Article https://doi.org/10.1038/s41467-023-35854-0 Cryo-EM structure of hnRNPDL-2 fibrils, a functional amyloid associated with limb- girdle muscular dystrophy D3 Received: 5 August 2022 Javier Garcia-Pardo 1, Andrea Bartolomé-Nafría 1, Antonio Chaves-Sanjuan 2,3, Marcos Gil-Garcia 1, Cristina Visentin2,4, Accepted: 4 January 2023 Martino Bolognesi 2,3, Stefano Ricagno 2,4 & Salvador Ventura 1 Check for updates hnRNPDL is a ribonucleoprotein (RNP) involved in transcription and RNA- 1234567890():,; 1234567890():,; processing that hosts missense mutations causing limb-girdle muscular dys- trophy D3 (LGMD D3). Mammalian-specific alternative splicing (AS) renders three natural isoforms, hnRNPDL-2 being predominant in humans. We present the cryo-electron microscopy structure of full-length hnRNPDL-2 amyloid fibrils, which are stable, non-toxic, and bind nucleic acids. The high-resolution amyloid core consists of a single Gly/Tyr-rich and highly hydrophilic filament containing internal water channels. The RNA binding domains are located as a solenoidal coat around the core. The architecture and activity of hnRNPDL-2 fibrils are reminiscent of functional amyloids, our results suggesting that LGMD D3 might be a loss-of-function disease associated with impaired fibril- lation. Strikingly, the fibril core matches exon 6, absent in the soluble hnRNPDL-3 isoform. This provides structural evidence for AS controlling hnRNPDL assembly by precisely including/skipping an amyloid exon, a mechanism that holds the potential to generate functional diversity in RNPs. Human heterogeneous ribonucleoprotein D-like (hnRNPDL) belongs than hnRNPDL-2 and is present mainly in brain and testis6. hnRNPDL-3 to a class of conserved nuclear RNA-binding proteins (RBPs) that is a shorter and minor isoform of 244 amino acids that lacks the N- and assemble with RNA to form ribonucleoproteins (RNPs). hnRNPDL acts C- terminal LCDs but conserves the PY-NLS7. as a transcriptional regulator and participates in the metabolism and A point mutation in hnRNPDL exon 6 causes autosomal dominant biogenesis of mRNA1–4. limb-girdle muscular dystrophy D3 (LGMD D3)8–10, a rare disease Three isoforms of hnRNPDL are produced by alternative splicing characterized by slowly progressive proximal muscle weakness11,12. (AS): hnRNPDL-1, hnRNPDL-2, and hnRNPDL-35 (Fig. 1a and Supple- This mutation changes the conserved Asp259 in the C-terminal LCD to mentary Fig. 1). hnRNPDL-2 is a 301 residues protein and the pre- either Asn or His. Similarly, mutation of specific Asp residues to Asn or dominant isoform in human tissues6. It consists of two consecutive Val in the LCDs of hnRNPA1 and hnRNPA2 are linked to amyotrophic globular RNA recognition motifs (RRM1 and RRM2), followed by a lateral sclerosis (ALS) and multisystem proteinopathy (MSP)13,14. How- C-terminal low-complexity domain (LCD) that maps at residues ever, unlike ALS and MSP patients, where hnRNPA1 and hnRNPA2 ~201–285 and a nuclear localization sequence (PY-NLS) comprising accumulate in cytoplasmic inclusions in muscular fibers14,15, most residues 281-301. hnRNPDL-1 is a longer isoform of 420 amino acids LGMD D3 patients do not exhibit nuclear or cytoplasmatic protein containing an additional Arg-rich N-terminal LCD5; it is less abundant inclusions8. 1 Institut de Biotecnologia i de Biomedicina (IBB) and Departament de Bioquímica i Biologia Molecular, Universitat Autònoma de Barcelona, 08193 Bellaterra, Barcelona, Spain. 2Dipartimento di Bioscienze, Università degli Studi di Milano, 20133 Milano, Italy. 3CRC Fondazione Romeo e Enrica Invernizzi and NOLIMITS, Università degli Studi di Milano, 20133 Milano, Italy. 4Institute of Molecular and Translational Cardiology, I.R.C.C.S. Policlinico San Donato, piazza Malan, 2, 20097 San Donato Milanese, Italy. e-mail: Salvador.Ventura@uab.cat Nature Communications | (2023)14:239 1 Structures of the amyloid fibrils formed by hnRNPA1 and protein region responsible for amyloid formation in hnRNPDL-2. In hnRNPA2 LCDs have been recently determined16,17. For both fibrils, the fact, exon 6 is missing in hnRNPDL-3, and the hnRNPDL-1 N-terminal PY-NLS was embedded within the ordered core. This led to suggest Arg-rich LCD may effectively counteract exon 6 amyloidogenic pro- that, under physiological conditions, binding of the import receptor pensity, likely by diverting the assembly towards phase-separated karyopherin-β2 (Kapβ2) to PY-NLS18 would impede fibrillation, whereas condensates3. its exposure under pathological conditions would allow amyloid Once formed, hnRNPDL-2 fibrils are stable and insensitive to salt formation16. However, in these studies, constructs of the LCD, alone or or temperature (Supplementary Fig. 7). Remarkably, these in vitro fused to a fluorescent protein, were used to form the fibrils, and it is formed structures are devoid of toxicity for different human cell lines unknown whether the observed assemblies would match those of the up to 0.4 mg/ml (Supplementary Fig. 8), which is consistent with the natural full-length proteins’ fibrils. observation that >30% of endogenous hnRNPDL-2 accumulates into Here we present the cryo-electron microscopy (cryo-EM) struc- the detergent-insoluble fraction of wild-type HeLa cells (Supplemen- ture of the fibrils formed by the full-length hnRNPDL-2 isoform at an tary Fig. 9). overall resolution of 2.5 Å. These fibrils are stable and bind oligonu- cleotides, with the associated RRM domains building an exposed Cryo-EM structure determination of hnRNPDL-2 amyloid fibrils solenoidal coat wrapping the structured fibril. The fibrils’ core is We used cryo-EM to investigate the molecular structure of hnRNPDL-2 formed by a single highly hydrophilic filament encompassing LCD amyloid fibrils (Fig. 2 and Supplementary Fig. 10). Two-dimensional residues 226-276, including Asp259. Modeling suggests that disease- (2D) classification yielded one single species of twisted fibrils. About associated mutations at this residue may have limited impact on fibril 54,500 fibril segments from 1,114 micrographs were selected for helical stability. Importantly, the fibril core does not include the PY-NLS. The reconstruction in RELION 3.121,22, allowing us to obtain a three- hnRNPDL-2 fibril core precisely matches exon 6 (residues 224-280), dimensional (3D) density map of hnRNPDL-2 at an overall resolution which is alternatively spliced (AS) in a mammalian-specific manner19. of 2.5 Å (Fig. 2b, Fig. 2c and Supplementary Fig. 11). The fibril consists of These AS events are frequent in RNPs, especially at their Y/G-rich a single protofilament where hnRNPDL-2 subunits stack along the fibril LCDs19 and they have been proposed to be a way to regulate protein (Z) elongation axis. The fibril forms a left-handed helix with a full pitch function by controlling the formation of high-order assemblies20; our of ~357 Å, a helical rise of 4.82 Å, and a helical turn of −4.86°, featuring a results provide structural evidence supporting this hypothesis. typical cross-β amyloid structure (Figs. 2c–f). A molecular model could Overall, we describe the structure of a full-length RNP in its be unambiguously built de novo for the fibrils core, spanning fibrillar functional conformation, providing insight into the molecular hnRNPDL-2 residues 226-276 (Figs. 2d–e) and matching strikingly the bases of LGMD D3, and illustrating how AS can control RNPs assembly region of exon 6 (residues 224-280), which is absent in the non- by including/excluding amyloidogenic exons at their LCDs. amyloidogenic hnRNPDL-3 isoform (Fig. 1a). The fibril core hosts six individual β-strands (β1 to β6) connected by β-turns and loops, yield- Results ing a sort of M-shaped fold, with approximate overall footprint of hnRNPDLs localize at the nucleus and exon 6 is key for their 55 Å x 40 Å (Figs. 2c–e). Data collection and refinement statistics are compartmentalization summarized in Supplementary Table 1. Human hnRNPDL AS renders three naturally occurring transcripts (Fig. 1a and Supplementary Fig. 1). When transfected into a HeLa Structural features of hnRNPDL-2 amyloid fibrils hnRNPDL knockout (KO) cell line (Supplementary Fig. 2), the three The core of hnRNPDL-2 amyloid fibrils, as well as exon 6, map to the ectopically expressed isoforms accumulated in the nucleus (Fig. 1b and C-terminal Y/G-rich LCD, comprising residues ~201–285. Accordingly, Supplementary Fig. 3), consistent with sharing a functional PY-NLS7. the filament amino acid composition is skewed towards Gly (27%) and hnRNPDL-1 and hnRNPDL-2 exhibited a granulated nuclear distribution Tyr (25%), which comprise >50% of the residues in the structure, and and were excluded from the nucleolar regions. In contrast, hnRNPDL-3 neutral-polar residues, with Asn, Gln, and Ser summing to 32%. Thus, was homogeneously distributed in both the nucleolus and the each layer of the amyloid core is essentially