Computational Insights into the Aggregation Mechanism and Amyloidogenic Core of Aortic Amyloid Medin Polypeptide

Abstract

Medin amyloid, prevalent in the vessel walls of 97% of individuals over 50, contributes to arterial stiffening and cerebrovascular dysfunction, yet our understanding of its aggregation mechanism remains limited. Dividing the full-length 50-amino-acid medin peptide into five 10-residue segments, we conducted individual investigations on each segment’s self-assembly dynamics via microsecond-timescale atomistic discrete molecular dynamics (DMD) simulations. Our findings showed that medin_1–10 and medin_11–20 segments predominantly existed as isolated unstructured monomers, unable to form stable oligomers. Medin_31–40 exhibited moderate aggregation, forming dynamic β-sheet oligomers with frequent association and dissociation. Conversely, medin_21–30 and medin_41–50 segments demonstrated significant self-assembly capability, readily forming stable β-sheet-rich oligomers. Residue pairwise contact frequency analysis highlighted the critical roles of residues 22–26 and 43–49 in driving the self-assembly of medin_21–30 and medin_41–50, acting as the β-sheet core and facilitating β-strand formation in other regions within medin monomers, expecting to extend to oligomers and fibrils. Regions containing residues 22–26 and 43–49, with substantial self-assembly abilities and assistance in β-sheet formation, represent crucial targets for amyloid inhibitor drug design against aortic medial amyloidosis (AMA). In summary, our study not only offers deep insights into the mechanism of medin amyloid formation but also provides crucial theoretical and practical guidance for future treatments of AMA.

Graphical Abstract

Introduction.

The abnormal aggregation of specific amyloid peptides into insoluble fibril deposits within certain human organs or tissues is associated with a broad spectrum of degenerative diseases[1–3], including the formation of amyloid-β (Aβ) plaques and tau tangles seen in Alzheimer’s disease (AD)[3, 4], the accumulation of human islet amyloid polypeptide (hIAPP) in type 2 diabetes mellitus (T2D)[1, 2, 5], and the localized deposition of medin in aortic medial amyloidosis (AMA)[6, 7]. Mounting evidence has demonstrated that the pathological aggregation of amyloid peptides disrupts normal cellular function and contributes to tissue damage and dysfunction[2, 8, 9]. Therefore, molecular insight into the molecular mechanisms driving these disease-related protein aggregations, particularly identifying key fragments that initiate fibril formation and understanding the aggregation tendencies and structural properties of the entire peptide, is essential for developing targeted therapies and interventions aimed at preventing or slowing disease progression[10, 11].

Medin, a 50-amino-acid internal cleavage product of the milk fat globule-EGF factor 8 (MFG-E8) protein, is the principal protein component of aortic medial amyloid, which is the most common type of amyloid observed in the vasculature of the almost all (97%) individuals over 50 years of age[12, 13]. Despite the fact that the role of medin in the development of AMA remains unknown, emerging evidence suggests that the presence of medin amyloid weakens and degenerates the arterial wall, leading to arterial stiffening and cerebrovascular dysfunction[7, 13–15]. The medin-deficient mice, in which the gene encoding the medin-containing C2 domain of MFG-E8 was truncated, did not exhibit vascular aggregates or age-associated decline in cerebrovascular function, indicating the crucial role of medin amyloid in vascular dysfunction with aging[7]. Larsson et al. found that the formation of medin amyloid in the medial layer of the aorta ultimately interfered with aortic function by increasing the tensile strength of the aorta[16]. In addition, high expression of MFG-E8 resulting in increased levels of medin exhibited a positive correlation with aortic stiffness and impairment of renal function[17]. Younger et al. discovered that the aggregation of medin, forming β-barrel pores, a potential toxic intermediate in amyloidosis[4, 5, 18], can induce membrane leakage, proposing it as a potential mechanism underlying vascular dysfunction[19]. Apart from vascular disease, increased levels of medin may also enhance the risk for other amyloid diseases through potential cross-talk (e.g., co-aggregation and cross-seeding) with other amyloid peptides[20, 21]. For example, prior studies have shown that medin fibrils can act as heterologous seeds for the aggregation of serum amyloid A in inflammatory bowel diseases (IBD)[22] and Aβ in AD[23, 24].

