Differential binding and conformational dynamics of tau microtubule-binding repeats with a preformed amyloid-β fibril seed

Abstract

Both senile plaques formed by amyloid-beta (Aβ) and neurofibrillary tangles (NFTs) comprised of tau are pathological hallmarks of Alzheimer’s disease (AD). The accumulation of NFTs better correlates with the loss of cognitive function than senile plaques, but NFTs are rarely observed without the presence of senile plaques. Hence, cross-seeding of tau by preformed Aβ amyloid fibril seeds has been proposed to drive the aggregation of tau and exacerbate AD progression, but the molecular mechanism remains unknown. Here, we first identified cross-interaction hotspots between Aβ and tau using atomistic discrete molecular dynamics simulations (DMD) and confirmed the critical role of the four microtubule-binding repeats of tau (R1-R4) in the cross-interaction with Aβ. We further investigated the binding structure and dynamics of each tau repeat with a preformed Aβ fibril seed. Specifically, R1 and R3 preferred to bind the Aβ fibril lateral surface instead of the elongation end. In contrast, R2 and R4 had higher binding propensities to the fibril elongation end than the lateral surface, enhancing β-sheet content by forming hydrogen bonds with the exposed hydrogen bond donors and acceptors. Together, our results suggest that the four repeats play distinct roles in driving the binding of tau to different surfaces of an Aβ fibril seed. Binding of tau to the lateral surface of Aβ fibril can increase the local concentration, while the binding to the elongation surface promotes β-sheet formation, both of which reduce the free energy barrier for tau aggregation nucleation and subsequent fibrillization.

Graphical Abstract

Introduction

Extracellular senile plaques and intracellular neurofibrillary tangles (NFTs), two neuropathological hallmarks of Alzheimer’s disease (AD), are amyloid fibrils primarily comprised of Aβ and tau, respectively.^1–4 Aβ and tau are two intrinsically disordered proteins (IDPs) found in the human central nervous system.^5–7 Aβ peptides are produced by proteolytic cleavage of the trans-membrane amyloid precursor protein (APP) into different isoforms.^8–10 The Aβ40 isoform is the most abundant, while Aβ42 is more amyloidogenic, toxic and corresponds to the core component of extracellular senile plaques in AD.^11,12 Meanwhile, tau441 is a multi-domain IDP, consisting of a charged N-terminal region followed by a proline-rich region (PRR), a microtubule-binding domain (MTBD) containing four repeat regions with similar sequences, and a C-terminal domain.^13,14 Tau also has multiple isoforms due to alternative splicing, and the isoforms with the length ranging from 352 to 441 amino acids can be categorized into two classes, one containing all four repeats (4R) and the other lacking the second repeat (3R).¹⁵ The 4R isoforms are found to cause greater neurodegeneration, while 3R isoforms may cause more axonal transport defects and locomotor impairments.¹⁶ Increasing evidence suggests that amyloid proteins associated with different diseases may interact with each other, leading to cross-talk between corresponding diseases.^17–21 In the case of AD, the accumulation of NFTs has shown a stronger association with the loss of cognitive function than senile plaques while the appearance of plaques is not necessarily responsible for developing dementia.^22,23 However, NFTs have rarely been observed in the cortex without senile plaques present.^24–26 Moreover, Aβ42 forms amyloid fibrils in vitro more readily than tau.²⁷ Treating cells with Aβ oligomers promotes intracellular tau seeding.²⁸ Hence, Aβ aggregation is thought to occur first at the early stage of AD, and tau aggregation, despite its better correlation with AD symptoms, is likely triggered and facilitated by Aβ aggregation and further dominates the disease progression.^29–33 However, the molecular mechanism by which Aβ and the corresponding amyloid aggregation influence tau seeding, specifically how aggregated species of Aβ contributes to this process, is largely unknown.

