RNA Seeds Higher Order Assembly of FUS Protein

INTRODUCTION

Eukaryotic cells have many examples of self-assembled bodies, which are concentrated protein structures not bound by lipid membranes. Examples in the cytoplasm include P-bodies and stress granules (Decker and Parker, 2012). Nuclear assemblies include the nucleolus, Gemini of coiled bodies, Cajal bodies, histone locus bodies, PML bodies, paraspeckles and splicing speckles (Nizami et al., 2010). These structures are rich in RNA-binding proteins, and in some cases their assembly may be nucleated by RNA itself (Kaiser et al., 2008).

Some RNA-binding proteins that comprise self-assembled structures contain a Low Complexity (LC) domain, which in isolation has the ability to form amyloid-like structures (Sun et al., 2011; Han et al., 2012; Kato et al., 2012). One such protein is FUS, an abundant nuclear protein that affects multiple levels of RNA biogenesis including transcription, splicing, and mRNA transport (Yang et al., 2000; Wang et al., 2008; Polymenidou et al., 2012). Mutations in FUS cause 5% of familial and rarely sporadic ALS, a neurodegenerative disease leading to death of motor neurons (Kwiatkowski et al., 2009; Vance et al., 2009; Polymenidou et al., 2012). At autopsy, large cytoplasmic aggregates stain positive for FUS in the motor neurons of ALS patients with these mutations.

In cells, FUS binds RNAP2 (RNA polymerase II) and affects transcription (Yang et al., 2000; Das et al., 2007; Wang et al., 2008; Schwartz et al., 2012; Tan et al., 2012). The loss of FUS or overexpression of FUS leads to altered gene expression for thousands of genes in many different cell types (Hoell et al., 2011; Ishigaki et al., 2012; Lagier-Tourenne et al., 2012; Schwartz et al., 2012; Tan et al., 2012). In our previous work, we characterized FUS binding to RNAP2 through its C-terminal domain (CTD) (Schwartz et al., 2012). FUS binds RNAP2 at the transcription start sites of more than 50% of expressed genes in HEK293T/17 cells and orchestrates Ser2 phosphorylation. Purified FUS binds the CTD in an RNA-dependent manner (Schwartz et al., 2012).

Here we investigate the mechanism by which RNA potentiates the FUS-CTD interaction. We show that RNA stimulates the formation of higher order FUS structures. We isolate two domains within FUS that contribute to higher order assembly. Finally, the CTD of RNAP2 binds these FUS structures, suggesting a new model for RNA-dependent recognition of the CTD by FUS.

RESULTS

Highly Cooperative Binding of FUS to Multiple RNA Sequences

Although previous studies identified RNA sequence/structure motifs that bind FUS (Lerga et al., 2001; Hoell et al., 2012; Ray et al., 2013), any comprehensive model of FUS function must account for the broad RNA binding seen in CLIP-seq experiments (Schwartz et al., 2012). We therefore tested a 48 nt (nucleotide) RNA from the promoter of gene DNMT3b; this prD RNA, one of many sequences identified previously (Schwartz et al., 2012), contains none of the published FUS-binding motifs (Supp Fig 1A). Purified, monomeric FUS (Supp Fig 1B) bound this RNA with reasonably high affinity (K_d ^{= 97 ± 2 nM) and high cooperativity (apparent Hill coefficient} n = 4.0 ± 0.2) (Fig 1A). Because n is the minimum number of interacting binding sites in a system showing positive cooperativity, the simplest conclusion is that at least four FUS proteins are bound to prD RNA in complex 3.

Figure 1

FUS Binds RNA in a Highly Cooperative but Not Highly Sequence-specific Manner

(A) Left, trace prD RNA was incubated with increasing concentrations of full-length MBP-FUS protein (0 to 1200 nM) and binding analyzed by EMSA. Right, quantification of F_bound = RNA in complexes/total RNA in lane, as a function of FUS concentration. Data fit with a binding curve, Hill coefficient n = 4. Error bars (which are small) represent the range of data from two independent experiments. (B) EMSA experiments for the RNA variants shown in Supp Fig 1A, all binding to full-length FUS. Uncertainties listed for K_d ^and n represent the range of 2 or more replicates. (C) Domain structure of FUS protein and truncated versions of the protein that were tested for RNA binding. (D) Summary of RNA-binding data for truncation mutants of FUS. Uncertainties represent the range of 2 or more replicates. The LC domain bound RNA so weakly that binding parameters were not determined (N.D.). Samples of primary data are shown in Supplemental Fig. 2.

To assess whether some novel sequence/structure motif in prD RNA was responsible for its binding to FUS, we first replaced either its 5’ half or its 3’ half with poly(A) (Fig 1B). These 48 nt RNAs bound FUS with affinities similar to that of the parental prD RNA, indicating that the stem-loop structure in the 5’ half of prD was not necessary for binding. Poly(A) itself bound FUS weakly (Fig. 1B). Furthermore, three RNAs with abundant secondary structure – two group I intron domains (Bp P456 and Pne P456) and the Medaka telomerase RNA (Med TR) – all bound FUS with good affinity (K_d ^{= 46 to 111 nM) and high cooperativity (}n = 3 to 4) (Supp Fig 1B). GGUG RNA, containing a proposed FUS-binding motif (Lerga et al., 2001), and its negative control CCUC RNA bound with lower affinity (Supp Fig 1C). In conclusion, FUS binding is not highly specific for RNA sequence or structure.

