Subcellular membrane-less organelles consist of proteins with low complexity domains. Many of them, such as hnRNPA1, can assemble into both a polydisperse liquid phase and an ordered solid phase of amyloid fibril. The former mirrors biological granule assembly, while the latter is usually associated with neurodegenerative disease. Here, we observe a reversible amyloid formation of hnRNPA1 that synchronizes with liquid–liquid phase separation, regulates the fluidity and mobility of the liquid-like droplets, and facilitates the recruitment of hnRNPA1 into stress granules. We identify the reversible amyloid-forming cores of hnRNPA1 (named hnRACs). The atomic structures of hnRACs reveal a distinct feature of stacking Asp residues, which contributes to fibril reversibility and explains the irreversible pathological fibril formation caused by the Asp mutations identified in familial ALS. Our work characterizes the structural diversity and heterogeneity of reversible amyloid fibrils and illuminates the biological function of reversible amyloid formation in protein phase separation.
Membrane-less organelles are organizations of proteins, RNA, or DNA molecules that condense at specific subcellular loci to fulfill defined biological functions1,2. In contrast to steady membrane-bound organelles, many membrane-less organelles, such as stress granules and P-bodies, dynamically assemble and disassemble in response to cellular stress and other stimuli3,4. Extensive studies have identified that many RNA-binding proteins, such as FUS, hnRNPA1, and TDP43, largely contribute to this conditional cellular compartmentation5,6. These proteins commonly contain low complexity (LC) domains that can drive the full-length proteins to undergo liquid–liquid phase separation (LLPS) via weak multivalent interactions7,8,9. On the other hand, many LCs are also prone to amyloid aggregation10,11. Missense mutations of LCs identified in neurodegenerative diseases, including amyotrophic lateral sclerosis (ALS), multisystem proteinopathy (MSP), and frontotemporal dementia (FTD), can promote amyloid aggregation, which is closely associated with the pathogenesis of these diseases12,13,14.
Liquid-like droplets and amyloid fibrils are both higher-ordered protein assemblies, while are different in thermodynamics and in principle can form successively. Indeed, confocal microscopy has shown amyloid fibrils grow out from liquid-like droplets of hnRNPA1 and FUS15,16. In the sense that amyloid fibrils are highly stable and resistant to proteolysis and harsh conditions (e.g., SDS and heating), amyloid-prone property seems to be of a high risk for cells and might not be the basis for dynamic granule assembly17,18,19. However, it has been recently emerging that amyloid fibrils formed by LC may be metastable and even readily reversible. Mcknight et al. showed that RNA-binding proteins including FUS and hnRNPA1 can form SDS-sensitive metastable fibrils20,21,22. Eisenberg et al. observed temperature-sensitive fibrils formed by an artificial sequence derived from LC23. Our previous work on FUS LC identified two segments that form reversible fibrils regulated by temperature and phosphorylation24. These observations altered the view of amyloid fibrils as thermostable end-product of protein super-molecular assembly, and raised the possibility that amyloid formation may play a functional role in the assembly of membrane-less organelles.
In this work, we directly visualize temperature-sensitive amyloid fibrils in the liquid-like droplets of hnRNPA1 by transmission electron microscopy (TEM). The fibrils synchronize with the occurrence and disappearance of liquid-like droplets as we observe, while over time, irreversible fibrils form and accumulate. We identify a reversible amyloid core from hnRNPA1 LC (named hnRAC1) and monitor the real-time in situ self-disassembly of hnRAC1 fibrils by atomic force microscopy (AFM). We further determine the atomic structure of hnRAC1 by microelectron and X-ray diffraction to the resolution of 1.0 Å and 1.4 Å, respectively. The structure reveals a distinct feature of negatively charged Asp residues stacking along the fibril, which contributes to fibril instability. Based on this structure, we identify another two reversible amyloid cores from hnRNPA1 LC (named hnRAC2 and hnRAC3) and more from hnRNP families. Asp mutations found in familial ALS enhance the formation of irreversible fibrils. We show that hnRACs not only define the formation of reversible fibrils but also reinforce hnRNPA1 phase separation in vitro and involvement into stress granules in vivo. Our work bridges the liquid phase and solid phase of hnRNPA1 and illuminates the functional role of reversible amyloids in the assembly of stress granules and the pathological risk as intermediates en route to irreversible fibrils.
Multiple-state self-assembly of hnRNPA1
To observe the phase separation of hnRNPA1, we cooled down the hnRNPA1 solution from 25 °C to 4 °C. The solution turned cloudy in a few minutes. Differential interference contrast (DIC) microscopy showed spherical droplets of hnRNPA1 in the liquid phase (Fig. 1a). The liquid property of the droplets was shown as that two adjacent droplets fused spontaneously within seconds measured by optical tweezers (Fig. 1b). We deposited the phase-separated hnRNPA1 sample on TEM grids to check the content of the droplets by negative-staining TEM. Intriguingly, we observed bunches of amyloid fibrils in the droplets (Fig. 1a; Supplementary Fig. 1). Although we carefully monitored the size and shape of droplets during the TEM grid preparation to minimize the influence of drying, it is still possible that the observed fibrils are derived from a drying artifact. Thus, we further validated the amyloid fibril formation by monitoring Thioflavin T (ThT) fluorescence using confocal microscopy and spectrophotometry (Fig. 1c, d). The results showed a specific enhancement of ThT intensity in the droplets, which consistent with the TEM imaging, indicates amyloid fibril formation in the droplets. To rule out a possible artifact of ThT staining, the FUS RGG region (residues 371–526) was used as a control. FUS RGG can undergo LLPS, but contains no amyloid-forming sequence. The control sample showed no obvious ThT intensity comparing with the hnRNPA1 sample (Fig. 1c), and artificial enhancement of ThT intensity by adjusting the image contrast showed that ThT spread out in the control sample, rather than condensed in the droplets as that seen in the hnRNPA1 sample (Fig. 1c; Supplementary Fig. 2).
