Early-Stage Oligomerization of Prion-like Polypeptides Reveals the Molecular Mechanism of Amyloid-Disrupting Capacity by Proline Residues

Abstract

Proline cis/trans isomerization governs protein local conformational changes via its local mechanical rigidity. The amyloid-disrupting capacity of proline is widely acknowledged; however, the molecular mechanism is still not clear. To understand how proline residues in polypeptide chains influence amyloid propensity, we study several truncated sequences of the TDP-43 C-terminal region (287–322) and their triple proline variants (³⁰⁸PPP³¹⁰). We use coarse-grained molecular simulation to study the time evolution of the process of aggregation in the early stages in an effective high-concentration condition (∼25 mM). This ensures the long time scales for protein association at laboratory concentrations. We use several experimentally determined structure templates as initial structures of monomer conformations. We carry out oligomer size analysis and cluster analysis, along with several structural measures, to characterize the size distributions of oligomers and their morphological/structural properties. We show that average oligomer size is not a good indicator of amyloid propensity. Structural order and/or morphological properties are better alternatives. We show that proline variants can efficiently maintain the formation of large “ordered” oligomers of shorter truncated sequences, i.e., 307–322. This “order” maintenance is weakened when using longer truncated sequences (i.e., 287–322), leading to the formation of “disordered” oligomers. From an energy trade-off perspective, if the entropic effect is weak (short sequence length), the shape-complementarity of proline variants effectively guides the oligomerization process to form “ordered” oligomer intermediates. This leads to a distinct aggregation pathway that promotes amyloid formation (on-pathway). Strong entropic effects (long sequence length), however, would cause the formation of “disordered” oligomers. This in turn will suppress amyloid formation (off-pathway). The proline shape-complementary effects provide a guided morphological restraint to facilitate the pathways of amyloid formation. Our study supports the importance of structure-based kinetic heterogeneity of prion-like sequence fragments in driving different aggregation pathways. This work sheds light on the role of morphological and structural order of early-stage oligomeric species in regulating amyloid-disrupting capacity by prolines.

Introduction

Tar DNA-binding protein 43 (TDP-43, 414 a.a.) is an amyloid protein that participates in forming amyloids in spinal motor neurons of amyotrophic lateral sclerosis (ALS) patients.¹ TDP-43 contains a low-complexity (LC) sequence at the C-terminus (266–414 a.a.). The TDP-43 C-ter is primarily Q/N-rich and therefore prion-like. Structurally, the C-terminus is intrinsically disordered in its monomeric form, thus prone to aggregation. For example, researchers have studied several truncated TDP-43 C-terminal fragments that are responsible for the formation of pathogenic aggregates, gels, and other membrane-less assemblies.^2,3 Interestingly, mutations of these fragments can turn pathogenic fibrillar aggregates into functional amorphous aggregates, suggesting the role of the LC domain in mediating the formation of different aggregates and their conversion.⁴⁻⁶ In particular, the reversibility between pathogenic fibrils and amorphous, hydrogel-like forms can be quantified using a steric zipper-vs-larks model for short peptide segments ranging from only 6 to 11 a.a.³ This finding implies the presence of a sequence minimal size effect that plays a role in modulating the core process of aggregation. It is very likely that the low-complexity nature of the TDP-43 C-ter sequence drives a guided search for certain types of oligomer species, presumably through the modulation of nonspecific molecular interactions, such as hydrophobic contacts and volume-excluded effects of the sequence itself.

Specific mutation effects, on the other hand, can also play an important role in this regard. For example, proline has long been considered an amyloid disruptor;^7,8 however, whether it behaves as a disruptor or fibril stabilizing role depends on the actual microenvironment in forming amyloids, i.e., the local physicochemical contexts in the sequence. In some cases, proline mutants do inhibit amyloid formation, but there are also cases where the presence of trans-prolines helps stabilize the fibril structure.⁹ Removing proline, however, suppresses fibril formation in many cases, presumably due to the disruption of the complementary shape between neighboring chains in the fibril.¹⁰ Prolines mostly were found to be in the “trans” form in fibrillar aggregates of most amyloid proteins, except for tau protein.^11,12 Despite a relatively small population in general, the fraction of the “cis” form was found to be correlated with the size of the oligomer formed.¹³ Overall, the isomerization effects on aggregation depend on the context of the mechanochemical structure in the fibril.^14,15 The terminal effect of the proline-rich domain (PRD), for example, suppresses the mechanical stability of amyloid fibril.¹⁶ The change in the stability is belived to be due to its high PPII propensity.¹⁷ Interestingly, a length-dependence of oligoprolines on polyglutamine (P10 and P3) aggregation was also observed both in experimental¹⁸ and computational¹⁹ studies. These studies report a clear stoichiometric relationship between the sequence length and the critical nucleus size, suggesting the role of sequence length in the oligomer size.⁶⁰ To further the effect of proline mutation on aggregation, recent studies on TDP-43 C-ter truncated sequence (D1: ²⁸⁷GGFGNSRGGGAGLGNNQGSNMGGGMNFGAFSINPAM³²²) compared the aggregation behavior of two proline variants where the same triple proline mutation (³⁰⁸GGG³¹⁰ to PPP) was conducted in the sequence of different truncated forms. They observed that the high amyloid propensity of the D1-core sequence (MGGGMNFGAFSINPAM) remains intact with the presence of the PPP mutation (MPPPMNFGAFSINPAM).²⁰ However, the same PPP mutation being applied to the longer D1 sequence (GGFGNSRGGGAGLGNNQGSNMPPPMNFGAFSINPAM) completely disrupted amyloidogenesis and the overall aggregate morphology became amorphous.⁷ In other words, the conformational rigidity of proline clearly yields different aggregate outcomes in different sequence contexts. This proline-mediated aggregation phenomenon seems to be generic across different amyloid peptides, but the detailed mechanism is still not clear. The binding-related physicochemical interactions between prolines in different polypeptide chains as well as a complex shape-complementary effect make it a nontrivial problem to be addressed.

The effect of proline on protein conformation and folding is well-documented. The cis–trans isomerization of proline provides a unique structural constraint that determines not only equilibrium properties but also rate-limiting kinetic behavior in the process of folding–unfolding.²¹ The influence of such conformational change on protein binding kinetics is found to be comprehensive; it changes the rate of binding, the kinetic pathway, as well as the global geometry.²² Recent studies have also shown that proline cis–trans isomerization plays a key role in dimerization of cyclopeptides,²³ amyloid aggregation,^23,24 and the development of antibody biotheurapeutics.²⁵ These studies all highlight the effect of torsional conformational constraints on the shape-complementarity between protein partners, which is able to propagate to a protein global property change via mechanical coupling, thereby modulating binding thermodynamics and kinetics. In spite of the role of proline conformational rigidity in determining protein’s functional activities, the molecular mechanism underlying the phenomenon is not yet clear. Very often, researchers’ explanations contradict each other in different experimental contexts. For example, in the regulation of phase behavior, a synthetic proline-rich peptide with all-cis prolines was found to facilitate more compact structures than that with all-trans prolines.²⁶ This result implied a better packing role of cis prolines within intramolecular interactions. However, when forming a protein binding complex, mutation with trans-proline might provide a less-fluctuated and smaller dimension structural complex compared to that with a cis-proline mutation.²² Similar complexity can also be observed in cases where proline isomerization determines molecular chaperone activity in a nontrivial, dual functionality manner.¹⁵ Because of this, a unifying picture of the effect of proline is required in order to resolve seemingly contradictory explanations. Recent studies using model sequence offer a means for quantitatively assessing the preferable effects of trans-proline mutation in simple oligomerization processes, such as dimerization.²³ However, there is still a gap to be filled regarding how single- or multiple-proline mutation affects complex aggregation processes.²⁷ It is still challenging for experimental studies alone to single out an unequivocal pathway.

