Rational Design of TDP-43 Derived α-Helical Peptide Inhibitors: An In Silico Strategy to Prevent TDP-43 Aggregation in Neurodegenerative Disorders

Abstract

TDP-43, an essential RNA/DNA-binding protein, is central to the pathology of neurodegenerative diseases, such as amyotrophic lateral sclerosis and frontotemporal dementia. Pathological mislocalization and aggregation of TDP-43 disrupt RNA splicing, mRNA stability, and mRNA transport, thereby impairing neuronal function and survival. The formation of amyloid-like TDP-43 filaments is largely facilitated by the destabilization of an α-helical segment within the disordered C-terminal region. In this study, we hypothesized that preventing the destabilization of the α-helical domain could potentially halt the growth of these pathological filaments. To explore this, we utilized a range of in silico techniques to design and evaluate peptide-based therapeutics that bind to pathological TDP-43 amyloid-like filament crystal structures and resist β sheet conversion. Our computational approaches, including biophysical and secondary structure property prediction, molecular docking, 3D structure prediction, and molecular dynamics simulations, were used to assess the structure, stability, and binding affinity of these peptides in relation to pathological TDP-43 filaments. The results of our in silico analyses identified a selection of promising peptides which displayed a stable α-helical structure, exhibited an increased number of intramolecular hydrogen bonds within the helical domain, and demonstrated high binding affinities for pathological TDP-43 amyloid-like filaments. Molecular dynamics simulations provided further support for the structural and thermodynamic stability of these peptides, as they exhibited lower root-mean-square deviation and more favorable free energy landscapes over 300 ns. These findings establish α-helical propensity peptides as potential lead molecules for the development of novel therapeutics against TDP-43 aggregation. This structure-based computational approach for the rational design of peptide inhibitors opens a new direction in the search for effective interventions for ALS, FTD, and other related neurodegenerative diseases. The peptides identified as the most promising candidates in this study are currently subject to further testing and validation through both in vitro and in vivo experiments.

Introduction

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disorder characterized by the loss of motor neurons in the brain and spinal cord, leading to progressive muscle weakness, paralysis, and eventually death.^1,2 A critical feature of ALS pathophysiology is the nuclear clearance and misfolding of essential RNA-binding proteins, resulting in the formation of insoluble aggregates.^3,4 Nearly all ALS cases (97%) show pathologic aggregation and nuclear depletion of transactive response DNA-binding protein of 43 kDa (TDP-43), a highly conserved RNA/DNA-binding protein that plays a crucial role in RNA splicing and repression of cryptic exons, regulation of mRNA stability, and transport.⁵⁻¹⁰ Familial forms of ALS (5–10% of cases) have been linked to mutations in several genes, including TARDBP that encodes TDP-43.^11,12 Pathological mislocalization and aggregation of TDP-43 are also hallmarks of limbic-predominant age-related TDP-43 encephalopathy (LATE),¹³ and ∼50% of frontotemporal lobar degeneration (FTLD-TDP) and has been observed in a subset of Alzheimer’s disease (AD) cases.^3,14−17 The broad relevance of TDP-43 pathology to multiple age-related neurodegenerative disorders underscores the need to elucidate molecular or structural alterations that trigger TDP-43 aggregation and to develop therapies targeting these pathways.

TDP-43 is a multidomain DNA/RNA-binding protein that comprises a folded N-terminal domain (residues 1–102) important for nuclear localization, oligomerization, and mediating protein–protein interactions;¹⁸⁻²² two RNA recognition motifs (RRM1, residues 104–176; and RRM2, residues 192–262) critical for binding to UG-rich RNA sequences;^23,24 and a predominantly disordered C-terminal low complexity domain (residues 274–414).^25,26 The C-terminal low-complexity domain (LCD) of TDP-43, also referred to as a prion-like domain, is enriched in glycine, glutamine, and asparagine residues and is a hotspot for ALS- and frontotemporal dementia (FTD)-associated mutations.²⁷ This region regulates TDP-43 solubility and serves as a crucial site for protein–protein interactions necessary for splicing, autoregulation, and other physiologic functions.^6,19,28−32 Though predominantly disordered, the LCD also contains an α-helical domain (aa 320–340) divided into two helical segments: residues 321–330 (AMMAAAQAAL) and 335–343 (GMMGMLASQ) (Figure 1A). Intermolecular interactions of the helical domain play an important role in driving TDP-43 liquid–liquid phase separation and assembly within physiologic ribonucleoprotein condensates in cells.³³⁻³⁶ However, under pathological conditions, the LCD is abnormally cleaved to form insoluble granulofilamentous aggregates and significantly contributes to the process of TDP-43 aggregation.^{17,25,37−41}

Figure 1.

(A) Schematic of TDP-43’s amyloidogenic core region, highlighting crucial residues within the α-helical domain that are involved in the misfolding and aggregation of TDP-43. (B) Schematic outline of the step-by-step approach used to develop peptides with enhanced helical propensity and analysis methods to assess peptide inhibitor stability and the ability to resist beta-strand conversion.

Within the LCD, several studies have identified an amyloidogenic core region that includes the α-helical domain (Figure 1A).⁴²⁻⁴⁴ This core region (residues 311–360) is necessary and sufficient for TDP-43 aggregate formation in cells and in vivo and is able to template the aggregation of full-length TDP-43, suggesting a key role in regulating the structural transition of TDP-43 to amyloid-like conformations.^42,45 Indeed, cryo-electron 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 amyloidogenic core residues and the α-helical domain.⁴⁶⁻⁴⁸ Cryo-EM of patient-derived aggregates revealed that the ordered filament core, composed of stacked TDP-43 molecules, adopts a double-spiral fold that is formed by the amyloidogenic core and flanking residues (282–360). Furthermore, the nucleus of the double-spiral fold contains several β-strands formed by hydrophobic residues of the α-helical domain.⁴⁶