As temperature reverted back to 25 °C, the hnRNPA1 solution became clear again, and neither droplet nor amyloid fibril was observed, indicating that the fibrils are reversible (Fig. 1a, c). As we estimated based on the ThT intensity, <10% of the total proteins in molecules formed reversible fibrils (Supplementary Fig. 3). Note that ThT fluorescence revealed a small amount of amyloid aggregates remained as temperature increased back to 25 °C (Fig. 1c), which indicates that a portion of fibrils maturated to relatively irreversible fibrils. Moreover, as we kept the cloudy hnRNPA1 solution at 4 °C for hours, the solution spontaneously became clear and the amount of reversible fibrils decreased accordingly (Fig. 1a; and Supplementary Fig. 3). Overnight, a significant amount of irreversible amyloid fibrils emerged which are stable as temperature increased to 25 °C for elongated time (Fig. 1a).
In these experiments, we observed three states of hnRNPA1 assemblies: liquid-like droplets, reversible fibrils, and irreversible fibrils. Among them, irreversible fibrils are known to be constructed by folded and highly ordered stacking of proteins, such that they are the most thermodynamic and hard to dissemble. Liquid-like droplets and reversible fibrils appear to be dynamic assemblies that can self-dissociate into soluble proteins or over time, proceed to enter the thermostable state of irreversible fibrils depending on energy input (e.g., temperature).
Identification of a reversible amyloid core of hnRNPA1
We next sought to investigate the mechanism underlying the reversible amyloid formation of hnRNPA1. To identify the reversible amyloid core (RAC) of hnRNPA1, we synthesized a series of segments selected from the LC domain with different lengths from 5 to 12 residues. The segments contain [S/G]Y/F[S/G] and/or RGG motifs that are characteristic in LC21,25,26,27 (Supplementary Fig. 4). The segments were screened for reversible amyloid formation. The result showed that segment 209GFGGNDNFG217 (named hnRAC1), but not the others, formed hydrogel at 4 °C. The hydrogel was composed of amyloid fibrils observed by TEM (Fig. 2a). As temperature increased to 25 °C, the fibrils disassociated spontaneously, resembling the behavior of full-length hnRNPA1 (Figs. 1a, 2a). Moreover, we utilized AFM to monitor the real-time and in situ self-disassembly of hnRAC1 fibrils at a single-fibril level. We observed that hnRAC1 fibrils gradually melt down in a timescale of minutes upon temperature increased from 4 °C to room temperature, whereas irreversible fibrils stayed unchanged (Fig. 2b; Supplementary Fig. 5). This observation emphasizes the remarkably dynamic property of reversible fibrils that is distinct from irreversible fibrils extensively studied previously18,19,28,29, although X-ray fibril diffraction showed that reversible hnRAC1 fibrils also feature a typical cross-β architecture as seen in irreversible amyloid fibrils19,28, with an interstrand spacing of ~4.8 Å and an inter-sheet spacing of ~11 Å (Supplementary Fig. 6).
Atomic structures of hnRAC1
To reveal the structural basis underlying the reversible amyloid formation of hnRAC1, we determined the atomic structure of hnRAC1 in the fibrillar form to the resolution of 1.0 Å by micro-electron diffraction (micro-ED) and to 1.4 Å by micro-focus X-ray diffraction (Table 1). HnRAC1 formed needle-like crystals that diffracted both electron and X-ray beams well. Especially for electron diffraction, the highest resolution reached a sub-angstrom of 0.82 Å (Fig. 3a; Supplementary Fig. 7a). The structural models obtained from the two different technologies are identical, yet micro-ED is superior to X-ray diffraction on structure resolution in this case (Fig. 3a; Table 1).
The structure revealed a cross-β architecture (or steric zipper18) with hydrophilic sheet interface composed by N213 and N215 (Fig. 3b). Hydrophilic interface has been mainly seen in amyloids formed by prion-like domains and LCs, and contributes to fibril instability, in contrast to the hydrophobic interface that is abundant in irreversible pathological fibrils18,19,24,28 (Supplementary Fig. 7b). Moreover, the hnRAC1 structure exhibits a distinct feature of negatively charged D214 continuously stacking along the parallel in-register β-sheets (Fig. 3b). Stacking-D engenders instability to the fibril architecture due to electrostatic repulsion and hence is barely seen in pathological fibrils (Supplementary Fig. 8). Note that in the local chemical environment of the hnRAC1 crystal, D214 is in contact with the amino terminus of hnRAC1. To evaluate whether this contact exists in fibrils and whether the behavior of D214 is relevant to that of the full-length protein, we modified the N-terminus of hnRAC1 with acetylation and performed D214 mutagenesis on both hnRAC1 and full-length hnRNPA1. The result showed that N-terminally acetylated hnRAC1 formed reversible amyloid fibrils with a similar diffraction pattern to that of the non-acetylated hnRAC1 by X-ray diffraction (Supplementary Fig. 9), which indicates that the amino terminus of hnRAC1 is dispensable for the fibril structure and reversibility. In contrast, as we mutated D214 to V or N, although the resulted peptides still formed amyloid fibrils with no profound structural change (Supplementary Fig. 10), unlike the WT fibril, the mutant fibrils were stable and did not dissociate as temperature increased (Fig. 3c; and Supplementary Fig. 5). Consistently, D214V mutation also resulted in the irreversible fibril formation of full-length hnRNPA1 (Fig. 3c).
Another structural feature of hnRAC1 structure lies on a sharp kink at G211. Kinked structure has been shown to widely present in various LCs for protein network assembly23,30. Similarly, in the hnRAC1 structure, kink at G211 enables the aromatic ring of F210 to reach out and interact with the Phe residues from neighboring molecules via π–π stacking (Fig. 3d). The intermolecular π–π interactions may cross-link single fibrils to form hydrogels. To further assess the role of F210 and F216, we mutated them to Ala, respectively. The result showed that the mutations impaired hydrogel formation of hnRNPA1, whereas did not significantly affect the formation of reversible amyloid fibrils (Fig. 3e). In addition, we mutated G211 to Val to constrain the geometry of F210. Similar to the result of F210 mutation, G211 mutation failed in gelation, while retained the ability of reversible amyloid formation (Fig. 3e). Thus, (G)F(G) motif is important for inter-fibrillar interactions and may also contribute to recruiting other molecules into stress granules.
As a control, we also determined the atomic structure of segment 234GGGYGGS240 by X-ray diffraction (Table 1). This segment does not form fibrils or hydrogels and thus may not be an amyloid core (Supplementary Fig. 4b). Indeed, the structure showed an unfolded conformation with no fibrillar packing (Supplementary Fig. 4b).
