From monomer to fibril: Abeta‐amyloid binding to Aducanumab antibody studied by molecular dynamics simulation

Alzheimer's disease is one of the most common causes of dementia. It is believed that the aggregation of short Aβ‐peptides to form oligomeric and protofibrillar amyloid assemblies plays a central role for disease‐relevant neurotoxicity. In recent years, passive immunotherapy has been introduced as a potential treatment strategy with anti‐amyloid antibodies binding to Aβ‐amyloids and inducing their subsequent degradation by the immune system. Although so far mostly unsuccessful in clinical studies, the high‐dosed application of the monoclonal antibody Aducanumab has shown therapeutic potential that might be attributed to its much greater affinity to Aβ‐aggregates vs monomeric Aβ‐peptides. In order to better understand how Aducanumab interacts with aggregated Aβ‐forms compared to monomers, we have generated structural model complexes based on the known structure of Aducanumab in complex with an Aβ2 − 7‐eptitope. Structural models of Aducanumab bound to full‐sequence Aβ1 − 40‐monomers, oligomers, protofilaments and mature fibrils were generated and investigated using extensive molecular dynamics simulations to characterize the flexibility and possible additional interactions. Indeed, an aggregate‐specific N‐terminal binding motif was found in case of Aducanumab binding to oligomers, protofilaments and fibrils that is located next to but not overlapping with the epitope binding site found in the crystal structure with Aβ2 − 7. Analysis of binding energetics indicates that this motif binds weaker than the epitope but likely contributes to Aducanumab's preference for aggregated Aβ‐species. The predicted aggregate‐specific binding motif could potentially serve as a basis to reengineer Aducanumab for further enhanced preference to bind Aβ‐aggregates vs monomers.