Emerging evidence suggests that Aβ can also accumulate intracellularly due to either proteolytic processing of APP in intracellular membrane or reuptake of extracellular Aβ, and the intraneuronal Aβ aggregation might contribute to disease progression.³⁴ Given the possibility of intraneuronal colocalization between Aβ and tau, many studies have investigated their direct interactions and possible co-aggregation or cross-seeding at the molecular level.^33,35 For example, Aβ and tau were found to be able to bind each other and form hetero-complexes, and the binding affinity was almost 1,000-fold higher than that of tau self-association.³⁶ The cross-interaction hotspots in Aβ and tau sequences were also identified by the peptide membrane array approach. Another recent experimental study revealed that the Aβ central hydrophobic core (¹⁶KLVFFA²¹) directly interacted with specific regions of tau and facilitated cross-seeding,³⁰ and the structure-based inhibitors of Aβ core also interfered with tau seeding. In addition to in vitro and in vivo evidence, preformed Aβ fibril seeds were able to accelerate tau aggregation in a cell-free assay, and immuno-electron microscopy imaging suggested that Aβ fibril seeds provided the nucleus for subsequent tau fibril formation, supporting the heterotypic nucleation of tau aggregation by preformed Aβ seeds.³⁷ An Aβ fibril seed features two distinct interfaces, the fibril elongation end and the lateral surface, which have different physicochemical properties. The fibril elongation surfaces at both ends feature exposed hydrophobic cores and also unsaturated backbone hydrogen bond donors and acceptors. In contrast, the lateral surface of fibrils resembles the surface of folded proteins, containing hydrophobic patches scattered among a majority of hydrophilic residues. Moreover, the four MTBD repeats, especially two nucleating six-residue segments ²⁷⁵VQIINK²⁸⁰ and ³⁰⁶VQIVYK³¹¹, called paired helical filaments hexapeptides PHF6 and PHF6* respectively in the second and third repeats,^38,38–40 have been identified as the amyloidogenic regions of tau aggregation. Hence, in order to understand the effects of Aβ fibril seeds on tau cross-seeding and aggregation, it is necessary to uncover details of the binding of tau protein, especially the amyloidogenic repeats in MTBD, with different surfaces of the Aβ fibril seed and characterize the corresponding structures and dynamics.

Here, we applied atomistic discrete molecular dynamics (DMD) simulations to study the binding of tau with a preformed Aβ fibril seed and also the binding-induced conformational changes. DMD is a rapid and predictive molecular dynamics algorithm that has been widely used by our group and others to study protein-protein interaction, protein folding and amyloid aggregation.^41–43 However, due to a large size of tau441, it is computationally expensive to directly study full-length tau interacting with Aβ or Aβ fibrils in DMD simulations. Therefore, we adopted a divide-and-conquer approach to first identify the cross-interaction hotspots of tau and Aβ42 by mimicking the peptide membrane array approach in experimental identification of cross-interaction hotspots.^36,44 Our simulation results demonstrated that four MTBD repeats indeed had high propensities to interact with Aβ42, and formed more intermolecular β-sheets than other regions, in agreement with prior experiments.³⁶ We next investigated the interaction of each MTBD repeat with a preformed Aβ42 fibril seed in DMD simulations without a priori assumption of the complex structures. All tau repeats showed an enhancement of β-sheet formation to different degrees in the presence of the Aβ fibril compared to isolated monomeric repeats. R1 and R3 preferred to bind the lateral surface rather than the elongation surface of the Aβ fibril, preserving their native secondary structure compositions. In contrast, R2 and R4 had a higher binding propensity to the elongation surface than the lateral surface of the fibril, forming extensive β-sheets with the fibril by forming hydrogen bonds with the exposed unsaturated hydrogen bond donors and acceptors. Altogether, our results suggest that four MTBD repeats play distinct roles in driving the binding of tau to two different surfaces of an Aβ fibril seed. Binding of tau in the lateral surface usually does not change tau secondary structures but can increase the local concentration of tau proteins, which promotes the nucleation of tau fibril seeds. On the other hand, binding of tau to the Aβ fibril elongation surface promotes β-sheet formation, significantly reducing the free energy barrier for tau seeding and subsequent fibrillization. Our computational study offers a molecular insight to the experimentally observed cross-seeding of tau aggregation by preformed Aβ fibril seeds, and helps better understand the pathological synergy of Aβ and tau aggregation in the etiology of AD.

