Computational Investigation of Co-Aggregation and Cross-Seeding between Aβ and hIAPP Underpinning the Crosstalk in Alzheimer’s Disease and Type-2 Diabetes

Abstract

The coexistence of Amyloid-β (Aβ) and human Islet Amyloid Polypeptide (hIAPP) in the brain and pancreas is associated with an increased risk of Alzheimer’s disease (AD) and type-2 diabetes (T2D) due to their co-aggregation and cross-seeding. Despite this, the molecular mechanisms underlying their interaction remain elusive. Here, we systematically investigated the cross-talk between Aβ and hIAPP using atomistic discrete molecular dynamics (DMD) simulations. Our results revealed that the amyloidogenic core regions of both Aβ (Aβ_10–21 and Aβ_30–41) and hIAPP (hIAPP_8–20 and hIAPP_22–29), driving their self-aggregation, also exhibited a strong tendency for cross-interaction. This propensity led to the formation of β-sheet-rich hetero-complexes, including potentially toxic β-barrel oligomers. The formation of Aβ and hIAPP hetero-aggregates did not impede the recruitment of additional peptides to grow into larger aggregates. Our cross-seeding simulations demonstrated that both Aβ and hIAPP fibrils could mutually act as seeds, assisting each other’s monomers in converting into β-sheets at the exposed fibril elongation ends. The amyloidogenic core regions of Aβ and hIAPP, in both oligomeric and fibrillar states, exhibited the ability to recruit isolated peptides, thereby extending the β-sheet edges, with limited sensitivity to the amino acid sequence. These findings suggest that targeting these regions by capping them with amyloid-resistant peptide drugs may hold potential as a therapeutic approach for addressing AD, T2D, and their co-pathologies.

Graphical Abstract

1. Introduction

The aberrant aggregation of amyloidogenic proteins and peptides into fibrillar aggregates in the form of insoluble deposits is a hallmark feature of amyloid diseases^1–4. Typically, each amyloid disease is linked to the pathological aggregation of one or two specific amyloid proteins^{2, 5}. For instance, Alzheimer’s disease (AD) involves the aggregation of amyloid-β (Aβ) and tau^{4, 6}, while type 2 diabetes (T2D) is associated with the assembly of human islet amyloid polypeptide (hIAPP)^{2, 7, 8}, and Parkinson’s disease (PD) is characterized by α-synuclein aggregation^{9, 10}. However, it is common to find the co-presence of multiple amyloid disease-related peptides within the same pathological tissues or organs (e.g., hIAPP in AD plaques¹¹, Aβ in PD Lewy bodies¹²), as well as the co-occurrence of multiple amyloid diseases in the same individual (e.g., AD with PD¹³, T2D with AD¹⁴). For example, while hIAPP is primarily associated with T2D, it is also found in the cerebrospinal fluid and brains of individuals with AD^15–17. Furthermore, hIAPP has been identified in the grey matter of the temporal lobe in diabetic patients, and deposits of hIAPP mixed with Aβ were often observed¹¹. Extensive clinical and epidemiological studies have demonstrated that individuals with T2D are at an significantly elevated risk of developing AD¹⁸. Increasing evidence suggests that the cross-talk between the co-localized Aβ and hIAPP might play a pivotal role in linking AD and T2D clinically^{11, 14, 19, 20}.

Aβ peptides, primarily derived from the cleavage of the amyloid precursor protein by β- and γ-secretases in the human brain, mainly constitute senile plaques in AD⁵. These plaques predominantly comprise Aβ40 and Aβ42 isoforms, with Aβ42 exhibiting heightened aggregation and toxicity^{4, 18}. Meanwhile, 37-amino acid hIAPP originates from the proteolytic processing of the 89-amino acid islet amyloid precursor protein⁷. This peptide is co-synthesized, co-secreted, and co-stored with insulin in the pancreatic β-cells found in the islets of Langerhans. Although Aβ and hIAPP have distinct primary sequences, they share a common all-or-none sigmoidal aggregation kinetics, characterized by three phases: the initial lag phase due to the nucleation of oligomers and proto-fibrils, the rapid growth phase via fibril elongation, and the final saturation phase after forming mature fibrils^{2, 5, 6, 8, 21}. Experimental findings also suggest that both Aβ and hIAPP fibrils share a common cross-β structure, where β-strands are aligned perpendicular to the fibril axis and pack against each other to form the hydrophobic cross-β core^{22, 23}. Similar to other amyloid peptides, soluble β-sheet-rich oligomers of Aβ and hIAPP formed in early aggregation are significantly more toxic than their mature fibril counterparts^{6, 24, 25}. While the precise structure of toxic oligomers and the corresponding toxicity mechanism are yet to be definitively determined, recent experimental and computational data suggest that β-barrel intermediates might act as the cytotoxic species in amyloidosis^{22, 23}. For both Aβ and hIAPP, whether in fragments and full-length forms, the formation of β-barrel intermediates during their aggregation have been reported^{4, 6, 8, 24, 26}.

Considering the co-existence of hIAPP and Aβ in the human brain and pancreas^{11, 15–17}, alongside the clinical overlaps between AD and T2D^{15, 19}, extensive studies have investigated the interplay between these peptides, encompassing co-aggregation and cross-seeding phenomena^{19, 27}. hIAPP aggregates more rapidly than Aβ at equivalent concentrations, displaying significantly shorter lag phases in vitro^{20, 28}. Co-aggregation experiments reveal that a mixture of soluble Aβ and hIAPP exhibits a shorter aggregation lag phase than Aβ alone, albeit slightly longer than IAPP alone at the same total peptide concentrations²⁹. Fibrils formed by both Aβ and hIAPP can act as seeds, mutually promoting each other’s amyloid aggregation, though less efficiently than their self-seeding processes^{30, 31}. Computational simulations have extensively explored the molecular interactions underlying this cross-talk, including possible interactions between Aβ and hIAPP fibrils^{27, 32–34}, hetero-dimerization of Aβ and hIAPP^{28, 35}, and the co-aggregation of amyloidogenic fragments from both peptides^{28, 35}. Notably, hot-spot binding regions between Aβ and hIAPP have been identified in vitro and in silico, involving Aβ residues 11–21 and 30–42, and hIAPP residues 8–20 and 21–37^{28, 35, 36}. Moreover, recent computational simulations have shown that a mixture of Aβ_16–22 and hIAPP_20–29 fragments can readily form β-barrel oligomers, prompting inquiries about whether full-length Aβ and hIAPP can also form similar β-barrel structures³⁷. Previous computational studies on Aβ and hIAPP co-aggregation primarily focused on either the structures and dynamics of full-length hetero-dimers^{28, 35} or the co-oligomerization of multiple short fragments^{37, 38}. As we have showed previously, the structural dynamics of oligomers in amyloid aggregation is highly dependent on the oligomer size^38–40 – e.g., hIAPP_8–20 underwent α-helix to β-sheet transition only when the aggregates were at least hexamers⁴⁰. Hence, further investigation of the co-aggregation of multiple full-length Aβ and hIAPP molecules, as well as their cross-seeding interactions, is crucial for a more detailed understanding of their cross-talk underpinning the co-pathology of AD and T2D.

