Mutations in fused in sarcoma (FUS) are a cause of familial amyotrophic lateral sclerosis (fALS). Patients carrying point mutations in the C‐terminus of FUS show neuronal cytoplasmic FUS‐positive inclusions, whereas in healthy controls, FUS is predominantly nuclear. Cytoplasmic FUS inclusions have also been identified in a subset of frontotemporal lobar degeneration (FTLD‐FUS). We show that a non‐classical PY nuclear localization signal (NLS) in the C‐terminus of FUS is necessary for nuclear import. The majority of fALS‐associated mutations occur within the NLS and impair nuclear import to a degree that correlates with the age of disease onset. This presents the first case of disease‐causing mutations within a PY‐NLS. Nuclear import of FUS is dependent on Transportin, and interference with this transport pathway leads to cytoplasmic redistribution and recruitment of FUS into stress granules. Moreover, proteins known to be stress granule markers co‐deposit with inclusions in fALS and FTLD‐FUS patients, implicating stress granule formation in the pathogenesis of these diseases. We propose that two pathological hits, namely nuclear import defects and cellular stress, are involved in the pathogenesis of FUS‐opathies.
There is a Have you seen ...? (August 2010) associated with this Article.
ALS, also known as Lou Gehrig's disease, is an incurable, severely disabling condition that is characterized by the degeneration of both upper and lower motor neurons. Loss of motor neurons leads to progressive muscle weakening, atrophy and spasticity. The majority of patients succumb to the disease within 1–5 years after disease onset, typically because of respiratory failure (Boillee et al, 2006). Although the majority of ALS cases are sporadic (sALS), about 10% are inherited in a dominant manner (fALS) (Boillee et al, 2006; Valdmanis and Rouleau, 2008). Of these, about 5–10% are caused by mutations in the TAR DNA‐binding protein 43 (TDP‐43) gene on chromosome 1 (Gitcho et al, 2008; Kabashi et al, 2008; Mackenzie and Rademakers, 2008; Sreedharan et al, 2008) or the FUS gene on chromosome 16 (Kwiatkowski et al, 2009; Vance et al, 2009). Although these genes account for only a small number of fALS cases, their gene products seem to have a crucial function in the pathogenesis of the majority of ALS cases, including sALS, as well as the related disorder frontotemporal lobar degeneration with ubiquitin‐positive inclusions (FTLD‐U). Both neurodegenerative diseases are characterized by the presence of neuronal and/or glial TDP‐43 or FUS inclusions. TDP‐43 inclusions are found in most ALS cases, with the exception of fALS caused by mutations in the Cu/Zn superoxide dismutase (SOD1) gene (Mackenzie et al, 2007). Moreover, TDP‐43‐positive inclusions are also found in >90% of FTLD‐U patients (Arai et al, 2006; Neumann et al, 2006), now renamed as FTLD‐TDP (Mackenzie et al, 2010). FUS inclusions are present in the remaining 10% of atypical TDP‐43‐negative FTLD‐U cases (aFTLD‐U) (Neumann et al, 2009a) and in other rare cases of FTLD, such as basophilic inclusion body disease (BIBD) (Munoz et al, 2009) and neuronal intermediate filament inclusion disease (NIFID) (Neumann et al, 2009b), now subsumed as FTLD‐FUS (Mackenzie et al, 2010), as well as in fALS patients carrying FUS mutations (Kwiatkowski et al, 2009; Vance et al, 2009). These diseases are now commonly termed FUS‐opathies (Munoz et al, 2009). The discovery of TDP‐43 and FUS inclusions in both ALS and FTLD has led to the concept that ALS and FTLD are related diseases and that the same proteins are involved in their pathogenesis (Neumann et al, 2009a). This is further supported by the fact that up to 50% of ALS patients show cognitive impairment and a significant portion of FTLD patients develop motor neuron disease (Talbot and Ansorge, 2006).
Both FUS and TDP‐43 are DNA‐ and RNA‐binding proteins that shuttle continuously between the nucleus and cytoplasm (Zinszner et al, 1997b; Ayala et al, 2008) and are involved in multiple steps of gene expression, such as transcriptional regulation, pre‐mRNA splicing and microRNA processing (Buratti and Baralle, 2008; Lagier‐Tourenne and Cleveland, 2009). In addition, FUS has been implicated in mRNA export and mRNA transport to neuronal dendrites (Fujii and Takumi, 2005; Fujii et al, 2005). Although FUS and TDP‐43 normally reside and function predominantly in the nucleus, pathological FUS and TDP‐43 inclusions are mostly observed in the cytosol, and inclusion‐bearing cells often show a reduction of nuclear staining (Arai et al, 2006; Neumann et al, 2006, 2009a; Kwiatkowski et al, 2009; Vance et al, 2009). It is completely unclear, how cytosolic FUS and TDP‐43 inclusions arise and apart from p62 and ubiquitin, no other cellular markers or co‐aggregating proteins have been detected within these inclusions (Neumann et al, 2006, 2007, 2009a; Kwiatkowski et al, 2009; Vance et al, 2009). As the inclusions occur predominantly in the cytosol, defects in nucleocytoplasmic transport or enhanced aggregation in the cytosol may lead to cytoplasmic mislocalization of TDP‐43 and FUS. This may interfere with their physiological nuclear function or cause a toxic gain‐of‐function because of excessive accumulation in the cytoplasm.
