Structural Mechanism of the Interaction of Alzheimer Disease Aβ Fibrils with the Non-steroidal Anti-inflammatory Drug (NSAID) Sulindac Sulfide*

Background: The mechanism of interaction between small molecules and amyloid-β fibrils is unknown. Results: Molecular modeling on the basis of solid-state NMR reveals that sulindac sulfide intercalates between β-strands of amyloid-β fibrils. Conclusion: Sulindac sulfide interacts with amyloid-β fibrils in a specific manner and binds to hydrophobic cavities in the core of the fibrils. Significance: Unraveling how small molecules interfere with amyloidogenic deposits will assist structure-based drug design for neurodegenerative disorders. Alzheimer disease is the most severe neurodegenerative disease worldwide. In the past years, a plethora of small molecules interfering with amyloid-β (Aβ) aggregation has been reported. However, their mode of interaction with amyloid fibers is not understood. Non-steroidal anti-inflammatory drugs (NSAIDs) are known γ-secretase modulators; they influence Aβ populations. It has been suggested that NSAIDs are pleiotrophic and can interact with more than one pathomechanism. Here we present a magic angle spinning solid-state NMR study demonstrating that the NSAID sulindac sulfide interacts specifically with Alzheimer disease Aβ fibrils. We find that sulindac sulfide does not induce drastic architectural changes in the fibrillar structure but intercalates between the two β-strands of the amyloid fibril and binds to hydrophobic cavities, which are found consistently in all analyzed structures. The characteristic Asp23-Lys28 salt bridge is not affected upon interacting with sulindac sulfide. The primary binding site is located in the vicinity of residue Gly33, a residue involved in Met35 oxidation. The results presented here will assist the search for pharmacologically active molecules that can potentially be employed as lead structures to guide the design of small molecules for the treatment of Alzheimer disease.

The self-assembly of amyloidogenic proteins into fibrils and oligomers plays a pivotal role in various diseases (1). The deposition of fibrils formed by the amyloid-␤ peptide (A␤) 3 into plaques in brain tissue is a major pathological hallmark in the progression of neurodegeneration in Alzheimer disease. A␤ peptides are generated through sequential proteolytic cleavages of the amyloid precursor protein (APP) by the ␤and ␥-secretases (2,3). This results in the production of A␤ peptides of differing lengths (4), mainly A␤ 1-40 and A␤   (5), and shorter variants, such as A␤   (6). Soluble oligomers formed by A␤ 1-42 represent the toxic species responsible for the decline in cognitive function associated with neurodegeneration (7). Several studies have demonstrated that small molecules can interfere with the solubility of amyloid proteins and are, therefore, potential drug candidates (8,9). Chronic inflammation significantly enhances Alzheimer disease pathogenesis (10). Fibrillar ␤-amyloid deposits co-localize with numerous chronic inflammatory mediators and activated microglia in the brain (11). The relation to inflammation suggests that non-steroidal anti-inflammatory drugs (NSAIDs) might be beneficial for the treatment of Alzheimer disease. In fact, epidemiological studies demonstrate a link between the use of anti-inflammatory drugs and the prevalence of Alzheimer disease (12). This work aims to investigate the interaction mechanism of the NSAID sulindac sulfide (Fig. 1a) with Alzheimer peptide A␤ fibrils. In addition to sulindac sulfide (13), NSAIDs, including ibuprofen, indomethacin (13), and flurbiprofen (14), have been identified as ␥-secretase modulators. ␥-Secretase modulators interfere with APP processing and modify the relative A␤ 1-42 population. In particular, sulindac sulfide decreases the relative amount of the amyloid-prone A␤  , whereas the production of shorter, less amyloidogenic A␤ peptides is increased (13)(14)(15)(16).
Solution-state NMR structures of the APP-TM (transmembrane) dimer have been solved for the wild-type (17) and a familial mutant (18). NMR and EPR experiments have revealed a potential cholesterol-binding site within the C-terminal Cys 99 sequence and highlight the significance of the GXXXG segments for binding (19). It has been suggested that NSAIDs can interact with lipids to form phospholipid complexes (20 -22). This may provide a general mechanism for the interaction of APP with small molecules. Reports on the interaction between sulindac sulfide and the APP transmembrane sequence are, however, contradictory. Sulindac sulfide, among other ␥-secretase modulators, binds to the APP transmembrane sequence of the Cys 99 motif (23) and to Cys 100 dimers in the presence of SDS micelles (24). Bacterial reporter assays show that ␥-secretase modulators, including sulindac sulfide, bind to the GXXXG dimerization motif and, thereby, attenuate the dimerization of the APP transmembrane sequence (25), a process necessary for proteolytic cleavage (26). However, colloidal aggregation of sulindac sulfide in aqueous solutions can potentially induce nonspecific binding (27,28). Contradicting data have been reported for the influence of sulindac sulfide on the A␤ peptide itself (29,30).