From the energy landscape perspective, there are several specific pathways in the landscape where only monomers in aggregation-prone conformations would actively bump into each other to form relevant oligomers, e.g., dimer, trimer, and tetramer. These aggregation-prone conformations usually are energetically not stable and meanwhile beta-strand-rich in their monomeric form; they are structurally different from those with the lowest energy and therefore most probable. Energetically favorable monomer conformations involve only the formation of disordered aggregates; therefore, they are not of interest here, although some ordered oligomer aggregates may form via subsequent conformational conversion, i.e., off pathways. To specifically study the effect of an active conformation or misfolded conformation in early-stage aggregation requires several additional treatments in order to enhance the population of the active species in experiments, for example, using denaturing agents. Molecular simulations, in this regard, are useful for studying putative systems where important factors can be systematically controlled one at a time.

The objective of the study is to develop a working simulation protocol that allows a quantitative analysis of sequence mutational effects on aggregation propensities in the early stages of aggregation. We use AWSEM-IDP,²⁸ a coarse-grained protein force field with the enhancement in protein’s intrinsically disordered properties, to study early-stage processes of TDP-43 aggregation. The aggregation propensity may depend on a range of factors at different stages of aggregation in terms of different time scales. We narrow the time scale of interest to be explored to an extremely early stage, where instantaneous confirmations of a monomer still couple with the earliest peptide-binding events (e.g., contacts). That is to say, the conformation of a monomer does not have sufficient time to sample its conformations before having the very first few contacts with any other peptides—memory effect. In this work, we set up an extremely high initial peptide concentration in our simulations. By doing so, we intentionally remove the influence of the distribution of the monomer structural ensemble in the process of early-stage aggregation. This approach allows us to look into aggregation pathways on a truly “structure-driven” basis.

Methods

Sequences and Structures of the TDP-43 C-Terminal Prion-like Domain

The D1 sequence (287–322) is one of the four sequence fragments (D1–D4) in the TDP-43 low-complexity domain (LCD). This domain is also considered “prion-like”. The sequences used in the present study are listed in Table 1. The structures of these sequences were taken from the PDB website. The cryoEM amyloid structures 7KWZ(29) and 6N3C(2) were used; these structures represent two different fibril polymorphic forms. Out of the structural templates studied, 6N3C exhibits a distinct solvation property which is considered to be pathologically relevant.² In addition, a fragment-based structure using homology modeling (MODELLER 9.22),³⁰ was used (using both 2N2C and 6N3C as the structural templates). 2N2C is a membrane-embedded structure in its monomeric form. The PDB IDs used, along with their structures in cartoon diagrams, are summarized in Table 2.

Table 1. C-Terminal Truncated Sequences of TDP-43 and Their Proline Variants Used in This Work.

TDP-43 C-ter fragments	sequence
D1 (287–322)	287GGFGNSRGGGAGLGNNQGSNMGGGMNFGAFSINPAM322
D1 (GGG308PPP) or D1-G3308P3 in short	287GGFGNSRGGGAGLGNNQGSNMPPPMNFGAFSINPAM322
D1-core (307–322)	307MGGGMNFGAFSINPAM322
D1-core (GGG308PPP) or D1-core-G3308P3 in short	307MPPPMNFGAFSINPAM322

Table 2. Structure Templates of TDP-43 C-Terminal Fragments Used in This Work.

^aThe protein conformation (double helix formed by a single polypeptide chain) is obtained from simulated annealing of a single D1 polypeptide chain. Note that the color scheme used is the same as in Figure 1a; the black color represents the sequence that is not used in this work.

Coarse-Grained Molecular Dynamics Simulation

In this study, we use the AWSEM-IDP (AWSEM-Intrinsically Disordered Protein) coarse-grained protein force field²⁸ to study the process of aggregation of TDP-43 C-terminal fragments. AWSEM-IDP is an IDP-enhanced version of the AWSEM (Associative-Memory, Water-Mediated, Structure and Energy Model),³¹ built on the LAMMPS simulation platform. In the AWSEM force field, each amino acid residue is represented by three beads: C-alpha, C-beta, and O atoms, while their respective residues maintain an ideal peptide bond geometry. This three-atom per residue scheme allows efficient simulations for an accurate description of protein folding dynamics, structure prediction,³¹⁻³³ and protein aggregation.^19,34,35 In addition, the energy function of AWSEM is featured by its transferable and physically motivated potentials that embody complex physicochemical properties of residues, for example, hydrophobicity, electrostatic interactions, and specific contacts, in the context of protein secondary and tertiary interactions. These physicochemical interactions are exerted on C-beta atoms. Thus, the enthalpic contributions of protein–protein contacts are implicitly included in the AWSEM contact energies (including proline residues), which are divided into protein-medicated and water-mediated interactions. In simulating the structural properties of IDP, both the bioinformatic fragment memory energy term and the radius of gyration (R_g) biasing potential term were employed; see the section below. Proline residues were set to be in trans-form throughout the study.

The structures of the TDP-43 C-terminal sequence fragment (the D1 sequence, 36 a.a.) and its truncated core sequence (the D1-core sequence, 16 a.a.) were prepared using the VMD program.³⁶ For each of the sequences, we carried out two sets of simulations; one set of simulations was carried out in a simulation box with its initial protein concentration ∼25 mM (2 × 2 × 3 = 12 monomers; calculated based on a virtual box size of 80 × 80 × 120 Å, Figure 1b), and the other set was carried out in much lower concentration (300 μM) with the box size (349 × 349 × 524 Å). The periodic boundary condition (PBC) was used. In each set of simulations, 10 independent simulation trajectories were conducted. In the high-concentration condition, the simulation trajectory was run for 11 million simulation time steps, which yields 11 000 frames (output 1 frame every 1000 time steps). In the low-concentration condition, the simulation trajectory was run for 61 million simulation time steps, which yields 61 000 frames.

Figure 1.