For TDP‐43, a classical NLS in the N‐terminal domain has been identified and experimentally confirmed (Winton et al, 2008). However, none of the over 30 mutations identified in TDP‐43 so far affect the NLS, and it is still unclear whether any of the mutations functionally affect nucleocytoplasmic transport. Moreover, expression of TDP‐43 in yeast and neuroblastoma cells suggested that the fALS‐associated mutations might increase the aggregation propensities of TDP‐43 rather than impairing its nuclear transport (Johnson et al, 2009; Nonaka et al, 2009). For FUS, it has been described that some of the fALS‐associated mutations in the C‐terminal region lead to an accumulation of the protein in the cytosol (Kwiatkowski et al, 2009; Vance et al, 2009). However, the underlying cellular mechanism is unknown, and it is not clear whether disturbed nuclear transport or aberrant cytoplasmic aggregation of mutant proteins leads to the cytosolic redistribution of mutant FUS. A non‐classical R/H/KX2−5PY‐NLS has been predicted in the FUS C‐terminal region (Lee et al, 2006). However, experimental evidence is missing that this sequence is required for nuclear import of FUS and the function of the predicted NLS is controversial. For example, a homologous motif in the related Ewing sarcoma protein (EWS), which belongs to the same transcription factor family (Law et al, 2006; Zakaryan and Gehring, 2006), was shown to be necessary, but not sufficient for nuclear import of EWS (Zakaryan and Gehring, 2006). Furthermore, an N‐terminal fragment, but not a C‐terminal fragment of FUS has earlier been shown to localize to the nucleus (Zinszner et al, 1997b). Nevertheless, 12 out of 22 FUS mutations identified in fALS patients are concentrated within the predicted NLS (Figure 1A) (Belzil et al, 2009; Chio et al, 2009; Kwiatkowski et al, 2009; Ticozzi et al, 2009; Vance et al, 2009; Corrado et al, 2010) and the motif is highly conserved during evolution (Figure 1B). We, therefore, assessed whether the predicted NLS was functionally relevant and whether it would be impaired by fALS‐associated FUS mutations.
C‐terminal region of FUS is necessary and sufficient for nuclear import
To test whether the C‐terminal domain of FUS is required for nuclear import, we generated a deletion mutant lacking the C‐terminal 13 amino acids (Δ514–526) (Figure 1A) and analysed its localization in HeLa cells. In contrast to wild‐type (WT) FUS, which was located almost exclusively in the nucleus, the deletion mutant showed a predominantly (72±4%) cytoplasmic localization (Figure 1C, see D for quantification). This shows that the C‐terminal domain (amino acids 514–526) is necessary for efficient nuclear import of FUS. We next investigated whether the three arginine residues most frequently mutated in FUS‐positive fALS patients (R521, R522, R524; see Figure 1A) are critical for nuclear import. When we changed these three arginines to alanine (R521A/R522A/R524A), the localization of the triple point mutant was indistinguishable from that of the deletion mutant (73±3% cytosolic; Figure 1C and D), showing that at least one of these arginines is essential for nuclear localization of FUS (for individual arginine mutations, see Figure 2). In addition, the two arginine residues further upstream (R514 and R518) are also required for nuclear import, as an R514A/R518A mutant showed a predominantly (65±5%) cytosolic localization (Figure 1C and D). These findings suggest that fALS‐associated mutations may affect a functionally active NLS located in the C‐terminus of FUS.
To test whether the C‐terminal tail of FUS is not only necessary but also sufficient for active nuclear import, we transplanted this domain onto the C‐terminus of the cytosolic reporter protein GST‐GFP. Owing to its size (55 kDa), the GST‐GFP fusion protein is largely excluded from the nucleus (Figure 1E) and requires active nuclear import mediated by an NLS (Iijima et al, 2006; Terry et al, 2007). Similar to the well‐characterized classical NLS of the SV40 large T antigen (SV40‐NLS) (Kalderon et al, 1984), the C‐terminal 13 amino acids of FUS (FUS514−526) mediated an almost exclusive nuclear localization of the reporter protein (Figure 1E, see F for quantification). In contrast, the same amino acids arranged in random order (FUS514−526 scrambled) were not able to mediate import of GST‐GFP (Figure 1E and F), showing that the FUS NLS activity requires a specific sequence motif rather than the random presence of several positively charged arginines. Taken together, these experiments demonstrate that the C‐terminal tail of FUS is necessary and sufficient for active nuclear import and thus constitutes a bona fide NLS.
fALS‐associated point mutations in the C‐terminal domain disrupt nuclear import of FUS
After functionally identifying the NLS of FUS, we analysed four fALS‐associated point mutations that occur within this domain at evolutionarily conserved residues (R521G, R522G, R524S, P525L; see Figure 1B) (Chio et al, 2009; Kwiatkowski et al, 2009) and asked whether they may disrupt nuclear import. Indeed, all four point mutations showed a varying degree of cytosolic accumulation, ranging from a mild mislocalization for R521G and R524S (16±10% and 21±7% cytosolic, respectively) over an intermediate phenotype for R522G (45±9% cytosolic) to a severe mislocalization for P525L (65±5% cytosolic) (Figure 2A, see B for quantification, note that mutations were ordered according to their strength/age of onset). R521H and R521C, the fALS‐associated FUS mutations identified most frequently (Belzil et al, 2009; Drepper et al, 2009; Kwiatkowski et al, 2009; Ticozzi et al, 2009; Vance et al, 2009; Corrado et al, 2010; Lai et al, 2010; Suzuki et al, 2010), also caused a mild nuclear import defect, similar to R521G (Supplementary Figure S1A and B). The fact that none of the point mutants are completely excluded from the nucleus is consistent with the finding that in fALS patients with FUS mutations neurons with cytoplasmic FUS inclusions still show some immunolabelling of the nucleus (Kwiatkowski et al, 2009; Vance et al, 2009; Rademakers et al, 2010). The observed cytoplasmic mislocalization cannot be attributed to a higher expression level of the point mutants, as similar expression levels were observed for all mutants (Figure 2C). Thus, R522 and P525 are important residues in the C‐terminal NLS of FUS, whereas R521 and R524 have a less important function for NLS activity. Another important residue is the highly conserved tyrosine at the C‐terminus, which is predicted to have an important function in a PY‐NLS (Lee et al, 2006), as its mutation to an alanine results in a dramatic relocalization of FUS (Supplementary Figure S1C and D). Furthermore, it is remarkable that the P525L and R522G mutations, which show the strongest degree of cytosolic mislocalization, were reported to cause especially aggressive forms of fALS, with a mean disease onset at 24 and 28.5 years, respectively (Chio et al, 2009; Kwiatkowski et al, 2009). Thus, the degree of cytosolic mislocalization is inversely correlated to the age of disease onset (Figure 2B).