Results and discussion

Tau MTBD repeats were hotspots for the cross-interaction with Aβ42

We adopted a computational peptide array method as used in experiments^45–47, which was able to recapitulate the hotspots for cross-interaction between Aβ and hIAPP⁴⁴ (details see Methods). Here, we split the tau into overlapping 14-residue fragments and modeled their interactions with the full-length Aβ42 peptide using DMD simulations. By averaging over simulations with overlapping sequences, we estimated the binding probability of each tau residue with Aβ42 (Fig. 1a). There were three regions in PRR, R2, and R3 that had high contact frequencies (> 0.9) with Aβ42. In particular, the highest Aβ-binding regions in tau were R2 and R3, corresponding to PHF6 and PHF6* respectively, which are known to be important for tau aggregation. Within the MTBD of tau, R1 and R4 were also relatively prone to interact with the Aβ peptide but with weaker affinities than R2 and R3. The N-terminus of tau displayed a lower Aβ-binding proclivity than the C-terminus of tau. Likewise, we observed a similar trend for the intermolecular backbone hydrogen bond formation between Aβ and tau fragments (Fig. S3). Overall, the tau MTBD displayed a high propensity of forming β sheets in the presence of Aβ, especially for tau PHF6 and PHF6* located in R2 and R3 (Fig. 1b). In addition, tau220–227, within the tau PRR, also displayed a significantly high propensity for β-sheet formation. Residues at the C-terminus of tau exhibited a high propensity for helical structures, consistent with its weak binding to the Aβ peptide. Next, we combined all the binding simulations of tau fragments with Aβ and computed the contact frequency map between Aβ and tau residues (Fig. S4). For the Aβ-binding hotspot regions, including the four repeats in MTBD and PRR, interaction patterns consistent with either parallel or anti-parallel intermolecular β-sheets with Aβ16–22 or the C-terminus – two known amyloidogenic regions of Aβ^44,48 – were observed.

Figure 1.

Secondary structure and inter-molecular contacts derived from the computational peptide array simulations. a) Contact frequency of each tau residue with Aβ42. Error bars represent standard error of mean from the ensemble average over all trajectories. b) Secondary structure propensity of tau residues in Aβ-tau interactions. Vertical lines (black) labels R1-R4 in both a) and b).

Our computational peptide array results agreed with the experimental measurements by Guo et al.³⁶ For instance, the membrane peptide array experiments demonstrated strong binding of Aβ with tau216–226, located in PRR of tau. Likewise, tau249–259, tau267–281, tau309–321 and tau337–368, which belong to the tau MTBD, also displayed strong binding with Aβ42. Furthermore, tau369–383 and tau393–407, located in R’ region of tau⁴⁹, also exhibited strong binding with Aβ42 in both tau peptide array membrane and our computational results. Together, our computational results confirmed the four MTB repeats as the binding hotspots of tau with Aβ. Given their important role in the amyloid aggregation of tau,^27,34 we next investigated the interaction of four MTB repeats with a preformed Aβ fibril seed in DMD simulations.

Secondary structure changes of tau MTB repeats upon binding to the Aβ fibril

For each of the four tau repeats, we performed DMD simulations in the presence of a preformed Aβ fibril seed. In each case, 30 independent simulations were run, starting with the tau peptide in the coil conformation and randomly positioned away from the fibril. The Aβ fibril was kept static during simulations and each independent run lasted 1 μs. Control simulations of each tau repeat monomer were also conducted for comparison. The dynamics of the simulations were monitored by the time evolution of several parameters, including the potential energy, the radius of gyration (Rg), and the number of atomic contacts with the Aβ fibril averaged over independent simulations (Figs. S5–S7). All four tau repeats showed significant potential energy gain in the presence of Aβ fibril seed. The increasing atomic contacts between repeat peptides and the fibril over time also confirmed the binding between the peptides and the fibril seed. More importantly, larger Rg values of tau repeats in the presence of the fibril than that of the isolated peptides suggested that the repeats underwent conformational alterations and became more extended while binding to the Aβ fibril. The time-evolution of computed parameters also suggested that our simulations had reached equilibrium after 500 ns (Figs. S5–S7). Using equilibrated trajectories, the averge binding potential energy of each tau repeat with the fibril was obtained (Table S1, Fig. S6), suggesting that R4 had the strongest binding, followed by R2, R1, and R3 accordingly.