So far, it is not understood how sulindac sulfide interacts with amyloids. NMR is a suitable technique to study A␤-ligand interactions for various A␤ aggregation states (31). Solutionstate NMR can be employed to study interactions of A␤ monomers and small molecule (32) or peptide inhibitors (33), nanoparticles (34), and various others. Besides monomers and fibrils, oligomeric intermediates formed by A␤ in solution constitute potential drug targets (35,36). However, oligomeric intermediates and insoluble fibrils are not detectable by solution-state NMR because their lines are broadened beyond detection. Solid-state NMR spectroscopy is a powerful tool that allows the study of A␤-small molecule interactions at atomic resolution. In the past, this technique has been applied successfully for the characterization of the interaction between A␤ and the polyphenol epigallocatechin gallate (37), curcumin (38,39), and catechol-type flavonoids (40) and to study the interface of Congo red and amyloids formed by the prion domain of the HET-s protein (41). In this work, we investigate the interaction between sulindac sulfide and A␤ fibrils using solid-state NMR spectroscopy. On the basis of the gathered NMR data, we employ docking to derive a model for the intercalation of sulindac sulfide with A␤ fibrils.

A␤ Expression and Purification and Sample Preparation-
The uniformly 15 N or 15 N-13 C-labeled A␤ 1-40 peptide was recombinantly expressed in Escherichia coli inclusion bodies and purified via reverse-phase chromatography as described previously (42). The construct contains an N-terminal methionine but shows the same biochemical properties as the wildtype peptide (43). To obtain monomeric A␤ in solution, the lyophilized peptide was initially dissolved in 10 mM NaOH, sonicated for 10 min in an ultrasonic bath, and centrifuged for 10 min at 14,800 rpm to remove potential nucleation seeds. The solution was then diluted in 2ϫ buffer (100 mM sodium phosphate and 100 mM NaCl buffer (pH 7.3)) to yield the respective A␤ concentration.
Preparation of NSAIDs-Stock solutions of sulindac sulfide were prepared in dimethyl-sulfoxide (DMSO), and the respective amount of NSAID stock was added to A␤. The final concentration of DMSO in aqueous solutions was 1%.
Solid-state NMR Sample Preparation-A␤ fibrils were obtained according to a protocol described previously (44). Briefly, monomeric A␤ at a concentration of 50 M was seeded with sonicated fibrils (10% w/w) and incubated under agitation until the completion of fibrillation. This step was repeated for 11 generations. For solid-state NMR measurements, the last generation was allowed to fibrilize for 2 days before sulindac sulfide was added to the sample. Approximately 10 mg of A␤ fibrils was incubated with a 5-fold molar excess of sulindac sulfide (250 M, 1% DMSO). After mixing, the sample was kept quiescent for 1 h at room temperature. As a reference, A␤ fibrils were incubated under the same conditions with 1% DMSO. The sulindac sulfide-incubated fibrils were sedimented into a 3.2-mm rotor. The reference fibrils were packed into a 4.0-mm rotor.
Solid-state NMR Measurements-13 C-detected assignment experiments were carried out using a Bruker Avance wide-bore spectrometer operating at a 1 H Larmor frequency of 700 MHz (16.5 tesla). The spectrometer was equipped with a triple-resonance MAS probe ( 1 H, 13 C, 15 N). All measurements were recorded at a MAS rotation frequency of 17 kHz for 3.2-mm rotors or 13 kHz for 4.0-mm rotors and a temperature of 12°C. In all experiments, 1 H-13 C magnetization transfer was achieved through cross-polarization. Two-dimensional spectra were recorded using proton-driven spin diffusion (45, 46) for 13 C-13 C magnetization transfer with a mixing time of 200 ms or transferred echo double-resonance (TEDOR) (47, 48) for 13 C-15 N magnetization transfer. The three-dimensional NCACX and NCOCX (49) experiments were recorded employing TEDOR for 13 C-15 N magnetization transfer and dipolar assisted rotational resonance (50) for 13 C-13 C mixing.
