The accumulation of aggregated β-Amyloid (Aβ) in the brain is a hallmark of Alzheimer's disease and is thought to play a role in the neurotoxicity associated with the disease. The mechanism by which Aβ aggregates induce toxicity is uncertain. Nonetheless, several small molecules have been found to interact with Aβ fibrils and to prevent their toxicity. In this paper we studied the binding of these known toxicity inhibitors to Aβ fibrils, as a means to explore surfaces or loci on Aβ aggregates that may be significant in the mechanism of action of these inhibitors. We believe knowledge of these binding loci will provide insight into surfaces on the Aβ fibrils important in Aβ biological activity. The program DOCK was used to computationally dock the inhibitors to an Aβ fibril. The inhibitors docked at two shared binding loci, near Lys28 and at the C-termini near Asn27 and Val39. The docking predictions were experimentally verified using lysine specific chemical modifications and Aβ fibrils mutated at Asn27. We found that both Congo red and Myricetin, despite being structurally different, bound at the same two sites. Additionally, our data suggests that three additional Aβ toxicity inhibitors may also bind in one of the sites. Identification of these common binding loci provides targets on the Aβ fibril surface that can be tested in the future for their role in Aβ biological activity.
The accumulation and aggregation of β-amyloid (Aβ), a 39–43 residue peptide cleavage product of the transmembrane amyloid precursor protein, are believed to play a key role in Alzheimer's disease (AD). Although it is established that Aβ is toxic only when aggregated,1, 2 the mechanism by which Aβ aggregates interact with cells and lead to neurodegeneration is controversial. Efforts targeted toward understanding the toxicity mechanisms of Aβ have suggested that certain Aβ residues are important in those interactions. Aβ binding to the cell membrane via positively charged amino acids was proposed.3 Specifically R5, K16, and K28 were reported to be essential for Aβ toxicity.4 Instead, M35 was suggested to be involved in the formation of free radicals and oxidative stress in Aβ42.5 F20 has been implicated in Aβ42 fibril-induced lipid peroxidation, possibly because of the role of F20 in an electron transfer from M35 to a Cu(II) bound at the N-terminus.6 Alternatively, residues H13–H14 have been hypothesized to play a role in Aβ-induced membrane pore formation7, 8 and in binding of metal ions implicated in AD.9 In addition, exposed hydrophobic regions of Aβ aggregates are thought to interact with the membrane directly and cause changes in membrane properties, such as fluidity and conductance.10, 11 Clearly, there is neither consensus regarding the mechanism by which Aβ induces neuronal death nor about the Aβ amino acid residues that may be involved in such a mechanism.
Despite the limited understanding of Aβ neurotoxicity, a variety of small molecule inhibitors have been found to interact with fibrils or aggregated Aβ and attenuate Aβ toxicity. We hypothesized that by elucidating the binding sites of these toxicity inhibitors on aggregated Aβ, we could identify residues or surfaces on Aβ aggregates that are important for Aβ biological activity.
Toward that end, in this article, we describe a study aimed at finding the binding sites of previously reported toxicity inhibitors onto an Aβ fibril. We used computational docking to predict the binding sites of the inhibitors, and then tested some of the predictions experimentally. Five toxicity inhibitors were investigated in this study: Congo red (CR),12–14 Myricetin (Myr),15, 16 melatonin,17, 18 nicotine,19–21 and curcumin.22, 23 Because only aggregated Aβ species are toxic, we included only inhibitors that were reported to bind or interact with Aβ fibrils or other β-sheet aggregates. For example, CR binds to Aβ aggregates24 and was shown to prevent the toxicity of preformed Aβ aggregates.12, 13 Myr16 and nicotine20 were reported to bind to Aβ aggregates or fibrils, and melatonin was shown to prevent the toxicity of preformed Aβ aggregates.18 Although some of these inhibitors have been reported to bind to Aβ monomers or small oligomers and prevent Aβ aggregation, in this study we specifically examined their reported ability to bind to Aβ fibrils.
Our results suggest that CR and Myr, despite being structurally distinct, shared two binding sites on Aβ fibrils; near Lys28 and at the C-termini near Asn27 and Val39. Furthermore, our data suggests that three additional toxicity inhibitors; melatonin, nicotine, and curcumin, may also bind near Lys28. Our identification of common binding loci of several toxicity inhibitors may suggest that these sites are important in the biological activity of Aβ fibril. This work contributes to our understanding of the mechanism of action of Aβ toxicity inhibitors and highlights potential targets for the development of novel therapeutics for Aβ toxicity prevention.