The aggregation kinetics of medin in the Thioflavin T (ThT) fluorescence spectrum assay exhibit a typical sigmoidal curve [8, 19, 24], wherein monomers first accumulate into oligomers and then undergo a series of complex conformational conversions to form well-ordered proto-fibrils, followed by the rapid growth of proto-fibrils and fibrils, ultimately leading to the formation of stable mature fibrils. This aggregation pattern is commonly observed in the aggregation of the other amyloid peptides (e.g., Aβ, hIAPP, and α-synuclein)[2, 4]. Experimentally determined atomistic structures of medin monomers, oligomers, and fibrils remain elusive. Insights from circular dichroism (CD) spectroscopy and nuclear magnetic resonance (NMR) measurements suggest that medin monomers predominantly adopt random coil and β-sheet structures[25–28], with evidence of a relatively small amount of helical structure also observed[25, 26]. The presence of negatively charged lipid membranes would induce medin monomers to adopt more helical conformations and accelerate fibril formation[27]. NMR spectroscopy chemical shifts suggest the presence of β-sheet structures involving residues 8–14, 21–23, and 39–50, with the strongest β-sheet tendency observed around the C-terminal 18–19 residues, while the N-terminal residues 1–10 display a weak helical tendency[28]. In prior aggregation studies with truncated medin peptides, it was observed that segments medin_32–41 or medin_42–49 exhibited rapid self-assembly into amyloid fibrils, whereas medin_1–12, medin_14–22, medin_16–24, and medin_1–25 did not demonstrate amyloid-like fibril formation under similar experimental conditions[6]. Solid-state NMR and X-ray assays have revealed that within medin_42–49 fibrils, it adopts an extended and in-register parallel cross-β structure, with the β-strands arranged in a face-to-back orientation[29, 30]. These studies suggest that the last 18–19 C-terminal residues promote the abnormal aggregation of medin[6, 30]. Additionally, a cyclic medin-derived peptide connecting two heptapeptides of medin_19–25 and medin_31–37 can aggregate into a β-sheet structure forming interpenetrating cubes[31], indicating that residues other than those at the C-terminus may also participate in β-sheet formation within medin fibrils. Despite extensive studies on the aggregation kinetics and conformational characteristics of full-length medin and its fragments using various techniques (e.g., ThT fluorescence, NMR, and CD assays), key residues identified as amyloid-prion regions remain inadequately understood. Currently, the atomistic structure of full-length medin and the critical core regions responsible for triggering its aggregation into amyloid fibrils are poorly defined.

To elucidate the aggregation mechanism and identify critical regions driving medin aggregation, the 50-amino-acid full-length medin peptide was evenly divided into five 10-residue segments (Seg-1: medin_1–10, Seg-2: medin_11–20, Seg-3: medin_21–30, Seg-4: medin_31–40, and Seg-5: medin_41–50). Each segment underwent individual investigation for its self-assembly conformational dynamics using ten isolated peptides in 30 independent 1000 ns atomistic discrete molecular dynamics (DMD) simulations. Our simulation results demonstrated that medin_1–10 and medin_11–20 were too weak to form stable aggregates and mostly assumed isolated monomeric forms. The medin_31–40 segment could aggregate into dynamic β-sheet-rich oligomers but featured frequent association and dissociation. Only medin_21–30 and medin_41–50 could aggregate into stable and extended β-sheet-rich oligomers. The self-assemblies of medin_41–50 showed slightly weaker stability compared to medin_21–30 but significantly greater stability than the oligomers formed by medin_31–40. Residue pairwise contact frequency suggested that residues 22–26 and 43–49 not only played a critical role in driving the self-assembly of medin_21–30 and medin_41–50 but also acted as the β-sheet core, assisting other regions (residues from medin_11–20 and medin_31–40) in forming β-strands through a capping mechanism within medin monomers. The self-assembly tendencies of each segment[6, 31], as well as their corresponding secondary structures within the medin monomer[28], were consistent with prior experimental studies. The atomistic conformations for the self-assemblies of each segment and the full-length medin monomer, along with the critical amyloidogenic core regions, were well illustrated by our systematic DMD simulations. In addition, regions of residues 22–26 and 43–49 with significant self-assembly abilities to form β-sheet aggregates, and the capability to assist other regions in forming β-sheets in monomers, as well as potentially in oligomers and fibrils, may serve as the critical targets for amyloid inhibitor drug design against AMA.

Materials and Methods

Molecular Systems.

To unravel the self-assembly mechanism and identify the specific role of each region of medin, the full-length medin peptide was evenly divided into five 10-residue length segments, including Seg-1 (medin_1–10: ⁰¹RLDKQGNFNA¹⁰), Seg-2 (medin_11–20: ¹¹WVAGSYGNDQ²⁰), Seg-3 (medin_21–30: ²¹WLQVDLGSSK³⁰), Seg-4 (medin_31–40: ³¹EVTGIITQGA⁴⁰), and Seg-5 (medin_41–50: ⁴¹RNFGSVQFVA⁵⁰). Given that the experimentally determined structures of medin monomer, oligomer, and fibrils are still elusive, and our prior folding and dimerization simulation of full-length medin indicated that the continuous β-sheet structure was interrupted by loop residues 16–21, 28–31, and 40–43[23, 32], we chose to evenly divide the medin peptide into five 10-residue segments to avoid disrupting the continuous β-sheet regions. This method of segmentation allows us to identify specific regions within the medin peptide that have a higher propensity for self-assembly into amyloid structures. By analyzing each segment individually, we can better understand the contribution of different regions to the overall aggregation process while minimizing the confounding effects of inter-segment interactions. This approach has been effectively utilized in previous amyloid studies to pinpoint the amyloidogenic cores of Aβ and hIAPP, as well as the hot-spot regions driving their co-aggregation in vitro[33].

For each segment, we conducted thirty independent DMD simulations with up to ten simulated peptides per simulation. Each DMD run lasted 1000.0 ns. The initial structure of each segment was set to a fully extended state for their corresponding self-assembly simulations. To avoid potential biases from identical initial states, ten peptides were randomly placed within a 9.0 nm cubic simulation box, ensuring distinct inter-peptide distances and orientations for each independent DMD run. The minimum distance between any pair of peptides in each initial state was maintained at no less than 1.5 nm, exceeding the cutoff for non-bonded interactions. Apart from studying the self-assembly conformational dynamics of each segment, we also investigated the structural dynamics of the monomer to understand the effects of inter-segment interactions on the conformational dynamics within the full-length medin. To achieve this, we conducted thirty independent medin DMD simulations, each lasting 1000.0 ns and starting with different initial velocities. All the details of the simulations are summarized in Table 1.