To explore the cross-talk between Aβ and hIAPP in amyloid aggregation, we conducted computational studies on the co-aggregation of three Aβ peptides with three hIAPP peptides, and the cross-seeding dynamics of Aβ monomers interacting with hIAPP fibrils, and vice versa, using atomistic discrete molecular dynamics (DMD) simulations. DMD, a rapid and predictive molecular dynamics algorithm, has been extensively employed in studying protein folding, misfolding and amyloid aggregation^41–47. Our simulations revealed that three Aβ and three hIAPP peptides spontaneously co-aggregated into β-sheet-rich hetero-complexes, including β-barrel oligomers. Intriguingly, the co-aggregation of Aβ and hIAPP did not hinder their ability to recruit additional peptides, thereby facilitating the growth of larger aggregates rich in β-sheet structures. Hot-spot regions driving the self-assembly of Aβ and hIAPP also promoted their co-aggregation. Cross-seeding simulations between Aβ and hIAPP demonstrated that their fibrils could mutually act as templates, assisting each other’s monomers in converting into β-sheets at the fibril elongation ends. The phenomenon of dock-lock fibril growth in the seeding process should be template-dependent but not highly sensitive to the amino acid sequence of the template peptides. Therefore, targeting the elongation edges around the amyloidogenic core region could serve as a promising strategy to modulate the pathological aggregation of amyloid proteins in degenerative diseases, since the binding-induced extending or capping is not highly sensitive to the amino sequence of the template peptides^{44, 48, 49}.

2. Methods and materials

2.1. Molecular systems setup.

To explore the conformational dynamics of the co-aggregation of multiple Aβ and hIAPP molecules, we conducted computational simulations involving three Aβ_1–42 peptides (referred to as Aβ) mixed with three IAPP_1–37 peptides (referred to as hIAPP). The amino acid sequences and initial structures of Aβ (PDB ID: 1z0q⁵⁰) and hIAPP (PDB ID: 5mgq⁵¹), characterized by NMR measurements and widely used as the initial monomer structures in previous computational simulation studies^{6, 8}, are illustrated in Figure 1. Three Aβ and three hIAPP peptides were randomly placed in a 10.0 nm cubic simulation box in various orientations, ensuring a minimum intermolecular atomic distance of at least 1.5 nm. Sixty independent simulations were conducted, each starting with different initial states, including coordinates, orientations, and velocities. To prevent potential bias, a minimum inter-peptide distance of no less than 1.5 nm was maintained for each distinct initial structure. Each independent simulation lasted for a duration of up to 600 ns.

Figure 1. Amino acid sequences and initial structures of Aβ and hIAPP used in our simulations.

The amino acid sequences of Aβ and hIAPP are depicted (a). Residues with high sequence identity and similarity between Aβ and hIAPP are shaded in blue and purple, respectively. Experimentally determined structures of monomers and fibers of Aβ_1–42 and hIAPP_1–37 utilized in our simulations are shown (b&c). Both Aβ and hIAPP proto-fibrils are composed of 20 peptides. Side-chains are represented as sticks and colored by residue type: hydrophobic residues (white), negatively charged residues (red), positively charged residues (blue), and polar residues (green).

To explore cross-seeding between Aβ and hIAPP, we conducted simulations for two molecular systems: one involving a preformed Aβ fibril interacting with an isolated hIAPP, and the other with a preformed hIAPP fibril interacting with an isolated Aβ. For each system, we ran fifty independent DMD simulations, each lasting 600 ns. The fibril models used in our simulations were derived from the full-length Aβ proto-fibril determined by the Schröder group using cryo-EM (PDB ID: 5oqv²²) and the full-length IAPP protofibrils⁵² proposed by the Tycko group through solid-state NMR. The reason we chose this hIAPP fibril model⁵², rather than the other recent cryo-EM determined models^{53, 54}, is because only this model included all of the hIAPP residues, while the other model missing the dynamic residues. Based on the aforementioned models, initial structures for cross-seeding simulations were constructed with 20 peptides comprising Aβ and hIAPP fibril seeds. The isolated Aβ and hIAPP monomers were randomly placed more than 1.5 nm away from the hIAPP and Aβ fibril seeds in a 12.0 nm cubic simulation box, respectively, as the initial states. For each cross-seeding molecular system, fifty different initial structures were generated to avoid potential bias from one specific initial structure. Detailed information regarding simulations of each molecular system can be found in Table 1.

Table 1.

Comprehensive details of each molecular system involved in our DMD simulations are provided, including the number of Aβ and hIAPP peptides in co-aggregation and cross-seeding simulations (Num. Peptide), dimensions of the cubic simulation box (Box size), the quantity of independent DMD simulation trajectories conducted (DMD run), the duration of each DMD simulation (Time), and the cumulative total simulation time (Total time).