Recent NMR studies and biophysical assays have examined how the α-helical region within the amyloidogenic core interacts with neighboring residues and changes structure during the early stages of aggregation.^42,49 Specifically, residues 331–334 (QSSW) (Figure 1A) that divide the α-helical domain into two helical segments play a key role in the conversion of intramolecular helix–helix contacts (i.e., interactions between amino acid side chains) to helix-β sheet contacts. Solid-state NMR studies identified a low propensity for helical conformational chemical shift at the Ser 332 residue,⁵⁰ and S332 in particular was implicated in many inter-residue contacts that led to fibrillar conversion of helix to β sheet fibrils.⁴⁹ In further support of this notion, phosphorylation of serine residues (S332, S333) also leads to helical breakage and TDP-43 misfolding and aggregation.⁴⁹ Taken together, these data suggest that the lack of helix stabilizing intramolecular contacts between S332, S333, and neighboring residues within the adjacent helical segments contributes to protein misfolding and structural transformation into amyloid-like fragments.^42,49

Based on these findings and cryo-EM structures of TDP-43 protofilaments, in this study, we used computational and rational design techniques to generate novel peptides with enhanced helical propensity that bind to TDP-43 amyloid-like filaments but resist β sheet conversion. This strategy effectively caps the growing tips of filaments (see the graphical abstract), and similar methods have been used to design peptide-based inhibitors of tau, β amyloid, and synuclein.⁵¹ Using the sequence of TDP-43’s amyloidogenic core (residues 311–360) as a starting point, we generated 400 novel peptides (50 residues in length) with mutations that stabilize the α-helical region. We hypothesized that peptides with mutations of S332 and/or S333 that enhance α-helix stability would bind TDP-43 amyloid-like filaments but resist conversion to β-sheet structure. We then employed multiple computational methodologies, including sequence-based secondary structure property prediction, three-dimensional (3D) structure modeling of peptides, docking analysis, and molecular dynamics, to rank the novel peptides according to their helical secondary structure, binding efficiency with TDP-43 amyloid-like fibrils, and ability to resist structural transformation (Figure 1B). Our approach takes advantage of recent advances in computing power and innovations in in silico techniques to rationally design and test peptide inhibitors with therapeutic potential against TDP-43 aggregation-related disorders. This strategy may be used more broadly for de novo therapeutic peptide design.

Results and Discussion

Computational Design Pipeline

Considering the critical role of the amyloidogenic core and α-helical domain residues in initiating TDP-43 aggregation, we used an in silico approach to investigate whether peptide inhibitors with enhanced α-helical propensity are a potential therapeutic strategy to inhibit the growth of pathological TDP-43 filaments (Figure 1B). We retrieved the wild-type sequence of TDP-43 (residues 311–360) from the Uniprot database (Q13148) and generated a library of unique peptides, each with a length of 50 residues. Given the helix destabilizing effects of serine residues 332 and 333,⁵⁰ we focused on peptides with modifications of these polar residues. The helical secondary structure is characterized by robust intramolecular interactions; therefore, we reasoned that substitutions at S332 and/or S333 could potentially increase interactions that contribute to the stability of the helical domain. This led us to concentrate on 400 distinct mutant peptide variants with diverse secondary structural properties. RaptorX, an ultradeep convolutional residual neural network that makes structure predictions from the primary sequence, was used to predict the secondary structure of each variant in our peptide library.⁵² Using this approach, we generated a ranked list of peptides according to their helical propensity; this list was refined further based on helical penalty scores, disorder prediction, and amino acid biochemical properties (Tables S1 and S2).^53,54 We selected peptides with substitutions of S332A/S333W, S332R/S333L, S332A, S332R, and S333L that exhibit enhanced helical propensity compared to the wild-type amyloidogenic core peptide sequence and low helical penalty scores. Considering that the TDP-43 C-terminal domain is predominantly disordered,^46,55 we also included a few variants, such as S332V/S333V and S332V, that are more ordered compared to wild type but have only slightly increased helical propensity scores (Tables S1 and S2). These top peptide candidates were analyzed further with 3D modeling, docking analysis, and molecular dynamics.

Increased Intramolecular Interactions Stabilize the Helical Structure in Top Hit Peptide Variants

To gain a deeper understanding of the structural characteristics of the top peptide hits, we used Rosetta comparative modeling (RosettaCM) to generate 3D models of backbone and side-chain atom topologies.⁵⁶ The predicted 3D structures were used to analyze the number and stability of intramolecular interactions, structural behavior, and differential energy profiles at each residue (Figure 2). Figure 2A depicts the 3D structural coordinates of the top peptide hits, highlighting their increased helical propensity compared to that of the wild-type model. To understand the basis of the enhanced helical structure of our top peptide hits, we analyzed the number of hydrogen bonds and intramolecular interaction energies for each peptide. Notably, the helical propensity variants displayed an increased number of intramolecular hydrogen bonds compared to that of the wild-type peptide (Figure 2A). Overall, 22 hydrogen bonds were maintained in the wild-type helical domain, whereas ≥30 hydrogen bonds (range 30–37) were maintained in the helical propensity peptides’ helical domain (Table 1).

Figure 2.

Comparative analysis of wild-type (Wt) and helical propensity variant peptide structures and intramolecular interaction energies. (A) 3D structure representations of Wt and top hit peptides showing enhanced helical propensity. Comparison of the number of intramolecular hydrogen bonds within the helical region underscores the structural differences between Wt and the top helical propensity peptides. (B) Intramolecular interaction energy is plotted for Wt (black) and helical propensity peptides at each residue. The higher (more negative) differential energy profiles for the top hit peptides offer insight into their stability and illustrate different energetic contributions at specific residues.