To identify 13 C atoms in A␤ that are located in the vicinity of the 19 F atom of sulindac sulfide, we recorded 13 C-19 F rotational echo double resonance (REDOR) experiments (supplemental Fig. S8  a) (51). These were recorded on a Bruker Avance III wide-bore spectrometer operating at a 1 H Larmor frequency of 600 MHz (14 tesla). The spectrometer was equipped with a triple-resonance cross-polarization MAS probe ( 1 H, 13 C, 19 F). All measurements were recorded at a MAS rotation frequency of 14.6 kHz and a temperature of 12°C. In the REDOR experiments, 13 C-19 F dipolar dephasing was preceded by a 13 C-13 C protondriven spin diffusion mixing step. REDOR dephasing times were set to 1.1, 2.2, and 4.4 ms. REDOR spectra recorded with and without dephasing pulses were subtracted to identify fluorine-coupled carbons. In Fig. 2a, REDOR spectra are represented in one-dimensional mode. No 13 C-13 C cross peaks could be observed in the three-dimensional REDOR experiments because the sensitivity was too low. To detect the characteristic salt bridge between the carboxyl of Asp 23 and the amine of Lys 28 of mature A␤ fibrils, TEDOR experiments (supplemental Fig. S8 b) were carried out on the basis of the pulse sequence described by Jaroniec et al. (52). All spectra were recorded on a Bruker Avance III narrow-bore spectrometer operating at a 1 H Larmor frequency of 750 MHz (17.63 tesla) equipped with a triple-resonance MAS probe ( 1 H, 13 C, 15 N) at a MAS rotation frequency of 11 kHz and 4°C. In the experiment, the TEDOR mixing time was set to 7.27 and 15.72 ms.
Transmission Electron Microscopy (TEM)-Images were recorded on an EM 900 from Carl Zeiss SMT. Samples were stained with 4% uranyl acetate solution on formvar/carboncoated grids.
Molecular Modeling-The 2-fold symmetric (PDB code 2LMN) (53) and the 3-fold symmetric (PDB code 2LMP) (54) A␤ 1-40 fibril NMR structural models were employed in molecular modeling studies. To identify internal cavities, each model of the A␤ fibril was analyzed separately. For the calculation of internal cavities, van der Waals volumes were derived using the Voronoi cell method (55,56). Atomic volumes are calculated on the basis of the allocation of space among atoms using hyperbolic surfaces (applying a cubic lattice of 0.1-Å grid width) and applying the ProtOr (57) atom radius set for protein atoms that were determined analytically from reference structures. Internal cavities are determined analytically from each model structure by a Delaunay triangulation (58). An internal cavity is defined as the buried space within a structure that is big enough to accommodate at least a 1.4-Å radius water-sized probe. The size of a cavity is estimated from the average distance of the cavity center to the neighboring atoms and depicted by the radius of a sphere. The polarity of the cavities is assessed with the program DOWSER (59), which calculates potential positions for water not resolved by the original structure determination approach. This program assesses the hydrophilicity of protein cavities by determining the interaction energy between a water molecule and its surrounding atoms. Water molecules with interaction energies of less than Ϫ10 kcal/mol are considered "low-energy water molecules" and selected for output. Internal cavities harboring such an internal water molecule are denoted as polar (blue) and the remaining cavities as hydrophobic (gray) ( Fig. 3

and supplemental Figs. S4 and S5).
For induced fit docking (60,61), the region between the two most hydrophobic cavity clusters (termed #1-#5) near Ile 32 (#2) and Val 36 (#3) from the first NMR model was targeted. Docking was performed using the Schrödinger Maestro suite (62) following a standard protocol. Subsequently, a grid with autoassigned box size was generated. The resulting box is centered near Leu 34 for the 3-fold symmetric structure and Ile 32 and Val 36 for the 2-fold symmetric structure. Induced fit docking was done with the program Glide with single precision and with no constraints applied (63). To simulate receptor flexibility, side chain orientations were optimized to take into account ligand binding. At last, a Glide redocking was applied for the refined structure. Docking poses are ranked and filtered by means of the Glide score. This score assesses the binding probability and accounts for steric clashes, van der Waals and coulomb energy, lipophilicity, H-bonding, and bond rotation ability, among others. Only the energetically most favorable docking poses (with theoretical energies Ͻ30 kcal/mol higher than the best pose) are considered for further analysis.