Identification of hnRAC2 and hnRAC3 in hnRNPA1
According to the structural characteristics of hnRAC1 fibrils, we revisited hnRNPA1 LC for segments that contain (N)D(N) and (G)F/Y(G) motifs. Notably, in addition to hnRAC1, we found another two candidates, 246GFGNDGSNF254 (named hnRAC2) and 260YNDFGNY266 (named hnRAC3) (Fig. 4a). We synthesized the two segments and observed that both of them indeed formed reversible fibrils and hydrogels, similar to the behavior of hnRAC1 (Fig. 4b). The reversible fibrils formed by hnRAC2 and hnRAC3 exhibited typical cross-β architectures by X-ray fibril diffraction (Supplementary Fig. 10).
Based on the atomic structure of hnRAC1 and the distance restrains of hnRAC2 and hnRAC3 derived from X-ray fibril diffraction, we built the structural models of hnRAC2 and hnRAC3 by Rosetta modeling (Fig. 4c; Supplementary Note 1). The structure of hnRAC2 showed that the side chains of N249 and N253 face inward the sheet interface and the side chain of D250 facing outward and stacking along the fibril axis. Mutation of D250V impaired the reversibility of fibrils formed by both hnRAC2 and full-length hnRNPA1 (Fig. 4b). Similar to that of hnRAC1, the steric zipper of hnRAC2 kinked at G248, which enables F247 to reach out for intra-fibrillar interactions. Indeed, mutation of F247A inhibited hnRAC2 to form hydrogels with no significant effect on the formation of reversible fibrils (Fig. 4b). As well, the N-terminal acetylation showed no influence on the structure and reversible fibril formation of hnRAC2 (Supplementary Fig. 9).
HnRAC3 fibril spine features a similar architecture to those of hnRAC1 and hnRAC2. The sheet interface is composed of the side chains of N261, F263, and N265 (Fig. 4c). The side chain of D262 facing outward the sheet interface presenting as a signature stacking-D. Importantly, mutations of D262V and D262N, which have been identified in familial ALS patients12, resulted in both hnRAC3 and full-length hnRNPA1 to form irreversible fibrils that were not able to disassemble as temperature increased (Fig. 4b). Taken together, our structural studies reveal a key role of stacking-D in the reversibility of amyloid formation, and explain that the disease mutations on the Asp residues may impair the reversibility of hnRNPA1 fibrils and hence enhance irreversible amyloid aggregation and pathogenesis of ALS.
In addition, we surveyed the LC domains of other members in the hnRNP family, including hnRNPA2, hnRNPDL, hnRNPH, hnRNPK, and hnRNPR, which were identified to be closely associated with different diseases, such as ALS, spinal muscular atrophy (SMA), and cancer31,32,33,34. We found several potential RAC segments and further characterized their capability of reversible amyloid fibril formation. As a result, for each protein, we were able to identify RAC segments that can form reversible amyloid fibrils (Supplementary Fig. 11). Notably, the sequence of hnRAC3 in hnRNPA1 also exists in hnRNPA2, and even more, mutation of Asp in this segment of hnRNPA2 has also been identified in ALS familial mutations12. This result indicates that RACs with stacking-D are widely adopted by proteins in the hnRNP family for the formation of reversible amyloid fibrils.
The function of hnRNPA1 reversible amyloid formation
To assess the role of reversible amyloid formation in phase separation and the involvement of stress-granule assembly of hnRNPA1, we first examined the influence of hnRACs to the phase-separation diagram of hnRNPA1. We deleted each individual RAC from the full-length hnRNPA1, respectively. Phase separation diagrams of the variants showed a significant deficiency in phase separation with lower clouding points and smaller amounts and sizes of droplets (Fig. 5a, b). As a control, deletion of 234GGGYGGS240 showed no significant effect (Supplementary Figs. 4b, 12). Reversely, we inserted an additional hnRAC1 into hnRNPA1 at the position between residues 198 and 199. This variant showed a significant promotion of phase separation with an increase of clouding point from 12 °C to 15 °C (Fig. 5a, b).
To assess the effect of amyloid formation on the physical property of the droplets, we next measured the fusion time of hnRNPA1 droplets by optical tweezers. The data showed that hnRAC1 deletion caused the droplets to fuse more quickly than that of the wild-type (Fig. 5c). In contrast, D214V mutation that promotes irreversible amyloid formation, caused the droplets to take longer to fuse (Fig. 5c). Then, we monitored the fluorescence recovery after photobleaching (FRAP) of the droplets. We photobleached a small portion in the center of a droplet and observed that the fluorescence signal of the droplets formed by hnRNPA1 with hnRAC1 deletion, recovered more quickly than that of the wild-type, while D214 mutation resulted in a slower recovery than that of the wild-type (Fig. 5d). Taken together, these data indicate that reversible amyloid formation decreases the fluidity and mobility of hnRNPA1 droplets, and hence reinforces the phase separation.
We next investigated the role of hnRACs in the assembly of stress granules in vivo. We observed that in response to stress, Flag-hnRNPA1 was released from the nuclei to the cytosol and recruited into stress granules (Fig. 5e). Deletion of either hnRAC1 or hnRAC2 significantly diminished the recruitment of hnRNPA1 into stress granules, while insertion of hnRAC1 reversely enhanced it (Fig. 5e). Note that deletion of hnRAC3 is not applicable for this experiment since it resulted in leaking of hnRAC3 from the nucleus to the cytosol when there is no stress due to a potential overlap of hnRAC3 with the nuclear localization sequence (NLS)35 (Supplementary Fig. 13). The effects of hnRACs on stress granule assembly in vivo is consistent with that observed on hnRNPA1 phase separation in vitro, which together provide evidence for the important functional role of reversible amyloid formation in hnRNPA1 phase separation and stress-granule formation.
