS alters splicing of key components of the cytoskeleton and related proteins that promote axonal maintenance and repair. The next challenge will be to confirm these findings in ALS and FTLD patients and relate them with the pathology and symptoms. Only then, we can begin to translate these findings into therapeutic approaches for these devastating diseases.