System	Num. Peptide	Box size (nm)	DMD run	Time (μs)	Total time (μs)
Aβ & hIAPP	3+3	10.0	60	0.60	36.0
Aβ fibril & hIAPP	20+1	12.0	50	0.60	30.0
hIAPP fibril & Aβ	20+1	12.0	50	0.60	30.0

2.2. Discrete molecular dynamics (DMD) simulation.

All simulations were carried out at 300 K using the atomistic DMD algorithm^{55, 56} coupled with the Medusa force field⁵⁷. The Medusa force field has been benchmarked to ensure accurate predictions of protein stability changes due to mutations and protein–ligand binding affinities^56–58. Both DMD and MD adhere to the same physical laws, with their main distinction lying in the discretization of interatomic interaction potentials derived from MD force fields. Bonded interactions, including bonds, bond angles, and dihedrals, were modeled using infinite square wells. Specifically, covalent bonds and bond angles typically had a single well, while dihedrals could exhibit multiple wells corresponding to cis- or trans-conformations. The van der Waals parameters utilized were sourced from the CHARMM force field⁵⁹, while solvation effects were simulated through the EEF1 implicit solvent model⁶⁰. Explicit handling of hydrogen bond interactions was achieved via a reaction-like method⁵⁶. Electrostatic interactions were simulated with the Debye–Hückel approximation, where the Debye length was set to approximately 10 Å to account for screened interactions. Temperature control was maintained using Anderson’s thermostat algorithm⁶¹. Thorough validation of the accuracy of the Medusa force field, coupled with the EEF1 implicit solvation model, has been conducted by comparing it with standard MD simulations utilizing various force fields (e.g., GROMOS96, OPLS-AA, AMBER99SB-ILDN, and CHARMM36m), particularly focusing on investigating the aggregation behaviors of both functional suckerin amyloid peptides⁶² and pathological hIAPP peptides²⁶. Due to its notable sampling efficiency for studying protein folding and amyloid aggregation processes, the DMD algorithm has been widely employed by our group^41–44 and others^45–47.

2.3. Analysis methods

The dictionary of secondary structures of proteins (DSSP) method⁶³ was used for the secondary structures analysis. Consistent with our prior research, a hydrogen bond was deemed formed when the distance between backbone N and O atoms was ⩽ 3.5 Å, and the N – H • • • O angle measured ⩾ 120 °^{26, 49}. Residues were considered in contact if the minimum heavy atom distance between non-sequential residues was less than 0.55 nm. The frequency of residue-pairwise contact maps is depicted as a two-dimensional matrix, showcasing the propensity between each pair of residues forming contacts across all conformations utilized for data analysis. Peptides linked by at least one inter-peptide contact were grouped as part of the same oligomer. A two-dimensional (2D) free energy surface was crafted using the formula -RTlnP(x, y), where P(x, y) represents the likelihood of specific reaction coordinates (x and y). Oligomers were categorized as β-barrel structures if the β-strand segments within them could form a closed loop, with each β-strand linked to two neighboring β-strands by a minimum of two main-chain hydrogen bonds^{6, 8}.

3. Results and discussion

3.1. Aβ and hIAPP co-aggregated into β-sheet-rich hetero-complexes by continuously recruiting additional peptides.

The co-aggregation dynamics of three Aβ peptides mixed with three hIAPP peptides were monitored by tracking the time evolution of the oligomer size for each peptide aggregated, the number of inter-peptide contacts and hydrogen bonds, along with the ratio of each secondary structure (Figure 2a–c). The analysis of co-aggregation dynamics revealed that irrespective of the initial dimer formation by two Aβ, two hIAPP, or one Aβ mixing with one hIAPP, during the early stages of aggregation, all three Aβ and three hIAPP peptides co-aggregated into a single β-sheet-rich hetero-hexamer. Given the amyloidogenic nature of both Aβ and hIAPP peptides^{4, 64}, with the ability to self-aggregate into oligomers and fibrils, the observation of Aβ and hIAPP homo-oligomers or hetero-oligomers recruiting isolated Aβ or hIAPP to grow into large β-sheet oligomers indicated that the co-aggregates shared a similar aggregation capability, recruiting isolated peptides to facilitate oligomerization into β-sheet aggregates. Such co-aggregation phenomena were also observed in fragment simulations of Aβ_16–22 mixing with hIAPP_22–29³⁸, as well as in full-length simulations of Aβ mixing with hIAPP²⁸. Prior electrospray ionization-ion mobility spectrometry-mass spectrometry (ESI-IMS-MS) measurements revealed that the oligomeric species co-aggregated by hIAPP and Aβ at a 1:1 molar ratio were highly heterogeneous at early aggregation stages²⁹. This heterogeneity stemmed from the ability of both peptides to recruit themselves and each other, occurring in both monomeric and oligomeric states as observed in our simulations (Figure 2).

Figure. 2. Co-aggregation dynamics and conformational analysis for simulations of three Aβ mixing with three hIAPP peptides.

The co-aggregation dynamics and conformations are monitored by the time evolution of the oligomer size for each peptide aggregated into (first row), the number of inter-peptide contacts and hydrogen bonds (second row), and the average ratio of each secondary structure (third row). Four representative snapshots corresponding to the initial co-aggregation and saturated stages are shown, with the corresponding simulation time stamped below (fourth row). Three trajectories illustrating the co-aggregation initiated by forming Aβ homo-dimer (a), hIAPP homo-dimer (b), and Aβ-hIAPP hetero-dimer (c) are randomly selected from the simulation pool of 60 independent DMD runs. For clarity, the peptides of Aβ and three hIAPP in the co-aggregates are colored blue and red, with the first Cα atom of each peptide highlighted.

The formation of inter-peptide contacts was accompanied by an increase in inter-chain backbone hydrogen bonds and a growth of the β-sheet ratio (Figures 2a–c and S1a–c), indicating that inter-peptide interactions drove Aβ and hIAPP co-aggregation into β-sheet conformations. All three Aβ and three hIAPP peptides spontaneously coalesced into a stable hetero-complex, which showed no significant changes in the total number of inter-peptide contacts and hydrogen bonds, as well as the average content of each type of secondary structures (including the random coil and bend composing unstructured conformations, β-sheet, and helix) during the last 200 ns. These results suggested that the co-aggregation simulations reached a steady state, further supported by the final snapshots without major conformational changes observed during the last 200 ns for six randomly selected trajectories out of the pool of 60 independent DMD runs (Figures 2 and S1). The time evolution of the number of inter-peptide contacts and hydrogen bonds among Aβ (Aβ-Aβ), among hIAPP (hIAPP-hIAPP), and between Aβ and hIAPP (Aβ-hIAPP) for each independent trajectory also displayed signs of reaching equilibrium during the last 200 ns (Figure S2). Therefore, the simulation data from the last 200 ns of each independent trajectory were used for the conformational analysis of the hetero-complex co-aggregated by three Aβ mixed with three hIAPP.

