De novo design of protein binders to stabilize monomeric TDP-43 and inhibit its pathological aggregation

Significance

Pathological aggregation of TDP-43, predominantly driven by its low-complexity domain, is a central driver of many neurodegenerative diseases, including amyotrophic lateral sclerosis and frontotemporal lobar degeneration, yet no effective therapies exist to directly halt this process. A central challenge lies in the structural flexibility of the TDP-43 low-complexity domain, especially in its region that drives the aggregation process. Here, we bridge this gap by leveraging de novo protein design tools based on AI to stabilize the monomeric TDP-43 and maintain its natural state. This approach may present a potential therapeutic strategy for treating diseases linked to TDP-43.

Abstract

Pathological aggregation of transactive response DNA binding protein of 43 kDa (TDP-43), primarily driven by its low-complexity domain, is closely associated with various neurodegenerative diseases, including amyotrophic lateral sclerosis (ALS) and frontotemporal lobar degeneration (FTLD). Despite the therapeutic potential of preventing TDP-43 aggregation, no effective small molecule or biomacromolecule therapeutics have been successfully developed so far. Here, we introduce a protein design strategy that yields de novo designed proteins capable of stabilizing the key amyloidogenic region of TDP-43 in its native helical conformation with nanomolar binding affinity. The binding mechanism was further characterized by the NMR and mutagenesis study. More importantly, we demonstrated that our designed protein binders efficiently reduced TDP-43 amyloid aggregation both in vitro and in cells. Our work provides a strategy for designing protein stabilizer of the native conformation of pathological proteins for preventing its amyloid aggregation, shedding light on the development of potential therapeutic approaches for ALS, FTLD, and other protein aggregation-associated diseases.

TDP-43 is essential for regulating RNA metabolism, including mRNA splicing, RNA transport, translation, and microRNA synthesis (1–3). Pathological aggregation of TDP-43 is not only the hallmark of ALS and FTLD but also the key pathological entity involved in the initiation and progression of these diseases (4–7). Recent studies have demonstrated that the low-complexity domain (LCD) of TDP-43 is the principal driver of its pathogenic aggregation, leading to TDP-43 mislocalization and toxic cytoplasmic inclusions (8). In disease states, TDP-43 is depleted from the nucleus and forms hyperphosphorylated or aggregated cytoplasmic inclusions, which are observed in about 97% of ALS patients and approximately 50% of FTLD patients (9, 10). TDP-43 is a 43 kDa protein consisting of an N-terminal domain (NTD) and two tandem RNA recognition motifs, RRM1 and RRM2, followed by a C-terminal glycine-rich region where most disease-associated mutations converge (11, 12). TDP-43 binds to nucleic acids via its RRM domains (Fig. 1A) and contributes to RNA processing, including but not limited to splicing, translation, and cytoplasmic stress granule response, in protein complexes that sequester mRNAs to minimize stress-related damage (13–15).

Fig. 1.

Design strategy of TDP-43 binding protein. (A) Schematic of the main functional domains and predicted secondary structures of TDP-43. (B) Structural model of TDP-43 illustrating the transition from soluble to aggregated states. The alphafold3 predicted full-length TDP-43 monomer, showing its N-terminal domain (NTD), RNA recognition motifs (RRM1 and RRM2), and C-terminal low-complexity domain (LCD) including conserved region (CR). The CR-helix (residues 319–335) assembly facilitating packed fibril structures in the insoluble state. Top and side views of the fibril structure derived from ALS patient samples are shown (PDB code: 7py2). (C) Design pipeline for CR-helix-targeting binders to prevent TDP-43 aggregation. Fifty RFdiffusion steps with enhanced interface contact potential are applied for scaffold generation. ProteinMPNN is used for sequence optimization, three sequences for each scaffold. Binder candidates are identified using AlphaFold2 initial predictions and filtered with Rosetta based on PAE_interaction ≤ 6 and ΔΔG ≤ −60 kJ/mol.

TDP-43 has been shown to form a variety of pathological assemblies with distinct morphologies, including diffuse, granular, compact, and skein-like inclusions, depending on the cellular context and disease stage (16–19). TDP-43’s inherent tendency to misfold and aggregate has posed significant challenges to develop therapies targeting its native state. Within the LCD of TDP-43, an amyloidogenic core region (residues 318–360) has been identified as critical for aggregation. In this region, an evolutionarily conserved segment, known as “conserved region” (CR, residues 319–335), adopts an α-helical conformation under physiological conditions. Pathological aggregation is driven by structural reorganization of the CR assembly through a conformational transition from helix to β-strand-enriched assemblies (20–23). This region is sufficient to drive the formation of TDP-43 aggregates in vitro, in cells, and in vivo, underscoring its central role in regulating the phase transition of TDP-43 to amyloid-like conformations (24–26). Indeed, cryoelectron microscopy (cryo-EM) studies of TDP-43 fibrils as well as pathological TDP-43 aggregates from the cortex of ALS/FTLD patients support the importance of the CR segment in forming the fibril core structure (27–29) (Fig. 1B). The cryo-EM structures of patient-derived TDP-43 fibrils revealed that the ordered filament core, comprising stacked TDP-43 molecules, adopts a characteristic double-spiral fold formed by the amyloidogenic core spanning residues 282–360. Within this core, multiple β-strands are contributed by hydrophobic residues from the CR region. Notably, the CR undergoes distinct conformational changes in different disease subtypes. Residues 326–327 and 332–337 form β-strands in TDP-43 filaments from ALS with FTLD. Residues 321–331 and 333–336 form β-strands in FTLD-TDP type A, whereas residues 321–326, 328–334 adopt β-strands structures in FTLD-TDP type C. Moreover, even within the same pathological subtype, structurally heterogeneous fibrils have been observed, highlighting the conformational plasticity of the CR segment and its propensity to form diverse β-strand-enriched assemblies.