Table 1.

Details of each molecular system, including self-assembly simulations of each segment and folding simulations of medin monomer, along with the corresponding amino acid sequence (System), the number of simulated peptides in each molecular system (Num. Pep.), the total number of residues in each corresponding system (Num. Res.), the dimension size of each cubic simulation box (Box), the duration of each independent DMD simulation (Time), the total number of independent DMD simulations performed for each molecular system (DMD Runs), and the cumulative total simulation time for each corresponding system (Total Time).

System		Num. Pep.	Num. Res.	Box (nm)	Time (μs)	DMD Run	Total Time (μs)
Segment	Sequence
Seg-1	01RLDKQGNFNA10	10	100	9.0	1.0	30	30.0
Seg-2	11WVAGSYGNDQ20	10	100	9.0	1.0	30	30.0
Seg-3	21WLQVDLGSSK30	10	100	9.0	1.0	30	30.0
Seg-4	31EVTGIITQGA40	10	100	9.0	1.0	30	30.0
Seg-5	41RNFGSVQFVA50	10	100	9.0	1.0	30	30.0
Medin: Seg-1~5		1	50	12.0	1.0	30	30.0

DMD simulations.

The simulations were carried out using an implicit-solvent atomistic DMD approach[34, 35] coupled with the Medusa force field[36, 37], operating within the canonical NVT ensemble at a temperature of 300 K (room temperature). The DMD algorithm is a unique type of molecular dynamics (MD) algorithm where continuous inter-atom potential functions used in classic MD simulations are optimized using a series of discrete step functions[38]. The system’s dynamics are governed by a sequence of collision events where two atoms interact at discrete energy steps, altering their velocities based on conservation laws[39]. DMD achieves enhanced efficiency by focusing on updating only the colliding atom pairs, predicting subsequent collisions with neighboring atoms, and determining the next collision using quick sort algorithms. This approach improves sampling efficiency compared to traditional MD simulations that require frequent force and acceleration calculations (typically every ~1–2 fs). The DMD software is available for academic researchers through Molecules In Action (www.moleculesinaction.com). In our simulations, we used the following units: mass (1 Da), time (50 fs), length (1 Å), and energy (1 kcal/mol). A brief summary and explanation of the DMD simulation process and the subsequent result analysis are also presented in Figure S1.

In the Medusa force field[36, 37], bonded interactions, including bonds, bond angles, and dihedrals, are represented as infinite square wells. Covalent bonds and bond angles typically have a single well, while dihedrals may have multiple wells to account for cis or trans conformations. Nonbonded interactions, such as van der Waals forces, solvation, hydrogen bonding, and electrostatics, are depicted as a series of discrete energetic steps that decrease in magnitude with increasing distance until they reach zero at the cutoff distance. The van der Waals parameters are derived from the CHARMM force field[40], and the EEF1 implicit solvent model is employed for solvation modeling[41]. Hydrogen bond formation is explicitly modeled using a reaction-like algorithm[34]. Electrostatic interactions between charged atoms are calculated using the Debye–Hückel approximation[42] with a Debye length of approximately 10 Å under physiological conditions. The efficacy of the Medusa force field in conjunction with the EEF1 implicit solvation model has been well-demonstrated, including accurate predictions of native states during protein folding[34], generation of conformational ensembles consistent with experimental data from single-molecule FRET measurements of multi-domain protein dynamics[43, 44], and successful reproduction of experimentally observed variations in amyloid tendencies and secondary structures of calcitonin (hCT, sCT, phCT, TL-hCT, and DM-hCT)[45–47] and amylin (hIAPP, hIAPP(S20G), and rIAPP)[5, 48] peptides. This approach’s accuracy was further evaluated by comparing DMD simulations using the Medusa force field for the aggregation of functional amyloid suckerin peptides[49] and pathological hIAPP peptides[48] against standard MD simulations with an explicit solvent model, as detailed in our previous work. Due to its rapid computational speed and improved sampling efficiency, DMD has been widely embraced by our research team[32, 50, 51] and other members of the scientific community[52–55] for investigating protein folding and aggregation.

Analysis methods.

The secondary structure analysis was conducted using the Dictionary of Protein Secondary Structure (DSSP) program[56]. Hydrogen bond formation was identified based on specific criteria: a N···O distance of less than 3.5 Å and a N–H···O angle exceeding 150°[57]. Additionally, pairwise residue contacts were defined as interactions between the heavy atoms of two non-sequential sidechains or main chains with interatomic distances within 0.65 nm. Two peptides were considered part of the same oligomer if they were linked by at least one intermolecular contact. The size of an oligomer was determined by the total number of peptides it contained. Peptides were recognized as forming a β-sheet when at least two consecutive residues from each β-strand conformation were connected by more than two backbone hydrogen bonds[58]. A β-sheet oligomer was described as multiple β-sheets interconnected by at least one pair of heavy atoms[57]. The total number of β-strands in a β-sheet oligomer and a β-sheet layer was defined as the β-sheet oligomer size and the β-sheet size, respectively. The β-strand length referred to the number of consecutive residues adopting a β-sheet structure in each peptide. The two-dimensional free energy surface, also known as the potential mean force[59] (PMF), was calculated using the formula -RTlnP(x, y), where P(x, y) represents the probability of a conformational state having specific parameter values of x and y. To assess the conformation of the medin monomer, cluster analysis was performed using the Daura algorithm[60] with a deviation cutoff of 0.50 nm for backbone atoms.

