Short RNA chaperones promote aggregation-resistant TDP-43 conformers to mitigate neurodegeneration

Structured Abstract

Introduction

Aberrant aggregation of TDP-43, an RNA-binding protein with a prion-like domain, is a pathological hallmark of amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). TDP-43 mislocalization from the nucleus to the cytoplasm and subsequent aggregation drive neuronal dysfunction and death across these fatal neurodegenerative diseases. TDP-43 pathology also occurs frequently in Alzheimer’s disease (AD), where it exacerbates cognitive decline, yet effective therapeutic strategies remain limited across all TDP-43 proteinopathies. Deliverable agents that prevent and reverse aberrant TDP-43 aggregation and restore functional TDP-43 to the nucleus in degenerating neurons could provide a therapeutic solution. Short RNA chaperones can solubilize TDP-43 and are highly deliverable to the central nervous system. However, critical mechanistic, translational, and functional barriers must be overcome to advance short RNA chaperones as therapeutics for TDP-43 proteinopathies.

Rationale

Our rationale was to address several unanswered questions about short RNA chaperones as TDP-43 therapeutics. We asked how short RNAs engage TDP-43 to antagonize aberrant assembly and how short RNAs remodel TDP-43 structure to prevent aggregation. We further asked whether short RNAs prevent aggregation of diverse disease-linked TDP-43 variants and whether optimized RNA sequences with enhanced chaperone activity can be identified. We investigated whether short RNAs mitigate aberrant TDP-43 phenotypes in optogenetic human cell models, iPSC-derived motor neurons under oxidative stress, and ALS patient-derived motor neurons. Finally, we asked whether short RNA chaperones could ameliorate established aberrant TDP-43 phenotypes in a mouse model of TDP-43 proteinopathy where cytoplasmic TDP-43 aggregation, loss of TDP-43 function, and progressive motor neuron degeneration are already underway.

Results

We elucidate how short, specific RNAs solubilize TDP-43. These short RNAs engage and stabilize the TDP-43 RNA-recognition motifs, which allosterically destabilizes a conserved helical region in the prion-like domain, thereby promoting aggregation-resistant conformers. Sequence-space mining yielded short RNA chaperones with enhanced activity against wild-type TDP-43 and a broad spectrum of disease-linked variants. Critically, an enhanced short RNA chaperone directly dissolves preformed TDP-43 condensates and aggregates. Furthermore, enhanced short RNA chaperones suppress cytoplasmic TDP-43 aggregation in an optogenetic human cellular model of disease. Importantly, lead short RNA chaperones do not interfere with TDP-43 function. Rather, short RNA chaperones antagonize loss of TDP-43 function in iPSC-derived motor neurons experiencing oxidative stress. Additionally, lead short RNAs restore physiological nuclear TDP-43 localization in ALS patient-derived motor neurons. Finally, in mice experiencing cytoplasmic TDP-43 aggregation, loss-of-function, and motor neuron degeneration, an enhanced short RNA chaperone reverses pathological TDP-43 aggregation, restores TDP-43 function, and confers neuroprotection.

Conclusion

We define a mechanistic framework for how short RNA chaperones counter TDP-43 aggregation through allosteric modulation of the prion-like domain. Our results establish enhanced short RNA chaperones as therapeutic candidates for TDP-43 proteinopathies, including ALS, FTD, and AD. This work provides a rational foundation for RNA-based therapeutic strategies targeting TDP-43 and demonstrates the feasibility of engineered short RNAs for mitigating fatal neurodegenerative diseases driven by pathological protein aggregation.

Graphical Abstract

Short RNA chaperones antagonize TDP-43 aggregation through allosteric mechanisms and confer neuroprotection. (Left) RNA binding stabilizes TDP-43 RNA-recognition motifs, allosterically destabilizing the prion-like domain α-helix and preventing aggregation. (Right) Enhanced short RNA chaperones solubilize disease-relevant TDP-43 variants, restore nuclear TDP-43 localization in ALS patient iPSC-derived motor neurons, and reverse pathological TDP-43 aggregation, restore TDP-43 function, and confer neuroprotection in mice. Generated using BioRender.

Aberrant aggregation of the prion-like RNA-binding protein TDP-43 drives several fatal neurodegenerative proteinopathies, including amyotrophic lateral sclerosis (ALS). Here, we define how short, specific RNAs solubilize TDP-43. These short RNAs engage and stabilize the TDP-43 RNA-recognition motifs, which allosterically destabilizes a conserved helical region in the prion-like domain, thereby promoting aggregation-resistant conformers. Sequence-space mining identified short RNA chaperones with enhanced activity against TDP-43 and disease-linked variants. Enhanced short RNA chaperones mitigated aberrant TDP-43 phenotypes in optogenetic models and in ALS patient–derived and control motor neurons. In mice with cytoplasmic TDP-43 aggregation and motor neuron loss, an enhanced short RNA chaperone reduced pathological aggregation, restored TDP-43 function, and conferred neuroprotection. These results define a mechanistic and therapeutic framework for RNA-based strategies to counter TDP-43 proteinopathies.

There are no effective therapeutics for fatal TDP-43 proteinopathies, including amyotrophic lateral sclerosis (ALS), frontotemporal dementia (FTD), limbic-predominant age-related TDP-43 encephalopathy (LATE), Alzheimer’s disease (AD), and chronic traumatic encephalopathy (CTE) (1–5). A unifying feature of degenerating neurons in the vast majority of ALS cases, substantial fractions of FTD and AD cases, all LATE cases, and advanced CTE cases is the aberrant cytoplasmic mislocalization and aggregation of TDP-43 (1–8). TDP-43 is an essential and predominantly nuclear RNA-binding protein (RBP) with a prion-like domain (PrLD) (1, 2), which plays critical roles in RNA processing, splicing, and polyadenylation (9, 10). An aberrant phase transition of TDP-43 in the cytoplasm is a key pathological event that is difficult for neurons to reverse (2, 6, 7, 11–16). Deliverable agents that prevent and reverse the aberrant phase transitions of TDP-43 and restore functional TDP-43 to the nucleus in degenerating neurons could provide a therapeutic solution (2, 11). Indeed, such agents would eliminate any toxic gain-of-function of aberrant TDP-43 conformers in the cytoplasm and any toxic loss-of-function caused by depletion of TDP-43 from the nucleus (2, 11).

TDP-43 contains two RNA recognition motifs (RRMs), which preferentially engage UG-rich RNA (Fig. 1A) (17). TDP-43 also harbors an intrinsically disordered PrLD, which includes a short, conserved region (CR) with transient α-helical structure (Fig. 1A) (1, 2, 18, 19). The CR plays a pivotal role in TDP-43 phase separation and aggregation (6, 7, 18–20). Typically, wild-type (WT) TDP-43 aggregates in disease, but rare forms of disease are connected with TDP-43 missense variants that are frequently found in the PrLD (1). Aberrant post-translational modifications (PTMs) of TDP-43 occur in disease, including hyperphosphorylation and lysine acetylation (8, 21, 22). A broad-acting therapeutic should be effective against diverse, disease-relevant forms of TDP-43.

Fig. 1. Clip34 is an allosteric antagonist of TDP-43 aggregation.

(A) Domain map of TDP-43 indicating five Phe-to-Leu mutations within the RRMs, and the A326P PrLD mutation. (B) RNA sequence of Clip34, a 34nt RNA derived from the 3’ UTR of TARDBP RNA, which TDP-43 binds to regulate its expression. (C) Area under the curve (AUC) of standardized aggregation turbidity data for each deletion construct, normalized to WT TDP-43. Data are mean ± SEM (n=3; one-way ANOVA with Dunnett’s correction comparing to WT; *p < 0.05). (D-I) Turbidity AUC for each variant normalized to its respective No RNA control; domain maps are shown above each graph. For (D, F-I), No RNA conditions are based on the same data shown in (C). Data are mean ± SEM (n=3; n=13 for (E); one-way ANOVA with Dunnett’s correction comparing to No RNA; **p < 0.01, ****p < 0.0001). (J) Bound 5’ 6-FAM Clip34 signal for EMSAs with indicated TDP-43-MBP-His variants or MBP-His. Data are mean ± SEM (n=3 for TDP-43 variants; n=2 for MBP-His; shown is the nonlinear regression: [agonist] vs. response with variable slope, of combined replicates). (K) TDP-43^5FL turbidity AUC data normalized to the No RNA control. Data are mean ± SEM (n=3; one-way ANOVA with Dunnett’s correction comparing to No RNA; ****p < 0.0001). (L) Apparent K_D values calculated from the bound signal of individual replicates of EMSAs performed with 5’ 6-FAM Clip34 and WT TDP-43-MBP-His or the indicated variant. WT data is the same data as in (J). Data are mean ± SEM (n=3; one-way ANOVA with Dunnett’s correction comparing to WT; *p< 0.05; **p< 0.01; ***p < 0.001; ****p < 0.0001). (M) Clip34 secondary structure predicted by RNAstructure. Text color indicates the probability for each nucleotide. (N) EC₅₀ values calculated from individual replicates of relative fluorescence intensity for 5’ 6-FAM Clip34 3’ BHQ1 with indicated TDP-43-MBP-His variants. Data are mean ± SEM (n=3; 100 nM RNA; one-way ANOVA with Dunnett’s correction comparing to WT; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

RNA solubilizes TDP-43 in vitro (13, 14, 23). Indeed, one short 34-nucleotide (nt) RNA derived from the 3’UTR of the TARDBP mRNA, termed Clip34 (Table S1 and Fig. 1B), which TDP-43 binds to regulate its own expression, can prevent and reverse WT TDP-43 phase separation and aggregation (13, 14). Given the therapeutic success of short oligonucleotides (for example, antisense oligonucleotides) for neurodegenerative disease (24), similarly sized short RNA chaperones like Clip34 warrant investigation as TDP-43 proteinopathy therapeutics.

Several critical barriers, however, limit development of short RNA chaperones like Clip34. Key unknowns include: (a) how short RNAs must engage TDP-43 to antagonize aberrant assembly; (b) how short RNAs alter TDP-43 structure to prevent aggregation; (c) whether short RNAs can prevent aggregation of diverse disease-linked TDP-43 variants, including missense mutants and TDP-43 bearing disease-linked PTMs; (d) whether more potent short RNA chaperones against TDP-43 exist beyond Clip34; and (e) whether short RNAs mitigate aberrant TDP-43 phenotypes in human motor neurons, including ALS patient-derived motor neurons, and mouse models of disease. Here, we addressed these pressing issues, and thereby defined a mechanistic and therapeutic framework for RNA-based strategies to treat TDP-43 proteinopathies.

Results

Clip34 is an allosteric antagonist of TDP-43 aggregation

We first mapped the domains of TDP-43 required for Clip34 chaperone activity by purifying full-length TDP-43 and constructs lacking the N-terminal domain (NTD), RRM1, RRM2, RRM1/2, PrLD, or NTD/PrLD, each with a C-terminal MBP tag (Fig. 1A and fig. S1, A and B). Specific MBP tag removal with TEV protease triggered rapid TDP-43 aggregation, whereas the MBP-tagged TDP-43 remained soluble (fig. S1C and Fig. 1, C to I) (13, 25). Upon MBP tag removal, full-length TDP-43, TDP-43^ΔNTD, TDP-43^ΔRRM1, and TDP-43^ΔRRM2 aggregated robustly, whereas TDP-43 lacking the PrLD did not (Fig. 1C), as expected (20). Unexpectedly, TDP-43^ΔRRM1/2 exhibited reduced aggregation (Fig. 1C), indicating that the RRMs contribute to the aggregation propensity of TDP-43.

We assessed the ability of Clip34 to antagonize aggregation of these TDP-43 constructs. At a 1:4 RNA:TDP-43 ratio, Clip34 abolished TDP-43 aggregation (Fig. 1D), demonstrating substoichiometric efficacy. This effect was specific, as Clip34 did not reduce phase separation of FUS (fig. S1, D and E), another ALS/FTD-linked RBP with a PrLD (2). Moreover, the UG-deficient RNA (AC)₁₇, which does not bind TDP-43, failed to inhibit aggregation (Fig. 1E). Thus, specific RNA sequences are required for chaperone activity.

Clip34 inhibited aggregation of TDP-43^ΔNTD (Fig. 1F), indicating that Clip34 binding to the NTD is not required for inhibition. Clip34 also effectively inhibited TDP-43^ΔRRM1 aggregation (Fig. 1G), but exhibited reduced activity against TDP-43^ΔRRM2 (Fig. 1H), demonstrating that RRM2 plays an important role. Clip34 failed to prevent aggregation of TDP-43^ΔRRM1/2 (Fig. 1I) or the isolated PrLD. Thus, Clip34 does not inhibit TDP-43 aggregation via direct interactions with the PrLD, which drives aggregation (6, 7, 20). Rather, Clip34 must engage the RRMs to antagonize TDP-43 aggregation. These results suggest that Clip34 binding to the TDP-43 RRMs elicits an allosteric effect on other domains of TDP-43, which precludes TDP-43 aggregation.

Revealing allosteric crosstalk between the TDP-43 RRMs, PrLD, and RNA

To explore how Clip34 promotes aggregation-resistant TDP-43 conformers, we explored how TDP-43 binds to Clip34. TDP-43 bound Clip34 cooperatively, with a hill slope (h) of ~2.4 and a K_D of ~0.49 μM (Fig. 1J and fig. S2, A to C). By contrast, Clip34 did not bind strongly to TDP-43^5FL, which bears F147L, F149L, F194L, F229L, and F231L mutations in the RRMs that impair RNA binding (Fig. 1J and fig. S2B) (26, 27). Indeed, Clip34 failed to inhibit TDP-43^5FL aggregation (Fig. 1K). Thus, Clip34 engages the RRMs to abrogate TDP-43 aggregation and must exert allosteric effects that prevent intermolecular contacts between PrLDs.

We next assessed the contribution of each RRM. TDP-43^ΔRRM1 binding to Clip34 was reduced by ~1.6-fold in terms of B_max (the maximum specific binding) compared to TDP-43, whereas TDP-43^ΔRRM2 binding was reduced by ~1.1-fold (Fig. 1J and fig. S2B). TDP-43^ΔRRM1 bound Clip34 with reduced cooperativity (h~1.2) and a K_D of ~6.4 μM, whereas TDP-43^ΔRRM2 bound Clip34 cooperatively (h~2.5) with a K_D of ~0.9 μM (Fig. 1J and fig. S2C). Thus, RRM1 contributes more to tight, cooperative Clip34 binding than RRM2, yet Clip34 still effectively inhibited TDP-43^ΔRRM1 aggregation and was less effective against TDP-43^ΔRRM2 (Fig. 1, G and H). Binding affinity therefore does not fully predict chaperone activity, indicating that Clip34 must engage TDP-43 in a specific manner to prevent aggregation, consistent with an allosteric mechanism.

We next explored the role of the NTD in binding to Clip34. The NTD negatively regulated Clip34 binding, as TDP-43^ΔNTD bound Clip34 with a K_D of ~0.37 μM, representing an ~1.3-fold increase in affinity compared to TDP-43 (Fig. 1L and fig. S2D). This finding is consistent with an allosteric connection between the NTD and RRMs (28). However, Clip34 still effectively inhibited TDP-43^ΔNTD aggregation (Fig. 1F), indicating that the NTD is not required for Clip34 chaperone activity.