Therapeutic strategies to inhibit TDP-43 fibrillization remain limited, primarily due to the challenges posed by its highly unstructured C-terminal domain (CTD) in the soluble state and its tendency to form polymorphic fibrils upon aggregation. These characteristics make the rational design of effective molecules difficult (30). However, de novo protein design has emerged as a transformative approach to overcome these challenges. Recently, Sahtoe et al. developed scaffolds with deep peptide-binding clefts to stabilize disordered proteins in β-strand and β-hairpin conformations, effectively preventing aggregation (31). Meanwhile, Torres et al. utilized AI–driven tools such as RFdiffusion and ProteinMPNN to design binders that target short α-helical peptides, demonstrating the potential of these methods to engage specific structural motifs (32).

Here, we sought to design high-affinity binders capable of stabilizing the aggregation core of TDP-43 in an α-helical conformation, thereby inhibiting its pathological aggregation. Among dozens of designed candidates, one CR-helix binder demonstrated a high binding affinity to the TDP-43 LCD and effectively inhibited aggregation in both in vitro assays and cellular models. Our findings suggest that precisely designed protein binders targeting the CR helix represent a promising potential approach for mitigating TDP-43-related pathologies observed in ALS and FTLD.

Results

Design Strategy for TDP-43 Inhibitory Binders.

Previous study demonstrated that the CR containing 319-335 of TDP-43, which adopts a helical conformation (CR-helix) in its native state, undergoes conformational change to form beta-strand for mediating amyloid aggregation of TDP-43 under diseased condition (24). Therefore, pharmacologically stabilizing the CR-helix may offer a promising therapeutic strategy to prevent this structural transition and inhibit the formation of pathogenic fibrils. To design binders that stabilize the CR-helix and prevent the conversion of TDP-43 CTD into amyloid fibril, we designed a pipeline to generate scaffolds capable of tightly grasping the CR-helix with good shape complementarity to support the alpha-helical conformation of CR residues (Fig. 1C). Using RFdiffusion (33), we tested both fold conditioned and unconditioned methods to generate scaffolds and finally yield 10,369 scaffolds with suitable lengths and stable alpha-helical conformations (SI Appendix, Figs. S1 and S2). Based on the aggregation score of CR-helix, four hydrophobic residues including Met322, Met323, L330, and W334 were selected as hot spot residues during the diffusion process. Amino acid–biased optimization with ProteinPMNN (34) was then applied to the scaffolds to enhance interactions with CR residues, resulting in 31,107 binder sequences. Various Rosetta filters (35) and Alphafold2 initial guess metrices (SI Appendix, Fig. S3) were subsequently used to screen the generated sequences as described previously (36), yielding 14 final binder designs with favorable interaction energy, minimal unsatisfied buried polar atoms, and low predicted aligned error of the interface (PAE_interaction), which were selected for further experimental characterization (Fig. 1C and SI Appendix, Tables S1–S3).

Characterization of Inhibitory Activity of Designed Binders.

The selected 14 designs were encoded in synthetic genes with a C-terminal polyhistidine affinity tag, expressed in Escherichia coli, and subsequently purified. Eight of the fourteen designs (SI Appendix, Figs. S4–S6) showed discrete monomeric peaks at the expected elution volumes in size-exclusion chromatography (SEC) and can be concentrated to high levels without obvious precipitation; thus, these proteins were selected for further evaluation.