The time evolution of the largest oligomer size for the mixture of three Aβ peptides and three hIAPP peptides in each independent DMD simulation showed that all the peptides readily aggregated into a single hetero-hexamer within 100 ns, although an extremely small proportion of trajectories exhibited transient dynamic dissociation and association (Figure 3a). The mass-weighted oligomer size distribution for the saturated states demonstrated that three Aβ and three hIAPP peptides predominantly aggregated into one single hetero-hexamer with a propensity greater than 90.4% (Figure 3b).

Figure. 3. Co-aggregation dynamics and the free energy landscape, along with structural analysis, in simulations involving three Aβ peptides mixed with three hIAPP peptides.

The time evolution of the largest oligomer size for three Aβ and three hIAPP peptides aggregated into each independent simulation (a). The mass-weighted oligomer size probability distribution for the hetero-complex formed by three Aβ and three hIAPP peptides during the saturated states of the last 200 ns from 60 independent 600 ns DMD trajectories (b). The aggregation free energy landscape as a function of total potential energy against the total number of inter-peptide contacts (left panel) and hydrogen bonds (right panel) (c). Four snapshots labeled 1–4 on the free energy landscape surface illustrate the co-aggregation process of forming a high content of β-sheet conformations from isolated monomer states. For clarity, the Aβ and hIAPP in the co-aggregates are colored blue and red, with the first Cα atom of each peptide highlighted. To capture the entire co-aggregation process, all 600 ns simulation data from each trajectory are used for the above aggregation free energy analysis. The radius distribution function (RDF) for the Cα atoms from Aβ and hIAPP peptides corresponding to the geometry center of their hetero-aggregates based on the last 200 ns simulation data (d). The average secondary structure content of Aβ and hIAPP within their co-aggregates during the saturated states (e). The propensity of each residue from Aβ (upper) and hIAPP (bottom) assumes unstructured, β-sheet, and helix conformations within their saturated co-assemblies (f).

The aggregation potential of mean force (PMF), also known as the free energy landscape, was analyzed based on the total potential energy against the corresponding number of inter-peptide contacts and hydrogen bonds, using the entire dataset from 60 independent 600 ns DMD simulations to capture the entire aggregation process (Figure 3c). The aggregation free energy landscape revealed that an increase in inter-peptide contacts led to a decrease in potential energy, accompanied by an increase in forming inter-peptide backbone hydrogen bonds (corresponding to the formation of inter-peptide β-sheets, as illustrated by snapshots 1–4 in Figure 3c). Additionally, conformations with a higher number of inter-peptide hydrogen bonds showed lower potential energy, suggesting that the conversion to β-sheets was energetically favorable. The aggregation free energy landscape, as a function of potential energy and average β-sheet ratio, further confirmed that the formation of β-sheets results in a decrease in potential energy (Figure S3). Conformations of Aβ and hIAPP hetero-aggregates located in the lowest free energy basin with the β-sheet ratio ~33%−55% for Aβ and ~18%−40% for hIAPP (Figure S3). However, conformations with an exceptionally high content of β-sheets exhibited lower potential energy but higher free energy (Figures 3c&S3). This could be attributed to the reduction in potential energy during the formation of ordered β-sheets not being sufficient to offset the loss of entropy in free energy before overcoming the barrier for amyloid fibril formation.

The radial distribution functions (RDF) of Cα atoms from either Aβ or hIAPP (Figure 3d) suggested that residues of Aβ were more buried than those of hIAPP inside the hetero-complex. Interactions among peptides within the hetero-complex were thoroughly investigated by analyzing the distributions of Aβ-Aβ, hIAPP-hIAPP, and Aβ-hIAPP contacts and hydrogen bonds (Figure S4). Notably, Aβ-hIAPP interactions were more prevalent than Aβ-Aβ interactions, with interactions of hIAPP-hIAPP being the least frequent. The conformational free energy landscape, depicted as a function of backbone hydrogen bonds to estimate the inter-peptide β-sheet formations, suggested that the Aβ-hIAPP and Aβ-Aβ interactions played a crucial role in the formation of β-sheets in the co-aggregates (Figure S5).

The average secondary structure content of the saturated hetero-complex indicated that both Aβ and hIAPP mainly adopted unstructured and β-sheet conformations, but Aβ exhibited a much stronger β-sheet propensity compared to hIAPP (Figure 3e). Specifically, the average unstructured and β-sheet propensity of Aβ was ~27.1% and ~53.2%, respectively, while for hIAPP, these values were ~39.2% and ~32.8%, respectively. The observation that Aβ displayed a significantly stronger β-sheet tendency than hIAPP (Figures 3e&S5) aligns with previous studies on the self-aggregation and hetero-dimerization of Aβ and hIAPP in silico^{6, 8, 35, 65, 66}. For example, prior all-atom explicit-solvent replica exchange molecular dynamics (REMD) simulations suggested that the β-sheet population of Aβ was notably greater than that of hIAPP, both in homo- and hetero-dimers³⁵.

The average secondary structure propensity suggested that β-sheet conformations were primarily formed by residues 10–21 and 30–42 of Aβ and residues 10–30 of hIAPP (Figure 3f). Intriguingly, the experimentally determined fibril model also exhibited a similar β-sheet tendency^{53, 54, 67, 68}. For instance, residues 1–10 were too dynamic to be determined in most Aβ fibril models^{67, 68}. Recently, cryo-EM-determined hIAPP models were mainly composed of residues 10–37, as the missing residues were too flexible to be measured^{53, 54}. Additionally, the β-sheet-populated regions of Aβ and hIAPP observed in our DMD simulations were consistent with prior classical MD and REMD simulations^{65, 69–71}. Overall, our simulation results demonstrated that residues of Aβ and hIAPP involved in forming the β-sheet core of their corresponding amyloid fibrils also participated in forming β-sheet structures in their co-oligomerization.