As deletion or mutation of the C‐terminal NLS did not eliminate nuclear import completely (see Figures 1B and 2B), we investigated whether any of the point mutations located in the N‐terminal SYGQ‐ and G‐domains (Figure 1A) might affect a putative second NLS. However, all of the investigated N‐terminal point mutants (G156E, R216C, G225V, R234C, R244C) showed an almost exclusive nuclear localization, indistinguishable from the WT protein (Figure 2D, see E for quantification). Furthermore, N‐terminal point mutations did not further aggravate mislocalization of the P525L mutant (Supplementary Figure S2A and B), making it unlikely that they affect a second, weaker NLS, which might become important when the C‐terminal NLS is impaired. Together, these findings show that C‐terminal fALS‐associated FUS mutations affect the protein's major NLS and thus impair its nuclear import, whereas the so far identified N‐terminal FUS mutations do not impair nuclear localization. The latter suggests that these mutations cause disease through a different cellular mechanism.
FUS‐P525L mutation disrupts nuclear import in primary neurons and prevents import of a cytosolic reporter in vitro and in vivo
As fALS‐associated FUS mutations specifically cause degeneration of neurons in the cortex and spinal cord (Kwiatkowski et al, 2009; Vance et al, 2009), we wanted to confirm that these mutations also affect FUS import in neuronal cells. We, therefore, prepared primary neuronal cultures from embryonic (E19) rat hippocampus and frontal cortex and examined the subcellular localization of the WT protein and the strong P525L mutant on transfection of the corresponding cDNA constructs. As in HeLa cells, FUS‐WT was located almost exclusively in the nucleus, whereas the P525L mutant was strongly redistributed to the cytosol, extending even into the neuronal processes in both cortical and hippocampal neurons (Figure 3A, see B for quantification). Thus, fALS‐associated C‐terminal FUS mutations disrupt nuclear import not only in transformed cell lines, but also in primary neuronal cells, the cell type primarily affected in FUS‐opathies.
To further support that cytosolic mislocalization of the fALS mutants was due to a disrupted NLS activity, the P525L mutation was introduced into the GST‐GFP‐FUS514−526 reporter construct and analysed for its effect on nuclear import. Consistent with the data shown in Figures 2A and 3A, the proline mutation efficiently blocked nuclear import of the reporter protein, as its localization was indistinguishable from that of the FUS514−526 scrambled reporter (Figure 3C and D). This further supports that fALS‐associated point mutations disrupt the FUS514−526 NLS. To provide in vivo evidence for these findings, we injected the same reporter constructs into zebrafish eggs and analysed their subcellular localization in zebrafish embryos on day 2 post‐fertilization. In both myocytes and spinal cord neurons, the FUS514−526 WT sequence mediated efficient nuclear import, whereas reporter proteins carrying the P525L mutation or scrambled NLS remained cytosolic and were detectable in axonal processes of spinal cord neurons (Figure 3E). These data corroborate that fALS‐associated point mutations in the FUS‐NLS disrupt nuclear import in a living animal in different cell types, including spinal cord neurons.
Nuclear import receptor Transportin is required for nuclear import of FUS
Next, we searched for the cellular mechanism of nuclear transport affected by the fALS mutations described above. The NLS of FUS bears some homology to other NLSs with a PY motif, which have been shown to be recognized by the nuclear transport receptor Transportin (Trp), also known as Karyopherin β2 (Lee et al, 2006; Imasaki et al, 2007). As FUS has been shown to interact with Trp in in vitro pull down assays (Guttinger et al, 2004; Lee et al, 2006), we speculated that Trp may mediate nuclear import of FUS by binding to its C‐terminal NLS and that fALS‐associated point mutations might interfere with this import pathway. To test whether Trp is responsible for nuclear import of FUS, we performed siRNA‐mediated knockdown of the two Trp homologues, Trp1 and Trp2, two closely related proteins shown to have redundant function as nuclear import receptors (Guttinger et al, 2004; Rebane et al, 2004). Whereas knockdown of either Trp1 or Trp2 alone had not effect on nuclear import of FUS (data not shown), knockdown of both Trp variants significantly impaired nuclear import of endogenous FUS, as about 25% of the protein was found outside the nucleus on Trp1/2 silencing (Figure 4A, see B for quantification). The fact that substantial amounts of FUS are still detectable in the nucleus may be due to residual Trp1 remaining in siRNA‐transfected cells (Figure 4C, middle panel) or could indicate that additional import receptors are involved in nuclear import of FUS. To test whether Trp is the predominant import receptor for the C‐terminal NLS, we expressed the two FUS mutants R521G and P525L in cells that had been transfected with Trp1/2‐specific siRNAs. As a control, we examined the behaviour of HA‐tagged WT FUS. Similar to endogenous FUS (Figure 4A), HA‐tagged FUS‐WT showed a moderate redistribution to the cytoplasm on Trp knockdown (Figure 4D and E, see F for knockdown efficiency and HA‐FUS protein levels). Interestingly, the FUS‐R521G mutant, which showed only a mild cytosolic mislocalization in non‐siRNA transfected (Figure 2A) or control siRNA‐transfected cells (Figure 4D, upper panel), was strongly affected by Trp silencing, as >60% of the protein redistributed to the cytosol under these conditions (Figure 4D and E). This shows that the R521G mutation indeed interferes with the Trp pathway and strongly suggests that Trp is the major import receptor for the C‐terminal NLS of FUS. In line with this, the severe cytosolic mislocalization of the P525L mutant was only slightly, if at all, aggravated by Trp silencing (Figure 4D and E). This further corroborates that Trp is the major import receptor recognizing the FUS PY‐NLS.