We next considered whether Clip34 binding to the RRMs allosterically affects the PrLD. The PrLD drives TDP-43 aggregation (6, 7, 20), but in cells also promotes binding and regulation of a subset of RNA targets, which contain >100-nt binding regions composed of dispersed motifs, including the 3’UTR of TARDBP mRNA (29). Most studies of TDP-43 binding to RNA have employed isolated RRMs rather than full-length TDP-43 (17). Thus, the impact of the PrLD on RNA binding has remained unclear. We found that the PrLD negatively regulates Clip34 binding to the RRMs, as PrLD deletion enhanced Clip34 binding (Fig. 1L and fig. S2D). TDP-43^ΔPrLD bound Clip34 more cooperatively (h~2.7) with a K_D of ~0.32 μM, representing an ~1.5-fold affinity increase relative to TDP-43 (Fig. 1L and fig. S2D). Thus, in addition to forming intermolecular contacts that drive aggregation (6, 7, 20), the PrLD indirectly promotes TDP-43 insolubility by reducing the apparent RRM affinity for RNA. Indeed, RNA-binding deficient TDP-43 is highly aggregation-prone in cells (14).

To determine whether this inhibitory effect stems from a specific PrLD region, we tested TDP-43 variants with specific deletions within the PrLD (29). Deletion of the extreme C-terminal portion of the PrLD (TDP-43^ΔIDR2(G/S)) slightly enhanced binding to Clip34, with an ~1.2-fold affinity increase relative to TDP-43 (Fig. 1L and fig. S2D). Deletion of the α-helical CR of the PrLD strongly enhanced binding to Clip34, indicated by an ~2.1-fold affinity increase for TDP-43^ΔCR, and a helix-breaking mutation within the CR, TDP-43^A326P, similarly enhanced binding (Fig. 1L and fig. S2D and E). Thus, negative regulation of RNA binding by the PrLD is mediated, at least in part, by the a-helicity of the CR, which is critical for TDP-43 phase separation via helix-helix interactions, and aggregation via intermolecular β-sheet interactions (6, 7, 16, 29, 30).

To explore whether the inhibitory effect of the PrLD on RNA binding impacts Clip34 chaperone activity, we assessed TDP-43 variants bearing specific PrLD deletions (29). Clip34 exhibited enhanced ability to antagonize aggregation of all partial PrLD deletion variants tested (fig. S3; see Supplementary Text). Thus, the PrLD antagonizes the ability of Clip34 to reduce TDP-43 aggregation.

TDP-43 remodels Clip34 by unfolding stem-loop structure

TDP-43 RRMs frequently engage single-stranded, UG-rich RNA, which is often found in introns (17, 31, 32). However, Clip34 is predicted to form a stem-loop structure (Fig. 1M) (33). To assess this prediction, we utilized Clip34 bearing a 5’ fluorophore and a 3’ quencher, exploiting the predicted proximity of the 5’ and 3’ ends (Fig. 1M) (33). Low basal fluorescence indicated close apposition of the 5’ and 3’ ends (Fig. 1N and fig. S2F, red arrow). TDP-43 bound Clip34 cooperatively (Fig. 1J), which may enable unfolding of the stem-loop structure. Upon addition of TDP-43, fluorescence increased strongly (Fig. 1N and fig. S2F), indicating that TDP-43 remodels Clip34 in a manner that increases the distance between the 5’ and 3’ ends. RRM1 deletion slightly impaired remodeling, whereas deletion of both RRMs strongly impaired remodeling (Fig. 1N and fig. S2F). Conversely, deletion of the PrLD or the CR helix-breaking TDP-43^A326P mutation enhanced remodeling (Fig. 1N and fig. S2F). Thus, TDP-43 remodels the Clip34 stem-loop in an RRM-dependent manner that is negatively regulated by the PrLD. This finding raised the possibility that the energetics of TDP-43:Clip34 binding might alter TDP-43 structural dynamics.

Clip34 remodels TDP-43 by stabilizing the RRMs and destabilizing the PrLD CR

We examined how Clip34 affected TDP-43 native structure via hydrogen/deuterium-exchange mass spectrometry (HXMS). HXMS measures the exchange of backbone amide hydrogens over time after dilution in D₂O-based buffer (34). When backbone hydrogens make hydrogen bonds, they exchange more slowly with deuterium (34). As backbone hydrogens make hydrogen bonds involved in protein secondary and tertiary structure, the kinetics by which the hydrogens exchange reports on the stability of structure in that region of the protein (34). Thus, we can establish how Clip34 might alter TDP-43 structural dynamics to preclude aggregation.

We performed HXMS across a wide timescale (1 s-14.5 h) with TDP-43 (with a C-terminal MBP tag to ensure solubility) in the absence or presence of excess Clip34 to saturate binding (Table S2 and fig. S4, A to C). We achieved ≥87.9% sequence coverage of TDP-43 under all conditions (Table S2 and fig. S4, A and B). For each peptide at each timepoint, we calculated the percentage difference in deuterium exchange between Clip34-bound and free states, then derived consensus values for individual TDP-43 residues (Fig. 2, A and B, and fig. S5).

Fig. 2. Clip34 remodels TDP-43 by stabilizing the RRMs and destabilizing the PrLD CR.

(A) Aligned with the TDP-43 domain map, for each timepoint, residue color corresponds to the consensus percentage difference in exchange between Clip34-bound (2:1::[Clip34]:[TDP-43]) and free states as in the legend. White spaces are coverage gaps. (B) Consensus percentage difference in exchange at 4.5 h shown again as in (A). Aligned beneath it are peptides analyzed at 4.5 h, with percentage differences in exchange between bound and free states colored as in (A). Amino acid number is indicated on the axis below. (C) Consensus percentage difference in exchange between bound and free states at 4.5 h shown on the structure of the RRMs bound to AUG12 RNA (PDB: 4BS2), colored as in (A) and generated in PyMOL. RRMs are represented as a cartoon, whereas RNA is represented as a stick. (D) HX for a representative RRM1 peptide. The dashed line represents the fully-deuterated condition. Data are mean ± SD (n=3–7 replicates run on MS per timepoint; some error bars too small to visualize; Welch’s t-test comparing bound and free at each timepoint; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). (E) Consensus percentage difference in exchange at the 2 s timepoint shown again as in (A). Aligned beneath it are peptides analyzed at 2 s, displayed as in (B). (F) Consensus percentage difference in exchange between bound and free states at 2 s shown on the cartoon representation of the AlphaFold structure of WT TDP-43 (Uniprot: Q13148), downloaded from the AlphaFold Protein Structure Database, colored as in (A) and generated in PyMOL. (G) Raw mass spectra at the 2 s timepoint for a representative peptide located in the CR. Signal corresponding to this peptide is colored red, whereas noise from overlapping peptide(s) is colored black. The blue dashed line indicates the monoisotopic peak, and the purple dashed line indicates the centroid value of the peptide in the fully deuterated sample.

Exchange in the NTD was similar between Clip34-bound and free states, indicating that Clip34 binding does not impact NTD structure (Fig. 2A and fig. S6, A to C), consistent with our finding that the NTD is dispensable for Clip34 chaperone activity (Fig. 1F). By contrast, extensive decreases in exchange occurred in the RRMs in the presence of Clip34, particularly at later timepoints, indicating that Clip34 stabilizes RRM1 and RRM2 structure (Fig. 2, A to D, and fig. S6, D to I). Previous NMR studies revealed the structure of the TDP-43 RRMs in complex with the 12-nt RNA AUG12 (Table S1) (17). Sub-localizing exchange differences revealed that Clip34 extensively stabilized both RRMs, including throughout the β-sheet RNA-binding surface (Fig. 2C), with particularly strong stabilization of RNA-binding residues F149 in RRM1 and F229 and F231 in RRM2 (Fig. 2B and fig. S5) (17). Additionally, at late timepoints Clip34 exerted one of its strongest stabilizing effects on the extreme C-terminal portion of RRM1 (Y155-D174; Fig. 2A), which likely prevents localized unfolding that may contribute to aggregation (16).

Exchange was rapid and unaffected by Clip34 across the majority of the PrLD, including IDR1 and IDR2, as expected for an intrinsically disordered domain (Fig. 2A and fig. S6, J to L). There was, however, one important exception. Exchange in the CR increased substantially in the presence of Clip34, particularly at early timepoints (1–18 s), indicating destabilization (Fig. 2A, E to G, fig. S6, M and N, and fig. S7, A to C). This destabilization spans residues M323 within the predicted α-helix through L340 at the end of the transient helical region (Fig. 2F) (18, 19, 35). Sub-localized analysis revealed strong destabilization at Q331 and S332 in the minor helical region (Fig. 2E and fig. S5). Thus, Clip34 induces a CR-specific allosteric destabilization spanning both major and minor helical segments, revealing an allosteric effect on the PrLD expected to antagonize aggregation.

CR peptide mass spectra exhibited bimodality at early timepoints, suggesting two TDP-43 populations with distinct CR structures (Fig. 2G and fig. S7, A to C). The slow-exchanging, more stabilized, population is strongly represented in spectra at early timepoints in the free state, but poorly represented in the Clip34-bound state (Fig. 2G and fig. S7, A to C). Thus, Clip34 binding decreases the probability of a more stabilized CR structure. Since the CR forms a transient α-helical structure (18, 19), these data suggest that in the absence of RNA, the CR forms a transient α-helix, whereas upon Clip34 binding, the CR is destabilized to disfavor α-helicity. Given the important role of CR α-helical structure in phase separation and aggregation (fig. S3B) (16, 18, 19, 29), this observation helps explain how Clip34 prevents TDP-43 aggregation. Specifically, Clip34 binding induces an aggregation-resistant form of TDP-43 with stabilized RRMs and a destabilized CR in the PrLD. Consistent with this mechanism, Clip34 failed to inhibit TDP-43^5FL aggregation (Fig. 1K) and exhibited reduced capacity to stabilize the RRMs and disrupt CR bimodal exchange in this variant (fig. S8 to S10; see Supplementary Text).

To complement these findings, all-atom molecular dynamics simulations of TDP-43 with the AUG12 RNA chaperone revealed that RNA engagement disrupts CR helicity and shifts the PrLD toward a more disordered ensemble (fig. S11 to S14; see Supplementary Text). This allosteric remodeling occurs without requiring direct RNA–CR contacts, indicating that conformational changes are propagated through indirect mechanisms. Thus, RNA binding constrains RRM conformational heterogeneity while reorganizing intramolecular interactions, providing strong support for the allosteric mechanism defined by HXMS.

Enhancing Clip34 activity against diverse disease-linked TDP-43 variants

For maximal therapeutic deployability, short RNA chaperones should mitigate aggregation of diverse disease-linked TDP-43 variants, including missense variants that cause disease, as well as TDP-43 bearing pathological PTMs. We assessed Clip34 chaperone activity against ALS/FTD-linked missense variants in RRM1 (P112H), the RRM1-RRM2 linker (K181E), or the PrLD (G295R, G298S, A321V, Q331K, M337V, A382T), as well as pathological PTM mimetics including phosphorylation (S292E, S409/410E, S292/409/410E), lysine acetylation (K145/K192Q), and physiological arginine methylation (R293F) (Fig. 3A and fig. S15, A and B) (1, 22, 36–39). These TDP-43 variants aggregated to a similar extent (fig. S15, C and D). Clip34 prevented aggregation of all disease-linked TDP-43 variants, with half-maximal inhibitor concentration (IC₅₀) values ranging from ~0.12 μM-0.69 μM (Fig. 3, B and F, and fig. S15E). For a subset of TDP-43 variants Clip34 prevented aggregation more effectively than for WT TDP-43 (IC₅₀~0.5 μM), including TDP-43^P112H (IC₅₀~0.28 μM) and TDP-43^K181E (IC₅₀~0.12 μM; Fig. 3B). This result is intriguing as these mutations have been suggested to reduce RNA binding (37–39), indicating that Clip34 can overcome this deficit. Clip34 also exhibited lower IC₅₀ values against phosphomimetic variants TDP-43^S409/410E (IC₅₀~0.29 μM) and TDP-43^{S292/409/410E} (IC₅₀~ 0.19 μM; Fig. 3B), which mimic phosphoforms of TDP-43 that accumulate in pathological inclusions (36, 40). Thus, Clip34 exhibits broad chaperone activity against diverse disease-linked forms of TDP-43.

Fig. 3. Clip34 and Clip34_UG6 effectively prevent aggregation of diverse disease-linked TDP-43 variants.

(A) Domain map of TDP-43 to scale (excluding MBP-His solubility tags). Missense mutants (top) and post-translational modification mimetics (bottom) investigated in this study are indicated. (B, C) TDP-43-MBP-His (5 μM) was incubated with TEV protease in the presence or absence of Clip34 (B) or Clip34_UG6 (C) for 16 h, measuring turbidity every minute in a plate reader. Standardized turbidity data was normalized to the No RNA condition for that replicate; the AUC of this data was utilized to calculate an IC₅₀ value (nonlinear regression: [inhibitor] vs. normalized response with variable slope). Data are mean ± SEM (n=5–8; one-way ANOVA with Dunnett’s correction comparing to WT; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). (D) RNA sequences of Clip34 and its variants. Red text represents nucleotides in Clip34, whereas blue text represents nucleotides in Clip34_UG6. (E, F) IC₅₀ values for Clip34 and Clip34 variants with WT TDP-43 (E) or TDP-43^K145/192Q (F). Data are mean ± SEM (n=3–4; n.c. indicates that an IC₅₀ value could not be accurately calculated). (G, H) Apparent K_D values calculated from the bound signal from individual replicates of EMSAs performed with 5’ 6-FAM Clip34 and the indicated TDP-43 protein (G), or TDP-43^K145/192Q with 5’ 6-FAM Clip34 or Clip34_UG6 (H). Data shown for TDP-43^K145/192Q with Clip34 is the same in both figure parts, and WT TDP-43 with Clip34 as in Fig. 1, J and L. Data are mean ± SEM (n=3; one-way ANOVA with Dunnett’s correction comparing to WT (G), or unpaired t-test (H); ***p< 0.001, ****p < 0.0001).

Clip34 was less effective against the lysine acetylation mimetic TDP-43^K145/192Q (IC₅₀~0.69 μM) than WT TDP-43 (Fig. 3F and fig. S15, E to G). K145Q:K192Q mutations reduce RNA binding (22, 41–43), which likely reduces Clip34 efficacy. This issue is problematic for Clip34 because TDP-43 acetylated at K145 accumulates in pathological inclusions in ALS (22).

To enhance activity against TDP-43^K145/192Q, we engineered Clip34_UG6, which harbors (UG)₄ in place of CAGAGACU in the middle of the sequence (Fig. 3, C and D). Clip34_UG6 binds to the isolated TDP-43 RRMs with higher affinity than Clip34 (44), and prevented aggregation of diverse disease-linked TDP-43 variants with IC₅₀ values ranging from ~0.16 μM-0.45 μM (Fig. 3C). Like Clip34, Clip34_UG6 was more effective against RRM1 variant TDP-43^P112H (IC₅₀~0.23 μM), linker variant TDP-43^K181E (IC₅₀~0.16 μM), and phosphomimetics TDP-43^S409/410E (IC₅₀~0.28 μM) and TDP-43^{S292/409/410E} (IC₅₀~0.21 μM) than WT TDP-43 (IC₅₀~0.45 μM; Fig. 3C and fig. S15F). Unlike Clip34, Clip34_UG6 was substantially more effective against the PrLD variant TDP-43^A321V (IC₅₀~0.25 μM) and the arginine methylation mimetic TDP-43^R293F (IC₅₀~0.26 μM) than WT TDP-43 (IC₅₀~0.45 μM; Fig. 3C). Critically, Clip34_UG6 prevented TDP-43^K145/192Q aggregation with efficacy similar to WT TDP-43 (IC₅₀~0.35 vs ~0.45 μM; Fig. 3C). This enhanced activity against TDP-43^K145/192Q could not be recapitulated by introducing (UG)₂ at different positions in the central portion of Clip34 (Fig. 3, D to F, and fig. S15G). Overall, Clip34_UG6 has broader applicability than Clip34 and is likely less affected by pathological K145/K192 acetylation.