3.2. The amyloidogenic core regions, crucial for the self-aggregation of Aβ and hIAPP, acted as the hot-spots driving their co-aggregation.

The structural characteristics of Aβ and hIAPP within their hetero-aggregates were further examined by computing the residue-pairwise contact frequency map among main-chain atoms (Figure 4). Within the hetero-complex, Aβ exhibited three distinct β-hairpin contact patterns perpendicular to the diagonal in the intra-chain contact frequency map, characterized by a strand-turn-strand motif, mainly stabilized by interactions among hydrophobic residues (snapshots 1–3 in Figure 4a). One prominent long β-hairpin motif with β-strands around residues 10–21 and 30–41, alongside two less populated β-hairpin motifs formed by residues 1–18 and 29–42. Such a typical β-hairpin structure formed by a similar region was also observed in simulations of Aβ monomers and oligomers^{6, 49}. An extended β-hairpin structure formed around residues 17–35 was also supported by experimental measurements⁷². The inter-peptide interactions between Aβ peptides exhibited various parallel (Aβ_30–41 vs Aβ_31–42, Aβ_15–18 vs Aβ_39–42) and anti-parallel (Aβ_12–19 vs Aβ_18–25, Aβ_38–42 vs Aβ_31–35, Aβ_29–41 vs Aβ_8–20) inter-peptide β-sheets stabilized by hydrophobic residues around Aβ_10–21 and Aβ_30–42 (snapshots 4–8 in Figure 4a). The average number of Aβ-Aβ inter-peptide contacts within the hetero-complex for residues around Aβ_10–21 and Aβ_30–42 was greater than in other regions, indicating that these regions were the hot-spots driving the aggregation among Aβ (Figure S6a).

Figure 4. Residue-pairwise interaction analysis among Aβ and among hIAPP within their hetero-aggregates.

The residue-pairwise contact frequency between inter-chain (top diagonal) and intra-chain (bottom diagonal) main-chain heavy atoms for three Aβ peptides (a) and three hIAPP peptides (b) within their co-aggregates. Representative structured motifs with high contact frequency patterns, mainly corresponding to helices or β-sheets, are labeled as 1–8 and displayed on the right. The contact frequency map is generated using the last 200 ns trajectories of 60 independent DMD simulations after reaching a steady state. To enhance clarity, Aβ peptides and hIAPP peptides are colored in blue and red, respectively. The amino acid sequence representing the motifs is color-highlighted by residue type: hydrophobic residues (black), negatively charged residues (red), positively charged residues (blue), and polar residues (green).

Intra-peptide interaction analysis of hIAPP within the hetero-hexamer revealed a β-hairpin motif involving residues 7–18 vs 21–32, along with a partial short-helix contact pattern around N-terminal residues 1–10 and C-terminal residues 30–37. This β-hairpin motif was also observed in previous hIAPP monomer and oligomer simulations, suggesting its role as the amyloidogenic precursor of hIAPP, both in silico and in vitro^{73, 74}. Inter-peptide residue-pairwise interactions among hIAPP peptides within the hetero-hexamer exhibited diverse inter-peptide parallel (e.g., hIAPP_21–27 vs hIAPP_23–29, hIAPP_8–14 vs hIAPP_21–27, hIAPP_11–23 vs hIAPP_11–23) and anti-parallel (e.g., hIAPP_14–23 vs hIAPP_24–33, hIAPP_11–18 vs hIAPP_15–22) β-sheets (snapshots 4–8 in Figure 4b). Residues 10–29 featured a greater average number of inter-peptide hIAPP-hIAPP contacts than other regions, indicating that this region served as the hot-spot driving the aggregation of hIAPP (Figure S6b). The independent formation of β-sheet-rich fibrillar structures for the segments of hIAPP_8–20^{40, 75}, hIAPP_15–25^{76, 77}, hIAPP_19–29^{76, 77}, hIAPP_22–29³⁸, both in vitro and in silico, also suggested the propensity of these regions to contribute to the overall aggregation process.

To identify the key interactions driving the co-aggregation between Aβ and hIAPP, the inter-peptide residue-pairwise contact frequency between Aβ and hIAPP was also analyzed (Figure 5). The hot-spot regions, Aβ_10–21 and Aβ_30–42, which drove the self-aggregation of Aβ, also displayed a strong tendency to interact with the amyloidogenic core regions of hIAPP (e.g., hIAPP_8–20 and hIAPP_22–29) to form various inter-peptide β-sheets (Figure 5a&b). The residue-pairwise interaction analysis for the representative β-sheet contact patterns suggested hydrophobic interactions between residues ¹⁷LAFFA²¹ and ³⁰AIIGLMVGGVVIA⁴² of Aβ with residues ¹²LANFLV¹⁷ and ²³FGAIL²⁷ of hIAPP, which were the main forces stabilizing the inter-chain β-sheet between Aβ and hIAPP (snapshots 1–10 in Figure 5b). The average number of inter-chain Aβ-hIAPP contacts for each residue suggested that Aβ residues 9–22 and 29–42, as well as hIAPP residues 8–29, served as the hot-spots for the co-aggregation between Aβ and hIAPP (Figure 5c). Notably, segments pivotal in the self-aggregation of Aβ and hIAPP also played a crucial role in their co-aggregation. Interestingly, the identified hot-spot regions facilitating the cross-association of Aβ and hIAPP in our simulations aligned with experimentally identified binding-affinity regions for Aβ-hIAPP interaction³⁶.

Figure. 5. The inter-peptide interaction analysis between Aβ and hIAPP in the co-aggregates formed by three Aβ and three hIAPP peptides.

The inter-peptide residue-pairwise contact frequency between Aβ and hIAPP within their hetero-aggregates is computed based on the last 200 ns trajectories of 60 independent DMD simulations after reaching a steady state (a). The representative binding motif segments, labeled as 1–10, corresponding to the parallel or antiparallel β-sheet patterns highlighted by boxes in the contact frequency map, are also presented (b). For clarity, the Aβ and hIAPP in the co-aggregates are colored in blue and red, respectively. The sequence of Aβ and hIAPP is tagged in blue and pink, respectively. The amino acid sequences in the representing motifs are color-highlighted by residue type: hydrophobic residues (black), negatively charged residues (red), positively charged residues (blue), and polar residues (green). To identify the hotspot binding region between Aβ and hIAPP, the average total number of intermolecular contacts between Aβ and hIAPP for each residue is also computed by integrating the corresponding pairwise contact frequency map (c).