stabilized through intra- nucleoplasm. This indicates that exon 6 (residues 224-280), shared by subunit hydrogen bonds involving buried polar side chains, main- hnRNPDL-1 and hnRNPDL-2 and absent in hnRNPDL-3, is responsible chain peptide groups, and ordered solvent molecules (Fig. 3a). The for the differential intranuclear compartmentalization of the isoforms, individual protein subunits pack on each other through an extensive delimiting regions with higher and lower protein concentration. hydrogen-bonding network holding the in-register β-strands together, in the typical cross-β arrangement (Supplementary Fig. 12a). Addi- hnRNPDL-2 forms stable and non-toxic amyloid fibrils tionally, the Tyr aromatic side chains build ladders of π-π stacking We purified monomeric hnRNPDL-1, hnRNPDL-2, and hnRNPDL-3 by interactions along the fibril elongation direction, stabilizing the inter- size-exclusion chromatography. Upon incubation, hnRNPDL-2 forms subunit association interface. Tyr residues are approximately equally amyloid fibrils (Supplementary Fig. 4), in a reaction that is highly distributed between the filament interior and the surface. The core sensitive to the solution ionic strength (Supplementary Fig. 5b). At pH region N-end contains two additional aromatic residues, W227 and 7.5 and low salt (50 mM NaCl), aggregation was fast, rendering large F231, which also form individual ladders of π-π stacking interactions assemblies that precipitated as the reaction progressed. In contrast, at along the fibrillar axis; surprisingly, F231 is exposed on the fibril sur- 300 mM NaCl, the reaction exhibited characteristic sigmoidal kinetics, face, and not buried within the core. Other hydrophobic residues such with the formation of highly ordered individual amyloid fibrils that as Val, Leu, and Ile are absent, whereas they are frequently observed in strongly bind Thioflavin-T (Th-T) (Figs. 1c, e). A further increase in ionic the structure of pathological fibrils23. Gly residues are dispersed strength (i.e. 600 mM NaCl) drastically delayed and reduced hnRNPDL- throughout the sequence, with their peptide units also involved in 2 amyloid formation (Supplementary Fig. 5b). Thus, the fibrils formed hydrogen bonding. The main chain of each individual subunit stacked at pH 7.5 and NaCl 300 mM were selected for structural studies. in the fibril does not lay in a plane, with a 10 Å distance between the top Importantly, under the same conditions, monomeric hnRNPDL-1 and of β1 strand and the edge of the turn immediately after β4 (Fig. 2c). As a hnRNPDL-3 did not aggregate into Th-T positive assemblies after 24 h result, each subunit (i) not only interacts with the layer directly above (Supplementary Fig. 4b), 48 h (Fig. 1c) or even after incubation for (i + 1) and below (i − 1), but also with layers (i + 3) and (i − 3). This is best 7 days (Supplementary Fig. 6). Additionally, these two isoforms did not exemplified by the formation of a strip of hydrogen bonds involving form amyloids upon incubation in the presence of 50 mM or 600 mM the side chains of Q237 in layer i and Y258 in layer i-3 (Supplementary NaCl (Supplementary Fig. 5a, c). These results point to exon 6 as the Fig. 12b-c). Nature Communications | (2023)14:239 2 Article https://doi.org/10.1038/s41467-023-35854-0 a b DL-1 DL-2 NLS NL Y/G-rich hnRNPDL-1 1 R-rich LCD RRM RRM LCD 420 NLS NL Y/G-rich hnRNPDL-2 1 RRM RRM LCD 301 10 μm 10 μm NLS hnRNPDL-3 1 RRM RRM 244 DL-3 GFP Skipped Exon 6 224 - GQNWNQGFNNYYDQGYGNYNSAYGGDQNYSGYGGYDYTGYNYGNYGYGQGYADYSGQ - 280 10 μm 10 μm d e f c 200 nm 200 nm 200 nm Fig. 1 | hnRNPDL-2 aggregates into ordered amyloid fibrils. a Domain organi- nuclear distribution of GFP-hnRNPDL fusions (green). Nuclear DNA was stained zation of human hnRNPDL isoforms. The RNA recognition motifs are labelled as with Hoechst (blue). White arrows indicate the location of the nucleolus. In all RRM (pink) and N- and C-terminal low complexity domains (LCD) are indicated as cases, the nucleus contour has been indicated with a white dashed line. Scale bar, R-rich LCD (orange) and Y/G-rich LCD (blue), respectively. The region and sequence 10 μm. c Th-T binding to hnRNPDL-1 (orange line), hnRNPDL-2 (blue line) and of exon 6 is also indicated, according to hnRNPDL-2 numbering. D259 disease- hnRNPDL-3 (purple line) after incubation at 37 °C and 600 rpm, pH 7.5 and 300 mM associated amino acid is shown in red. NLS indicates the shared nuclear localization NaCl for 2 days. Representative negative staining TEM micrographs of the incu- signal. b Representative confocal images of HeLa hnRNPDL KO cells transiently bated solutions of (d) hnRNPDL-1, (e) hnRNPDL-2 and (f) hnRNPDL-3. Scale bar, transfected with the different hnRNPDL isoforms (DL-1, DL-2 and DL-3). The empty 200 nm. Note that only in (e) individual amyloid filaments are evident. In (b) and GFP vector (GFP) was transfected as control. The images depict the characteristic (d–f), results are representative from three independent experiments. Unlike most amyloid structures, the conformation of hnRNPDL-2 with the values observed for disease-associated fibrils (Supplementary residues N226-D276 outlines pores that define two internal water Table 3). channels. The major of such channels spans the fibril’s entire length The proportion of exposed polar atoms and polar buried residues and contains additional densities that correspond to two ordered in the inner layers of hnRNPDL-2 are also higher than in the hnRNPA1 water molecules per layer, that are H-bonded to nearby polar groups of and hnRNPA2 LCDs fibrils. Overall, hnRNPDL-2 appears to assemble Y239, N241, and with peptide N-atoms of G256 (Supplementary into one of the most hydrophilic fibrils described so far, providing Fig. 12d). According to our experimental estimates, the cryo-EM den- donor and acceptor groups for potential interactions. sity around the channel shows one of the highest resolutions within the We calculated the solvation free energy of folding (ΔG) and ΔG/ entire fibril structure, indicating high local stability (Supplementary residue for hnRNPDL-2 fibrils (Supplementary Fig. 14 and Supple- Fig. 11a). On average, each fibril layer encloses ten ordered water mentary Table 4). These values are significantly lower than those of molecules. disease-associated fibrils, independently if they were obtained in vitro, hnRNPDL-2 fibrils are mainly composed of hydrophilic amino ex vivo or upon in vitro seeding with ex vivo fibrils, indicating that acids and display more polar surfaces than the fibrils of typical despite their irreversibility, hydrophilic hnRNPDL-2 fibrils are less pathogenic proteins (Supplementary Fig. 13). The solvent-accessible stable. surface area (SASA) of the hnRNPDL-2 fibril upper layer is 3458 Å2, 48% Importantly, in contrast to the hnRNPA1 and hnRNPA2 LCD of which is covered by polar atoms, with a percentage of polar residues fibrils16,17, the PY-NLS in hnRNPDL-2 fibrils is adjacent to, but not part of in the buried area of 64%; both values being significantly higher than in the structural core (Supplementary Fig. 15), suggesting that binding of disease-associated fibrils (Supplementary Table 2). The SASA of inter- Kapβ2 would not necessarily hamper fibrillation. nal hnRNPDL-2 layers is, on average, 1294 Å2, reflecting a burial of 63% of the surface relative to the end solvent-exposed layers, 57% of Disease-causative hereditary mutations in the hnRNPDL-2 fibril exposed atoms being polar. This endorses the lateral surface of the structure fibril with a high hydrophilic character (Supplementary Fig. 13a). In Two missense mutations in hnRNPDL exon 6, D259N and D259H, are contrast, in the lateral surfaces of pathogenic fibrils, non-polar atoms linked to LGMD D38–10. D259 is strictly conserved in the hnRNPDL are predominant (Supplementary Fig. 13). The percentage of polar C-terminal LCD of vertebrates (Fig. 3b) and maps at the end of the loop residues in the buried area (67%) is exceedingly high when compared connecting β4 to β5 in the hnRNPDL-2 fibril (Fig. 3c). This residue is Nature Communications | (2023)14:239 3 Article https://doi.org/10.1038/s41467-023-35854-0 a b 50 nm Pitch = 357 Å c β1 β4 β5 N226 D276 β2 β3 β6 4.82 Å 10 Å d 224 - GQNWNQGFNNYYDQGYGNYNSAYGGDQNYSGYGGYDYTGYNYGNYGYGQGYADYSGQ - 280 β1 β2 β-t β3 β-t β4 β5 β6 e D259 f 90 ° D276 Ct N226 Nt Fig. 2 | Structure of hnRNPDL-2 amyloid filaments. a Cryo-electron micrograph three stacked rungs comprising residues N226-D276 (right panel). The distance of hnRNPDL-2 filaments. Representative image from 1114 micrographs. The inset between the stacks is indicated in Å. d Sequence of hnRNPDL exon 6, according to shows a representative reference-free 2D class average image of the hnRNPDL-2 hnRNPDL-2 numbering. The observed β-strands that build the hnRNPDL-2 fibrils filament. Scale bar, 50 nm. b Side view of the three-dimensional cryo-EM density