The electrophoretic mobilities of the FUS-RNA complexes gave additional information. When poly(A) was substituted for either half of prD, binding was less cooperative (n = 1.7 to 2.0) and the intermediate Complex 1 accumulated much more than with prD RNA (Fig. 1B). This suggests that when half of prD is attached to the weak-binding poly(A) sequence, the prD portion binds two FUS molecules cooperatively (Complex 1) and only at high [FUS] does FUS also bind the poly(A) segment to give Complexes 2 and 3. The observation of such discrete complexes indicates well-folded protein. We conclude that diverse RNA sequences can bind multiple FUS proteins in a highly cooperative manner.

RNA Interactions Promote FUS Binding to the CTD of RNAP2

We investigated the RNA dependence of FUS binding to the CTD of RNAP2. Titration of the prD RNA showed that FUS pulled down with GST-CTD at a half-maximum RNA concentration similar to the K_d ^{of this RNA for FUS (}Fig 2A). Multiple RNAs with sequences from gene promoters bound by FUS in HEK293T/17 cells, along with a variety of other noncoding RNAs, also promoted FUS pulldown with GST-CTD (Fig 2B). In multiple experiments, similar CTD binding occurred with MBPFUS, containing an N-terminal maltose binding protein solubility tag, as with H6-FUS (data not shown). Not all RNAs were able to promote FUS-CTD binding. Non-human sequences including the P456 domain of the Tetrahymena group I intron were active. Interestingly, a point mutant in P456 disrupting the RNA tertiary structure abolished its ability to promote FUS to bind the CTD. Despite showing no activity in promoting FUS-CTD interaction, RNAs Bp P456, Pne P456, and Med TR had high affinity for FUS binding. Thus, it appears that some RNA sequence/structure may be necessary for FUS to bind the CTD at the low concentrations tested.

Figure 2

RNA Promotes FUS Binding to the CTD of RNAP2

(A) Left, a representative western from pulldown of MBP-FUS by GST-CTD, stimulated by titrating in the prD RNA. Right, MBP-FUS binding to GST-CTD as a function of prD RNA concentration averaged from 3 replicates (error bars represent standard error). (B) The ability of various RNAs to promote FUS binding to the CTD, including sequences from FUS-regulated gene promoters (CIT, CYC1, ExoSC4, MCM7), RMRP, telomerase RNAs (yeast mini-TLC1, Tetrahymena (Tet TR), zebrafish (Zeb TR), or medaka (Med TR)) (Xie et al., 2008), and various group I intron domains from Tetrahymena (P456 with defined tertiary structure, Bp P456 with disrupted tertiary structure, P3P9) or Pneumocystis (carinii) jirovecii (Pne P456). The CIT RNA did not promote FUS to bind GST alone (left lane) Sequences GGUG and CCUC (Lerga et al., 2001) were too small to visualize on this Sybr stained gel. (C) Binding curve of GST-CTD for the prD RNA shows that CTD does not bind RNA. (D) EMSA of FUS for RNA with (dashed line) or without (dotted curve) the presence of 1 μM GST-CTD reveals no change in the K_d ^{of FUS for RNA.}

One simple model for the FUS-RNA-CTD interaction would involve the CTD binding directly to RNA (Kaneko et al., 2005) and the RNA then bridging FUS and CTD. However, the CTD itself showed no stable binding to prD RNA (Fig 2C). Alternatively, the CTD could simultaneously contact RNA and FUS, in which case the CTD should lower the K_d of FUS for RNA. However, FUS-RNA binding curves were unchanged in the presence of CTD (Fig. 2D, Supp Fig 2).

RNA Seeds FUS Assembly Formation

FUS has some ability to form higher order assemblies without any RNA, especially at high concentrations (~100 μM) (Sun et al., 2011; Han et al., 2012; Kato et al., 2012). We found that assemblies were visible as an opalescent turbidity in solution, which could be quantified by measuring the absorbance at 405 nm. Addition of prD RNA promoted more assembly formation (Fig 3A). Low concentrations of FUS in the presence of RNA were able to form assemblies to the same extent as higher concentrations of FUS without RNA. High concentrations of FUS upon addition of RNA eventually precipitated out of solution, preventing measurement (vertical dashed line in Fig 3A).

Figure 3

RNA Seeds FUS Assembly

(A) FUS protein at various concentrations was incubated with or without RNA and absorbance at 405 nm was measured (turbidity). Dashed line indicates that at higher RNA concentrations, 25 μM and 50 μM FUS samples precipitated in a large opalescent gel, which prevented turbidity measurement. (B) FUS incubated with or without 3 μM prD RNA in PBS pH 6.0 and crosslinked with formaldehyde. Protein assemblies were resolved by SDS-PAGE and stained with Coomassie Blue. (C) FUS was boiled at 95 °C for 5 min and treated with formaldehyde as in B. (D) Domain LC or heat denatured LC (Heat DN) was crosslinked with UV and resolved by SDS-PAGE. (E) The RNA-binding domains del-LC and RGG-Zn-RGG (20–26 μM) were incubated with various concentrations of prD RNA and crosslinked with formaldehyde. Numbers below lanes indicate fraction of the total protein that was crosslinked.