The goal of our investigation was to elucidate binding sites on the Aβ40 fibril important in toxicity inhibitor binding. To accomplish this goal, we first used computational docking to identify possible binding sites for the inhibitors, which were then tested experimentally by altering an amino acid at the predicted sites. For the docking step, we considered several binding sites on Aβ based on previous reports of possible binding sites of the inhibitors. Additionally, as we hypothesized that common binding sites of inhibitors may exist at loci that are essential for Aβ activity, we also considered as binding sites Aβ residues that were proposed to be important for Aβ toxicity (or other biological activity). As a result, the inhibitors were docked at five different locations on the Aβ fibril: (1) residues His13–His14, (2) residues Lys16–Phe20, (3) residue Lys28, (4) residue Met35, and (5) residues Val36–Val40 at the C-terminus. Table I summarizes the specific reasoning behind the selection of these sites and Figure 1 illustrates their location on an Aβ fibril segment.
Table I. Initial Guesses for Binding Sites of Inhibitors on Aβ Fibrila
Binding sites were considered either because they were proposed to be a binding location for a toxicity inhibitor, or because they were hypothesized to be important for Aβ interaction with cells.
Relatively solvent accessible although hydrophobic29 hypothesized to play a role in membrane binding30
Docking was performed using a set of 20 Aβ fibril NMR structures as a receptor.31 The NMR Aβ fibril structures were 12-mers of Aβ (9–40) that represent roughly a 25 Å segment of a fibril. The 20 structures included two possible staggers in the fibril orientation and 10 possible side chain configurations for each stagger, which were determined based on the best fit of NMR data to the experimental constraints.31 Four toxicity inhibitors were used as ligands: CR, Myr, Melatonin, and Nicotine. To cover a diverse range of inhibitors, we included only inhibitors that were structurally dissimilar. For example, among several sulfonated dyes that can attenuate Aβ toxicity,12 we included only CR, which is the most commonly used dye. Additionally, as DOCK works better with ligands with up to seven rotatable bonds, inhibitors such as curcumin with a higher degree of flexibility were not docked.
The DOCK program was designed and validated for docking at a defined binding pocket with limited volume.32 We therefore docked the inhibitors against the five different binding pockets individually (Table I), and later integrated and ranked the best predicted energy scores for all the sites. Figure 3 in the Supporting Information indicates that the five pockets overlapped such that they covered the entire Aβ surface. To ensure the validity of this approach and test whether the true binding site can still be found among several pockets, we first tested the methodology on a validation set of 114 receptor–ligand complex structures. For each complex structure, the ligand was docked at multiple randomly chosen receptor binding pockets, such that when combined the binding pockets covered the entire receptor surface. The energy score of DOCK at the true binding site was among the top two scored ligand positions in nearly 86% of the receptor–ligand pairs. The high success rate of DOCK at finding the correct ligand position indicates that by docking a ligand at multiple partially overlapping binding pockets, DOCK is capable of finding the true binding site. Because the volume of each pocket in DOCK is limited, this method becomes useful when the region of the ligand binding is unknown and multiple binding pockets need to be considered. The full description of the docking method and results of the validation set are available in Figures 1 and 2 of the Supporting Information.
The docking procedure was repeated for all 20 Aβ fibril structures. The docking results of the inhibitors at the five sites we considered were visually inspected to eliminate unreasonable poses. Poses where inhibitors bound at the end surfaces of the fibril (top two and bottom two Aβ chains) were discarded, as well as poses where an inhibitor interacted with the artificial G9 N-termini (formed because residues 1–8 were missing in the NMR structure). Binding sites were considered to be the same if the receptor–ligand interactions were similar The symmetry of the fibrils along the z-axis sometimes resulted a in transposed ligand position with a large RMSD even though detailed inspection revealed that the ligand maintained the interactions with the same amino acids on the fibril surface. For this reason, RMSD and volume Tanimoto coefficient are not reported. No differences were observed in docking poses and energies when inhibitors were docked to fibrils of different staggers.
Table II summarizes the average top scored sites among the 20 fibril structures for each inhibitor, and the amino acids at the binding site for each inhibitor. CR, Myr, and melatonin shared two binding sites among the top three best-scored positions; at the β-turn between residues A21–A30, and at the nook located near the C-termini, between amino acids V39–V40 of one fibril layer to amino acids N27 and G29 of the other fibril layer (Fig. 2). The 20 Aβ fibril models we used differ slightly from each other in their β-turn cavity size and configuration, which led to some variability in the best-scored poses of the inhibitors within the β-turn cavity. In some cases, the inhibitors docked near F19, A21, and I32, whereas in other cases, the inhibitors were closer to the β-turn itself, and docked between A21 and A30, while having electrostatic interaction of charged or polar atoms with the ε-amino group of K28, or with the carboxyl group of D23. In both cases, whether docked closer to K28 or to F19, the inhibitors docked parallel to the fibril axis, such that each inhibitor interacted with 2–4 stacks of Aβ monomer chains, depending on the size of the inhibitor. The exact orientation and configuration of the inhibitors varied slightly between the different fibril structures, yet the energy scores did not change significantly. The differences between the inhibitors' poses when docked with different Aβ structures were mostly in rotamer angles between aromatic rings of the inhibitors. However, rather than revealing the exact pose of the inhibitors, our main goal was to identify the amino acids on the fibril adjacent to the binding sites of the inhibitors, which remained fairly invariant (Table II). Representative images of CR and Myr docked inside the β-turn are shown Figure 2(A–B).