amyloid core are indicated. e Cryo-EM density map of a layer of hnRNPDL-2 amyloid map showing the structured core of an individual hnRNPDL-2 filament. The fibril core. Fitting of the atomic model for residues N226-D276 is shown on top. D259 pitch is indicated. c Detailed view of the cryo-EM density map showing the layer disease-associated amino acid is highlighted in red. Water molecules are shown as packing in the hnRNPDL-2 filament (left panel). The filament rise/subunit is indi- red spheres. f Representative 3D class average image of the hnRNPDL-2 amy- cated in Å. Rendered side view of the secondary structure elements accounting for loid fibril. solvent exposed, with 54% of its surface accessible in the inner fibril This distinguishes hnRNPDL-2 from hnRNPA1 and hnRNPA2 LCDs fibril layers, yielding a negatively charged ladder along the fibril surface cores, with calculated pIs of 6.0 and 8.4, respectively. (Figs. 3c and 3d). Apart from D259, the fibril core contains three As in hnRNPDL-2, mutation of a conserved Asp at the fibrillar core additional exposed Asp residues, D236, D249, and D276, and no of hnRNPA2 LCD to Val (D290V) or to Asn and Val in hnRNPA1 LCD positively charged amino acid, which results in a calculated single-layer (D262N/V) are disease-associated. Virtual mutations of the respective pI of 3.3 and a highly anionic patch extended along the fibrillar axis. amyloid cores16,17 indicated that they would render more stable fibrils Nature Communications | (2023)14:239 4 Article https://doi.org/10.1038/s41467-023-35854-0 Fig. 3 | Overall structure of the hnRNPDL-2 fibril core. a Schematic representa- the charge of an electron and a T is 298 K). The location of the six β-strands (β1 to tion of one cross-sectional layer of the hnRNPDL-2 fibril core. The location of the β- β6) is indicated. The side chain of D259 is shown as blue sticks and labeled. d Stick strands is indicated with thicker arrows. Polar and hydrophobic residues are model of one hnRNPDL-2 fibril rung showing the position of evolutionary con- colored in green and white, respectively. Glycine residues are colored in yellow. served residues (colored in pink). D259 disease-associated residue is highlighted in Aspartic residues are colored in red. b Sequence alignment of hnRNPDL ortholo- red. e Upper panel, top view of the Y255 to Y260 segment showing the interlayer gues showing evolutionary conservation of residues within exon 6. The β-strands interactions between Y260 and G256. Lower panel, close-up view perpendicular to that build the hnRNPDL-2 fibrils amyloid core are indicated. The mutated D259 and the fibril axis showing the Y260 packing and their interactions with the main chain surrounding conserved residues are shown in red and blue, respectively. Con- of G256. Y260 side chains are shown as sticks (colored in cyan) over the cryo-EM served residues are indicated with an asterisk. c Surface representation showing the map sown as a grey mesh. f Effect of LGMDD3-associated mutations on hnRNPDL-2 electrostatic surface potential of the hnRNPDL-2 fibril at pH 7, with ribbon repre- fibril formation. The WT hnRNPDL-2 and D259H and D259N mutant proteins were sentation of one subunit on top. Electrostatic potential maps were calculated using incubated under the same conditions and subjected to separation on a glycerol APBS server and visualized using the APBS plugin in PyMol (Schrödinger, NY, cushion and SDS–PAGE analysis (fractions from top to bottom, 1–6). Bottom USA)52,59. Negative and positive map potential values are colored in red and blue, panels, representative NS-TEM images of amyloid filaments from the bottom respectively, according to the kbTec−1 unit scale (kb is the Boltzmann’s constant, ec is fraction. This experiment has been performed twice with similar results. by removing charge repulsions along the structure, explaining the the impact of the D259H mutation in the hnRNPDL-2 fibril predicts that presence of inclusions of these RNPs variants in the tissues of ALS or it would be destabilizing, whereas the stabilization predicted for MSP patients14. However, this mechanism would not necessarily apply D259N is lower than for the D262N/V or D290V mutations in hnRNPA1 to hnRNPDL-2; firstly, because the mutation causing more severe and and hnRNPA2 fibrils, respectively (Supplementary Table 5). It has been earliest LGMD D3 onset, D259H, does not neutralize the negative recently proposed that mutations promoting the conversion of low- charge but reverts it, and secondly because the D259N mutation still complexity amyloid-like kinked segments (LARKS) into steric zippers leaves three other Asp residues exposed to solvent, the core surface underlie irreversible aggregation and disease24. The impact of the remaining highly acidic, with a calculated pI of 3.3. Indeed, modeling D290V mutation in hnRNPA2 LCD seems to respond to this Nature Communications | (2023)14:239 5 Article https://doi.org/10.1038/s41467-023-35854-0 mechanism24. However, in hnRNPDL, D259H/N mutations do not to bind nucleic acids, in keeping with a potential functional role for increase the propensity to form steric zippers above the detection these self-assembled structures. We have shown that the amyloid fibril threshold (Supplementary Fig. 16). These analyses suggested that in surface is strongly acidic; we therefore expect the globular RRM LGMD D3, disease-associated variants may not necessarily act by domains that decorate the fibril to be responsible for such activity. increasing the fibril stability or amyloid propensity. We produced and purified monomeric hnRNPDL-2 D259H and Discussion D259N variants and incubated them under the same conditions used to hnRNPDL-2 is the major hnRNPDL isoform in human tissues6 and, as we produce the wild-type (WT) fibrils described in the previous section. show here, the only one forming ordered amyloid fibrils under When the species in the incubated samples were separated on a gly- physiologic-like conditions. This property can unequivocally be cerol cushion, it could be observed that the formation of high- attributed to exon 6, which also accounts for the granular and het- molecular weight species was dramatically reduced in the disease- erogeneous protein distribution in the cell nucleus. Importantly, associated mutants, compared with WT (Fig. 3f). Moreover, TEM hnRNPDL-1 does not form fibrils, although it phase separates under the imaging of the high-molecular-weight fraction evidenced that, in same conditions3. Thus, hnRNPDL constitutes a notable exception to contrast to the copious and long fibrils present in WT, the mutants the general rule considering liquid-liquid phase separation (LLPS) and exhibited scarce and significantly shorter fibrils (Fig. 3f), displaying amyloid formation as two interconnected phenomena27,28. This might poor binding to Th-T (Supplementary Fig. 17). WT hnRNPDL-2 exhib- be a way to evolve two structurally different assemblies, condensates ited a strict bimodal distribution, being present only in the low- and and fibrils, each associated with an hnRNPDL isoform, skipping high-molecular weight fractions, indicating that they constitute the potential pathogenic transitions between these conformational states. predominant metastable states of hnRNPDL-2. These results question The in vitro cryo-EM hnRNPDL-2 fibrillar structure we report here an amyloid origin of LGMD D3 and are consistent with the absence of exemplifies the amyloid assembly of a full-length RNP, whereas pre- mutated protein inclusions in the atrophied muscle of patients8. vious fibrils were solved starting from the LCD alone16,29,30 or the LCD fused to a fluorescent protein17,31. Thus, we think it should better reflect hnRNPDL-2 amyloid fibrils bind nucleic acids the in vivo fibrillar packing of this protein family. Indeed, the hnRNPDL-2 contains two N-terminal tandem RRM domains (RRM1 and hnRNPDL-2 fibril structure exhibits significant differences relative to RRM2) that are thought to be functional RNA/DNA-binding motifs25 those of hnRNPA1 and hnRNPA2 LCDs, even though the respective full- (Fig. 1a and Supplementary Fig. 18). As shown in Figs. 4a and 4b, the length soluble forms share the same overall molecular architecture32,33. RRMs were visible in hnRNPDL-2 fibrils 2D classes, when a small In particular, the hnRNPDL-2 amyloid core matches a vertebrate con- number of segments were averaged, as additional fuzzy globular served exon, whereas this is not the case for hnRNPA1 and hnRNPA2 densities around the filament core; such densities are averaged out in (Supplementary Fig. 20). Secondly, the hnRNPDL-2 fibril is significantly the 3D reconstruction due to irregular locations of the RRMs along the more hydrophilic and acidic on its surface, a property directly fibril. Importantly, the RRMs surrounding the fibrillar core could be impacting on the interaction with other (macro)molecules. Thirdly, evidenced by immunogold labeling in TEM images (Fig. 4a). We then hnRNPDL-2 fibrils are irreversible and cannot be disassembled at propose a model whereby the structured fibril core, built by exon 6 90 °C, whereas those of hnRNPA1 and hnRNPA2 LCDs are reversible16,17. residues, is decorated by a fuzzy coat of flexible RRMs (Fig. 4c). Indeed, ΔG and ΔG/residue values for hnRNPDL-2 fibrils are more In this respect, it should be noted that residue F231 appears negative than for hnRNPA1 and hnRNPA2 structures (Supplementary exposed to solvent, giving rise to a hydrophobic ladder along the fibril Table 4). In addition, our amyloid includes the complete protein, and hydrophilic surface. Closer inspection of the density map, however, contacts between the RRMs and the core may also contribute to sta- suggests that the N-terminal arm of the fibril core, including F231, bility. Indeed, reversibility of hnRNPA1 and hnRNPA2 fibrils is expected interacts with other regions of the hnRNPDL-2 assembly, likely the since, for them, LLPS and fibril formation are interrelated, and they can fuzzy RRMs (Fig. 4d), that would preclude F231 entropically disfavored potentially transition between the two states. Such connection does interaction with the solvent. not apply in the case of hnRNPDL isoforms, and because hnRNPDL-2 Previous studies have shown that hnRNPDL is a protein that does not phase separate, no back-transition to liquid droplets is pos- actively participates in transcription and AS regulation1–4. A prior sible. The equilibrium should be established between the amyloid and investigation indicated that this protein binds to oligonucleotides with monomeric states, which fits well with our sedimentation assays. the consensus sequence ACUAGC, as deduced from the screening of a A final and important difference between hnRNPA1, hnRNPA2, set of in vivo identified RNA ligands26. To confirm that soluble and hnRNPDL-2 fibril structures is that, in the first two proteins, the PY- hnRNPDL-2 can bind nucleic acids, we performed Electrophoretic NLS residues that bind to importin Kapβ2 establish stabilizing inter- Mobility Shift Binding Assays (EMSA). We found that the soluble pro- actions and are buried inside the fibril core16,17; this does not occur in tein binds to a 7-mer fluorescently labeled ssDNA oligonucleotide hnRNPDL-2, where PY-NLS is adjacent but external to the amyloid core displaying the ACUAGC motif with an apparent dissociation constant (not mapped by density). The accessibility of PY-NLS in both mono- (Kd) of 5.9 ± 0.78 µM (4f, g), whereas the affinity of the protein for an meric and fibrillar states of hnRNPDL-2 is consistent with the obser- equivalent RNA sequence was lower (Supplementary Fig. 19). vation that, in humans, Kapβ2 co-localizes with both condensed and We incubated the same ssDNA oligonucleotide with preformed diffuse hnRNPDL nuclear regions8. This does not exclude that Kapβ2 hnRNPDL-2 amyloid fibrils, and we quantified the amount of oligonu- might still modulate hnRNPDL-2 fibril formation, sterically interfering cleotide bound to the fibrils after centrifugation. As shown in Fig. 4h, with the building of the amyloid core; however, this action would be the fibrils significantly bind ssDNA in a concentration-dependent mechanistically different from those exerted on hnRNPA1 and manner, showing a Kd of 2.1 ± 0.36 µM. This interaction was confirmed hnRNPA2, where Kapβ2 also acts as disaggregase34. by confocal microscopy, as incubated hnRNPDL-2 amyloids appeared hnRNPDL-2 fibrils share similarities with the 3D structures of highly fluorescent due to the incorporation of the fluorescein-labeled functional amyloids. Like most of them35–37, hnRNPDL-2 fibrils do not ssDNA (Fig. 4i). It could be that longer nucleic acids sequences, as exhibit polymorphism. This suggests that they may represent a global those found in nature, would result in tighter binding, because multi- free energy minimum, allowing us to speculate that the same structure valency allows weak interactions to collectively form much stronger would be adopted in the cell and that it represents the functional interactions, a phenomenon known as avidity. Unfortunately, specific conformation of the assembled state. In contrast, pathogenic fibrils are long RNA/DNA hnRNPDL targets remain to be identified. In any case, mostly polymorphic23. The hnRNPDL-2 fibril core is hydrophilic, and our data demonstrates that hnRNPDL-2 amyloid fibrils retain the ability the structure is stabilized by hydrogen-bonding networks, between Nature Communications | (2023)14:239 6 Article https://doi.org/10.1038/s41467-023-35854-0 a b c d Fibril axis RRM1 D276 RRM2 N226 Amyloid core 10 x F231 (48.2 Å) e RRM1 RRM2 Nt 50 nm Ct hnRNPDL-2 concentration (μM) f g A sDN 2 0.25 DL- F-s 0.5 1.5 2.5 3.5 7.5 10 15 20 40 60 1 5 C Kd = 5.9 ± 0.78 μM 7 nt – F h i + F-ssDNA Control Fibers 5 μm Kd = 2.1 ± 0.36 μM 40 μm 40 μm Fig. 4 | Structure of the RNA/DNA-binding domains from hnRNPDL-2 and their the N-terminal RNA-binding domains RRM1 and RRM2 from hnRNPDL-2 generated location in the amyloid filaments. a Top panel, representative 2D class average with AlphaFold260. f Electrophoretic mobility shift assay (EMSA) of soluble image showing hnRNPDL-2 fibrils with an approximate solenoid coat of globular hnRNPDL-2 with a Fluorescein-labelled oligonucleotide (F-ssDNA). The 7-mer domains. The arrows indicate the location of the density assigned to the RRMs. ssDNA was incubated with the soluble form of hnRNPDL-2 at the indicated protein Bottom panel, negative-stain electron microscopy (EM) micrographs of hnRNPDL-2 concentrations. EMSA has been performed three times with similar results. filaments bound to nanogold-antibodies (~10 nm, white arrow) targeting the loca- g Binding affinity of soluble hnRNPDL-2 to the 7-mer fluorescent ssDNA (F-ssDNA) tion of the hnRNPDL-2 globular domains. The N-terminal 6xHis tag was labelled determined by the EMSA assay. Data is shown as mean ± SEM (n = 3 independent using nanogold-conjugated secondary antibodies against anti-6xHis antibodies experiments). h Binding of the 7-mer fluorescent ssDNA (F-ssDNA) to preformed produced in mouse. Representative image from two independent experiments. hnRNPDL-2 amyloid filaments. Data is shown as mean ± SEM (n = 4 independent Scale bar, 50 nm. b 3D unsharpened density map reconstruction of hnRNPDL-2 experiments for all protein concentrations, except for 10 and 25 µM with n = 3 fibril. Densities for the amyloid core and putative RRMs are colored in blue and independent experiments). i Representative confocal microscopy image of pink, respectively. The 10x filament rise/subunit is indicated in Å. c Schematic hnRNPDL-2 amyloid fibrils bound to fluorescent ssDNA (+F-ssDNA). Control fibrils diagram showing the proposed organization of the RRM domains (pink) around the without F-ssDNA are shown as control condition. Representative image from three fibril core (cyan). d Cross-sectional view of the unsharpened cryo-EM map of independent experiments. In (g) and (h) data was fitted to one-site specific binding hnRNPDL-2 fibril, with the superimposed hnRNPDL-2 amyloid core in a ribbon mechanism with Hill slope using GraphPad Prism57. representation. The side chain of F231 is represented as sticks. e Model structure of Nature Communications | (2023)14:239 7 Article https://doi.org/10.1038/s41467-023-35854-0 residues and with water molecules, together with Tyr side chains π-π ability to form fibrils are also pathogenic40. Interestingly, these stacking and dipole-dipole interactions. The eminent polar nature of hnRNPA1 variants cause vacuolar rimmed myopathy, histologically the interactions that hold up the hnRNPDL-2 fibril and the abundance similar to LGMDD3. This suggests that RNP-associated diseases might of flexible Gly residues would allow protein chains to explore the not respond to a unique molecular mechanism, but somewhat differ- conformational space efficiently towards the final structure, without ent sequential/structural perturbations might elicit cellular dysfunc- being trapped into polymorphic local minima by stable hydrophobic tion and degeneration, potentially by affecting the function of LCD- interactions, as it often occurs for disease-associated fibrils23. In addi- containing RNPs in muscle regeneration, where they execute pre- tion, the hydrophilic nature of hnRNPDL-2 fibril surfaces may lay mRNA splicing, stabilize large muscle-specific transcripts and aid in behind their observed lack of toxicity, precluding interactions with their transport41. hydrophobic cellular membranes and their subsequent disruption. The most intriguing and unique feature of the hnRNPDL-2 fibril is hnRNPDL-2 fibrils consist of a single filament, whereas 74% of the the perfect overlap between the amyloid core and exon 6. AS patterns available amyloid structures have two or more protofilaments38. This is have diverged rapidly during evolution, with exons that were ances- not surprising, since most fibril interfaces involve hydrophobic inter- trally constitutive in vertebrates evolving to become alternatively actions between individual protofilaments38 and, as said, non-polar spliced in mammals, expanding the regulatory complexity of this amino acids are virtually absent in the hnRNPDL-2 fibril core. In addi- lineage42. These evolutionary changes impact all members of the tion, the surrounding RRMs coat would impair any filament-to-filament hnRNPD family, to which hnRNPDL belongs. Accordingly, the exon lateral association. 