Table 1. No. of Hydrogen Bonds and Contributing Atoms in the Helical Domain for Wt and Helical Propensity Variants.

peptides	intramolecular interaction in the helical domain				no. of hydrogen bonds
Wt	M13:H–P10:O	A14:H–P10:O	A15:H–A11:O	A16:H–M12:O	22
	Q17:H–M13:O	A18:H–A14:O	A18:H–A15:O	A19:H–A15:O
	A19:H–A16:O	L20:H–A16:O	Q21:H–A18:O	S23:H–L20:O
	W24:H–L20:O	W24:H–S23:OG	G25:H–L20:O	M26:H–Q21:O
	M27:H–W24:O	G28:H–G25:O	M29:H–G25:O	L30:H–M26:O
	P10:HD2–P10:O	G25:HA3–Q21:O
S332A S333W	A14:H–P10:O	A15:H–A11:O	A16:H–M12:O	A16:H–M13:O	31
	Q17:H–M13:O	A18:H–A14:O	A19:H–A15:O	L20:H–A16:O
	L20:H–Q17:O	Q21:H–Q17:O	Q21:H–A18:O	A22:H–A18:O
	A22:H–A19:O	W23:H–A19:O	W23:H–L20:O	W24:H–L20:O
	G25:H–Q21:O	M26:H–A22:O	M27:H–W23:O	G28:H–W24:O
	M9:H–G25:O	L30:H–M26:O	A31:H–M27:O	A31:H–G28:O
	S32:H–G28:O	Q33:H–M29:O	Q33:H–L30:O	Q34:H–L30:O
	W23:HD1–A19:O	G28:HA2–W24:O	S32:HB2–M29:O
S332R S333L	A14:H–P10:O	A15:H–A11:O	A16:H–M12:O	A16:H–M13:O	32
	A15:H–A11:O	A16:H–M12:O	A16:H–M13:O	Q17:H–M13:O
	A18:H–A14:O	A19:H–A15:O	L20:H–A16:O	Q21:H–Q17:O
	Q21:H–A18:O	R22:H–A18:O	L23:H–A19:O	W24:H–L20:O
	G25:H–Q21:O	M26:H–R22:O	M26:H–L23:O	M27:H–L23:O
	G28:H–W24:O	G28:H–G25:O	M29:H–G25:O	M29:H–M26:O
	L30:H–M26:O	A31:H–M27:O	A31:H–G28:O	S32:H–G28:O
	Q33:H–M29:O	Q33:H–L30:O	Q34:H–L30:O	Q34:H–A31:O
S332A	A14:H–P10:O	A15:H–A11:O	A16:H–M12:O	A16:H–M13:O	34
	A15:H–A11:O	A16:H–M12:O	A16:H–M13:O	Q17:H–M13:O
	A18:H–A14:O	A19:H–A15:O	L20:H–A16:O	Q21:H–Q17:O
	Q21:H–A18:O	R22:H–A18:O	L23:H–A19:O	W24:H–L20:O
	G25:H–Q21:O	M26:H–R22:O	M26:H–L23:O	M27:H–L23:O
	G28:H–W24:O	G28:H–G25:O	M29:H–G25:O	M29:H–M26:O
	L30:H–M26:O	A31:H–M27:O	A31:H–G28:O	S32:H–G28:O
	Q33:H–M29:O	Q33:H–L30:O	Q34:H–L30:O	Q34:H–A31:O
	G28:HA2–W24:O	S32:HB2–M29:O
S332R	A14:H–P10:O	A15:H–A11:O	A16:H–M12:O	A16:H–M13:O	37
	Q17:H–M13:O	A18:H–A14:O	A19:H–A15:O	L20:H–A16:O
	L20:H–Q17:O	Q21:H–Q17:O	Q21:H–A18:O	R22:H–A18:O
	S23:H–A19:O	S23:H–L20:O	S23:HG–A19:O	W24:H–L20:O
	G25:H–Q21:O	M26:H–R22:O	M26:H–S23:O	M27:H–S23:O
	G28:H–W24:O	G28:H–G25:O	M29:H–G25:O	M29:H–M26:O
	L30:H–M26:O	A31:H–M27:O	A31:H–G28:O	S32:H–G28:O
	S32:H–M29:O	S32:HG–M29:O	Q33:H–M29:O	Q33:H–L30:O
	Q34:H–L30:O	Q34:H–A31:O	A14:HA–Q17:OE1	S23:HB1–A19:O
	S23:HB1–L20:O
S332V S333V	A14:H–P10:O	A15:H–A11:O	A16:H–M12:O	A16:H–M13:O	33
	Q17:H–M13:O	A18:H–A14:O	A19:H–A15:O	L20:H–A16:O
	L20:H–Q17:O	Q21:H–Q17:O	Q21:H–A18:O	V22:H–A18:O
	V22:H–A19:O	V23:H–A19:O	V23:H–L20:O	W24:H–L20:O
	G25:H–Q21:O	G25:H–V22:O	M26:H–V22:O	M27:H–V23:O
	G28:H–W24:O	G28:H–G25:O	M29:H–G25:O	M29:H–M26:O
	L30:H–M26:O	A31:H–M27:O	A31:H–G28:O	S32:H–G28:O
	Q33:H–M29:O	Q33:H–L30:O	Q34:H–L30:O	Q34:H–A31:O
	S32:HB2–M29:O
S333L	A14:H–P10:O	A15:H–A11:O	A16:H–M12:O	A16:H–M13:O	36
	Q17:H–M13:O	A18:H–A14:O	A19:H–A15:O	A19:H–A16:O
	L20:H–A16:O	Q21:H–Q17:O	S22:H–A18:O	S22:H–A19:O
	S22:HG–A18:O	L23:H–A19:O	L23:H–L20:O	W24:H–L20:O
	G25:H–Q21:O	M26:H–S22:O	M27:H–L23:O	G28:H–W24:O
	M29:H–G25:O	M29:H–M26:O	L30:H–M26:O	A31:H–M27:O
	A31:H–G28:O	S32:H–G28:O	S32:H–M29:O	Q33:H–M29:O
	Q33:H–L30:O	Q34:H–L30:O	Q34:H–A31:O	A14:HA–Q17:OE1
	S22:HB1–A18:O	S22:HB1–A19:O	G28:HA2–W24:O	S32:HB2–M29:O
S332V	A14:H–P10:O	A15:H–A11:O	A16:H–M12:O	Q17:H–M13:O	32
	Q17:HE21–Q21:OE1	A18:H–A14:O	A19:H–A15:O	A19:H–A16:O
	L20:H–A16:O	L20:H– Q17:O	Q21:H–Q17:O	V22:H–A18:O
	V22:H–A19:O	S23:H–A19:O	S23:H–L20:O	W24:H–L20:O
	G25:H–Q21:O	M26:H–V22:O	M26:H–S23:O	M27:H–S23:O
	M27:H–W24:O	G28:H–W24:O	M29:H–G25:O	L30:H–M26:O
	A31:H–M27:O	A31:H–G28:O	S32:H–G28:O	Q33:H–M29:O
	A18:HA–Q17:OE1	S23:HB1–L20:O	S32:HB1–G28:O	S32:HB2–M29:O