Using formaldehyde crosslinking, we found that addition of prD RNA promoted FUS assembly with 70% of protein crosslinked in assemblies (Fig 3B). A distinct ladder consistent with up to five FUS proteins could be resolved by SDS-PAGE (Supp Fig 3A). Single-stranded DNA with the same sequence as the prD RNA also promoted FUS assembly (Supp Fig 3B). Furthermore, crosslinking FUS in the presence of RNA followed by nuclease treatment did not destroy the complexes, suggesting that they involve protein-protein crosslinks and not merely protein-RNA crosslinks (Supp 3C). No FUS crosslinking was observed if FUS was first heat denatured at 95 °C, indicating that FUS structure is required for assembly (Fig 3C).

Formaldehyde crosslinking failed to detect LC domain assemblies. This is likely due to the fact that LC contains no lysines. Therefore, we crosslinked LC assemblies using UV irradiation. The amount of LC domain assemblies detected was reduced following heat denaturation of the protein (Fig 3D). This suggested that LC possesses some structure required for assembly formation and this assembly does not require RNA.

Next, we tested the RGG-Zn-RGG domain and the del-LC domain for their ability to assemble. Between 30 and 40% of del-LC and of RGG-Zn-RGG (Supp Fig 3D and 3E) were crosslinked in higher order assemblies without RNA, whereas 80% of the protein was in higher order assemblies in the presence of the prD RNA or ssDNA. Other RNA sequences tested (Supp Fig 1A) also promoted assembly formation to a degree consistent with their measured K_d ⁽Fig 1B).

Substoichiometric levels of RNA promote assembly formation. RNA at concentrations 9- or 25-fold below that of the del-LC or RGG-Zn-RGG protein promoted the majority of the protein to form assemblies (Fig 3E, note Supp Fig 3F for no cross-link control). At an RNA/protein ratio of 0.11, even with four FUS bound to each RNA, more than half of the FUS would be non-RNA-bound. Stoichiometric levels of RNA (the same concentration as the protein or higher) reduced assemblies to a level less than those seen in the absence of RNA (Fig 3E). We performed the same experiments for full-length FUS and found a similar trend for RNA-dependent seeding and then inhibition of protein assembly at high [RNA] (Supp Fig 3G).

FUS Assemblies Bind the CTD of RNAP2

FUS assemblies were large enough to be visualized by light microscopy. Assemblies adhered to glass ranged from 1 to more than 5 μm across and appeared ropey (Supp Fig 4A). In solution, they adopted a spherical shape consistent with phase transitions previously reported for protein hydrogels (Li et al., 2012) (Supp Fig 4B). Protein constructs containing the assembly domains of FUS also readily formed microscopic hydrogels. These hydrogel assemblies weakly stained for the amyloid-specific fluorescent dye thioflavin T (ThT), suggesting the presence of some cross-β structure. ThT stains RNA so all assemblies, except LC, stained brightly for ThT in the presence of RNA (Supp Fig 4C). FUS is known to bind tightly to its own LC domain (Wang et al., 2008; Kato et al., 2012); therefore, we tested whether this domain would bind these FUS assemblies. We observed LC to remain stably associated with these assemblies after more than 48-hour incubation with fresh PBS (Supp Fig 4D).

We then tested the ability of the CTD of RNAP2 to associate with these assemblies. FUS assemblies were allowed to form at high concentrations without addition of RNA. GFP-tagged CTD (GFP-CTD) was incubated with the assemblies (Fig 4A). Two days after fresh buffer exchange, GFP-CTD remained bound to the FUS assemblies (Fig 4B, grey bars). Negative control protein (rabbit IgG) did not associate with FUS assemblies, even when monitored as early as two hours after changing the buffer (Fig 4B, black bars). Finally, LC-RRM and del-LC assemblies were incubated with GFP-CTD. After 48 hours, GFP-signal was still detected above background. The signal normalized to FUS signal was higher for the del-LC construct than for the LC-RRM (Fig 4B), suggesting that CTD associated more stably with the RGG-Zn-RGG domain of FUS.