6 sequence is conserved among vertebrates (Fig. 3b), an additional The structure of hnRNPDL-2 fibrils is reminiscent of that of HET-s evidence of its functionality, and alternatively spliced in mammals7,19. prion, where globular domains also hang from a single filament fibril in Mammalian-specific AS is especially frequent at the Y/G-rich LCDs of a solenoidal fashion35. A single filament allows for the decoration of RNPs19, suggesting that regulation of the number of GY motifs in these amyloid cores with regularly spaced globular domains following the regions confers fitness benefit20. Furthermore, elimination of these helical twist of the fibril, something hardly compatible with a multi- repeats through exon skipping results in dominant-negative RNPs that protofilament assembly. Indeed, the fibrils of hnRNPA217 and FUS31 bind nucleic acids but cannot form multimeric complexes through Tyr- LCDs fused to fluorescent proteins also exhibit a single protofilament, dependent interactions, which significantly modifies their gene reg- consistent with this being the preferred disposition when globular ulation activity and nuclear patterning20. This differential behavior is domains are adjacent to LCDs in the sequence. Instead, the fibrils of often attributed to the longer isoforms’ ability to undergo LLPS20. hnRNPA1 LCD alone involved two filaments16. However, in proteins of the Rbfox family, the splicing activity is con- The most substantial evidence of hnRNPDL-2 fibrils’ functionality tingent on the formation of Th-T positive fibrous structures, which is is their ability to bind small oligonucleotides, especially ssDNA, with an mediated by its LCD in a Tyr-dependent manner43, although it is affinity equivalent to that of the soluble counterpart. This indicates unknown if Rbfox fibers correspond to cross-β amyloids. Here we that the RRM domains wrapping around the structured fibrils are fol- provide high-resolution structural evidence supporting the role of ded and functional. mammalian specific-AS in controlling the assembly and nuclear dis- The functionality of hnRNPDL-2 fibrils provides the basis for tribution of RNPs, which in the particular case of hnRNPDL occurs by understanding the connection between D259 mutations and LGMDD3 precisely including/skipping a conserved amyloid-prone exon. and why, in contrast to MSP patients bearing similar mutations in Overall, this work presents the detailed cryo-EM structure of a full- hnRNPA1 and hnRNPA2 LCDs, whose muscle biopsies show cyto- length RNP in its fibrillar functional conformation. The structure of plasmic mislocalization and protein aggregation15, in LGMDD3, myo- hnRNPDL-2 exon 6 in its amyloid form provides critical insights into pathologic studies coincide in the absence of sarcoplasmic protein the molecular bases of LGMD D3 and the mechanism of AS-controlled aggregates8, although congophylic deposits were detected in some RNPs assembly in mammals. instances39. For hnRNPA1 and hnRNPA2, theoretical calculations indi- cated that Asp substitutions would stabilize the fibrils16,17 and facilitate Methods LARKS to steric zippers transitions24. This would thermodynamically Cell culture, plasmids and cell lines shift any potential droplet/fibril equilibrium towards the fibrillar state, The human HeLa (ATCC CCL-2) and SH-SY5Y (ATCC CRL-2266) cell reducing reversibility. For hnRNPDL-2, this equilibrium does not apply, lines were maintained in Dulbecco’s Modified Eagle Medium (DMEM) and the WT fibrils are already irreversible. The same calculations or minimum essential medium α (MEM-α) medium, respectively. indicate that D259H/N mutations do not significantly stabilize the fibril Media were supplemented with 10% (v/v) Fetal Bovine Serum (FBS). core and that no LARKS to steric zipper transition occurs. Indeed, the Both cell lines were grown under a highly humidified atmosphere of mutant proteins exhibit a low propensity to fibrillate, and the fibrils are 95% air with 5% CO2 at 37 °C. The HeLa hnRNPDL KO cell line was shorter and less organized. This is consistent with the lack of aggre- generated as described elsewhere3. HeLa cells have been authenticated gates in most patients’ muscular tissues8, suggesting that we might within the last 3 years by STR analysis. The HeLa and SH-SY5Y cell lines face a loss-of-function disease. It could be that hnRNPDL-2 fibrils were regularly tested for the presence of mycoplasma using qPCR cannot be efficiently formed in the affected muscle, and the soluble Detection Kit (SIGMA), and both cell lines are mycoplasma negative. protein cannot compensate for their activity, or that Kapβ2 cannot For recombinant protein expression, the genes encoding the properly transport the mutant proteins7. Alternatively, inefficient three isoforms of hnRNPDL (hnRNPDL-1, hnRNPDL-2 and hnRNPDL-3) fibrillation pathways of mutants might involve intermediates that are were inserted into pETite (Lucigen corporation) vector with a His- either degraded by the protein quality control machinery, decreasing SUMO N-terminal tag. For subcellular localization experiments, the the pool of active protein, or instead accumulate in myocytes, exerting same genes were cloned into pEGFP-C3 (Clontech). In all the cases, the toxicity. Indeed, knockdown of zebrafish hnRNPDL (85% identity with correctness of the DNA sequence was verified by sequencing. human hnRNPDL-2) using antisense oligonucleotides resulted in dose- dependent disorganization of myofibers, causing body shape defects Cell transfection and immunoblotting and restricted and uncoordinated movements, which is consistent Adherent cells were transfected with linear polyethylenimine (PEI; with a loss-of-function myopathy8. Polysciences) in a 1:3 DNA:PEI ratio. Cells were collected after 48 h, Our results are in keeping with the recent evidence that even for lysed in M-PER mammalian protein extraction reagent hnRNPA1, disease manifestation is not always associated with (Thermo Fisher Scientific) with 1/1000 of the EDTA-free protease increased fibrillation, and variants with LCD mutations displaying a low inhibitor cocktail Set III (Calbiochem), and centrifuged for 30 min at Nature Communications | (2023)14:239 8 Article https://doi.org/10.1038/s41467-023-35854-0 15,000 × g at 4 °C. The soluble and insoluble fractions were analyzed Bradford using a commercial Coomassie Protein assay reagent by SDS–PAGE and immunoblotting onto PVDF membranes (EMD Mil- (Thermo Fisher Scientific). lipore) using standard protocols44. hnRNPDL proteins were detected using different primary antibodies (anti-hnRNPDL antibody, Immunolabelling and negative staining electron microscopy HPA056820 from Sigma-Aldrich, dilution 1:500; anti-vinculin mono- Aliquots of hnRNPDL-2 amyloid fibrils were mixed with an anti-6xHis clonal antibody VLN01, MA5-11690 from Invitrogen, dilution 1:5000). (MA1-21315 from Thermo Fisher Scientific) primary antibody (1:10 The primary antibodies were detected with the appropiate HRP- dilution) in 50 mM HEPES, 150 mM NaCl, pH 7.5 buffer and the sample labelled secondary antibody (goat anti-Rabbit IgG (H + L), 31460 from was then incubated for 18 h at 4 °C. After incubation, an anti-mouse Invitrogen, dilution 1:2000 or goat anti-Mouse IgG (H + L), 31430 from 10 nm colloidal gold-linked secondary antibody (A-31561 from Thermo Invitrogen, dilution 1:2000), followed by chemiluminescence detec- Fisher Scientific) was added to the samples (dilution 1:100) and incu- tion using Immobilon Forte Western HRP substrate (Sigma-Aldrich). bated for 1 h at room temperature. For negative staining transmission electron microscopy (NS-TEM), 5 µl of each sample was incubated on a Protein expression and purification EMR 400 mesh carbon-coated copper grids (Micro to Nano Innovative Protein expression was performed in E. coli BL21(DE3) cells, induced Microscopy Supplies) for 5 min. After incubation, the grids were with 0.5 mM IPTG at a OD600 = 0.5. After incubation for 3 h at 37 °C and washed with 10 µl MQ water and stained for 1 min with 5 µl of 2% (w/v) 250 rpm, the cells were harvested by centrifugation for 15 min at uranyl acetate. The excess solutions from each step were removed with 4000 × g. Cell pellets were resuspended in Binding Buffer (50 mM filter paper. Each grid was allowed to dry before inspection using a HEPES, 1 M NaCl, 5% glycerol and 20 mM imidazole, pH 7.5), lysed by JEOL JEM-1400 Electron Microscope operating at 120 kV with a CCD sonication, and centrifugated at 30,000 × g for 30 min at 4 °C. The GATAN 794 MSC 600HP camera with Digital Micrograph 1.8 (GATAN) supernatant was filtered through a 0.45 µm filter and loaded into a software. 