Beyond the analysis of hydrogen bonds, our investigation sought to comprehensively assess the various energy interaction types within these peptides by utilizing an interaction energy matrix.⁵⁷ This encompassed a comprehensive prediction of intramolecular interaction energies, including all categories of covalent, polar, and nonpolar interactions. This analysis demonstrated distinct values within the helical domain (residues 320–340) for the helical propensity variants and the wild-type peptide. Specifically, the helical propensity variants exhibit an average energy of −96.119 kJ/mol, whereas the wild-type counterpart displays an average energy of −57.8 kJ/mol (Figure 2B).These findings suggest that increased intramolecular energy observed in the helical propensity variants plays an important role in maintaining the stability of the helical structure. Upon closer examination, we observed that the substitutions at S332 and/or S333 residues contributed a marked increase in the intramolecular interaction energy. Specifically, mutations to S332A/S332W yielded an intramolecular interaction energy of −90.14 kj/mol, whereas the wild-type model exhibited fewer intramolecular interactions at S332 and S333 (−33.2 kj/mol) (Figure 2B). These data indicate that alterations in the serine residues significantly influence the intramolecular interaction characteristics in the helical region (residues 320–340). These observations also suggest that the mutated serine residues in our top peptide hits demonstrate a more favorable intramolecular interaction energy profile with neighboring residues, thus contributing robust interactions that stabilize their helical secondary structure.

Analysis of Helical Propensity Peptide Stability through Molecular Dynamics

Comprehensive molecular dynamics (MD) simulations were performed to validate the folding and conformational stability of top hit peptides and to gain insight on their dynamic behavior and interplay among constituent residues compared to the wild-type peptide. We conducted MD simulations using GROMACS (version 2021). For the initial equilibrium system, we employed the Optimized GROMOS43a1 force field in conjunction with the Simple Point-Charge (SPCE) water model.⁵⁸ We analyzed the wild-type and top helical propensity monomer structures for up to 100 ns to understand the time evolution of their structural characteristics (Figure 3), intramolecular interactions (Figure 4), and free energy properties (Figure 5). Root mean square deviation (RMSD) analysis was employed to assess the stability and conformational changes of the proteins. Generally, lower RMSD values indicate a more stable conformation, whereas higher or fluctuating RMSD values suggest structural changes. Notably, the helical propensity variants exhibited lower RMSD values throughout the simulation, suggesting enhanced stability compared to the wild-type model (Figure 3A). Hydrogen bonds are an important source of intramolecular interactions that maintain stable helical structure. Therefore, we also examined the total number of hydrogen bonds in the helical domain (320–340) for the wild-type and helical propensity variants. The wild-type model exhibited an average of 10 to 15 hydrogen bonds, whereas the helical propensity variants maintained 15 to 20 hydrogen bonds over the simulation period (Figure 3B). Furthermore, we determined the secondary structure properties for the helical propensity peptides throughout the simulation period. Secondary structural analysis of the wild-type model revealed an unstable helical segment spanning residues 327 to 340 that transitioned into loop structures (Figure 3C). In contrast, the variants displaying an enhanced propensity for helical formation, specifically S332A/S333W, S332R/S333L, and S333L, maintained stable secondary structures characterized by well-defined helical segments throughout the simulation. S332V/S333V and S332V variants that were more ordered but had slightly lower helical propensity scores (Table S1) showed some fluctuations to coil and turn structures but still displayed more helical stability than the wild type-model (Figure 3C). Combining all of these results, MD data indicate that substitutions of serine residues in our top peptide hits substantially enhance helical conformational stability, in part by contributing robust intramolecular hydrogen bonds that stabilize their helical secondary structure.

Figure 3.

Molecular dynamics simulations were used to assess the conformational stability of helical propensity and wild-type peptides. (A) The RMSD values were calculated to measure the stability and conformational changes of helical propensity peptides and the wild-type model over a 100 ns simulation period. The plot shows the relative stability of the helical propensity variants compared to the wild-type peptide. (B) The total number of hydrogen bonds formed within the helical domain of each peptide was monitored throughout the 100 ns simulation period. This spline plot illustrates the variation in the hydrogen bonds for the helical propensity variants and the wild-type model, providing insights into the strength and stability of the intramolecular interactions. (C) Residue-level secondary structure analysis for each peptide over the simulation. Secondary structural properties (C, coil; I, 5turn helix; G, 3turn helix; H, 4turn helix; B, Beta bridge; E, β Sheet; T, turn) were analyzed for each residue in the helical propensity variants and the wild-type model over the course of the 100 ns simulation. This plot illustrates the stability of helical conformations in our top hit peptides relative to the wild-type amyloidogenic peptide.

Figure 4.

Heatmap for predicted pairwise interactions of residues 311–360 across molecular dynamics snapshots for the (A) wild-type and (B) S332A/S333W peptide. The color gradient indicates variation in intramolecular interaction energies (kj/mol). In each panel, the black dotted boxes highlight inter-residue interactions within the helical domain. The red dotted boxes highlight interactions between helical domain residues and residues in the flanking region (outside of the helical domain).