We first calculated the secondary structure propensities of each residue in a tau MTBD repeat with and without the presence of Aβ fibril seed (ordered second structures of helix and sheets in Fig. 2, coil and turn in Fig. S8). In all cases, coil was the dominant conformation for the tau repeats, consistent with their IDP nature. Despite their sequence similarity, isolated R1, R2 and R4 monomers contained residual helixes with little β-sheet content, while R3 formed partial β-sheets with little helixes. Upon binding Aβ fibril, all tau repeats showed increased the β-sheet contents, although the increases of R2 and R4 were much higher than those of R1 and R3. For R2 and R4, variations of helices and turn structures were smaller compared to the enhancement of β sheets, which mainly resulted from the disorder-to-order conversion of coils (Fig. S8). In the case of R1, the slight increase of β-sheet content coincided with the loss of helices upon binding with the Aβ fibril. R3, on the other hand, still had little helices upon binding the Aβ fibril and the increase of β-sheets resulted from the decrease of coils (Fig. S8). We next investigated what caused the differential secondary structure changes of tau repeats upon binding Aβ fibril.

Figure 2.

The probability of forming β-sheet and helix by each reside in tau MTBD repeats, R1 to R4, in the presence (solid line) and absence (dash line) of a preformed Aβ fibril seed. The error bar corresponds the standard error of mean, averaged over independent simulations.

Tau repeats bound to different surfaces of the Aβ fibril

The elongation and lateral surfaces of a fibril seed have distinct physicochemical properties. Peptides at the fibril elongation surface have unsaturated backbone hydrogen bond donors or acceptors exposed, enabling the formation of interchain hydrogen bonds. In contrast, the backbone hydrogen bond donors and acceptors of residues at the lateral surface of an amyloid fibril are saturated by adjacent peptides in the fibril structure. Therefore, we computed the distribution of the number of backbone hydrogen bonds formed between exposed Aβ peptides at the fibril elongation surface and tau repeats (Fig. 3a), and the contact frequency of each tau repeat residue with the fibril elongation (Fig. 3b) and lateral (Fig. 3c) surfaces.

Figure 3.

Interaction of tau repeat residues at two surfaces of the Aβ fibril a) Distribution of the number of interpeptide backbone hydrogen bonds between a tau repeat and the fibril (fibril-tau hydrogen bonds). The upper part of the panel along vertical axis from 0.2 to 1 was shown in the subplot. b-c) Contact frequency between each residue of a tau repeat (R1-R4) and Aβ peptides at the b) elongation surface c) lateral surface of the Aβ fibril. Error bars refer to the standard error of mean, averaged over independent simulations.

R1 displayed the lowest propensity to form interpeptide backbone hydrogen bonds with the Aβ fibril seed, followed by R3, R2, and R4 in the order of increasing propensity for hydrogen bonding to the fibril (Fig. 3a). For instance, R1 and R3 did not form any hydrogen bonds with the Aβ fibril for more than half of the time, while R2 and R4 hydrogen bonded with the fibril for the majority of the time in our equilibrium simulations (e.g., inset of Fig. 3a). A similar trend was also observed in the tau residue contact frequencies with the fibril elongation surface (Fig. 3b), where R4 and R2 had stronger binding than R3 and R1. For each of the tau repeats, the propensity of forming backbone hydrogen binds with the fibril coincided with the corresponding increase of β-sheets upon binding the fibril, suggesting that the increase of β-sheet content was mainly driven by the formation of inter-peptide hydrogen bonds with the exposed residues at the fibril elongation surface. At the elongation surface, the N-termini of R2 and R4 had similar binding to the fibril but the C-terminus of R4 had stronger binding than that of R2 (Fig. 3b). Comparison of sequences (Table 1) indicates that the C-terminal half of R2 is more charged as well as more hydrophilic than R4. For the binding of tau repeat residues with the fibril lateral surface (Fig. 3c), most of the peaks corresponded to positively charged Lys in all four peptides along with some hydrophobic or aromatic residues (e.g., residues Y5, V8, L10 in repeat R3; or residue L10 in repeat R1).

Table 1.

The amino acid sequence of each of the tau microtubule-binding repeats, R1-R4 as well as their sequence similarity using Clustal Omega.⁶⁶., : and * demonstrated an increasing degree of residue similarity

	Region	Sequence
R1	tau244–274	-QTAPVPMPDLKNVKSKIGSTENLKHQPGGGK
R2	tau275–305	VQIINKKLDL-SNVQSKCGSKDNIKHVPGGGS
R3	tau306–336	VQIVYKPVDL-SKVTSKCGSLGNIHHKPGGGQ
R4	tau337–368	VEVKSEKLDFKDRVQSKIGSLDNITHVPGGGN
Similarity		: : ..* ** ** *: * ****.