We next examined these eight designed protein binders on amyloid fibril formation of TDP-43. To this end, we tested TDP-43 LCD fibril formation using a thioflavin T (ThT) assay. Among the candidates, two binders including B1 and B9 demonstrated potent inhibitory activity. The ThT assay results revealed that adding B9 at 0.5 molar ratio significantly inhibited TDP-43 fibrillation, as evidenced by an extended lag phase (Fig. 2A) and a 60% reduction in ThT fluorescence intensity (Fig. 2B). B1 also exhibited a notable inhibitory effect at 0.5 molar ratio, reducing the ThT signal by approximately 40% (Fig. 2 A and B). Consistently, negative-staining transmission electron microscope (NS-TEM) imaging confirmed that TDP-43 LCD fibril formation was significantly reduced in the presence of B9 and B1 (Fig. 2C). To rule out the possibility of a monomer-oligomer equilibrium for the binders, we analyzed B1 and B9 using SEC. The results showed that both binders were monodispersive, as indicated by their sharp and single peaks in the SEC profiles (SI Appendix, Fig. S7). Further, circular dichroism (CD) measurements demonstrated that both designs adopted α-helical structures and possessed high thermal stability, with similar CD spectra observed at 99 °C compared to those recorded at 26 °C (SI Appendix, Fig. S8). Subsequently, a concentration gradient ThT assay was performed for both binders (Fig. 2D and SI Appendix, Fig. S9). The results showed that fibril formation was significantly inhibited by both B1 and B9 in a concentration-dependent manner. Overall, these results suggest that B1 and B9 strongly inhibited the amyloid aggregation of TDP-43 LCD in a dose-dependent manner. To further validate these effects on the full-length protein, we incubated full-length TDP-43 with or without binders and assessed its aggregation by NS-TEM, as its fibrils do not bind ThT (18). Consistent with the LCD results, B9 strongly inhibited fibril formation, and B1 showed a moderate effect (SI Appendix, Fig. S10). These results indicate that B1 and B9 are effective against both the LCD and full-length TDP-43.

Fig. 2.

Characterization of designed TDP-43 binders. (A) ThT screening assay for identical binder of TDP-43. 20 μM TDP-43 LCD monomers fibrillation in the presence of 10 μM different binder. Data correspond to the mean ± SEM, n = 3. (B) Quantification of ThT signal after 8 h from the commencement of the aggregation shown in A. Data are mean ± SEM (n = 3). Two-tailed unpaired t test. Significance: P < 0.05 (*), P < 0.01 (**), P < 0.001 (***), P < 0.0001 (****). (C) TEM images of the ThT samples in the presence and absence of binders at 8 h. (Scale bar, 200 nm.) The imaging was independently repeated three times with similar observations. (D) ThT fluorescence assay of 20 μM TDP-43 LCD monomers fibrillation in the presence of different concentrations of B9. Data correspond to the mean ± SEM, n = 3. Quantification of the ThT signal after 8 h from the commencement of the aggregation shown on the Right. Two-tailed unpaired t test. Significance: P < 0.05 (*), P < 0.01 (**), P < 0.001 (***), P < 0.0001 (****). (E) The binding kinetics of B9 with TDP-43 LCD measured by BLI assay.

To investigate the binding properties of B1 and B9 to the TDP-43 LCD monomer, Bio-Layer Interferometry (BLI) assays were used to measure binding affinities between binders and the TDP-43 LCD monomer. The binders directly interact with TDP-43 LCD and affinity of B9 to the TDP-43 LCD monomer was measured as 87.9 ± 2.7 nM (Fig. 2E), while B1 exhibited a lower affinity at 25 ± 2.1 μM (SI Appendix, Fig. S9). Lower binding affinity of B1 likely explains its comparatively weaker inhibitory effect relative to B9. These findings are consistent with he inhibitory activities observed in the ThT assay and NS-TEM imaging.

Structural Characterization of the Interaction Between TDP-43 and Binders.

We predicted the complex structures of designed B9 or B1 proteins in complex with Full-length TDP-43 using Alphafold3. The model predictions revealed very high iPTM scores, with 0.94 for B9 and 0.85 for B1 (Fig. 3A and SI Appendix, Fig. S11). In both B9 and B1, three helices grasp the CR-helix, exhibiting strong contacts driven by major hydrophobic interactions, as intended. Notably, the designed model aligns closely with the Alphafold3 predictions, showing a Cα RMSD of 0.71 Å for B9 and 0.92 Å for B1 (Fig. 3A and SI Appendix, Fig. S11). Furthermore, the key binding residues at the interface displayed nearly identical side-chain conformations in both the predicted structure and the design model.

Fig. 3.

Structural characterization of designed TDP-43 binders. (A) Comparison of the designed structure and AlphaFold3 prediction for the TDP-43 CR-helix and binder B9 complex. The TDP-43 CR-helix is shown in pink (design) and light blue (AlphaFold3 prediction), while binder B9 is depicted in purple (design) and teal (AlphaFold3 prediction). The structural similarity between the designed model and AlphaFold3 prediction is reflected by complex RMSD of 0.71 Å and an iPTM of 0.94. (B) Overlay of the 2D 1H-15 N HSQC spectra of 20 μM 15 N-TDP-43 LCD in the presence of B9. The peaks of target region are labeled and enlarged in (C). (D) Intensity changes (Right) of TDP-43 LCD titrated by B9 from (B), the target region is colored in peach puff.