Simulated early-stage aggregation kinetic profiles of the TDP43 C-terminal D1 sequence (287–322) and the D1-core sequence (307–322) are compared. (a) Two fibril polymorphic structures of the truncated TDP-43 C-terminal sequences, both taken from cryoEM (PDB ID: 7KWZ and 6N3C), are used as the structure template in the simulations. Note that the D1-core (307–322) sequence is shown in red. The blue surface area represents the location of the three consecutive glycine residues GGG (308–310) along the sequence. The sequence information is shown below using the same color scheme. (b) The simulation box contains a total of 12 monomers (2 × 3 × 2), which yields the concentration of ∼25 mM. (c) The average oligomer size of the D1/D1 core (solid blue line/dotted black line) is shown as a function of simulation time steps (7KWZ is used as the initial structure). The error bar is obtained by averaging over 10 independent simulation trajectories. (d) The final equilibrated, average oligomer sizes of D1 and D1-core are compared using different structure templates. These structure templates include 7KWZ, 6N3C, 2N2C + 6N3C (homology modeling), a simulated annealing structure of 7KWZ, and a dual set of 7KWZ and 6N3C (dual template). Their aggregation shows different propensities; they are classified into heterogeneous (Hetero.) and homogeneous (Homo.), accordingly. Note that the template of 2N2C + 6N3C is obtained using a homology modeling method (see Methods), while the making of the dual template is described in the main text.

In addition to AWSEM-IDP, we also performed MD simulations using standard AWSEM. We compared the simulation results of the two versions and summarized their structural analyses in the Supporting Information (Tables S1–S5).

IDP Fragment Memories and Radius of Gyration Biasing

The conformational dynamics of IDPs were simulated using AWSEM-IDP, developed by Wu et al.²⁸ In the IDP protocol, the IDP structural dynamics were achieved by using a “fragment memory library” and the protein’s radius of gyration as the local structural biasing and the global geometric restraint, respectively. The fragment memory structures were obtained from fully atomistic simulations using the NAMD 2.14 molecular dynamics simulation package³⁷ with the CHARMM36m force field (specifically for IDPs).³⁸ When generating fragment memories, D1 (287–322) and D1 (GGG308PPP), D1 core (307–322), and D1 core (GGG308PPP) in their monomeric form were prepared using VMD³⁶ and put in a simulation box of appropriate size, e.g., 43.1 × 43.2 × 42.5 Å for D1 (287–322). Each of the simulation boxes encompasses explicit water molecules and chlorine counterions (0.026 M for D1 (287–322) and 0.033 M for D1 (GGG308PPP)) so as to neutralize the total charge of the system (net charge = 0). All the simulations were carried out using a 2 fs time step. Periodic boundary conditions were used along with the particle mesh Ewald method for electrostatics calculation. The interaction cutoff was set to be 12 Å.

To ensure good quality of sampling, simulated annealing was scheduled as follows. First, the temperature of the system was set to be 500 K for a 4 ns simulation. The temperature of the system was then decreased by 50 K, and an additional 4 ns simulation was performed for structural relaxation purposes. We repeated the same 50 K decreasing steps until the system temperature reached 300 K. Finally, the simulation was further equilibrated at 300 K for 4 ns. The structure of the last snapshot in the simulation trajectory was used to represent one single fragment memory. A total of 50 independent simulations were carried out, which allowed us to collect a total of 50 fragment memories. These fragment memory structures, treated as equal weights, were then used to represent the structural diversity of the IDP.

In global structural modulation, we used a biasing potential in terms of radius of gyration²⁸ to regulate the extent of the morphological collapse of polypeptide chains.

1

R_g⁰ is the desired value of R_g to be biased, which quantifies the extent of being ellipsoid-like relative to a sphere. The value of R_g⁰ was calculated using the average R_g over 50 fragment memory structures obtained from the previous simulated annealing processes. N refers to the number of residues of the polypeptide chain; α and β are parameters used to modulate the width of the potential. More details about the potential can be found in the original paper by Wu et al.²⁸ These R_g⁰ values were 11.30, 12.97, 9.28, and 9.27 Å for D1 (287–322), D1 (GGG308PPP), D1 core (307–322), and D1 core (GGG308PPP), respectively. Note that R_g is defined as

2

where m_i is the mass of the i^th C-alpha atom and Δr_i refers to the distance between the i^th C-alpha atom and the center of mass of the protein molecule. Experimentally, R_g can be determined using spectroscopic techniques such as small-angle X-ray scattering (SAXS) and fluorescence resonance energy transfer (FRET). The AWSEM-IDP version has been used successfully used to investigate functional mechanisms of IDPs³⁹ and binding dynamics of a disordered linker with a chromosomal particle.⁴⁰

Oligomer Size Analysis

In analyzing the size of oligomers, we used the same algorithm as in the previous work.³⁴ An oligomer was identified if two monomers were in the proximity of 9.5 Å and so on and so forth. We specifically tracked the time evolution of oligomer size for each trajectory. For each output frame of a given simulation trajectory, we recorded the largest oligomer size formed and used it to represent the oligomer size of that particular frame. This frame-based analysis allows us to directly calculate the oligomer size as a function of time—the kinetic trajectory of oligomer size (see Figures S1 and S2 for the time evolution of the first, second, and third largest oligomers in the Supporting Information). The corresponding averaged kinetic curve was obtained by averaging over 10 simulation replicates. The distribution plot for the oligomer size was made using a specified range of data, spanning from 40 × 10⁵ to 110 × 10⁵ simulation time steps) where fluctuation of oligomer sizes is stabilized, reaching an equilibrium condition.

Structural Clustering Analysis

To quantify the differences and similarities of the structures of D1/D1-core and their variants, we have applied clustering analyses to subsets of structures from the simulations. We use mutual Q as a distance measure between two snapshots. Q refers to the fraction of the contacts that is the same between the two structures. The clustering analyses are carried out separately for the D1 sequence, its truncated form D1-core, and their proline variants, D1/D1-core/D1(G₃308P₃)/D1-core(G₃308P₃). Before doing clustering analysis, we need to preprocess our simulation data. For each sequence, we generate a set of structures for clustering. We select 1000 structures from the simulation trajectories by uniformly extracting 100 frames from each trajectory; we have a total of 10 simulation trajectories. This yields a total of 1000 frames. We then classify all 1000 frames into different groups according to their oligomer sizes: dimer, trimer, tetramer, up to decamer (if any). Note that the oligomer size is determined by the largest oligomer size of each frame (see previous section Oligomer Size Analysis and Figures S1 and S2 in the Supporting Information). There are N_i members collected in each group (i refers to the size of the group). In the process of clustering, we first construct a Q matrix (N_i × N_i) where each element refers to the pairwise mutual Q calculated for any two of the structures of the members within the group. We then performed hierarchical clustering for each group individually and identified the population (see Table S1 in the Supporting Information) and the centroid structures of each cluster (see Tables S2 and S3 in the Supporting Information). Since the specific chain order for two structures of the same size (e.g., both are trimers) may affect the Q value, all permutations of the chain order for each structure are used for the calculation. However, we find that this permutation effect is not significant in our simulations, possibly due to high-concentration conditions, in which oligomers mainly originate from specific simulated trajectories.

Measure of Excluded Distance, D̅

The distance, D̅, defined as the average of pairwise distance between the C_α (C-alpha) carbon of residue 309 in different polypeptide chains of a given oligomer, is calculated and shown as a function of the cluster database with different oligomer sizes (from 2mer to 12mer) at different substrate templates.

3

The N is the number of members in the size group b, and i, j refer to the chain indices of the oligomer. R_i^C_α and R_j represent the position vectors of C_α of residue 309 of chains i and j, respectively.

Nematic Order Parameter to Quantify Structural Order

To characterize the population of a key oligomeric intermediate, one needs to choose a structural measure that best reflects the conformational change criteria of oligomeric species of the same size. When an oligomer undergoes structural transition, it changes orientation of individual monomers while retaining its stacking morphology as it was. This is similar to anisotropic fluids such as liquid crystals. The nematic order parameter has been used to describe the orientation of liquid crystals. Caflisch et al. first used it to identify the oligomer intermediate of short peptides.⁴¹ It was also used to quantify the oligomer species in secondary nucleation.⁴²

The order parameter Q₀ is used to quantify the structural order of a system. A value between 0 and 1 describes the orientational order/disorder with “0” indicating a completely isotropic system and “1” representing a perfect ordering of an anisotropic system. The analysis of nematic order parameter (Q₀) was calculated using the compute_nematic_order function of MDTraj (version 1.98).⁴³ The C-alpha carbons were used to compute unit vectors of individual polypeptide chains. The result is shown as a function of oligomer size (from 2mer to 12mer).

Results

Aggregation of TDP-43 C-Terminal Disorder Region (287–322) Shows Oligomer Size Dependence on Different Truncated Sequences

The TDP-43 C-terminal sequence fragment 287–322 is considered a disorder region, as has been predicted using different algorithms (see Figure 1B in the reference indicated).⁴⁴ The typical experimentally observed lag time for protein aggregation kinetics ranges from ∼seconds to hours,⁴⁵ depending on the concentration regime explored. To computationally access the long time scales for association at laboratory concentrations,⁴⁶ we carry out molecular simulation at a high protein concentration (∼25 mM) and systematically monitor the time evolution of the formation of oligomers. Figure 1a shows the initial conformation of the sequence used in the simulation (PDB ID: 7KWZ). 6N3C is also shown below for comparison (see Methods for details). The initial configuration of peptides in the simulation box is schematically shown in Figure 1b. Using the template 7KWZ as the initial conformation, we calculate the average oligomer size as a function of time; the average is taken over 10 independent simulation trajectories. Figure 1c compares such an average kinetic profile of the D1 sequence (287–322) and its truncated form, the D1-core sequence (307–322). The high initial peptide concentration ensures significant protein–protein interaction as the simulation begins, as evidenced by a burst phase observed in their kinetic profiles. The burst phase is followed by a kinetic plateau, meaning the process of aggregation is directed toward an equilibrium. When the system reaches an equilibrium, the average oligomer size of D1 is 4–5, while the average oligomer size of D1-core decreases to 3 (see Figure 1c). This result indicates that D1-core has a slightly inhibitory effect on the aggregate size, as compared to D1.

However, we find that this sequence truncation effect might have a diverse influence on the oligomer size, depending on the initial conditions of the simulation, specifically, different initial structures used in the simulation. Figure 1d compares the final equilibrated average oligomer sizes of D1 and D1-core when different structure templates are used. The results show that the sequence truncation effect varies from template to template. Table 2 lists all the structure templates available in this work, including their exact sequences. For example, using a different structure template (PDB ID: 6N3C), the D1 to D1-core truncation effect on aggregation becomes promotive. The average oligomer size significantly increases from 4–5 to 9, highlighting the propensity of forming larger oligomers. Moreover, using a somewhat modified structure template obtained via homology modeling, combining both 2N2C and 6N3C templates yields a similar inhibitory effect as the first case, i.e., 7KWZ. Please see Figure S3 in the Supporting Information for the individual aggregation kinetic profiles using other templates. All these results suggest the critical role of initial structures in the aggregation kinetic profile in the early stages of aggregation. This phenomenon is exemplified as the “memory effect”, which is enhanced when the protein concentration is high. In this high-concentration regime, the instantaneous molecular contacts, arising from the initial structure of polypeptide chains, are rather heterogeneous, which overall result in diverse oligomer size distribution. As a result, the aggregation propensity becomes highly heterogeneous.

In contrast to the structure templates directly obtained from fibrils, we also carry out a different set of simulations using the initial structure taken from the most populated monomer conformation. The most populated monomer conformation is obtained by running an independent simulated annealing schedule for a single D1/D1-core (see Table 2 for the structure). This annealing set provides a reference value of the average oligomer size (∼3 both for D1 and D1-core). In this case, the heterogeneous interactions due to monomer structural relaxation are removed. In fact, the onset of disease-related aggregation concerns monomers in non-native conformations, usually energetically high and therefore less populated species. This justifies the use of high protein concentrations in the present study. Another way of removing the heterogeneous effect is to generate a hypothetical system with nonbiased configurations using a dual set of structure templates (dual template). For example, one can distribute a total of 12 monomers (six monomers with 7KWZ and the other six monomers with 6N3C) in the simulation box. To ensure compositional homogeneity, they are placed in an alternating fashion in all directions, such as 7KWZ, 6N3C, 7KWZ...in the x-, y-, and z-directions. As expected, the result shown in Figure 1d reports an roughly equal value of the average oligomer size ∼4.5, indicating there is no significant difference in the average oligomer size between the D1 and the D1-core.

Secondary Structure Content of Monomers Determines the Effect of Sequence Truncation on Oligomer Size and Its Distribution

We have observed the oligomer size dependence on different truncated sequences (D1 vs D1-core). To understand the oligomer size distribution in great detail, we carry out the size distribution analysis. Figure 2 compares the average oligomer size and distribution of the D1 and the D1-core sequences. Using the 7KWZ structure template as the initial structure, the result shows that the average oligomer size is 4–5 and 3 for D1 and D1-core, respectively (see Figure 2a). The difference in the average size results in a population shift in the distribution (see Figure 2b). For example, D1 tends to form a large size of oligomers (up to 7mer), while D1-core mainly forms smaller oligomers (less than 5mer) with the dimer as the most populated species. Figure 2c shows several representative centroid structures of the clusters of D1 (left) and D1-core (right). The D1 sequence tends to form specific structural elements, such as sheet structures or double helices (two strands wrapping around each other into a double helix) in small oligomers, e.g., dimers or trimers. When the size increases (e.g., a septamer), the resulting large aggregate becomes a mixture of both structural elements, more like an amorphous, disordered molecular assembly. On the other hand, the D1-core sequence prefers to form a double helix in a dimeric form. Multiple dimers then interact to form a large oligomer (i.e., tetramer); this process will be followed by a structural conversion into an ordered sheet structure as a pentamer. In other words, D1-core forms a large oligomer that is structurally ordered, i.e., the oligomerization process involves a clear disorder-to-order transition, whereas D1 tends to form large oligomers with multiple structural elements randomly stacked together; thereby, the structure overall is disordered. This finding indicates that different aggregation mechanisms are undertaken by D1 and D1-core. Structurally distinct molecular assemblies concerning different structural order are formed via different oligomerization pathways. The results clearly show that the variation in the size distribution provides molecular details that cannot be simply explained by the average oligomer size.

Figure 2.

Oligomer size distributions of D1 (287–322) and D1 core (307–322) are compared. (a–c) The 7KWZ structural template is used as the initial structure. The average kinetic profile (a), the size distribution (b), and the representative centroid structures of the clusters (c) of the D1 and the D1-core are compared. (d–f) The 6N3C structural template is used as the initial structure. The average kinetic profile (d), the size distribution (e), and the representative centroid structures of the clusters (f) of the D1 and the D1-core are compared. (g,h) The average kinetic profile and the corresponding size distribution, respectively, for the dual template.

To further quantify the structural change during time evolution, we track the time evolution of the secondary structure content of monomers using the DSSP (Dictionary of Secondary Structure of Proteins).⁶¹ Specifically, we look into monomer secondary structural preferences responsible for changes in size distribution. The fraction of beta-strand is shown as a function of time, as can be seen in Figure 2a (onto the other y-axis). As the oligomerization progresses, the beta-strand fraction curve saturates at a value that is slightly higher for the D1 sequence (∼0.1) than that of the D1-core sequence (∼0.07) (see Figure S4 for individual fractions of coil, strand, and helix in the Supporting Information). This result indicates a slightly different secondary structure content of the monomers not only can result in a different average oligomer size but also lead to a different size distribution as well as distinct aggregation structural features.

Using another structure template (PDB ID: 6N3C), the average kinetic profile is significantly different between D1 and D1-core. Figure 2d shows that the average oligomer size is 4–5 in the case of D1. However, the average oligomer size increases to 9 in the case of D1-core. The oligomer size distribution of D1 also exhibits a completely different profile, compared to that of the D1-core sequence. Figure 2e shows that the effect of sequence truncation (D1-core) causes a significant population shift of the distribution toward a unique distribution with a large average oligomer size (n = 9). D1 primarily tends to form small oligomers (up to 5mer, with 30% of 2mer, 5% of 3–5mer each, similar to the result of 7KWZ as described above), while the D1-core sequence generates a range of large oligomers (up to 12mer, with 20% of 2mer, 5–10% of 3–4mer each, 5% of 6–11mer each, and 10% of 12mer). This result indicates the significant role of the structure template used in modulating the effect of sequence truncation on aggregation. The structure template used might not affect the average oligomer size so much as in the case of the D1 sequence but can significantly influence the aggregation profile of the D1-core. The significant enhancement in forming large-size oligomers, presumably, is due to the high fraction of beta-strand in the D1-core sequence. The beta-strand fraction is ∼0.1 for D1; however, the fraction significantly increases to ∼0.3 for D1-core (see Figure S4 for individual fractions of coil, strand, and helix in the Supporting Information). Figure 2f shows the centroid structure of the clusters of D1 (left) and D1-core (right). The D1 sequence forms similar sheet structures and double helices in small oligomers, e.g., dimer or trimer. Their large oligomers (i.e., a pentamer) show similar amorphous structural features, as compared to 7KWZ. However, the truncated D1-core displays an extremely strong aggregation propensity. The oligomer size increases (up to 12mer). In particular, the structural motif almost always involves a “cylindrin-shaped” geometry in the oligomerization process. Similar prefibrillar structures also have been observed in the case of Abeta peptides.^34,47 This is reminiscent of disorder-to-order structural conversion, suggesting a functionally distinct aggregation pathway being carried out.

Finally, we examine the oligomer size distribution when the dual structure templates (7KWZ and 6N3C) are used. As has been seen in Figure 2g, also mentioned in the previous section, the dual set result exhibits no difference in the average oligomer size for both the D1 and the D1-core. This is because heterogeneous interactions are somehow partially removed. Interestingly, however, the size distribution of the D1-core shows a clear, gradual shift toward a population of larger oligomer sizes (see Figure 2h). This phenomenon arises from the reduction of heterogeneous interactions (via compositional homogeneity). From this result, we can see that aggregation propensity is better determined by the size distribution, rather than the average oligomer size. Experimentally, it is challenging to quantify such a heterogeneous effect. This may be one of the reasons for protein aggregation being stochastic in nature.⁴⁸ This series of analyses highlights the relationship between heterogeneous interactions of monomers and the resulting heterogeneous aggregation properties.

Proline Variants Can Modulate the Morphological and/or Structural Characteristics of Oligomers

Next, we explore the effect of proline mutation on aggregation using the 7KWZ template as an example. We specifically compare the aggregation profile of the D1 sequence and that of its proline variant D1-GGG308PPP (D1-G₃308P₃ in short). Figure 3a shows that the average oligomer size of D1 is n = 4–5, while the average oligomer size of its proline variant (D1-G₃308P₃) increases by ∼1 (n = 5–6). The size distribution of the proline variant shifts toward larger values (Figure 3b). Likewise, in the case of the D1-core sequence, Figure 3d shows that the average oligomer size is n = 3 while the average oligomer size of its proline variant (D1-core-G₃308P₃) increases by ∼1 (n = 4). The corresponding size distribution is shown in Figure 3e. The result indicates the generic effect of proline mutations on increasing oligomer size, independent of the different truncated sequences. However, if we study the structural features of the resulting oligomers of D1 and D1-core, they are quite different in terms of morphology and structural order. Figure 3c shows that disordered aggregates form as the oligomer size increases (6mer for D1 and 8mer for the D1 proline variant); their morphology is nearly spherical. In contrast, the D1-core oligomers display somewhat distinct structural characteristics. Figure 3f shows that as the oligomer size increases (from 5mer for D1-core to 7mer for its proline variant), the well-organized lamellar structures become more favorable, and their morphology becomes elongated in shape. This observation suggests that additional structural measures, rather than the oligomer size itself, are required to effectively quantify the morphological changes and changes in the structural order of the proline variants, which we will investigate further in the next section.

Figure 3.

Change in oligomer size distribution due to the proline mutation (GGG to PPP) is compared between the D1 sequence and the D1-core sequence, using the 7KWZ template as an example. (a–c) The D1 sequence (287–322) and its D1 proline variant. The averaged kinetic profile (a), the size distribution (b), and the representative centroid structures of the clusters (c) of D1 and its proline variant are compared. (d–f) The D1-core sequence (307–322) and its D1-core proline variant. The averaged kinetic profile (d), the size distribution (e), and the representative centroid structures of the clusters (f) of the D1-core and its proline variant are compared.

Proline Residues Tend to Be Mutually Excluded in Oligomers

To probe the structural changes due to the effect of proline mutations, we carry out a structural analysis that uses pairwise proline-to-proline distances between peptide chains within individual oligomers. Computing the average of these pairwise distances as a function of the oligomer size allows us to explore the relationship between this novel physical quantity, a measure of excluded distance (called D̅ hereafter), and the oligomer size. Figure 4a,b shows the distance-excluded measure as a function of the oligomer size for D1, D1-core, and their proline variants. The excluded distance D̅ is quantified by calculating the average of all the pairwise proline-to-proline (here residue 309 to residue 309) distances between polypeptide chains of a given oligomer. The value of D̅ represents the extent of distance separation due to changes in mutual PPP/GGG distance pairs between polypeptide chains in the aggregate. For the D1 (287–322) sequence, we observe a nearly monotonic increase in the value of D̅ as the oligomer size increases, from 16 Å (3mer–6mer) to 20 Å (>7mer). When the PPP mutations are introduced, the monotonic trend remains, while the value of D̅ becomes larger in general, from 10 Å (dimer), 20 Å (tetramer), to 25 Å (11mer). This result clearly shows that the distance-excluded effect caused by the proline variants is a general morphological property of the peptide sequence when forming oligomer aggregates, independent of the structure templates used and the sequence length. This finding implies there is a correlation between the local distribution of prolines in different chains and the size of the formed oligomers—an indicator of an increase in R_g. Figure 4c,d shows several representative centroid structures of oligomer of different sizes. Taking the pentamer as an example, we can clearly see that for the D1 sequence, it shows a rather compact shape (by looking at the mutual distance separation between the five prolines), while its proline variant displays a distinct pattern for the same five prolines (Figure 4c). This distance-excluded effect eventually can lead to significant changes in the global shape/morphology as a result (see the section below).

Figure 4.

Excluded distance effects of D1, D1-core, and their proline variants are compared as a function of the oligomer size. (a,c) D1 (287–322). (b,d) D1-core (307–322). The excluded distance measure, D̅, defined as the average of all pairwise proline-to-proline distances between different chains in the oligomer, is calculated and is shown as a function of the oligomer size (from 2mer to 12mer). In the calculation of the distances, the location of residue 309 is used, thus using the subscript 309–309. Triangles, squares, and circles represent the PDB IDs of the structure templates used, respectively: 7KWZ, 6N3C, and 2N2C + 6N3C. Gray color refers to the proline variants (···³⁰⁸PPP³¹⁰···), while purple color refers to the wild-type (i.e., ···³⁰⁸GGG³¹⁰···). The shaded area shows a range of variation in D̅ when different structure templates are used. The representative structures, the centroid structures of the selected two clusters, are shown in ribbon style (size 5 and size 7 in (c) and size 4 and size 7 in (d)). (top) The proline variant (···³⁰⁸PPP³¹⁰···). (bottom) The wild-type (···³⁰⁸GGG³¹⁰···). Note that the surface area represents the location of the three consecutive GGG/PPP residues.

Proline Variants Lead to Increased R_g of Oligomers

Based on our previous result on the distance-excluded effects by proline variants, we speculate that the overal shape of the oligomers would also change. It would be interesting to see if there exists such a correlation relationship. To achieve this, we specifically look into the radius of gyration (R_g) of the oligomers and calculate the R_g distribution. Figure 5 shows the comparison of the R_g distributions of the wild-type D1 and its proline variant of the pentamer. The R_g of the proline variant (∼16.5 Å), in general, is larger than that of the wild-type (∼15 Å), as shown in Figure 5a. This is generally true for the same R_g distribution plots obtained with other structure templates (see Figure S5a in the Supporting Information). The results clearly show that R_g of the pentamer becomes larger if D1 (···³⁰⁸GGG³¹⁰···) is mutated to D1 (···³⁰⁸PPP³¹⁰···). We further analyze the distribution of intermolecular number of contacts. Figure 5b shows that the wild-type has more contacts, ∼310, while the proline variant has less contacts, ∼260. The results show that the wild-type is energetically well-optimized, compared to that of the proline variant, regardless of the structure templates selected (see Figure S5b in the Supporting Information). This observation indicates steric repulsive effects between polypeptide chains, consistent with our previous result. Similar effects due to the presence of prolines were also observed in a recent single-molecule spectrospy experiment of an intrinsic disordered protein (NCBD with its binding partner ACTR).²² Our findings imply that proline mutation can generate long-range effects that perturb intermolecular interactions between chains in oligomers. Thus, it leads to overall shape changes and increased R_g. Figure 5c shows the general weak positive correlation of R_g against the number of contacts, as is observed in the same oligomer species. The Pearson correlation coefficient is 0.33 for the wild-type, while the value increases to 0.47 for the proline variant. The correlation is slightly higher, presumably, due to the long-range perturbative effects of proline conformational rigidity on intermolecular contacts.

Figure 5.

The distribution plots of the radius of gyration (R_g) and the number of intermolecular contacts of the pentamer are shown. (a) Radius of gyration. (b) Number of contacts (C_α–C_α < 6.5 Å). The probability distribution function (p.d.f.) of R_g is shown using histograms. Light red refers to the presence of proline residues, while light blue refers to the absence of proline residues (wild-type). The distribution in R_g is obtained by calculating R_g’s for all of the members in the pentamer cluster. The surface representation of the pentamer (wild-type, top-right) and its proline variant (middle-right) are shown on the sides, respectively (using 7KWZ). (c) The correlation plots of R_g against the number of contacts are shown for the wild-type (left) and the proline variant (right), respectively.

Proline Helps Maintain the Structural Order of Oligomers in Short Sequences

To gain more structural insight into the aggregation mechanism, we carry out structural order analyses for a range of oligomers. Specifically, we calculate the nematic order parameter Q̅₀ (see Methods for some details), which is used to quantify the orientational “order” and “disorder” of the oligomer aggregates studied. The value of Q̅₀ ranges from 0 to 1, which refers to disorder and order, respectively. Figure 6 presents Q̅₀ as a function of the oligomer size both for the D1 and the D1-core proline variants, using 7KWZ as an example (the results using other structure templates can be found in Figure S6 in the Supporting Information). By comparing the result from the D1 proline variant with that from the D1-core proline variant, we find that the D1 proline variant displays a significant lower value of Q̅₀ as the oligomer size increases, ∼0.5 for tetramer, ∼0.45 for pentamer, and <0.4 for hexamer and above. In contrast, the D1-core proline variant shows overall higher values of Q̅₀, ∼0.9 for dimer, ∼0.6 for trimer, and >0.55 for tetramer and above. The results show that the structural order is better maintained for D1-core-based oligomers as the size increases. From trimer to heptamer, the decreasing trend gradually reaches a plateau. The D1 proline variant also shows a similar trend but reaches a different plateau as the size increases. The effect of proline mutations on the structural order, therefore, are different in the cases of D1 and D1-core. Apparently, the presence of prolines helps D1-core to be more resistant to disorder when forming larger oligomers. This finding suggests that different mechanisms of amyloidogenesis may arise from different kinetic pathways involving formation of distinct types of oligomers. They can be classified using the concept of structural order. Moreover, the results also imply that prolines play a role in regulating the orientational order of the aggregates. The regulatory effect may be well-reflected in the sequence length. Similar sequence length effects have been employed to study “self-avoiding random walk” of polypeptides.⁴⁹ Our work provides a simulation evidence for extending the argument to explore the extent of orientational order of the oligomer aggregates.

Figure 6.

Nematic order parameter is used to quantify the structural order of D1/D1-core variants as a function of the oligomer size. The order parameter Q̅₀ of oligomers of the D1 proline variant are shown using black downward triangles, from tetramer to octamer, with a black solid line. The Q̅₀ of the D1-core proline variant is represented using gray downward triangles, from 2mer to 7mer, with a gray dashed line overlaid. The simulations use 7KWZ as an example.

Discussion

Distinct Aggregation Pathway Involving an Ordered Oligomer Intermediate Governs Amyloid Formation

The structural analysis of D1/D1-core formed oligomers carried out in this work allows us to look into the morphological changes, structures, as well as the compactness of the oligomer clusters. In particular, a direct comparison between the wild-type and its proline variant provides insights into many of the structural features above that can help distinguish different early-stage aggregation mechanisms. For example, D1/D1-core proline variants share similar mutually distance-excluded effects between proline residues of different chains in the oligomer aggregates. The distance-excluded effect is responsible for the overall shape change in the radius of gyration (R_g); the shape of the oligomer becomes ellipsoid-like both for D1 and D1-core. However, D1 and D1-core display a distinct structural property in terms of the structural order of their oligomers. This is an essential finding from the simulation because one can use the extent of “order” or “disorder” as a quantitative measure to classify a variety of oligomers. In general, a longer sequence, say D1, has a larger configurational entropy, the D1 sequence tends to form randomly disordered oligomers as its oligomer size increases. Since the contribution of the high entropy capacity of D1 outperforms the effect of proline mutation (which plays a major role in maintaining the structural order of the oligomer), the presence of proline mutations does not significantly change the structural features of the oligomer too much. Oligomers of the D1 proline variant thus exhibit a similar trend in their structural disorder as the oligomer size increases. However, when it comes to the D1-core, the presence of proline in the D1-core can help counteract the entropic effects created by the random stacking of oligomeric aggregates. We can say in this case that proline is more resistant to disorder as the size increases, presumably due to the shape-complementary effect by proline residues. From this perspective, the result is very useful for understanding the aggregation mechanism. There are many studies that have already pointed out different oligomerization pathways involving different types of oligomer intermediates. To classify them, one possible way is to use a measure of structural “order.” This is similar to the “order parameter” often used in condensed matter physics for order–disorder transition. Order parameter provides a quantitative measure for the structural order of molecular assemblies. Here, one plausible explanation is that the proline variants of D1 and D1-core in fact follow distinct aggregation pathways. The D1 proline variant undergoes an aggregation pathway involving amorphous oligomer intermediates, while the D1-core proline variant follows an alternative aggregation pathway involving more ordered oligomers formed. This argument can be used to explain the stronger amyloidogenic property of the D1-core. The presence of proline in D1 causes significant distance-excluded effects which hampers the conformational conversion of an amorphous oligomer into fibrillar oligomers, thus disrupting amyloidogenesis. However, the D1-core proline variant can better maintain the structural order of the formed oligomers due to the proline shape-complementarity; therefore, it does not alter much of the tendency for fibril formation. Figure 7 illustrates the mechanism of proline-mediated aggregation pathways.

Figure 7.

Schematic diagram for the proline-mediated aggregation pathway is shown. The proposed proline-mediated aggregation pathways help illustrate the mechanism of amyloid-disrupting capacity by prolines. The key mechanistic step in amyloid formation, the conformational conversion between prefibrillar oligomers to fibrillar oligomers, is highlighted. Formation of prefribrillar oligomers takes place in the early stages of aggregation. The structures of these D1/D1-core oligomers are determined by the energy trade-off between the configurational entropy of the sequence itself and the shape-complementary effect of prolines (if any). Whether the entropic effect is dominant would generate oligomers of different morphological and structural properties. The extent of the structural order of the resulting prefibrillar oligomer species correlates with the height of the energy barrier for the prefibrillar to fibrillar conformational conversion. The D1 sequence has high configurational entropy, which enhances the distance-excluded effect while depreciating proline shape-complementarity. Therefore, amyloid formation is suppressed due to the overall high energy barrier (in red). The D1-core sequence, however, has lower configurational entropy, which, along with the structural guidance provided by the proline shape-complementarity, facilitates the formation of “order” prefibrillar oligomer species. This also facilitates a distinct aggregation pathway featuring a lower energy barrier (in green), thereby promoting amyloid formation. The amyloid-disrupting capacity of prolines is attenuated as a result. Created with BioRender.com.

Double Helix of Polypeptide Chains as a New Structural Motif in Protein–Protein Interactions

The double helix is a natural structural characteristic observed in the structure of nucleic acids. A similar structural feature is rarely observed in polypeptides. However, there are some exceptions. Gramicidin, a peptide antibiotic, is one of the few cases that displays the double-helix feature.⁵⁰ In fact, forces that stabilize the double helix can be manifold, for example, synthetic molecular strands from the aromatic oligoamides,⁵¹ synthetic entwined covalent helical polymers via hydrogen bonding,⁵² and a double-helical assembly of heterodimeric nanoclusters driven by van der Waals interactions.⁵³ In the study, we also have observed similar double helices in the simulations; in particular, they are made of polypeptide chains. Interestingly, the double-helix motif displays a structural diversity when forming different types of molecular assemblies. For example, we have observed several aggregates with their conformations being a mixture of beta-sheet and double helix. When simulations are carried out using the IDP force field, we can see the formation of donut-like aggregates where multiple polypeptide chains interact with each other via cross-linking, a feature that propagates a network-like structure in a nonspecific way. We summarize the structural diversity in terms of large molecular aggregates that can possibly form (see Figure S7 in the Supporting Information). Most importantly, a polypeptide double helix, as a novel structural motif in protein–protein interactions, provides an opportunity for network cross-linking, a clue to the dynamic interaction and contact linkage among aggregates of different sizes and shapes. This dynamic interaction capacity by the double-helix motif can be the structural ingredient needed in liquid–liquid phase separation (LLPS). In fact, a recent computational study on nuclear body remodeling has revealed the behavior of phase separation in terms of nucleoli–chromatin interactions that induce nucleation due to the generic polymer nature of the chromatin network.⁵⁴ This network-based regulation of phase separation can be generic and transferable to our system, where the double-helix motif conforms to the structural diversity of forming a nucleoli-like assembly (cross-linking assembly) and chromatin network (dynamic long-range contacts). They exhibit structural diversity and the dynamic interaction needed for LLPS. The population of the double-helix motif and the like is not necessarily prevalent in liquid droplets but the structural elements in the maintenance of the outer boundary of droplets. Since the liquid boundary contains only a thin layer of peptides, the homogeneous interior contains major molecular components. Therefore, there are possibilities for bulk experimental studies not being able to detect the structure over the surface layer. Surface-enhanced experimental techniques are required to explore the detailed composition near the boundaries of liquid droplets.