The structures of several non‐classical PY‐NLSs bound to Trp have been solved and converge to a consensus‐binding geometry consisting of an N‐terminal hydrophobic/basic motif and the C‐terminal motif R/H/KX2−5PY (Lee et al, 2006; Cansizoglu et al, 2007; Imasaki et al, 2007). On the basis of these data, we modelled the FUS C‐terminal sequence as subtype RXXPY bound to Trp (Figure 4G). Interestingly, our model shows two distinct binding areas within this RXXPY motif: R522 makes strong charged H‐bond/ion‐pair interactions with the side chain carboxylates of E509 and D543 of Trp1. The second area of tight binding comprises residues P525 and Y526. As in the experimental structures, the proline allows a particular kinked main chain geometry between these two residues, enabling a specific surface recognition, consisting of a hydrophobic pocket that engulfs the proline side chain and the phenyl ring of Y526 and an H‐bond contact to Y526. In contrast, no specific interaction partner can be found for R521, resulting most likely in a somewhat flexible conformation, and only a weak attractive force between the receptor surface and R521 because of the negative electrostatic potential of the Trp surface in this region can be predicted. Similarly, R524 points away from the receptor surface and should not contribute significantly to recognition. Thus, our model is consistent with our analysis of the fALS‐associated FUS mutations (Figure 2A and B), as changes in amino acids that are responsible for strong contacts between the FUS C‐terminal sequence and Trp (R522 and P525) resulted in a severe cytosolic relocalization and early disease onset, whereas mutation of either R521 or R524 to glycine or serine only resulted in a mild mislocalization and later disease onset.
As siRNA‐mediated silencing of Trp did not result in a complete block of the Trp pathway (see residual Trp1 levels in Figure 4C and F) and only led to a moderate cytosolic redistribution of WT‐FUS, we searched for a more efficient way of blocking Trp‐mediated nuclear import. To this end, we took advantage of a Trp‐specific inhibitor peptide (M9M) designed to bind Trp with high affinity by joining the N‐terminal half of the hnRNP A1 NLS (called M9) with the C‐terminal half of the PY‐NLS of hnRNP M (Cansizoglu et al, 2007). By combining these two high‐affinity‐binding sites, the M9M peptide efficiently competes with natural substrates of Trp, such as hnRNP A1, HuR and hnRNP M (Cansizoglu et al, 2007). We reasoned that if Trp mediates nuclear import of FUS by binding to its C‐terminal PY‐NLS, the high‐affinity peptide inhibitor should compete for this interaction and thus should prevent or reduce nuclear import of endogenous FUS. Indeed, when we expressed a GFP‐M9M construct in cortical neurons (Figure 5A), hippocampal neurons (Supplementary Figure S3A) and HeLa cells (Figure 5B and C), the transfected cells showed a striking redistribution of endogenous FUS to the cytosol. In contrast, TDP‐43, which is imported through the classical NLS‐dependent importin α/β (karyopherin α/β1) pathway (Winton et al, 2008), was not affected by expression of the M9M inhibitor and remained nuclear even in cells with cytoplasmic FUS staining (Figure 5D). As an additional control, we expressed an importin α/β‐specific inhibitor construct (GFP‐Bimax2), designed to bind to importin α with high affinity (Kosugi et al, 2008). As expected, the importin α/β‐specific inhibitor interfered with the nuclear import of TDP‐43, but not of FUS (Supplementary Figure S3B). These data show that Trp, but not the classical import receptor importin α/β, is required for nuclear import of FUS. Taken together, our knockdown and competition experiments and structural modelling of the FUS PY‐NLS indicate that FUS is imported into the nucleus through the Trp receptor and suggest that the pathogenic mechanism underlying the C‐terminal FUS mutations is an impairment of Trp‐dependent nuclear import of FUS.
Redistribution of FUS into cytoplasmic stress granules
Interestingly, we noted that after GFP‐M9M expression, HeLa cells and primary neurons with cytosolic FUS redistribution often showed a punctuate localization pattern of FUS (magnifications in Figure 5A and B, see for quantification). As FUS is an RNA‐binding protein (Zinszner et al, 1997b), we wondered whether the observed puncta might be stress granules, cytoplasmic RNP structures that temporarily store translationally arrested mRNAs during cellular stress (Anderson and Kedersha, 2006). In addition to stalled mRNAs, stress granules contain characteristic proteins, such as proteins of the small ribosomal subunit, translation initiation factors including eIF3 and eIF4G and a large variety of RNA‐binding proteins, such as the PolyA‐binding protein (PABP‐1), the translational silencer T cell intracellular antigen‐1 (TIA‐1), TIA‐1‐related protein (TIAR) and the Ras‐GAP‐SH3‐binding protein (G3BP1). These proteins specifically associate with stress granules and not with other types of cytoplasmic RNA granules, such as processing bodies (P bodies) associated with mRNA decay (Kiebler and Bassell, 2006). Therefore, they are commonly used as specific markers of stress granules (Anderson and Kedersha, 2008). Co‐staining with antibodies specific to PABP‐1, TIA‐1, TIAR and G3BP1 showed that indeed FUS was localized to stress granules in GFP‐M9M‐transfected cells (Figure 6A, rows 1–4). In contrast, the P body‐specific marker protein Dcp1 did not co‐stain with FUS (Figure 6A, row 5), excluding that the FUS‐positive granules observed after GFP‐M9M expression correspond to P bodies. Furthermore, on addition of the polysome‐stabilizing drug cycloheximide, a well‐known inhibitor of stress granule assembly (Kedersha and Anderson, 2007), FUS remained diffusely distributed in the cytosol and no FUS‐positive G3BP1‐positive granules could be observed (Figure 6B). This confirms that the FUS‐positive granules observed after GFP‐M9M expression are stress granules. Together, our data suggest that cytoplasmically mislocalized FUS may be recruited into stress granules under conditions of cellular stress, such as strong inhibition of Trp‐dependent transport by GFP‐M9M over‐expression.
Pathologic inclusions in fALS and FTLD‐FUS patients contain marker proteins of stress granules
As stress granules dynamically grow and coalesce on prolonged stress exposure (Kedersha et al, 2000), we wondered whether the FUS‐containing stress granules might be related to the large FUS‐positive inclusions present in brains of fALS and FTLD‐FUS patients. We, therefore, analysed sections of post‐mortem brain and spinal cord tissue from an fALS case carrying an FUS‐R521C mutation and from cases with sporadic FTLD‐FUS, including aFTLD‐U (n=3), NIFID (n=3) and BIBD (n=1), by immunohistochemistry and double‐label immunofluorescence for the presence of the stress granule marker PABP‐1 in neuronal cytoplasmic inclusions (NCIs). Strikingly, all cases with FUS pathology revealed strong labelling for PABP‐1 in NCIs in affected brain regions such as spinal cord and hippocampus (Figure 7A, upper panels). This was further confirmed by double‐label immunofluorescence with anti‐p62 (green), a robust marker of NCIs in FUS‐opathies (Neumann et al, 2009a, 2009b; Munoz et al, 2009), and anti‐PABP‐1 (red), which showed a clear co‐localization in NCIs of all tested FUS‐opathy cases (Figure 7A, lower panels). Furthermore, NCIs in all cases examined showed an enrichment for another stress granule marker protein, eIF4G (Figure 7B). Notably, cases with FTLD‐TDP pathology (n=2) included as neurologic controls showed no staining of NCIs with PABP‐1 or eIF4G (Figure 7A and B). This suggests that co‐sequestration of stress granule‐associated proteins is a specific feature of FUS inclusions and that stress granule formation might be involved in inclusion body formation in FUS‐opathies.