To experimentally confirm the structural mechanism underlying the binder–TDP-43 interaction, we performed NMR spectroscopy by titrating B9 and B1 into ¹⁵N-labeled TDP-43 LCD, respectively. The 2D 1H–15 N HSQC spectra demonstrated a significant dose-dependent signal attenuation upon titration with B9 and B1 (Fig. 3 B–D and SI Appendix, Fig. S12). Notably, the residues showing substantial changes were clustered in the CR-helix. Adding 0.6 equivalents of B9 suppressed 60% of the signal, while 0.4 equivalents resulted in 40% suppression. This pattern indicates a consistent 1:1 binding interaction, as designed. Additionally, titration of TDP-43 LCD monomers with B1 revealed similar interactions between TDP-43 and B1 as observed with B9 (SI Appendix, Fig. S12).

Structural Model of the TDP-43–B9 Complex Validated by Mutagenesis Study.

We next focused on B9, which demonstrated the highest binding affinity and inhibitory activity to TDP-43 among all the designed binders. It appears that the CR-helix of TDP-43 forms a strong hydrophobic interaction with B9 (Fig. 4A). To validate the complex structure of B9–TDP-43 and explore the underlying binding mechanism, we examined two hydrophobic-to-polar mutations at the B9 interface (L15E, L112E), alongside two similar mutations in the TDP-43 LCD (L330R, M322D), in addition to a previously reported mutation (A326P) of TDP-43, which was known to disrupt the α-helical conformation of the CR(21).

Fig. 4.

Structural mutagenesis abolished both the specific interaction and the inhibitory effect. (A) Structural insights into the interaction between TDP-43 CR-helix and the designed binder B9. The CR-helix of TDP-43 (megenta) forms strong hydrophobic interactions with B9 (cyan), as shown in the interface. (B) The binding kinetics of B9 with TDP-43 LCD WT and mutants (A326P, M322D, and L330R) measured by BLI assay. (C) The binding kinetics of TDP-43 LCD with B9 WT and mutants (L112E and L15E) measured by BLI assay. (D–F) ThT fluorescence assay of 20 μM TDP-43 LCD mutant (A326P, L330R, and M322D) monomers fibrillation in the presence and absence of B9. Data correspond to the mean ± SEM, n = 3. TEM images of the ThT samples were shown on the Bottom. (Scale bar, 500 nm.) (G) ThT fluorescence assay of 20 μM TDP-43 LCD monomers fibrillation in the presence and absence of B9 mutant (L15E, L112E) and WT. Data correspond to the mean ± SEM, n = 3. TEM images of the ThT samples were shown on the Bottom. (Scale bar, 500 nm.)

Using BLI, we first assessed interactions between B9 and TDP-43 mutants. TDP-43 LCD mutants were biotinylated and immobilized on the sensor surface, while 1 µM B9 protein was introduced in the mobile phase. The results showed that none of the three TDP-43 mutants were able to bind B9 (Fig. 4B). Consistently, ThT assay and NS-TEM confirmed that B9 could not inhibit the aggregation of mutated TDP-43 LCD (Fig. 4 D–F).

We similarly tested the binding of B9 mutants to TDP-43 LCD. Using 1 µM TDP-43 LCD as the immobilized phase and introducing 1 µM B9 mutants as the mobile phase, we observed a significant reduction in binding affinity between B9 mutants and TDP-43 (Fig. 4C). Additionally, ThT assay and NS-TEM confirmed that B9 mutants were unable to prevent the aggregation of TDP-43 LCD (Fig. 4G). Collectively, these mutations of the key residues in either B9 or TDP-43 abolish binding and inhibitory activity, confirming that the binder B9 interacts with TDP-43 CR-helix in a mode as designed.

Designed Binders Are Capable of Preventing TDP-43 Aggregation in Cells.

Given the potent activity of B9 in preventing TDP-43 fibrillation in vitro, we next examined whether it could prevent pathological aggregation of TDP-43 in cells. We used two previously well-documented TDP-43 aggregation cell models including a nucleus TDP-43 aggregation model and a cytoplasm TDP-43 aggregation model (37). To confirm the nature of the aggregates, we performed pFTAA staining and observed colocalization with TDP-43 aggregates, indicating the presence of β-sheet–rich structures (SI Appendix, Fig. S13). As for the nucleus model, we cotransfected the TDP-43 K181E and binder plasmid into cells. Remarkably, cotransfection with B9 significantly reduced the proportion of HEK-293 T cells exhibiting TDP-43 nuclear aggregation from 64% to 16%, indicating a strong inhibitory effect on TDP-43 aggregation (Fig. 5 A and B). Cotransfection with B1 resulted in a more modest reduction, decreasing aggregation from 64 to 48%, aligning with the in vitro data (SI Appendix, Figs. S9 and S14). Since pTDP-43 at S409/410 is a disease-specific marker of TDP-43 aggregates, we performed a WB analysis to examine the expression and pTDP-43 (S409/410) levels of TDP-43 K181E in the presence of binders. Intriguingly, while the expression level of the TDP-43 K181E plasmid was unaffected by cotransfected binders (Fig. 5C and SI Appendix, Fig. S14), a significant decrease in pTDP-43 (S409/410) was observed in the presence of B9 (Fig. 5C), with a noticeable but less pronounced reduction in the presence of B1 (SI Appendix, Fig. S14). These findings suggest a modulatory role for the binders in regulating pTDP-43 (S409/410).