To understand differences in chaperone activity, we assessed whether differing IC₅₀ values reflect alterations in binding affinity by determining the K_D of select TDP-43 variants for Clip34. In some cases, K_D tracked closely with IC₅₀ (Fig. 3, G and H, and fig. S15, H and I). For example, TDP-43^G295R bound Clip34 with similar affinity as WT TDP-43, consistent with comparable IC₅₀ values (Fig. 3, B and G, and fig. S15H). Additionally, TDP-43^K145/K192Q exhibited impaired binding to Clip34 compared to Clip34_UG6, with affinity reduced ~2.9-fold relative to WT TDP-43 (Fig. 3, G and H, and fig. S15, H and I). This finding helps explain why Clip34 is less effective against TDP-43^K145/192Q. However, in other cases K_D did not precisely track with IC₅₀. Despite exhibiting lower IC₅₀ values with Clip34 than WT TDP-43, TDP-43^P112H and TDP-43^K181E did not show increased binding affinity to Clip34 (Fig. 3, B and G, and fig. S15H). In fact, TDP-43^P112H displayed impaired binding to Clip34 compared to WT TDP-43 (Fig. 3G and fig. S15H). We suggest that although a certain threshold of binding affinity (K_D < 1.4 μM) is critical for a short RNA to effectively chaperone TDP-43, other components of the interaction beyond simple binding affinity must contribute to chaperone activity.

Mining short RNA sequence space for enhanced TDP-43 chaperones

To expand our arsenal of short RNA chaperones, we identified additional RNAs that prevent aggregation of disease-relevant TDP-43 variants (Fig. 4A and fig. S16A). The synthetic RNA (UG)₁₇ is an extremely potent chaperone (IC₅₀~0.2 μM) for WT TDP-43 (fig. S16B) and effectively chaperoned all tested variants, including RRM missense mutants (TDP-43^P112H, TDP-43^K181E) and PrLD mutants (TDP-43^G295R, TDP-43^Q331K) (fig. S16B). Thus, simple repetitive UG sequences effectively chaperone TDP-43.

Fig. 4. Malat1_start RNA displays enhanced chaperone activity against diverse disease-linked TDP-43 variants.

(A) RNA sequences of tested RNAs. (B) Heatmap displaying mean values of the individual IC₅₀ data shown in Fig. 3 (B, C, F) and in fig. S16 (C–E). (C) Apparent K_D values calculated from bound 5’ 6-FAM signal of the indicated RNAs, from individual replicates of EMSAs performed with WT TDP-43. Clip34 data is the same as shown in Fig. 3G. Data are mean ± SEM (n=3–4; one-way ANOVA with Tukey’s correction; *p < 0.05; **p < 0.01). (D) The K_D,app for WT TDP-43 with each indicated RNA, as shown in (C), is plotted against the IC₅₀ for that RNA with WT TDP-43, as shown in (B) (Pearson correlation; not significant). (E) Overlay of ¹H-¹⁵N heteronuclear single quantum coherence (HSQC) spectra of TDP-43 RRMs with Clip34 RNA (green, 2:1::[Clip34]:[TDP-43]) and Malat1_start RNA (magenta, 2:1::[Malat1_start]:[TDP-43]). (F) ¹H and ¹⁵N chemical shift perturbations (Δδ¹H (top) and Δδ¹⁵N (middle), respectively), and intensity ratios (bottom) of TDP-43 RRMs upon binding of Clip34 (green) and Malat1_start (magenta) RNA. Domain map of TDP-43 RRMs shown at the bottom, aligned to x-axes of graphs.

We identified potent short RNA chaperones from natural TDP-43-binding sequences (Fig. 4A), including SATIII (from pericentromeric satellite III repeats), Malat1_start (from MALAT1 long non-coding RNA), and CLN6_middle (from the CLN6 protein-coding transcript) (29, 32, 45, 46). Each RNA effectively chaperoned WT TDP-43 and disease-linked variants, including TDP-43^K145/192Q (Fig. 4B and fig. S16, C to H), with response patterns resembling Clip34 and Clip34_UG6. However, compared to Clip34 variants, SATIII, Malat1_start, and CLN6_middle showed fewer significant IC₅₀ differences between WT and disease-associated variants (Fig. 3, B and C, and fig. S16, C to E), suggesting that they engage TDP-43 more uniformly across variants. Malat1_start emerged as the most potent natural RNA chaperone, with the lowest IC₅₀ values against WT TDP-43 (IC₅₀~0.36 μM) and disease-linked variants (IC₅₀~0.17 μM–0.44 μM) (Fig. 4B and fig. S17, A and B). These findings support development of Malat1_start as a therapeutic short RNA.

RNAs with less steep inhibition curves (lower cooperativity) tended to be more effective chaperones (fig. S17C; see Supplementary Text). Additional RNAs derived from MALAT1 and CLN6 also prevented TDP-43 aggregation, although with lower potency than Malat1_start and CLN6_middle, respectively (fig. S17, D to I; see Supplementary Text). Short G-quadruplex-forming RNAs, such as LTR-III derived from HIV-1 LTR, also effectively chaperone TDP-43, expanding the structural diversity of RNA chaperones (fig. S17, J and K; see Supplementary Text).

Effective RNA chaperones can engage the TDP-43 RRMs differently

Binding affinity did not correlate with IC₅₀ values across our RNA panel for WT TDP-43 or TDP-43^P112H (Fig. 4, C and D, and fig. S18, A to C), confirming that binding affinity alone does not determine chaperone activity. Another factor that may influence chaperone activity is exactly how each RNA engages the TDP-43 RRMs. To understand whether different short RNAs engage the TDP-43 RRMs in the same way, we conducted NMR experiments on the isolated RRMs of TDP-43 in solution.

We first performed NMR experiments with WT TDP-43 RRMs and TDP-43^5FL RRMs in the presence of increasing concentrations of the short A(GU)₆ RNA. Compared with WT, TDP-43^5FL RRMs displayed several broadened resonances and lacked chemical-shift perturbations for a subset of residues, indicating weaker binding and involvement of fewer residues in binding RNA (fig. S18, D to G). Thus, TDP-43^5FL exhibits defective RNA binding (Fig. 1J) (26).

NMR experiments with WT TDP-43 RRMs saturated with Clip34, Clip34_UG6, and Malat1_start revealed overall similarities in resonance shifts and broadening in RRM1 (residues 138–142 and 160–172), indicating that these regions are critical for binding independent of RNA sequence (Fig. 4, E and F, and fig. S18, H and I). However, focused regions showed RNA-sequence-dependent differences in interaction and conformational dynamics (Fig. 4, E and F, and fig. S18, H and I). Clip34 produced unique shifts and broadening in the region of residues 145–151 in the third β-strand of RRM1 compared to Malat1_start (Fig. 4, E and F). This region harbors K145 that can be acetylated to disrupt RNA binding (22), and F147 and F149 that stack to interact with a U or G base, respectively, and are essential for RNA binding in RRM1 (17). Broadening and unique shifts were also observed in the region of residues 104–106 on the adjacent first β-strand of RRM1 (Fig. 4, E and F). Additionally, Clip34_UG6 also showed distinct perturbations compared to Clip34 in the vicinity of K145, as well as large chemical shift perturbations at three positions between 130 and 140, distinguishing these two similar sequences (fig. S18, H and I). These differences may help explain why Clip34 is less effective in chaperoning the pathological acetylation mimic, TDP-43^K145/192Q (22), in comparison to Clip34_UG6 and Malat1_start, which are more effective (Fig. 4B). Overall, we suggest that binding affinity, the mode of RRM engagement, the cooperativity of binding (inversely correlated with IC₅₀), and the capacity to directly stabilize the RRMs while allosterically destabilizing the conserved helical region in the PrLD combine to influence short RNA chaperone efficacy.

Malat1_start reverses TDP-43 condensation and aberrant aggregation

Malat1_start rapidly solubilized preformed TDP-43 condensates, whereas the control RNA (AC)₁₇ was ineffective (fig. S19, A to C). Likewise, Malat1_start partially restored aggregated TDP-43 to the soluble fraction, whereas the control RNA (AC)₁₇ had no effect (fig. S20, A and B). Electron microscopy revealed that Malat1_start remodeled large aggregates into smaller structures, reducing aggregate size ~100-fold and decreasing aggregate area and density (fig. S20, C to F; see Supplementary Text). Thus, Malat1_start solubilizes preformed TDP-43 condensates and aggregates.

Short RNAs reduce cytoplasmic TDP-43 aggregation in an optogenetic model

We assessed short RNA efficacy in immortalized human (HEK293) cells using an optogenetic TDP-43 proteinopathy model, in which TDP-43-Cry2olig undergoes blue-light–induced homo-oligomerization to form cytoplasmic puncta with disease hallmarks including p62 and pTDP-43(pS409/410) colocalization (13, 14). We tested Malat1_start (IC₅₀~0.36 μM), (UG)₁₇ (IC₅₀~0.2 μM), and CLN6_middle (IC₅₀~0.42 μM) against a control RNA. Malat1_start and (UG)₁₇ substantially reduced cytoplasmic TDP-43 inclusion area per cell, whereas CLN6_middle did not (Fig. 5, A and B, and fig. S21, A to C). Cells treated with Malat1_start and (UG)₁₇ exhibited nuclear foci, resembling cells expressing optoTDP-43 but not exposed to blue light (dark) where optoTDP-43 forms nuclear condensates (fig. S21, A to C). Biochemical fractionation confirmed that Malat1_start solubilized optoTDP-43 in cells (fig. S21, D to F). Malat1_start and (UG)₁₇, the two most effective RNAs at reducing cytoplasmic puncta, also ranked highest at preventing aggregation in vitro, demonstrating that our biochemical assay provides a powerful platform for identifying RNAs with activity in human cells.

Fig. 5. Malat1_start RNA mitigates aberrant TDP-43 phenotypes in optogenetic human cell models and patient-derived neurons.

(A) Sequence of the control (CTR) RNA. (B) OptoTDP-43 stable HEK293 cells were treated with the indicated RNA, followed by blue light exposure to induce Cry2olig oligomerization, and imaged after fixation. The average area of cytoplasmic puncta per cell, normalized to the average of the CTR-treated condition. Data are mean ± SEM (n=3 biological replicates; one-way ANOVA with Dunnett’s correction comparing to CTR; *p < 0.05; **p< 0.01). (C) Representative images of C9orf72-ALS patient iPSC-derived neurons treated with the indicated RNAs, stained with DAPI and for TDP-43 and MAP2. Scale bar indicates 25 μm. (D) The average ratio of TDP-43 nuclear to cytoplasmic signal, normalized to healthy control iPSC-derived neurons without RNA treatment. Data are mean ± SEM (n=3 biological replicates, represented as the average of n=2 technical replicates each; one-way ANOVA with Dunnett’s correction comparing to CTR; *p < 0.05).

Clip34 and Malat1_start RNAs do not cause TDP-43 loss of function

A potential concern with employing short RNAs is that stable binding to TDP-43 might interfere with essential RNA-processing reactions. Using a CUTS biosensor to detect TDP-43 loss-of-function (47), we found that Clip34 and Malat1_start did not interfere with TDP-43 function, whereas (UG)₁₇ elicited undesirable TDP-43 loss-of-function (fig. S21G; see Supplementary Text). Thus, (UG)₁₇ was excluded from further development. These results validate Clip34 and Malat1_start as therapeutic leads.

Short RNA chaperones restore physiological TDP-43 localization in ALS patient-derived motor neurons

We evaluated short RNA therapeutic potential using iPSC-derived motor neurons from healthy control or C9-ALS patients (hexanucleotide repeat expansion in C9orf72). C9-ALS iPSC-derived motor neurons exhibit TDP-43 pathology characterized by decreased TDP-43 nuclear/cytoplasmic ratio (36, 48). Compared to healthy control motor neurons, untreated or control-RNA-treated C9-ALS neurons showed reduced TDP-43 nuclear/cytoplasmic ratio (fig. S22, A and B). Treatment with Clip34 or Malat1_start, but not control RNA, restored the TDP-43 nuclear/cytoplasmic ratio to a similar value as observed for healthy control neurons (Fig. 5, C and D, and fig. S22B). This rescue was not due to differential RNA localization, as control and Malat1_start RNAs exhibited similar localization patterns in motor neurons (fig. S22C). Thus, Clip34 and Malat1_start counteract TDP-43 cytoplasmic mislocalization and restore nuclear localization in C9-ALS patient-derived motor neurons experiencing nuclear-pore dysfunction (48, 49).

Malat1_start restores TDP-43 functionality in stressed iPSC-derived motor neurons

We assessed whether short RNA chaperones restore TDP-43 function under stress conditions by treating control iPSC-derived human motor neurons with sodium arsenite to induce TDP-43 nuclear depletion and loss of function (fig. S23A) (15, 50, 51). Sodium arsenite triggered cryptic splicing of STMN2 and KCNQ2, disrupting their normal TDP-43-dependent splicing, which is critical for axonal regeneration and neuronal excitability, respectively (52–54). Malat1_start markedly reduced cryptic splicing of both STMN2 and KCNQ2 compared to control RNA (fig. S23, B to E; see Supplementary Text). Thus, Malat1_start restores TDP-43 splicing function in stressed motor neurons.

Malat1_start does not disrupt physiological neuritic RNA granules

We confirmed that short RNA chaperones do not perturb physiological TDP-43 function in axonal RNA granules by analyzing TDP-43-containing granules in neurites of healthy control and C9-ALS iPSC-derived motor neurons (fig. S24 and fig. S25; see Supplementary Text) (55). TDP-43-positive and Staufen-1–positive granules were unaffected by Malat1_start relative to the control RNA treatment (fig. S24 and fig. S25, A to D) (56). Thus, Malat1_start mitigates aberrant TDP-43 phenotypes while preserving physiological localization of TDP-43 and STAU1 to neuritic RNA granules.

Malat1_start confers neuroprotection, reverses TDP-43 aggregation, and restores TDP_43 function in mice

We tested the ability of Malat1_start to mitigate aberrant TDP-43 phenotypes in vivo using an acute spinal expression paradigm in mice (Fig. 6A) (57). We generated Adeno-associated virus (AAV) 9 expressing YFP-tagged TDP-43^ΔNLS (58), which results in cytoplasmic YFP-tagged TDP-43 due to a mutated nuclear localization signal. This virus was bilaterally injected across six sites in the cervical spinal cord of p180 mice (Fig. 6A). At day 7 (D7), viral delivery to the ventral horn was highly efficient, with robust expression of TDP-43^ΔNLS (fig. S26A), which formed cytoplasmic puncta (fig. S26B). At D7, animals were treated with saline or Malat1_start.

Fig. 6. Malat1_start RNA mitigates neurodegeneration, TDP-43 aggregation, and TDP-43 dysfunction in a mouse model of TDP-43 proteinopathy.