Docking sites and energy scores as predicted by DOCK for CR, Myr, melatonin and nicotine. The rank of the sites was determined by analyzing the docked poses and interactions, the energy scores, and the recurrence of the pose and interactions amongst the 20 Aβ fibril structures. CR also docked at a third site with a score of −41.2 kcal/mol near the C-termini while interacting with N27, K28 backbone, G29, V39, and the C-terminus.
F19, A21, D23, K28, A30, I32
H14, K16, V18
N27, K28 backbone, G29, V39, C-terminus
A21, D23, K28, A30,
A21, D23, K28, A30
N27, K28 backbone, C-terminus
β-turn (near A21, D23, K28) and C-termini (near N27, V39, V40) with 20–25 kcal/mol and near H14 with 15–25 kcal/mol, with no site showing clear higher energy scores.
The second common binding site shared by the inhibitors was near the C-termini. The energy scores of the inhibitors at the C-termini were similar to their scores at the β-turn, except for CR, which docked at this location with slightly lower energy. The energy was composed mostly of Van der Waals (VdW) interactions between the inhibitors and V39–V40 of one fibril layer and the Cβ of N27 and the backbone amide hydrogen of G29 of the second layer. Electrostatic energy contribution associated with binding at this location came from hydrogen bonds with the side-chain amino group of N27, the backbone carbonyl oxygen of K28, and with the carboxyl group at the C-termini. Similar to the poses in the β-turn site, the inhibitors docked parallel to the fibril axis and interacted with 2–4 Aβ stacks. Again, the exact inhibitor configuration differed slightly when docked to the 20 fibril structures. Yet, the overall binding location was conserved. The amino acids that interacted with the inhibitors at this position include N27, G29, V39, V40, and the backbone carbonyl oxygen of K28. Representative images of CR and Myr docked at the C-terminal site are shown in Figure 2(C–D).
In addition to these two positions, CR docked successfully at two other sites, near K16–V18 and near H14–K16 [Fig. 2(E)]. When docked near K16–V18, the sulfonate oxygens of CR formed hydrogen bonds with the primary amine of K16, whereas the central hydrophobic biphenyl group had VdW interactions with V18. With other Aβ structures, CR docked between H14 and K16 side chains, forming hydrogen bonds with both residues. The electrostatic contribution in both poses was significantly higher than the electrostatic contribution to the binding energy at the two previously described sites and constituted nearly a third of the total energy score for CR binding. The apparent interaction with the histidine should be regarded carefully, as the assigned protonation state of its side-chain varied between the different structures. The docking of CR near K16, either between H14–K16 or K16–V18, had similar energy scores to the docking of CR within the β-turn near K28, and was observed more frequently than docking at the C-terminal site among the 20 different Aβ structures. The other inhibitors did not dock near K16. We were unable to elucidate a clear preferred binding site for nicotine as it docked nonspecifically with only modest binding energies (Table II). None of the inhibitors bound near M35, because of the low solvent accessibility of the methionine in all fibril structures. It is important to note that regardless of the initial docking location (i.e., near K16) the inhibitors consistently docked in the β-turn near K28 or at the C-terminal site. In addition, when we screened a library of nearly 2000 compounds from the NCI diversity set for docking against these two binding loci, only 2% of the library bound with energies comparable with the scores of CR. This result suggests that noninhibitors are unlikely to dock successfully to the Aβ fibril structure.
We experimentally tested the predictions of the computational docking using Aβ fibrils that were prepared analogously to methods used for NMR spectra collection.31 Fibrils were chemically modified or mutated at specific locations near the predicted binding sites. Equilibrium binding isotherms of CR and Myr with chemically modified, mutated, or wild type (WT) Aβ fibrils were obtained. Isotherms were then compared to determine if the amino acid modification or mutation perturbed the binding.
We used reductive methlylation, as described previously,33, 34 to modify the Aβ lysines to probe whether the inhibitors bound inside the β-turn of the fibril, near K28. Each exposed primary amine of an Aβ fibril typically adds two methyl groups under the reaction conditions used in this study.34 The methylation results in an increase in mass of 14 Da per methyl addition that can easily be followed by mass spectrometry. As seen in Figure 3(A), most of the Aβ added either 4 or 6 methyl groups, suggesting that all three primary amines, the N-terminus, K16, and K28, were at least partially modified. For the remainder of this article, we abbreviate this modification of Aβ as Aβ-K-CH3.