To visualize the binding of tau repeats onto the Aβ fibril surface, we computed the contact frequency of each residue in the Aβ fibril with tau repeats and colored all Aβ residues accordingly, constituting a binding heat map for tau repeats at the Aβ fibril surface (Fig. 4a). We also distinguished intermolecular Aβ-tau residue contacts with the lateral and elongation fibril surfaces and computed a two-dimensional potential mean force (PMF) as a function of the number of intermolecular residue-wise contacts at the two surfaces (Fig. 4b). The binding preferences of R1 and R3 with Aβ residues at the lateral fibril surface were evidenced by the red-colored high-frequency contact regions visible on the lateral fibril surface as well as the lowest free energy basins in the PMF with large number of lateral surface contacts but little contacts to the elongation surface. R2 and R4, on the other hand, tended to bind to the elongation surface of the fibril (Fig. 4a), corresponding to deep free energy basins in the PMF (Fig. 4b). The average binding probabilities of each repeat to two different fibrils surfaces are also summarized in Table S1.

Figure 4.

Contacts and secondary structure of tau repeats bound to different surfaces of the Aβ fibril. a) Contact map of tau repeats on the fibril surfaces. b) 2-dimentional PMF as a function of residue contacts for the tau repeat occurring at the elongation and lateral surfaces of the Aβ fibril. c) Residue contacts of simulation trajectories of R2 and R3 with two fibril surfaces as a function of time. Representative snapshots at different time along the trajectories were shown to the right. The Aβ fibril was show as carton in gray. The tau peptide was shown in stick and colored according to different secondary structures.

We also computed the average tau-binding frequency of Aβ residues with each tau repeat at the elongation and lateral fibril surfaces (Fig. S9). R1 bound primarily to Aβ residues F20 and E22, which were exposed to the solvent at one “edge” of the fibril lateral surface (Fig. 4a). R3 tended to contact the Aβ residue Y10 at the other exposed “edge” of the fibril lateral surface (Fig. 4a). The binding of both R2 and R4 with the elongation surface centered around Aβ11–21 and Aβ31–42 (Fig. S9), corresponding to the two amyloidogenic regions of Aβ42^44,48, consistent with our identification of the Aβ-tau hotspot region results (Fig. S4). Our analyses suggested that the binding of tau repeats with the fibril lateral surface was driven by electrostatic interactions since the Aβ peptide was negatively charged and the charged residues were exposed on the fibril lateral surfaces (e.g., especially with the same charged residues such as E22 aligned along the fibril lateral surface, Fig. 4). The aromatic interaction between the only Y5 in R3 of all tau MTBD repeats (Fig. 3) and the exposed Aβ Y10 (Fig. S9, aligned along the fibril lateral surface in Fig. 4) also played an important role for driving the differential binding PHF6* of R3 and PHF6 of R2 with the fibril, despite their sequence similarity. The binding of tau repeats with the elongation surface was mainly driven by hydrophobic interactions and the backbone hydrogen bonding.

For all four repeats, in addition to the two bound states at either the fibril elongation or lateral surfaces, there were also free energy basins in the PMF corresponding to intermediate states (Fig. 4b). As illustrated by the representative trajectories (R1, R4 in Fig. S10) and (R2, R3 in Fig. 4c), R1 and R3 mainly stayed on the lateral surface once bound. They could diffuse along the lateral surface to the elongation surface (e.g., 300 ns of the R1 simulation in Fig. S10), contributing to simultaneous contacts with both surfaces – i.e., the intermediate states. However, since their binding with the elongation surface was weak compared to with the lateral surface, the peptides usually did not remain in contact with the elongation surface and often diffused away (e.g., after 250 ns of the R3 simulation in Fig. 4c). R2, on the other hand, stayed at or near the elongation surface once it diffused along the lateral surface to the fibril end (e.g., after 150 ns in R2 simulation in Fig. 4c). With a weaker binding to the lateral surface, R4 could dissociate from the lateral surface and eventually remained bound the elongation surface (e.g., the R4 trajectory in Fig. S10).