(A) Schematic of experimental paradigm. On Day 0 (D0), animals undergo laminectomy and bilateral AAV9 viral injection across the C4–C6 region, to express TDP-43^ΔNLS throughout the ventral horns of cervical spine. On D7, animals undergo a second surgery to receive RNA or saline. Histology was assessed at days 3 (D10) and 5 (D12) following treatment. (B) Representative immunohistochemistry images of ventral horns at D12, stained for ChAT. 20x magnification z-stack confocal images; scale bar indicates 100 μm. (C) ChAT⁺ motor neurons were manually counted within the ventral horn of spinal cord sections. Data are mean ± SEM (n=10 animals per condition; shown: one-way ANOVA with Dunnett’s correction comparing to D7 TDP-43; ****p < 0.0001; not shown: two-way ANOVA with Šídák’s correction: ****p < 0.0001 for Malat1_start versus saline at D10 and D12). (D) Representative 60x magnification immunofluorescent images from 5-day (12-day expressing) saline-treated (left) and RNA-treated (right) animals. Scale bar indicates 20 μm. (E) TDP-43 positive puncta were assessed in ChAT⁺ motor neurons at 60x magnification in the ventral horn for each animal. The average puncta size per neuron was calculated for each animal. Data shown are mean ± SEM (n=10 animals per condition; average of 30 neurons per animal; shown: two-way ANOVA with Šídák’s correction; ****p < 0.0001; not shown: one-way ANOVA with Dunnett’s correction comparing to D7 TDP-43: *p < 0.05 for Malat1_start at D10). (F) The ratio of relative Sort1 transcripts containing exon 17b to canonical Sort1 transcripts (WT) per animal. Data shown are mean ± SEM (n=10 animals per condition; shown: two-way ANOVA with Šídák’s correction; ***p < 0.001; ****p < 0.0001; not shown: one-way ANOVA with Dunnett’s correction comparing to D7 TDP-43: ****p < 0.0001 for D10 Malat1_start and ***p < 0.001 for D12 Malat1_start).

We verified RNA penetration to the spinal cord ventral horn and observed partial colocalization of TDP-43 puncta with Malat1_start, indicating successful target engagement (fig. S26, B and C). Motor neurons were quantified using choline acetyltransferase (ChAT) for cholinergic identity and NeuN for neuronal nuclei. Saline-treated TDP-43^ΔNLS animals displayed progressive loss of ChAT⁺ motor neurons, whereas Malat1_start-treated animals maintained ChAT⁺ motor neuron numbers (Fig. 6, B and C, and fig. S26D). Similarly, ventral horn NeuN⁺ neurons progressively degenerated in saline-treated animals but were preserved in Malat1_start-treated animals (fig. S26E). Thus, Malat1_start mitigates TDP-43-driven neurodegeneration in vivo.

Three-dimensional image analysis revealed that average TDP-43 puncta size was reduced in Malat1_start-treated animals compared to saline controls at D10 and D12 (Fig. 6, D and E, and fig. S27A). Furthermore, the average puncta size was reduced at D10 in Malat1_start-treated animals compared to the D7 baseline (Fig. 6E). This partial reversal of TDP-43 aggregation indicates that Malat1_start acts on existing aggregates in vivo. Average TDP-43 puncta number per motor neuron increased progressively in saline-treated animals, but was reduced in Malat1_start-treated animals (Fig. 6D and fig. S27, A and B). Thus, Malat1_start reduces the size and number of TDP-43 aggregates in mouse motor neurons.

We evaluated TDP-43 functionality in vivo by quantifying aberrant exon 17b inclusion in Sort1 transcripts, a known TDP-43 splicing target in mice and humans (43, 59). Total Sort1 mRNA amounts were comparable across treatment groups (fig. S27C), but Malat1_start treatment reduced the exon 17b–containing isoform ratio by ~50% relative to saline controls (Fig. 6F). Thus, Malat1_start corrects splicing defects caused by loss of TDP-43 function in vivo. We conclude that a single dose of Malat1_start reduced motor neuron degeneration, reversed TDP-43 aggregation, and restored TDP-43 function in a mouse model of TDP-43 proteinopathy.

Discussion

We define a mechanistic and therapeutic framework in which short RNA chaperones reprogram the conformational landscape of TDP-43 to enhance solubility, restore function, and counter neurodegeneration. Central to this framework is allosteric crosstalk between the RRMs, PrLD, and RNA. Clip34 binding to the RRMs does not merely anchor TDP-43 to nucleic acid, but allosterically biases the CR α-helix within the PrLD toward disordered, aggregation-resistant conformations, thereby remodeling TDP-43 to depopulate aggregation-prone states. RRM2 plays a key role in transmitting RNA-binding effects to the PrLD. Moreover, Clip34 imposes allosteric effects beyond destabilizing the CR α-helix, broadly maintaining PrLD disorder even in CR-deletion constructs. Thus, short RNA chaperones modulate an extensive allosteric network coupling RRM occupancy to PrLD conformation, establishing a mechanistic paradigm for regulating prion-like RBPs.

Critically, regulation is bidirectional: the PrLD, particularly the α-helicity of the CR, negatively regulates RRM affinity for RNA. Consistent with prior findings that the PrLD modulates TDP-43 binding at endogenous RNA regions (29), our data position the PrLD as a regulatory hub for tuning RNA interactions. CR deletion causes impaired neuronal function and behavioral abnormalities in mice (60), reinforcing the critical regulatory role of the CR. We propose that interplay between the RRMs, PrLD, and RNA maintains precise balance between TDP-43 solubility and self-assembly propensity, with TDP-43 responding dynamically to RNA availability. RNA-depleted environments, such as the cytoplasm of aging neurons or the interior of aging stress granules or alternative condensates, place TDP-43 at risk for pathological aggregation (14–16, 61). Indeed, pathological TDP-43 inclusions are typically RNA-depleted (14), consistent with failure of this regulatory axis.

Clip34 prevents TDP-43 aggregation even at substoichiometric amounts, a therapeutically favorable property rationalized by multivalent binding, wherein a single 34-nt RNA accommodates up to three TDP-43 molecules (62), depleting aggregation-permissive conformers below the nucleation threshold. Molecular dynamics simulations support this mechanism. RNA engagement remodels the TDP-43 conformational ensemble without requiring direct RNA–CR contacts, allowing limiting RNA amounts to effectively solubilize TDP-43 by depopulating aggregation-prone states.

RNA chaperones with broad applicability to sporadic and familial ALS/FTD should exhibit potent activity against disease-linked missense variants and pathological PTM mimetics. Although Clip34 effectively chaperoned diverse ALS/FTD-linked TDP-43 variants and PTM mimetics, it showed reduced activity against the acetylation mimetic K145/K192Q, which is problematic, as K145-acetylated TDP-43 accumulates in ALS inclusions (22). Mining natural and synthetic short RNAs uncovered additional chaperones, with Malat1_start proving effective against all tested variants, including TDP-43^K145/K192Q. Importantly, not all effective chaperones are suitable therapeutics: (UG)₁₇ prevented aggregation but interfered with TDP-43 function, disqualifying this RNA from development and cautioning against simple affinity-based screens for tight TDP-43 binders. In contrast, neither Malat1_start nor Clip34 compromised function (13, 27), revealing a therapeutic window in which short RNA chaperones restore TDP-43 solubility without disrupting activity. Malat1_start also corrected TDP-43 mislocalization in patient-derived motor neurons and rescued cryptic splicing under oxidative stress in control motor neurons (15, 16, 51), demonstrating robust activity extending beyond disaggregation to restoration of nuclear TDP-43 localization and function.

To address RNA delivery challenges, we employed 2’OMe-modified Malat1_start with select phosphorothioate linkages, which penetrates into the neuronal cytoplasm and targets TDP-43 aggregates in vivo. A single dose administered to mice after TDP-43 aggregation was widespread, motor neurodegeneration was underway, and TDP-43 functionality was compromised reduced TDP-43 aggregation, restored function, and prevented further degeneration. The capacity of short RNA chaperones to reverse—not merely prevent—disease-relevant TDP-43 pathology in vivo carries direct translational implications for treating patients with existing symptoms.

Our studies reveal mechanisms by which short RNA chaperones restore TDP-43 solubility and function, establishing a foundation for therapeutics targeting fatal TDP-43 proteinopathies. We envision that short RNAs preferentially engage cytoplasmic TDP-43, where competition from endogenous RNA is reduced relative to the RNA-rich nucleus (23), shifting equilibrium toward soluble forms competent for nuclear import. Upon binding to nuclear-import receptors, solubilized TDP-43 ejects the therapeutic RNA, freeing the RNA for further rounds of cytoplasmic solubilization (63). This recycling enables substoichiometric RNA to sustain TDP-43 solubilization while nuclear-import receptors return TDP-43 to the nucleus, restoring function. Importantly, our short RNAs resemble FDA-approved antisense oligonucleotides in size and chemistry, supporting translational feasibility for CNS delivery (24).

Despite these advances, several limitations warrant consideration and suggest avenues for future investigation. While short RNA chaperones reverse aberrant TDP-43 aggregation in vitro, in human cells, and in a mouse model, validation in additional mouse models that recapitulate TDP-43 pathology will be essential to establish therapeutic generalizability. Similarly, although short RNA chaperones promote TDP-43 functionality and nuclear localization in control and ALS patient-derived motor neurons, broader assessment across genetically diverse patient lines is needed to confirm efficacy across the heterogeneous ALS/FTD population. Finally, our study establishes a mechanistic framework linking RNA sequence and structure to restoration of TDP-43 solubility and functionality, yet systematic exploration of RNA sequence, structure, and chemical modification space will be required to define robust design principles and optimize therapeutic candidates. Whether we have achieved maximal potency with current short RNA chaperones remains an open question that warrants continued optimization efforts. Nonetheless, our findings establish short RNA chaperones as a promising therapeutic modality for countering TDP-43 proteinopathies.

Materials and Methods

Animals

Experimental procedures were approved by the Institutional Animal Care and Use Committee (IACUC) at Thomas Jefferson University and were conducted in compliance with the Guide for the Care and Use of Laboratory Animals from the National Institutes of Health. Surgical methods and behavioral tests listed are under IACUC-approved protocol for Piera Pasinelli (01914). The Thomas Jefferson University PHS Approved animal welfare assurance ID from the NIH Office of Laboratory Animal Welfare is D16–00051 (A3085–01). Female non-transgenic C57BL/6J mice (Strain #: 000664; RRID: IMSR_JAX:000664) aged 180 days were acquired from Jackson Laboratories (https://www.jax.org/strain/000664) and housed in an animal facility with controlled humidity, temperature and light cycles, with access to ad libitum water and standard chow. As analgesics are delivered on the basis of animal weight, and to better control variance in animal starting weight, female mice were exclusively chosen for this present study.

This study represents a total of 64 animals having undergone the spinal surgeries described below. Initial characterization of viral expression used n=3 animals in sham surgery, and TDP-43^ΔNLS groups expressing virus for 7 days. Animal numbers chosen for our large-scale cohort were based on effect sizes of ChAT+ expression noted in our test cohort. As per this assessment, guided by a power analysis assuming a desired power value of 0.80, we determined that an estimated 16 animals per timepoint (8 in each group) would be required. We have therefore chosen to include 10 animals per timepoints per group. In the main cohort of animals, there were n=8 sham non-injected animals and 50 animals which received injections to express TDP-43^ΔNLS. After one week, all 58 animals underwent a second surgery. Sham animals again received no treatment, whereas TDP-43^ΔNLS animals were subdivided into saline and RNA treatment groups (n=20 for each). Of these two groups, n=10 were used for each of the 3-day and 5-day endpoints for immunostaining analysis. The n=8 sham animals were also collected at the 5-day endpoint.

Cell lines

Induced pluripotent stem cell (iPSC) lines CS15iCTR-5, CS29iALS-n1, and CS52iALS-n6A were obtained from the Cedars-Sinai RMI iPSC Core, and are male. Line JH034 was obtained from Johns Hopkins Hospital, and is female. The estimated G4C2 repeat expansion sizes are >2.5 kb for line JH034, and 6–8 kb for lines CS29iALS-n1 and CS52iALS-n6A.

iPSCs were differentiated into motor neurons following previously described protocols (64, 65). iPSCs were cultured in Matrigel (Corning) and mTeSR+ (StemCell Technologies) and kept in a humidified chamber with regulated CO₂ (5%) and temperature (37°C). For differentiation, 1×10^6 iPSCs were plated in 6-well plates. Once cells reached ~90% confluency, media was changed from mTeSR+ to N2B27 media (50% DMEM:F12, 50% Neurobasal, plus NEAA, Glutamax, N2 and B27; all from Gibco) plus 10 μM SB431542 (StemCell Technologies), 100 nM LDN-193189 (Sigma-Aldrich), 1 μM RA (Sigma-Aldrich) and 1 μM Smoothened-Agonist (SAG, Cayman Chemical). Media was changed daily for a total of 6 days. Cells were then switched to N2B27 including 1 μM RA, 1 μM SAG, 4 μM SU5402 (Cayman Chemical) and 5 μM DAPT (Cayman Chemical), and media was changed daily until day 13. Neurons were dissociated on day 14 using TrypLE and DNAse I, and plated in Matrigel-coated 24-well plates with glass coverslips for confocal imaging studies. Cells were fed every other day and maintained for 13 days after plating in Neurobasal media + NEAA, Glutamax, N2, B27, plus 10 ng/mL BDNF, GDNF, CNTF (all from PeproTech) and 0.2 μg/mL Ascorbic acid (Sigma-Aldrich). HEK293 cells were obtained from ATCC (catalog # CRL-1573).

Microbe strains

Escherichia coli BL21 (DE3)-RIL cells (Agilent 230245) and BL21 Star (DE3) Chemically competent cells (Thermo Fisher C601003) were utilized for protein purification, with growth conditions as described in the purification sections. Escherichia coli XL10-Gold Ultracompetent cells (Agilent 200314) were utilized for cloning and plasmid propagation, and were grown at 37°C with the appropriate antibiotic.

Cloning

pJ4M was from Addgene (plasmid #104480; http://n2t.net/addgene:104480; RRID:Addgene_104480) (66). All other TDP-43 constructs purified were generated using the pJ4M plasmid. Partial PrLD deletion plasmids were generated previously (29). TDP-43^5FL was generated previously (14). TDP-43^S292E, TDP-43^R293F, TDP-43^S409/410E, and TDP-43^{S292/409/410E} plasmids were generated previously (36). All other TDP-43 disease-relevant variants and domain deletion plasmids, as well as the MBP-His plasmid, were generated via QuikChange Site-Directed Mutagenesis (Agilent 210518). MBP-FUS was generated previously (67).

Purification of TEV protease

TEV protease was purified as previously described (13). His-TEV plasmid was transformed into BL21 (DE3)-RIL E. coli and grown on an LB-ampicillin plate at 37°C for 16 h. The cells were then transferred to a starter culture of LB containing 100 μg/mL ampicillin and 34 μg/mL chloramphenicol, and incubated at 37°C for 2 h while shaking at 250 rpm. After 2 h, the starter culture was diluted 1:100 into the main culture of LB containing 100 μg/mL ampicillin and 34 μg/mL chloramphenicol. The main culture was shaken at 37°C and 250 rpm until the OD₆₀₀ reached ~0.7, then stored at 4°C for ~30 min while the incubator cooled to 15°C. The culture was then induced with 1 mM IPTG (MilliporeSigma 420322), and grown shaking at 250 rpm for 16 h at 15°C. After 16 h, the culture was harvested by centrifugation at 4658 rcf at 4°C for 25 min. The pelleted cells were resuspended in 30 mL Lysis Buffer (500 mM NaCl, 25 mM Tris-HCl pH 8.0, supplemented with 10 mM β-mercaptoethanol and cOmplete, EDTA-free Protease Inhibitor Cocktail (MilliporeSigma (Roche) 5056489001) at 1 tablet/50 mL buffer). The resuspended cells were lysed on ice with 1 mg/mL lysozyme (MilliporeSigma L6876) for 30 min, then sonication. The lysate was then centrifuged at 30,597 rcf at 4°C for 20 min.