In this work, we observed a synchronous formation and deformation of hnRNPA1 droplets and reversible fibrils (Fig. 1), which indicate that the reversible fibrils may play a role in hnRNPA1 phase separation. By optical tweezers and FRAP experiments, we show that reversible fibrils solidify the liquid-like droplets, which by changing the mechanical property of the droplets, reinforces hnRNPA1 phase separation (Fig. 5c, d). Thus, our work suggests that reversible fibrils play an important functional role in stress granule assembly. From a physical view, membrane-less assemblies are condensed matter of biological macromolecules. Amyloid fibrils as super-molecular polymers could confer such condensed matter with a property of complex fluid that features plasticity and sensitivity to the microenvironmental change and molecular signals. Indeed, it has been reported that stress granules and nuclear paraspeckles consist of stable substructures with a hard core and a mobile shell, which indicates a complex state of molecular assembly with manifold interactions besides low-affinity multivalent interactions36,37,38. To this end, study of the physical and chemical properties of reversible fibrils and the transition of different assemblies of hnRNPA1 and others (e.g., FUS) would help to understand the hierarchical assemblies of membrane-less organelles and biological regulations to this process.
In addition, we also observed the influence of hnRAC deletions on the initiation of phase separation (Fig. 5a). This might reflect a mixed effect of both valence decrease and deficient fibril formation. Weak multivalent interactions are known to drive the phase separation of proteins39. While, amyloid fibrils may strengthen the low-affinity interactions by condensing valences40,41,42,43, and thus may enhance phase separation.
Amyloid fibrils as highly ordered protein super-molecular polymers are commonly thermostable and pathological that are related to numerous neurodegenerative diseases17,19,44. RNA-binding proteins, such as FUS, hnRNPA1, and TDP43, contain LC domains that are prone to form pathological amyloid fibrils associated with diseases, such as ALS6,10,45. While, recent studies have shown that LCs can also form metastable or reversible amyloid fibrils20,21,22,23,24. Thus, it is elusive how LCs form pathological irreversible amyloid fibrils as well as dynamic reversible fibrils, what is the structural difference between these two fibrillar assemblies, and what regulates their formation. In this work, we observed the state transition of hnRNPA1 regulated by temperature from dynamic liquid–solid co-existing phase to solid irreversible fibrils. Liquid droplets and reversible fibrils are dynamic states of protein assemblies with high free energies, whereas irreversible fibrils are more thermostable (Fig. 6). It appears that irreversible amyloid-forming proteins commonly use a continuous region to form amyloid fibril spine46,47 (Supplementary Fig. 8). In contrast, reversible amyloid-forming proteins, e.g., FUS and hnRNPA1, contain multiple amyloid-forming cores24. For example, hnRNPA1 contains three RAC segments that are separated by highly flexible R/G-rich regions (Fig. 4a), and thus it is conceivable that amyloid fibrils may form heterogeneously via different RACs and different numbers of RACs. As more RACs are involved in the fibril spine, the fibrils might turn more stable and difficult to reverse (Fig. 6). Indeed, we observed that deletion of hnRACs, especially hnRAC3, obliterated irreversible fibril formation (Supplementary Fig. 14). Certainly, the irreversible amyloid core, if exists, may also contribute to the irreversible amyloid formation. Moreover, once a reversible amyloid core lost its reversibility due to mutation, irreversible fibril formation could be largely promoted, which provides mechanistic explanation for the pathological condition of Asp mutations identified in familial ALS12 (Fig. 6). Therefore, these evidence suggest that reversible amyloid fibrils of hnRNPA1 are dynamic intermediates en route to irreversible fibrils, which is reminiscent of amyloid oligomers that are pathological intermediates of irreversible fibrils48,49,50, while reversible fibrils serve as a functional intermediate. To benefit from the reversible fibril formation and meanwhile encounter the risk of irreversible fibril formation, cells need to employ tight regulation on protein-phase separation, failure of which may lead to diseases.
Our previous work on RAC segments of FUS LC showed the structural mechanisms of two reversible amyloid fibril spines24. One exhibits an ordered-coil architecture distinct from the cross-β structure that usually seen in amyloid fibrils; the other features a hydrous sheet interface that is different from the dry interface of known amyloid fibrils (Supplementary Fig. 15). In this work, we present another different structural mechanism for reversible amyloid fibrils formed by RAC segments of hnRNPA1, which features stacking-D in the fibril core (Supplementary Fig. 15). A recent NMR study also suggests the involvement of D290 in destabilizing the self-association of hnRNPA2 LC domain51. In addition, kinked β-sheets (so called LARK) have also been suggested to destabilize fibril formation23,34. Taken together, it is plausible to demonstrate that structural diversity is an essential difference between reversible and irreversible amyloid fibrils. There are various mechanisms allowing reversibility, while irreversible fibrils commonly adopt a uniform structure of steric zipper with dry (usually hydrophobic) interface18,19,28. Steric zipper is generally a highly thermostable architecture of protein super-molecular assembly, whereas, many disturbances, such as hydrophilic and/or hydrous sheet interface, ordered-coil and kinked β-sheet (LARK) configurations, and stacking-D in the fibril core may introduce instability to amyloid fibrils and generate reversible amyloid fibrils. More importantly, these structural features provide bases for regulation of dynamic amyloid assembly, disruption of which thus may lead to pathogenesis of neurodegenerative diseases.
Protein expression and purification
Wild-type hnRNPA1 in vector pET-9d was purchased from Addgene (Plasmid #23026). Variants of hnRNPA1 were subcloned from the wild-type construct. Recombinant hnRNPA1 and variants were overexpressed in E. coli BL21(DE3) pLysS with the addition of 0.4 mM IPTG at OD600 = 1.0–1.2 and further incubation at 25 °C overnight. Cells, except for those expressing the D262V mutant, were harvested and lysed in buffer containing 50 mM Tris-HCl pH 7.5, 100 μg ml−1 RNAase (Roche), 2 mM DTT and 1 mM PMSF. The lysate was centrifuged at 30,966 g at 4 °C. The supernatant was loaded onto a 5-ml SP FF column connected with ÄKTA Purifier (GE Healthcare, USA). The proteins were eluted with a gradient mixing of buffer A (50 mM Tris-HCl pH 7.5 and 2 mM DTT) and buffer B (buffer A with 1 M NaCl). Fractions were analyzed by SDS-PAGE. Those mainly containing hnRNPA1 were collected and further purified by Superdex 75 16/60 column (GE Healthcare) pre-equilibrated with buffer containing 50 mM Tris-HCl pH 7.5, 500 mM NaCl, and 2 mM DTT. Fractions containing hnRNPA1 monomers were collected and concentrated. The D262V mutant was overexpressed and purified with the same procedure as the others, except that the buffer pH was adjusted to Bis-Tris pH 6.5. All purified proteins were changed to buffer containing 50 mM Tris-HCl pH 7.5, 100 mM NaCl for the following experiments.