Figure 4

Assemblies of FUS Bind the CTD of RNAP2

(A) FUS assemblies were allowed to form at high concentration without RNA and spotted on glass coverslips. Assemblies were visualized by phase-contrast (Bright, left) and fluorescence microscopy (right). Bar indicates 5 μm. FUS was visualized by immunofluorescence with anti-FUS antibodies. GFP-tagged CTD was incubated with FUS assemblies and remained bound up to 48 hours after incubation in PBS. (B) Following a 48-hour incubation, the signal from GFP-CTD or negative control Alexafluor633-conjugated rabbit IgG was measured and normalized by the signal for FUS, del-LC, or LC-RRM for each assembly (n > 30). Error bars represent standard deviation. (C) FUS (60 μM) assemblies were formed in the presence (left) or absence (right) of RNA (3 μM). Fluorescence indicates binding of GFP-CTD. Solid white scale bar indicates 10 μm. (D) MBP-tagged domains LC and RGG-Zn-RGG were stained with 1% uranyl acetate and imaged by transmission electron microscopy. LC and RGG-Zn-RGG both form polymeric fibers 70 to 150 nm long and either 30 nm (LC) or 10 nm (RGG-Zn-RGG) in diameter. (E) Untagged FUS was incubated with or without prD RNA and imaged by TEM. Without RNA, most FUS appears in discs ranging from 30 to 50 nm in diameter. In most fields of view, large assemblies with a clear ropey sub-architecture are present. (F) Model for RNA seeding of FUS assemblies. Multiple FUS proteins bind cooperatively along the RNA. The FUS-RNA complex forms a seed that organizes non-RNA bound FUS into a fibrous assembly capable of binding the CTD of RNAP2.

FUS assemblies in figures 4A and 4B were formed without RNA to demonstrate that once FUS assemblies are formed, they are capable of binding the CTD in the absence of RNA. However, in the presence of RNA, much larger FUS assemblies were formed that also bound GFP-CTD (Fig 4C).

Highly Cooperative Binding of FUS to Multiple RNA Sequences

Although previous studies identified RNA sequence/structure motifs that bind FUS (Lerga et al., 2001; Hoell et al., 2012; Ray et al., 2013), any comprehensive model of FUS function must account for the broad RNA binding seen in CLIP-seq experiments (Schwartz et al., 2012). We therefore tested a 48 nt (nucleotide) RNA from the promoter of gene DNMT3b; this prD RNA, one of many sequences identified previously (Schwartz et al., 2012), contains none of the published FUS-binding motifs (Supp Fig 1A). Purified, monomeric FUS (Supp Fig 1B) bound this RNA with reasonably high affinity (K_d ^{= 97 ± 2 nM) and high cooperativity (apparent Hill coefficient} n = 4.0 ± 0.2) (Fig 1A). Because n is the minimum number of interacting binding sites in a system showing positive cooperativity, the simplest conclusion is that at least four FUS proteins are bound to prD RNA in complex 3.

Figure 1

FUS Binds RNA in a Highly Cooperative but Not Highly Sequence-specific Manner

(A) Left, trace prD RNA was incubated with increasing concentrations of full-length MBP-FUS protein (0 to 1200 nM) and binding analyzed by EMSA. Right, quantification of F_bound = RNA in complexes/total RNA in lane, as a function of FUS concentration. Data fit with a binding curve, Hill coefficient n = 4. Error bars (which are small) represent the range of data from two independent experiments. (B) EMSA experiments for the RNA variants shown in Supp Fig 1A, all binding to full-length FUS. Uncertainties listed for K_d ^and n represent the range of 2 or more replicates. (C) Domain structure of FUS protein and truncated versions of the protein that were tested for RNA binding. (D) Summary of RNA-binding data for truncation mutants of FUS. Uncertainties represent the range of 2 or more replicates. The LC domain bound RNA so weakly that binding parameters were not determined (N.D.). Samples of primary data are shown in Supplemental Fig. 2.

To assess whether some novel sequence/structure motif in prD RNA was responsible for its binding to FUS, we first replaced either its 5’ half or its 3’ half with poly(A) (Fig 1B). These 48 nt RNAs bound FUS with affinities similar to that of the parental prD RNA, indicating that the stem-loop structure in the 5’ half of prD was not necessary for binding. Poly(A) itself bound FUS weakly (Fig. 1B). Furthermore, three RNAs with abundant secondary structure – two group I intron domains (Bp P456 and Pne P456) and the Medaka telomerase RNA (Med TR) – all bound FUS with good affinity (K_d ^{= 46 to 111 nM) and high cooperativity (}n = 3 to 4) (Supp Fig 1B). GGUG RNA, containing a proposed FUS-binding motif (Lerga et al., 2001), and its negative control CCUC RNA bound with lower affinity (Supp Fig 1C). In conclusion, FUS binding is not highly specific for RNA sequence or structure.

The electrophoretic mobilities of the FUS-RNA complexes gave additional information. When poly(A) was substituted for either half of prD, binding was less cooperative (n = 1.7 to 2.0) and the intermediate Complex 1 accumulated much more than with prD RNA (Fig. 1B). This suggests that when half of prD is attached to the weak-binding poly(A) sequence, the prD portion binds two FUS molecules cooperatively (Complex 1) and only at high [FUS] does FUS also bind the poly(A) segment to give Complexes 2 and 3. The observation of such discrete complexes indicates well-folded protein. We conclude that diverse RNA sequences can bind multiple FUS proteins in a highly cooperative manner.