3–5 micrographs were recorded for each sample at two dif- HisTrapTM FF Ni-column equilibrated with Binding Buffer. The bound ferent nominal magnifications (i.e. ×6000 and ×10000) and an esti- protein was eluted with an imidazole gradient starting from 0 to 100% mated defocus of about ±1–4 μm. of the Elution Buffer (50 mM HEPES, 1 M NaCl, 5% glycerol, 500 mM imidazole, pH 7.5). Afterwards, fractions containing purified proteins Cryo-EM sample preparation and data collection were pooled and loaded into a HiLoadTM 26/600 SuperdexTM 75 pg For cryo-EM, sample vitrification was carried out using a Mark IV column equilibrated with a 50 mM HEPES pH 7.5 buffer containing 1 M Vitrobot (Thermo Fisher Scientific). 3 μl hnRNPDL-2 amyloid fibrils NaCl and 5% glycerol. Finally, the proteins were concentrated using a diluted in MQ water at a final concentration of 0.25 mg/mL were 10 K Amicon (Merck-Millipore), flash-frozen in liquid nitrogen and applied to a C-Flat 1.2/1.3-3Cu-T50 grid (Protochips) previously glow- stored at −80 °C until use. discharged at 30 mA for 30 s in a GloQube (Quorum Technologies). Sample was incubated on grid for 60 s at 4 °C and 100% humidity, Fibril formation and in vitro aggregation kinetics blotted and plunge-frozen into liquid ethane. Vitrified samples were For aggregation experiments, the purified proteins corresponding to transferred to a Talos Arctica transmission electron microscope the different hnRNPDL isoforms and mutants were loaded into a PD-10 (Thermo Fisher Scientific) operated at 200 kV and equipped with a desalting column SephadexTM G-25 M to exchange the buffer. Sam- Falcon 3 direct electron detector (Thermo Fisher Scientific) and EPU ples were diluted to a final protein concentration of 50 µM in 50 mM 2.8 (Thermo Fisher Scientific) software. A total of 1114 movies were HEPES, 300 mM NaCl, pH 7.5. Then, the aggregation reactions were collected using EPU 2.8 (Thermo Fisher Scientific) in electron counting incubated at 37 °C with 600 rpm agitation in sealed Eppendorfs. mode with an applied dose of 40 e-/Å2 divided in 40 frames at a mag- Th-T binding to hnRNPDL aggregates was measured by recording nification of 120 kx. All the micrographs were acquired with a pixel size Th-T fluorescence in the range between 460 and 600 nm after exci- of 0.889 Å/pixel and a defocus range of −1.0 to −2.2 μm. tation with wavelength of 445 nm using a Jasco FP-8200 spectro- fluorimeter. The final Th-T and protein concentrations were 25 μM and Helical reconstruction 10 μM, respectively. All the samples were diluted in 50 mM HEPES Best movies (without the presence of artifacts, crystalline ice, severe buffer at pH 7.5 with 300 mM NaCl and this same buffer alone was used astigmatism or with obvious drift) were imported in RELION 3.1 for as a control. The light scattering of the reactions was measured in the further processing following a helical reconstruction pipeline22. All the range between 325 and 340 nm after excitation at 340 nm using a Jasco movies were motion-corrected and dose-weighted using FP-8200 spectrofluorimeter. MOTIONCOR245. Contrast transfer function (CTF) estimation was The aggregation kinetics of hnRNPDL isoforms were monitored performed on aligned, unweighted sum power spectra every 4 e/Å2 in 96 well plates by following increments in Th-T signal of 25 μM using CTFFIND446. Micrographs with a resolution estimate of 5 Å or protein samples. Plates were incubated at 37 °C under constant better were selected for further analysis. Fibrils were manually picked, shaking (100 rpm) using a Spark (TECAN) Fluorescence microplate and the segments were successively extracted using a box size of 380 reader. The Th-T fluorescence of each well was measured every pixel and inter-box distance of 33.2 Å, yielding a total of 158,493 seg- 30 min by exciting with a 445 nm filter and collecting the emission ments. Reference-free 2D classification was performed to identify with a 480–510 nm filter. Fluorescence emission of all the proteins homogeneous segments for further processing. in the absence of Th-T and the signal of the buffers alone with Th-T An initial model was generated using a single large class average were determined as control conditions. To test the effect of NaCl on similarly as described elsewhere47. In brief, the cross-over distance was the aggregation kinetics of hnRNPDL, aggregation reactions with measured from the class average image to be ~179 Å. The initial model different NaCl concentrations (50, 300 and 600 mM) were also was then re-scaled and re-windowed to match the unbinned particles prepared. with a 380-pixel box and low-pass filtered to 10 Å. A 3D classification with four classes, a regularization value of T = 4 and imposing a helical Ultracentrifugation rise of 4.82 Å and a helical twist of −4.86° was used to select the best To perform ultracentrifugation experiments, the samples were cen- class containing 54490 particles. trifuged at 120,000 × g for 2 h at 4 °C in a 20% (v/v) glycerol cushion. Particles were aligned with a 3D auto-refine job using a 10 Å low- After centrifugation, six fractions (70 μl each) were collected from top pass filtered map from the previous 3D classification, considering the to bottom, without disturbing the layers, and analyzed by SDS–PAGE. 30% central part of the box, and refining the helical twist and rise. The total protein concentration of each fraction was determined by Bayesian polishing48,49 and CTF refinement48 were performed to Nature Communications | (2023)14:239 9 Article https://doi.org/10.1038/s41467-023-35854-0 increase the resolution of the final reconstruction. Finally, the refined labeled RNA/ssDNA (F-GACUAGC) and increasing amounts of soluble 3D reconstructions were sharpened using the standard post- hnRNPDL-2. The samples were incubated for 24 h at RT before the processing procedures in RELION 3.1. Based on the gold-standard complexes were resolved on an 8% polyacrylamide gel at a constant Fourier shell correlation=0.143 criteria, the best map for the hnRNPDL- voltage. After electrophoresis, the gel was imaged using a ChemiDoc™ 2 fibril showed a final resolution of 2.5 Å. All the details of data acqui- MP Imaging System (Bio-Rad) and Image Lab Touch Software (Bio-Rad sition and processing are given in Supplementary Table S1. The Laboratories). Each experiment was performed in triplicate. The hnRNPDL-2 structure, and cryo-EM map have been deposited in the apparent dissociation constants (Kd) for specific binding of RNA or Protein Data Bank (PDB ID: 7ZIR) and Electron Microscopy Data Bank ssDNA to soluble hnRNPDL-2 were determined by fitting the data to (EMDB-14738), respectively. Raw images and raw processing data are one-site specific binding mechanism with Hill slope using GraphPad available at the Electron Microscopy Public Image Archive - EMBL-EBI Prism57. (EMPIAR-11064). Fluorescent RNA/DNA-binding assays Model building and refinement The ability of hnRNPDL-2 amyloid fibrils to bind ssDNA was studied by The hnRNPDL-2 amyloid fibril model was manually built into the 2.5 Å measuring the fluorescence of the fibrils bound to Fluorescein-labelled sharpened map of the hnRNPDL-2 fibril using COOT50. In brief, an initial ssDNA using a Jasco FP-8200 Spectrofluorometer (Jasco Corporation) polyalanine model was incorporated into the 2.5 Å sharpened map of and a fluorescence microscope (Leica Microsystems). the hnRNPDL-2 filament. Then, Ala residues were mutated to Tyr where For the fluorometric assays, 20 l reactions containing 10 µM of the electron density was clear enough. Based on the location of Tyr and hnRNPDL-2 amyloid fibrils and increasing amounts of Fluorescein- spacing pattern, the comprising residues 226 to 276 of hnRNPDL-2 labeled ssDNA (F-GACUAGC) at the following concentrations (0.05, were successfully assigned and modeled. Real-space refinement was 0.1, 0.5, 1, 2.5, 5, 10, 25, 50 and 100 µM) were prepared. After incubation accomplished with the phenix.real_space_refine module from Phenix51. for 24 h at RT, the samples were centrifuged at 16,000 × g for 40 min at Molecular graphics and structural analyses were performed with 4 °C. The pellet containing the hnRNPDL-2 amyloid fibrils bound to the Pymol52 and ChimeraX53. Statistics for the final model are provided in ssDNA was washed with 50 µl of 50 mM HEPES buffer (pH 7,5) with Supplementary Table 1 and the corresponding atomic coordinates 150 mM NaCl, centrifuged for 30 min and sonicated. Fluorescence have been deposited in the