The FUS RGG region (residues 371–526) was constructed in vector pET-28a and overexpressed in E. coli BL21(DE3) by induction at OD600 = 1.0–1.2 with 0.5 mM IPTG and further incubation for 24 h at 16 °C. Cells were harvested and lysed in buffer containing 50 mM Tris-HCl pH 8.0 and 6 M guanidine hydrochloride. The cell lysate was then centrifuged at 30,966 g for 1 h. The supernatant was loaded onto a 5-ml Ni column (GE Healthcare, USA). The bound proteins were eluted with buffer solution containing 50 mM Tris-HCl pH 8.0, 6 M guanidine hydrochloride, and 50 mM imidazole. The eluted proteins were further purified by HPLC (Agilent, USA) with the elution buffer containing 35% v/v acetonitrile. The purified protein was freeze-dried by FreeZone lyophilizers (Thermo, USA) and stored at −20 °C.
In total, 100 μM hnRNPA1 wild-type and variants were incubated at 6 °C and LLPS occurred in minutes. In all, 3 μl of protein samples were dropped on a glass slide. The slide was placed on a TP-CHSQ-C thermal stage that was maintained at 6 °C and checked by a Leica SP8 confocal microscope. Images were taken by a DMI8 camera with a 63 × 0.7 NA dry objective.
To observe the content of droplets, 80 μM hnRNPA1 was placed on ice for 20 min. In total, 4 μl of the sample was transferred to a glow-discharged TEM grid (Cu, 300 mesh; Beijing Zhongjingkeyi Technology Co., Ltd.). The grid was washed with 4 μl pre-cooled 3% w/v uranyl acetate, and then stained by 4 μl pre-cooled 3% w/v uranyl acetate for 45 s. During sample preparation, most droplets were washed off the grid or disrupted by uranyl acetate solution. To observe an integrate droplet, multiple grids were prepared in parallel and carefully checked by TEM. Since the thickness of droplet weakens electron transmission and blurs the imaging of fibrils inside, we also prepared grids with an additional step of washing with 4 μl pre-cooled ddH2O before staining to fully disrupt the droplets. Before acquiring TEM images, the excess buffer on grids was removed and dried in air. Grids were loaded on a FEI Tecnai T12 TEM working at 120 KV. The TEM is equipped with LaB6 filament and BM-eagle camera with the pixel of 4096 × 4096. Exposure time for each image was 2 s.
Amyloid fibril sample grids were prepared and observed by TEM following the same protocol. To check the reversibility of fibrils, the fibril samples were incubated at 25 °C for 20 min and then observed by TEM.
ThT fluorescence staining
Overall, 19 μl of 100 μM hnRNPA1 were placed on ice. The protein solution turned cloudy in seconds, followed by addition of 1 μl of 1 mM ThT and further incubation for 5 min. The sample was observed at 4 °C by a Leica SP8 confocal microscope equipped with a 458 laser and a TP-CHSQ-C thermal stage. ThT signals were detected by a HyD detector. To test the fibril reversibility, the stage temperature was gradually increased from 4 °C to 16 °C and further to 25 °C. The sample was incubated at each temperature for 2 min. The control sample of the FUS RGG region was prepared and observed following the same protocol.
Amyloid fibril formation
Reversible amyloid fibrils of full-length hnRNPA1 were observed as phase separation by placing 80 μM hnRNPA1 on ice for 20 min. To test the mutation of Asp residues to the reversibility of the full-length hnRNPA1 fibrils, hnRNPA1 proteins with mutations at the concentration of 100 μM were placed on ice for 20 min to form fibrils. The fibril reversibility was assessed by increasing the temperature from 4 °C to 25/37 °C.
Peptides were synthesized with free termini or N-terminal acetylation as indicated by ChinaPeptides Co., Ltd. All peptides were dissolved in ddH2O. The peptide sequences and conditions for gelation and fibrillation were listed in Supplementary Table 1. The reversibility was measured by increasing the temperature from 4 °C to 25 °C or 37 °C as indicated in figures for 20 min.
Liquid cell and freshly cleaved mica were pre-cooled at 4 °C. Overall, 30 μl of fibril samples were added into the liquid cell. Images were acquired in fluid using a MultiMode AFM instruments with NanoScope V system (Bruker). Measurements were carried out using SNL-10 silicon cantilevers with a spring constant of 0.24 N m−1 (Bruker) in scanasyst mode. Disassembly of fibrils was monitored by AFM with a scan rate of 0.9 Hz upon temperature increase from 4 °C to room temperature. All images were analyzed by NanoScope Analysis 1.5 (Bruker).
Crystallization and data collection
HnRAC1 was dissolved in ddH2O to 3 mg ml−1. Crystals were obtained by hanging drop in buffer containing 2.0 M ammonium sulfate, 0.1 M CAPS pH 10.5, 0.2 M lithium sulfate at 16 °C. Peptide 234GGGYGGS240 was dissolved in ddH2O to 20 mg ml−1. The crystals were obtained from the buffer containing 0.2 M ammonium sulfate, 30% PEG 4000 at 16 °C by hanging drop.
For Micro-ED data collection, the crystal drop was transferred to a freshly glow-discharged quantifoil EM grid (R2/2, Cu 300 mesh; QUANTIFOIL), followed by blotting from the bottom and washing twice with 5% v/v PEG 200. Then, the grid was vitrified by plunging into liquid ethane using Vitrobot Mark IV (FEI). The grid was then transferred using a Gatan 626 cryo-holder to FEI Tecnai F20 equipped with field-emission gun operating at 200 kV (wavelength = 0.0251 Å). Diffraction images were recorded using Gatan US4000 (895) camera with a sensor size of 4096 × 4096 pixels. The eTasED software (developed by Dr. Xueming Li and available at http://github.com/THUEM/eTasED) was used to perform the semi-automated Micro-ED data collection. The nominal camera length was set to 520 mm, and calibrated by using an oriented-gold-film standard sample. Exposure time for each image frame was 4.7 s. Selected area aperture (10 μm in diameter) was used to limit the sample area for diffraction. The measured electron dose exposed on the sample is ~0.008 e− Å−2 s−1. During the exposure, crystals were continuously rotated with a speed of 0.28° per second. Therefore, each image frame covered 1.4° wedge, and each dataset from a single crystal covered a total 40°–80° angle range in reciprocal space. XDS was used to index the diffraction data52. Datasets from four crystals were merged together and scaled by XSCALE52 (Table 1).