| INTRODUCTION
Alzheimer's disease (AD) is the most widespread lethal neurodegenerative disease in the world and its medical, social, and economic burden is steadily increasing due to our continuously growing and aging population. [1][2][3] Despite immense research efforts in the last decades, however, only a handful of symptomatic drugs are hitherto in clinical use, including acetylcholinesterase inhibitors like Donezepil and NMDA receptor antagonists like Memantine. [4][5][6][7][8] These provide limited cognitive improvement in mild to severe AD, but are far away from significantly interfering with disease progression, not to mention disease onset. [4][5][6][7][8] Due to the complexity of AD ranging from Aβ and Tau pathology to overall neuronal dysfunction and inflammation, many potential targets have been identified for future medical intervention, with a main focus on Aβ-amyloid formation due to its central pathological role according to the amyloid cascade hypothesis. [4][5][6]9,10 Amyloid-based treatment strategies range from reducing Aβ production, inhibiting Aβ aggregation to enhancing Aβ clearance. 4,11,12 The latter approach has reached several clinical phase II and III studies using passive immunotherapy with anti-amyloid antibodies. 11,13 After intravasal injection and blood-brain-barrier crossing, the antibodies bind to cerebral Aβ species and induce the patient's immune system to degrade Aβ via different mechanisms such as microglial phagocytosis. 11,14 Based on this immunotherapeutic strategy, anti-amyloid antibodies have been developed with varying affinity to Aβ from monomeric to aggregated forms. 13 However, several partially halted or entirely discontinued clinical studies have suggested restrictions on the benefit of such antibodies, as a general correlation between cerebral Aβ clearance and improvement of cognitive function could not be confirmed. 6,9,12,13,15,16 A multitude of potential explanations has been discussed in the literature, ranging from a partial or entire rejection of the amyloid cascade hypothesis over the need for an earlier, presymptomatic intervention with AD pathology to restrictions on the affinity profile required for clinically successful anti-amyloid antibodies. 4,12,13,[16][17][18] The latter is concluded from antibodies with negative or so far insignificant outcome like Bapineuzumab, Crenezumab, and Solanezumab which show similar or even higher selectivity for monomeric compared to aggregated forms. 6,9,12,13,15,16,19 However, such antibodies, injected in finite dose, may be captured away by the vast amount of monomers, leaving potentially neurotoxic species like oligomers and protofilaments undegraded. 20 In addition, targeting Aβ monomers may actually be harmful, as data suggest that these are involved in physiological function such as maintenance of the bloodbrain-barrier, antimicrobial and even neuronal protection. 21 In contrast, antibodies with ongoing clinical potential like Aducanumab (Biogen) and BAN2401 (Eisei/Bioartic Neurosciene) have considerably higher selectivity for aggregated vs monomeric Aβ forms (around 10 000-fold in case of Aducanumab). 9,15,19,[22][23][24] Although more data will be required, a reevaluation of initially discarded clinical late-stage studies provided hints that Aducanumab may indeed be able to combine a reduction in amyloid plaques load with cognitive improvement if applied in its highest tested dose of 10 mg/kg for around 80 weeks, which lead to Biogen's request for clinical approval in the end of 2019. 9,22,25 However, even if these data will be confirmed, various aspects need to be optimized; in particular, the right time point of intervention with AD pathology and, due to the presumably lifelong need for therapy, a reduction of necessary dose in order to remove side effects and costs. 20,24 The latter aspect may be addressed at a molecular level by further enhancing Aducanumab's blood-brain-barrier permeability as well as its selectivity for aggregated Aβ species vs monomers. A basis for quantitative optimization of epitope binding is provided by the X-ray crystal structure PBD 6co3 by Arndt et al which elucidates atomic interactions between the Fab region of Aducanumab (AduFab) and Nterminal Aβ residues 2 to 7. 19 However, enhancing Aducanumab's selectivity for aggregates requires binding characterization not only of the epitope, but of the entire (full) Aβ sequence, and most importantly a comparison of monomer binding vs aggregates like oligomers, protofilaments and mature fibrils. Due to the intrinsically disordered nature of monomers and highly dynamic oligomeric intermediates, such structural information is difficult to access by experiment. [26][27][28][29] In order to address these questions, molecular dynamics (MD) simulation is a suitable complementary tool with its simultaneous combination of atomic space and femtosecond time resolution. [26][27][28][29] For example, binding interactions from Aβ monomer to fibril were successfully modeled and simulated for Solanezumab, Crenezumab variants and the single-domain Gammabody, yielding insight into the antibodies' selectivity profile. [30][31][32] Besides computational epitope docking and alanine scanning, 19 comparable simulation of Aducanumab interacting with the entire Aβ sequence and/or oligomers, protofilaments and fibrils has not been performed to the best of our knowledge, yet is of significant relevance due to its planned clinical approval.
This work presents MD simulations of atomistic interaction models between AduFab and full-sequence Aβ 1 − 40 complexes of increasing oligomeric size in explicit solvent. Models are based on the crystal structure PDB 6co3, 19 that is, AduFab bound to the N-terminal Aβ 2 − 7 epitope, with the latter being extended to entire Aβ 1 − 40 monomers, oligomers, protofilaments and mature fibrils. From the vast polymorphism of resolved Aβ fibril structures, modeling is based on PDB 2m4j, which is a medically relevant Aβ 1 − 40 fibril structure derived from human Alzheimer's brain tissue. 33 This fibril structure is furthermore suitable due to its full-sequence resolution, in particular the N-terminal region which serves as epitope for Aducanumab. 19 After simulations of 250 to 1000 ns length, qualitative and quantitative analysis includes model stability and conformational order parameters of both AduFab and Aβ with a special focus on the AduFab-Aβ interaction surface. Binding motifs are compared between Aβ monomers vs oligomers, protofilaments and fibrils, providing potential for further improving the selectivity of Aducanumab for Aβ aggregates vs monomers.