A CV of 2.67 mL of Ni-NTA resin (QIAGEN 30250) was utilized per 1 L prep, and the Ni-NTA resin was equilibrated with 10 CV of MilliQ and 6 CV of Lysis Buffer. The clarified supernatant was rotated with Ni-NTA resin for 1.5 h at 4°C, then centrifuged at 179 rcf at 4°C for 5 min. The Ni-NTA resin was then washed with 25 CV of Wash Buffer (500 mM NaCl, 25 mM Tris-HCl pH 8.0, 25 mM imidazole, supplemented with 10 mM β-mercaptoethanol), with centrifugations performed at 179 rcf at 4°C for 2 min. The Ni-NTA resin was then resuspended in 2 CV of Wash Buffer and applied to a chromatography column. Protein was eluted with 5 CV of Elution Buffer (500 mM NaCl, 25 mM Tris-HCl pH 8.0, 300 mM imidazole, supplemented with 10 mM β-mercaptoethanol). Eluted protein was pooled and concentrated to ~10 mL utilizing an Amicon Ultra-15 Centrifugal Filter Unit, MWCO 30 kDa (Millipore UFC9030), by centrifugation at 716 rcf at 4°C. Concentrated protein was centrifuged at 716 rcf at 4°C for 3 min. Dialysis tubing was equilibrated in Dialysis Buffer (25 mM HEPES-NaOH pH 7.0, 5% glycerol, supplemented with 5 mM β-mercaptoethanol) for ~10 min. The protein was dialyzed in 5 L of Dialysis Buffer overnight, stirring at 4°C.

Dialyzed protein was centrifuged at 716 rcf at 4°C for 10 min. Filtered supernatant was purified using an FPLC with a HiTrap SP XL column, equilibrated in Low-Salt ion exchange (IEX) Buffer (25 mM HEPES-NaOH pH 7.0, 5% glycerol, supplemented with 5 mM DTT). The column was washed with 2 CV of Low-Salt IEX Buffer, then protein was eluted utilizing a 0–80% gradient with Low-Salt IEX Buffer as the base buffer, and High-Salt IEX Buffer (750 mM NaCl, 25 mM HEPES-NaOH pH 7.0, 5% glycerol, supplemented with 5 mM DTT) as the elution buffer. Based on the chromatogram, elution fractions were pooled and concentrated to ~40 mg/mL utilizing an Amicon Ultra-15 Centrifugal Filter Unit, MWCO 30 kDa (Millipore), by centrifugation at 716 rcf at 4°C. The concentrated protein was supplemented to 50% glycerol with 100% glycerol, then aliquoted, flash-frozen in liquid nitrogen, and stored at −80°C until use.

Purification of TDP-43-MBP-His

TDP-43-MBP-His, or MBP-His alone, plasmids were transformed into BL21 (DE3)-RIL E. coli and grown on LB-Kanamycin plates at 37°C for 16 h. The cells were then transferred to a starter culture of LB containing 50 μg/mL kanamycin and 34 μg/mL chloramphenicol, and incubated at 37°C for 4 h while shaking at 250 rpm. After 4 h, the starter culture was diluted 1:100 into the main culture of 1 L LB containing 50 μg/mL kanamycin, 34 μg/mL chloramphenicol, and 0.2% glucose. The main culture was shaken at 37°C and 250 rpm until the OD₆₀₀ reached ~0.25, then continued to grow while cooling to 16°C, and induced with 1 mM IPTG after reaching 16°C and an OD₆₀₀ of ~0.5–0.6. The induced culture was grown shaking at 250 rpm for 16 h at 16°C. After 16 h, the culture was harvested by centrifugation at 4658 rcf at 4°C for 20 min. The pelleted cells were resuspended in 20 mL of Resuspension/Wash Buffer (1 M NaCl, 20 mM Tris-HCl pH 8.0, 10% glycerol, 10 mM imidazole pH 8.0, supplemented with 1 mM DTT, 5 μM Pepstatin A, 100 μM PMSF, and cOmplete, EDTA-free Protease Inhibitor Cocktail at 1 tablet/50 mL buffer). The resuspended cells were lysed on ice with 1 mg/mL lysozyme for 30 min, then sonication. The lysate was then centrifuged at 30,966 rcf at 4°C for 20 min.

A CV of 5 mL of Ni-NTA resin (QIAGEN) was utilized per 1 L prep, and the Ni-NTA resin was equilibrated with 18 CV of MilliQ and 3 CV of Resuspension/Wash Buffer. The clarified supernatant was rotated with Ni-NTA resin for 1 h at 4°C, then centrifuged at 179 rcf (2000 rpm; 50 mL tubes) at 4°C for 4 min. The Ni-NTA slurry was then applied to a chromatography column, with the flow-through re-applied once. At 4°C, the column was washed with 10 CV of Resuspension/Wash buffer, then eluted in 3 CV of Nickel Elution Buffer (1 M NaCl, 20 mM Tris-HCl pH 8.0, 10% glycerol, 300 mM imidazole pH 8.0, supplemented with 1 mM DTT, 5 μM Pepstatin A, 100 μM PMSF, and cOmplete, EDTA-free Protease Inhibitor Cocktail at 1 tablet/50 mL buffer). Eluted fractions were stored overnight at 4°C.

Eluted fractions were pooled based on purity determined by SDS-PAGE. Approximate protein concentration was determined by Bradford, and 1 mL amylose resin (New England Biolabs E8021L) per 6 mg protein was utilized as the amylose resin CV. Amylose resin was equilibrated in ~10 CV MilliQ and 3 CV Resuspension/Wash Buffer. The protein was rotated with amylose resin at 4°C for 30 min, then centrifuged at 179 rcf at 4°C for 4 min. The amylose slurry was then applied to a chromatography column, with the flow-through re-applied once. At 4°C, the column was washed with 5 CV of Resuspension/Wash buffer, then eluted in 3 CV of Amylose Elution Buffer (1 M NaCl, 20 mM Tris-HCl pH 8.0, 10% glycerol, 10 mM imidazole pH 8.0, supplemented with 1 mM DTT, 5 μM Pepstatin A, 100 μM PMSF, 10 mM maltose). Eluted fractions were pooled based on purity determined by SDS-PAGE, then concentrated utilizing an Amicon Ultra-15 Centrifugal Filter Unit, MWCO 50 kDa (Millipore UFC9050), by centrifugation at 716 rcf at 4°C, until a concentration of ~150–200 μM was achieved. The protein was aliquoted, flash-frozen in liquid nitrogen, and stored at −80°C until use.

Purification of TDP-43-MBP-His utilized for condensation assays

Purification was performed based on a previously described protocol (66). TDP-43-MBP-His plasmid was transformed into One Shot BL21 Star (DE3) E. Coli (Thermo Fisher Scientific) and grown on LB-Kanamycin plates at 37°C for 16 h. The cells were then transferred to a starter culture of LB containing 50 μg/mL kanamycin, and grown at 37°C and 250 rpm. The starter culture was diluted into a main culture of 1 L LB containing 50 μg/mL kanamycin and 0.2% glucose, which was grown at 37°C and 250 rpm until reaching an OD₆₀₀ of ~0.5–0.6. The culture was then incubated at 4°C for 30–45 min, then induced with 1 mM IPTG and grown at 16°C and 250 rpm for 16 h. After 16 h, the culture was harvested by centrifugation at 4658 rcf at 4°C for 20 min. The pelleted cells were resuspended in 30 mL of Lysis Buffer (1 M NaCl, 20 mM Tris-HCl pH 8.0, 10 mM imidazole pH 8.0, 10% glycerol, and supplemented with 2.5 mM β-mercaptoethanol and cOmplete, EDTA-free Protease Inhibitor Cocktail at 1 tablet/50 mL buffer). The resuspended cells were lysed by sonication, then centrifuged at 48,384 rcf at 4°C for 1 h, then filtered.

The filtered lysate was purified using an FPLC with a XK 50/20 column (Cytiva) packed with Ni-NTA agarose beads (Qiagen), equilibrated in Lysis Buffer. The column was washed with 3 CV of Buffer A (Lysis Buffer without Protease Inhibitor Cocktail), then protein was eluted utilizing a 0–80% gradient with Buffer A as the base buffer, and Buffer B (1 M NaCl, 20 mM Tris-HCl pH 8.0, 500 mM imidazole pH 8.0, 10% glycerol, and supplemented with 2.5 mM β-mercaptoethanol) as the elution buffer. Desired elution fractions were pooled, concentrated with an Amicon Ultra-15 Centrifugal Filter Unit, MWCO 50 kDa (Millipore), and filtered. The filtered protein was then further purified using an FPLC with a 26/600 Superdex 200 pg column (Cytiva), equilibrated in SEC Buffer (300 mM NaCl, 20 mM Tris-HCl pH 8.0, and supplemented with 1 mM DTT). The fractions from the second out of three peaks, as determined by absorbance at 280 nm, were pooled and concentrated with an Amicon Ultra-15 Centrifugal Filter Unit, MWCO 50 kDa (Millipore) until a concentration of at least 250 μM. The protein was aliquoted, flash-frozen in liquid nitrogen, and stored at −80°C until use.

Purification of MBP-FUS

MBP-FUS was purified based on previous protocols (67). In brief, MBP-FUS plasmid DNA was transformed into One Shot BL21 Star (DE3) E. coli (Thermo Fisher Scientific) cells via heat shock, plated on LB agar plates containing 100 μg/ml ampicillin and incubated overnight at 37°C. The next day, bacterial cultures were scaled-up in LB media supplied with 100 μg/mL ampicillin and 0.2 % glucose and grown to an OD₆₀₀ of 0.6 at 37°C. Expression was induced by adding 1 mM IPTG followed by incubation for 16 h at 16°C and 250 rpm. Cells were harvested by centrifugation for 20 min at 4658 rcf and 4°C. Cells were resuspended in lysis buffer (20 mM HEPES, pH 7.4, 50 mM NaCl, 2 mM EDTA, 10% glucose, 2 mM DTT), supplied with 20 mg/mL lysozyme and incubated on ice for 30 min. After sonication, the lysate was centrifuged for 20 min at 30,966 rcf and 4°C.

The supernatant was pooled and added to 5 mL amylose beads (New England Biolabs) equilibrated with resuspension buffer and nutated for 2 h at 4°C. Subsequently, the beads were washed with resuspension buffer and eluted using the same buffer supplied with 10 mM maltose. For further purification and RNA removal, the eluted sample was loaded onto a Heparin column (HiTrap Heparin HP, Cytiva) equilibrated with resuspension buffer using an FPLC. The sample was eluted with a linear gradient ranging from 0–80% high-salt buffer (20 mM HEPES, pH 7.4, 1 M NaCl, 2 mM EDTA, 10% glucose, 2 mM DTT) over 90 mL. Protein-containing fractions were pooled, concentrated using an Amicon spin concentrator (Merck Millipore, MWCO 50 kDa) and flash-frozen in liquid nitrogen.

Purification of TDP-43 RRMs utilized for NMR experiments

The TDP-43^5FL RRMs plasmid was synthesized in the pJ411 vector by GenScript. WT TDP-43 RRMs (102–269) was expressed via a codon-optimized sequence in the pJ411 vector. Protein growth and purification protocols were adapted from the literature (17). The protein was grown in BL21 Star (DE3) E. coli cells in M9 minimal media supplemented with ¹⁵NH₄Cl for isotopic labeling. Bacterial cultures were grown at 37°C with agitation at 200 rpm to an optical density of 0.8. The cultures were induced with 1 mM IPTG and grown for 4 additional hours before harvesting by centrifugation at 6000 rpm and 4°C for 15 min. and resuspended in lysis buffer (20 mM HEPES pH 7.5, 1 M NaCl, 30 mM imidazole, 1 mM DTT) supplemented with an EDTA-free protease inhibitor cocktail (Roche). The cells were lysed using an EmulsiFlex C3 homogenizer (Avestin), and the lysate was cleared by centrifugation at 20,000 rpm and 4°C for 1 h. The protein was eluted via a nickel HisTrap HP column (GE Healthcare) by affinity chromatography with a linear gradient of elution buffer (20 mM HEPES pH 7.5, 1 M NaCl, 300 mM imidazole, 1 mM DTT). The hexahistidine tag was cleaved by overnight dialysis with 0.03 mg/mL TEV protease at room temperature (20 mM HEPES pH 7.5, 500 mM NaCl, 1 mM DTT). The tag and TEV protease were removed with an additional elution over the HisTrap HP column. The purified protein was buffer exchanged into NMR buffer (20 mM KPi pH 6.8, 1 mM DTT), concentrated to ~1 mM, flash frozen, and stored at −80°C.

RNA oligonucleotides

All RNA oligonucleotides utilized were purchased from Integrated DNA Technologies (IDT) or Horizon Discovery. All RNAs utilized for in vitro pure protein assays were unmodified and purified with standard desalting (except for RNAs utilized for NMR, which were HPLC purified), and were resuspended in RNase-free water, with nanodrop measurement performed to calculate the RNA concentration. RNAs utilized for cellular experiments were HPLC purified and were fully 2’OMe modified; the Clip34 RNA also had five phosphorothioate backbone modifications on each end of the RNA. RNAs utilized for mouse experiments were purified with in vivo HPLC, were fully 2’OMe modified, and had five phosphorothioate backbone modifications on each end of the RNA. A subset of RNA utilized for mouse experiments contained a 5’ Cy5 fluorophore.

In vitro TDP-43 aggregation prevention assay

RNA was thawed on ice, then serially diluted in water to achieve the desired working concentrations. Protein was thawed on ice, then centrifuged at 21,300 rcf for 10 min. at 4°C. Protein was then buffer exchanged into aggregation assay buffer (166.66 mM NaCl, 22.22 mM HEPES-NaOH pH 7.0, 1.11 mM DTT) using Micro Bio-Spin Chromatography Columns (BIO-RAD 7326200), following manufacturer’s instructions. After buffer exchange, nanodrop measurements were performed to calculate protein concentration. Aggregation assay buffer, and subsequently protein, were added to the tubes containing water/RNA, in order to achieve sample reactions with final concentrations of 5 μM TDP-43, 150 mM NaCl, 20 mM HEPES-NaOH pH 7.0, 1 mM DTT, with varying concentrations of RNA. The reactions were incubated for 15 min at RT after the addition of protein. To a 96-well nonbinding plate (Greiner Bio-One 655906), 0.25 μg TEV protease was added for a final concentration of 2.5 μg/mL, or TEV protease elution buffer for the No TEV control. The reactions were then added to the 96-well plate. The 96-well plate was sealed using parafilm. Turbidity was measured at absorbance 395 nm in a Tecan plate reader (Infinite M1000 or Safire2) for 16 h, measuring every 1 min. Plate reader measurements were conducted at ambient temperature, typically ~25–30°C.

For quantification, turbidity data was first standardized by setting the initial value for each well to 0. For the standardized data, any negative values were also set to 0. For normalization, the maximum value of the standardized No RNA condition data for a replicate was set to 100, with all other conditions for that protein in the replicate normalized based on this. Area under the curve (AUC) was calculated for the normalized data. To then normalize the AUC data, the AUC for the No RNA condition was set to 100. This analysis was performed separately for each replicate. The normalized AUC was then used to calculate an IC₅₀ value for each replicate, utilizing nonlinear regression: [inhibitor] vs. normalized response with variable slope. The IC₅₀ value for each replicate was then combined to generate summary data.