For X-ray diffraction data collection, the crystals were diffracted on the micro-focus beam line BL-17U at Shanghai Synchrotron Radiation Facility (wavelength = 0.9791 Å). The crystals were cryo-cooled (100 K) during data collection. Data were processed by using XDS52 (Table 1). One crystal (one spot per crystal) was used for data collection.
For hnRAC1, direct method by SHLEXD53 was used to determine the phase information from electron diffraction data, and molecular replacement by Phaser54 from Phenixprogram package was used to determine the phase information from X-ray data. The initial searching model consists of a geometrically ideal β-strand composed of four alanine residues. For 234GGGYGGS240, phase information was determined by direct method using SHELXT/L55,56. Structure refinement was performed by using phenix.refine57. The atomic models were built by COOT58. The statistics of structure refinement were reported in Table 1. The structures were illustrated by PyMol (Schorödinger, LLC).
X-ray fibril diffraction
In total, 5 μl of fibril samples were deposited between two sealed capillaries and allowed fibril samples to align and dry at 4 °C overnight. The aligned fibrils were diffracted by an in-house Rigaku FR-E generator (wavelength 1.5418 Å) equipped with an R-AXIS IV + + imaging plate detector.
The structural models of hnRNPA1 hnRAC2 and hnRAC3 were built using Rosetta59 (Supplementary Note 1). The hnRACs were modeled as a β-sheet. The β-sheets were assembled by exploring all possible arrangements (different steric zipper classes28). The assembled fibril structures were then filtered and refined by simultaneously optimizing the rigid-body degree of freedom between symmetrical copies, side chain, and backbone torsions of hnRACs subunits, guided by full-atom Rosetta energy functions60. The intra-sheet and inter-sheet distances of hnRACs fibrils obtained from fibril diffraction were input as restrains. The top-ranking models based on Rosetta energy were selected.
Phase diagram measurement
In total, 100 μl of 100 μM hnRNPA1 samples were loaded into a 0.5 mm Cuvette. To measure the phase diagram, the temperature of the cuvette was gradually decreased from 20 °C to 4 °C at a rate of 1 °C min−1. Meanwhile, OD600 values were measured every minute during the temperature decrease by using Chirascan™ CD Spectrometer equipped with a temperature controlled device.
Stress-granule formation and immunofluorescence imaging
HeLa cell (Cell Bank of the Chinese Academy of Sciences, Shanghai, Cat. Lot: SCSP-504) culture was plated on coverslips for 18–24 h, and then transfected with hnRNPA1 plasmids with a pAGGS vector in DMEM by polyjet (SignaGen Laboratories) for 6 h. Forty-eight hours after transfection, 0.5 mM sodium arsenate was incubated with the cell culture for 1 h to induce stress granules. For immunofluorescence imaging, cells were fixed with a mixture of 4% paraformaldehyde and 4% glucose in PBS for 15 min, and permeabilized in 0.15% triton for 15 min. After blocking with 5% BSA, cells were co-incubated with anti-Flag antibody (Sigma-Aldrich, Cat. Lot: F1804) and anti-TIAR antibody (Cell Signaling Technology Cat. Lot: 8509 S) overnight at 4 °C. Secondary antibodies were conjugated to Alexa Fluor 488 (ThermoFisher, Cat. Lot: A-11029) or Alexa Fluor 594 (ThermoFisher, Cat. Lot: A-11012). The antibodies used were 1000 times diluted. Coverslips were mounted using ProLong™ Gold Antifade Mountant with DAPI (ThermoFisher Cat. Lot: P36935). Images were collected on a SP8 Leica microscope with a DMI8 camera and further analyzed by a Leica LAS AF Lite software. The capability of the recruitment of hnRNPA1 into stress granules was measure by calculating the percentage of the cells with Flag-hnRNPA1 in stress granules in the total number of cells with transfected Flag-hnRNPA1. A total number of ~300 hnRNPA1-transfected cells were calculated for each sample.
An optical tweezer microscope C-trap™ from LUMICKS (Amsterdam, The Netherlands) with two steerable traps was used to perform controlled fusion of droplets16. Sample chambers were prepared by first applying two thin pieces of double-stick transparent tape (~0.1 mm thick) in parallel within 5 mm separation to a coverslip, followed by placing a glass slide on top of the coverslip. Liquid droplets of hnRNPA1 and variants were formed with 180 μM of proteins at an environmental temperature of 13 °C. These droplets were flowed into the chamber just before data acquisition. A 1064 nm laser with low light intensity (< 0.5 W) was applied to minimize heating. One droplet was held in place by a trap, and the other steerable trap was used to capture other droplets and bring them toward the stationary droplet with a velocity of 0.04 μM s−1 until the two droplets surfaces touched. Fusion events were recorded simultaneously with high laser signal (50 kHz) and low video signal (30 Hz). Fusion times were determined via analysis from the laser signal and confirmed with video signal.
WT and mutants were labeled with Alexa Fluor 647 carboxylic acid, tris(triethylammonium) salt (ThermoFisher LOT: A33084) following the manufacturer’s instructions. For the FRAP assay, unlabeled 50 μM WT and mutants (25 mM Tris, 50 mM NaCl, 10% PEG 3350, pH 7.5) mixed with fluorophore-labeled proteins at a molar ratio of 50:1 were incubated at room temperature for 20 min. In all, 6 μl of the samples were loaded on an ultra-thin glass bottom dish. FRAP experiments were performed by using a Leica SP8 microscope with a ×100 oil objective and a 633 laser line. Approximately a 2.5 μm in diameter circular region within the droplets in size of 4–6 μm was bleached for 2.5 s with full laser power. Post-bleach images were acquired at 2.5 s per frame. Images were analyzed with the Leica Application Suite X software. For each time point, the intensity was corrected by the intensity of a neighboring unbleached region. Fluorescence recovery fraction was calculated with the formula (It − Iminimum)/(Ibefore bleaching − Iminimum). It is the intensity at individual time points.