Figure 5.

Comparative Gibbs free energy analysis from molecular dynamics simulations of wild-type (Wt) and helical propensity peptides. (A–C) 3D graphical representation of calculated Gibbs free energy for (A) Wt, (B) S332A/S333W, and (C) and S333L peptides, providing insights into the thermodynamic stability of these structures (blue color ranges indicate a favorable, high Gibbs free energy, and red color ranges indicate low Gibbs free energy). (D–F) Selected snapshots from molecular dynamics simulations at various time points for (D) Wt, (E) S332A/S333W, and (F) S333L peptides, illustrating the dynamic behavior and conformational changes over time. (G) The cumulative contribution of intramolecular interaction energies between residues at 332 and 333 position and neighboring residues (A329, L330, Q331, W334, G335, and M336) observed over the course of 100 ns MD. Analysis performed for the wild-type (S332, S333) model and peptide variants with mutated residues S332A/S33W and S333L.

We went on to examine pairwise interactions between the residues comprising TDP-43’s amyloidogenic core region (Figure 4). In the wild-type peptide, we observed 25 distinct interactions between helical domain residues and residues outside of the helical region during the 100 ns simulation. In particular, residues S332 and W334 within the helical domain interact with residues G357 and N358 outside of the helical domain (Figure 4A, red dotted box). These observations underscore the relative lack of helix-stabilizing interactions between S332 and neighboring residues within the helical domain. These findings are also consistent with prior studies that show that S332 and W334 participate instead in helix-destabilizing interactions that initiate helix to β sheet conversion.^42,49 In contrast, pairwise interactions for the S332A/S333W helical propensity peptide are most robust within the helical regions and contribute to stable intramolecular interaction energies (Figure 4B, black dotted box). Thus, relative to the S332A/S333W peptide, the wild-type peptide exhibits an elevated frequency of crosstalk interactions between helical domain residues and residues outside of this region, which contributes to the destabilization of the α-helical secondary structure.

Next, we predicted the thermodynamic stability of the wild-type and top two helical propensity peptides over the simulation period. From the MD simulations, Gibbs free energy heatmaps were generated (Figure 5A–C). We found distinct clusters representing different conformational states of the peptides, with cooler blue colors indicating relatively stable states. The helical propensity variants, S332A/S333W and S333L, exhibited stable helical conformations throughout the simulation (Figure 5E,F), as indicated by the favorable Gibbs free energy values within their respective clusters (Figure 5B,C). In contrast, the wild-type model showed structural changes at 1.5 Å RMSD, at which point it reached an unstable state with reduced free energy (Figure 5A,D). This instability was mainly attributed to the presence of polar residues, such as S332 and S333, in the helical domain, resulting in fewer hydrogen bonds and an unstable intramolecular interaction energy. Conversely, the helical propensity variants maintained stable conformations with stable intramolecular interaction energies (Figure 2B; Figure S1).

The Role of Serine 332 and 333 Residues in TDP-43 Fibrillization: Insights into the Importance of Neighbor Interactions in the Helical Domain

Our data (Figures 3 and 4) agree with previous investigations pinpointing inter-residue interactions, particularly interactions involving S332 and W334, that contribute to structural conformational changes of TDP-43’s amyloidogenic core and subsequent fibril formation.^49,59 It has been suggested by Zhuo et al. that the lack of intramolecular interactions between S332 and neighboring residues has a potential impact on fibril initiation.⁴⁹ SSNMR with proton detection also revealed a close interaction between the indole Nε1–Hε1 of W334 and the side-chain carbonyl of Q343. Given the fundamental requirement of three to four residues for sustaining a single turn within the helical structural motif, we investigated how substitutions of S332 and/or S333 impact the cumulative intramolecular interaction energies between those residues and their three neighboring residues in the helical domain (e.g., S332/3 with A329, L330, Q331, W334, G335, and M336). Over the course of a 100 ns simulation period, it became evident that S332A/S333W and S333L substitutions resulted in more favorable (more negative) cumulative intramolecular interaction energies compared to the wild-type model (Figure 5G). These findings support the view that robust intramolecular neighbor residue interactions seen with S332A, S333W, and S333L substitutions likely contribute to strong helix-stabilizing interactions in our top hit helical propensity peptides.

To test this idea further, we analyzed the probability of helical secondary structure for wild-type and our helical propensity peptide monomers throughout the simulation period. We found that the wild-type model has reduced helical probability compared to our top hit peptides (Figure S2A). We also plotted the residue-wise probability (i.e., helical probability at each residue) of the wild-type and helical propensity peptides (Figure 6A;Figure S2B–G). We found that S332A/S333W, S333L, S332R, and S332V/S333V peptides display higher helical probability at S332/3 neighbor residues (331–335) compared to the wild-type model. Therefore, these probability data provide additional evidence that serine residues in the helical domain play a major role in helix destabilization and/or initiating misfolding. Based on these findings, we focused on the S332A/S333W helical propensity variant for further analysis, including docking studies and complex simulations with amyloid-like fibrils.

Figure 6.

Predicted residue-wise and overall α-helical or β-sheet probability for the wild-type peptide and the S332A/S333W helical propensity peptide in two different states: (A–D) peptide monomer state and (B–H) complex state (peptide bound to TDP-43 amyloid-like fibrils). (A, E) Residue-wise predicted helical probability and (C, G) β sheet probability in the monomer and complex state for the wild-type and S332A/S333W peptides. (B, F) Overall α-helical probability for the (B) monomer state and (F) complex state; two-tailed t test, ****p < 0.0001. (D, H) Overall β sheet probability for the (D) monomer state and (H) complex state; two-tailed t test, ****p < 0.0001. These figures collectively provide a comprehensive analysis of the α-helical and β sheet probabilities for the wild-type and helical propensity peptides in both the monomer and complex states. The data are presented as means with standard error of the mean from n = 2 technical replicates.