Fig. 5.

Designed binders inhibit the aggregation of TDP-43 and reduce the pTDP-43 (S409/410) levels in the nucleus and cytoplasm. (A) Representative images of HEK293 T cells expressing TDP-43 K181E EGFP and B9 Myc. The imaging was independently repeated three times with similar observations. (B) Quantitative analysis of the number of cells with aggregates for images (A). Data correspond to the mean ± SD, n ~ 200 cells. Two-tailed unpaired t test. Significance: P < 0.05 (*), P < 0.01 (**), P < 0.001 (***), P < 0.0001 (****). (C) Western blot for the expression of EGFP, pTDP-43 (S409/410) and Myc and in the transfected 293 T cells. GAPDH serves as a loading control. The imaging was independently repeated three times with similar observations. (D) Representative images of HEK293T cells expressing TDP-43 CTF EGFP and B9 Myc. The imaging was independently repeated three times with similar observations. (E) Quantitative analysis of the number of cells with aggregates for images (D). Data correspond to the mean ± SD, n ~ 200 cells. Two-tailed unpaired t test. Significance: P < 0.05 (*), P < 0.01 (**), P < 0.001 (***), P < 0.0001 (****). (F) Western blot for the expression of EGFP, pTDP-43 (S409/410), and Myc and in the transfected 293T cells. GAPDH serves as a loading control. The imaging was independently repeated three times with similar observations.

In the cytoplasmic TDP-43 aggregation model, cotransfection of cells with the TDP-43 C-terminal fragment (CTF) and binder plasmids showed that B9 significantly reduced the proportion of cells with cytoplasmic aggregates from 71 to 21%, while increasing the proportion of cells displaying cytoplasmic dispersion (Fig. 5 D and E). B1 exhibited a weaker yet still noticeable effect, reducing cytoplasmic aggregates from 71 to 54%. WB analysis confirmed that neither B9 nor B1 impacted the expression level of the TDP-43 CTF plasmid. However, B9 resulted in a more substantial decrease in the pTDP-43 (S409/410) level compared to the modest reduction observed with B1 (Fig. 5F and SI Appendix, Fig. S14). In summary, these findings underscore the superior capability of B9 to efficiently inhibit fibrillation and diminish disease-associated pTDP-43 (S409/410) in both the nucleus and cytoplasm.

Discussion

Recent studies indicate that the TDP-43 LCD is the main driver of its pathogenic aggregation, which leads to TDP-43 mislocalization and aggregation in the cytoplasm (8), resulting in toxic gains of function that drive neurodegeneration in ALS/FTLD (4, 9, 38). By contrast, the structured NTD mediates physiological homodimerization essential for efficient RNA binding and splicing, and disruption of this NTD-mediated dimerization results in nuclear depletion and cytoplasmic aggregation (39). Importantly, experiments have shown that disrupting LCD-driven phase separation (for example by mutating key hydrophobic motifs) abrogates aberrant droplet formation without impairing TDP-43’s splicing activity (40). Interestingly, a recent work (41) demonstrated that the CR region is crucial for pathological aggregation of TDP-43 demixed from dynamic stress granule, but not for recruitment of TDP-43 into functional stress granule. This observation suggests that targeting CR may have a beneficial effect without alternating its function involved in stress granule. Therefore, our design, specifically targeting the CR region within the LCD, enables selective inhibition of pathological self-association while preserving the physiological NTD-mediated homodimerization that is essential for RNA binding and splicing (42).

While several heat-shock proteins (HSPs), such as HSP70 and HSPB1, have been shown to inhibit TDP-43 amyloid aggregation (14, 43, 44), their clinical applicability remains limited due to their broad substrate specificity and pivotal roles in cellular protein quality control systems. Additionally, earlier strategies such as antibody-mediated TDP-43 polyubiquitination and degradation (45), peptide-based aggregation inhibition targeting the CTD (46), and small-molecule interventions modulating TDP-43 RNA-binding and aggregation (47), did not directly target the TDP-43 LCD aggregation core region due to the structural complexity of TDP-43 LCD.

Despite decades of unsuccessful clinical trials and the absence of effective biologic therapies for neurodegenerative diseases, recent approvals of monoclonal antibodies (Aducanumab, Lecanemab, and Donanemab) targeting amyloid protein aggregates represent significant progress. These developments provide considerable motivation for further advancing biologic therapeutic strategies, particularly monoclonal antibodies, for neurodegenerative disorders.