Results
To probe the interaction between A␤ fibrils and sulindac sulfide (Fig. 1a), we titrated sulindac sulfide to preformed fibrils. Fibrillar A␤ is observed by TEM in the presence of the NSAID (Fig. 1b). The resulting 13 C-15 N (Fig. 1c) and 13 C-13 C (supplemental Fig. S1) correlation spectra show well dispersed peaks. The three-dimensional NCACX and NCOCX experiments allowed sequential assignment of resonances for residues Gln 15 -Val 40 for both samples (representative strip plots are shown in supplemental Fig. S2; all assigned resonances are listed in supplemental Table S1). We detected only one set of resonances for both samples, indicating that the fibrils exist in one conformation.
Sulindac sulfide has no significant effect on the fibrillar structure of A␤ because the spectra of both samples are relatively similar ( Fig. 1c and supplemental Fig. S1). However, small but defined chemical shift perturbations (CSPs) are observed in the presence of sulindac sulfide, indicating specific interactions of the NSAID with the fibrils. This is remarkable because a nonquantitative and nonspecific binding of sulindac sulfide to A␤ fibrils would result in peak-splitting and line-broadening. Fig.  1d shows CSPs, ⌬␦ (parts per million), upon addition of sulindac sulfide. We observe changes in chemical shift in particular for side chain resonances of Lys 16 , Val 18 -C␤, Phe 19 -C␤, Phe 20 -C␤, Asn 27 -C␥, and Met 35 -C␤ as well as for the backbone resonances of Phe 19 , Phe 20 , Ala 21 and Gly 33 . CSPs reflect ligand binding but could as well be a consequence of local or global structural rearrangements. To unambiguously probe ligand binding, we recorded 13 C- 19 F REDOR experiments employing the NMR-active properties of the 19 F atom of sulindac sulfide. In general, only aliphatic resonances of A␤ could be detected. For comparison, a one-dimensional 13 C spectrum containing assignments for all aliphatic resonances is represented (Fig. 2a). The strongest dephasing effects are observed for the C␥ resonances of Val 18 or Val 39 . However, because of spectral overlap, these two peaks cannot be discriminated. Smaller signal attenuations are observed for C␥ of Val 24 and Val 36 , C␤s of Ala 21 or Ala 30 (overlap) as well as for C␥2 and C␦1 of Ile 32 at longer mixing times. These resonances exhibit the most severe dephasing effects. The respective residues must therefore be located close to the fluorine atom of sulindac sulfide.
To gain further information on the potential effects of sulindac sulfide on the A␤ fibril structure, we investigated the effect on the salt bridge, which is typically formed between the side chains of residues Asp 23 and Lys 28 (53,54). From sequential assignments, the chemical shift of the carboxylic group of Asp 23 in the presence of sulindac sulfide was assigned to 177.8 ppm and to 178.0 ppm for the reference fibrils. On the basis of onedimensional 15 N spectra, the chemical shift of Lys 28 -N was found to be 33.9 ppm and 34.3 ppm for the two preparations, respectively. In both samples, a cross-peak between Lys 28 -N and Asp 23 -C␥ was detected, implying the presence of a salt bridge in both cases (Fig. 2b).
In the following, the CSPs in the two-dimensional 13 C-13 C proton-driven spin diffusion and 13 C-15 N TEDOR as well as the 13 C-19 F REDOR contacts were used as restraints to derive a model for sulindac sulfide in complex with A␤ 1-40 fibrils. We used both a 2-fold (53) and 3-fold symmetric (54) A␤ 1-40 fibril NMR structure as reference structures for modeling and docking experiments because they show the highest correlation with our chemical shifts (supplemental Fig. S3) compared with all structures and models analyzed (44, 53, 54, 64 -68). The NMR structures contained 10 models each. The architecture of the A␤ 1-40 fibrils obviously differs comparing the 2-fold and 3-fold-symmetric structures. In each structure, A␤ adopts a   Table S2. Recoupled and reference spectra were subtracted to identify affected 13 C resonances (red). For reference, a one-dimensional 13 C experiment without 19 F recoupling, including all aliphatic resonances, is shown (gray). b, analysis of the salt bridge involving residues Asp 23 and Lys 28 . 13 C-15 N TEDOR spectra show cross-peaks between Lys 28 -N and Asp 23 -C␥ in the presence (red) and absence of sulindac sulfide (black). A TEDOR mixing time of 15.72 ms is employed. One-dimensional traces extracted at the N chemical shifts are superimposed with the one-dimensional 13 C reference spectra of the respective sample (gray).