Tau conformational dynamics at different fibril surfaces

To further characterize how different Aβ fibril surfaces affected the conformational dynamics of the tau repeats, we computed their average secondary structure contents when they were bound to the lateral or elongation surfaces and compared the results with those of isolated monomers in solution (Fig. 5a & Fig. S11). For all four tau repeats, binding with the Aβ fibril elongation surface promoted β-sheet formation through hydrogen bond formation with the fibril. The binding with Aβ fibril lateral surface only slightly increased the β-sheet formation for R1, R2 and R3, but decreased the β-sheet content for R4 instead. Compared to their isolated monomeric state in solution, the helices of R2 and R4 were reduced when they were bound to the fibril elongation surface, but were enhanced when they were bound to the lateral fibril surface. R3 formed little helices, as in the monomeric state, regardless of the presence of the Aβ fibril and where it was bound. The helical content of R1 was slightly decreased upon binding Aβ fibril, with the extent larger at the elongation end than at the lateral surface.

Figure 5.

Conformational analysis of tau repeats a) Helix and β sheet contents of tau repeats at the elongation, lateral surfaces of the Aβ fibril or the isolated repeat peptides. The formation of the fibril-tau hydrogen bonds differentiated which surfaces of the fibril that tau repeats bound to and simultaneously filtered secondary structure contents at the two surfaces. Error bars represent stander errors from average calculation of equilibrium samples. b) The 2D PMF of tau repeats R1-R4 as the function of Rg and the number of the fibril-tau hydrogen bonds, including representative snapshots of conformations from the corresponding potential basins (The Aβ fibril was demonstrated as carton in gray. The tau peptide was colored according to different secondary structures, β sheet in red, helices in cyan, coils in magenta).

We also computed the two-dimensional PMF as a function of the number of hydrogen bonds between tau and Aβ fibril vs the tau repeat Rg (Fig. 5b) or β-sheet content (Fig. S12). The analyses were done using the equilibrated last 500ns of trajectories for all independent simulations. Representative structures corresponding to free energy basins were also shown alongside the PMF plots. The deepest free energy basin of R1 had zero backbone hydrogen bonding with the fibril, where the peptide showed large conformational flexibility with a broad range of Rg and occasionally formed transient β-sheets. Two other shallow basins of R1 represented a weakly populated binding state at the elongation ends. Similarly, the lateral-binding state of R3 had the lowest free energy, which had a narrower range of Rg (Fig. 5b R3) and a high possibility of forming β hairpins (Fig. S12c). The fibril binding profiles of R2 and R4 were similar and both had deep free energy basins when bound at the elongation surface (Fig. 5b R2&R4). R2 had a prominent intermediate conformation in which R2 was bound to both surfaces of the fibril (e.g., also shown in Fig. 4c). Whereas R4 preferred to bind to the elongation surface of the fibril while intermediate binding states were less populated.

Aβ fibrils are known to be highly polymorphic.⁵⁰ Comparison of different Aβ fibril structures indicates that they all tend to bury hydrophobic residues inside the fibril core and expose hydrophilic residues on the fibril lateral surface, while the same residues are exposed at the elongation ends. Therefore, we expect different Aβ fibrils have similar binding trend and preference for each tau repeat, although subtle differences and the dependence on the fibril morphology remain to be determined in future studies. In a prior computational study³¹ where tau repeats of R2-R4 were assumed to already form β-sheet rich oligomers with fibril-like structures similar to an Aβ40 fibril (PDB ID: 2BEG⁵¹) and bind to the Aβ40 fibril via the alignment of β-sheets, only R2 was found to have strong binding due to better alignments in the turn region and the β-structure domain. Here, we did not presume the binding sites and the corresponding bound structures of tau repeats on the preformed Aβ fibrils, and focused on the early binding of tau with Aβ fibrils. Hence, our results along with the prior study suggest that R2 not only prefers to bind the elongation surfaces of Aβ fibrils but may also form aligned β-sheets. R4, on the other hand, might not be able to co-aggregate with the Aβ fibrils by forming aligned β-sheets although the repeat also prefers to bind the fibril elongation surfaces.

Taken together, R1 and R3 displayed stronger binding with the Aβ fibril lateral surface, which induced minor conformational changes compared to the solution structure, where R1 was more extended and R3 was more compacted as a result of β hairpins formation. The binding of tau at the lateral surface could enhance the local tau concentration in solution and might further play an important role in their self-aggregation via secondary nucleation. In addition, increased diffusion along the 1D fibril lateral surface than 3D in solution could also promote self-association of tau. In contrast, R2 and R4 showed stronger binding with the Aβ fibril elongation surface and enhanced β-sheet formation by forming fibril-tau hydrogen bonds. The binding of tau at the Aβ fibril and the subsequent promotion of β-sheet conformation could reduce the free energy barrier associated with conformational nucleation of tau.