Results and discussions.

Medin_21–30, medin_31–40, and medin_41–50 readily self-assembled into β-sheet aggregates with distinct conformational dynamics, while medin_1–10 and medin_11–20 predominantly assumed unstructured and isolated states.

The self-assembly conformational dynamics of each of the ten medin segments, starting from isolated monomeric states, were investigated by examining the time evolution of the largest oligomer size, the total number of inter-peptide main-chain hydrogen bonds and heavy atom contacts, along with the average ratio of unstructured and β-sheet conformations (Figure 1). For each molecular system, we randomly selected one trajectory from a pool of thirty independent DMD simulations. The medin_1–10 and medin_11–20 segments predominantly formed small, dynamic oligomers with a relatively low number of inter-peptide hydrogen bonds and contacts, displaying frequent dissociation and reassociation (Figure 1a&b). Analysis of secondary structure and representative snapshots over simulation time revealed that both the medin_1–10 and medin_11–20 segments predominantly adopted unstructured conformations, with β-sheets sparsely populated and rarely observed (Figure 1a&b). The time evolution of the largest oligomer size for each of the ten segments aggregated in each DMD simulation trajectory further confirmed the relatively weak self-assembly capability of medin_1–10 and medin_11–20. The medin_1–10 and medin_11–20 segments only formed transient small-size oligomers (mostly less than four), with the ensemble average oligomer size fluctuating ~1–2 (Figure S2a&b).

Figure 1. Self-assembly dynamics of individual medin segments.

For each molecular system, one single trajectory is randomly selected out of the pool of 30 independent DMD simulations to investigate the self-assembly dynamics of each medin segment, including Seg-1 of medin_1–10 segment a), Seg-2 of medin_11–20 segment b), Seg-3 of medin_21–30 segment c), Seg-4 of medin_31–40 segment d), and Seg-5 of medin_41–50 segment e). The corresponding amino acid sequences of each segment from the full-length medin are displayed at the top. The first column showcases the time evolution of the largest oligomer size formed by each set of ten aggregated fragments. The second column illustrates the progression of intermolecular hydrogen bonds and contacts over time. The third column provides the averaged proportions of unstructured and β-sheet conformations as a function of simulation time. Snapshots captured at 500, 750, and 1000 ns along the simulation trajectory are presented on the right.

Differently, the ten peptides of medin_21–30, medin_31–40, and medin_41–50 readily self-aggregated into oligomeric states exhibiting varying stability in oligomer size (Figure 1c–e). The self-assemblies of medin_21–30 exhibited the greatest stability in size due to forming the largest number of inter-peptide hydrogen bonds and contacts. The stability of medin_41–50 was slightly weaker than that of medin_21–30 but much stronger than that of medin_31–40, as confirmed by the corresponding number of intermolecular hydrogen bonds and contacts. The time evolution of the secondary structure and the corresponding snapshots during the last 500 ns suggested that all three peptides could aggregate into conformations abundant in β-sheets, with the average β-sheet ratio fluctuating around ~30% for medin_21–30 and medin_41–50, and ~20% for medin_31–40 (Figure 1c–e). Additionally, the time evolution of the largest oligomer size from each independent DMD trajectory and the ensemble mass-weighted average oligomer size over thirty DMD simulations for each of the ten segments further suggested that the aggregation tendency of medin_41–50 was slightly weaker than that of medin_21–30 but much stronger than that of medin_31–40, as illustrated by the dynamic size of each type of self-assemblies (Figure S2c–e). For instance, the average oligomer size was ~6.9 for medin_21–30, ~3.2 for medin_31–40, and ~5.8 for medin_41–50 (Figure S2c–e).

The self-assembly of each segment was further investigated by analyzing the aggregation free energy landscape as a function of the oligomer size against the average number of residues assuming β-sheet structures per chain within the corresponding oligomers (Figure 2). This analysis was accompanied by monitoring the self-assembly conformation dynamics through the time evolution of the largest oligomer size, the largest β-sheet oligomer size, the largest β-sheet layer size, as well as the average mass-weighted β-sheet layer size (Figure 2). The aggregation free energy landscapes of medin_1–10 and medin_11–20 revealed a narrow and deep global energy basin, with the oligomer size ranging ~1–3 and an average of 0.0 β-sheet residues per peptide (Figure 2a&b). These results suggested that both fragments predominantly existed as isolated monomers or formed extremely small-size oligomers (e.g., dimers and trimers). Conformations featuring β-sheets exhibited high free energy, indicating that β-sheet formations were unfavorable for both medin_1–10 and medin_11–20. Further analysis of aggregation dynamics illustrated that both segments mainly adopted transient, dynamic, small-size oligomers and lacked β-sheet structures (Figure 2a&b). Our observation that medin_1–10 and medin_11–20 exhibited an extremely weak aggregation tendency aligns with prior experimental studies, which found that truncated segments like medin_1–12, medin_14–22, and medin_16–24 did not form amyloid-like fibrils[6]. In addition, medin1–10 exhibited a very weak β-sheet tendency, which agreed with experimental measurements and classical MD simulations that found the N-terminus of medin was highly dynamic and lacked β-sheet formations[28]. The seeding efficiency of medin_11–50 fibrils was stronger than medin_1–50 fibrils but weaker than medin_31–50 fibrils for the aggregation of medin_1–50, indicating that medin_1–10 and certain residues from medin_11–30 hindered the aggregation of full-length medin[6].