| Simulation models
In this work, the interaction between the Fab-region of Aducanumab (AduFab) and Aβ 1 − 40 peptides is modeled and simulated in all-atom resolution and explicit solvent, ranging from Aβ monomers over short oligomers and protofilaments to an entire fibril segment.
Models are based on the X-ray crystal structure PDB 6co3, which resolves an Aβ 2 − 7 peptide fragment bound to AduFab. 19 In agreement with Arndt et al, AduFab is shortened to the relevant paratopeforming variable parts V L and V H of the light and heavy chain, respectively, in order to enhance simulation performance (see Figure 1A). 19 Aβ fibril structures underly a vast polymorphism, ranging from intrasample heterogeneity over the dependence on experimental conditions such as initial concentration, buffer composition and sample agitation to differences between in vitro and in vivo samples. [34][35][36][37] In order to study the medically relevant interaction between AduFab and Aβ-amyloid, the models presented in this work are based on a disease-relevant Aβ 1 − 40 fibril structure extracted from the brain of Alzheimer's patients, which was determined by solid-state NMR and electron microscopy (PDB 2m4j). 33 Of additional relevance, this structure resolves the entire Aβ 1 − 40 sequence, including the N-terminal region which is relevant for binding to AduFab.
AduFab-Aβ start model complexes were created by fitting an Aβ monomer, dimer and a hexameric protofilament from PDB 2m4j to the resolved Aβ 2 − 7 fragment bound to AduFab in PDB 6co3 (see Figure 1B-D). This is possible with little sterical strain that can be removed by energy minimization (see "Section 4" for further details).
In order to study if a protofilament in complex with AduFab is able to continue growth at its bound tip and the binding site can thus be located in the middle of a protofilament, the AduFab-Aβ hexamer model is elongated by an additional monomer, dimer and pentameric protofilament, respectively, which is denoted as "hexamer + 1", "hexamer + 2," and "hexamer + 5" in the following (see Figure 1E). In a final step, the AduFab-Aβ undecameric protofilament (hexamer + 5) is completed to an entire fibril consisting of three parallel, undecameric protofilaments (see Figure 1F). All MD simulations to refine the model complexes and study the conformational flexibility are performed at all-atom resolution and including explicit aqueous solvent (see "Section 4" for further details). Simulation times range from 250 to 1000 ns as summarized in Table 1.

| Complex stability
Within 250 to 1000 ns, all simulated AduFab-Aβ complexes remain stably bound. However, there are striking size-dependent differences F I G U R E 1 Modeling the interaction between AduFab and Aβ 1 − 40 amyloid. A, Crystal structure of Aβ 2 − 7 peptide fragment bound to AduFab (PDB 6co3). The Aβ fragment is shown in red-colored line representation. AduFab is depicted in van-der-Waals surface representation, with the paratope-forming variable parts V L and V H of the light and heavy chain being colored light and dark blue, respectively. The transparent domains below correspond to the constant regions of AduFab, which are discarded in order to enhance simulation performance. B-D, The bound Aβ fragment is completed to a full-length Aβ 1 − 40 monomer, dimer and hexamer based on the disease-relevant Aβ fibril structure PDB 2m4j. The Aβ peptides are shown in cartoon representation colored in gray and red, with red corresponding to monomers closer to the AduFab crystal-binding site. E, The bound Aβ hexamer is elongated by an additional monomer (hexamer + 1), dimer (hexamer + 2), and pentamer (hexamer + 5), respectively, in order to study growth of an already bound Aβ protofilament. F, The Aβ undecameric protofilament (hexamer + 5 model) is extended to an entire fibril bound to AduFab, consisting of three parallel, undecameric protofilaments in Aβ mobility with respect to the surface of AduFab, as measured in terms of the Aβ center-of-mass (COM) motion in Figure 2A and as visualized in Figure 2B-H.
The bound Aβ monomer and dimer show the highest mobility, which is lowered in case of the larger hexameric protofilament and the hexamer elongated by an additional monomer (hexamer + 1). In contrast, the mobility is drastically reduced in case of the models hexamer + 2 and hexamer + 5 and remains low in case of the bound fibril due to additional stabilizing contacts formed between N-terminal Aβ fragments and the AduFab surface.
The high monomeric and dimeric mobility can be further distinguished: While the Aβ monomer refolds from its initial β-hairpin structure to a more compact conformation, its C-terminal strand and loop regions start binding to the V L chain of AduFab, which provides stabilizing contacts in addition to the monomer's N-terminus in the AduFab crystal-binding site. This reduces the monomeric COM mobility after around 600 ns and results in a stably bound complex until the end of the simulation.