Efficiency of Helical Propensity Peptide Binding with TDP-43 Amyloid-like Structures

The binding efficacy of the S332A/S333W helical propensity peptide with TDP-43 amyloid-like fibrils was evaluated by using three distinct TDP-43 filament crystal structures. The first structure was derived from aggregated TDP-43 located in the frontal and motor cortices of a patient diagnosed with ALS/FTLD.⁴⁶ Two additional structures were elucidated through crystallographic analysis.^47,48 Each of these structures displays unique β sheet conformations. The patient-derived structure 7PY2 contains 10 short beta sheets, with the longest one extending from serine 332 to leucine 340. Meanwhile, structures 7KWZ and 6N37 exhibit 10 and 3 beta sheets, respectively, each adopting distinct β sheet conformations. Interestingly, the helical domain within these structures forms a lengthy β sheet conformation. As indicated by prior studies,^42,49 the transition step of the helical domain serves as the initial phase of aggregate formation. Taken together with the patient-derived cryo-EM and crystal-solved structures, the data suggest that helical domains participate in forming the most elongated, stable β sheets within these three structures. Therefore, when these reference TDP-43 fibrillar structures are complexed with our top hit helical propensity peptides, we predicted that the S332A/S333W peptide inhibitor will bind amyloid-like fibrils but resist structural transformation compared to the wild-type peptide; furthermore, we speculated that it would prevent the addition of TDP-43 CTD monomer fragments to amyloid-like fibrils.

Previous findings have suggested that TDP-43 CTD needs interacting partners (additional C-terminal fragment) to undergo fibrillation.³¹ As a result, in our MD simulation results with only monomers, we did not observe any stable β sheet formation (Figure 3). Therefore, we tested our top helical propensity peptides and the wild-type model bound to several crystal structures of C-terminal TDP-43 fibrils, including 7KWZ, 7PY2, and 6N37.⁴⁶⁻⁴⁸ The amyloid-like fibrils and peptides were docked using HDOCK,⁶⁰ a protein–protein docking tool that allows amino acid sequences to be used as inputs and combines homology search, template-based modeling, structure prediction, and macromolecular docking functions (Figure S3). The docking analysis revealed that the helical propensity variants and wild-type peptide achieved similar scores (range: −685.51 to −759.83 kcal/mol) (Table 2). Importantly, for each TDP-43 fibril structure we tested, the S332A/S333W helical propensity peptide binds to the helical domain residues (320–340), which participate in β sheet formation (Figure S3A). These data suggest that the S332A/S333W peptide has favorable binding affinity with TDP-43 amyloid-like fibrils and has potential to act as a TDP-43 C-terminal fragment mimetic. To further investigate the stability and structural changes between the complexes, we selected top poses with favorable interactions for the molecular dynamics simulation analysis.

Table 2. Docking Scores of Helical Propensity Peptide Variants with Amyloid-like Fibrils of TDP-43.

variants	7KWZ48	7PY246	6N3747
Wt	–699.60	–622.41	–416.96
S332A/S333W	–685.51	–599.62	–670.91
S332R/S333L	–709.39	–638.95	–653.31
S332V/S333V	–717.35	–662.35	–660.23
S332V	–689.09	–587.00	–629.61
S333L	–759.83	–701.79	–666.20
S332A	–725.48	–629.92	–697.64
S332R	–712.70	–647.64	–720.48

Helical Propensity Peptides Maintain Helical Structure and Resist β-Sheet Transformation When Bound to TDP-43 Fibrils

As a next step in testing our top peptide inhibitors, we were interested in assessing the structural stability of helical propensity peptides after binding to TDP-43 amyloid-like fibrils. From this analysis, we aimed to find the probability of helix to β sheet transformation of (i) peptide monomers and (ii) peptides bound to TDP-43 amyloid-like fibrils. To elucidate the molecular interaction between the complexes, we used top poses from the docking analysis for the MD simulation. To minimize bias of the starting position, simulations were carried out for 100 ns (monomer) and 300 ns (complexes), and dual technical replicates were performed for increased reliability. The simulation results indicate that the peptide/fibril complex formed a stable structure and maintained favorable interactions with the TDP-43 fibrils. In the monomer state, the S332A/S333W helical propensity showed a greater than 2-fold increase in helical probability compared to the wild-type peptide (Figure 6A,B). As we mentioned previously, monomers do not readily fibrillize, but we still found that the S332A/S333W peptide shows a small decrease in β sheet probability compared to the wild-type monomer state (Figure 6C,D). Consistent with previous studies of the known amyloidogenic core, the wild-type peptide bound fibril model showed a significant increase in β sheet probabilities and a corresponding reduction in helical probability, (Figure 6E–H). Residue-wise analysis showed that S332 and neighbor residues, such as 334 to 360 AAs, had a marked increase in β sheet probability and decreased helical probability (Figure 6E,G). On the other hand, the helical propensity peptide (S332A/S333W) maintains stable interaction with neighbor residues, and we found very low values for β sheet probability in both monomer and complex states (Figure 6D,H). To gain a structural view, snapshots of different time points from the MD simulations were examined for complexes of the wild-type peptide or S332A/S333W peptide with 7KWZ (peptide-7KWZ complexes). Our data indicate that the wild-type model bound with amyloid-like fibrils is unable to maintain helical conformations, and most of the residues from the helical domain transition to loops or β sheets (Figure S4A). In contrast, the helical propensity peptides maintained the helical conformation throughout the simulation period (Figure S4B,C). We also observed some conformational changes in the helical propensity of peptides during the simulation, suggesting their flexibility and potential for adopting different conformations.