Figure 2. Self-assembly free-energy landscape and aggregation conformational dynamics analysis for each medin segment.

The aggregation free energy landscapes for each of the medin segments a–e) are depicted on the left, illustrating their relationship with oligomer size (n_{oligomer size}) against the average number of residues per chain adopting the β-sheet conformation (n_β-residue). Data from all 30 independent 1000 ns simulations are utilized for this analysis to capture the entire self-assembly conformations during the aggregation process. On the right, one randomly selected trajectory from the 30 independent DMD simulations illustrates the self-assembly conformational dynamics of the corresponding medin segment. Self-assembly dynamics are monitored through the time evolution of the largest oligomer size, the largest β-sheet oligomer size, the largest β-sheet size, and the average β-sheet oligomer size. Snapshots of the assemblies are provided in the inset on the right, with their corresponding states labeled in the free energy landscape.

A broad and deep free energy basin was observed in the aggregation free energy landscapes of medin_21–31 and medin_41–50, with the oligomer size ranging ~7–10 and an average of ~3–5 β-sheet residues per peptide (Figure 2c&e). Consistent with the analysis of self-assembly conformational dynamics, both the medin_21–31 and medin_41–50 peptides spontaneously assembled into β-sheet-rich oligomers, with the oligomer size mostly fluctuating ~7–10. The self-assemblies of medin_41–50 exhibited slightly greater fluctuation than those of medin_21–30, indicating that the aggregates of medin_21–30 were slightly more stable than those of medin_41–50, as illustrated by the deeper and smaller free energy basin observed for medin_21–30 compared to medin_41–50. Despite the aggregation of the fragment containing residues 21–30 not being studied experimentally, a cyclic peptide comprising medin_19–25 and medin_31–37 spontaneously aggregated into β-sheet structures[31], suggesting that residues from medin_21–30 might be amyloid-prone. Our observation that medin_41–50 exhibited strong aggregation capability aligned with previous assays demonstrating that medin_42–49 could independently form amyloid fibrils[61]. Additionally, residues 21–26 and 42–49 showed stronger β-sheet chemical shift conformations than other regions of the full-length medin monomer, further supporting our findings[28].

Meanwhile, the medin_31–40 segments could also self-assemble into β-sheet oligomers but exhibited frequent dissociation and reassociation, with the sizes of the largest oligomer and β-sheet oligomer fluctuating ~3–10 (Figure 2c&e). The aggregation free energy landscape of medin_31–40 displayed a flat and broad energy basin with oligomer sizes ranging from ~3–8 and ~2–5 β-sheet residues per peptide, indicating diverse β-sheet aggregates. Interestingly, a similar fragment, medin_32–41, was also found to form amyloid fibrils in vitro[6]. These observations suggested that the aggregation tendency and β-sheet formation of medin_31–40 were moderate compared to those of medin_21–30 and medin_41–50. Overall, our simulation results suggested that the aggregation tendencies of medin_1–10 and medin_11–20 were too weak to form stable oligomers, while medin_21–30, medin_31–40, and medin_41–50 exhibited a significant capability to aggregate into β-sheet-rich oligomers. Among these β-sheet-rich aggregates, the self-assemblies of medin_41–50 showed slightly weaker stability compared to medin_31–40, but significantly greater stability than the oligomers formed by medin_31–40. The β-sheet formations of medin_31–40 were more dynamic than those of medin_21–30 and medin_41–50, supported by NMR measurements showing that the chemical shifts of residues 21–30 and 41–50 indicated a stronger β-sheet tendency than those of residues 31–40 in full-length medin[28].

The β-sheet residues were primarily located at the N-terminus of medin_21–30, the center of medin_31–40, and the C-terminus of medin_41–50 within their respective self-assemblies.

The equilibrium and convergence of each molecular system during the self-assembly of the ten segments were systematically examined by analyzing the time evolution of various structural parameters (Figures S3&S4), including radius of gyration (Rg), the number of inter-peptide main-chain hydrogen bonds and heavy atom contacts, and the content of unstructured and β-sheet conformations. One trajectory was randomly selected from each simulated system, chosen from a pool of thirty independent simulations. It was observed that all systems reached steady states within the final 400 ns for the monitored parameters (Figures S3). Additionally, the ensemble-averaged time evolution of these parameters across all thirty independent trajectories showed minimal changes during the last 400 ns, indicating that all molecular systems were adequately equilibrated and had reached a steady state (Figures S4). Consequently, simulation data from the last 400 ns of each trajectory were used for the comprehensive conformational and structural analysis.