C‐terminal fALS mutations favour recruitment of FUS into stress granules
It is important to note that in our cellular models FUS‐positive stress granules were observed after GFP‐M9M expression (Figure 6), but not on expression of mutations interfering with the NLS of FUS, despite their substantial redistribution to the cytosol (Figures 1 and 2). This suggests that cytosolic mislocalization by itself is not sufficient for the formation of FUS‐containing granules, but that additional cellular stress (e.g. strong inhibition of Trp‐dependent transport by GFP‐M9M expression) is required for this to occur. To test this hypothesis, we expressed FUS‐WT and the above‐examined C‐terminal FUS mutants (R521G, R522G, R524S, P525L) in HeLa cells (Figure 8A) and primary neurons (Figure 8B), subjected them to heat shock and analysed whether the proteins would be recruited to stress granules. In the absence of heat stress, the expressed proteins showed the above‐described localization pattern and no stress granules could be detected with a PABP‐1 or TIAR antibody (Figure 8A and B, left panels). In contrast, after heat shock, all FUS mutants localized to cytoplasmic stress granules (Figure 8A and B, right panels, for additional stress granule markers, see Supplementary Figure S4A), confirming our hypothesis that additional cellular stress is required for the recruitment of FUS into stress granules. The presence of cycloheximide during heat shock completely prevented the granular localization of the FUS‐P525L mutant (Supplementary Figure S4B), confirming that the FUS mutant‐containing granules observed after heat shock correspond to stress granules. Consistent with the GFP‐M9M experiment (Figure 5D) and the pathology data shown in Figure 7, TDP‐43 was not detectable in FUS‐P525L‐positive stress granules after heat shock (Figure 8C). Interestingly, FUS‐WT also remained exclusively nuclear after heat shock and was not detectable in cytoplasmic stress granules (Figure 8A and B, right panels). This suggests that only cytoplasmically mislocalized FUS is recruited into stress granules. In line with this, the amount of mutant FUS in stress granules correlated with the degree of cytoplasmic mislocalization of the point mutants and inversely with their age of disease onset (Figure 8A). This suggests that cytosolic mislocalization strongly facilitates the formation of FUS‐positive stress granules.
In conclusion, our data suggest that two pathological hits, namely cytosolic mislocalization of FUS and cellular stress, are required for the formation of FUS‐positive stress granules, both in peripheral and neuronal cells (Figure 9). Furthermore, recruitment of cytoplasmic FUS into stress granules might be an important cellular mechanism leading to FUS pathology, as stress granule marker proteins also co‐deposit with FUS in brains of fALS and FTLD‐FUS patients (Figure 7).
Our data show that the C‐terminal domain of FUS harbours the protein's major NLS, which mediates Trp‐dependent nuclear import. Furthermore, we show that several fALS‐associated point mutations in the C‐terminus of FUS disrupt this import mechanism, leading to cytoplasmic mislocalization of the protein. As the loss of nuclear import correlates with the age of disease onset of the individual point mutations, it seems likely that the nuclear import defect is causally linked to the disease. However, we could also show that additional cellular stress, such as heat shock, is necessary to cause a clustering of cytosolic FUS in the form of stress granules. As we consistently found the stress granule markers PABP‐1 and eIF4G co‐deposited with FUS inclusions in the whole spectrum of FUS‐opathies, including fALS‐FUS, aFTLD‐U, NIFID and BIBD, we implicate cellular stress in the pathogenesis of these FUS‐opathies and propose that two pathological hits, namely a nuclear import defect and cellular stress, are involved in the pathogenesis of FUS‐associated diseases.
The identified NLS in the C‐terminus of FUS corresponds to the class of non‐classical PY‐NLSs, which are recognized by the nuclear transport receptor Trp (Lee et al, 2006). PY‐NLSs typically are 20–30 residue signals with a C‐terminal R/H/KX2−5PY consensus motif preceded by a hydrophobic or basic motif, which both make important contacts with Trp (Lee et al, 2006; Imasaki et al, 2007). The C‐terminal tail of FUS fulfils the criteria of this consensus motif and indeed we could show that import of FUS depends on Trp (Figures 4 and 5). Our mutational analysis suggests that R522, P525 and Y526 are the important residues of the R/H‐X2−5PY consensus motif and that R514 and/or R518 may constitute the N‐terminal basic motif required for interaction with Trp (Figures 1 and 2; Supplementary Figure S1). This is supported by our three‐dimensional model of the FUS‐NLS bound to Trp (Figure 4G), explaining why mutations in residues R522, P525 and Y526 specifically affect Trp interaction and, therefore, show the most severe impairment of nuclear transport. To our knowledge, this is the first case in which mutations within a PY‐NLS cause a human disease. The only other known example of disease‐causing mutations in an NLS is Swyer syndrome, in which mutations in the classical NLS of SRY, the testes‐determining transcription factor encoded by the human Y chromosome, lead to a reduced activation of testes‐specific genes and thus male‐to‐female sex reversal (Li et al, 2001; Harley et al, 2003). Considering the large number of important nuclear proteins, it seems likely that other human diseases can be attributed to nuclear import defects because of mutations within different types of NLSs.