To distinguish between the affect of modification of K28 and K16, we needed to selectively modify only one of the lysine amino acids. To that end, we reductively ethylated Aβ and assumed that K28, which has been shown to be less solvent accessible,34, 35 would be less reactive toward the larger aldehyde.36 The extent of ethylation of the Aβ fibrils was again followed by MALDI TOF, where an increase in mass of 28 Da indicated the addition of one ethyl group. As seen in Figure 3(A), only 2 ethyl groups were added to Aβ fibrils at the completion of the reaction, one presumably at the unstructured N-terminus, and the other at either K16 or K28. We used tryptic digestion followed by mass spectrometry of the chemically modified Aβ fibril to determine the locations of the added ethyl groups [Fig. 3(B)]. Aβ fragment 6-28, which contains both lysines, appeared at 2672.5 Da, instead of at the theoretical mass of 2643.3 Da, indicating that one ethyl group had been added. In contrast, fragment 17–28, which includes only K28, appeared unmodified at 1325.1 Da. Hence, the modification with acetaldehyde added an ethyl group to K16, whereas K28 was not modified. We designated this modified fibril for the remainder of this article as Aβ-K16-C2H5.
In the second binding site predicted by docking, between the C-terminus of one Aβ layer and the outside of the β-turn of the second layer [Figs. 2(C–D)], the inhibitors interacted with amino acids V39, V40, N27, G29, and the backbone of K28. To experimentally test the binding of the inhibitors at this location, we used a single amino acid Aβ mutant, in which N27 was replaced with proline. We designated this mutant as N27P for the remainder of this article.
Before testing the binding of inhibitors to the different Aβ fibrils; WT, Aβ-K-CH3, Aβ-K16-C2H5, and N27P fibrils, we confirmed using TEM that the fibril structures were not disrupted. As shown in Figure 4, the structures of chemically modified and mutated fibrils were similar to that of wild type fibrils. Circular dichroism (CD) spectra indicated that no changes in the secondary structure composition occurred as a result of the methylation of Aβ fibrils, as spectra of both WT and Aβ-K-CH3 fibrils had a minimum at ∼219 nm characteristic of a β-sheet motif. The secondary structure of N27P fibrils, as determined by CD, was typical of a β-sheet and comparable with WT fibrils. Nonetheless, the CD of N27P fibrils showed lower light scattering below 210 nm (evident by noise) compared with WT and Aβ-K-CH3 fibrils, possibly because of N27P formed shorter fibrils (Figure 4, Supporting Information). Analysis of the amount of insoluble aggregates versus soluble peptides using a BCA assay indicated that the local changes induced by chemical modification (both methylation and ethylation) and mutation did not alter the equilibrium between the insoluble fibrils and other soluble forms of Aβ (data not shown). Preliminary examination of fibril unfolding in the presence of denaturant indicated that there were no differences in fibril stability between WT and the methylated fibrils (data not shown). Lastly, the fibril content of the different fibril samples were qualitatively compared using the ThT fluorescence37 and CR absorbance shift24 assays. Both assays indicated that the fibril content of all Aβ types did not change as a result of the chemical modifications or mutation compared to WT fibrils (data not shown). Taken together, our data suggest that the reductive alkylation of K28 and/or K16, or the mutation of N27, did not disrupt the fibril structure and solubility, and if any structural changes were associated with the amino acid alterations, the changes were localized to the environment of the modified amino acid.
We then tested the binding of CR and Myr to the four fibril types. Representative equilibrium binding isotherms for CR and Myr bound to WT, Aβ-K-CH3, Aβ-K16-C2H5, and N27P fibrils are shown in Figure 5. We compared the dissociation constant (Kd) and maximal binding sites availability (Bmax) of all isotherms to identify differences in the binding of CR and Myr. We did not identify significant differences in Kd values of CR or Myr. Figure 5 summarizes the Bmax values of CR and Myr to the four different fibril types.
At high concentrations of ligand, both CR and Myr bound less to the Aβ-K-CH3 fibrils than to WT fibrils, suggesting that their binding was partially blocked by the methylation of lysines in the Aβ-K-CH3 fibrils. This result was more pronounced in the Myr-Aβ binding isotherms [shown in Fig. 5(A)] than in the CR-Aβ isotherms (isotherm not shown), where the difference in binding at high concentrations was only significant at P = 0.1. The binding isotherms of both CR and Myr to Aβ-K16-C2H5 fibrils were similar to the binding isotherms to WT fibrils, suggesting that the addition of an ethyl group onto K16 did not affect ligand binding. In summary, the modification of the N-terminus, K16, and K28 lowered the binding capacity of Aβ fibrils for both CR and Myr, whereas the modification of only the N-terminus and K16 did not affect CR and Myr binding. The difference between the binding of CR and Myr to Aβ-K-CH3 and Aβ-K16-C2H5 fibrils suggests that both inhibitors bound near K28. The binding isotherms of CR and Myr to N27P fibrils demonstrate that these fibrils had an altered binding capacity for both inhibitors compared with WT fibrils. Although binding of Myr to N27P was enhanced [Fig. 5(A)], binding of CR was reduced relative to WT fibrils [Fig. 5(B)]. The molecular interactions that led to the different effects of the N27P mutation on CR and Myr binding remain unclear. Yet, the data suggest that both Myr and CR binding were affected by the mutation of N27, consistent with the docking predictions.