The mass-weighted oligomer size probability distribution revealed distinct self-assembly behaviors among different segments of medin. Specifically, medin_1–10 and medin_11–20 segments predominantly remained as isolated monomers, with relatively low populations of dimers and trimers (Figure 3a). Larger aggregates composed of more than 4 peptides were not observed in these self-assemblies. In contrast, medin_21–30 and medin_41–50 segments predominantly aggregated into β-sheet oligomers with sizes ranging 7–10, while medin_31–40 assembled into smaller β-sheet oligomers composed of 2–7 β-strands (Figure 3a&b). The β-sheet layer size distribution indicated that ~3–6 β-strands forming β-sheet layers were the most populated species within medin_21–30 and medin_41–50 self-assemblies, comprising nearly half of their β-sheet oligomer size distribution (Figure 3b&c). These results suggest that the β-sheet oligomers formed by medin_21–30 and medin_41–50 may contain more than one β-sheet layer. On the other hand, the distribution of β-sheet oligomer sizes within medin_31–40 aggregates was similar, indicating that medin_31–40 predominantly aggregated into single-layer β-sheet oligomers (Figure 3a–c).

Figure 3. Equilibrium conformational and structural analyses for the self-Assemblies formed by each medin segment.

Probability distributions of the oligomer size a), β-sheet oligomer size b), and β-sheet layer size c) for the self-assemblies formed by each of the ten individual medin segments are presented. To minimize initial state bias, only conformations from the last 400 ns of all independent simulations are included for the conformational analysis. Average secondary structure contents, encompassing unstructured, β-sheet, helix, and turn formations, for the aggregates of each medin segment are shown e). Propensities of each residue to adopt coil and bend, β-sheet, helix, and turn conformations for the self-assemblies of medin segment-3 f), segment-4 g), and segment-5 h) are presented. A final snapshot for each segment, randomly selected from 30 independent DMD trajectories, is also displayed i).

The average secondary structure analysis revealed that all medin segment self-assemblies were predominantly unstructured, with notably high populations observed for medin_1–10 (~83.7%) and medin_11–20 (~92.4%) aggregates (Figure 3d). In contrast, medin_1–10 exhibited ~5.3% helical structured conformations, while the self-assemblies of medin_21–30, medin_31–40, and medin_41–50 exhibited β-sheet formation with probabilities of approximately ~35.5%, ~22.2%, and ~33.1%, respectively (Figure 3d). Specifically, the β-sheet formations of medin_21–30 were predominantly formed by their N-terminal residues (L22, Q23, V24, D25, and L26) with ratios ranging ~47.9%-78.4% (Figure 3e). In contrast, the β-sheet structures of medin_31–40 were primarily formed by their central residues (G34, I35, I36, T37, and Q38) with probabilities ranging ~25.9%-46.1% (Figure 3f). The β-sheet formations of medin_41–50 were mainly attributed to their C-terminal residues (S45, V46, Q47, F48, and V49) with probabilities ranging ~45.2%-63.8% (Figure 3g). The β-sheet populated regions of medin_21–30 and medin_41–50 contained three hydrophobic residues, whereas medin_31–40 only contained two hydrophobic residues (Figure 3e–g). This difference in hydrophobic content may contribute to the observed greater stability in β-sheet formations for medin_21–30 and medin_41–50 compared to medin_31–40.

Taken together, the conformational and structural analysis suggested that medin_1–10 and medin_11–20 mainly preferred to adopt an isolated monomer structure, while medin_21–30 and medin_41–50 exhibited a strong aggregation tendency, forming multiple β-sheet layer composed oligomers (Figure 3h). On the other hand, medin_31–40 featured a moderate aggregation tendency, primarily forming single-layer dynamic β-sheet aggregates (Figure 3h). These computational investigations were in line with prior experimental measurements of medin segment amyloid aggregation [6, 29, 30]. For example, no amyloid aggregates were observed in the aggregation of medin_1–12, medin_14–22, medin_16–24, and medin_1–25, while segments medin_32–41 or medin_42–49 exhibited rapid self-assembly into amyloid fibrils[6]. β-sheet dominant aggregates were also observed in the two heptapeptides of medin_19–25 and medin_31–37[31], but not in medin_16–24 and medin_1–25 aggregation[6], indicating that residues from medin_1–20 suppressed the aggregation of residues 21–25. Similar inhibitory effects were also observed in the aggregation of α-synuclein when two amyloid-resident repeats were included[50, 62].

The interactions among hydrophobic residues are driving the self-assembly of medin_21–30, medin_31–40, and medin_41–50, leading to the formation of extended β-sheet structures.

The analysis of β-sheet lengths revealed that the β-strands of medin_21–30 peaked at lengths of 3–6 residues, while those of medin_31–40 and medin_41–50 peaked at lengths of 4–7 residues (Figure 4a). Additionally, the distribution of the distance between N-terminal and C-terminal Cα atoms (End2End distance) for each peptide indicated that the conformations of medin_21–30, medin_31–40, and medin_41–50 predominantly assumed extended conformations compared to medin_1–10 and medin_11–20 (Figure 4b). The extended cross-β structure formed by medin_42–49 was also supported by NMR and X-ray assays[29, 30]. Residue-pairwise contact analysis for the self-aggregates revealed that interactions between residues 22–26 drove medin_21–30 to assume either parallel or anti-parallel β-sheet structures, stabilized by hydrophobic interactions among residues L22, V24, and L26 (Figure 4c). Additionally, electrostatic repulsion between D28 residues favored the formation of anti-parallel β-sheets over parallel β-sheets in medin_21–30 self-assemblies. In contrast, medin_31–40 self-assemblies predominantly exhibited an anti-parallel β-sheet pattern, mainly stabilized by hydrophobic residue interactions between L35 and L36 (Figure 4d). Moreover, interactions among hydrophobic residues F43, V46, F48, and V49 drove regions of residues 43–49 to form both parallel and anti-parallel β-sheet formations (Figure 4e). Notably, the number of hydrophobic residue-pairwise contacts associated with parallel β-sheets was more abundant than those associated with anti-parallel β-sheets in medin_41–50 self-assemblies, with parallel β-sheets being slightly more populated than anti-parallel ones. Our residue-pairwise contact analysis for the β-sheet-rich aggregates formed by each medin segment revealed that residues 22–26 and 43–49 played a critical role in driving the abnormal aggregation of medin, while residues L35 and L36 were also involved but contributed to a lesser extent.