In vitro FUS phase separation prevention assay

RNA was thawed on ice and diluted in water and 2x LLPS buffer to achieve the desired RNA concentration in 1x FUS LLPS buffer (20 mM HEPES-NaOH pH 7.4, 1 mM DTT). Protein was thawed on ice, then centrifuged at 21,300 rcf for 5 min at 4°C. Protein was diluted to 6 μM in FUS elution buffer (570 mM NaCl, 20 mM HEPES-NaOH pH 7.4, 2 mM EDTA, 10% glycerol, 0.5 mM DTT). TEV protease was diluted to 0.12 mg/mL in 1x FUS LLPS buffer. Equal volumes of protein and RNA were mixed, then transferred to a 384-well glass bottom plate (Azenta MGB101-1-2-LG-L). Diluted TEV protease was then added directly to the samples in the 384-well plate, to achieve final concentrations of 2 μM FUS protein, varying concentrations of RNA, 0.04 mg/mL TEV protease, 190 mM NaCl, 20 mM HEPES-NaOH pH 7.4, 0.83 mM DTT, 0.67 mM EDTA, and 3.33% glycerol. Turbidity was measured at absorbance 395 nm in a BMG Labtech plate reader (CLARIOstar Plus) for ~2–2.5 h at 26°C, measuring every 1 min. At the endpoint of turbidity measurements, samples were imaged within the plate by brightfield microscopy with a 100x objective (EVOS M5000).

Electrophoretic mobility shift assay

Protein was thawed on ice, then centrifuged at 21,300 rcf for 10 min at 4°C. Protein was then buffer exchanged into 150 mM NaCl, 20 mM HEPES-NaOH pH 7.0 (or pH 6.0 where indicated), 10% glycerol, 1 mM DTT using BIO-RAD Micro Bio-Spin Chromatography Columns, following manufacturer’s instructions. After buffer exchange, nanodrop measurements were performed to calculate protein concentration. Protein was diluted in buffer to achieve a working concentration of 50 μM, in EMSA assay buffer (150 mM NaCl, 20 mM HEPES-NaOH pH 7.0 (or pH 6.0 where indicated), 10% glycerol, 1 mM DTT, 20 ng/μL bovine serum albumin (BSA; Thermo Fisher 23209), 2.5 ng/μL yeast tRNA (Thermo Fisher AM7119), 0.4 U/μL RNasin (Promega N2511)). Protein was then serially diluted in EMSA assay buffer to achieve a range of protein concentrations. 20 μM 5’ 6-FAM RNA resuspended in RNase-free water was diluted to 1 μM in EMSA assay buffer (10x working concentration). 10x RNA was then added to protein samples to achieve 100 nM (1x) RNA and a range of protein concentrations in EMSA assay buffer. Samples were incubated at RT for 30 min. During this incubation, 6% DNA Retardation gels (Thermo Fisher EC63655BOX) were pre-run in 0.5x TBE buffer at 150 V for ~20 min. 1x dye was prepared by dilution of 5x dye (20 mM EDTA, 50% sucrose, 0.25% bromophenol blue) in EMSA assay buffer. After incubation, heparin (MilliporeSigma H3393) was added to each sample to achieve a final concentration of 0.5 mg/mL heparin. 15 μL of 1x dye was loaded in the first lane to monitor sample progression, while 15 μL of undyed sample was loaded in remaining lanes. Gels were run at 150 V for 40 min. Gels were then imaged on a Typhoon Scanner using FAM fluorescence measurement. The signal of bound TDP-43 in each lane was quantified utilizing Image Studio Lite.

SDS-PAGE

Samples were diluted in 3x sample buffer (187.5 mM Tris-HCl, 6% SDS, 30% glycerol, 0.05% bromophenol blue, pH 6.8, 1.42 M β-mercaptoethanol) and boiled at 95°C for 5 min. Precision Plus Protein Dual Color Standard (BIO-RAD 1610374) and samples were loaded on Tris-HCl gels (4–15% or 4–20% as indicated) (BIO-RAD 3450027, 3450033), and run at 175 V for 1 h 15 min. Gels were stained with Coomassie Brilliant Blue, followed by incubation with Destain I (40% methanol, 7% acetic acid), then Destain II (5% methanol, 7% acetic acid) overnight before imaging.

Fluorescence 5’ 6-FAM Clip34 3’ BHQ1 assay

Protein was thawed on ice, then centrifuged at 21,300 rcf for 10 min at 4°C. Protein was then buffer exchanged into fluorescence assay buffer (150 mM NaCl, 20 mM HEPES-NaOH pH 7.0, 1 mM DTT) using BIO-RAD Micro Bio-Spin Chromatography Columns, following manufacturer’s instructions. After buffer exchange, nanodrop measurements were performed to calculate protein concentration. Protein was diluted in fluorescence assay buffer to achieve a working concentration of 50 μM. Protein was then serially diluted in fluorescence assay buffer to achieve a range of protein concentrations. 20 μM 5’ 6-FAM 3’ BHQ1 RNA resuspended in RNase-free water was diluted to 1 μM in fluorescence assay buffer (10x working concentration). 10x RNA was then added to a 96-well nonbinding plate (Greiner). Protein samples were then also added to the 96-well plate, to achieve final concentrations of 100 nM 5’ 6-FAM Clip34 3’ BHQ1, and a range of protein concentrations, in 150 mM NaCl, 20 mM HEPES-NaOH pH 7.0, 1 mM DTT. Samples were incubated at RT for 30 min.

Fluorescence was measured in a Tecan plate reader (Spark) at 25°C. Excitation: 475 nm; bandwidth: 15 nm. Emission: 520 nm; bandwidth: 20 nm. A gain value of 80 was used for all trials. The turbidity value at 30 min was utilized. For each protein variant, it was validated that the signal was stable at the 30 min timepoint by measuring after sample addition to the plate, for 1 h every 1 min, for at least one replicate. For quantification, (F-F₀)/F₀ values were calculated for each condition: the average signal for the RNA alone (no TDP-43) condition was subtracted from the signal for a condition, which was then divided by the average signal for the RNA alone (no TDP-43) condition. EC₅₀ values were determined from this data, by performing nonlinear regression: [agonist] vs. response with variable slope for each replicate.

Hydrogen/deuterium-exchange mass spectrometry (HXMS)

RNA was thawed on ice where needed. Protein was thawed on ice, then centrifuged at 21,300 rcf for 10 min at 4°C. For “free” conditions, protein was buffer exchanged into Non-Deuterated Buffer (150 mM NaCl, 20 mM HEPES-NaOH pH 7.0, 1 mM DTT) using BIO-RAD Micro Bio-Spin Chromatography Columns, following manufacturer’s instructions. After buffer exchange, nanodrop measurements were performed to calculate protein concentration. Protein was then diluted in Non-Deuterated Buffer to make a protein sample consisting of 20 μM TDP-43-MBP-His in 150 mM NaCl, 20 mM HEPES-NaOH pH 7.0, 1 mM DTT. Deuterium on-exchange was performed at 25°C by mixing 10 μL of sample with 40 μL of deuterium on-exchange buffer (D₂O-based; 150 mM NaCl, 20 mM HEPES-NaOD pH 7.0, 1 mM DTT), resulting in a D₂O concentration of 80%. At the indicated timepoint, the exchange reaction was quenched by addition of 10 μL of ice-cold 250 mM phosphoric acid, to achieve a final pH of pH 2.5. For non-deuterated samples, 10 μL of sample was mixed with 40 μL of Non-Deuterated Buffer, then quenched by addition of 10 μL of ice-cold 250 mM phosphoric acid. For the fully deuterated sample, 10 μL of sample was mixed with 40 μL of on-exchange buffer, incubated at 30°C for ~18 h, then quenched by addition of 10 μL of ice-cold 250 mM phosphoric acid. For “Clip34-bound” conditions, all procedures were the same, except that the protein was buffer exchanged into 166.67 mM NaCl, 22.22 mM HEPES-NaOH pH 7.0, 1.11 mM DTT, then diluted into the same buffer along with RNA and water, to achieve final sample concentrations of 20 μM TDP-43-MBP-His and 40 μM Clip34 in 150 mM NaCl, 20 mM HEPES-NaOH pH 7.0, 1 mM DTT.

HX measurements from 20 s to 14.5 h were performed at pH 7.0. In order to measure less protected, faster exchanging parts of the protein, another set of measurements was performed at pH 6.0. Due to the direct dependence of the intrinsic exchange rate on OH⁻ concentration, these measurements can be put on the same time axis as the pH 7.0 measurements by dividing the actual exchange time by 10 (68). For the subset of timepoints done with pH 6-based buffer, all procedures were the same, except that the HEPES-NaOH component of both the Non-Deuterated Buffer and on-exchange buffer was at pH 6.0, and the quench reagent utilized was ice-cold 145 mM phosphoric acid (to achieve a final pH of pH 2.5). All 1 s, 2 s, 6 s, and 18 s timepoints were collected utilizing pH 6.0 buffer; 1 min and 3 min timepoints were collected with some replicates utilizing pH 6.0 buffer and others utilizing pH 7.0 buffer; 20 s, and 10 min and longer timepoints were collected utilizing pH 7.0 buffer. For example, the “1 min” timepoints were measured by 1 min of on-exchange at pH 7.0, or 10 min of on-exchange at pH 6.0. The agreement between these duplicated replicates indicates that the protein structural stability measured by HX is not affected by the pH change. In addition, WT TDP-43 and Clip34 were confirmed to maintain binding at pH 6.0.

For MS analysis, the sample was digested by loading 50 μL onto a homemade pepsin column maintained at 0°C, where pepsin was immobilized by coupling to POROS 20 AL support (Applied Biosystems) and packed into a column housing of 2 mm × 2 cm (64 μL) (Upchurch) (69). The protease-generated fragments were then collected onto a TARGA C8 5 μM Piccolo HPLC column (1.0 × 5.0 mm, Higgins Analytical) and separated on a C8 analytical column utilizing a shaped 10–45% Buffer B gradient at 8 μL/min (Buffer A: 0.1% formic acid; Buffer B: 0.1% formic acid, 99.9% acetonitrile). The effluent was electrosprayed into the mass spectrometer. Peptides were identified from non-deuterated samples by MS/MS (Thermo Q Exactive), by analyzing MS/MS data using SEQUEST Proteome Discoverer (ThermoFisher). Peptide identification was performed separately for WT TDP-43 and TDP-43^5FL. Deuterated samples, and additional non-deuterated reference samples, were analyzed by MS (Thermo Q Exactive or Thermo Exactive Plus EMR).

HDExaminer software was utilized to process and analyze the HXMS data. The timepoints for samples performed at pH 6.0 were input as one-tenth of the actual on-exchange time. ExMS2, a MATLAB-based program, was used to prepare the peptide pool used by HDExaminer, from the SEQUEST output files for MS/MS data analysis. HDExaminer uses a non-deuterated sample as the reference for identifying deuterated peptides. Manual adjustment of retention times and m/z windows was performed as needed to correct any initial errors. Each deuterated peptide is corrected for back exchange after quenching, by normalizing to the maximal deuteration of that peptide as detected in the fully deuterated sample. For calculating the peptide deuteration at each timepoint, HDExaminer identifies the peptide envelope centroid values for both the non-deuterated and deuterated peptides.

HXMS data visualization

For visualizing the difference in peptide deuteration for each peptide, the HDExaminer data was visualized using MATLAB. At each timepoint, the average deuteration percent for a peptide in either the free or bound condition was calculated by taking the average deuteration percent of all replicates for the peptide at that timepoint that were identified with medium or high confidence. These values were then analyzed in MATLAB by subtracting the average deuteration percent of the peptide in the Clip34-bound state from the average deuteration percent of the peptide in the free state. This data is plotted in MATLAB according to the colors shown in the color legends in the figures (as in Fig. 2). Peptides of the same sequence but different charge states are plotted to allow visualization of the agreement across separate charge states for a unique peptide sequence.

To generate plots of consensus exchange difference for each timepoint, the exchange differences of the peptides at that timepoint were manually analyzed, with the consensus exchange difference determined based on the average classification of all peptides including a specific residue. This was done via a scoring system, where a peptide with a difference of less than 10% receives a score of 0, a peptide with a difference of ≥ 10% (light blue according to legend) receives a score of −1, a peptide with a difference of ≥ 20% (medium blue) receives a score of −2, a peptide with a difference of ≥ 30% (dark blue) receives a score of −3, a peptide with a difference of ≤ −10% (light red) receives a score of +1, a peptide with a difference of ≤ −20% (medium red) receives a score of +2, and a peptide with a difference of ≤ −30% (dark red) receives a score of +3. As peptides were binned according to this scoring system, the displayed consensus percentage differences in exchange do not report the exact value of the percentage difference for each peptide, and do not report the proximity of each peptide’s behavior to the cutoff value for each score. The average score for each residue was rounded to the nearest whole number, and this data was plotted in GraphPad Prism, with the rounded score value for each residue colored according to the same scoring system as described for the manual analysis above.

To generate plots of HX data for representative peptides displaying exchange as the number of deuterons, the HDExaminer output of this data was visualized using GraphPad Prism. For a peptide, the values for the number of deuterons from HDExaminer was taken for each replicate with high or medium confidence at each timepoint, for both free and bound states. This was then plotted in GraphPad Prism.

To generate plots of mass spectra, HDExaminer output was visualized in GraphPad Prism. The raw mass spectrum data for a particular replicate of a specific timepoint in either the free or bound state was copied into GraphPad Prism. This data was manually analyzed to determine the signal that corresponded to the desired peptide, based on the signal corresponding to the appropriate m/z values. All signal is displayed in the mass spectra plots, but the signal determined to correspond to the correct peptide is colored red, for ease of visualization. Dashed guidelines for visualization are also displayed, with the blue line corresponding to the value of the monoisotopic peak for that peptide, and the purple line corresponding to the centroid value of the peptide in the fully deuterated condition. Representative spectra for each state at each timepoint were chosen by determining the average value of the centroid for all replicates with peptides of high or medium confidence, for that state and timepoint. The spectrum displayed as the representative spectrum corresponds to the replicate with the centroid value closest to the average of these centroid values.

Simulation system preparation

The all-atom model of full-length and ΔNTD TDP-43 in complex with RNA was constructed using MODELLER, based on multiple experimentally resolved structures (for NTD PDB: 5MDI, for RRM1/2 PDB:4BS2, for CTD PDB:2N3X) and a well-tempered ensemble for the PrLD generated in our previous work (70) as templates. To create the TDP-43^5FL construct, five Phe residues (positions 147, 149, 194, 229, and 231) were mutated to leucine using the “swapaa” module in ChimeraX. The resulting structures were solvated using GROMACS v2022.5. Specifically, the initial model was placed in an octahedral simulation box (l = 15.0 nm for FL TDP-43 and l = 14 nm for TDP-43^ΔNTD) and solvated with explicit water molecules. Na⁺ and Cl⁻ ions were added to achieve a salt concentration of 100 mM, along with additional counterions to ensure overall charge neutrality.

MD simulations used AUG12 RNA, a 12-nucleotide sequence derived from TDP-43 CLIP data and designed by Lukavsky et al. (17), rather than Clip34. Since experimental data show that neither the NTD nor PrLD directly involve RNA binding, the 12-nt sequence captures the essential RRM-RNA interactions, and we determined that AUG12 inhibits TDP-43 aggregation, this sequence is sufficient to investigate allosteric regulation upon RRM binding.