3.3. Co-aggregation of Aβ and hIAPP led to the formation of potentially toxic β-barrel hetero-complexes.

The β-barrel oligomer, initially discovered in an 11-residue segment of αB crystallin and frequently observed in toxic amyloid fragments⁷⁸ and full-length peptide aggregation simulations^{6, 26, 79, 80}, was proposed as a potentially cytotoxic oligomer in amyloidosis. Our prior simulation studies have demonstrated that the β-barrel oligomer was a common intermediate in the fibrillization of toxic amyloid segments (e.g., NACore⁸¹, hIAPP_8–20⁴⁰, hIAPP_19–29⁷⁶, SOD1_28–38³⁹), but absent in the aggregation of non-toxic amyloid fragments (e.g., hIAPP_15–25⁷⁶, SOD1_28–38^G33W39, SOD1_28–38^G33V39). A recent study directly illustrated that the creation of a β-barrel oligomer caused membrane leakage in in silico⁸². The formation of β-barrel intermediates was also supported in the aggregation of full-length Aβ^{6, 24}, hIAPP^{8, 26}, human calcitonin (hCT)^{42, 79}, and medin⁸⁰ in silico and in vitro. The β-barrel structure formed by full-length Aβ was also determined by cryo-EM assay²⁴. The prevalence of β-barrel intermediates in wild-type Aβ exceeded that of the AD-protective A2T substitution but fell short of the AD-causing mutations D7N and E22G⁶. Meanwhile, the S20G substitution in hIAPP heightened its amyloidogenicity and cytotoxicity, significantly enhancing β-barrel formation compared to the wild-type peptide, while no β-barrels were observed in the non-toxic rat IAPP²⁶. Considering the co-pathology of AD and PD, as well as AD with aortic medial amyloidosis (AMA), the co-aggregation of Aβ with αS¹², as well as Aβ with medin⁴⁴, also resulted in the formation of a β-barrel in silico. These studies suggested that the β-barrel might act as the toxic species in amyloidosis.

The co-localization of Aβ and hIAPP, resulting in their cross-talk in amyloid aggregation, raises a question of whether the full-length Aβ and hIAPP could also form β-barrel formations. In computational simulations exploring the co-aggregation of full-length Aβ and hIAPP through their hetero-dimerization^{28, 35}, no β-barrel intermediates were identified due to the limited size of the simulated peptides, which prevented them from attaining the necessary minimum size for a β-barrel structure^{6, 8, 28}. Interestingly, the β-barrel intermediates observed in the self-assembly of Aβ and hIAPP were also evident in their co-aggregations (Figure 6a&b). Within the β-barrel hetero-hexamer, each β-strand formed by either Aβ or hIAPP was inter-connected by two neighboring β-strands, creating a closed cycle. The probability of β-barrel formations in each independent DMD trajectory was very heterogeneous, with more than 10 out of 60 trajectories exhibiting a population of β-barrel formation greater than 5.0% (Figure 6c). The distribution of β-sheet hetero-aggregates on the surface of their co-aggregation free energy landscape showed that the presence of β-sheet formations mainly occurred in the process of Aβ and hIAPP converting to high β-sheet conformations (Figure S7). Within the β-barrel hetero-hexamer, Aβ peptide exhibited a much stronger β-sheet tendency than hIAPP (Figure 6d). Our direct observation of the potentially toxic β-barrel hetero-complex formed by the interaction of Aβ with hIAPP may offer insight into clinical observations indicating an increased risk of developing AD among individuals with T2D¹⁸.

Figure. 6. Analysis of β-barrel formation during the co-aggregation of three Aβ peptides mixed with three hIAPP peptides.

Two of the most populated β-barrel formation trajectories are selected from 60 independent trajectories to illustrate the co-aggregation dynamics (a&b). The dynamics of co-aggregation are monitored through the time evolution of the number of inter-peptide Aβ-Aβ, hIAPP-hIAPP, and Aβ-hIAPP hydrogen bonds (first row), along with the average ratio of each secondary structure (second row). The regions with the formation of β-barrel structures are illustrated by a gray shadow according to the simulation time. Snapshots for each trajectory are also presented every 150 ns (last row). For clarity, the peptides of Aβ and hIAPP in the co-aggregates are colored in blue and red, with the first Cα atom of each peptide highlighted. The frequency of β-barrel oligomers observed in the 60 independent DMD trajectories is sorted in descending order based on the probability of β-barrel formation (c). The probability distribution function (PDF) of the β-sheet ratio for each Aβ and hIAPP in the β-barrel structure is shown (d). All 600 ns of simulation data are utilized for the β-barrel analysis, as β-barrel formation is observed before reaching the saturation stage.

3.4. Preformed fibrils of Aβ and hIAPP assisted each other’s monomers in converting into β-sheets at the fibril elongation ends during cross-seeding.

The concurrent presence of Aβ and hIAPP^{11, 15–17} facilitated their co-aggregation and initiated cross-seeding between them^29–31. Subsequent simulations aimed to further investigate their cross-seeding. In one scenario, a preformed Aβ fibril interacted with an isolated hIAPP monomer (Figure 7), while in another, a preformed hIAPP fibril interacted with an isolated Aβ monomer (Figure 8). To minimize potential biases from the initial state, 50 distinct initial structures were generated for each molecular system, with the isolated monomer randomly positioned more than 1.5 nm away from the fibril.

Figure 7. Binding and conformational dynamics of an hIAPP monomer to a preformed Aβ fibril seed.