By employing advanced AI-based protein de novo design techniques, our study highlights a promising approach to modulate TDP-43 LCD-induced aggregation. De novo binders have several potential advantages over antibodies, including lower molecular weight, potentially enhanced stability, and greater flexibility to precisely modulate binding properties and specifically target desired epitopes informed by structural priors.

Nevertheless, translating mini protein binders into effective therapies faces significant delivery and safety challenges. Antineurodegeneration protein drugs must cross the blood–brain barrier (BBB) and enter neurons, yet the BBB remains highly impermeable to large molecules (48). Receptor-mediated transcytosis strategies have been successfully utilized for transporting antibodies and Fc-fusion proteins into the central nervous system (CNS) (49). De novo protein binders have also demonstrated potential for BBB penetration when collaborating with transferrin (50). Additionally, emerging strategies, including conjugation with peptides and nanoparticles or employing Trojan horse vectors that leverage endogenous transport mechanisms, show promise (51). Gene- and mRNA-based delivery systems such as AAV and lipid nanoparticles have also been explored for CNS protein therapy, yielding sustained CNS expression (52, 53).

The study presented here demonstrates that LCD CR-helix is important to prevent aggregation of TDP-43, and the de novo binders can effectively suppress TDP-43 aggregation, thereby maintaining TDP-43 in its physiological monomeric or dimeric states within nucleus of neural cell. However, when considering the therapeutic potential of such binders there are challenges that must be addressed, such as rapid protein degradation resulting in limited circulation half-lives, as well as immunogenicity concerns. Approaches such as optimizing minimally immunogenic protein sequences, incorporating humanized or glycosylation motifs, and employing PEGylation or polypeptide fusions to extend circulating half-life and improving protein stability and minimizing digestibility in vivo to evade immune detection are useful strategies for enhancing the feasibility of potential therapeutic usage. Meanwhile, although targeting the CR within the LCD does not appear to impair splicing activity and may have beneficial effect without alternating its function involved in stress granule, the possibility that it may affect TDP-43’s physiological function cannot be fully excluded and should be carefully evaluated in future studies.

Moving forward, the design strategy presented here may also prove broadly applicable to other amyloidogenic proteins implicated in neurodegenerative diseases. If the aggregation driver regions of these proteins possess amino acid sequences with intrinsic α-helical propensities, similar de novo protein binders could be designed to stabilize their native conformations, effectively inhibiting pathological amyloid formation, thus providing therapeutic potential across a spectrum of amyloid-related disorders.

Materials and Methods

Binder Protein Backbone Generation.

We explored the scaffold target CR-helix motif by RFdiffusion fold conditioned and unconditioned method simultaneously. First, we used miniprotein scaffolds to guide the diffusion process. During diffusion, Met322, Met323, L330, and W334 were designated as hotspot residues, while the loop regions of the miniprotein templates were masked to achieve more diverse scaffold types. In the second method, we constrained the length of the generated proteins to 80–120 amino acids, again designating Met322, Met323, L330, and W334 as hotspot residues with increased contact potential between the scaffold and the target. Ultimately, by selecting appropriate interaction areas and radii of gyration, we obtained a total of 10,369 scaffold proteins.

Binder Protein Sequence Design.

The scaffolds generated by RFdiffusion were sequence-optimized using ProteinMPNN with up-weighted polar amino acids to enhance the polarity of the noninteraction interface, thereby improving protein stability. The optimized results were further subjected to energy minimization using Rosetta, and the binding and free energy (ddG) between the binder and target were evaluated based on the beta_nov16 scoring function. Additionally, various properties of the binder–target interface were computed, including interaction surface area, the number of unsatisfied hydrogen bonds, and shape complementarity. Using the AlphaFold2-initial guess method, designs with interface PAE < 6, binder PAE < 5, and pLDDT > 85 were selected for experimental validation.

Protein Expression and Purification.

Designed binders were cloned into pET21a vector with an 8-His tag (GGGYSHHHHHHHH) at the C-terminus. The binders were overexpressed in E. coli BL21(DE3) (TransGen Biotech, CD601-02) by adding 1 mM IPTG when the OD600 reached 1.2 and incubating at 20 °C for 16 h. The cells were then harvested by centrifugation (3,500×g, 4 °C, 15 min) and resuspended in a buffer containing 20 mM Tris-HCl, pH 8.0, and 150 mM NaCl. The cells were lysed by a high-pressure homogenizer under the conditions of 800 atm for 3 min. The suspension was collected by centrifugation (30,000×g, 4 °C, 1 h) and loaded onto a Ni column. The protein was washed using washing buffer containing 20 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 30 mM imidazole and eluted by elution buffer containing 20 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 300 mM imidazole. The proteins were further purified by size exclusion chromatography (Superdex increase 75 10/300 GL, Cytiva) and the monodispersed samples were collected.