␤-strandOturnO␤-strand fold. Fibrils are stabilized by hydrogen bonds connecting individual ␤-strands along the fibril axis. The 3-fold symmetric fibril structure contains three stacks of A␤ molecules in a triangular form. The 2-fold symmetric fibril structure is built from two antiparallel stacks of A␤ molecules. To detect potential binding sites for the hydrophobic sulindac sulfide, we performed a packing density analysis for each A␤ structure. In fact, all models revealed large packing defects. On average, we found eight cavities per stack of four A␤ molecules in both structures (supplemental Figs. S4 and S5), clustering in five different regions within one stack (clusters #1-#5) (Fig. 3). To dissect polar from unpolar cavities, we applied the tool DOWSER (59) to calculate positions of internal water molecules that are not resolved in the NMR structures. We found that the most hydrophobic cavities in each structure clustered in the rigid core region, with calculated water occupancies of Ͻ10% in cluster #2 and Ͻ30% in cluster #3 that are located on both sides of Phe-19 (supplemental Table S3). The 3-fold symmetric structure seemed to be more tightly packed around clusters #2 and #3, featuring fewer cavities in comparison with the 2-fold symmetric structure. At the same time, other clusters in the 3-fold symmetric structure seemed to contain a larger numbers of cavities. Clusters of polar cavities were found in the flexible turn region between residues Glu 22 and Ile 31 (cluster #1) and toward the termini of the peptide sequence (cluster #4). The two structures differed in their interface architecture between the termini to the adjacent loop region. Although, in the 3-fold symmetric structure, cavities of cluster #4 were stabilized through interactions with the loop, in the 2-fold symmetric structure, fewer polar cavities were found because the structure opened and cavities could become exposed. Cluster #5 (around M35) with polar cavities was located at the intersheet/contact region between termini and the loop region (3-fold symmetric structure) or at the interface of the antiparallel ␤-sheets (2-fold symmetric structure). In the intersheet region, where the contact interface between 3-fold and 2-fold symmetric structure was apparently different, the amount of cavities per stack (of four A␤ molecules) and their polarity did not differ. The amount of cavities and the respective water contents per cluster #1-#5 are shown in supplemental Table S3.
Clusters #2 and #3 contain large hydrophobic cavities, which could, in principle, bury a sulindac sulfide molecule. The exact size of the individual cavities depends on the rotameric state of Phe 19 . The 3-fold symmetric and the 2-fold symmetric structures differ slightly in the position of Phe 19 with respect to the registry of the second ␤-sheet containing residues Ile 32 to Val 36 (supplemental Fig. S6). In the 3-fold symmetric structure, Phe 19 is oriented more toward Ile 32 , whereas, in the 2-fold symmetric structure, Phe 19 is pointing toward Leu 34 and is, therefore, located more central in the core region. Surprisingly, the size, polarity, and distribution of the detected cavities is very similar in all analyzed structural models even though there are slight differences between the two models.
To test whether the hydrophobic cavities in clusters #2 and #3 are indeed suitable binding sites for sulindac sulfide, we applied an induced fit docking approach where the sulindac sulfide and side chains of A␤ were kept flexible. The size and accessibility of these cavities depends on the rotameric state of Phe 19 . In fact, docking into cluster #3 of the 3-fold symmetric structure allows identification of two scenarios that fit to the distance restraints obtained from the NMR analysis. In particular, we find that the distance of the sulindac sulfide 19 F atom to the methyl groups of Ile 32 , Leu 34 , or Val 36 is smaller than 6 Å. In pose 1, sulindac sulfide lies in the Gly 33 groove and is orientated along the fibril axes, being close to Ile 32 and Leu 34 . The aromatic side chains of Phe 19 are directed toward Ala 30 (Fig. 4a, top  panel). In pose 2, sulindac sulfide has rotated into a position with its conjugated ring system parallel to the ␤-sheets and orthogonal to Phe 19 so that the 19 F atom approaches Leu 34 and Val 36 . (Fig. 4a, bottom panel). Both poses suggest an aromatic stacking interaction between the conjugated ring system of sulindac sulfide and the Phe 19 side chain. Similar interactions are obtained from docking to cluster #2 of the 2-fold symmetric structure (Fig. 4b). In pose 1, sulindac sulfide is orientated along the fibril axes, with its conjugated ring system parallel to Phe 19 , analogous to pose 1 in the 3-fold symmetric structure. Again, the 19 F atom is positioned in the Gly 33 groove close to Ile 32 and Leu 34 . In contrast to the 3-fold symmetric structure, Phe 19 is positioned here on the other side of sulindac sulfide, close to Leu 34 , thereby shielding Val 36 so that sulindac sulfide cannot change into a position to contact Val 36 (Fig. 4b, top panel). An additional docking to cluster #3 yields poses 2 and 3, where the normal of the conjugated ring system of sulindac sulfide is orientated perpendicularly to the fibril axes. Thereby, sulindac sulfide extends into the terminal region (#4). The conjugated ring system is positioned parallel to the ␤-sheets and orthogonal to Phe 19 . Sulindac sulfide is either orientated in a way that the 19 F atom faces Phe 19 , Leu 34 , and Val 36 (Fig. 4b, center panel) or that the 19 F atom is oriented toward the fibril exterior, contacting Val 36 and Val 39 (Fig. 4b, bottom panel).