| Conformational order parameters
In the following, conformational order parameters are analyzed for AduFab and Aβ.

| AduFab
AduFab is a dimer consisting of the paratope-forming variable parts This can be quantified in more detail by calculating the rootmean-square fluctuation (RMSF) for each AduFab residue (see Figure 3B). While there are only minor fluctuations of <1 Å averaged over all AduFab residues in the crystal complex, the overall RMSF steadily increases from bound Aβ monomer to hexamer and hexamer + 1. This effect is especially pronounced in the V L domain, as the majority of additional contacts between AduFab and the growing hexamer are formed here (see Figure 2D,E).
With increasing Aβ size for hexamer + 2 and larger models, the mobility with respect to the AduFab surface decreases (see Figure 2A) and so does the overall RMSF. However, fluctuations remain significantly elevated compared to the crystal complex due to additional binding contacts formed with the increasingly growing amyloid entities.
By coloring AduFab according to RMSF values sampled in complex with the Aβ hexamer, it can be seen that the internal core of the AduFab dimer remains stable, while fluctuations occur in loop regions and outer β-sheet regions, where additional and partially fluctuating contacts with Aβ are formed (see Figure 3B).

| Bound crystal fragment
For all simulated models, the binding site and binding arrangement seen in the crystal structure are conserved in agreement with experiment. 19 However, there are model-dependent differences in the mobility of the bound N-terminal Aβ fragment.
For the isolated Aβ 2 − 7 fragment (as seen in the crystal structure), the average Cα-backbone RMSD and in particular its standard deviation (SD) is considerably higher compared to all other models, due to the lack of stabilizing neighboring Aβ residues (see Figure 3C). The corresponding per-residue RMSF shows fluctuations >4 Å for the terminal residues 2Ala and 3Glu, while the residues 4Phe, 5Arg, 6His, and 7Asp remain fixed in the binding site with an RMSF of around ≤2 Å (see RMSF plot and illustration i) in Figure 3D).
For large amyloid entities having a high mobility with respect to the AduFab surface (dimer, hexamer and hexamer + 1), the RMSF increases up to 8 Å due to drift motions of the N-terminal fragment in the binding site, which are induced by the global amyloid motion ( Figure 3D, illustration ii)). For the immobile hexamer + 2, hexamer + 5 and fibril model, the RMSF decreases to 2 to 4 Å for most residues and the N-terminal fragment is tightly confined in the crystal-binding site ( Figure 3D, illustration iii)).

| Amyloid cross-β structure
During simulation time, the cross-β structure is conserved from Aβ dimer to fibril, with an average β-sheet content of 37% and a coil content of 38.5% (for details see SI Table A.1). In contrast, the Aβ monomer refolds from its initial β-hairpin structure into a collapsed coil on the AduFab surface with a dominating average coil content of 46% and a minor β-sheet content of 13% (SI Table A.1).
As observed for isolated Aβ protofilaments and fibrils, also the bound Aβ entities from dimer to protofilament show a lengthdependent twisting motion, which can be quantified by a dihedral angle spanned between the two monomers at opposing protofilament tips (see Figure 3E). The average twist increases from 13 in case of the bound dimer to 102 for the undecameric protofilament

| The AduFab-Aβ interface
In the following, the interaction interface between AduFab and Aβ is analyzed in more detail.