Contributions of FUS Domains to RNA Binding

The FUS protein has an N-terminal low-complexity (LC) domain, capable of forming fibrous structures in isolation (Kato et al., 2012), followed by two RNA-binding domains – an RRM and a zinc finger (Fig. 1C). RGG domains surround each RNA-binding domain. Both the RRM-containing and zinc finger-containing (RGG-Zn-RGG) domains bound prD RNA (Fig. 1D), as reported previously (Iko et al., 2004). The LC domain did not bind RNA (Supp Fig 1D). We found that the binding affinities of the two domains were very similar to each other and to that of the full-length protein; this may indicate that only one RNA-binding domain is available for binding in the full-length protein. Full binding cooperativity required both RNA-binding/RGG domains (see full-length FUS and del-LC); RGG-Zn-RGG bound with intermediate cooperativity, and LC-RRM bound with full affinity but reduced or no cooperativity.

RNA Interactions Promote FUS Binding to the CTD of RNAP2

We investigated the RNA dependence of FUS binding to the CTD of RNAP2. Titration of the prD RNA showed that FUS pulled down with GST-CTD at a half-maximum RNA concentration similar to the K_d ^{of this RNA for FUS (}Fig 2A). Multiple RNAs with sequences from gene promoters bound by FUS in HEK293T/17 cells, along with a variety of other noncoding RNAs, also promoted FUS pulldown with GST-CTD (Fig 2B). In multiple experiments, similar CTD binding occurred with MBPFUS, containing an N-terminal maltose binding protein solubility tag, as with H6-FUS (data not shown). Not all RNAs were able to promote FUS-CTD binding. Non-human sequences including the P456 domain of the Tetrahymena group I intron were active. Interestingly, a point mutant in P456 disrupting the RNA tertiary structure abolished its ability to promote FUS to bind the CTD. Despite showing no activity in promoting FUS-CTD interaction, RNAs Bp P456, Pne P456, and Med TR had high affinity for FUS binding. Thus, it appears that some RNA sequence/structure may be necessary for FUS to bind the CTD at the low concentrations tested.

Figure 2

RNA Promotes FUS Binding to the CTD of RNAP2

(A) Left, a representative western from pulldown of MBP-FUS by GST-CTD, stimulated by titrating in the prD RNA. Right, MBP-FUS binding to GST-CTD as a function of prD RNA concentration averaged from 3 replicates (error bars represent standard error). (B) The ability of various RNAs to promote FUS binding to the CTD, including sequences from FUS-regulated gene promoters (CIT, CYC1, ExoSC4, MCM7), RMRP, telomerase RNAs (yeast mini-TLC1, Tetrahymena (Tet TR), zebrafish (Zeb TR), or medaka (Med TR)) (Xie et al., 2008), and various group I intron domains from Tetrahymena (P456 with defined tertiary structure, Bp P456 with disrupted tertiary structure, P3P9) or Pneumocystis (carinii) jirovecii (Pne P456). The CIT RNA did not promote FUS to bind GST alone (left lane) Sequences GGUG and CCUC (Lerga et al., 2001) were too small to visualize on this Sybr stained gel. (C) Binding curve of GST-CTD for the prD RNA shows that CTD does not bind RNA. (D) EMSA of FUS for RNA with (dashed line) or without (dotted curve) the presence of 1 μM GST-CTD reveals no change in the K_d ^{of FUS for RNA.}

One simple model for the FUS-RNA-CTD interaction would involve the CTD binding directly to RNA (Kaneko et al., 2005) and the RNA then bridging FUS and CTD. However, the CTD itself showed no stable binding to prD RNA (Fig 2C). Alternatively, the CTD could simultaneously contact RNA and FUS, in which case the CTD should lower the K_d of FUS for RNA. However, FUS-RNA binding curves were unchanged in the presence of CTD (Fig. 2D, Supp Fig 2).

RNA Seeds FUS Assembly Formation

FUS has some ability to form higher order assemblies without any RNA, especially at high concentrations (~100 μM) (Sun et al., 2011; Han et al., 2012; Kato et al., 2012). We found that assemblies were visible as an opalescent turbidity in solution, which could be quantified by measuring the absorbance at 405 nm. Addition of prD RNA promoted more assembly formation (Fig 3A). Low concentrations of FUS in the presence of RNA were able to form assemblies to the same extent as higher concentrations of FUS without RNA. High concentrations of FUS upon addition of RNA eventually precipitated out of solution, preventing measurement (vertical dashed line in Fig 3A).

Figure 3

RNA Seeds FUS Assembly

(A) FUS protein at various concentrations was incubated with or without RNA and absorbance at 405 nm was measured (turbidity). Dashed line indicates that at higher RNA concentrations, 25 μM and 50 μM FUS samples precipitated in a large opalescent gel, which prevented turbidity measurement. (B) FUS incubated with or without 3 μM prD RNA in PBS pH 6.0 and crosslinked with formaldehyde. Protein assemblies were resolved by SDS-PAGE and stained with Coomassie Blue. (C) FUS was boiled at 95 °C for 5 min and treated with formaldehyde as in B. (D) Domain LC or heat denatured LC (Heat DN) was crosslinked with UV and resolved by SDS-PAGE. (E) The RNA-binding domains del-LC and RGG-Zn-RGG (20–26 μM) were incubated with various concentrations of prD RNA and crosslinked with formaldehyde. Numbers below lanes indicate fraction of the total protein that was crosslinked.