Our data show that mutations in R522 and P525 lead to a strong cytosolic mislocalization of FUS, however, without a complete exclusion of mutant proteins from the nucleus (Figure 2A and B). This is consistent with the observations that neurons of fALS patients with FUS mutations still show some immunolabelling in the nucleus (Kwiatkowski et al, 2009; Vance et al, 2009; Rademakers et al, 2010). Although the reported number of patients is still small, it is striking that mutations in R522 and P525 cause an especially aggressive form of ALS. The five reported patients with a P525L mutation all had a very early onset of disease (mean age of onset: 24 years) and an unusually rapid disease progression, with death in <12 months (Chio et al, 2009; Kwiatkowski et al, 2009). R522G (n=2) was the second most aggressive mutation reported by Kwiatkowski et al (2009) (mean age of onset: 28.5 years; mean duration: 25 months). Thus, the two mutations that showed the most severe nuclear import defect (Figure 2A and B) and were most readily recruited into stress granules on heat shock (Figure 7A) caused the most aggressive disease course of all reported FUS mutations. Furthermore, the R521G (n=19) and R524S (n=1) mutations, which showed a much weaker import defect (Figure 2A and B) and were recruited into heat‐induced stress granules to a lesser degree (Figure 7A), on average caused disease at a much later age (average age of onset: 43 and 34 years, respectively). Interestingly, the weak R521G mutation showed incomplete penetrance, as two members of an R521G family lived well past the average age of onset without developing disease (Kwiatkowski et al, 2009). This is consistent with our hypothesis that environmental stress contributes to disease initiation. Different exposure of individuals to environmental stress might also explain the variation in the age of onset reported for other fALS‐associated FUS mutations (Kwiatkowski et al, 2009). Although the number of FUS mutation carriers reported to date is still small, Alzheimer's disease (AD) research has shown that the in vitro effects of presenilin mutations can be clearly correlated with the age of disease onset (Duering et al, 2005; Page et al, 2008), and in some cases, cell culture experiments have even predicted familial AD mutations and their disease onset. Given the severe effect of the Y526A mutation on nuclear import of FUS, it seems possible that fALS patients carrying FUS‐Y526 mutations may be identified in the near future and one would predict a similarly early age of onset as for the P525L mutation. Consistent with this hypothesis, a novel frameshift mutation, which leads to a premature stop codon and thus truncation of the C‐terminal 60 amino acids of FUS, was recently reported with a disease onset of 20 years (DeJesus‐Hernandez et al, 2010).
In addition to mutations in the C‐terminal tail of FUS, several point mutations have been identified in the N‐terminal SYGQ‐ or G‐rich domains of FUS (Belzil et al, 2009; Kwiatkowski et al, 2009; Ticozzi et al, 2009; Corrado et al, 2010) (Figure 1A). As deletion of the C‐terminal NLS did not completely eliminate nuclear import (Figure 1C and D) and an N‐terminal fragment comprising the SYGQ‐ and G‐domains was described to localize to the nucleus (Zinszner et al, 1997a, 1997b), it seemed possible that the N‐terminal fALS‐associated mutations disrupt a putative second NLS. However, our mutational analysis (Figure 2D and E; Supplementary Figure S2A and B) suggests that N‐terminal fALS mutations act through a different cellular mechanism. As the SYGQ‐ and G‐rich domains are involved in transcriptional activation and have been shown to interact with RNA polymerase II and various transcription factors (Zinszner et al, 1994; Yang et al, 2000; Law et al, 2006), the mutations instead might modulate FUS function as a transcriptional activator, either impairing or aberrantly activating transcription of neuronal target genes. The finding that different clusters of mutations within one gene may affect different cellular mechanisms is not too surprising, given the fact that familial AD‐causing mutations in three different regions within and around the amyloid β (Aβ) peptide domain have fundamentally different consequences on Aβ metabolism and aggregation (Haass, 2004).
Strikingly, we found a co‐deposition of cytosolic FUS with various stress granule marker proteins in cells subjected to cellular stress as well as in brains of fALS‐FUS and FTLD‐FUS patients. However, some differences exist between the FUS‐positive stress granules observed in our cellular models (primary neurons and HeLa cells) and the FUS inclusions present in fALS and FTLD‐FUS patients: GFP‐M9M‐expressing cells or cells exposed to heat shock contain multiple small FUS granules in the perinuclear region or in neuritic processes (Figures 6 and 8). In contrast, NCIs in brains of FTLD‐FUS patients are usually much larger (see Figure 7; Neumann et al, 2009a). However, the granules in our in vitro cultures and the inclusions in brains of patients also share important properties: first, they resemble each other ultrastructurally, as both stress granules and inclusions in BIBID and NIFID patients appear as fibrillogranular aggregates in electron microscopy (Munoz‐Garcia and Ludwin, 1984; Mosaheb et al, 2005; Souquere et al, 2009). A second common property is their protein composition: they both contain FUS and the stress granule markers PABP‐1 and eIF4G, but not TDP‐43. Interestingly, before the discovery of FUS in ALS and FTLD, Fujita et al (2008) described that the basophilic inclusions in BIBD patients contain markers of stress granules. Stress granules are known to be dynamic entities that can enlarge and coalesce on prolonged stress exposure (Kedersha et al, 2000). Hence, it seems possible that in the presence of chronic stress, such as oxidative stress, viral infections or proteasome inhibition (Anderson and Kedersha, 2008), small stress granules give rise to larger granules and eventually to large, insoluble inclusions. It will, therefore, be interesting to see if indeed RNA and further RNA‐binding proteins commonly found in stress granules in in vitro cultures are present in FUS inclusions in fALS and FTLD‐FUS patients.
Several reports recently described that TDP‐43 can be found in stress granules under various experimental conditions (Colombrita et al, 2009; Moisse et al, 2009; Volkening et al, 2009; Freibaum et al, 2010). First, sciatic axotomy in adult mice was reported to cause a marked increase of cytoplasmic TDP‐43 and its co‐localization with the RNA‐binding proteins Staufen and TIA‐1 (Moisse et al, 2009). In cellular models, oxidative stress and proteasome inhibition were reported to lead to a partial recruitment of TDP‐43 into stress granules (Colombrita et al, 2009; Freibaum et al, 2010), and Freibaum et al identified numerous components of stress granules as TDP‐43‐interacting proteins (Freibaum et al, 2010). Although our data clearly show that the FUS‐containing stress granules observed after GFP‐M9M expression or heat shock do not contain TDP‐43 (Figures 5D and 8C), this might be explained by the fact that TDP‐43 was predominantly nuclear under these experimental conditions. According to our two hit model, it seems possible that similar to FUS, cytosolic relocalization of TDP‐43 is a prerequisite for efficient stress granule recruitment. Thus, it can be speculated that in the studies mentioned above, TDP‐43 was at least partially mislocalized to the cytosol before cellular stress.