Protein Data Bank (PDB accession num- emission of the ssDNA bound to fibrils was measured at an excitation ber: 7ZIR). and emission wavelength of 480 nm and 520 nm, respectively. Control samples with only fluorescent ssDNA were used as controls conditions. Fibril stability and solvation free energy calculations Each experiment was performed at least in triplicate. The apparent The free energy of mutation calculations were determined using the dissociation constant (Kd) for specific binding of ssDNA to hnRNPDL-2 module ddg_monomer from Rosetta54. Delta delta G (ΔΔG) energy amyloid fibrils was determined by fitting the data to one-site specific values for WT hnRNPDL-2 and mutants (D259H, D259N and D259V) binding mechanism with Hill slope using GraphPad Prism57. fibril structures were determined following a standard high-resolution For confocal microscopy analysis, 20 µl reactions with hnRNPDL-2 protocol as described elsewhere16. In brief, we first performed energy amyloid fibrils at a concentration of 20 µM were incubated for 24 h at minimization on the hnRNPDL-2 fibril structure containing five suc- RT in the presence of 30 µM of Fluorescein-labelled ssDNA (F- cessive rungs. The resultant restraint file from this step was used in the GACUAGC). After centrifugation at 16,000 × g for 40 min, the pre- subsequent processing. Standard Van der Waals and solvation energy cipitated fraction was placed on a microscope slide and sealed. Con- parameters (i.e. cut off value of 9 Å) were applied. For comparative focal fluorescence images were obtained with Leica SP5 confocal purposes, the ΔΔG values for WT and relevant hnRNPA1 or hnRNPA2 microscope (Leica microsystems). hnRNPDL-2 amyloid fibrils and mutant fibrils were also determined using the same Rosetta protocol. labeled ssDNA alone were imaged as control samples. The percentage of buried polar residues and the exposed surfaces of hnRNPDL-2, hnRNPA2, hnRNPA1, hSAA1, Aβ−42 and α-synuclein Cytotoxicity assays fibrillar structures were calculated using PDBePISA55. Exposed surfaces The cytotoxicity of the hnRNPDL-2 amyloid fibers toward SH-SY5Y and were calculated as the difference between the total and the buried HeLa cells was evaluated using a resazurin-based assay58. Briefly, the surface. The percentage of exposed polar surface was calculated using cells were seeded in 96-well plates at a concentration of 3 × 103 cells per the total and apolar surface values predicted by GetArea56. The solva- well and incubated for 24 h. Afterwards, the cells were treated with tion free energy of folding (ΔG in kcal/mol) values for all the structures different hnRNPDL-2 fiber concentrations ranging from 0.004 to were calculated using PDBePISA55. 0.4 mg/mL (w/v). After 48 h of incubation, aliquots of 10 μL of the PrestoBlue™ (Thermo Fisher Scientific) cell viability reagent solution In vivo cell imaging were added to each well. After 1 h incubation at 37 °C in the absence of For in vivo cell imaging experiments, adherent HeLa cells were grown light and in the presence of a highly humidified atmosphere of 95% air on 35 mm glass-bottom culture dishes at a density of 1 × 105 cells per with 5% CO2, fluorescence emission (ex/em=531/572 nm) of each well dish. After 24 h incubation, the cells were transiently transfected with was measured using a fluorescence microplate reader (PerkinElmer GFP-tagged versions of the hnRNPDL-1/-2/-3 isoforms and linear poly- Victor 3 V) and PerkinElmer 2030 software. Cell cytotoxicity was ethylenimine (PEI; Polysciences, Eppelheim, Germany) in a 1:3 DNA:PEI determined in terms of cell growth inhibition in treated samples and ratio. After 24 h incubation, the cells were washed with fresh medium expressed as percentage of the control condition. (2 × 1 mL), and then cell nuclei were stained with Hoechst (Invitrogen) for 10 min at 37 °C and 5% CO2. In vivo cell imaging was performed at Reporting summary 37 °C using a Leica TCS SP5 confocal microscope and a 63 × 1.4 Further information on research design is available in the Nature numerical aperture Plan Apochromat oil-immersion lens. All the con- Portfolio Reporting Summary linked to this article. focal images were processed using Bitplane Imaris 7.2.1 software. Data availability Gel Electrophoretic Mobility Shift Binding Assays (EMSA) The hnRNPDL-2 amyloid fibril structure, and cryo-EM map have been The ability of hnRNPDL-2 amyloid fibrils to bind RNA/ssDNA was stu- deposited in the Protein Data Bank and Electron Microscopy Data Bank died by performing a Gel Electrophoretic Mobility Shift Binding Assays under the accession codes 7ZIR and EMDB-14738, respectively. Raw (EMSA). We prepared 10 µl reactions containing 50 nM of Fluorescein- images and raw processing data have been deposited in Electron Nature Communications | (2023)14:239 10 Article https://doi.org/10.1038/s41467-023-35854-0 Microscopy Public Image Archive with accession code EMPIAR-11064. 20. Gueroussov, S. et al. Regulatory Expansion in Mammals of Multi- Other structures referenced in this work are available under the PDB valent hnRNP Assemblies that Globally Control Alternative Spli- accession codes 6MST, 5KK3, 6OSJ, 7Q4M and 7NCK. Other data cing. Cell 170, 324–339 e23 (2017). supporting the findings of this study are available within the article and 21. Scheres, S. H. RELION: implementation of a Bayesian approach its associated supplementary information files. Source data are pro- to cryo-EM structure determination. J. Struct. Biol. 180, vided with this paper. 519–530 (2012). 22. He, S. & Scheres, S. H. W. Helical reconstruction in RELION. J. References Struct. Biol. 198, 163–176 (2017). 1. Dreyfuss, G., Matunis, M. J., Pinol-Roma, S. & Burd, C. G. hnRNP 23. Sawaya, M. R., Hughes, M. P., Rodriguez, J. A., Riek, R. & Eisenberg, proteins and the biogenesis of mRNA. Annu Rev. Biochem. 62, D. S. The expanding amyloid family: Structure, stability, function, 289–321 (1993). and pathogenesis. Cell 184, 4857–4873 (2021). 2. Li, R. Z. et al. hnRNPDL extensively regulates transcription and 24. Murray, K. A. et al. Identifying amyloid-related diseases by mapping alternative splicing. Gene 687, 125–134 (2019). mutations in low-complexity protein domains to pathologies. Nat. 3. Batlle, C. et al. hnRNPDL Phase Separation Is Regulated by Alter- Struct. Mol. Biol. 29, 529–536 (2022). native Splicing and Disease-Causing Mutations Accelerate Its 25. Tsuchiya, N., Kamei, D., Takano, A., Matsui, T. & Yamada, M. Cloning Aggregation. Cell Rep. 30, 1117–1128 e5 (2020). and characterization of a cDNA encoding a novel heterogeneous 4. Kemmerer, K., Fischer, S. & Weigand, J. E. Auto- and cross- nuclear ribonucleoprotein-like protein and its expression in myeloid regulation of the hnRNPs D and DL. RNA 24, 324–331 (2018). leukemia cells. J. Biochem. 123, 499–507 (1998). 5. Kamei, D., Tsuchiya, N., Yamazaki, M., Meguro, H. & Yamada, M. Two 26. Kamei, D. & Yamada, M. Interactions of heterogeneous nuclear forms of expression and genomic structure of the human hetero- ribonucleoprotein D-like protein JKTBP and its domains with high- geneous nuclear ribonucleoprotein D-like JKTBP gene (HNRPDL). affinity binding sites. Gene 298, 49–57 (2002). Gene 228, 13–22 (1999). 27. Molliex, A. et al. Phase separation by low complexity domains 6. Akagi, T. et al. Molecular characterization of a mouse hetero- promotes stress granule assembly and drives pathological fibrilli- geneous nuclear ribonucleoprotein D-like protein JKTBP and its zation. Cell 163, 123–133 (2015). tissue-specific expression. Gene 245, 267–273 (2000). 28. Patel, A. et al. A Liquid-to-Solid Phase Transition of the ALS Protein 7. Kawamura, H. et al. Identification of the nucleocytoplasmic shut- FUS Accelerated by Disease Mutation. Cell 162, 1066–1077 (2015). tling sequence of heterogeneous nuclear ribonucleoprotein D-like 29. Murray, D. T. et al. Structure of FUS Protein Fibrils and Its Relevance protein JKTBP and its interaction with mRNA. J. Biol. Chem. 277, to Self-Assembly and Phase Separation of Low-Complexity 2732–2739 (2002). Domains. Cell 171, 615–627 e16 (2017). 8. Vieira, N. M. et al. A defect in the RNA-processing protein HNRPDL 30. Li, Q., Babinchak, W. M. & Surewicz, W. K. Cryo-EM structure of causes limb-girdle muscular dystrophy 1G (LGMD1G). Hum. Mol. amyloid fibrils formed by the entire low complexity domain of TDP- Genet. 23, 4103–4110 (2014). 43. Nat. Commun. 12, 1620 (2021). 9. Starling, A., Kok, F., Passos-Bueno, M. R., Vainzof, M. & Zatz, M. A 31. Sun, Y. et al. Molecular structure of an amyloid fibril formed by FUS new form of autosomal dominant limb-girdle muscular dystrophy low-complexity domain. iScience 25, 103701 (2021). (LGMD1G) with progressive fingers and toes flexion limitation maps 32. Geuens, T., Bouhy, D. & Timmerman, V. The hnRNP family: to chromosome 4p21. Eur. J. Hum. Genet. 12, 1033–1040 (2004). insights into their role in health and disease. Hum. Genet. 