ThT kinetic assay
Overall, 60 μl of samples containing 50 μM hnRNPA1 or variants and 50 μM ThT were shook at 900 rpm at 25 °C. ThT fluorescence was excited at 440 nm, and the emission wavelength of 485 nm was recorded per 5 ms by a Thermo Scientific Varioskan Flash spectrophotometer. Four replicates for each sample were performed.
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
The PDB accession code of hnRAC1 solved by Micro-ED is 6J60 and by X-ray diffraction is 5ZGD. The PDB accession code of non-RAC segment GGGYGGS is 5ZGL. The source data underlying Figs. 1d, 5a, 5c–e and Supplementary Figs. 2b, 3, 12, and 14 are provided as a Source Data file. A reporting summary for this article is available as a Supplementary Information file. All other data supporting the findings of this study are available from the corresponding authors on reasonable request.
Banani, S. F., Lee, H. O., Hyman, A. A. & Rosen, M. K. Biomolecular condensates: organizers of cellular biochemistry. Nat. Rev. Mol. Cell Biol. 18, 285–298 (2017).
Kato, M. & McKnight, S. L. A solid-state conceptualization of information transfer from gene to message to protein. Annu. Rev. Biochem. 87, 351–390 (2018).
Protter, D. S. & Parker, R. Principles and properties of stress granules. Trends Cell Biol. 26, 668–679 (2016).
Shukla, S. & Parker, R. Hypo- and hyper-assembly diseases of RNA-protein complexes. Trends Mol. Med. 22, 615–628 (2016).
Bentmann, E. et al. Requirements for stress granule recruitment of fused in sarcoma (FUS) and TAR DNA-binding protein of 43 kDa (TDP-43). J. Biol. Chem. 287, 23079–23094 (2012).
Ramaswami, M., Taylor, J. P. & Parker, R. Altered ribostasis: RNA-protein granules in degenerative disorders. Cell 154, 727–736 (2013).
Molliex, A. et al. Phase separation by low complexity domains promotes stress granule assembly and drives pathological fibrillization. Cell 163, 123–133 (2015).
Burke, K. A., Janke, A. M., Rhine, C. L. & Fawzi, N. L. Residue-by-residue view of in vitro FUS granules that bind the C-terminal domain of RNA polymerase II. Mol. Cell 60, 231–241 (2015).
Pak, C. W. et al. Sequence determinants of intracellular phase separation by complex coacervation of a disordered protein. Mol. Cell 63, 72–85 (2016).
King, O. D., Gitler, A. D. & Shorter, J. The tip of the iceberg: RNA-binding proteins with prion-like domains in neurodegenerative disease. Brain Res. 1462, 61–80 (2012).
Aguzzi, A. & Altmeyer, M. Phase separation: linking cellular compartmentalization to disease. Trends Cell Biol. 26, 547–558 (2016).
Kim, H. J. et al. Mutations in prion-like domains in hnRNPA2B1 and hnRNPA1 cause multisystem proteinopathy and ALS. Nature 495, 467–473 (2013).
Sun, Z. et al. Molecular determinants and genetic modifiers of aggregation and toxicity for the ALS disease protein FUS/TLS. PLoS Biol. 9, e1000614 (2011).
Johnson, B. S. et al. TDP-43 is intrinsically aggregation-prone, and amyotrophic lateral sclerosis-linked mutations accelerate aggregation and increase toxicity. J. Biol. Chem. 284, 20329–20339 (2009).
Lin, Y., Protter, D. S., Rosen, M. K. & Parker, R. Formation and maturation of phase-separated liquid droplets by RNA-binding proteins. Mol. Cell 60, 208–219 (2015).
Patel, A. et al. A liquid-to-solid phase transition of the ALS protein FUS accelerated by disease mutation. Cell 162, 1066–1077 (2015).
Eisenberg, D. & Jucker, M. The amyloid state of proteins in human diseases. Cell 148, 1188–1203 (2012).
Nelson, R. et al. Structure of the cross-beta spine of amyloid-like fibrils. Nature 435, 773–778 (2005).
Riek, R. & Eisenberg, D. S. The activities of amyloids from a structural perspective. Nature 539, 227–235 (2016).
Kato, M. et al. Cell-free formation of RNA granules: low complexity sequence domains form dynamic fibers within hydrogels. Cell 149, 753–767 (2012).
Han, T. W. et al. Cell-free formation of RNA granules: bound RNAs identify features and components of cellular assemblies. Cell 149, 768–779 (2012).
Kato, M. & McKnight, S. L. Cross-beta polymerization of low complexity sequence domains. Cold Spring Harb. Perspect. Biol. 9, a023598 (2017).
Hughes, M. P. et al. Atomic structures of low-complexity protein segments reveal kinked beta sheets that assemble networks. Science 359, 698–701 (2018).
Luo, F. et al. Atomic structures of FUS LC domain segments reveal bases for reversible amyloid fibril formation. Nat. Struct. Mol. Biol. 25, 341–346 (2018).
Kwon, I. et al. Phosphorylation-regulated binding of RNA polymerase II to fibrous polymers of low-complexity domains. Cell 155, 1049–1060 (2013).
Nott, T. J. et al. Phase transition of a disordered nuage protein generates environmentally responsive membraneless organelles. Mol. Cell 57, 936–947 (2015).
Martin, E. W. & Mittag, T. Relationship of sequence and phase separation in protein low-complexity regions. Biochemistry 57, 2478–2487 (2018).
Sawaya, M. R. et al. Atomic structures of amyloid cross-beta spines reveal varied steric zippers. Nature 447, 453–457 (2007).
Rodriguez, J. A. et al. Structure of the toxic core of alpha-synuclein from invisible crystals. Nature 525, 486–490 (2015).
Guenther, E. L. et al. Atomic structures of TDP-43 LCD segments and insights into reversible or pathogenic aggregation. Nat. Struct. Mol. Biol. 25, 463–471 (2018).