Figure 4. The β-sheet structure analysis for the self-assemblies formed by each medin segment.

Probability distributions of the β-strand length for each peptide within the self-assemblies formed by each of the ten medin segments are presented a). Probability distributions of end-to-end (End2End) distances for each medin segment within their aggregates are shown b). Residue-pairwise inter-peptide contact frequency maps between main-chain atoms for the β-sheet-rich self-assemblies of medin segment-3 c), segment-4 d), and segment-5 e). Representative parallel and anti-parallel β-sheet contact motifs are displayed on the right. Only the data from the last 400 ns of each 1000 ns independent simulation are used for the analysis.

Residues 22–26 and 43–49 acted as the β-sheet core, assisting medin monomers in forming a three to four β-strand structure.

To explore the inter-segment interactions governing the conformational dynamics within medin monomers, we conducted comprehensive folding dynamics simulations of the full-length medin peptide using thirty independent 1000.0 ns DMD trajectories. During the full-length medin folding simulation, snapshots and secondary structure analysis revealed that the medin monomer primarily formed a β-sheet structure with three or four β-strands between residues 11–40, while the medin_1–10 segment predominantly exhibited unstructured or helical conformations (Figure 5a&b). The average secondary structure propensity for each residue indicated that residues 20–26, 33–39, and 44–49 exhibited a pronounced β-sheet tendency in the medin monomer, with a propensity mostly greater than 50% (Figure 5c). Additionally, residues 8–16 showed a 25%-50% β-sheet tendency. Dynamic helices were mainly formed by residues 2–15. The observation of dynamic weak helical tendencies around residues 1–10, coupled with predominant β-sheet formation in the 20–49 regions, aligns with prior NMR measurements[28]. The average β-sheet tendency within the medin monomer was greater than in their self-assemblies (Figures 3e–g&5c), suggesting that intra-peptide interactions among medin segments could assist medin folding into β-sheet formations. Furthermore, the root mean square fluctuation (RMSF) analysis of the medin monomer revealed that high amyloidogenic regions observed in the segment self-assembly simulations (e.g., medin_22–26, medin_34–37, and medin_43–49) were much more stable than other regions (Figure 5d). Additionally, the central structures of the top 10 most populated medin monomers showed that segments with relatively weak β-sheet propensity in their self-assembly assumed β-sheet conformations when capping onto the high amyloidogenic regions (Figure 5e).

Figure 5. Conformational dynamics analysis for the medin monomer.

The time evolution of the secondary structure of each residue within the medin monomer, with corresponding snapshots presented every 100 ns a-b). Two trajectories are randomly selected from a pool of thirty independent simulation runs. The average secondary structure propensity for each residue within the medin monomer c). The average root-mean-square-fluctuation (RMSF) for each residue within the medin monomer d). The central structures and corresponding populations of the top 8 most populated medin monomer conformations e). Only the last 400 ns of saturated simulation data are used for the above conformational analysis in c-e). For clarity, the N-terminal Cα atom is highlighted by a bead, and the structure of each segment is colored according to the color bar stamped at the bottom.

The residue-pairwise contact frequency analysis within the medin monomer revealed distinct structural patterns, including one helical and five β-sheet contact patterns (Figure 6a). The helical contact pattern was prominent among residues 2–19, with a highly populated helical region concentrated within residues 2–10 (labeled as contact pattern and structure 1 in Figure 6a). Interactions between neighboring segments drove the formation of three evident β-hairpin structures, characterized by anti-parallel β-sheet formations: residues 9–16 vs 19–26, 23–28 vs 32–37, and 32–38 vs 44–50 (labeled as contact patterns and structures 2–4 in Figure 6a). Notably, the β-hairpin between medin31–40 and 41–50 exhibited the highest contact frequency, highlighting its importance as a dominant structural motif within the medin monomer. Furthermore, apart from interacting with medin_31–40, the segment of medin_41–50 also engaged with medin_21–30 and medin_19–20 to form additional β-sheet formations, specifically involving residues 8–14 vs 43–49 and 19–25 vs 43–49 (labeled as contact patterns and structures 5–6 in Figure 6a).

Figure 6. Conformational analysis of each residue within the medin monomer.

Residue-residue contact frequency among heavy atoms from the main-chain (lower diagonal) and side-chain (upper diagonal) within the medin monomer a). Representative structured contact patterns labeled as 1–6 on the residue-pairwise contact frequency map are presented on the right. The average accessible surface area (SASA) of each residue within the medin monomer b). The distance distribution of each Cα atom of each residue corresponding to the geometric center of each medin monomer c). All the above conformational analyses are calculated using the last 400 ns of simulation data from 30 independent 1000.0 ns DMD trajectories.