What remains controversial is the data on stress granule markers in patients with TDP‐43 pathology. One study reported a stronger TIA‐1 (stress granule) and XRN‐1 (P body) staining in sALS patients compared with healthy controls (Volkening et al, 2009), whereas another study found a lack of stress granule markers in TDP‐43 inclusions of sALS patients (Colombrita et al, 2009). Consistent with the latter study, inclusions of FTLD‐TDP patients were consistently negative for the stress granule markers PABP‐1 and eIF4G in our study (Figure 7). Further studies with larger number of patients are needed to clarify this issue.
Although we did not observe HA‐tagged WT‐FUS (Figure 8A and B) or endogenous FUS (data not shown) in stress granules on heat shock, it is possible that very small amounts of FUS, undetectable by our antibodies, are present in stress granules even in untransfected or FUS‐WT‐transfected cells. That this might be the case is suggested by the report of Andersson et al (2008), showing a recruitment of endogenous FUS to stress granules on exposure of cell lines to oxidative stress. Nevertheless, our analysis of fALS‐associated FUS mutations strongly suggests that cytosolic mislocalization of FUS favours recruitment of the protein into stress granules, as the degree of stress granule recruitment correlated with the cytoplasmic mislocalization of the individual point mutations (Figure 8A). Thus, our data suggest that cytosolic FUS is recruited to stress granules, and furthermore, imply that an increased presence of FUS in the cytosol favours the formation of FUS‐positive stress granules. In the light of these data, one may postulate that at least small amounts of FUS have to accumulate in the cytosol to trigger the onset of sporadic FTLD‐FUS. As not only fALS patients with FUS mutations but also sporadic FTLD‐FUS patients show a neuronal cytoplasmic redistribution of FUS (Munoz et al, 2009; Neumann et al, 2009a, 2009b) and a co‐deposition of FUS and stress granule marker proteins in NCIs (this study), the question arises what causes the abnormal cytoplasmic accumulation of FUS in cases without FUS mutations. Subtle alterations in the Trp pathway, for example caused by reduced Trp expression or post‐translational modifications of FUS, might lead to an increase of FUS in the cytosol even in the absence of FUS mutations. Indeed, weak cytosolic FUS staining is consistently observed in post‐mortem brain tissue from healthy controls (Neumann et al, 2009a), indicating that cytosolic FUS may accumulate during ageing. In combination with environmental stress, a small increase in cytosolic FUS may then be sufficient to initiate clustering of FUS in stress granules and eventually larger inclusions (Figure 9).
Materials and methods
Cell culture and transfection
Human cervical carcinoma cells (HeLa) were cultured in Dulbecco's modified Eagle's medium with Glutamax (Invitrogen) supplemented with 10% (vol/vol) foetal calf serum (FCS, Invitrogen) and penicillin/streptomycin (PAA). Transfection of HeLa cells was carried out with Fugene 6 (Roche) or Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Hippocampal and cortical neurons were cultured from embryonic day 19 rat embryos as described earlier (Tada et al, 2007). Neurons were transfected on day in vitro (DIV) 5 using Lipofectamine 2000 (Invitrogen) and were analysed on DIV 7. When mCherry was used as a filler to visualize neuron morphology, the mCherry:HA‐FUS DNA ratio was 1:10.
Antibodies and inhibitors
A complete list of all antibodies used can be found in the Supplementary data. Cycloheximide (Sigma) was used at a concentration of 20 μg/ml.
cDNA constructs and primers
The cDNA sequence of human FUS (NM_004960) was amplified from a human brain cDNA library and was cloned by XhoI/BamHI restriction digest into the pcDNA3.1/Hygro(−) vector (Invitrogen) with an N‐terminal HA‐tag. HA‐FUS‐P525L was generated from the FUS‐WT construct by QuikChange mutagenesis (Stratagene), all other FUS mutations were introduced through conventional PCR primers and PCR products were cloned into pcDNA3.1/Hygro(−) through restriction digest. The pGST‐EGFP‐C1 vector was generated by inserting the GST sequence with a Kozak sequence into the NheI/AgeI restriction sites of the pEGFP‐C1 vector (Clontech). NLS reporter constructs were generated by ligating annealed oligos into the XhoI/BamHI restriction sites of pGST‐EGFP‐C1. The nuclear import inhibitor constructs (GFP‐M9M, GFP‐Bimax2) were generated by ligating annealed oligos into the XhoI/BamHI restriction sites of pEGFP‐C1. For all constructs, sequence integrity was verified by sequencing. Oligonucleotides sequences are available on request.
Zebrafish husbandry and embryo injection
All of the experiments were performed in compliance with the guidelines of the German Council on Animal Care. WT AB zebrafish were kept at 28°C and raised and mated as described (Mullins et al, 1994). GST‐EGFP constructs (25 ng/μl) were injected into fertilized eggs at the one‐cell stage; 48 h old embryos were stained as described in the Supplementary data.
Human post‐mortem tissue
Cases with confirmed FUS pathology used in this study have been earlier described and included a case of fALS with an FUS‐R521C mutation (Rademakers et al, 2010) and cases of aFTLD‐U (n=3) (Neumann et al, 2009a), NIFID (n=3) (Neumann et al, 2009b) and BIBD (n=1) (Munoz et al, 2009). In addition, FTLD‐TDP (n=2) as neurologic controls and cases with no history of neurologic diseases (n=2) were included.