| Interface statistics
The total average contact interface between AduFab and the fibril amounts to 14.8 ± 1.2 nm 2 , with an average number of 4552 ± 393 F I G U R E 3 Conformational order parameters for AduFab and Aβ. A, Boxplot of AduFab Cα-backbone root-mean-square deviation (RMSD) for all simulation models. Mean and median values are depicted as yellow stars and yellow horizontal lines, respectively. Boxes contain 50% of the data, the outer 50% are depicted as whiskers and red dotted points. B, Root-mean-square fluctuation (RMSF) per AduFab residue for each simulation model. The vertical black line splits the AduFab sequence into the light chain V L (residues 1-109) and the heavy chain V H (residues 110-235). Illustrated below the plot is an RMSF projection onto both AduFab domains in cartoon representation for the hexamer model. As depicted in the colorbar, low RMSF values are colored in blue, high values in red. The loop involved in the crystal-binding site is denoted as (*) in both plot and illustration. C, Boxplot of Cα-backbone RMSD of the crystal fragment Aβ 2 − 7 for all simulation models and, D, corresponding RMSF plot. The latter is illustrated by a time superposition of the bound Aβ 2 − 7 fragment in the crystal (i), hexamer (ii), and hexamer + 5 (iii) simulation. Aβ 2 − 7 is shown in cartoon representation and colored as a function of time from red to blue, while AduFab is depicted as gray van-der-Waals surface. E, Boxplot of the twist angle within the N-terminal β-sheet from dimer to fibril. The twist angle definition is illustrated in the plot inset: It is defined by four Cα atoms (blue spheres) of residues 12Val and 18Val of two monomers at opposing protofilament tips (see also magnified view in SI Figure A.3). The resulting twist is illustrated for the undecameric protofilament in the models hexamer + 5 (i) and fibril (ii) [Color figure can be viewed at wileyonlinelibrary.com] contacts and 14 ± 3 hydrogen bonds being formed (see SI

| AduFab surface occupancy
In order to find model-dependent differences in Aβ-AduFab interaction, an analysis of the AduFab surface occupancy is performed, that is, the maximum fraction of simulation time each AduFab residue is involved in a contact with Aβ (see Figure 4A). Contact-relevant surface regions and residues of AduFab are shown in Figure 4B and model-specific occupancies are illustrated in Figure 4C-F, together with maximum cluster representatives of Aβ resulting from interface clustering (see "Section 4").
The binding site defined by the crystal complex is highly preserved for all models (colored red and pink in Figure 4A,B). It can be further differentiated into a core-binding site with an AduFab residue occupancy >50% (colored red), while the surrounding region corresponds to weaker crystal contacts <50% which tighten with increasing Aβ size due to the binding of additional Aβ N-termini (colored pink).
F I G U R E 4 Occupancy of Adufab residues in complex with Aβ. A, For each simulation model, the occupancy of each AduFab residue is depicted, that is, the maximum fraction of simulation time of an existent contact with Aβ . Occupancies are colored from blue, that is, no contact, to red, that is, contact in 100% of simulation frames. On top of the plot, contact-relevant AduFab surface regions are summarized as bars colored according to the visualization in B. Here, AduFab is depicted as gray van-der-Waals surface, with the crystal core-binding site colored red and contact regions for larger Aβ models colored in pink, light, and dark blue. AduFab residues involved in crystal binding contacts and further stable binding motifs are labeled in black (for a detailed contact analysis, see Besides the crystal-binding site, two additional contact-forming regions can be identified: Binding of additional N-terminal Aβ fragments can be found on the heavy chain V H , which becomes relevant for the hexamer and larger Aβ entities (dark blue region in Figure 4A,B and illustration in Figure 4E,F as well as SI Figure A.6B-D). Another binding region can be found on the light chain V L , with diffuse contacts predominantly formed by the collapsed Aβ monomer and hexamer + 1 (see light blue region in Figure 4A,B and illustration D as well as SI Figure A.6B).
Differences in AduFab occupancy can be related to the modeldependent differences in Aβ mobility discussed in the context of  Figure 4A,B and illustration E,F). This complex stabilization increases with increasing Aβ size and so does the contact stability on the heavy chain V H (dark blue region in Figure 4A,B), while formation of contacts onto the V L chain decreases (light blue region in Figure 4A,B).

| An aggregate-specific binding motif
A medically relevant improvement of Aducanumab may comprise a further increase of AduFab affinity to Aβ oligomers, protofilaments and fibrils compared to monomers in order to decrease the necessary dose and side effects in clinical applications. It is hence of interest to detect aggregate-specific contacts in addition to the experimentally observed crystal-binding site, which is occupied unspecifically from Aβ monomer to fibril.
While the entire monomer sequence collapses onto the Adufab light chain V L (see light blue monomer in Figure 5A and corresponding light blue surface region in Figure 4B), larger Aβ protofilament entities predominantly bind with their N-terminal sequence regions to AduFab, while the loop-regions and C-terminal sheets mainly remain solvent-exposed (see Figure 2).
In order to identify possible aggregate-specific binding motifs, an overlay of all bound N-terminal Aβ fragments is shown in Figure 5A for the maximum cluster representatives of each simulation model. In agreement with the quantitative surface occupancy analysis in Figure 4, the crystal-binding site is conserved for all models (see red fragments in Figure 5A). The majority of further N-terminal fragments are found to bind in an unspecific, diffuse manner mostly to the V H domain (see blue fragments in Figure 5A and blue surface region in Figure 4B). However, in front of the crystal-binding site, an additional N-terminal binding motif can be identified in case of the models hexamer + 1/2/5 and fibril (see pink fragments in Figure 5A Figure 6C vs E as well as SI Figure A.7). The latter is the case for the models hexamer + 5 and fibril, in which the fragment's tip residues 1Asp-4Phe detach from AduFab in order to form contacts, in particular hydrogen bonds, with neighboring Aβ strands. Only the residues 6His and 7Asp and to a certain amount 5Arg of the fragment remain involved in a contact network with AduFab, including the surface residues 28Ser, 30Ser, 92Tyr, and 93Ser (see Figure 6A,E).
This contact network is also established in a slightly weakened form for the models hexamer + 1/2. In addition, as there are fewer surrounding intraaggregate Aβ strands to compete with, the fragment's tip residues 1Asp-4Phe form contacts with AduFab, in particular between the oppositely charged residues 1Asp of Aβ and 174Lys of AduFab (see Figure 6A,C).

| Analysis of the crystal-binding site
While the crystal-binding site is occupied in all simulated models, a detailed contact analysis suggests slight changes in several crystal binding contacts in case of bound Aβ dimers to fibrils compared to F I G U R E 6 Analysis of the aggregate-specific binding motif. A, Contact maps (y-axis) between the bound Aβ fragment and AduFab are calculated for each model (x-axis), with contact occupancies colored from white, that is, no contact, to dark blue, that is, contact existing in 100% of simulation time. Average interaction energies per Aβ fragment residue are calculated in case of the models hexamer + 2, B, and hexamer + 5, D. The histograms show time average and standard error (SE) of the total interaction energy per Aβ residue (gray) as well as van-der-Waals (blue) and electrostatic Coulomb contributions (red). Corresponding model-dependent contact differences are visualized in case of the models hexamer + 2, C, and hexamer + 5, E. Illustrations of the simulated models correspond to maximum cluster representatives. AduFab is shown in van-der-Waals surface representation, colored according to the physicochemical property of surface amino acids, that is, hydrophobic (white), polar (light blue), positively and negatively charged (blue and red). The same color coding applies for the bound Aβ fragments, which are depicted in backbone and side chain representation. Aβ fragment residues are labeled in black, contact-relevant AduFab residues in gray. Sterically not visible AduFab contact residues 91Ser, 161Trp, 211Ile, 215Arg, 216Gly, and 217Pro are preponderantly located at the bottom of the crystal-binding site and are illustrated in Figure 4B Figure 7E). This is associated with an unfavorable increase in average Coulomb energy for 5Arg from −1.7 kcal/mol to +2.1 kcal/mol and, including slight changes in van-der-Waals interaction, an overall increase in average interaction energy from −8.4 to −3.9 kcal/mol (compare Figure 7B,D).
Compared to the interaction hotspots 4Phe and 6His, fewer contacts, reduced contact occupancy and lower average interaction energies are found for Aβ residues 1 to 3 and 8 to 10, in agreement with the experimental suggestion that these residues are disordered. Aducanumab's high affinity to aggregates, as binding is directly possible without global, energy-intensive conformational rearrangements of Aβ or the AduFab surface. This is in contrast to antibodies like the discontinued Ponezumab, which binds epitopes located in the Cterminal Aβ 30 − 40 sequence. [13][14][15][16][17][18][19] Due to its location inside the hydrophobic fibril core, the epitope accessibility drastically decreases with increasing aggregate size, which may be correlated to the considerably higher antibody selectivity to monomers compared to aggregates.

| DISCUSSION
Antibodies like Solanezumab and Crenezumab with epitopes Aβ 16 − 26 and Aβ 13 − 24 being located at the inner Aβ N-terminal strand and loop region show higher or at most comparable selectivity for monomers and aggregates. [13][14][15][16][17][18][19] Although these epitope regions are located at least partially at the exterior fibril surface (in the example of PDB 2m4j fibril structure), they are involved in the tight β-sheet stacking of the central hydrophobic fibril core, which may reduce their flexibility to adapt to the antibody surface compared to Aducanumab's epitope Aβ 2 − 7 being located at the flexible tips of the N-terminal β-sheet.
As discussed in Arndt et al, 19  The hexamer + 5 model with an undecameric protofilament bound to AduFab was extended by adding two more protofilaments in order to form a mature fibril according to the quaternary structure of PDB 2m4j. This was sterically enabled due to quaternary fibril contacts not affecting the N-terminal β-sheet tips, which remain solvent-exposed and accessible for binding to AduFab.

| Simulation details
The Charmm22* force field was used for model parametrization, enabling the use of neutral termini for all amino acids in order to prevent potential charged-induced instability of the Aβ protofilament and fibril models. [44][45][46] This force field has been used successfully in previous studies on disordered proteins and Aβ aggregation 47 Table 1. Simulation frames are saved every 20 ps.

| Simulation analysis
Simulation analysis was performed using tools from Gromacs 2018 [40][41][42][43] and Amber cpptraj version 18.00. 59 The mobility of Aβ with respect to the AduFab surface is characterized by the three-dimensional COM of the respective Aβ entity after fitting the trajectory to the initial AduFab conformation in order to remove global translation and rotation. For AduFab and the crystal Aβ 2 − 7 fragment, root-mean-square deviation (RMSD) values are calculated including all Cα-backbone atoms. In case of AduFab, the RMSD is calculated for the entire dimer (1Asp-235Ser) and separately for both domains (1Asp-109Thr, 110Gln-235Ser). The average mobility of each AduFab residue and each residue in the Aβ 2 − 7 fragment is quantified by the root-mean-square fluctuation (RMSF) over the entire trajectory, after fitting the trajectory to the initial AduFab and Aβ 2 − 7 conformation, respectively. Aβ conformations are characterized by the average secondary structure content calculated using the DSSP algorithm. 60 Relevant secondary structure elements are coil, bend, turn, and β-sheet content, while α-helical content is of minor relevance for the cross-β structure. In addition, the twisting motion along the Aβ protofilament axis is quantified using a dihedral angle within the N-terminal β-sheet, which is larger and more stable than the C-terminal β-sheet. The angle is spanned between four Cα-atoms of residues 12Val and Val18 within two monomers at the opposing protofilament tips. For hexameric and larger protofilaments, the peripheral monomers directly at the tips are omitted due to considerable conformational fluctuations. The AduFab-Aβ interface is characterized by the interface area, the number of interface contacts and hydrogen bonds. The interface area is obtained by calculating the solvent-accessible surface area (SASA) according to Eisenhaber et al with standard Gromacs parameters for the AduFab-Aβ complex (A total ) and the isolated AduFab and Aβ structure (A AduFab , A Aβ ). 61 The interface area results from the subtraction A AduFab + A Aβ -A total . Furthermore, all SASA contributions are split into hydrophilic and hydrophobic parts. For the contact analysis, an interface contact between AduFab and Aβ is defined to exist if the interatom distance between two residues is <0.6 nm. The number of interface hydrogen bonds is defined according to standard Amber definitions, corresponding to a donor-acceptor distance cutoff <3.0 Å and an angle cutoff >135 . 59 Statistics on each individual interface contact is collected from frame-wise contact maps using the Conan analysis tool. 62

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1002/prot.25978.