Using formaldehyde crosslinking, we found that addition of prD RNA promoted FUS assembly with 70% of protein crosslinked in assemblies (Fig 3B). A distinct ladder consistent with up to five FUS proteins could be resolved by SDS-PAGE (Supp Fig 3A). Single-stranded DNA with the same sequence as the prD RNA also promoted FUS assembly (Supp Fig 3B). Furthermore, crosslinking FUS in the presence of RNA followed by nuclease treatment did not destroy the complexes, suggesting that they involve protein-protein crosslinks and not merely protein-RNA crosslinks (Supp 3C). No FUS crosslinking was observed if FUS was first heat denatured at 95 °C, indicating that FUS structure is required for assembly (Fig 3C).

Formaldehyde crosslinking failed to detect LC domain assemblies. This is likely due to the fact that LC contains no lysines. Therefore, we crosslinked LC assemblies using UV irradiation. The amount of LC domain assemblies detected was reduced following heat denaturation of the protein (Fig 3D). This suggested that LC possesses some structure required for assembly formation and this assembly does not require RNA.

Next, we tested the RGG-Zn-RGG domain and the del-LC domain for their ability to assemble. Between 30 and 40% of del-LC and of RGG-Zn-RGG (Supp Fig 3D and 3E) were crosslinked in higher order assemblies without RNA, whereas 80% of the protein was in higher order assemblies in the presence of the prD RNA or ssDNA. Other RNA sequences tested (Supp Fig 1A) also promoted assembly formation to a degree consistent with their measured K_d ⁽Fig 1B).

Substoichiometric levels of RNA promote assembly formation. RNA at concentrations 9- or 25-fold below that of the del-LC or RGG-Zn-RGG protein promoted the majority of the protein to form assemblies (Fig 3E, note Supp Fig 3F for no cross-link control). At an RNA/protein ratio of 0.11, even with four FUS bound to each RNA, more than half of the FUS would be non-RNA-bound. Stoichiometric levels of RNA (the same concentration as the protein or higher) reduced assemblies to a level less than those seen in the absence of RNA (Fig 3E). We performed the same experiments for full-length FUS and found a similar trend for RNA-dependent seeding and then inhibition of protein assembly at high [RNA] (Supp Fig 3G).

FUS Assemblies Bind the CTD of RNAP2

FUS assemblies were large enough to be visualized by light microscopy. Assemblies adhered to glass ranged from 1 to more than 5 μm across and appeared ropey (Supp Fig 4A). In solution, they adopted a spherical shape consistent with phase transitions previously reported for protein hydrogels (Li et al., 2012) (Supp Fig 4B). Protein constructs containing the assembly domains of FUS also readily formed microscopic hydrogels. These hydrogel assemblies weakly stained for the amyloid-specific fluorescent dye thioflavin T (ThT), suggesting the presence of some cross-β structure. ThT stains RNA so all assemblies, except LC, stained brightly for ThT in the presence of RNA (Supp Fig 4C). FUS is known to bind tightly to its own LC domain (Wang et al., 2008; Kato et al., 2012); therefore, we tested whether this domain would bind these FUS assemblies. We observed LC to remain stably associated with these assemblies after more than 48-hour incubation with fresh PBS (Supp Fig 4D).

We then tested the ability of the CTD of RNAP2 to associate with these assemblies. FUS assemblies were allowed to form at high concentrations without addition of RNA. GFP-tagged CTD (GFP-CTD) was incubated with the assemblies (Fig 4A). Two days after fresh buffer exchange, GFP-CTD remained bound to the FUS assemblies (Fig 4B, grey bars). Negative control protein (rabbit IgG) did not associate with FUS assemblies, even when monitored as early as two hours after changing the buffer (Fig 4B, black bars). Finally, LC-RRM and del-LC assemblies were incubated with GFP-CTD. After 48 hours, GFP-signal was still detected above background. The signal normalized to FUS signal was higher for the del-LC construct than for the LC-RRM (Fig 4B), suggesting that CTD associated more stably with the RGG-Zn-RGG domain of FUS.