TDP-43 LCD was cloned into the pET28a vector with a 6-His tag at the N-terminus. For the purification of the TDP-43 LCD construct, TDP-43 LCD was overexpressed in E. coli BL21(DE3) cells (TransGen Biotech, CD601-02) by adding 1 mM IPTG and incubating at 37 °C for 12 h. The cells were then harvested and lysed in a buffer containing 50 mM Tris-HCl, pH 8.0, and 100 mM NaCl. Cell pellets were collected by centrifugation (30,000×g, 4 °C, 1 h) and subsequently resuspended in a denaturing buffer containing 50 mM Tris-HCl, pH 8.0, 6 M guanidine hydrochloride with sonication. The resuspended protein was filtered and then loaded onto a Ni column (GE Healthcare, USA). The protein was eluted using a denaturing elution buffer containing 50 mM Tris-HCl, pH 8.0, 6 M guanidine hydrochloride and 100 mM imidazole and was further concentrated into over 30 mg/mL proteins, flash-frozen, and stored at −80 °C. Before experiments, the protein was desalted using a desalting column (GE Healthcare, USA) into a buffer containing 20 mM MES, pH 6.0. The protein was not concentrated after desalination and ThT experiment was carried out immediately.

Full-length TDP-43 (FL) was expressed as a fusion protein with a SUMO tag in E. coli ER2566 supercompetent cells (Beyotime Technology, D1039S). Expression was induced using 0.5 mM IPTG at 16 °C for 16 h. Bacterial cells were harvested and subsequently lysed in a buffer consisting of 50 mM Tris-HCl (pH 7.5), 500 mM NaCl, 25 mM imidazole, 4 mM β-mercaptoethanol, 10% glycerol, and 2 mM PMSF. The clarified lysate was loaded onto a Ni-NTA affinity chromatography column, and the target protein was eluted using a gradient of imidazole. The SUMO tag was then cleaved with Ulp1 protease, and native TDP-43 FL protein was collected from the flow-through fraction upon reapplication to the Ni-NTA column. Purified protein was finally stored in buffer containing 30 mM Tris-HCl (pH 7.5), 100 mM NaCl, and 2 mM DTT until use.

ThT Fluorescence Kinetic Assay.

The ThT fluorescence kinetic assay, owing to its specific affinity for β-sheet amyloid fibril structures, was employed to observe the progression of amyloid fibrils. Measurements were taken from a 50 µL sample housed in a NUNC 384-well plate. The fluorescence signal was captured using a microplate reader operating at excitation and emission wavelengths of 440 and 480 nm, respectively. The TDP-43 LCD ThT samples were incubated under conditions of 37 °C, with a 20 µM concentration of TDP-43 LCD, in a buffer solution (20 mM MES, pH 6.0). Each condition was reproduced in three separate experimental replicates for consistency.

BLI Assays.

To determine the interaction between TDP-43 LCD and binders and their variants using BLI assay, proteins of biotinylated TDP-43 LCD with 6 × His tag and binders with 8 × His tag were prepared. The BLI assays were performed using an Octet RED96 instrument (ForteBio) and streptavidin (SA) biosensor (Sartorius). SA biosensors were pre-equilibrated in buffer including 20 mM MES (pH 6.0), 100 mM NaCl, 10 mg/mL BSA for 1 h at room temperature. After a baseline step of 120 s, biotinylated TDP-43 or variants were immobilized on the SA biosensors as the ligands. After another baseline step, SA biosensors were then dipped into wells of binders with different concentrations for binding measurements. The concentration gradients of TDP-43/B9 in the BLI assays were 729, 243, 81, 27, and 9 nM. The concentration gradients of TDP-43/B1 in the BLI assays were 30, 10, 3, and 1 μM. The concentration of TDP-43 variants and B9 variants in the BLI assays was 1 μM. Assays were performed at 30 °C. The binding affinities were determined using OCTET Data Analysis 10.0.

CD Measurements.

CD wavelength scans were performed using a Chirascan-Plus spectrometer (Applied Photophysics). Protein samples were prepared at concentrations ranging from 0.1 to 0.2 mg/mL in 20 mM MES buffer (pH 6.0) containing 5 mM NaCl. Spectra were recorded from 260 to 190 nm with a scanning increment of 1 nm, measured in triplicate, and subsequently averaged. Thermal denaturation experiments were carried out by monitoring the CD signal at 220 nm, with temperature increased in 2 °C increments at a rate of 2 °C/min. Wavelength scan spectra were acquired at three time points: initially at 20 °C, at 99 °C, and after cooling back to 20 °C.

NS-TEM.

A drop of 5 μL aliquots sample was adsorbed onto a freshly glow-discharged grid with 200 mesh carbon support film (Beijing Zhongjingkeyi Technology Co., Ltd.) for 45 s. Then, the grid was washed with 5 μL of ddH2O and followed by another wash with 5 μL of 3% w/v uranyl acetate. The grid was further stained with 3% (w/v) uranyl acetate for 45 s. After removing the excess buffer with filter paper, the grid was dried with an infrared lamp. TEM images were acquired on a Tecnai T12 microscope (FEI Company) operated at 120 kV.