Selection of Protein and RNA Force Fields

Proteins were modeled using the Amber03ws force field with TIP4P/2005 water and improved ion parameters from Luo and Roux (71). This force field was chosen for its optimized protein-water interactions that prevent artifactual compaction in intrinsically disordered regions (72, 73). RNA was modeled using the χOL3 force field with refined glycosidic torsions (74), adjusted phosphate oxygen radii (75), and scaled RNA-water interactions to address known RNA simulation artifacts including unrealistic ladder-like structures and overestimated electrostatic interactions. Force field compatibility was validated in our prior work (76).

Simulation Protocol

The solvated protein systems were first subjected to energy minimization using the steepest descent algorithm in GROMACS (77). This was followed by an initial configuration relaxation, consisting of a 100ps NVT equilibration using the V-rescale thermostat (78) at a 2 fs time step, and an additional 100ps NPT equilibration using the Parrinello–Rahman barostat (79) with isotropic coupling and a pressure relaxation time of 5 ps. Following equilibration, the topology(.top) and coordinate(.gro) files generated by GROMACS were converted into AMBER-compatible input formats (.parm7 and .rst7) using the ParmEd module in AmberTools22 (80). To enable a 4 fs time step during production runs, hydrogen atom masses were increased to 1.5 amu (81).

Production simulations were performed in AMBER22 under constant pressure (1 bar) and temperature (300 K) conditions (80). Temperature was maintained using Langevin dynamics with a friction coefficient of 1 ps⁻¹, while pressure was controlled using a Monte Carlo barostat with isotropic coupling (relaxation time of 1.0 ps) (82). Short-range nonbonded interactions were calculated with a cutoff of 0.9 nm, and long-range electrostatic interactions were treated using the particle mesh Ewald (PME) method (83). All bonds involving hydrogen atoms were constrained using the SHAKE algorithm (84).

Sedimentation analysis

At the end timepoint of in vitro TDP-43 aggregation prevention assays (t = 16 h), a portion of select conditions was transferred to a tube. Tubes were spun at 21,300 rcf for 10 min at RT to sediment the pellet. The supernatant was transferred to a fresh tube. The pellet was resuspended in assay buffer of equal volume. Equal volumes of supernatant from multiple conditions were then run on SDS-PAGE gels. To determine the relative amounts of protein in the supernatant, Image Studio Lite was utilized to quantify the TDP-43 band signal for supernatant samples for each condition.

In vitro TDP-43 condensate reversal assay

TDP-43 was thawed on ice and centrifuged for 10 min at 16,000 rcf at 4°C. TDP-43 and TEV protease were diluted into PS buffer (150 mM NaCl, 20 mM HEPES-NaOH pH 7.4, 1 mM DTT), then mixed and incubated at room temperature for 75 min (reaction concentrations: 4.22 μM TDP-43, 150 mM NaCl, 20 mM HEPES-NaOH pH 7.4, 1 mM DTT, 10.56 μg/mL TEV protease). Portions of solution were then transferred to wells of a UV-transparent half-area 96-well plate (Greiner) or a glass slide, and allowed to settle. After 15 additional minutes (90 min total incubation), the pre-addition sample on the slide was imaged by brightfield microscopy with 100x objective (EVOS M5000). The solution in the 96-well plate was scanned once at 350 nm in a BMG Labtech plate reader (CLARIOstar Plus), then RNA or buffer was added to the solution in the wells or tubes for final concentrations of 0 or 2 μM RNA, 4 μM TDP-43, 150 mM NaCl, 20 mM HEPES-NaOH pH 7.4, 1 mM DTT, 10 μg/mL TEV protease. Turbidity was then measured at 350 nm once per minute for 60 min at 25°C. After 1 h of incubation, samples in tubes were imaged by brightfield microscopy with 100x objective (EVOS M5000). For quantification of the turbidity data, pre-addition readings at t=0 were standardized by subtracting the turbidity value of a sample of PS buffer alone. For each condition, values were then normalized to set the value at t=0 to 100 for each condition.

In vitro TDP-43 disaggregation assay

TDP-43 was thawed on ice and centrifuged for 10 min at 21,300 rcf at 4°C. TDP-43 was buffer exchanged into 166.66 mM NaCl, 22.22 mM HEPES-NaOH pH 7.0, 1.11 mM DTT (BIO-RAD Micro Bio-Spin Chromatography Columns, following manufacturer’s instructions) and concentration was determined via NanoDrop, e₂₈₀ = 114250 cm⁻¹M⁻¹. TDP-43 was diluted into buffer and RNase-free water to achieve a final concentration of 4 μM TDP-43, 150 mM NaCl, 20 mM HEPES-NaOH pH 7.0, 1mM DTT. TEV protease was added at a final concentration of 7.5 μg/mL. A Safire2 Tecan plate reader was used to assess turbidity once per minute at 395 nm in a nonbinding 96 well plate (Greiner) over 4 h at approximately 25–30°C. After 4 h, turbidity readings were paused. RNA (or water for controls without RNA) was added to samples, resulting in final concentrations of 40 μM RNA (for samples with RNA), 3.648 μM TDP-43, 136.8 mM NaCl, 18.24 mM HEPES-NaOH pH 7.0, 0.912 mM DTT. Turbidity readings in the Tecan plate reader were resumed for an additional 16 h after addition of RNA or water. Sedimentation was performed at the end timepoint of the assay, as described above. At the end timepoint of the assay, samples were also prepared for electron microscopy.

Transmission electron microscopy

300-mesh carbon-coated copper grids (Electron Microscopy Sciences) were glow-discharged. 5 μL of sample from the end timepoint of disaggregation assays was added to the grid and incubated for 40 s. Grids were blotted dry with filter paper. 5 μL of 1% uranyl acetate was added to the grid, then immediately blotted dry with filter paper. Grids were then stored at room temperature until imaging. Samples were viewed and imaged using a JEOL JEM-1011 electron microscope. Quantification of electron micrographs was performed with ImageJ. The image scale in pixels per μm was set based on the scale bar. Images were inverted to have dark backgrounds, then thresholded to determine regions of interest (ROIs) corresponding to aggregates. Particle analysis was constrained to particles ≥ 2 pixels. ROIs definitively corresponding to the scale bar and any broken remnants of grid were manually excluded from the quantification calculations. Quantification was reported as the values determined for each micrograph, with 4–6 micrographs of the same magnification quantified per condition. Parameters measured per micrograph were the average size of aggregates in μm², the percentage of micrograph area occupied by aggregates, and the average integrated density of aggregates.

NMR data collection and processing

All NMR experiments were heteronuclear single quantum coherence (HSQC) spectra conducted on Bruker Avance 600 MHz ¹H Larmor frequency spectrometers with HCN TCI z-gradient cryoprobe at 298K. TDP-43 RRMs NMR samples contained 50 μM protein in 20 mM KPi pH 6.8, 1 mM DTT, 5% D₂O (v/v). NMR samples of RRMs with RNA included 100 μM (2x molar equivalent) RNA. Backbone chemical shift assignments were transferred from BMRB deposited data (BMRB ID 27613). NMR data were processed and analyzed with Bruker TopSpin, NMRPipe (85), and CCPNMR (86). Chemical shift perturbations were quantified by comparison of the ¹H–¹⁵N cross-peak measurements in the RNA-containing and RNA-free HSQCs. Intensity ratios were calculated from the intensity of the ¹H–¹⁵N cross-peaks with the formula I/I_o where I is the RNA-containing sample and I_o is the RNA-free control sample.

G-quadruplex RNA annealing

RNA was thawed on ice. RNA was diluted to achieve working concentrations of 20 μM RNA, 150 mM NaCl, 20 mM HEPES-NaOH pH 7.0, 1 mM DTT. RNA was then annealed in a PCR machine by heating at 95°C for 2 min, followed by decreasing temperature at a rate of 1°C per minute, until reaching RT. RNA was added to the desired assay within a maximum of 30 min after the end of the annealing process.

Circular Dichroism

RNA was first prepared as described in the above “G-quadruplex RNA annealing” methods section. RNA was then diluted to achieve a final concentration of 5 μM RNA, 150 mM NaCl, 20 mM HEPES-NaOH pH 7.0, 1 mM DTT. The absorbance spectra were recorded in a 1 mm pathlength cuvette at 25°C with an Aviv Circular Dichroism Spectrometer, Model 202. Parameters for measuring the spectra were a measurement range of 220–320 nm, a bandwidth of 2 nm, a wavelength step of 2 nm, and an averaging time of 60 s. Data was standardized by subtracting the absorbance spectrum of the blank. The standardized data was then normalized utilizing the equation: Δε(M⁻¹cm⁻¹) = θ/(32980*c*l), where θ is the reported CD signal in millidegrees, c is the RNA molar concentration, and l is the pathlength in cm.

HEK293 cell oligonucleotide treatment

Glass bottom 24-well plates were coated with 50 μg/mL collagen overnight. OptoTDP-43 stable HEK293 cells were plated at either 150,000 cells/well for imaging, or 1,000,000 cells/well in 6-well plates for western blot analysis, in DMEM (Fisher Scientific) with 10% BGS (Hyclone; Fisher Scientific) and 1% Glutamax (Thermo Fisher). 16 hours after plating, optoTDP-43 expression was induced by media change to phenol-free DMEM/10%BGS/1%Glutamax with 750 ng/mL (for imaging) or 1000 ng/mL (for western blot) doxycycline-hyclate. Immediately after the media change, oligo treatments were started. 2’OMe_RNA oligos were transfected using lipofectamine RNAiMAX according to manufacturer’s instructions (Invitrogen). Briefly, 500 nM of each oligo was diluted in OptiMEM (Thermo Fisher) and mixed with 1 μL lipofectamine per well, incubated at room temperature for 10 minutes, and added dropwise to the cells. Plates were loosely wrapped in aluminum foil to prevent light exposure and subsequent light-induced TDP-43 oligomerization. 43 hours after doxycycline-hyclate induction, plates were removed from the foil and placed on an LED array (Amuza) for blue light stimulation (465 nm) for 5 hours at 37°C. After blue light stimulation, cells were pelleted for downstream protein analysis, or for imaging cells were washed once with PBS and fixed with 4% PFA (in PBS) for 20 minutes at room temperature. Cells were permeabilized in 0.3% Triton X-100 in PBS and stained with Hoechst (1:1000; Thermo Fisher) overnight.

HEK293 cell imaging and analysis

Image acquisition was performed using a Nikon Eclipse Ti2 Inverted Microscope with a 40X air objective. 20 fields of view (FOVs) were randomly selected by the NIS-Elements software per well. All image visualization and quantification were performed using NIS-Elements AR1 Analysis 4.51. The microscopy images were collected across three independent experiments and maximum intensity projection images were used for analysis. Binary thresholds (594 nm and 405 nm channel) and spot detection were used to capture and separate nuclei and puncta objects. Puncta overlapping with the nuclear signal were removed from the analysis, leaving only cytoplasmic puncta for quantification. Puncta area (μm²) was divided by nuclei count and expressed as puncta area per cell. Puncta area/cell was normalized against the average value for the control (CTR) oligo. Out-of-focus images were removed, and FOVs with mean puncta area >100 μm²/cell were excluded and considered outliers. Twenty FOVs were analyzed per well, and at least two to three wells were imaged per experiment. Mean values per experiment were normalized to control oligonucleotide treatment and considered a biological replicate, and the mean values of three biological replicates (n=3 experiments) were used to analyze the effect of the oligonucleotide.

Stable HEK293 CUTS cell line

Stable HEK293 cells expressing CUTS were generated as previously described (47). Briefly, HEK293 cells were seeded in 6-well plates and transfected at roughly 70% confluency with 2.5 μg of PiggyBac plasmids encoding CUTS along with 0.5 μg of a Super PiggyBac Transposase Expressing plasmid (PB200PA-1), using Lipofectamine 3000 (Invitrogen) as per the manufacturer’s instructions. A control group lacking the transposase plasmid was included. After 48 hours, cells were subjected to selection with 5 μg/mL puromycin (Sigma, P8833), with media being refreshed every two days. Non-transfected control cells typically died within five days under selection. Surviving cells were expanded and cultured in media containing a reduced puromycin concentration (2.5 μg/mL) to establish stable cell lines. Successful transgene expression was validated through live imaging.

Live Confocal Microscopy of HEK CUTS cell line

Live-cell imaging was carried out using a Nikon A1 laser-scanning confocal microscope equipped with a 10X objective lens. Environmental conditions during imaging were maintained using a Tokai HIT stage-top incubator. Images were acquired and analyzed using Nikon Elements software. Representative images were selected from a minimum of three independent experiments to ensure reproducibility.

siRNA Reverse Transfection and RNA oligonucleotide transfection of HEK CUTS cell line

siRNA-mediated gene knockdown was performed via reverse transfection using Lipofectamine RNAiMAX reagent (Invitrogen, 13778150), following the manufacturer’s instructions. To reduce TDP-43 expression, ON-TARGETplus SMARTpool siRNA targeting TARDBP (Dharmacon, L-012394-00-0005) was employed. Non-targeting siRNA (Dharmacon, D-001206-13-05) served as a control in these experiments. RNA oligonucleotide transfections were performed using Lipofectamine RNAiMAX reagent (Invitrogen) in accordance with the manufacturer’s protocol.

Detergent Solubility Assay

The detergent solubility assay was performed as previously described (14). In brief, cells were collected in RIPA buffer, incubated on ice for 10 minutes, and sonicated. Samples were centrifuged at 100,000 rcf for 1 hour at 4°C. The supernatant was collected and labeled as the detergent-soluble fraction. Protein concentrations were determined using Pierce BCA protein assay (Thermo Fisher). The remaining pellet was resuspended in RIPA buffer, briefly sonicated, and centrifuged at 100,000g for 30 minutes at 4°C. Supernatant was removed, and the remaining cell pellet was resuspended in urea buffer, sonicated, and centrifuged at 100,000 rcf for 1 hour at room temperature. The final supernatant was collected as the detergent-insoluble, urea-soluble fraction. Protein from each fraction was separated using SDS-PAGE and analyzed by western blot analysis.

SDS-PAGE/Western Blotting

Protein samples were prepared in 4x Laemmli buffer (BIO-RAD) with β-mercaptoethanol and boiled at 95°C for 10 minutes. Precision Plus Protein Western C ladder (BIO-RAD) and samples were separated via SDS-page (4–20% Mini-PROTEAN TGX precast gels, BIO-RAD) and transferred to nitrocellulose membranes (BIO-RAD) at 10 V for 90 minutes in mini-gel tanks (Invitrogen). Membranes were then incubated in Ponceau for 10 minutes and imaged. Membranes were washed and blocked for 1 hour at room temperature in 5% milk in TBST, then incubated with primary antibody overnight at 4°C. Following TBST washes, membranes were incubated at room temperature for 1 hour with secondary antibody and streptactin HRP-conjugate (BIO-RAD 1:10000). All western blot images were taken on the GE Amersham ImageQuant 800. Membranes were stripped for 10 minutes (Restore PLUS western blot stripping buffer; Thermo Fisher) and reblotted as needed.