Figure 4

Assemblies of FUS Bind the CTD of RNAP2

(A) FUS assemblies were allowed to form at high concentration without RNA and spotted on glass coverslips. Assemblies were visualized by phase-contrast (Bright, left) and fluorescence microscopy (right). Bar indicates 5 μm. FUS was visualized by immunofluorescence with anti-FUS antibodies. GFP-tagged CTD was incubated with FUS assemblies and remained bound up to 48 hours after incubation in PBS. (B) Following a 48-hour incubation, the signal from GFP-CTD or negative control Alexafluor633-conjugated rabbit IgG was measured and normalized by the signal for FUS, del-LC, or LC-RRM for each assembly (n > 30). Error bars represent standard deviation. (C) FUS (60 μM) assemblies were formed in the presence (left) or absence (right) of RNA (3 μM). Fluorescence indicates binding of GFP-CTD. Solid white scale bar indicates 10 μm. (D) MBP-tagged domains LC and RGG-Zn-RGG were stained with 1% uranyl acetate and imaged by transmission electron microscopy. LC and RGG-Zn-RGG both form polymeric fibers 70 to 150 nm long and either 30 nm (LC) or 10 nm (RGG-Zn-RGG) in diameter. (E) Untagged FUS was incubated with or without prD RNA and imaged by TEM. Without RNA, most FUS appears in discs ranging from 30 to 50 nm in diameter. In most fields of view, large assemblies with a clear ropey sub-architecture are present. (F) Model for RNA seeding of FUS assemblies. Multiple FUS proteins bind cooperatively along the RNA. The FUS-RNA complex forms a seed that organizes non-RNA bound FUS into a fibrous assembly capable of binding the CTD of RNAP2.

FUS assemblies in figures 4A and 4B were formed without RNA to demonstrate that once FUS assemblies are formed, they are capable of binding the CTD in the absence of RNA. However, in the presence of RNA, much larger FUS assemblies were formed that also bound GFP-CTD (Fig 4C).

FUS Forms Fibrous Assemblies in an RNA-dependent Manner

We investigated the structure of FUS assemblies by transmission electron microscopy after staining with uranyl acetate. Consistent with previous literature (Sun et al., 2011; Kato et al., 2012), the LC domain formed amyloid-like fibers approximately 30 nm in diameter and 50 to 100 nm long (Fig 4D). Interestingly, the RGG-Zn-RGG domain also formed fibrous assemblies approximately 10 to 15 nm wide and 100 nm or more long.

To observe the sub-architecture of these branched assemblies, FUS was purified, the MBP-tag cleaved by PreScission protease and then assemblies were allowed to form by incubation at room temperature overnight with shaking. Upon investigation by TEM, FUS in the absence of RNA had some ability to spontaneously form ropey assemblies but mostly was found in round disk-shaped assemblies ranging from 30 to 50 nm in diameter. Upon addition of RNA, large branched assemblies were formed and sub-architecture could be observed that appeared as bundles of fibers forming the branches of these ropey structures (Fig 4E). We tested whether heat-denatured FUS could produce assemblies and did not observe any higher-order assemblies for denatured full length FUS or LC-RRM.

We used circular dichroism spectroscopy to assess the secondary structure of FUS and domains of FUS. The domains LC-RRM and del-LC had spectra comprised almost entirely of β-sheet and random coil (Supp Fig 4E). Domain RGG-Zn-RGG remained soluble in a buffer (14 mM NaH₂PO₄, 130 mM NaF) that allowed spectra to be collected for lower wavelengths (190-200 nm range). This spectrum revealed RGG-Zn-RGG to be 49% β-sheet and only 33% random coil. Domain LC was also largely β-sheet (Supp Fig 4E). These results support the conclusion that two domains of FUS, LC and RGG-Zn-RGG, form higher order assemblies rich in β-sheet structures.

DISCUSSION

RNA as a Regulator of FUS Assembly

Here we show that FUS is triggered to form higher order assemblies by binding to RNA. FUS assemblies utilize two domains common to proteins involved in cellular assemblies: the low-complexity (LC) domain (Kato et al., 2012) and the arginine-glycine rich (RGG) domain (this study). Each of these domains forms fibrous assemblies largely comprised of β-sheet secondary structure. In the absence of RNA binding, assembly requires very high protein concentrations. We find that FUS binds RNA with high cooperativity, implying the formation of a close-packed array of proteins on the RNA. We propose that this initial RNP then nucleates assembly formation, allowing it to occur at protein concentrations that are physiologically relevant (Fig 4F). These assemblies function to selectively bind the CTD of RNAP2 and the LC domain of FUS. This suggests that FUS assemblies have the potential to affect transcription by forming a protein scaffold that recruits RNAP2. Kwon et al. (Kwon et al., 2013) have presented direct evidence that FUS assemblies can stimulate transcription in human cells and that this activity depends on the ability for FUS to form fibrous assemblies.

We find that RNA has the ability to both (1) promote FUS assembly, at low RNA/protein ratios, and also (2) inhibit assembly, at high RNA/protein ratios (Fig 3E). Such behavior is consistent with a model in which a FUS-RNA complex nucleates additional FUS proteins to be added. At high RNA ratios, all FUS is bound to RNA in nucleation complexes, and there are no free FUS proteins to polymerize. We find that FUS assemblies undergo significant structural reorganization upon binding to RNA. The full-length protein in the absence of RNA exists in disc-shaped assemblies. The size of these discs and the fact that their diameter varies (30 to 50 nm) suggests that multiple proteins comprise these structures. However, the molecules of FUS present in these assemblies are not in a structure that is efficiently crosslinked by formaldehyde.

While the primary result of this paper is that RNA stimulates FUS assembly, an important question remaining is whether fibrous assemblies contribute to FUS-RNAP2 interactions in vivo. FUS affects RNAP2 recruitment, the efficiency of transcription, and the phosphorylation status on the CTD of RNAP2 (Yang et al., 2000; Schwartz et al., 2012, Kwon et al., 2013). An attractive model is that the efficiency of CTD recruitment to a FUS fibrous assembly would localize the polymerase to a greater extent than single FUS-CTD interactions. However, this model remains to be critically tested in vivo.