Immunocytochemistry and immunohistochemistry
For immunocytochemistry on HeLa cells and neurons, cells were fixed for 15 min in 4% paraformaldehyde in PBS, permeabilized for 5 min in 0.2% Triton X‐100 with 50 mM NH4Cl and subsequently blocked for 20–30 min in blocking buffer (5% goat serum or 2% BSA in PBSS=PBS with 0.1% saponin). Cells were stained with the indicated primary and secondary antibodies diluted in blocking buffer for 30 min and were washed 5 × in PBSS. To visualize nuclei, cells were stained with TO‐PRO‐3 iodide (Invitrogen) for 15 min and were washed 3 × in PBS. Coverslips were mounted onto glass slides using ProLong Gold Antifade Reagent (Invitrogen). All steps were carried out at RT.
Immunohistochemistry on human post‐mortem material was performed on 5 μm thick sections of formalin fixed, paraffin‐embedded tissue from spinal cord, medulla or hippocampus with the indicated antibodies (after microwave antigen retrieval) and the avidin–biotin complex detection system (Vector Laboratories) with 3,3′‐diaminobenzidine as chromogen. Double‐label immunofluorescence for PABP‐1 and p62 (after microwave antigen retrieval) was performed using Alexa‐488 and ‐594‐conjugated secondary antibodies; 4′‐6‐diamidino‐2‐phenylindol (Vector Laboratories) was used for nuclear counter‐staining.
Image acquisition and quantification
Confocal images of HeLa cells, primary neurons and zebrafish embryos were obtained with an inverted laser scanning confocal microscope (Zeiss Axiovert 200M) with a 63 × /1.4 NA oil immersion lens, using a pinhole diameter of 1 Airy unit. Pictures were taken and analysed with the LSM 510 confocal software (Zeiss), and, if necessary, for printing, brightness and contrast were linearly enhanced using the LSM image browser (Zeiss). For HeLa cells, single confocal images were taken in the plane of the largest cytosolic area. For neuronal cultures and zebrafish embryos, a series of images along the z axis was taken and projected into a single image using the maximal projection tool of the LSM 510 software. Immunofluorescence images of brain sections were obtained by wide‐field fluorescence microscopy (BX61 Olympus with digital camera F‐view, Olympus).
Nuclear and cytosolic localization was quantified with the LSM 510's co‐localization tool as follows: total fluorescence intensities of the green channel were calculated from the mean fluorescence intensity and the number of pixels. Pixels that were co‐localized with TO‐PRO‐3 were considered ‘nuclear’ and pixels that did not overlap with TO‐PRO‐3 were considered ‘cytosolic’. For each sample, 7–12 randomly selected fields were analysed, containing a total of 50–100 transfected cells. Means across all fields were calculated and s.d. are indicated by error bars. Pictures and quantification shown are from one experiment, but are representative of several experiments.
siRNA‐mediated knockdown of Trp1 and Trp2
Trp1/2 knockdown was achieved using two different siRNA pools: Trp1/2 pool #1 consisted of the Trp1‐specific siGENOME siRNA D‐011308‐01 (target sequence: 5′‐guauagagaugcagccuua‐3′) and the Trp2‐specific siGENOME SMART pool M‐020491‐01 (target sequences: 5′‐gggcagagaugcagccuua‐3′; 5′‐gcaguucucugagcaauuc‐3′; 5′‐aaacaggagugucucaaca‐3′; 5′‐gcgcugauggacaauauug‐3′), both from Dharmacon. Trp1/2 pool #2 consisted of the Trp1‐specific siGENOME siRNA D‐011308‐04 from Dharmacon (target sequence: 5′‐caauuggucgucuugguua‐3′) and the Trp2‐specific siGENOME SMART pool M‐020491‐01 (see above). A non‐targeting (NT) siRNA (ON‐TARGET plus NT siRNA #3, D‐001810‐03 from Dharmacon) was used as a negative control. Cells were reverse transfected using a total of 50 pmol siRNA and 5 μl Lipofectamine 2000 (Invitrogen) per six well. Medium was changed 4–6 h post‐transfection and effect of knockdown was analysed 48–72 h post‐transfection.
Cell lysates and immunoblotting
Cells were washed twice in PBS, scraped off and pelleted at 1000 g, 5 min. Total cell lysates were prepared in ice cold RIPA buffer freshly supplemented with complete EDTA‐free protease inhibitor cocktail (Roche). After 15 min lysis on ice, lysates were sonicated in a bioruptor (Diagenode, 45 s on high) and protein concentration was determined by BCA protein assay (Pierce); 4 × SDS–PAGE sample buffer was added and samples were boiled for 5 min. Proteins were separated by SDS–PAGE, transferred onto a PVDF membrane (Immobilon‐P, Millipore) and analysed by immunoblotting using the indicated antibodies. Bound antibodies were detected with the chemiluminescence detection reagents ECL (Amersham) or Immobilon (Millipore).
The C‐terminal amino‐acid residues D520–Y526 of FUS were modelled manually into the Trp‐binding pocket. Residues of the non‐classical NLSs of subtype RXXPY (pdb‐entries 2OT8 (Cansizoglu et al, 2007) and 2Z5K (Imasaki et al, 2007) differing between the experimental structures and the FUS C‐terminal sequence were exchanged obeying standard amino‐acid conformations (Engh and Huber, 1991) and the resulting structure was locally energy minimized using MAIN (Turk, 1992), keeping the Trp molecule rigid.
Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).
Conflict of Interest
The authors declare that they have no conflict of interest.
We thank Drs Harald Steiner, Richard Page and Stefan Lichtenthaler for critical discussion, Dr Sven Lammich for providing human brain cDNA and the Hans and Ilse Breuer Foundation for the confocal microscope. This work was supported by the Center for Integrated Protein Science Munich (CIPSM), the Competence Network of Neurodegenerative Diseases (KNDD) of the Bundesministerium für Bildung und Forschung (BMBF), the Sonderforschungsbereich Molecular Mechanisms of Neurodegeneration (SFB 596) and an EMBO postdoctoral fellowship (to DD). DE is supported by the Helmholtz Young Investigator Program. MN is supported by the Stavros‐Niarchos Foundation and the Synapsis Foundation. IRAM is supported by the Canadian Institutes of Health Research and the Pacific Alzheimer's Research Foundation. CH is supported by a ‘Forschungsprofessur’ of the Ludwig‐Maximilians University.
- Copyright © 2010 European Molecular Biology Organization