135, 10. Sun, Y. et al. Limb girdle muscular dystrophy D3 HNRNPDL related 851–867 (2016). in a Chinese family with distal muscle weakness caused by a 33. Krecic, A. M. & Swanson, M. S. hnRNP complexes: composition, mutation in the prion-like domain. J. Neurol. 266, 498–506 (2019). structure, and function. Curr. Opin. Cell Biol. 11, 363–371 (1999). 11. Liewluck, T. & Milone, M. Untangling the complexity of limb-girdle 34. Guo, L. et al. Nuclear-Import Receptors Reverse Aberrant Phase muscular dystrophies. Muscle Nerve. 58, 167–177 (2018). Transitions of RNA-Binding Proteins with Prion-like Domains. Cell 12. Batlle, C. & Ventura, S. Prion-like domain disease-causing mutations 173, 677–692 e20 (2018). and misregulation of alternative splicing relevance in limb-girdle 35. Wasmer, C. et al. Amyloid fibrils of the HET-s(218-289) prion form a muscular dystrophy (LGMD) 1G. Neural Regen. Res. 15, beta solenoid with a triangular hydrophobic core. Science 319, 2239–2240 (2020). 1523–1526 (2008). 13. Harrison, A. F. & Shorter, J. RNA-binding proteins with prion-like 36. Seuring, C. et al. The three-dimensional structure of human domains in health and disease. Biochem J. 474, 1417–1438 beta-endorphin amyloid fibrils. Nat. Struct. Mol. Biol. 27, (2017). 1178–1184 (2020). 14. Kim, H. J. et al. Mutations in prion-like domains in hnRNPA2B1 and 37. Wu, X. et al. The structure of a minimum amyloid fibril core formed hnRNPA1 cause multisystem proteinopathy and ALS. Nature 495, by necroptosis-mediating RHIM of human RIPK3. Proc. Natl. Acad. 467–473 (2013). Sci. 118, e2022933118 (2021). 15. Purice, M. D. & Taylor, J. P. Linking hnRNP Function to ALS and FTD 38. Louros, N., van der Kant, R., Schymkowitz, J. & Rousseau, F. StAmP- Pathology. Front Neurosci. 12, 326 (2018). DB: A platform for structures of polymorphic amyloid fibril cores. 16. Sun, Y. et al. The nuclear localization sequence mediates hnRNPA1 Bioinformatics 38, 2636–2638 (2022). amyloid fibril formation revealed by cryoEM structure. Nat. Com- 39. Vicente, L. M. et al. HNRNPDL-related limb girdle muscular dys- mun. 11, 6349 (2020). trophy in a Spanish family with scapulo-peroneal phenotype, the 17. Lu, J. et al. CryoEM structure of the low-complexity domain of first family in Europe. J. Neurol. Sci. 414, 116875 (2020). hnRNPA2 and its conversion to pathogenic amyloid. Nat. Commun. 40. Beijer, D. et al. Characterization of HNRNPA1 mutations defines 11, 4090 (2020). diversity in pathogenic mechanisms and clinical presentation. JCI 18. Xu, D., Farmer, A. & Chook, Y. M. Recognition of nuclear targeting Insight. 6, e148363 (2021). signals by Karyopherin-beta proteins. Curr. Opin. Struct. Biol. 20, 41. Vogler, T. O. et al. TDP-43 and RNA form amyloid-like myo-granules 782–790 (2010). in regenerating muscle. Nature 563, 508–513 (2018). 19. Buljan, M. et al. Tissue-specific splicing of disordered segments that 42. Barbosa-Morais, N. L., Carmo-Fonseca, M. & Aparicio, S. Systematic embed binding motifs rewires protein interaction networks. Mol. genome-wide annotation of spliceosomal proteins reveals differ- Cell 46, 871–883 (2012). ential gene family expansion. Genome Res. 16, 66–77 (2006). Nature Communications | (2023)14:239 11 Article https://doi.org/10.1038/s41467-023-35854-0 43. Ying, Y. et al. Splicing Activation by Rbfox Requires Self- Spanish Ministry of Science and Innovation via doctoral grants, FPU16/ Aggregation through Its Tyrosine-Rich Domain. Cell 170, 312–323 02465. S.R. was funded by Fondazione ARISLA (project TDP- 43- e10 (2017). STRUCT) and by Ricerca Corrente funding from Italian Ministry of Health 44. Batlle, C. et al. MED15 prion-like domain forms a coiled-coil to IRCCS Policlinico San Donato. The authors thank the Microscopy responsible for its amyloid conversion and propagation. Commun. Service, Universitat Autònoma de Barcelona, for their exceptional Biol. 4, 414 (2021). technical support in electron microscopy and confocal imaging, and the 45. Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam- University of Milan Unitech NOLIMITS Center for granting access to the induced motion for improved cryo-electron microscopy. Nat. cryo-EM facility. We thank Prof. J. Paul Taylor for providing the hnRNPDL Methods 14, 331–332 (2017). KO HeLa cell line and Dr. Tim Schulte for technical assistance. 46. Rohou, A. & Grigorieff, N. CTFFIND4: Fast and accurate defocus estimation from electron micrographs. J. Struct. Biol. 192, Author contributions 216–221 (2015). J.G-P., M.B., S.R. and S.V. designed the conceptual framework of the 47. Swuec, P. et al. Cryo-EM structure of cardiac amyloid fibrils from an study and experiments. J.G-P., A.B-N., M.G-G. and C.V. performed the immunoglobulin light chain AL amyloidosis patient. Nat. Commun. experiments and contributed to data acquisition, data analysis, and 10, 1269 (2019). preparation of manuscript figures. A.C-S., J.G.-P., and M.G-G. con- 48. Scheres, S. H. W. Amyloid structure determination in RELION-3.1. tributed to cryo-EM data acquisition and cryo-EM structural character- Acta Crystallogr D. Struct. Biol. 76, 94–101 (2020). ization of the amyloid fibrils. J.G-P. and S.V. wrote the manuscript with 49. Zivanov, J., Nakane, T. & Scheres, S. H. W. A Bayesian approach to contributions and comments from all authors. beam-induced motion correction in cryo-EM single-particle analy- sis. IUCrJ 6, 5–17 (2019). Competing interests 50. Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and The authors declare no competing interests. development of Coot. Acta Crystallogr D. Biol. Crystallogr. 66, 486–501 (2010). Additional information 51. Liebschner, D. et al. Macromolecular structure determination using Supplementary information The online version contains X-rays, neutrons and electrons: recent developments in Phenix. supplementary material available at Acta Crystallogr D. Struct. Biol. 75, 861–877 (2019). https://doi.org/10.1038/s41467-023-35854-0. 52. Schrödinger, L. The PyMOL Molecular Graphics System. 2.0 edn. 53. Pettersen, E. F. et al. UCSF ChimeraX: Structure visualization for Correspondence and requests for materials should be addressed to researchers, educators, and developers. Protein Sci. 30, Salvador Ventura. 70–82 (2021). 54. Kellogg, E. H., Leaver-Fay, A. & Baker, D. Role of conformational Peer review information Nature Communications thanks Meytal Landau, sampling in computing mutation-induced changes in protein Juan Vilchez, and the other, anonymous, reviewer(s) for their contribu- structure and stability. Proteins 79, 830–838 (2011). tion to the peer review of this work. Peer reviewer reports are available. 55. Krissinel, E. & Henrick, K. Inference of macromolecular assemblies from crystalline state. J. Mol. Biol. 372, 774–797 (2007). Reprints and permissions information is available at 56. Fraczkiewicz, R. & Braun, W. Exact and efficient analytical calcula- http://www.nature.com/reprints tion of the accessible surface areas and their gradients for macro- molecules. J. Comp. Chem. 19, 319–333 (1998). Publisher’s note Springer Nature remains neutral with regard to jur- 57. GraphPad Prism. 6.01 edn (GraphPad software, San Diego, Cali- isdictional claims in published maps and institutional affiliations. fornia USA, 2012). 58. Garcia-Pardo, J. et al. Bioinspired Theranostic Coordination Polymer Open Access This article is licensed under a Creative Commons Nanoparticles for Intranasal Dopamine Replacement in Parkinson’s Attribution 4.0 International License, which permits use, sharing, Disease. ACS Nano. 15, 8592–8609 (2021). adaptation, distribution and reproduction in any medium or format, as 59. Jurrus, E. et al. Improvements to the APBS biomolecular solvation long as you give appropriate credit to the original author(s) and the software suite. Protein Sci. 27, 112–128 (2018). source, provide a link to the Creative Commons license, and indicate if 60. Jumper, J. et al. Highly accurate protein structure prediction with changes were made. The images or other third party material in this AlphaFold. Nature 596, 583–58 (2021). article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not Acknowledgements included in the article’s Creative Commons license and your intended This work was funded by European Union’s Horizon 2020 research and use is not permitted by statutory regulation or exceeds the permitted innovation programme under GA 952334 (PhasAGE), by the Spanish use, you will need to obtain permission directly from the copyright Ministry of Science and Innovation (PID2019–105017RB-I00) to S.V. and holder. To view a copy of this license, visit http://creativecommons.org/ by ICREA, ICREA-Academia 2015 and 2020 to S.V. J.G-P. was supported licenses/by/4.0/. by the Spanish Ministry of Science and Innovation with a Juan de la Cierva Incorporación IJC2019-041039-I. M.G-G. was supported by the © The Author(s) 2023 Nature Communications | (2023)14:239 12