The FUS RGG-containing Domain Forms Fibers

Here we show that the RGG-Zn-RGG domain unexpectedly forms fibrous assemblies, in addition to those formed by the LC domain. RGG motifs are frequently found in RNA-binding proteins and are well known to bind RNA (Thandapani et al., 2013). However, one study has suggested that the majority of RNA-binding for FUS in vivo may be through the RRM (Daigle et al., 2013).

There is precedence for implication of RGG motifs in formation of protein assemblies. The RGG domain at the C-terminus of nucleolin has been shown to mediate protein interactions (Bacharach et al., 2000). The formation of translation-repressive mRNPs is regulated by RGG domains present in certain proteins that form assemblies with eiF4E and 4G (Rajyaguru et al., 2012). The first RGG domain in RGG-Zn-RGG of FUS has been predicted to be prion-like (King et al., 2012). We find the domain RGG-Zn-RGG to have an unexpectedly high content of β-strand secondary structure. We further note that the crystal structure of an antifreeze protein of the snow flea, which possesses a highly similar amino acid motif XGG, reveals an unusual β-sandwich structure stabilized by backbone hydrogen bonding rather than a more traditional hydrophobic core (Pentelute et al., 2008).

Implications of FUS Assemblies in Neurological Disease

Although higher-order polymer assemblies formed by proteins are fundamental to cellular function, higher-order assemblies can also contribute to disease pathology. The principles of RNA-promoted FUS assembly may apply more generally to the assembly of other RNA-rich bodies in the cell. Prion or low complexity domains contained in RNA-binding proteins have been proposed to drive RNP granule assembly (Gilks et al., 2004; Decker et al., 2007). Likely candidates that may have similar propensity for RNA-promoted assembly include: (1) those that have high sequence similarity – EWSR1 and TAF15; (2) those with related domain architectures – CIRBP and hnRNP proteins such as A1, A2B1, U, A0; and (3) RGG-domain containing proteins – FXRP1, nucleolin, LSM4, and fibrillin. An emerging theme is that perturbation of the biologically important assembly of these RNA-binding proteins can be pathogenic (Bosco et al., 2010; Ju et al., 2011; Kim et al., 2013; Ramaswami et al., 2013).

EXPERIMENTAL PROCEDURES

Protein Purification

Recombinant MBP-FUS or H6-FUS was expressed in and purified from BL21 E. coli. Briefly, cells were lysed by sonication in lysis buffer (1 M KCl, 50 mM Tris pH 7.4, 10 mM imidazole, 5% glycerol, 1% NP40, 1.5 mM β-mercaptoethanol, either with or without 1 M urea). Protein was extracted from the supernatant with Ni-sepharose beads. Beads were washed 4 times in wash buffer (1 M KCl, 50 mM Tris pH 7.4, 10 mM imidazole, 1.5 mM β-mercaptoethanol, either with or without 1 M urea). Protein was eluted in either wash buffer supplemented with 250 mM imidazole or PBS supplemented with 250 mM imidazole. See Extended Experimental Procedures for details.

FUS-RNA Binding Curves

In a 20 μl reaction, a trace amount of [5’-^{P] prD-RNA was incubated with MBP-FUS in binding buffer (50 mM Tris-HCl pH 7.4, 150 mM KCl, 2 mM MgCl}₂, 2 mM DTT, 0.1 mg/ml yeast tRNA, 0.1 mg/ml BSA and trace amount of orange dye) at room temperature for 30 min. 10 μl of each reaction was loaded onto a 4-20% TBE (Invitrogen) gel and separated at room temperature at 150 V for 1.5 hours. The ^{P signal was detected using phosphorimager screens and quantified using ImageG software. Curves and binding constants were calculated using MATLAB software. MBP-FUS was pre-treated with micrococcal nuclease to remove any nucleic acid that might have co-purified.}

CTD Pull-down Assays

FUS (0 to 4 μM) and GST-CTD (0 to 1 μM) were incubated with different concentrations of prD-RNA in 35 μl reaction buffer (50 mM Tris-HCl pH 7.4, 150 mM KCl, 2 mM MgCl₂, 2 mM DTT, 0.1 mg/ml BSA, 0.1 mg/ml yeast tRNA) for 30 min at room temperature. Complexes were immobilized using Glutathione Sepharose™ 4B beads (GE Healthcare). After washing, proteins were eluted with 35 μl elution buffer (reaction buffer without BSA and yeast tRNA but containing 30 mM glutathione, pH to 7.4 by KOH). Proteins were resolved on a 4-20% NuPAGE Bis-Tris Gel (Invitrogen), followed by western blotting with either anti-FUS (4H11; Santa Cruz) or anti-GST (ab3416; Abcam).

For more description, please refer to the Extended Experimental Procedures.

Supplementary Material

01

02

ACKNOWLEDGEMENTS