Promising Capping of TDP-43 Amyloid-like Fibrils by Helical Propensity Peptides

Finally, we tested the thermodynamic stability of complexes of either wild-type (Figure 7A,B) or helical propensity peptide bound to TDP-43 amyloid-like fibrils (Figure 7C,D). To perform this analysis, we predicted the free energy landscape as a function of time on the x axis and plotted helical or β sheet probabilities on the y axis. We sought to understand whether helical propensity peptides bound to TDP-43 fibrils display a more thermodynamically stable helical secondary structure than the wild-type peptide/fibril complex. We found that the S332A/S333W peptide exhibits a higher, more stable helical probability in the complex state (45–50%), with more favorable (dark blue) free energy clusters, compared to the wild-type peptide/fibril complex (Figure 7A,C). Moreover, the β sheet probability of the S332A/S333W peptide/fibril complex is extremely low throughout the simulation period, with minimal favorable free energy clusters (Figure 7D). As we expected, the wild-type peptide/fibril complex displays increasing β sheet probability as the simulation progresses and reaches ∼20% β sheet probability within favorable free energy clusters by 250–300 ns (Figure 7B). Overall, our data suggest that the S332A/S333W helical propensity peptide has strong binding efficiency with amyloid-like fibrils, forming a thermodynamically stable complex structure while still maintaining stable helical propensity. Thus, this rationally designed helical propensity peptide displays promising features, as it has potential to bind and cap TDP-43 amyloid-like fibrils and act as a mimetic of TDP-43’s amyloidogenic core region but resists β sheet transformation. Overall, our study provides insights into the binding mechanism and stability of TDP-43 C-terminal fragment peptide mimetics, which could be useful for developing new therapies combating TDP-43 proteinopathies.

Figure 7.

Predicted free energy landscape of (A, B) wild-type or (C, D) S332A/S333W peptide-bound amyloid-like fibrils, with time (ns) plotted along the x axis and the helical or β sheet probabilities represented on the y axis. Each chain of the amyloid-like fibrils is color-coded with distinct colors, whereas the peptides are highlighted in gray. The free energy values are depicted using a rainbow color scheme, where favorable high-energy regions are represented in blue or violet and unfavorable lower-energy regions are shown in orange or red. This landscape provides insights into the conformational stability and dynamics of the peptides within the amyloid-like fibril structure, showcasing conversion from helical to β sheet conformations and their associated free energy variations. The helical probability and β sheet probability data were obtained from the average of two technical replicates, each conducted over a 300 ns simulation period.

Conclusions

By utilizing advanced in silico techniques, we uncover a new direction for the development of therapeutic strategies against TDP-43 aggregation-related disorders, such as ALS and FTD. Our work specifically targets the destabilization of the α-helical domain in the C-terminal region of TDP-43, a process that is critical to the formation of pathological amyloid-like filaments. These helical domains act as crucial sentinels, effectively safeguarding against the formation of amyloid-like fibrils. Simultaneously, it will be important to explore whether the intrinsic flexibility of TDP-43’s α-helical region, which is split into two short helices, plays a physiological role and to ensure that our helical propensity peptides do not disrupt endogenous TDP-43 splicing function, mRNA transport, and/or hinder essential protein–protein interactions. Considering the various constraints of this study, including the absence of conclusive evidence regarding the peptide ability to bind to the lateral faces of aggregates and the exclusive use of a single force field and docking tool, subsequent experiments will be essential to validate our findings. Therefore, our helical propensity peptides are now undergoing further evaluation for specificity and target engagement in vitro and in cellular assays.

Nevertheless, the peptides highlighted in our computational analysis, notable for their binding affinity and improved stability when complexed with TDP-43 amyloid-like fibrils, show the potential to bind and cap the growing tips of filaments. If validated through ongoing investigations of target engagement, efficacy, and specificity in vitro and in vivo, these peptides will bring the field closer to developing novel therapeutic strategies for TDP-43 aggregation-related disorders. The potential of in silico methodologies to accelerate the discovery of therapeutic leads represents a significant contribution to the field of drug discovery for neurodegenerative diseases. Our findings underscore the prospect of targeting the α-helical domain in TDP-43 to prevent the formation of pathological filaments, a strategy that could potentially revolutionize the treatment of conditions such as ALS and FTD.

Materials and Methods

Protein Data Set Preparation

Starting structures for docking analysis and molecular dynamics simulations were extracted from the Protein Data Bank.⁶¹ We retrieved three TDP-43 crystal structures (PDB ID: 7KWZ, 6N37, and 7PY2).⁴⁶⁻⁴⁸ These included structures of ALS/FTLD patient-derived TDP-43 aggregates from the frontal and motor cortices as well as two structures elucidated through crystallographic analysis. The selected PDB structures were prepared for computational approaches using the Preparation Wizard of Schrödinger.⁶² The preparation of proteins included adding hydrogens, ionizing metals, and fixing missing residues and atoms. To mimic the neutral pH conditions of the physiological environment, modifications were made such that the N-terminus was positively charged (NH3⁺) and the C-terminus was negatively charged (COO−). The prepared structures were used for docking and molecular dynamics analysis.

Peptide Library Design

We generated a library of approximately 400 unique peptides derived from the helical domain of TDP-43, with a length of 50 residues each, using the wild-type sequence [Uniprot (Q13148)]. Mutations were introduced at residue S332 or S333 in this region to generate 400 unique peptides. We utilized RaptorX (http://raptorx.uchicago.edu/StructurePropertyPred/predict/) to predict the secondary structure properties of the peptides.⁵² RaptorX Property is an advanced web server that can predict structural properties of a protein sequence without relying on any template information. This server excels in performance, particularly for proteins that lack closely related sequences in the Protein Data Bank (PDB) or have limited sequence profile information available. The peptides were then ranked based on their helical propensity, and this list was refined further based on the helical penalty score⁵³ and disorder prediction⁵⁴ (Tables S1 and S2). The five top-ranked peptides with the highest helical propensity were selected for further analysis.