The interactions between the hIAPP monomer and the Aβ fibril are tracked through the time evolution of the number of backbone hydrogen bonds and residue-pairwise contacts between hIAPP and Aβ (left panel), along with the secondary structure of each residue from the hIAPP monomer (middle panel) (a–c). The corresponding snapshots at 0, 200, 400, and 600 ns are displayed on the right panel. Three representative trajectories are randomly selected from 50 independent DMD runs. The potential of mean force (PMF) is presented as a function of the number of residue-pairwise contacts and backbone hydrogen bonds formed between hIAPP and Aβ (d). Four representative structures labeled 1–4 in the PMF are also depicted below. To encompass the entire cross-seeding process, all simulation data are used for the PMF analysis. The binding interaction between hIAPP and the Aβ fibril is depicted by the residue-pairwise contact frequency map based on the last 200 ns simulation data from all independent trajectories (e). For clarity, the Aβ fibrils are shaded light blue to represent the environment, while the hIAPP monomer is highlighted in pink.

Figure 8. Binding and conformational dynamics of Aβ with a preformed hIAPP fibril seed.

The interactions between one isolated Aβ monomer and preformed hIAPP fibril are monitored through the time evolution of the number of backbone hydrogen bonds and residue-pairwise contacts between Aβ and hIAPP (left panel), along with the secondary structure of each residue from the Aβ monomer (middle panel) (a–c). The corresponding snapshots at 0, 200, 400, and 600 ns are displayed on the right panel. Three representative trajectories are randomly selected from 50 independent DMD runs. The potential of mean force (PMF) is presented as a function of the number of residue-pairwise contacts and backbone hydrogen bonds formed between Aβ and hIAPP (d). Four representative structures labeled 1–4 in the PMFs are also shown below. To capture the whole cross-seeding process, all simulation data are used for the PMF analysis. The binding interaction between Aβ and the hIAPP fibril is characterized by the residue-pairwise contact frequency map according to the last 200 ns simulation data from all the independent trajectories (e). For clarity, the hIAPP fibrils are colored pink to represent the environment, while the Aβ monomer is highlighted in blue.

The examination of interaction and conformational dynamics of the hIAPP monomer with a preformed Aβ fibril seed involved analyzing the time evolution of the number of inter-peptide contacts and hydrogen bonds, as well as the secondary structure of each residue within the hIAPP monomer, along with snapshots captured during the binding process (Figure 7a–c). Due to differences in the initial conditions across each independent simulation, the isolated hIAPP monomer exhibited two primary binding scenarios: it either initially adhered to the lateral surface and then transitioned to the elongation surface of the Aβ fibril, or directly attached to the elongation surface. Ultimately, the isolated hIAPP monomer underwent a conformational change to form β-sheets, thereby extended to the Aβ fibril elongation ends and contributing to the formation of hetero-fibrillar aggregates. The PMF was analyzed in relation to the total number of inter-peptide heavy atom contacts versus inter-molecular backbone hydrogen bonds formed between hIAPP and the Aβ fibril throughout the entire binding process (Figure 7d). Two distinct low-energy basins characterized the cross-seeding process. Initially, a narrow and shallow basin was observed with a small number of inter-peptide contacts (~50–100) but lacking inter-peptide backbone hydrogen bonds (~0–2) (labeled as 1 on the PMF surface in Figure 7d). Subsequently, a broader and deeper basin emerged, featuring a higher number of inter-peptide contacts (~80–130) and more inter-peptide backbone hydrogen bonds (~7–15). This indicated a favorable formation of inter-peptide β-sheets of hIAPP with the Aβ fibril at the elongation ends, driven by lower free energy (labeled as 2 on the PMF surface in Figure 7d). Conformations characterized by ~15–25 inter-peptide backbone hydrogen bonds, suggesting extensive interaction between hIAPP residues and Aβ at the fibril end in the form of β-sheets (identified as 3 and 4 on the PMF surface in Figure 7d), were noted to have slightly elevated free energy, possibly attributable to the loss of entropy as hIAPP adopted a more structured conformation.

The residue-pairwise contact frequency was also analyzed based on the last 200 ns of saturated conformational data (Figure 7e). Residues from the amyloidogenic core regions of hIAPP (including hIAPP_8–20⁴⁰ and hIAPP_22–29³⁸) displayed a strong tendency to extend the Aβ fibril ends around Aβ_10–21 and Aβ_30–42 as β-sheet formations through hydrophobic interactions (Figure 7e). Additionally, the electrostatic attraction among charged residues could also induce the hIAPP N-terminal residues to assume a helical structure attaching to the Aβ fibril lateral surface.

The conformational free energy landscape suggested that an increase in the number of inter-peptide contact resulted in the formation of more inter-chain backbone hydrogen bonds and a decrease in free energy (Figure 7d). This could be attributed to the fact that the formation of inter-peptide contacts and backbone hydrogen bonds may result in a decrease in potential energy (Figure S8a). Despite the penalty for the loss of entropy from hIAPP becoming highly ordered β-sheet conformations, which may result in an increase in free energy, on average, the extension of hIAPP to the Aβ fibril ends exhibited more inter-peptide contacts and backbone hydrogen bonds with lower potential energy. Thus, binding to the elongation edges was more favorable than lateral surface bindings for the interactions between hIAPP monomer and Aβ fibrils.

Similar to how the Aβ fibril seeded the hIAPP monomer (Figure 7a–c), the Aβ monomer binding to the hIAPP fibril growth edges also exhibited the assumption of stable β-sheet conformations (Figure 8a–c). In this scenario, Aβ extended to the hIAPP fibril ends by forming backbone hydrogen bonds, participating in the growth of the fibril. Different from the metastable lateral surface binding and the predominant preference for attaching to the elongation end in the Aβ fibril seeding hIAPP monomer, both the lateral surfaces and elongation edges bindings of the Aβ monomer to the hIAPP fibril were favorable, as both corresponding states exhibited stable dynamic performances. The free energy landscape for the entire binding process exhibited the lowest free energy basin for conformations with a number of inter-peptide contacts ranging ~70–130 and a number of inter-peptide backbone hydrogen bonds ranging ~0–5 (state 1 in Figure 8d). Conformations with a number of inter-peptide backbone hydrogen bonds ranging ~5–15 (states 2–3 in Figure 8d), indicating the formation of inter-peptide β-sheet hydrogen bonds, also exhibited low free energy, albeit slightly higher than the states dominated by lateral surface binding (state 1 in Figure 8d). Conformations featuring the majority of residues from Aβ extending onto the ends of the hIAPP fibril as β-sheets were also noted, yet these displayed a high free energy state (labeled as 4 in Figure 8d).