Conclusion

In summary, we computationally identified the cross-interaction hot-spots between Aβ and tau by mimicking the experimental peptide array method. Using all-atom DMD simulations, the binding of full-length Aβ with a large number of sequence-overlapping fragments of tau was investigated. The computationally derived cross-interaction hotspots that featured strong Aβ-binding via inter-peptide hydrogen bonding agreed with experimental measurements.³⁶ In particular, the MTBD repeats of tau that have been found experimentally to be responsible for the formation of tau fibrils were also the hotspots to bind Aβ. The binding patterns were mainly driven by the hydrophobic effect and electrostatic interactions. We further investigated the binding and conformational dynamics of tau MTBD repeats with a preformed Aβ fibril seed. R1 and R3 tended to bind to the lateral surface of the fibril. R2 and R4, on the other hand, preferred to bind the Aβ fibril elongation surface and formed extensive interpeptide hydrogen bonds with the fibril. The distinctive binding patterns of tau repeats could contribute to different ways of promoting tau nucleation and aggregation. R2 and R4 could drive tau monomers to the elongation surface of Aβ fibril seeds, and the formation of β-sheets with the exposed fibril ends could reduce the aggregation free energy barrier associated with conformational nucleation. R1 and R3 could drive tau monomers to the lateral surface of Aβ fibril seeds. The increased the local tau concentration on the lateral surface and accelerated self-association with enhanced diffusion in one-dimension could also promote the secondary nucleation of tau. Together, our computational study offers a novel molecular insight to the experimentally observed cross-seeding of tau aggregation by preformed Aβ fibril seeds, which may help to better understand the pathological synergy of Aβ and tau aggregation in the etiology of AD and contribute to the future development of AD medical treatments.

Method and materials

Discrete molecular dynamics simulation

DMD is one type of molecular dynamics (MD) algorithms,⁵² where stepwise functions that mimic the continuous interaction potentials are applied to model interatomic interactions.⁵³ The dynamics in DMD are characterized a series of atomic collisions governed by the discontinuous potentials. Atomic positions and velocities at each collision are updated according to conservation laws of energy and momenta. DMD achieves improved sampling efficiency by only solving for the trajectories of colliding atoms rather than continuously integrating the forces and velocities as in traditional MD. DMD is widely used by our group and others in studying protein folding, amyloid aggregation, and nanoparticle–protein interactions.^54–57 In our atomistic DMD simulations, we used implicit solvent, united-atom models for biomolecules, and non-bonded interaction potentials including van der Waals (VDW), solvation, electrostatic, as well as hydrogen bonding. The VDW parameters were taken from CHARMM19.⁵⁸ The corresponding solvation energy was computed according to the Lazaridis-Karplus effective energy function, EEF1.⁵⁹ A reaction-like algorithm was used to model the distance- and angular-dependent hydrogen bond interactions.⁶⁰ Screened electrostatic interactions were evaluated using the Debye-Hückel approximation, in which a Debye length of 1 nm was used by assuming the water dielectric constant of 80 and a monovalent ion concentration of ~0.1 M. Counter ions (Cl− or Na+) were added accordingly to maintain a zero net charge of a simulated system and account for the counterion condensation effect⁶¹ in highly charged systems, such as Aβ fibrils with many changed residues aligned along the surface. The Andersen thermostat was utilized to maintain the constant temperature of 300 K.⁶² The units of mass, time, length, and energy were 1 Dalton, ~50 fs, 0.1 nm, and 1 kcal/mol, respectively.

Computational peptide array method

Given the large size of tau, we investigated the interactions between tau fragments and full-length Aβ42 to identify the cross-interactions hotspots between Aβ and tau. We intended to adopt a similar idea as the peptide array method^45–47 experimentally used to characterize protein-protein cross-interaction hotspots. The full-length tau441 with the sequence taken from the Uniprot database (P10636)⁶³ was split into a series of overlapping 14-residue fragments by shifting every seven residues along the sequence. For each tau fragment, we performed 15 independent DMD simulations at 300 K inside the simulation box with the dimension of 10 nm starting from differently randomized position, orientation and velocity. The minimum initial separating intermolecular distance was set to be 1.5 nm and each simulation lasted ~0.5 μs.