As a control, we examined the effect of chemical modification of histidines on the Aβ fibril on Myr binding, an amino acid residue not predicted by docking to be near the ligand binding site. As expected, histidine modification did not significantly alter Myr binding to Aβ fibrils (data not shown).
The binding of nicotine and melatonin, the two other inhibitors we used for docking, was experimentally examined indirectly. WT Aβ fibrils were methylated as described, however the extent of the modification was compared in the presence or absence of the inhibitors during the course of the reaction. Curcumin, an inhibitor we were unable to dock because of its high flexibility, was also included in this experiment. Aβ fibrils were preincubated with nicotine, melatonin, curcumin, or Myr, and then reacted with cyanoborohydride and formaldehyde to modify the lysines. Figure 6 shows representative mass spectra obtained after 90 min of reaction. As shown, there are significant differences in the number of modified sites between Aβ fibrils in the presence or absence of the inhibitors. After 90 min, very little of the Aβ in the absence of an inhibitor remained unmodified, with most of the Aβ adding mass equivalent to three or more methyl groups. In contrast, in the presence of inhibitors, a significant fraction of the Aβ remained unmodified or increased in mass equivalent to one or two methyl groups. Similar trends were observed at later time points (data not shown). The presence of nicotine, melatonin, curcumin, and Myr did not affect the methylation of insulin (data not shown), suggesting that the effect was specific to Aβ fibrils.
A plethora of reported biological phenomena is associated with Aβ interactions with cells, some of which were linked to specific Aβ residues. For example, Aβ-inducted oxidative stress38 was associated with Y10, F20, G33, and M35,5, 6, 39, 40 Aβ interactions with the negatively charged cell membrane were associated with R5, K16, or K28,3, 4 whereas histidines were associated with membrane pore formation.7, 8 Given this apparent wide range of possible Aβ-cell interactions, and the nonspecificity of some Aβ interactions,27 it has proven challenging to elucidate residues on Aβ associated with the Aβ-cell interaction directly. We proposed here an alternative approach, in which knowledge of the binding site of Aβ toxicity inhibitors may serve as an indirect assessment of Aβ amino acids that are crucial for Aβ biological activity. We assume that inhibitors act by binding to Aβ at locations that are meaningful in terms of Aβ biological activity, and in this manner attenuate Aβ toxicity. As a first step in testing that hypothesis, we report here a residue level assessment of the binding of small molecular weight toxicity inhibitors to Aβ fibrils. Ultimately, we will examine the role of those residues in Aβ cell interactions and toxicity.
Our methodology for identifying binding sites of toxicity inhibitors on Aβ fibrils was comprised of two steps. First, we computationally explored the binding by docking the inhibitors onto an Aβ fibril structure. Then, we experimentally tested the sites predicted by docking, using Aβ fibrils chemically modified or mutated at specific amino acid residues.
The only available Aβ40 structures were determined using solid state NMR,31, 41 which is not ideal for molecular level computational modeling because of the relative low accuracy of the of side chain positions in the structures.31 Yet, in the absence of available X-ray structures, NMR structures have previously been used successfully for molecular docking.42–44 To address some of the uncertainty derived from the limited accuracy of the Aβ40 fibril structure, we docked the inhibitors against 20 fibril structures, 10 of each stagger type (see Ref.31 for full details). By docking against a set of NMR fibril structures, we intended to cover a wide range of possible conformations of Aβ fibrils. We experimentally tested only sites that showed recurring Aβ-inhibitor binding among the different fibril conformations. It is important to note that the NMR structure used in this study does not include the flexible N-terminus, so interactions with these residues were not considered in this study. Yet, the possible role of residues 1–8 in Aβ biological activity or in ligand binding cannot be ruled out. For example, N-terminal residues including H6 are involved in copper binding, which may be linked to Aβ induced oxidative stress.9
Another possible inadequacy of using an Aβ fibril NMR structure arises from the question: are we addressing the right target? Aβ can aggregate to form a variety of morphologies, which induce different levels of neurotoxicity. Although Aβ fibrils are toxic to cells, it is widely accepted that rather than the Aβ fibril, an aggregation intermediate often referred to as Aβ oligomer, is the species manifesting the highest biological activity and neurotoxicity,11, 45, 46 and is believed to be connected to the root cause of AD.47 Yet, the structure of the toxic Aβ oligomer has been much more challenging to elucidate because of its transient nature and difficulties associated with its isolation. Available structural information for the Aβ oligomer is not yet sufficient such that computational docking or molecular dynamic simulations tools could be used to elucidate ligand/toxicity inhibitor binding sites on the oligomer.
Although it is clear that docking of ligands to an Aβ oligomer is not yet possible, we believe that it is still of value to dock toxicity inhibitor ligands to an Aβ fibril as a means to explore residues on Aβ associated with biological activity. First, it appears that there are some structural similarities between Aβ intermediates that are associated with toxicity and Aβ fibrils. The toxic intermediate has β-sheet fibril-like secondary structure48 and is recognized by several fibril-specific antibodies.49 Although there may be differences in tertiary structure,48 backbone solvent accessibility,35, 29 or stability between the toxic intermediate and the fibril,46 it appears that there are at least some molecular level structural similarities between the toxic intermediate and the mature fibril. Second, Aβ fibrils are still toxic46 and have been shown to have a wide range of biological activities.46, 50 Thus, in the absence of a high resolution structure of an Aβ aggregation intermediate structure, we believe that the fibril structure could serve as a reasonable approximation for other β-sheet fibril-like aggregates. Although other Aβ aggregates such as oligomers may interact differently with the inhibitors we studied, our findings may provide a good starting point when more structural information becomes available.
Our docking results suggest that three of the four inhibitors; CR, Myr, and melatonin, although structurally different, all shared the best scored docking loci; at the β-turn between residues A21–A30, and at the nook located near the C-terminus, between amino acids V39–V40 of one fibril layer to the outer side of the β-turn of the other fibril layer, next to amino acids N27 and G29 (Fig. 2). Docking of nicotine did not as clearly converge to a specific binding site. Our experimental studies supported these findings. Binding of both CR and Myr were perturbed upon methylation of K16 and K28, whereas neither was affected when only K16 was ethylated (Fig. 5). Methylation of K28 does not change the charge of the lysine residue but could potentially disrupt hydrogen bonds with an inhibitor and sterically hinder inhibitor binding. In addition, the presence of inhibitors (Myr, curcumin, nicotine, and melatonin) attenuated the methylation of lysines on Aβ (Fig. 6). These inhibitors had no effect on a control protein, suggesting that the inhibitors altered the reactive environment or solvent accessibility of either K16 or K28 of Aβ fibrils. Based on the binding isotherms and the docking results, we presume that the inhibitors altered the environment of K28.
The N27P Aβ40 mutation had been shown previously to form fibrils similar to WT51 and was unlikely to perturb the β-turn.52, 53 However, the N27P mutation led to significant changes in binding capacity of fibrils for CR and Myr, consistent with docking results. It is notable, however, that while the binding capacity of the N27P fibrils for CR decreased, the Myr binding capacity increased. It is not clear why the replacement of N27 with a proline induced a different effect on CR and Myr. One possibility is that different molecular interactions are responsible for the binding of CR and Myr to Aβ fibrils at this position. There are many examples where point mutations in an enzyme or receptor led to altered selectivity of the protein for its substrate or ligand, respectively.54–57 It is possible that the asparagine to proline mutation altered the selectivity of this binding site on the Aβ fibril, decreasing its affinity and capacity for CR, while increasing its affinity/capacity for Myr. Although our data would suggest that there are at least two binding sites for CR and Myr on Aβ fibrils, the uncertainty in the data did not allow us to explore multiple binding site models.
Our CR docking and experimental data are in agreement with some previous proposals for CR binding to Aβ. Hydrophobic interactions were hypothesized to play a key role in the burial of the hydrophobic center of CR, whereas the negatively charged sulfonate groups were thought to interact with a positively charged amino acid.58, 59 Additionally, based on geometric constraints it has been proposed that CR is aligned parallel to the fibril axis and perpendicular to the β-strands, while interacting with four stacks of Aβ monomers.60 Such a binding mode was observed in all three docking sites of CR (Fig. 2). Other studies have suggested that CR may bind near a histidine amino acid,61 however our docking near H13–H14 was not reproducible. The existence of multiple binding sites of CR on Aβ has been proposed before.59, 62 Our data would suggest that there are at least two potential binding sites for CR on Aβ fibril, the two perturbed by the K28 modification and N27P mutation, and potentially a third binding site that gives rise to the shift in absorbance used for amyloid detection. For all fibril samples, the characteristic shift in CR absorbance with fibril binding was still observed (data not shown). To our knowledge, our results are the first reported evidence for the identity of the binding loci of Myr on Aβ fibrils. In addition, we believe this study represents the first report to identify the loci on the Aβ fibril near K28 and near N27/the C-terminus as important for potential inhibitor interactions.
The Aβ coordinates we used for docking and our experimental fibril preparation method are associated with specific fibril morphology, a striated fibril with twofold symmetry. Other fibril morphologies41, 63 may yield different results. Recent evidence suggests that Aβ fibril morphology in vivo is heterogeneous, with some fibrils sharing structural similarities to the fibrils used in this study.64 We believe the strategy outlined here will be useful in elucidating binding sites on Aβ fibrils of other morphologies, and other Aβ aggregate structures.
In summary, we have shown the power of computational docking at elucidating possible binding sites of small molecule toxicity inhibitors to Aβ. We have experimentally verified the binding of two structurally dissimilar toxicity inhibitors at two sites and demonstrated indirectly that another three toxicity inhibitors may bind at one of those sites. The identification of these sites now allows us the opportunity to explore the biological relevance of these common binding loci on Aβ fibrils. Furthermore, these results provide us with a target for in silico screening of libraries of compounds that bind to Aβ fibrils that could later be tested in vitro for Aβ toxicity prevention.
The program DOCK 632, 65 was used to predict inhibitors binding sites. The receptor in docking simulations was an NMR Aβ40 fibril structure made up of 12 Aβ (9–40) monomer units, where residues 1–8 were unstructured and therefore omitted (coordinates from solid state NMR, provided by R. Tycko by personal communication). The NMR structure includes two different categories of fibrils; stagger (+2) and stagger (−2).31 Ten structures of each category were used, representing different possible backbone and side-chain configurations. Hydrogens missing from Aβ were added using PDB2PQR66 using the Amber force field and assigned pH of 7.4 using PROPKA.67 Coordinates of inhibitors were obtained from the Zinc or ChemDB databases.68, 69 Aβ and inhibitor's charges were added using Antechamber70 embedded in Chimera.71 Binding sites were defined by stripping all receptor hydrogens, and calculating the accessible receptor surface using the program DMS embedded in DOCK. Sets of overlapping spheres were generated using SPHGEN (embedded in DOCK) to create a negative image of the receptor surface and to define potential binding sites. The number of spheres in a site did not exceed 200. Energy and contact grids were generated using the default settings within a box surrounding the binding site. Energy scores were calculated using the grid based score with a flexible ligand and rigid receptor using default settings except for the following: minimal anchor size was six atoms, maximum number of orientations was 3000, bump filter was 2 (2 events of 75% VdW radius overlap), internal ligand energy was included, and 1000 iterations were used during ligand minimization. Volume Tanimoto coefficients (T) were calculated using ROCS.72 Docking results were inspected visually using Chimera.71
All chemicals were purchased from Fisher (Pittsburgh, PA) or Sigma Aldrich (St. Louis, MO), unless otherwise specified.
Aβ fibril preparation
Aβ40 stock solutions were prepared by dissolving Aβ (Anaspec, San Jose, CA) at 10 mg/mL in ddH2O with 0.1% TFA for 45–60 min. The stock was diluted in PBS (138 mM NaCl, 10 mM Na2HPO4, 2.7 mM KCl, 1.8 mM KH2PO4) to a final concentration of 100 μM and was gently mixed using a rotator for 24 h at room temperature. We found that fibrils formed using this protocol were reproducible in terms of morphology, size, CR binding, and stability.46 As much as possible, a single lot of Aβ40 was used for all experiments. N27P Aβ40 mutants (Custom synthesized by Anaspec, at > 95% purity by RP-HPLC) were prepared similarly except that N27P Aβ was first aliquoted into single use sample sizes by dissolving in HFIP, aliquoting out into microcentrifuge tubes, then evaporating the HFIP from the Aβ under vacuum. Aliquots were then stored as HFIP-free films at −80°C until use. Total amount of protein in an Aβ sample was determined using the BCA assay according to the manufacturer's instructions (Pierce, Rockford, IL). Total amount of insoluble fibril in a sample was estimated by first separating soluble Aβ from insoluble Aβ fibril via centrifugation of samples at 23,100 × g for 5 min, then by quantifying the amount of protein in soluble and insoluble fractions using the BCA assay.
Modification of lysine amino acids
The modification of lysine residues was performed after fibril formation using reductive alkylation with formaldehyde or with acetaldehyde at 25°C. Sodium cyanoborohydride was used as a reducing agent. The reaction with formaldehyde was carried for 6 h at 88 μM Aβ, 880 μM formaldehyde and an excess of sodium cyanoborohydride (88 mM). The reaction modified the N-terminus, K16, and K28, as indicated by MALDI TOF MS of tryptic fragments of Aβ. The modification with acetaldehyde was carried at similar conditions with 83 μM Aβ fibrils, 1.66 mM acetaldehyde and an excess of sodium cyanoborohydride (83 mM), and was carried for 8 days. The reaction modified the N-terminus and K16, but did not modify K28, as indicated by MALDI TOF of tryptic fragments of the peptide. For the modification of Aβ fibril in the presence of inhibitors, 20 μM Aβ fibrils were incubated with Myr, nicotine, melatonin, or curcumin at 30 μM. Nicotine, melatonin, and curcumin stock solutions were prepared in ethanol at high concentrations (mM range), and were diluted to a final concentration in PBS. Myr was dissolved in DMSO and then diluted similarly in PBS. Control samples without an inhibitor, but with PBS containing same amount of ethanol/DMSO were used to confirm that residual solvent did not interfere with the reaction. After 30 min of incubation with Aβ, 200 μM formaldehyde and 20 mM sodium cyanoborohydride were added. The extent of the methylation reaction in the presence of inhibitors was examined using MALDI TOF every 90 min up to 6 h.
Before peptide digestion, the reductive alkylation was quenched by an addition of ammonium bicarbonate (83 mM) to avoid alkylation of the N-termini of Aβ fragments. Aβ fibrils were digested using immobilized trypsin (Pierce, Rockford, IL) according to the manufacture instructions. Briefly, immobilized trypsin was washed with 0.1M NH4HCO3 then suspended in 200 μL of 0.1M NH4HCO3. A total of 200 μL of 50 μM Aβ fibrils were then added to the trypsin bead suspension. The reaction was carried for 45 min at 37°C under gentle agitation. Trypsin beads were removed using centrifugation (23,100 × g, 5 min). No attempt was made to recover all the possible Aβ fragments associated with the trypsinized fibrils. Sample was analyzed using MALDI TOF.
Matrix-assisted laser desorption/ionization time of flight—Mass spectrometry (MALDI TOF)
MALDI TOF was performed using Bruker Daltonics Autoflex MALDI-TOF MS (Billerica, MA). Samples were desalted using 0.6 μL C18 resin ZipTip® (Millipore, Billerica, MA) according to the manufacture instructions. Briefly, the tips were washed in acetonitrile and then loaded with the sample. The sample was desalted by washing the pipette loaded tip in ddH2O. The Aβ sample was then redissolved and eluted off the C18 tip using 100% acetonitrile. We have found under these conditions, Aβ fibril samples unfold sufficiently for MALDI-TOF analysis.33 Saturated Sinapic acid in a solution of 2:1 (v/v) ddH2O:acetonitrile was used as the matrix. The mass spectrometer was run in positive ion mode. Laser power was optimized on a sample-by-sample basis and generally fell between 30 and 50% of the maximum power afforded by the instrument.
CR and Myr binding to Aβ fibrils
A stock solution of Myr was prepared in DMSO at 11.3 mM and stored at 4°C. CR was dissolved in 90% PBS and 10% ethanol to a concentration of 300 μM and then filtered using 0.22 μm filter to eliminate CR microaggregates. Final concentration of CR after filtration was determined by diluting CR to about 15 μM in 60/40 (v/v %) mixture of sodium phosphate (1 mM , pH 7.0) and ethanol, and the absorbance was measured at 505 nm using ε = 5.93 × 104 M−1 cm−1.24 CR and Myr were diluted in PBS to the final desired concentrations before each assay. Different concentrations of CR or Myr were incubated with 20 μM Aβ fibrils, either WT, mutated, or chemically modified, for 30 min at room temperature. Final ethanol/DMSO concentrations were less than 3%. The samples were then centrifuged at 23,100 × g for 5 min to separate the insoluble fibrils with the bound CR or Myr from unbound inhibitor. Unbound inhibitor concentrations were determined by measuring the absorbance of the supernatant (505 nm for CR, 324 nm with ε = 7000 M−1 cm−1 for Myr). Calibration curves of samples prepared identically but without Aβ were used to correct the extinction coefficient of CR and Myr for pH and other solvent effects. Absorbance was read using a Cary 50 WinVC Spectrophotometer (Varian, Palo Alto, CA). The concentration of inhibitor-Aβ fibrils complex was calculated by subtracting the unbound from the total inhibitor concentration. Chemical modification reagents did not interfere with the absorbance of CR or Myr. For each binding assay, fibril content in the sample was estimated using the BCA assay. Binding data were then normalized by total fibril content. A single site equilibrium binding isotherm shown in Eq. (1) was fit to the binding data using a nonlinear least squares regression in KaleidaGraph (Reading, PA), where B is the concentration of bound ligand, L is the concentration of soluble ligand, Bmax is the maximum binding capacity to fibrils, and Kd, the equilibrium dissociation constant.
Transmission electron microscopy
Aβ fibrils were prepared for Transmission electron microscopy on carbon-coated Formvar grids, 400 mesh (TedPella, Redding, CA). Grids were dipped in a sample of Aβ fibrils, washed in ddH2O, and dipped in staining solution of 1% ammonium molybdate in ddH2O. Grids were viewed using a Zeiss-1OCA microscope (Zeiss, Germany) equipped with an Olympus Morada digital camera (Melville, NY), using an accelerating voltage of 60 kV. Images were taken at the Keith R Porter Imaging Facility at UMBC.
The authors are grateful to Robert Tycko (NIH) for providing us with the fibril coordinates used for docking.