NMR Spectroscopy.

The NMR experiments were conducted using a Bruker Avance 900 MHz spectrometer equipped with a cryogenically cooled probe. The backbone assignment was accomplished according to previous publications (21). We prepared each NMR sample to a total volume of 500 μL, incorporating 20 μM of 15 N-TDP-43 LCD in 20Â mM MES (pH 6.0), 10% D2O, either with or without binders at concentrations of 16 μM, 12 μM, 8 μM, and 4 μM for titration experiments. We collected the 2D 1H-15 N HSQC spectra using the Bruker standard pulse sequence (hsqcetfpf3gpsi) with 16 scans at 298 K. For both 1H (16 ppm) and 15 N (19 ppm) dimensions, the data matrix was set to 2,048 × 160 complex points. The experimental setup, including NMR buffer compositions and data collection parameters, was in substantial accordance with those described in prior publications, ensuring the accuracy of the transferred chemical shifts in our study. Signal intensity changes were calculated by the ratio I/I0, where I denotes the intensity in the HSQC spectrum of TDP-43 LCD with binders, and I0 indicates the intensity in the HSQC spectrum of TDP-43 LCD alone. Data acquisition was performed using Topspin (3.5pl5), while NMR data analysis was performed using NMRViewJ54 (9.2.0) and SPARKY55 (3.113).

Cell Culture and Transfection.

We sourced cells from the Shanghai Cell Bank of the Chinese Academy of Sciences: 293 T cells (SCSP-5035). All cell types were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS, VISTECH) and 1% penicillin, under a controlled environment of 37 °C with 5% CO2. Cells were seeded 18 to 24 h prior to transfection to achieve a monolayer density of ~70 to 80%. Plasmids were introduced into cells using PolyJetTM reagent (SignaGen Laboratories), and after 6 h, we replaced the transfection medium with fresh DMEM. Unless otherwise specified, all cells were allowed a 24-h posttransfection period before undergoing any further drug treatments or assessments.

Preparation and Immunofluorescence Staining Assay.

We fixed the cells with 4% paraformaldehyde for 30 min at room temperature, followed by three washes with PBS. The cells were then permeabilized using 0.5% Triton X-100 in PBS for 30 min and subsequently blocked with a PBS buffer containing 3% goat serum and 0.1% Triton X-100 (PBST) for 1 h at room temperature. We incubated the cells with the primary antibody overnight at 4 °C. After three washes with PBST, the cells were treated with secondary antibodies for 1 to 2 h at room temperature. To pFTAA staining, which preferentially recognizes the β-sheet-rich protein aggregates (54, 55), fixed cells were bathed in 20 μM of pFTAA (in PBS) at room temperature for 30 min. pFTAA was prepared as previously described (55, 56). For mounting, we used ProLongTM Gold Antifade Mountant with DAPI (Thermo Fisher Cat #P36935). Fluorescent images were captured using a SP8 Leica microscope equipped with a DMI8 camera and further analyzed using Leica AF Lite software. The following antibodies were utilized: primary antibody: anti-Myc Tag (Cat#MA1-21316, Thermo Fisher), anti-pS409/410-TDP-43 (Cat#66318-1-lg, Proteintech. The dilution ratio for above antibodies was 1:1000. GAPDH (Cat#AT0002, Engibody). The dilution ratio for these antibodies was 1:10,000. Fluorescent secondary antibodies: donkey anti-rabbit Flour 568 (Life Technologies, A10042), goat anti-mouse-Alexa Flour 647 (Life Technologies, A21235). The dilution ratio for these antibodies was 1:1000. The antibodies are well validated for the indicated use by the manufacturer available on their websites.

Western Blotting.

Cultured cells’ total protein was extracted using a 4% extraction buffer (100 mM Tris pH 6.8, 4% SDS, 1% mercaptoethanol, 40% glycerol, and 0.04% bromophenol blue) that included a protease inhibitor cocktail and phosphatase inhibitor (Beyotime Biotechnology, China). Subsequently, the protein samples were subjected to a 10-min boiling period at 100 °C, followed by separation via 10 or 12.5% sodium dodecyl sulfate-polyacrylamide gel electrophoresis. The separated proteins were transferred to a nitrocellulose filter membrane and blocked using 5% skimmed milk in Tris-buffered saline with Tween 20 (TBST). Following this, the membranes were incubated with primary antibodies at 4 °C overnight. After three wash cycles with TBST solution, the bound primary antibodies were detected with a 1:2000 dilution of HRP-conjugated secondary antibody, and protein signals were visualized using the BeyoECL Star analysis reagent (Beyotime Biotechnology, China).

Statistical Analysis.

GraphPad Prism and Microsoft Excel software were used for statistical analysis. All experiments were repeated independently more than three times. The statistical significance in this study is determined by a two-tailed t test at *P < 0.05, **P < 0.01, and ***P < 0.001, ns: no significant.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix.