Parallel In-Register Contact Propensity Predicts the Amyloidogenicity of ADan and ABri in Familial Dementias

Abstract

Familial Danish dementia (FDD) and British dementia (FBD) are rare neurodegenerative diseases caused by stop-codon mutations in the Bri2 gene, leading to amyloid deposits formed by the aggregation of mutant ADan and ABri peptides, respectively. FDD symptoms usually manifest decades earlier than those of FBD. Probably due to the rarity of these conditions, the aggregation mechanisms of ADan and ABri and the molecular basis for FDD’s earlier onset remain underexplored. Here, we computationally investigated the conformational dynamics of monomeric wild-type Bri23 and the two mutants ADan and ABri, as well as their self-assembly dynamics from dimers to hexamers using atomistic discrete molecular dynamics simulations. Our results aligned with earlier experimental work on the monomeric structures. We also confirmed that Bri23 is less amyloidogenic, showing significantly lower self-assembly propensity compared to ADan and ABri, which both favored forming β-sheet-rich oligomers. In amyloid aggregation, the critical nucleation event involves the transition from unaligned or misaligned oligomers to well-aligned fibril seeds featuring parallel in-register (PAIR) intermolecular β-sheets — the structural hallmark of mature fibrils. Notably, ADan showed a higher propensity than ABri to form longer PAIR intermolecular β-sheets within oligomers, suggesting a lower fibril nucleation barrier (i.e., higher amyloidogenicity) and thereby explaining FDD’s earlier onset. We further identified nucleation “hotspots” with high PAIR propensities in both ADan and ABri, which may serve as potential therapeutic targets. Together, these insights into the early aggregation dynamics of these rare dementias enhance our understanding of the aggregation-nucleation process and disease mechanisms.

Graphical Abstract

Introduction

Familial Danish dementia (FDD) and British dementia (FBD) are two rare, inherited neurodegenerative diseases characterized by widespread amyloid depositions due to mutations in the Bri2 gene^1,2. FDD patients starts to show symptoms in their twenties with cataracts, followed by impaired hearing, cerebellar ataxia, paranoid psychosis and dementia^3–5. FBD manifests later,withpatients reaching their fifties before experiencingpersonality changes, followed by memory impairment and progressive dementia, spasticity, and ataxia¹. In both FDD and FBD patients, widespread amyloid depositions are observed in blood vessel walls and areas of the central nervous system (CNS), including the cerebral cortex, choroid plexus, cerebellum, leptomeninges, brain stem and white matter^5,6. ADan and ABri, two 34-residue C-terminal fragments of the mutated Bri2 protein, are identified as the corresponding components of FDD and FBD lesion amyloid deposititions^7,8. The wild-type Bri2 gene encodes a 266-residue transmembrane protein. Cleavage by a furin-like protease between positions 243 and 244 releases a 23-residue fragment, Bri23. Mutant Bri2 genes in both FDD and FDB result from a stop-codon mutation at position 267 and a 10-nucleotide duplication insertion at the C-terminus, producing two elongated 277-residue precursor proteins, correspondingly. Cleaved at the same position as the wild-type precursor, these mutant Bri2 proteins release ADan and ABri peptide fragments, sharing the same first 22 residues in their sequences with the wild-type Bri23^1,2.

Both ADan and ABri showed a high tendency to self-assemble into β-sheet-rich amyloid fibrils in vitro, as confirmed by Western blot analysis, thioflavin- and ANS-binding as well as electron microscopy studies. Whereas wild-type Bri23 exhibited no detectable amyloidogenicity^9–11. Like many other amyloid proteins¹², the fibrilization process of Bri23 mutants follows a typical all-or-none sigmoidal kinetics. During the initial lag-phase, proteins form soluble oligomers and the rate-limiting step corresponds to the nucleation process, where the oligomers undergo nucleated conformational conversion to form the fibril seeds^13,14. With increasing high-resolution fibril structures solved to-date by advanced methods like cryo-EM, amyloid fibrils of most full-length proteins are comprised of parallel in-register (PAIR) intermolecular β-sheets, where the residues with the same positions between neighboring peptides are in contact^15–17. Therefore, the nucleation process marks the conformational transition from oligomers with random intermolecular interactions to well-aligned fibril seeds, in which all neighboring peptides participate in PAIR β-sheets that define the cross-β structure. Associated with a high free energy barrier, the pre- and post-nucleation aggregation intermediates around the free energy barrier of fibrilization are thus weakly populated, rendering both experimental and computational characterization of such nucleated conformational change challenging^18,19.

Multiple experiments investigated the amyloidogenicity and cytotoxicity of ADan and ABri in vivo and in vitro. In a transgenic mice model expressing the Danish mutant form of Bri2, significant vascular amyloid deposition, parenchymal ADan accumulation, intra- and intercellular deposition of oligomeric forms were observed²⁰. With overexpressed ADan precursor protein in the transgenic mice²¹, age-dependent behavioral deficits, weight loss, and increasing anxiety were observed together with age-related aggregations in the hippocampus and meningeal vessels. In contrast, the disease development in a transgenic mice model carrying the FBD mutation was subtle. Reported lines with high expression levels of mutant Bri2 displayed no detected brain pathology²². In a knock-in mouse of FBD, memory impairment was linked to a loss of Bri2 protein function²³. Unlike the hallmarks of perivascular amyloid aggregates and plaques mainly localized to the hippocampus and cerebellum observed in human cases²⁴, further in vivo studies are needed to fully characterize the effects of the British mutation on the Bri2 protein. Following a report of the neurotoxic effect of ABri peptide²⁴, multiple in vitro studies have investigated the toxicities of ADan and ABri oligomeric aggregates. Soluble oligomeric species of both ADan and ABri are shown to be neurotoxic to human neuroblastoma SH-SY5Y cells, whereas wild-type Bri23 exhibits minimal toxicity^10,11,25. Additionally, in a transgenic Drosophila model, ADan and ABri – but not Bri23 – were found to be neurotoxic in the central nervous system, even at low expression levels²⁶. A widely accepted hypothesis for amyloid-induced cytotoxicity is that these amyloid proteins may form oligomers in the form of nanopores that can be embedded in the membrane during their early stage of amyloid aggregation, disrupting membrane integrity and permeability^27–29. However, due to the rarity of these familial dementia cases, many aspects of ADan and ABri aggregation, including the molecular mechanism of amyloid aggregation, the molecular insights of the aggregation and toxicity of ADan and ABri oligomers, and molecular bases for the differential age-of-onset in corresponding familial dementia of FDD and FBD, remain unknown.

Here, we systematically studied wild-type Bri23 and two mutants ADan and ABri in silico by comparing their monomer structures and self-assembly dynamics up to hexamers using all-atom discrete molecular dynamics (DMD) simulations. DMD is a rapid and predictive molecular dynamics algorithm that enables efficient study of protein folding, protein interactions and amyloid aggregation of intrinsically disordered proteins (IDPs)^30–39. Our simulation result showed that the monomeric Bri23 was mostly unstructured with residual helices. ADan had a higher tendency to form helix structures, while ABri preferred to adopt a 3-stranded S-shaped β-sheet structure. In dimerization simulations, we confirmed that Bri23 was less favored to form intermolecular β-sheets compared to ADan and ABri, making it less likely to aggregate. During the oligomerization process, ADan rapidly formed large aggregates abundant in α-helices and underwent conformational change to form β-sheet-rich oligomers. For ABri, however, the stable intra-domain S-shaped β-sheet structure made it difficult to break the intra-molecular interactions, delaying the formation of extensive inter-peptide interactions. As a result, ABri underwent a dynamic aggregation process towards forming larger oligomers. In higher-order oligomers, both ADan and ABri formed β-barrel nanopores as intermediate states, which may contribute to amyloid toxicity.

Since both ADan and ABri could self-assemble into β-sheet-rich oligomers, alternative order parameters (or reaction coordinates) beyond the β-sheet content of aggregates should be used to distinguish their differential amyloidogenicity. Naturally, the extent of intermolecular PAIR alignment (i.e., the number of same-residues forming intermolecular PAIR contacts) between self-assembling peptides could be used to monitor fibril formation^40–43. Indeed, we were able to directly observe differential PAIR β-sheets formation in oligomerization simulations of ADan and ABri with different numbers of peptides by running many independent simulations. To account for different combinations of peptide pairs in computing the averaged PAIR alignment of oligomers with varying sizes, we proposed to weight the PAIR contact of each peptide pair in the oligomer by their total intermolecular contacts. Using this weighting quantification method, we found that ADan exhibited a stronger tendency to form longer PAIR β-sheets than ABri. Additionally, ADan displayed an increased PAIR propensity with increased oligomer sizes, while ABri did not. These results suggested that ADan has a lower free energy barrier for fibril nucleation than ABrithus providing a possible molecular basis for FDD’s earlier onset than FBD. Furthermore, we identified nucleation “hotspots” for ADan and ABri as corresponding regions with high PAIR propensities in oligomer simulations, which may be targeted for mitigating their pathological aggregation. Together, our study provides first molecular insights into the nucleation process of ADan and ABri in the early stage of aggregation, with important implications for therapeutic research in the two rare familial dementias.

Methods

Molecular Systems

The sequences of wild type Bri23, mutated ADan and ABri peptides used in our simulation were as depicted in Fig. 1, which have been widely employed in previous studies^1,2. To investigate the monomeric structures, we performed 500 ns replica exchange simulations in an 8 nm cubic simulation box, starting from diverse initial states to ensure sampling. Due to the absence of available solution structures, simulations were initiated with fully extended conformations. To study the thermodynamic properties of dimers in a 10nm cubic box, we also conducted replica exchange simulations with 2-peptide systems. The details of the molecular systems were summarized in Table 1. Initial structures were randomly selected from the top 10 most populated monomeric structures by clustering analysis from all the snapshots at T = 300K after systems reached equilibrium. For monomeric and dimeric systems for each peptide, temperatures for 12 replicas were evenly distributed from 270 K to 380 K. To further explore higher-order oligomers kinetics for ADan and ABri, we conducted constant temperature simulations of systems with 2, 4 and 6 peptides at T = 300 K. To maintain similar association and disassociation rates across each system, the concentration was kept around 0.33 nM by adjusting box sizes (Table 2). Initial structures were selected from the centroid nodes of the top 10 most populated clusters in the monomeric simulations. For sufficient conformational sampling, we performed 50 independent DMD simulations, varying initial states in terms of velocities, orientations and coordinates. The simulation for each system was run for over 750 ns.

Figure 1.

Conformational dynamics of monomers in replica exchange DMD simulations. (A) The amino acid sequences of Bri23, ADan and ABri are shown, with negatively charged residues colored in red, positively charged residues in blue, and hydrophobic residues in green. (B) Heat capacity ($C_v$) and secondary structure contents as a function of temperature are computed using WHAM. for Bri23 (C, D), ADan (E, F) and ABri (G, H), secondary structure propensities per residue (Top), contact frequency map, and representative monomeric conformations of the top four most-populated clusters are calculated. The structures are displayed in cartoon representation and colored according to secondary structures. Positively and negatively charged residues involving in salt-bridge formation are depicted as sticks, colored in blue and red, respectively. The N-terminal Cα atoms are showed as a sphere.

Table 1.

Details of the molecular systems for Replica Exchange simulations for Bri23, ADan and ABri.

Peptides	Peptide Numbers	Dimension (nm)	Simulation time (ns)
Bri23	1	8.0	500
	2	10..0	500
ADan	1	8.0	500
	2	10.0	500
ABri	1	8.0	500
	2	10.0	500

Table 2.

Details of the molecular systems for constant temperature simulations under T = 300K for ADan and ABri.

Peptides	Peptide Numbers	Dimension (nm)	DMD runs	Simulation time (ns)
ADan	2	10.0	50	800
	4	12.6	50	770
	6	14.4	50	767
ABri	2	10.0	50	768
	4	12.6	50	708
	6	14.4	50	773

DMD simulations

The simulations were performed using a rapid and predictive all-atom DMD algorithm utilizing Medusa force field, which has been extensively benchmarked for its accuracy in the field of protein conformational change, protein-ligand binding, protein interacting with nanoparticles with charged surfaces, as well as RNA structure predictions^44–48. Unlike classical molecular dynamics (MD), DMD employs optimized discrete step functions instead of continuous potential functions, and it includes both bonded interactions (i.e., covalent bonds, bond angles and dihedrals) and non-bonded interactions (i.e., van der Waals solvation, hydrogen bonds and electrostatic interactions)^44,49. The simulations used an EEF1 implicit solvent model⁵⁰. The constant temperature was controlled and monitored at around 300K by Anderson thermostat. Hydrogen bond interactions were modeled by a reaction-like algorithm^44,51. The screened electrostatic interactions between charged atoms were determined by the Debye–Hückel approximation with the Debye length set to approximately 10 Å. Atomic collision equations were resolved by solving the motions based on the conservation laws of energy, momentum and angular momentum. DMD software is accessible via Molecules In Action, LLC (www.moleculesinaction.com).

Weighted histogram analysis method (WHAM)

In replica exchange simulations, data collected under different temperatures were analysed using the weighted histogram analysis method (WHAM)⁵², which enables accurate reconstruction of the potential of mean force (PMF) and other thermodynamic properties of the simulated systems. The principle idea of WHAM is to obtain the density of states, $\rho \left(E\right)$, which is computed self-consistently from simulation trajectories⁵². Once $\rho \left(E\right)$ is determined, the partition function can be written as

$$Z=\int \rho \left(E\right)e^{-\frac{E}{k_{B}T}}dE.$$ #(1)

From this expression, the probability $P\left(A\right)$ of observing a structural parameter $A$ can then be evaluated. Here, the parameter A may represent a single-component scalar, multi-component vector, or a higher-dimensional tensor. The potential of mean force (PMF) of observing $A$ can therefore be calculated as

$$PMF\left(A\right)=-\ln P\left(A\right)=-\ln \left(\int P\left(A|E\right)\rho \left(E\right)e^{-\frac{E}{k_{B}T}}dE\right)+C,$$ #(2)

where $P\left(A\right|E)$ is the conditional probability of observing $A$ within the energy range of ($E,E+dE$) over all trajectories, and $C$ is a constant. WHAM also allows calculation of the heat capacity at constant volume $C_v$. By definition,

$$C_v=\frac{d<E>}{dT}.$$ #(3)

The ensemble-averaged energy at temperature $T,{<E>}_T$, is given by:

$$<E{>}_T=\frac{1}{Z}\int E\rho \left(E\right)e^{-\frac{E}{k_{B}T}}dE$$ #(4)

Combining Eqs. (1), (3), and (4), the heat capacity can be expressed as

$$C_v=\frac{<E^2{>}_T-<{E>}_T^2}{k_B{T}^2},$$ #(5)

in which

$$<E^2{>}_T=\frac{1}{Z}\int E^2\rho \left(E\right)e^{-\frac{E}{k_{B}T}}dE$$

Analysis methods

Hydrogen bonds were identified if the distance between backbone N and O atoms was less than 0.35 nm and the N-H…O angle was at least 120°⁵³. Secondary structure information was calculated using the dictionary secondary structure (DSSP) algorithm⁵⁴. Two peptides were considered in contact if they had at least one pair of intermolecular atomic distance smaller than the cutoff distacen of 0.65 nm. A peptide was considered part of the oligomer if it was in contact with one of component peptides of the complex. A two-dimensional PMF was derived from the probability distribution function, i.e., $-k_{B}TlnP(x,y)$, where $k_B$ represents for Boltzmann constant, and ($x,y$) is the two reaction coordinates of the system. To identify representative structures, cluster analysis was performed with a clustering oc program based on the root-mean-square distance (RMSD) for all pairs of atoms in two different conformations. β-barrel structures were identified when each β-strand in the oligomer formed at least two inter-peptide hydrogen bonds with its two neighboring β-strands to form a closed cycle^55,56. The cumulative distribution function of $N_{PAIR}$ is $CP\left(x\right)={\int }_x^{34}P\left(x\text{'}\right)dx\text{'}$, where the $P\left(x\right)$ indicated the probability of $x$ residues adopting PAIR contacts.

Intermolecular contact-weighted methods of multiple-molecular system

Two groups of atoms were considered to be in contact if the minimum atomic distance between them was at most 0.65 nm. In the simulations involving 4 or 6 peptides for ADan and ABri, each peptide could not simultaneously interact with all other peptides in the aggregates. Therefore, the number of atomic intermolecular contacts between two chains was used as a weight to average the contact parameters within the aggregates. Thus, the time-averaged probability for the number of residues forming parallel in-register intermolecular contacts was calculated from a Kronecker Delta function:

$$P\left(n\right)=<{\frac{\sum _{\alpha =1}^C\sum _{\beta <\alpha }{\delta }_{n,N_{PAIR}\left(\alpha ,\beta \right)}·N\left(\alpha ,\beta \right)}{\sum _{\alpha =1}^n\sum _{\beta <\alpha }N\left(\alpha ,\beta \right)}>}_{time}$$ #(6)

where n is the number of residues forming parallel in-register intermolecular contacts, C represents the total number of peptides in each aggregate, $N_{PAIR}(\alpha ,\beta )$ denotes the number of residues forming parallel in-register intermolecular contacts between chains $\alpha$ and $\beta$, and $N(\alpha ,\beta )$ is the number of atomic intermolecular contacts between chains $\alpha$ and $\beta$. Similarly, the residue-wise intermolecular contact frequency within each aggregate was calculated as:

$$f\left(R_i,R_j\right)=<\frac{\sum _{\alpha =1}^C\sum _{\beta <\alpha }\left[C\left(R_i^{\alpha },R_j^{\beta }\right)orC\left(R_i^{\beta },R_j^{\alpha }\right)\right]·N\left(\alpha ,\beta \right)}{\sum _{\alpha =1}^n\sum _{\beta <\alpha }N\left(\alpha ,\beta \right)}{>}_{time}$$ #(7)

where $C\left(R_i^{\alpha },R_j^{\beta }\right)$ indicates whether $R_i$ in chain $\alpha$ and $R_j$ in chain $\beta$ in contact, and the logical or operation accounts for the arbitrary order of chains. This approach ensures accurate weighting of intermolecular interactions and captures the main features of residue contacts in aggregates.

Results and Discussion

Folding dynamics of monomeric Bri23 and two mutants of ADan and ABri

The amino acid sequences of Bri23, ADan, and ABri^1,2 are shown in Fig. 1A. The three peptides have identical sequences in the first 22 residues. In the mutated C-terminus, ADan is rich with hydrophobic and aromatic residues (²³F, ²⁵LFL²⁷, ³⁴Y), while ABri contains more charged residues, including ²⁴R, ²⁷KK²⁸ and ³²EE³³. We first investigated the monomeric conformational dynamics of the three peptides using replica exchange DMD simulations. We started simulations with the peptides in fully extended conformations. For each peptide, twelve replicas with temperatures ranging from 270K to 380K were used, and the simulation of each replica lasted at least 500 ns with an accumulative simulation time of ~6 μs (details in Methods). The equilibration of the simulations was assessed by calculating the time-evolution of replica-averaged potential energies and the structural parameters such as β-sheet content (Fig. S1), which suggested that steady states for all three systems were achieved after ~200 ns. Therefore, simulations in the last 300ns of each replica were used for further analyses.

We first computed the heat capacity ($C_v$) as a function of temperature using the weighted histogram analysis method (WHAM, Methods)⁵² (Fig. 1B). Each peptide exhibited a major $C_v$ peak, corresponding to significant energy changes associated with conformational transitions. To clearly interpret the physical events underlying each peak, it is necessary to examine the $C_v$ profile together with changes in other structural parameters. For the monomers, the peaks corresponded to the temperature at which the peptides began to lose their ordered secondary structures and became fully unstructured, as shown by the sharp decrease in β-sheet and α-helical contents together with a corresponding increase in random coils. The $C_v$ peak temperature of ABri was 314 K, higher than those of Bri23 (304 K) and ADan (308 K), suggesting that ABri might have the highest thermostability. To characterize these conformational changes, we calculated the secondary structure contents as a function of temperature. At 300 K, Bri23 was predominantly unstructured with residual helices. This largely unstructured property was consistent with CD spectra from a prior experimental study, which showed that Bri23 remained in an unordered conformation at 0 h in a physiological-salt buffer, likely reflecting its monomeric state⁹. ADan had similar secondary structure contents as Bri23, with a slightly higher helical content. ABri, on the other hand, tended to adopt more β-sheet structures than Bri23 and ADan, as observed experimentally⁵⁷. For all three peptides, their $C_v$ peaks were associated with the loss of ordered secondary structures to disordered coils and with significantly increasing radius of gyration (Fig. S2).

For a more detailed structural characterization, we calculated secondary structure propensities per residue, and residue-wise contact frequency maps at 300 K. We also selected representative structures using the hiarchical clustering analysis with a 0.4 nm root-mean-square distance (RMSD) cutoff (Fig. 1C–H). Bri23 (Fig. 1C, D) was largely disordered at both termini but formed a dynamic helix centered around residues 8–17, and occasionally formed β-hairpins that were stabilized by hydrophobic interactions and a salt bridge between arginine 9 (⁹R) and glutamate 18 (¹⁸E). For ADan (Fig. 1E, F), the addition of 11 residues in the C-terminal did not significantly affect the overall secondary structure propensities of the first 23 residues, and the additional residues predominantly formed helical structures. However, in the case of ABri, the additional 11 residues in the C-terminal changed the structural preference of the N-terminal 23 residues from helix and coil to β-sheet (Fig. 1G, H). The peptide preferred to form S-shaped β-sheet structures, composed of residue fragments 3–11, 14–21 and 26–31. The abundance of charged residues in the C-terminal, particularly these double charges of ²⁷KK²⁸ and ³¹EE³², facilitated the formation of multiple salt bridges between neighboring β-strands and stabilized the β-sheet structures. Due to the intrinsically disordered propensity in the N-terminal, the S-shaped β-sheet structure of ABri could be disrupted, rendering a dynamic structural ensemble. In short, while all three peptides remain largely disordered, their monomeric ensembles contain diverse transient structural motifs shaped by their underlying sequence properties.

Earlier studies have reported the formation of disulfide bonds between 5C and 22C in all three peptides under oxidizing environments^25,58–60. Based on secondary structure prediction using multi-sequence alignment and a constraint of an intra-molecular disulfide bond between cysteines 5 and 22 (⁵C-²²C)⁶¹, an earlier molecular model of ABri also featured an S-shape β-sheet structure stabilized by a salt-bridge between ⁹R and ¹⁸E, along with electrostatic interactions involving lysine 14 (¹⁴K)⁶⁰. Interestingly, the most populated structural cluster in our simulation – conducted without imposing disulfide bond constraints – closely resembles this earlier model. These results indicate that ABri’s S-shaped β-sheet structure may represent a typical monomeric configuration in both reducing and oxidizing environments. In contrast, Bri23 and ADan did not show cysteines ⁵C and ²²C in proximity within their β-hairpins in our simulations (Fig. 1D, F), suggesting that their monomeric structures might differ under oxidative environments.

Bri23 has significantly lower dimerization propensity than ADan and ABri

We next investigated the dimerization dynamics of each peptide using replica exchange simulations, starting with two randomly separated monomers in fully extended conformations. The simulation of each replica lasted approximately 500 ns. Systems reached equilibrium for three peptides after ~250 ns, as suggested by the time evolution of replica-averaged potential energies, β-sheet and α-helix contents (Fig. S1). To minimize potential biases from the initial structures, only the data from the final 250 ns of the simulations were utilized for further analyses.

We first calculated the temperature dependence of $C_v$ and binding frequencies ($P_{\mathit{Bound}}$) using WHAM to examine the dimer thermostabilities (Fig. 2A). Under low temperatures, dimer populations were high for all three peptides, approximately 95% for Bri23 and 99% for both ADan and ABri. As temperature increased, the dimers started to dissociate. As indicated by the $C_v$ peaks and the sharp drop in $P_{\mathit{Bound}}$ corresponded to the dimer dissociation, the Bri23 dimer exhibited significantly lower dissociation temperature, and thus, lower thermostability and weaker binding affinity than these of ABri and ADan dimers. Next, we analyzed secondary structure changes upon dimerization by calculating averaged secondary structure content at 300K (Fig. 2B). All three peptides underwent structural changes upon dimerization, characterized by increased β-sheet content and reduced α-helices and unstructured coils. Among them, Bri23 exhibited lower β-sheet content than its two mutants, suggesting a weak tendency to form β-sheet–rich dimers, consistent with the faint ThT fluorescence observed in an earlier experimental study⁶². Residue-wise secondary structure propensities and their changes relative to monomer simulations were further examined at 300K (Fig. 2C). In Bri23, the increase in β-sheet content was primarily observed in residues 4–11 and 17–22. These regions in ADan also exhibited increased β-sheet propensity, but to a much greater extent. Additionally, residues 22–26 in ADan’s C-terminal region exhibited a significant rise in β-sheet content, while the rest of C-terminal residues retained some helical structures. ABri showed notable increases in β-sheet propensity of residues 12–14 and 22–25, which mainly adopted turns or coils linking two strands in its monomeric S-shaped β-sheets. Its C-terminal region also displayed enhanced β-sheet formation upon dimerization.

Figure 2.

Dimerization dynamics probed by replica exchange DMD simulations. (A) Heat capacity (Cv) (Top) and contact frequency ($P_{\mathit{Bound}}$) (Bottom) are analyzed as functions of temperature. (B) Secondary structure propensities for the three peptides are calculated at 300K. (C) The residue-wise secondary structure contents (C_SS) and differences in β-sheet contents between monomeric structures and dimeric systems (Δβ) are computed for Bri23 (Top), ADan (Middle) and ABri (Bottom). (D) The potential of mean force (PMF) is calculated as the function of the number of residues in intermolecular contacts ($N_{\mathit{inter}}$) and the number of residues in intra-molecular contacts ($N_{\mathit{intra}}$) for Bri23 (Top), ADan (Middle) and ABri (Bottom) using WHAM. (E) The PMF is computed as the function of the number of residues adopting β-sheet ($N_{\mathit{Sheet}}$) and the number of interchain hydrogen bond (N_HB) for Bri23 (Left), ADan (Middle) and ABri (Right) using WHAM.

To further investigate dimerization dynamics, we calculated the two-dimensional potential of mean force (2D-PMF, i.e., the effective free energy landscape) at 300K as a function of the number of inter- and intra-molecular residue-wise contacts ($N_{\mathit{inter}}$, $N_{\mathit{intra}}$) (Fig. 2D) and also as a function of the number of residues forming β-sheets ($N_{\mathit{Sheet}}$) and inter-peptide hydrogen bonds (N_HB) (Fig. 2E). Both PMF plots of B23 dimerization featured two basins, corresponding to the unbound state with zero intermolecular contacts and hydrogen bonds and the bound state, correspondingly. Compared to Bri23, the unbound states for ADan and ABri became less populated with the corresponding basins being narrower and shallower, while the bound states formed more intermolecular contacts and hydrogen bonds. While the analyses confirmed that both mutants had a significantly higher dimerization propensity than the wild type Bri23, their bound states displayed distinct features in the PMF plots – e.g., ADan had one major dimer basin, while ABri had two separated dimer basins (Fig. 2D, E). To examine the structural differences, we computed the intra- and intermolecular contact frequency maps along with representative dimer structures as the centroid clusters from clustering analysis (Fig. S3). The contact maps indicated that the dimers of Bri23 were less populated with lower β-sheet content than those of its mutants, and these Bri23 dimers were formed dynamically through weak nonspecific contacts (e.g., conformation η in Fig. S3A). Most ADan dimers featured composite monomers forming a β-hairpin while two monomers interacted with each other along either one of the two β-strands (e.g., conformations a, b and c in Fig. S3B). Compared to ADan, ABri dimers were structurally more diverse, where each monomer could either retain the S-shaped β-sheet or have one or both β-hairpins unfolded, two monomers forming different extents of intermolecular β-sheets between the exposed β-strands (Fig. S3C), resulting into two distinct basins in the PMF plots (Fig. 2D,E). These diverse ABri dimers were stabilized by inter- and intermolecular salt-bridges between charged residues rich in the mutant sequence in the C-terminal (Fig. 1A).

Given that previous studies reported heterozygosity in both FDD and FBD patients^63,64, wild-type Bri23 can coexist with the mutated peptides in vivo. Therefore, we further examined the potential formation of heterodimers, specifically Bri23–ADan and Bri23–ABri dimers (Figs. S4 & S5). To enable direct comparison with the homodimer simulations, we used the same replica-exchange simulation setup as the ADan or ABri homodimers, performing 500 ns simulations and analyzing the data from the last 250 ns. As indicated by the temperature-dependent binding frequencies, both ADan and ABri were able to form dimers with Bri23. However, the sharp decrease in binding frequency at the temperatures corresponding to the $C_v$ peaks suggested that the dissociation temperatures of the heterodimers were lower than those of the homodimers. This indicated that homodimers formed by the mutated peptides possessed higher thermostabilities than the corresponding heterodimers with Bri23, meaning that the mutated peptides were more likely to associate with themselves than with wild-type Bri23 (Figs. S4A & S5A). To further investigate heterodimer structures, we computed two-dimensional PMFs as functions of the number of residues adopting β-sheets ($N_{\mathit{Sheet}}$) and the number of intermolecular hydrogen bonds (N_HB) (Figs. S4B & S5B). Both ADan and ABri exhibited PMF basins corresponding to states with no intermolecular hydrogen bonds, while additional basins reflected dimers stabilized by intermolecular hydrogen bonds. We also constructed intermolecular contact-frequency maps to examine the detailed binding structures and hotspot regions. These analyses revealed that Bri23 preferentially interacted with the mutated C-terminal regions of ADan and ABri by forming intermolecular β-sheets with these segments. In summary, even in the presence of wild-type Bri23, the mutated peptides remained the pereference of self-aggregation. Meanwhile, they could also form intermolecular β-sheets with wild-type Bri23 and potentially interfere with its normal function.

Both ADan and ABri formed β-sheet-rich oligomers

While our replica exchange simulations showed that both ADan and ABri preferred to form β-sheet-rich dimers and that ABri dimers had a higher structural diversity than ADan dimers, it was not clear how this behavior would affect the self-assembly of lager oligomers. Hence we investigated the aggregation dynamics of two, four and six ABri and ADan peptides using constant temperature simulations at 300 K, in order to capture the time-dependent aggregation kinetics, which cannot be readily obtained from replica exchange simulations. To ensure sufficient sampling, we performed 50 independent DMD simulations for each molecular system. Each independent simulation started with two, four or six monomers adopting equilibrated structures (selected from the centroid nodes of the top 10 populated clusters of the snapshots at T = 300 K in the replica exchange simulations) randomly positioned and kept from each other by a minimal distance of 1.5 nm in the simulation box, and lasted at least 700 ns (Fig. S6, S7 & S8). Molecular systems with different numbers of peptides were kept at the same peptide concentration (~3.3 mM) by adjusting the size of corresponding simulation boxes (Table 1).

We first compared the steady-state structural properties of ADan and ABri by averaging over the last 250 ns trajectories of all independent simulations. Both ADan and ABri formed β-sheet-rich aggregates in 2-, 4-, and 6-peptide simulations (Fig. S6–S8). Examination of the secondary structure propensity per residue demonstrated that ADan residues 11–16, connecting two β-strands in replica exchange simulations of ADan dimerization (Fig. 2C), underwent a conformational transition from unstructured turns and coils to β-sheet structures with increasing aggregate sizes (Fig. 3B). Additionally, ADan residues in the C-terminal also displayed a helix-to-sheet transition with the increase of aggregate sizes. ABri, on the other hand, exhibited very little size-dependent changes in secondary structure contents with 2, 4, or 6 peptides (Fig. 3A, Right). Notably, the averaged secondary structure content observed in constant-temperature 2-peptide simulations closely resembled the corresponding results in replica exchange dimerization simulations (Fig. 2C), underscoring the reliability of the structural ensembles captured in the constant-temperature simulations with multiple independent runs.

Figure 3.

Conformational analysis for self-assembly of the molecular systems containing 2, 4, or 6 ADan and ABri. (A) Averaged contents of secondary structures are calculated for ADan and ABri. (B) Residue-wise secondary structure propensities are computed for ADan (Left) and ABri (Right). Error bars depict the standard deviations of the mean from all the 50 trajectories. The potential of mean force (PMF) is calculated as the function of the average number of hydrogen bonds per chain for inter-peptide (N_HB - Inter) and intra-peptide (N_HB - Intra) with representative structures for ADan (C) and ABri (D) The corresponding structures are colored by chains.

To further characterize the aggregate structures of different numbers of peptides simulated, we computed the 2D-PMFs as a function of the average number of inter-peptide (N_HB-Inter) and intra-peptide (N_HB-Intra) hydrogen bonds per chain, using the last 250 ns trajectories of all independent simulations (Fig. 3C & 3D). As the oligomer size increased, free-energy basins shifted rightward for both ADan and ABri, indicating the formation of more intermolecular hydrogen bonds in larger oligomers. In 2-peptide simulations, both mutants exhibited basins centered at N_HB-Inter = 0, corresponding to unbound states or dimers dynamically formed by weak nonspecific contacts (Fig. 3C & 3D, Fig. S9A & S10A). Additionally, each mutant displayed a primary bound-state basin corresponding to dimers forming intermolecular β-sheets, along with a minor basin corresponding to dimers with more extensive intermolecular hydrogen bonds. As the peptide number grew, the zero N_HB-Inter basins disappeared for both mutants. For ADan, the central primary basin shifted continuously to higher N_HB-Inter values, merging with the minor basin and extending into a wider basin. This behavior corresponded to the formation of larger β-sheet rich aggregates stabilized by extensive intermolecular hydrogen bonds with increasing aggregate sizes. In contrast, ABri displayed only a modest rightward shift of the central basin, while the minor basin with high N_HB-Inter configurations remained weakly populated with inceasing aggregate sizes (Fig. 3D & S10). By computing the averaged intrapeptide contact maps (Fig. S11 & S12), the analysis showed that ADan’s intramolecular contact patterns became less pronounced as the aggreagte size increased, whereas ABri’s intramolecular contact maps remained largely unchanged across all simulation systems. Together, these findings indicated that as oligomer size increased, ADan transitioned from intradomain to interdomain interactions, while the stable intradomain interactions of ABri monomers hindered the formation of extensive intermolecular hydrogen bonds in their aggregates.

Differential aggregation pathways of ADan and ABri

We next characterized the oligomerization processes for both ADan and ABri by examining the largest simulation system with six peptides. We first compared the distribution of oligomer sizes within the last 250 ns of all independent simulations. ADan notably favored the formation of hexamers than ABri (Fig. 4A). As shown by the time-evolution of the largest aggregate size and the total β-sheet content averaged over independent simulations, ADan formed hexamers more rapidly and showed a greater increase in the β-sheet content than ABri (Fig. S8). We also analyzed the probability distribution of β-strand lengths, defined as the number of consecutive residues adopting β-sheet structures (Fig. 4B). The β-strand length distribution of ABri showed a sharp, narrow peak, with 7 residues being most dominant in the aggregates. In contrast, ADan displayed a broader distribution of β-strand lengths, with a major peak around 10 residues and a minor peak around 19 residues, reflecting a significantly higher propensity to form longer β-strands.

Figure 4.

Conformational analysis of the oligomer systems. (A) Weighted oligomer size distribution is computed for ADan and ABri. Error bars represent the standard deviations of the mean from all the 50 trajectories. (B) Probability distribution of β-strand length for ADan and ABri. PMF as the function of oligomer size and β-sheet content is calculated, with representative structures for ADan (C) and ABri (D). β-barrel probabilities are calculated for (E) ADan and (F) ABri in molecular systems containing 2, 4, or 6 peptides. Error bars represent the standard deviations of the mean across all the 50 trajectories. Representative β-barrel homo-hexamers for ADan and ABri are shown in both side and top views, colored by chains. Peptides in all the panels are shown in cartoon representation and colored by chain.

We then calculated the aggregation free energy landscape, approximated by the 2D-PMF as a function of oligomer size and the corresponding β-sheet content (Fig. 4C & 4D). We used all trajectories in 50 independent simulations in order to capture the assembly process from non-interacting monomers. Although both ADan and ABri exhibited the deepest energy basins corresponding to β-sheet-rich hexamers, two mutants displayed distinctive aggregation pathways towards such stable aggregates from non-interactive monomers with low β-sheet content (i.e., the lower left conner basin in the 2D-PMF in Fig. 4C,D). For ADan, the most probable pathway with lowest free energies in the PMF featured the rapid assembly into large, coil- and helix-rich oligomers (state a in Fig. 4C with a representastion snapshot at 93ns of a typical trajectory shown in Fig. S13A). Transition to β-sheet–rich aggregates required overcoming a free-energy barrier to disrupt intramolecular interactions and form extensive intermolecular hydrogen bonds (e.g., states b-d in Fig. 4C). In contrast, the aggregation pathway of ABri followed the initial formation of β-sheet–rich dimers (state α in Fig. 4D), which futher self-assembled into larger aggregates, corresponding to a broad, flat basin 4 to 6 chains (states β, γ in Fig. 4D) with all these states having similar free energies. Examination of representative structures revealed the persistent intra-domain S-shaped β-sheets, consistent with the high probability of short β-strands observed in the aggregates (Fig. 4B). As shown in the representative trajectories (Fig. S13B), the hexamers underwent frequent association and dissociation, likely due to the limited number of hydrogen bonds between short β-strands, which lead to less stable aggregates and a more dynamic aggregation process. Consequently, the aggregation of ABri had higher population of small oligomers than that in ADan (Fig. 4A). Together, our results suggest that ADan and ABri followed distinct early aggregation pathways.

Both ADan and ABri could form β-barrel oligomers as intermediate states.

In many amyloid proteins, small oligomers in the early stages of fibrillization are considered to be the most toxic species, often leading to cell dysfunction or death in amyloidosis. A leading amyloid toxicity hypothesis proposes that amyloid proteins can form nanopores within the cell membranes, leading to the ion leakage or disruption of membrane intergrity^19–21. Supporting this hypothesis, a fragment of αB-crystalline was first found to form a cylindrical barrel (e.g., β-barrel) composed of antiparallel β-sheets, displaying toxicity in a cell viability assay⁶⁵. The formation of β-barrel structures has also been observed in other amyloid peptides, including Aβ and IAPP, supported by both experimental and computational studies^66–71. An earlier computational study examining Aβ16–22 oligomerization further demonstrated that β-barrel structures can act as metastable intermediates on the aggregation free-energy landscape, separated from double-layer β-sheet structures by a free-energy barrier⁷². This work provides a useful conceptual framework for understanding how transient β-barrel species may contribute to oligomer toxicity within the aggregation pathway. Interestingly, β-barrel intermediate states were also observed in the aggregation process of both ADan and ABri in our simulations, as long as the number of simulated peptides exceeded two (Fig. 4E, F). For both mutants, the probability of observing the β-barrel structures increased with increasing number of peptides simulated. These findings may shed light on the potential role of β-barrel oligomers formed by both ADan and ABri in contributing to their amyloid toxicity and pathogenesis in FDD and FBD.

ADan preferred to form longer PAIR β-sheets than ABri in oligomers

From the above analyses, although ADan and ABri followed different early aggregation pathways, both peptides ultimately favored β-sheet–rich aggregates and formed β-barrel intermediates with comparable probabilities. In order to account for FDD’s earlier onset compared to FBD at the molecular level, further investigation was needed. Given that the formation of parallel in-register or PAIR inter-domain β-sheets, defined by the one-to-one alignment of the same sequences between neighboring chains, are crucial for the structural transition from oligomers to protofibrils⁷³, we proposed to evaluate the fibril nucleation tendencies by calculating the tendency to form PAIR contacts. Specifically, a PAIR contact was defined when the sidechain of a residue in one chain was in contact with the sidechain of the same residue in its neighboring peptide. In dimer simulations, calculating the number of PAIR contacts between peptides ($N_{\mathit{PAIR}}$) was straightforward, as only one peptide pair was present. However, in larger systems, such as the 4- and 6-peptide simulations, multiple peptide pairs existed with varying degrees of intermolecular interactions – e.g., some peptide pairs had many intermolecular interactions while others had little, even zero contacts. To take such differences into account, we weighted each peptide pair’s $N_{\mathit{PAIR}}$ value by the total number of intermolecular residue contacts within the pair, N_contacts. Similarly, we also computed the probability distribution of $N_{\mathit{PAIR}}$ in a given oligomer, by weighting the probability of a specific $N_{\mathit{PAIR}}$ value between a peptide pair by its corresponding N_contacts, as shown in Eq. (6) (Methods). Averaging across all sampled oligomers yielded the final $N_{\mathit{PAIR}}$ probability distribution (Fig. S14). This approach enabled direct comparison between simulation systems with different numbers of peptides.

To better illustrate the trend of PAIR β-sheet formation in different molecular systems, we also computed the cumulative probability distribution of $N_{\mathit{PAIR}}$, defined as the probability of forming at least $N_{\mathit{PAIR}}$ PAIR contacts (Fig. 5A). In 2-peptide simulations of ADan and ABri, $N_{\mathit{PAIR}}$ probability peaks around 10 and 16 occurred at frequencies below 0.01, indicating that PAIR contacts were rare and less favored for dimers (Fig. S14). This observation was consistent with replica exchange simulations of dimers, which also showed minimal PAIR contacts in the $N_{\mathit{inter}}$ -$N_{\mathit{PAIR}}$ PMF for ADan (Fig. S15). As the oligomer size increased, ADan exhibited a marked shift toward more residues forming PAIR contacts – e.g., the $N_{\mathit{PAIR}}$ distributions of both tetramers and hexamers peaked near 16, while hexamers had an additional peak at 23 (Fig. 5A & S14). In contrast, ABri tetramers and hexamers showed no significant peak beyond 20 residues. Hence, our direct observation of PAIR contact formation in oligomers provided additional molecular insights beyond the β-sheet content into the nucleation process in the early stage of aggregation. Additionally, the observation that ADan more readily formed longer PAIR contacts in larger oligomers than ABri, suggesting a lower nucleation barrier associated with the critical conformational transition between oligomers and fibril seeds. With a lower fibril nucleation barrier, ADan is expected to be more amyloidogenic than ABri, which is consistent with the clinically observed earlier onset of FDD than that of FBD. Interestingly, we also identified trajectories in which ADan and ABri started from monomers and eventually formed β-sheet–rich hexamers that contained PAIR β-sheet segments (Fig. S16). Along this pathway, β-barrel intermediates also appeared. Capturing both of these rare events within a single trajectory further highlighted the sampling capability of the DMD algorithm.

Figure 5.

The aggregate formation of parallel in-register intermolecular contacts. (A) The cumulative probability distributions for the number of residues in intermolecular parallel in-register contact ($N_{\mathit{PAIR}}$) is calculated for ADan (Top) and ABri (Bottom) in molecular systems containing 2, 4 or 6 peptides. (B) Intermolecular contact frequency maps for systems with 2 (Left), 4 (Middle) and 6 (Right) peptides are plotted for ADan (Top) and ABri (Bottom). The probability of PAIR contact formation for each residue is projected in 2-, 4-, and 6-peptide systems for ADan (Top) and ABri (Bottom). (C) Hexamers with parallel in-register β-sheets for ADan (Top) and ABri (Bottom) are depicted in cartoon style, colored by chains, with the parallel in-register β-sheet motifs emphasized.

To identify potential fibril nucleation sites, we next computed the residue-wise intermolecular contact frequency maps for the 2-, 4-, and 6-peptide systems using the same weighting method as in the calculation of $N_{\mathit{PAIR}}$ contacts, as shown in Eq. (7) (Methods). To better visualize the hotspot regions involved in forming PAIR contacts, we projected the diagonal of the contact frequency maps as a 1D probability distribution, where residues with higher PAIR alignment probabilities could stand out (Fig. 5B). As the system size increased, the fibril nucleation hotspots with high PAIR propensities for ADan first emerged from its mutated hydrophobic C-terminal region ²⁰LICFNLFL²⁷. Interestingly, this primary nucleation hotspot we identified aligns well with a recent study reporting that the amyloidogenic core of ADan lies within residues 20–26⁷⁴, further highlighting the sampling capability of the DMD algorithm. The PAIR β-sheet formation region further extended to the adjacent sequence of ¹⁵FAVETLICFNLFL²⁷ with a secondary site in the N-terminal, ⁶FAIRHFE¹². In these hotspots, hydrophobic interactions facilitated PAIR contacts. The lower PAIR propensity of the hydrophobic segment ⁶FAIRHFE¹² was likely due to electrostatic repulsion between the closely positioned charged sidechains of 9R and 12E, making the PAIR contacts less stable. Unlike ADan, the mutated C-terminal region of ABri was less hydrophobic and highly charged, making it unlikely to contribute to PAIR formation. As a result, the PAIR contacts in ABri were primarily formed in the N-terminal hydrophobic fragment ⁶FAIRHFE¹², with a secondary hotspot in the center, ¹⁵FAVETLI²¹, observed consistently across all simulated systems. Notably, charged residues ²⁷KK²⁸ and ³²EE³³ facilitated the formation of anti-parallel β-sheets in ABri aggregates, leading to same-residue contact around residue 30, creating a “false-positive” spike in the PAIR contact probability (Fig. 5B). Interestingly, ABri’s identified fibril nucleation hotspot regions overlapwith findings from a previous study showing important roles of ⁶FAIRHF¹¹ and ¹¹FEKFAV¹⁷ in the fibrillization⁷⁵. Given their importance in driving fibril nucleation,the identified hotspots in ADan and ABri could serve as potential targets in future therapeutic designs.

Since Cys5 and Cys22 were found in close position in ABri monomeric structures predicted under both our reduced-environment simulations and the oxidized conditions reported previously, we further examined their intra- and intermolecular contact frequencies across variant cysteine positions (Fig. S17). In ADan, intramolecular ⁵Cys–²²Cys contacts decreased as the number of simulated peptides increased, consistent with the reduced intramolecular interactions observed as system sizes enlarged (Fig. 3C & S11). In contrast, ABri maintained similar levels of intramolecular ⁵Cys–²²Cys contacts across all system sizes, in agreement with its persistent intramolecular contact patterns and hydrogen bonds (Fig. 3D & S12). For intermolecular contacts between cysteines at the same positions in neighboring chains, ADan showed a higher propensity for Cys22 than for Cys5, consistent with nucleation hotspot regions originating in the mutated hydrophobic C-terminal segment (Fig. 5B). As these PAIR-contact hotspot regions expanded toward the N-terminal residues in larger systems, Cys5 increasingly participated in intermolecular contacts with Cys5 from its neighbouring chains (Fig. S17A). In comparison to ADan, ABri exhibited lower intermolecular contact frequencies for both Cys5 and Cys22. Consistent with nucleation hotspots located near the N-terminus, ABri showed a higher intermolecular contact propensity for Cys5 than for Cys22 (Fig. 5B). Together, under reduced conditions, the intramolecular cysteine contact propensities were closely linked to the amyloidogenicity of the peptides, while the intermolecular contact propensities of cysteines at equivalent positions were strongly governed by their adjacent nucleation hotspot regions. Our observations further suggested that disulfide-bond formation under oxidizing conditions might modulate the aggregation pathways of these peptides, providing a potential direction for future investigation.

Based on the above analyses, we propose an order parameter, Q_fibril, defined as the average $N_{\mathit{PAIR}}$ normalized by the sequence length L – Q_fibril = <N_PAIR>/L, to quantify the conformational transition from monomers and oligomers to fibrils. The Q_fibril quantifies the fraction of residues adopting PAIR contacts in an aggregate. A fibril with perfect alignments has <N_PAIR> = L and Q_fibril = 1. Conversely, completely misaligned aggregates yield Q_fibril = 0. Thus, a higher Q_fibril value indicates a stronger fibrillar order within the configuration. For the two representative hexamer structures in Fig. 5C selected based on peak values in the N_PAIR probability distribution – i.e., 23 PAIR contact residues for ADan; 17 PAIR contact residues for ABri (Fig. S14), the corresponding Q_fibril values were 0.262 and 0.189. This result underscores the efficiency of the proposed order parameters in differentiating fibril orders. Hence, such a configuration-specific, multi-peptide-system applicable, single-valued order parameter ranging from 0 to 1, can be used in the future study to quantitatively assess the progression of fibril nucleation. Together, our results highlighting the distinct nucleation pathways of ADan and ABri, driven by differences in their mutated C-terminal sequences, might offer new perspectives for investigating other disease-related amyloid aggregation at molecular level.

Conclusions

Using atomistic DMD simulations, we systematically investigated the conformational dynamics and early aggregation behavior of wild-type Bri23 and its dementia-associated mutants, ADan and ABri. Our simulations revealed that Bri23 monomers remain largely unstructured with residual helices and exhibit minimal self-assembly propensity, consistent with prior experimental studies reporting its lack of amyloidogenicity. In contrast, ABri adopts meta-stable S-shaped β-sheet conformations, while ADan displays a more dynamic structural ensemble, including transient α-helices and β-hairpins, particularly in the hydrophobic C-terminal extension. These results recapitulate previous structural observations and highlight the distinct conformational landscapes of the mutant peptides.

In oligomerization simulations, both ADan and ABri form β-sheet–rich aggregates and transient β-barrel intermediates, potentially relevant to cytotoxicity. However, only ADan exhibits a marked propensity to form longer parallel in-register (PAIR) β-sheet alignments – an essential structural hallmark of fibril nucleation. This trend was robust across oligomer sizes (2, 4, and 6 peptides) and was quantified using the probability distribution of contact-weighted $N_{\mathit{PAIR}}$, the number of residues forming PAIR contacts. We also introduced a novel order parameter, Q_fibril, which measures the fraction of residues engaged in PAIR contacts. Our analyses suggested a significantly lower fibril nucleation barrier for ADan. These findings are consistent with clinical observations of FDD’s earlier onset compared to FBD and support previous experimental evidence of ADan’s higher aggregation propensity and toxicity. We further identified sequence-dependent nucleation “hotspots” that correlate with hydrophobic and charged residue distributions. The primary nucleation hotspot of ADan was located at the hydrophobic C-terminal residues 20–27, then extends to 15–27 with a secondary nucleation hotspot in residues 6–12. In ABri, however, its mutated C-terminal was highly charged, thus residues 6–12 became the primary nucleation hotspot. Another hydrophobic region of residues 15–21, however, was less dominant in PAIR β-sheet formation, likely because it remained buried in ABri’s S-shaped monomeric β-sheets. Thus, ABri’s higher β-sheet content in its monomeric states did not necessarily indicate greater amyloidogenicity. Instead, its stable intramolecular contacts likely created additional barriers for ABri to overcome the fibril nucleation process. Together with its sequence properties, ABri was less favorable than ADan in forming extended intermolecular PAIR β-sheets and progressing toward fibril formation.

Taken together, this work provides atomistic insight into how point mutations modulate early-stage aggregation dynamics and nucleation kinetics in rare familial dementias. By integrating predictive, efficient DMD simulations with a novel, quantitative fibril order parameter, we establish a generalizable framework for dissecting sequence–structure–aggregation relationships in amyloid diseases. These findings underscore the biological and therapeutic significance of early oligomer structure and alignment in dictating disease onset and progression.

Highlights.

• Weighted intermolecular PAIR contacts is proposed to quantify oligomer structures

• PAIR-contact propensity correlates with ADan and ABri amyloidogenicity

• Residues with high PAIR-contact propensity are amyloid aggregation hotspots

• Normalized PAIR contacts can serve as an order parameter for oligomer nucleation

Data availability

The raw data for plotting figures and corresponding protein structure Pymol session files are available at Zenodo (10.5281/zenodo.17784594).