Role of Proline Shape-Complementarity in Multiplex Framework and Its Implication in Cell Toxicity: An Experimental Proposal

We examine several structural polymorphs of TDP-43 fibrils and compare their local structural features at different mutation sites, bundling stoichiometry and cell toxicity. The result is summarized in Table 3. We find that the shape-complementarity of specific key sequence segments showing a turn or loop correlates with cell toxicity. This observation suggests that the formation of the shape-complementary motif in the protofibril is necessary for displaying cell toxicity. The SegB, R-shaped polymorphic structure of TDP-43 (286–331) (PDB ID 6N3C, cryoEM) shows a 4-fold bundled fibrillar topology with one specific bundling interface involving a loop-and-turn shape-complementarity (²⁹⁴GGG against ³⁰⁸GGG). The relevant solvation energy at the interface also exhibits additional stabilization.² In contrast to the other three polymorphs, which do not show such shape-complementarity, 6N3C exhibits clear cell toxicity. This result prompted us to carry out a survey over similarly shaped complementary motifs across different TDP-43 polymorphic structures. Likewise, a recently solved TDP-43 structure (PDB ID 7PY2, cryoEM) with cell toxicity also displays a similar turn-based shape-complementarity on (³⁰⁸GGG). Interestingly, in this case, the shape-complementarity takes place within the same protein chain. This suggests two different types of shape-complementary scenarios, one arising from within the intramolecular context and the other arising from within the intermolecular context, an analogy of the multiplex framework that was used to explain the topological beta-sheet architecture of the protein nucleus size.⁵⁵ We, therefore, bring this issue up for discussion and propose that the shape-complementarity to turn or loop in the polymorphic fibril structure is the key structural feature that correlates with the cell toxicity.

Table 3. Shape-Complementary Survey over TDP43 Polymorphic Fibrillar Structures.

PDB ID	toxicity	fold	shape comple. (S.C.)	inter. or intra.	proline	ref
6N3C	yes	4	yes	inter.	no	(2)
7PY2	yes	1	yes	intra.	320,340	(59)
7KWZ	-	1	-	-	280,320,349,363	(29)
6N3A	-	2	-	-	320,349	(2)
6N3B	-	2	-	-	320,349	(2)
6N37	-	2	-	-	320	(2)

To justify our hypothesis on the relationship between cell toxicity and shape-complementarity, here we propose an experimental protocol for evaluating the structure–toxicity relationship for proline-mediated amyloidogenesis. This protocol contains two experimental methods which are used to confirm the formation of a shape-complementarity and assess cell toxicity/infectivity, respectively. First, a cross-linker coupled with a mass spectrometer provides a rapid way to examine the structure of the shape-complementarity of interest. Second, for cell toxicity assay, protein misfolding cyclic amplification (PMCA) can be used to replicate the pathogenic agents, where the structure of the amyloid aggregates is akin to that found in the brains of patients. PMCA is a popular in vitro assay that can efficiently mimic the amyloidogenic toxicity/infectivity of in vivo experiments.^56,57 We anticipate that our predictive model can be well-confirmed using the experimental pipeline described above.

Conclusions

The prion-like properties of the TDP43 low-complexity domain (LCD) exhibit diverse structural properties such as amyloids, liquid phases, and gel-like liquid droplets. Their interconversions are reversible, suggesting potential therapeutic strategies to reverse pathogenic processes. However, the molecular mechanism regarding how amyloidogenesis is regulated is still not clear. We tackle the fundamental question of how the mechanical restraint of prolines disrupts amyloid formation. Understanding the amyloid-disrupting capacity of prolines requires detailed investigation in the early stages of aggregation, which is usually difficult to access via experimental methods alone. In this work, we use computational methods to investigate the aggregation of TDP-43 C-terminal sequence fragments. We have carried out coarse-grained molecular dynamics simulation using AWSEM-IDP for a series of polypeptide chains and their proline variants. To ensure long time scales for protein association at laboratory concentrations, we perform simulations in a high-concentration condition (∼25 mM). We use several experimentally determined polymorphic fibrillar structures as the initial structures in the simulations to diversify our aggregation sampling, in particular, under a high-concentration regime, where different initial configurations of proteins lead to heterogeneous aggregation properties. We carry out oligomer size and clustering analyses and use several structural measures to characterize oligomer size distributions as well as morphological and structural properties of the polypeptides studied. We find that the oligomer size dependence on different truncated sequences is determined by the secondary structure content of monomers. We also have calculated the polypeptides’ morphological (radius of gyration and distance measure D) and structural (nematic order parameter) properties. We have compared the results of D1/D1-core and their proline variants. The results show that proline variants can modulate morphological properties of the formed oligomer by excluding mutual distances between prolines in different chains. The radius of gyration of the oligomers increases (becomes more ellipsoid-like) as a result. However, proline residues within the oligomers behave differently in terms of their structural order. For the D1 sequence, its high configurational entropy blocks the shape-complementary effect by proline residues and therefore destroys an oligomer’s structural order, leading to the formation of “disordered” oligomer species. Interestingly, such a structural-order-disrupting capacity is suppressed in the D1-core sequence due to its shorter sequence length (low configurational entropy). This finding allows us to correlate the effect of structural order disrupting with the amyloid-disrupting capacity by prolines. A recent coordinate–free energy landscape visualization method⁵⁸ is suitable for quantifying such conformational phase space of the simulated structures. We plan to employ this method to delineate and visualize relevant free energy basins to analyze their pathways and energy barriers along aggregation free energy landscapes. The present study provides insights into the role of morphological and structural order of early-stage oligomeric species in driving different aggregation pathways.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpcb.2c05463.

• Additional information related to the summary of clustering results (Table S1); detailed structural features (Tables S2–S5); time evolution of (average) oligomer size obtained from molecular dynamics simulations (Figures S1–S3); fractions of secondary structure (Figure S4); distribution of the radius of gyration (R_g) and the numbers of contacts for differernt structure templates (Figure S5); nematic order parameters of D1/D1-core and their variants (Figure S6); and schematic diagram about structural diversity of protein double-helix molecular construct (Figure S7) (PDF)

Author Present Address

^§ Department of Chemistry and Biochemistry, National Chung Cheng University, Chiayi, Taiwan 621301

The authors declare no competing financial interest.

Special Issue

Published as part of The Journal of Physical Chemistry virtual special issue “Jose Onuchic Festschrift”.