Geuens, T., Bouhy, D. & Timmerman, V. The hnRNP family: insights into their role in health and disease. Hum. Genet. 135, 851–867 (2016).
Fukuda, T., Naiki, T., Saito, M. & Irie, K. hnRNP K interacts with RNA binding motif protein 42 and functions in the maintenance of cellular ATP level during stress conditions. Genes Cells 14, 113–128 (2009).
Dombert, B., Sivadasan, R., Simon, C. M., Jablonka, S. & Sendtner, M. Presynaptic localization of Smn and hnRNP R in axon terminals of embryonic and postnatal mouse motoneurons. PLoS ONE 9, e110846 (2014).
Lee, Y. B. et al. Hexanucleotide repeats in ALS/FTD form length-dependent RNA foci, sequester RNA binding proteins, and are neurotoxic. Cell Rep. 5, 1178–1186 (2013).
Michael, W. M., Choi, M. & Dreyfuss, G. A nuclear export signal in hnRNP A1: a signal-mediated, temperature-dependent nuclear protein export pathway. Cell 83, 415–422 (1995).
Wheeler, J. R., Matheny, T., Jain, S., Abrisch, R. & Parker, R. Distinct stages in stress granule assembly and disassembly. Elife 5, e18413 (2016).
Jain, S. et al. ATPase-modulated stress granules contain a diverse proteome and substructure. Cell 164, 487–498 (2016).
West, J. A. et al. Structural, super-resolution microscopy analysis of paraspeckle nuclear body organization. J. Cell Biol. 214, 817–830 (2016).
Li, P. et al. Phase transitions in the assembly of multivalent signalling proteins. Nature 483, 336–340 (2012).
Li, J. et al. The RIP1/RIP3 necrosome forms a functional amyloid signaling complex required for programmed necrosis. Cell 150, 339–350 (2012).
Wu, H. & Fuxreiter, M. The structure and dynamics of higher-order assemblies: amyloids, signalosomes, and granules. Cell 165, 1055–1066 (2016).
Dai, B. et al. Tunable assembly of amyloid-forming peptides into nanosheets as a retrovirus carrier. Proc. Natl Acad. Sci. USA 112, 2996–3001 (2015).
Li, D. et al. Structure-based design of functional amyloid materials. J. Am. Chem. Soc. 136, 18044–18051 (2014).
Knowles, T. P. J., Vendruscolo, M. & Dobson, C. M. The amyloid state and its association with protein misfolding diseases. Nat. Rev. Mol. Cell Biol. 15, 384–396 (2014).
Li, Y. R., King, O. D., Shorter, J. & Gitler, A. D. Stress granules as crucibles of ALS pathogenesis. J. Cell Biol. 201, 361–372 (2013).
Tuttle, M. D. et al. Solid-state NMR structure of a pathogenic fibril of full-length human alpha-synuclein. Nat. Struct. Mol. Biol. 23, 409–415 (2016).
Qiang, W., Yau, W. M., Lu, J. X., Collinge, J. & Tycko, R. Structural variation in amyloid-beta fibrils from Alzheimer’s disease clinical subtypes. Nature 541, 217–221 (2017).
Stroud, J. C., Liu, C., Teng, P. K. & Eisenberg, D. Toxic fibrillar oligomers of amyloid-beta have cross-beta structure. Proc. Natl Acad. Sci. U. S. A. 109, 7717–7722 (2012).
Liu, C. et al. Out-of-register beta-sheets suggest a pathway to toxic amyloid aggregates. Proc. Natl Acad. Sci. USA 109, 20913–20918 (2012).
Laganowsky, A. et al. Atomic view of a toxic amyloid small oligomer. Science 335, 1228–1231 (2012).
Murray, D. T. et al. Structural characterization of the D290V mutation site in hnRNPA2 low-complexity-domain polymers. Proc. Natl Acad. Sci. USA 115, E9782–E9791 (2018).
Kabsch, W. XDS. Acta Crystallogr. D: Biol. Crystallogr. 66, 125–132 (2010).
Usón, I. & Sheldrick, G. M. Advances in direct methods for protein crystallography. Curr. Opin. Struct. Biol. 9, 643–648 (1999).
McCoy, A. J. et al. Phaser crystallographic software. J. Appl. Crystallogr. 40, 658–674 (2007).
Sheldrick, G. M. SHELXT integrated space-group and crystal-structure determination. Acta Crystallogr. A: Found. Adv. 71, 3–8 (2015).
Sheldrick, G. M. Crystal structure refinement with SHELXL. Acta Crystallogr. C: Struct. Chem. 71, 3–8 (2015).
Afonine, P. V. et al. Towards automated crystallographic structure refinement with phenix.refine. Acta Crystallogr. D: Biol. Crystallogr. 68, 352–367 (2012).
Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr. D: Biol. Crystallogr. 66, 486–501 (2010).
Leaver-Fay, A. et al. ROSETTA3: an object-oriented software suite for the simulation and design of macromolecules. Methods Enzymol. 487, 545–574 (2011).
Andre, I., Bradley, P., Wang, C. & Baker, D. Prediction of the structure of symmetrical protein assemblies. Proc. Natl Acad. Sci. USA 104, 17656–17661 (2007).
We thank assistance from staff in Shanghai Synchrotron Radiation Facility BL17U for X-ray data collection, Tsinghua University Branch of China National Center for Protein Sciences Beijing for providing facility supports on CryoEM, and Dr. Zhipu Luo from Soochow University for his help on optimizing Micro-ED data processing. This work was supported by the Major State Basic Research Development Program (2016YFA0501902 to C. Liu and X. Li, and 2016YFA0501902 to X. Li), Shanghai Pujiang Program (18PJ1404300 to D. Li), “Eastern Scholar” project supported by Shanghai Municipal Education Commission to D. Li, Innovation Program of Shanghai Municipal Education Commission 2019-01-07-00-02-E00037 to D. Li, the National Natural Science Foundation (NSF) of China (31872716, 91853113 to C. Liu and D. Li), Shanghai Sailing Program (16YF1414500 to B. Dai).
The authors declare no competing interests.
Journal peer review information: Nature Communications thanks Alf Honigmann and the other anonymous reviewer(s) for their contribution to the peer review of this work.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.