The binding dynamics and conformational free energy landscape suggested that the states of Aβ binding to the lateral surface and growth edges of an IAPP fibril were favorable (Figure 8a–d). This was because the negatively charged residues D1, E3, D7, E11, E22, and D23 of Aβ were easily attracted by the positively charged residues R11 on the fibril surface grooves of hIAPP, with the other hydrophobic and aromatic residues of Aβ attracting with the F15 located as the neighbor of R11 on the hIAPP fibril surface (Figure 8e). The residues from Aβ_10–21 or Aβ_30–41 could attract the amyloidogenic core regions of hIAPP, especially hIAPP_22–29, to extend the fibril ends by forming inter-peptide β-sheet through hydrophobic interactions (snapshots 4–6 in Figure 8e). The PMF analysis, as a function of potential energy against the number of inter-peptide contacts and backbone hydrogen bonds, revealed that an increase in inter-peptide contacts resulted in a decrease in potential energy (Figure S8b). However, an increase in inter-chain hydrogen bonds did not induce significant changes in potential energy (Figure S8b). This finding supported the conclusion that the binding of Aβ to the hIAPP fibril was favorable with both lateral surface and growth edge regions binding, in contrast to the hIAPP monomer predominantly attaching to the Aβ fibril elongation edges.

3.5. The dock-lock fibril growth process seemed to depend on the existence of a template but showed limited sensitivity to the amino acid sequence of the template peptides.

Based on kinetic assessments of amyloid aggregation across different amyloid peptides, a dock-lock mechanism has been proposed for the addition of monomers at the ends of preexisting amyloid fibrils⁸³. For instance, increasing the amyloid fibril elongation edges via sonication significantly enhances the efficiency of fibril seeding, indicating a critical role of the fibril ends in the rapid elongation process of amyloid aggregation⁸⁴. This dock-lock mechanism is a common phenomenon observed during the rapid elongation stage, extensively studied both in vitro and in silico^{44, 83–85}. Our direct observation of hIAPP monomer extending to Aβ fibrils, as well as Aβ monomer binding to hIAPP fibrils, suggested that fibril seeds induced dock-lock behavior, forming inter-peptide β-sheets, should be template-dependent and not highly sensitive to the sequence details of the template peptides. For instance, Xi et al. demonstrated that regardless of the template formed by Aβ, hIAPP, or polyalanine, an enhancement of β-sheet formations was observed for Aβ monomers in silico⁴⁸. Additionally, our recent studies also showed that Aβ fibril templates could assist medin and tau peptides in converting into β-sheet conformations^{44, 86}. Indeed, cross-seeding between different amyloid-disease related peptides⁴⁴, as well as the triggering of pathological amyloid peptide abnormal aggregation by bacterial biofilms⁸⁷, also indicated that the seeding growth mechanism of amyloid fibrils should be template-dependent, but not highly sensitive to the amino sequence of the template peptides.

The inherent property of amyloid fibrils to attract isolated peptides, extending or capping them at the elongation ends, without high sensitivity to the amino acid sequence, also presents a promising strategy for the development of amyloid inhibitor designs. For example, αB-crystallin has been identified as an inhibitor of Aβ aggregation by effectively capping the β-sheet elongation edge^{88, 89}. The capping of SEVI (semen-derived enhancer of viral infection) at the Aβ fibril edge has proven to efficiently prevent the seeding efficiency of Aβ fibrils⁴⁹. Furthermore, a comparable strategy, known as “negative design,” has been utilized in the development of amyloid inhibitors by strategically positioning amyloid-resistant fragments along the edges of β-sheets^{90, 91}. Therefore, targeting the elongation edges could serve as a novel therapeutic strategy to modulate the pathological aggregation of amyloid proteins in degenerative diseases, as the binding-induced extending or capping is not highly sensitive to the amino sequence of the template peptides.

Conclusion.

Given the co-existence of hIAPP and Aβ in the human brain and pancreas^{11, 15–17}, along with the clinical overlaps between AD and T2D^{15, 19}, we systematically investigated the cross-talk between Aβ and hIAPP, including their co-aggregation and mutual cross-seeding, through comprehensive atomistic DMD simulations. Our simulation results demonstrated that the cross-interaction between Aβ and hIAPP drove them to readily aggregate into β-sheet-rich structures, including potentially toxic β-barrel oligomers. Interestingly, the formation of Aβ and hIAPP hetero-aggregates did not impede their ability to recruit additional peptides, thereby facilitating the growth of larger aggregates rich in β-sheet structures. The cross-seeding simulations between Aβ and hIAPP demonstrated that their fibrils could mutually act as templates, assisting each other’s monomers in converting into β-sheets at the fibril elongation ends. The phenomenon of dock-lock fibrillar growth in the seeding process was expected to be template-dependent but showed limited sensitivity to the amino acid sequence of the template peptides. Additionally, the amyloidogenic core region that drove the self-assembly of Aβ (e.g., Aβ_10–21 and Aβ_30–42) and hIAPP (e.g., hIAPP_8–20 and hIAPP_22–29) also promoted their co-aggregation of forming β-sheet aggregates and cross-seeding of forming fibrillar hetero-complexes. Since the amyloidogenic core regions of Aβ and hIAPP in both oligomeric and fibrillar states can recruit isolated peptides to extend along the β-sheet edges without being highly sensitive to the amino sequence, targeting these regions by inducing amyloid-resistant drug peptides may serve as a promising therapeutic strategy to modulate the pathological aggregation of amyloid proteins in degenerative diseases. Overall, our comprehensive simulations suggest that the interplay between Aβ and hIAPP, including their co-aggregation and cross-seeding, may contribute to their pathological aggregation, which in turn shed light on the co-occurrence of AD and T2D, providing valuable insights for future therapeutic development.

Data and Software Availability.

DMD simulation engine is available at Molecules In Action, LLC. (www.moleculesinaction.com). Initial conformations, input parameter and topology files for DMD simulation, and representative DMD output trajectories for each system are available (https://zenodo.org/records/11206527).