Primary antibodies included: mCherry 1C51 (mouse; Novus Biologicals, Cat: NBP1–96752; 1:1000; RRID: AB_11034849); TDP-43 (rabbit; Proteintech, Cat: 10782–2-AP; 1:2500; RRID: AB_615042); GAPDH (mouse; Proteintech, Cat: 60004–1-IG; 1:10000; RRID: AB_2107436). Secondary antibodies included: Donkey Anti-Mouse IgG (H+L)-HRP Conjugate (Invitrogen, Cat: SA1100; 1:10000); Goat Anti-Rabbit IgG (H+L)-HRP Conjugate (Jackson Immuno Research, Cat: 111035046; 1:10000).

iPSC-derived neuron treatment and immunostaining

RNA treatments started on day 13 after plating (DIV27) and lasted 24 h. RNAs were transfected using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer’s instructions. Briefly, each RNA was diluted in OptiMEM (Gibco) and combined with 1 μL Lipofectamine per well, also diluted in OptiMEM. The mixture was incubated at RT for 10 min, and then added dropwise to the cells with each RNA at a final concentration of 500 nM. Neurons were fixed 24 h after RNA treatment on day 14 after plating (DIV28).

On DIV28, cells were washed once in PBS (Gibco) and fixed in 4% paraformaldehyde (PFA) (Electron Microscopy Sciences) immediately after treatments ended. Cells were kept in PFA for 20 min, then washed three times in PBS and blocked with 5% Donkey Serum (Jackson ImmunoResearch) + 0.3% TX-100 (Sigma-Aldrich) in PBS for 30 min at RT. Primary antibodies (goat MAP2 1:1000, Phosphosolutions 1099, RRID: AB_2752241; rabbit TDP-43 1:300, Proteintech 10782–2-AP, RRID: AB_615042) were diluted in blocking solution and incubated overnight at 4°C. Secondary antibodies (donkey Alexa Fluor, Jackson ImmunoResearch) were used at 1:1000 dilution in blocking solution and incubated for 60 min at RT. All treatments and cell lines were treated and probed simultaneously to decrease variability. Coverslips were mounted on slides using Prolong Glass mounting media (Invitrogen).

Images were acquired (20 per group) using an A1R Nikon Confocal Microscope and fields of view (FOV) were processed for analyses using Nikon NIS Elements Software. Settings were kept consistent across treatments. Within each FOV, neurons that were isolated (not over glial cells or other neurons) were selected and nuclear TDP-43 signal was measured by overlaying an ROI using DAPI as a guide. Cytosolic area was hand-drawn using MAP2 signal as a guide. Raw intensity values for nuclear signal in the 488 channel (TDP-43) were normalized against cytosolic intensity values and the output was referred to as “nuclear/cytosolic ratio.”

For analysis of neuritic puncta, a Cy5-labeled Malat1_start and CTR oligo were used at the same concentration and timing as described above. We used Staufen-1, another RBP, as a control to assess possible interactions between TDP-43 and Malat1_start. Staufen-1 antibody was from Proteintech (14225-1-AP; RRID: AB_2302744) and was used at a 1:300 dilution. After imaging, Malat1_start and CTR oligo puncta were detected using ROI autodetection on NIS-Elements. Pearson coefficient between Malat1_start/CTR oligo and either TDP-43 or Staufen-1 was calculated for each ROI, and we considered Pearson ≥ 0.5 as colocalizing. The number of total Malat1_start, CTR oligo, TDP-43, and Staufen-1 puncta were expressed as puncta/100 μm, same as the number of puncta within each group with Pearson ≥ 0.5. Values were graphed as the average of 20–25 neurites in 4 technical replicates of the same lines.

Sodium arsenite treatment and immunofluorescence

Control iPSC-derived MNs were plated on coverslips at a density of 150,000 cells/well. On DIV27 cells received either no treatment (NT), CTR RNA, or Malat1_start RNA. RNAs were delivered at a concentration of 500 nM with RNAiMAX Lipofectamine. On DIV28, 22 h after the RNA treatments, neurons were exposed to 250 μM sodium arsenite for 2 h. Cells were then fixed in 4% PFA, washed 3 times in PBS, blocked in 0.3% TX-100 + 5% Normal Donkey Serum in PBS, and incubated overnight at 4°C in primary antibodies diluted in blocking solution at the following concentrations: TDP-43 (Proteintech 10782–2-AP; RRID: AB_615042) at 1:300; G3BP1 (Santa Cruz sc-365338; RRID: AB_10846950) at 1:100; MAP2 (PhosphoSolutions 1099-MAP2; RRID: AB_2752241) at 1:1000. Coverslips were then washed in PBS and incubated in secondary antibodies (Jackson ImmunoResearch) at a concentration of 1:1000 in blocking solution for 90 min at RT, then washed in PBS 3 times and mounted on glass slides with Prolong Glass with NucBlue mounting media (Invitrogen P36983). Images were acquired on a confocal Nikon A1R confocal microscope.

RT-PCR

Total RNA from differentiated iPSC-derived motor neurons was extracted using the RNeasy Mini Kit (Qiagen, 74106) according to the manufacturer’s instructions. RNA concentration and purity were assessed with a NanoDrop ND-1000 spectrophotometer (Thermo Fisher). Complementary DNA (cDNA) was synthesized from total RNA using the iScript Reverse Transcription Supermix (BIO-RAD, 1708841). RT-PCR was conducted on 5 ng of RNA per reaction. Primers utilized to detect KCNQ2 cryptic splicing: KCNQ2-CE-F, 5′- TATGCCCACAGCAAGATCAC-3′; KCNQ2-CE-R, 5′-AGACACCGATGAGGGTGAAG-3′. As a loading control, 18S was amplified using the following primers: 18S-FWD 5’ GCAGAATCCACGCCAGTACA and 18S-REV 5’ TTCACGGAGCTTGTTGTCCA. For STMN2 we utilized previously described primers (87); we ran the full length transcript using the following primers: STMN2-F,5′- AGCTGTCCATGCTGTCACTG-3′;STMN2-R, 5′- GGTGGCTTCAAGATCAGCTC −3′ and its truncated version:STMN2a-F, 5′- GGACTCGGCAGAAGACCTTC −3′; STMN2a-R, 5′- GCAGGCTGTCTGTCTCTCTC-3′, and values were normalized against β-actin ACTB-F, 5’-TTGTTACAGGAAGTCCCTTGCC-3’; ACTB-R, 5’-ATGCTATCACCTCCCCTGTGTG-3’. For RT-PCR, PCR products were amplified by Quick-Load Taq 2X Master Mix (NEB, M0271L) using the following PCR program in S1000 Thermal Cycler (BIO-RAD): 95°C for 30 s, followed by 40 cycles of 95°C for 15 s, 55°C for 30 s, 68 °C for 30 s and extension time of 68 °C for 5 min. PCR products were separated by agarose gel electrophoresis, and the bands were visualized with Amersham ImageQuant 800 GxP biomolecular imager system. Band intensity was quantified utilizing ImageLab from BIO-RAD.

Virus Production

AAV9 virus was generated by Vector Biolabs using a plasmid designed as follows. The CMV-promoter driven pcDNA3.2 TDP-43 NLS1 YFP plasmid, a gift from Aaron Gitler (Addgene plasmid #84912; RRID:Addgene_84912) (58), was packaged into AAV9 viral particles.

Intraspinal delivery of AAV9 Virus

Intraspinal delivery of AAV9 in p180 mice was carried out as previously described (57). Mice deeply under anesthesia underwent an incision of their dorsal skin and underlying muscle with retraction, revealing the spinous processes between vertebrae C2 and T1. Following laminectomy at spinal levels C4, C5, and C6, six total bilateral injections were given across this area. Each injection contained 1×10¹¹ GC of the AAV9-TDP-43 NLS1 virus (TDP-43^ΔNLS) in a 1 μL total volume. A gas-tight Hamilton syringe mounted on a UMP3 electronic micropump (World Precision International) was used for these injections, with a 33-gauge 45° beveled needle. Targeting of injections was guided on the lateral axis by the midpoint of each spinal segment, and on the rostral-caudal axis by the location of dorsal root entry for C4, C5, and C6. The needle was lowered to a depth 0.8 mm below the dorsal surface for ventral horn targeting, with injections then delivered over a 5-minute interval at a constant rate. Sham surgery control animals underwent identical procedures and laminectomies, as well as needle placement and insertion. In sham animals, the Hamilton syringe was filled with sterile PBS and the micropump was not initiated. Following the final injection, the dura was removed from the dorsal spinal cord of the injection region, and a non-adhering dressing (Adaptic non-adhering dressing by Systagenix) was applied. Overlying muscles were then closed in layers, using sterile silk sutures. The skin incision was also closed using both sutures and sterile wound clips. Animals recovered on a heating pad until awake, and were then returned to their home cage. To minimize pain and distress, at the time of surgery and at 12-hour intervals for the first 24 hours following surgery, animals were given subcutaneous sterile saline for fluid balance, buprenorphine analgesic (0.05 mg/kg), and cefazolin antibiotic (10 mg/kg). Animals were monitored daily and were checked for signs of pain and/or distress, as well as ambulatory potential and ability to obtain food/water.

Spinal delivery of RNA

After one week of viral expression, animals were again deeply anesthetized. The original surgical incision was reopened and skin and muscle retracted to expose the spinal cord. The non-adhering dressing was removed from the spinal cord surface, and was replaced with a pre-saturated gelfoam sponge (sterile gelfoam Dental sponge, Pharmacia & Upjohn). TDP-43^ΔNLS-expressing animals were randomized into saline control, or RNA treatment groups, with the sponge being pre-soaked either in sterile saline solution, or sterile saline reconstituted RNA at a 100 μg/mouse dosing. This dosing penetrates to the ventral spinal cord by 3-days post-application in a robust and reproducible manner, and causes no detrimental side-effects when evaluated out to two-weeks post-administration. Sham surgery control animals underwent identical procedures, and received a saline-saturated gelfoam sponge. Following this application, overlying muscles and skin were again closed with sterile silk sutures, with the skin also being bound with sterile wound clips. Animals again recovered on a heating pad until awake, prior to return to home cages. Animals received the same set of compounds to minimize pain and distress, at the time of surgery and at 12-hour intervals for the first 24 hours following surgery: subcutaneous sterile saline for fluid balance, buprenorphine analgesic (0.05 mg/kg), and cefazolin antibiotic (10 mg/kg). Animals were again monitored daily for signs of pain and/or distress, as well as ambulatory potential and ability to obtain food/water. Assessment of motor neuron numbers was carried out at 3 and 5 days following RNA application, to determine potential beneficial therapeutic effects and longevity.

Animal Harvesting for Spinal Cord Immunofluorescence

Mice were euthanized using carbon dioxide asphyxiation, and perfused and fixed following standard laboratory procedures. A perfusion needle connected to a peristaltic pump was inserted into the left ventricle of the heart, and animals were then perfused with approximately 20 mL of PBS followed by 25 mL 4% paraformaldehyde. The animal was dissected to obtain the cervical spinal cord, which was then placed in 4% paraformaldehyde overnight. Paraformaldehyde was briefly rinsed off the tissue with PBS, with spinal cords then placed in 30% sucrose until tissue sinking (24–48 hours). Spinal cords were frozen into Tissue-Tek OCT solution and were subsequently sectioned using a Cryostar NX50 cryostat (Epredia) at a section depth of 30 μm. Sections were placed onto charged glass slides. Tissue blocking, permeabilization, and staining were performed according to laboratory standard protocols and according to antibody manufacturer recommendations. 30 μm spinal cord sections on slides were heated overnight at 55°C and were then rinsed with PBS. Sections were next blocked in 5% BSA for 1 hour at room temperature, and then incubated in primary antibodies at 4°C overnight (NeuN) or for 48 hours (ChAT). Primary antibodies included: anti-ChAT (Millipore RRID:AB_2079751, 1:1,000) and anti-NeuN (Cell Signaling Cat# 24307, RRID:AB_2651140, 1:400). Following this incubation and PBS washing, secondary labeling for visualization was attained with AlexaFluor594 (Life Technologies). To label cell nuclei, Hoechst stain (ThermoFisher) was used. Slides were mounted with coverslips using Citifluor AF3 (Electron Microscopy Sciences). Microscopy imaging was accomplished using a Nikon A1+ confocal microscope and NIS-Elements software. ChAT⁺ and NeuN⁺ cells were assessed using bilateral ventral horn images with manual counting by a blinded assessor. TDP-43 puncta, visualized using the YFP tag of the virally expressed protein, and colocalization of these puncta with Cy5-tagged Malat1_start molecules were assessed using NIS-Elements software.

RT-qPCR analysis of Sort1 transcripts

RNA was extracted from fixed mouse spinal cords using the Thermofisher PureLink RNA Mini Kit. Reverse transcription was accomplished using the Qiagen QuantiTech Reverse Transcription Kit. cDNA samples were then prepared for qPCR assessment using SYBR Green qPCR master mix from Thermofisher, and were evaluated using the QuantStudio 5 Real-Time PCR System. Samples were measured in triplicate for each transcript of interest, with data normalized to GAPDH transcript expression. The ratio of the misspliced variant (Sort1-ex17b) to Sort1-WT transcripts is graphically represented, as has been reported in previous studies (43, 59). Fold change between groups for total Sort1 is also represented.

The previously described and validated primer pairs for total, WT, and TDP-43 misspliced (Sort1-ex17b) mouse Sort1 transcripts (43, 59) were obtained from Integrated DNA Technologies (IDT). Primer pairs are as follows:

Sort1_Total fwd: CGTGTTCCCTGGAGGACTTCCT;

Sort1_Total rev: TTCAGGCTGCTCCACGCACT;

Sort1_WT fwd: CCCCACAAAGCAGAATTCCAAGTC;

Sort1_WT rev: TGACAAGCATCAGTCCCACGAT;

Sort1_ex17b fwd: AAATCCCAGGAGACAAATGC;

Sort1_ex17b rev: GAGCTGGATTCTGGGACAAG;

GAPDH fwd: AACAGCAACTCCCACTCTTC;

GAPDH rev: CCTGTTGCTGTAGCCGTATT

Animal Data Analysis Considerations

Animals were numbered according to surgery order and were randomized into treatment groups. Counting of motor neuron numbers was performed by a second, blinded individual. All surgical animals were considered for inclusion and assessment based on confirmation of YFP fluorescence in spinal cord motor neurons. No animals reached IACUC-based endpoint criteria for early euthanasia and study exclusion. All 10 animals per timepoint per group were considered.

Quantification and Statistical Analysis

All statistical details of experiments can be found in the figure legends. Data visualization and statistical analyses were performed with GraphPad Prism (GraphPad Software Inc.; La Jolla, CA, USA). Quantification of gel images was performed with Image Studio Lite (LI-COR Biosciences; Lincoln, NE, USA). Quantification of micrographs was performed with ImageJ (National Institutes of Health; Bethesda, Maryland, USA). HX analysis was performed with HDExaminer (Trajan Scientific and Medical; Melbourne, Australia). MATLAB was utilized for some HX data visualization (MathWorks; Natick, MA, USA). ExMS2, a MATLAB-based tool, was utilized to prepare the peptide pool for HX (88). Immunofluorescence analysis was performed utilizing NIS-Elements (Nikon; Minato City, Tokyo, Japan).

Supplementary Material

Supplementary Text

Figs. S1 to S27

Tables S1 to S4

Data and materials availability:

Plasmids generated in this study will be made readily available to the scientific community. All requests will be honored in a timely manner. Material transfers will be made with no more restrictive terms than in the Simple Letter Agreement or the Uniform Biological Materials Transfer Agreement and without reach through requirements. All data used for this study are available in the manuscript, the supplementary material or deposited at indicated data repositories. The mass spectrometry proteomics data for HXMS experiments have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (89) with the dataset identifier PXD071117. All tabulated data underlying the figures is deposited at Dryad (90). This paper does not report original code.