The solvent accessible surface area (SASA) of each residue within the medin monomer was further analyzed (Figure 6b). Residues from regions with significant aggregation tendencies in the segment self-assembly simulations (e.g., medin_22–26, medin_34–37, and medin_43–49) exhibited relatively smaller SASA values compared to other regions, likely due to binding interactions with neighboring residues (Figure 6a). Moreover, these high amyloidogenic regions were predominantly buried inside the medin monomer, as illustrated by the radius distribution of the Cα atom of each residue relative to the geometric center (Figure 6c), with regions such as medin_22–26 and medin_43–49 being closest to the center. Overall, our systematic analysis underscored the critical roles of residues from medin_20–31 and medin_41–50 in maintaining the β-sheet structure of the medin monomer.

Future amyloid inhibitor drug design against AMA disease should focus on the computationally determined amyloidogenic core regions around medin_21–30 and medin_41–50.

Prior computational simulations and experimental studies on fibrils of Aβ, hIAPP, and α-synuclein have suggested that regions exhibiting dynamic β-sheet structures in monomers or oligomers [4, 5, 63, 64] also act as the β-sheet core in their corresponding fibrils[65–67]. This common behavior of β-sheet structures in amyloid peptides suggests that the β-sheet regions observed in medin monomers (Figure 5c) are expected to be present in medin fibrils, although the specific fibril structures of medin have yet to be fully established. Despite residues 8–49 displayed strong β-sheet tendencies in medin monomer, and also expected as in the fibrils, only medin_11–20 and medin_41–50 could independently self-assembly into well-ordered stable β-sheet formations, indicating this region should sever as the amyloidogenic core. Since the other segments either mostly maintained in isolated monomer (e.g., medin_1–10 and medin_11–20) or dynamic β-sheet oligomers (e.g., medin_31–40) (Figures 1–3). A typical example is α-synuclein fibrils, where the β-sheet structure is typically formed by residues 38–95[67], but only segments encompassing residues 31–41 and 68–78 (known as NACore, non-amyloid component core) can independently self-assemble into well-ordered β-sheet aggregates[50]. Due to the absence of the NACore region in β-synuclein, it is unable to aggregate into amyloid fibrils, unlike the amyloidogenic peptide α-synuclein[68]. Human amylin and calcitonin peptides readily form amyloid deposits, whereas mutations occurring around the amyloidogenic core region render rat amylin and salmon calcitonin as amyloid-resistant peptides[45, 46, 48]. This evidence[45, 46, 48, 68] suggests that targeting the amyloidogenic core region, rather than all fibril β-sheet regions, could efficiently prevent the abnormal aggregation of amyloid peptides. For instance, prior studies have shown that capping the amyloidogenic core regions of Aβ with αB-Crystallin and SEVI (semen-derived enhancer of viral infection) peptides can effectively mitigate the abnormal aggregation of Aβ[39, 69–71]. Therefore, in future amyloid inhibitor drug designs targeting AMA disease, emphasis should be placed on residues 22–26 and 43–49. These regions not only promoted the independent aggregation of medin_21–30 and medin_41–50 into stable β-sheet aggregates but also facilitated β-sheet formation in other regions within the medin monomer, which was expected to trigger the oligomerization and fibrillization of medin.

For instance, our prior dimerization simulations of medin have demonstrated that regions observed as intra-chain β-sheet could also present as inter-chain β-sheet, but the corresponding participating partners were from different chains[23, 32]. This domain-swap phenomenon suggested that the same interactions stabilizing intra-peptide structures also contribute to the formation of inter-peptide structures in protein-protein interactions, ultimately leading to the formation of high-order complexes.

Conclusion

Medin amyloid formation is closely linked to endothelial dysfunction and vascular inflammation, contributing to the pathogenesis of various vascular diseases. Our study aimed to elucidate the molecular mechanisms driving medin aggregation, focusing on identifying key fragments that initiate fibril formation and on understanding the structural properties of the entire peptide. Our multiple microsecond-timescale simulations revealed that the medin_1–10 and medin_11–20 segments exhibited a weak aggregation tendency, predominantly assuming unstructured isolated monomers, while the medin_31–40 segment formed dynamic β-sheet-rich oligomers characterized by frequent association and dissociation. Both the medin_21–30 and medin_41–50 segments exhibited stable aggregation into β-sheet-rich oligomers, with the self-assemblies of medin_41–50 showing slightly weaker stability compared to medin_21–30, but significantly greater stability than those of medin_31–40 alone. Residue pairwise contact frequency analysis indicated residues 22–26 and 43–49 played critical roles in driving the self-assembly of medin_21–30 and medin_41–50, acting as the β-sheet core and facilitating the formation of β-strands in other regions (such as residues from medin_11–20 and medin_31–40) through a capping mechanism within medin monomers, as they are expected to do in oligomers and fibrils. Given that the core regions of amyloid peptides, including medin, can recruit isolated peptides and extend β-sheet edges with limited sensitivity to the amino acid sequence[23, 72, 73], targeting these theoretically predicted amyloidogenic core regions by capping them with amyloid-resistant peptide drugs may offer a promising therapeutic approach for AMA. Prior experimental studies have shown that negatively designed β-sheet peptide drugs targeting amyloidogenic core regions effectively inhibit edge-to-edge aggregation of Aβ, hIAPP, and α-synuclein[10, 11, 69, 70, 74]. Overall, our study not only advances the understanding of medin amyloid formation but also identifies potential target regions for the rational design of future amyloid inhibitors against AMA.

Data availability

Data will be made available on request.