Ab Initio 3D Structure Prediction and Preparations

To validate sequence-based secondary structural properties, the top five peptide hits with the highest helical propensity were selected for 3D structure prediction and further analysis. Peptide structures were predicted using comparative modeling with Rosetta comparative modeling (RosettaCM method).⁵⁶ The wild-type TDP-43 amyloidogenic core region solution structure (PDB ID: 2N3X)⁴² was used as a template for the comparative modeling. This region spans 50 amino acids and encompasses residues 311–360 of the TDP-43 C-terminal LCD. The residues from 311 to 320 (MNFGAFSINP) and 344 to 360 (QNQSGPSGNNQNQGNMQ) are disorder coils, and residues 321–330 (AMMAAAQAAL) and 335–343 (GMMGMLASQ) are known as conserved helical regions. The predicted peptide structures were prepared for computational approaches using the Preparation Wizard of Schrödinger as described.⁶²

Docking Studies

Docking analysis was undertaken with the top five peptide candidates and amyloid-like filament structures (PDB ID: 7KWZ, 7PY2, and 6N37) (Figure S3B). This study was facilitated using the locally installed HDOCK software package following standard procedures.⁶⁰ Compared to other molecular docking tools, the HDOCK server is a comprehensive tool for protein–protein docking that integrates homology search, template-based modeling, structure prediction, and macromolecular docking functions.⁶⁰ It uniquely allows for amino acid sequences as inputs and employs a hybrid docking algorithm combining template-based and template-free docking. This hybrid approach integrates experimental data about protein–protein binding sites and small-angle X-ray scattering data. Compared to other docking tools such as HADDOCK^63,64 and Patch dock,⁶⁵ HDOCK is adept at handling novel peptides and mutated sequences where limited structural information is available.⁶⁰ Therefore, HDOCK was well-suited for our experimental approach. Based on the positions of peptides, docking scores, and favorable interactions with amyloid-like fibrils, we selected the top complex poses for molecular dynamics simulation.

Molecular Dynamics Simulation

We performed molecular dynamics simulations utilizing GROMACS 2021.3 complemented by PLUMED 2 for advanced analysis.^66,67 We considered numerous studies comparing the effects of different force fields on simulating monomeric and fibrillar forms of aggregation prone proteins.⁶⁸⁻⁷² Among the primary force fields frequently used in disordered protein and fibril experiments^68,71—AMBER99SB-ILDN, CHARMM22, and GROMOS43a1⁶⁹⁻⁷² —we adopted the GROMOS43a1 force field based on several factors: (i) GROMOS43a1, similar to AMBER99SBILDN, has demonstrated exceptional stability in maintaining fibril structures, a crucial aspect for our study. This stability is in contrast to force fields like CHARMM36M, which have shown tendencies to allow dissociation of edge chains in fibril systems.⁶⁹ (ii) The capability of GROMOS43a1 to accurately simulate intramolecular CH-π interactions was a significant factor in its selection.^69,73,74 These interactions are vital in the transition from monomeric to fibrillar forms, which is a key process in understanding helical propensity peptide conformation.⁷⁵ (iii) GROMOS43a1 shows a propensity for supporting β-strand-like structures.^70,71 Taking into account these factors, we selected the GROMOS43a1 force field for our protein modeling and simulation work.⁶⁸⁻⁷² Water molecules were represented using the SPC model.^76,77 Our simulations focused on both individual monomers and complexes, including fibrils and monomers. Each was centrally placed in a truncated octahedral box, maintaining a minimum distance of 1.2 nm from the protein to the box wall. The box, with its narrowest cross section being 8.09 nm, was populated with water molecules; Na⁺ ions and NaCl were introduced to replicate physiological ionic conditions at 150 mM. The systems underwent initial relaxation through the steepest descent method. This was followed by NVT and NPT equilibrations maintained at 310 K and 1 bar, employing velocity-rescaling and the Parrinello–Rahman methods.⁷⁸ We constrained bond lengths using LINCS⁷⁹ for proteins and SETTLE⁸⁰ for water molecules, applying a 2 fs integration step. Both electrostatic and van der Waals interactions were computed with a 1.2 nm cutoff utilizing the PME method.⁸¹ For comprehensive analysis, we conducted two sets of simulations for the monomer and the complex, each spanning durations of 100 and 300 ns.

Analysis Methods

The root mean square deviation (RMSD), root mean square fluctuation (RMSF), and secondary structural properties were calculated from trajectory data. To analyze the free energy landscape (FEL) and generate 3D figures, the following steps were employed. First, the structural coordinates, including RMSD, radius of gyration (RG), helical probability, and β sheet probability, were extracted from trajectory data. Next, input files for sham analysis were generated using the “sham.pl” Perl script. The sham analysis was then performed using the “gmx sham” command, which produced the free energy landscape stored in the “free-energy-landscape.xpm” file. To further analyze, the “FEL.xpm” file was converted to a text format using the “xpm2txt.py” Python script. Finally, Mathematica 12.1 and GraphPad Prism were used to visualize the FEL, allowing for the generation of informative 3D figures. These methods provided a robust approach for investigating the free energy landscape and visualizing the results for comprehensive analysis. To predict the intramolecular interaction energy, we used the interaction energy matrix tool.⁵⁷ Multiple structural conformations from various snapshots were retrieved from molecular dynamics trajectory data and used for predicting intramolecular interaction energies. Statistical tests were performed in GraphPad Prism. A two-tailed t test (for normally distributed data) or Mann–Whitney U test (for non-normally distributed data) was used to compare two groups. One-way analysis of variation with Tukey’s post hoc test was used to compare multiple groups of normally distributed data. The Kruskal–Wallis test with Dunn’s correction was performed to compare multiple groups of non-normally distributed data.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acschemneuro.3c00659.

• Overall helical probability for wild-type and other helical propensity residues (311–360 AA); docking poses of the S332A/S333W variant with the 7KWZ protein structure and crystal structures employed in the molecular docking; and predicted secondary structural properties through sequences and molecular dynamics (PDF)

Author Contributions

M.R.S. and P.P.G. designed the project, analyzed the results, and wrote the manuscript. M.R.S. carried out the experiments.

This research was supported by the National Institute of Neurological Disorders and Stroke/NIH under Award R01NS122907 (to P.G.).

The authors declare no competing financial interest.