Before we applied the computational peptide array method to the Aβ-tau hotspot identification, we firstly applied it to the interaction between Aβ40 and IAPP to examine the idea in silico because the interaction has been systematically characterized in experiments⁴⁴ as the benchmark. Adopting a molecular setup similar to that used by Andreetto et. al.⁴⁴, the two amyloid proteins were split into continuously overlapping 10-residue fragments by sliding along the sequence of every amino acid. Each fragment of Aβ40 or IAPP was subjected to DMD binding simulations at 300K with the target protein of full-length IAPP or Aβ40, correspondingly. For each molecular system, 20 independent trajectories were simulated, starting with randomized initial positions and velocities. The starting configuration had the intermolecular atomic distances at least 2.0 nm. A cubic box with the dimension of 10 nm and periodic boundary condition was used, and each binding simulation lasted ~0.5 μs. Our simulation results were consistent with experimental measurements (Fig. S1). Specifically, the cross-interaction propensities of fragments from Aβ40 and IAPP derived from our simulations were directly comparable to the experimental results. As shown in Fig. S1a, fragments #10-#15 and #25-#31of Aβ40 displayed a strong binding with the IAPP monomer; meanwhile, fragments #9-#15 and #20-#25 of IAPP exhibited frequent interactions with Aβ40 monomer.

To evaluate molecular interactions and further indicate the bindings, we utilized reside-wise intermolecular contact frequencies. Two heavy atoms formed an atomic contact if their distance was less than the cut-off distance of 0.55 nm. Two residues or peptides contacted each other if they had at least one atomic contact. For each fragment split from the full-length peptide, we computed the residue-wise or inter-peptide contact frequencies by averaging over all independent simulations. According to how we split the IAPP or Aβ40 peptide into fragments, a residue could appear in up to 10 fragments interacting with the target peptide. Combining simulations of all the fragments, we pieced together a full contact frequency map between two cross-interacting peptides where the contact frequency of each pair was averaged over the number of molecular systems sampling their interaction. The corresponding contact frequencies were indicated by the grayscale, as shown in Fig. S1b. Although the contact intensities differed according to the full-length target protein, we could observe four regions in the constructed contact frequency maps, which corresponded to interactions among cross-interaction hotspots identified above (Fig. S1a). The computationally identified binding hotspots also agreed with the existing evidence both in vitro⁴⁴ and in silico.⁶⁴ Together, our analysis suggested that the cross-interaction hotspots in both Aβ and IAPP could be identified either by simulating Aβ fragments interacting with full-length IAPP or IAPP fragments interacting with full-length Aβ.

Interaction between tau microtubule-binding repeats and an Aβ protofibril seed

Our computational peptide array study confirmed that four tau repeats (Table 1) are indeed the cross-interaction hotspots with Aβ42. We chose an Aβ42 fibril structure recently resolved by cryo-EM experiment (PDB ID: 5OQV)⁶⁵ to model a preformed Aβ42 fibril seed (Fig. S2), since it was one of the most well-defined and high-resolution fibril structures of full-length Aβ42. The initial structures of tau repeats were fully extended, which quickly relaxed to coil conformations in DMD simulations. For each of the four repeats, the fragment and the preformed Aβ fibril seed along with counter ions were placed in a cubic box with the dimension of 15 nm. 30 independent trajectories were performed for each repeat at 300 K, starting from randomized initial positions and orientations with an initial 1.5 nm minimum intermolecular distance separating tau and the fibril. Each independent simulation trajectory lasted ~1 μs. For each tau repeat, control simulations in the absence of Aβ42 fibril seed were also performed for comparison.

Computational analysis

The secondary structures were computed using the DSSP algorithm.⁶⁷ Two atoms were in contact if their distance was less than the cutoff of 5.5 Å. Two residues formed a contact if they had at least one atomic contact between them. A two-dimensional potential mean force (PMF) was defined as –k_BT ln P(x, y) where P(x, y) was the probability of finding a conformation with parameters, x and y, during simulations. The formation of a backbone hydrogen bond between two residues was recognized if the N⋯O distance was less than 3.5 Å and the N−H⋯O greater than 150°.