Discussion
Sulindac sulfide-incubated A␤ fibrils are highly similar to control fibrils, implying that sulindac sulfide has no significant effect on fibril structure. Both the NMR chemical shift patterns ( Fig. 1c and supplemental Fig. S1) and the morphology in TEM images (Fig. 1b) are maintained. Analysis of 13 C chemical shifts predicts ␤-strands as the main secondary structural element in both fibril preparations (supplemental Fig. S7). In addition, chemical shift analysis by torsion angle likeliness obtained from shift and sequence similaritiesϩ (TALOSϩ) (69) yields secondary structural propensities that predict the presence of two ␤-strands typically observed in specific regions within A␤ 1-40 fibrils (44, 53, 54, 67, 70) and oligomers (71,72). Furthermore, we find, by TEDOR experiments, that the Asp 23 -Lys 28 salt bridge remains intact in the presence of sulindac sulfide (Fig.  2b). Even though the overall fibrillar character is maintained, we observe defined chemical shift changes, potentially indicating local conformational changes (Fig. 1d). Previous reports have stated that small molecules such as curcumin (39) are able to disrupt the characteristic salt bridge in A␤ 1-42 fibrils. Judging from EM data, curcumin has a more drastic effect on the general fibril architecture compared with sulindac sulfide because it destroys A␤ 1-42 fibrils (39). We note that only one set of resonances is observed for the sulindac sulfide-incubated A␤ fibrils. No resonance splitting is observed for cross-peaks. Rather, resonances move to new positions, indicating that each peptide in a fibril interacts specifically with one or more NSAID molecules.
Upon addition of sulindac sulfide, CSPs are observed in particular for hydrophobic residues such as Val 18 , Phe 19 and Phe 20 , Gly 33 , and Met 35 but also for the polar side chain of Lys 16 . The most dramatic CSPs are detected for Lys 16 and Gly 33 (Fig. 1d). 13  To identify trends and avoid bias, we use both the 2-fold symmetric (PDB code 2LMN) (53) and the 3-fold symmetric (PDB code 2LMP) (54) A␤ 1-40 fibrils as reference structures for modeling and docking experiments. The polymorphism of these structures may differ from the fibrils investigated in this study. However, they show the highest similarities of all three wild-type A␤ 1-40 fibril structures currently available (53,54,64). Within the two A␤ fibril structural ensembles, several hydrophobic cavities are detected that are large enough to potentially host a sulindac sulfide molecule. We did not find significant deviations in the distribution, size, and polarity of the cavities between the analyzed structural ensembles and models. Therefore, changes in fibril polymorphism do not significantly affect the distribution of hydrophobic patches in amyloid fibrils. We conclude that the two employed structures provide a suitable basis for docking experiments, which is also supported by experimental NMR data, although high-resolution structures of a better defined polymorph will allow more accurate docking in the future.
An induced fit docking reveals that the NSAID can interact with fibrils in three different ways, and all models fulfill the NMR restraints. The apparently ambiguous REDOR restraints suggest that more than one sulindac sulfide molecule might be involved in binding. In the docking poses that are in best agreement with the NMR restraints, sulindac sulfide intercalates between the two ␤-strands of the A␤ fibril, with the long axis of the molecule either parallel (Fig. 4, a and b, pose 1) or perpendicular (Fig. 4, a, pose 2, and b, poses 2 and 3) to the fibril axis. Therefore, residue Phe 19 seems to play a crucial role in sulindac sulfide binding because its rotameric state has an influence on the size and shape of cavities in clusters #2 and #3. Furthermore, the aromatic side chain is involved in stacking with the conjugated ring system of the NSAID.
Pose 1 of the docking approach suggests that sulindac sulfide fits into the groove formed by Gly 33 (Fig. 4, a and b, top panels). Therefore, the large CSPs may be attributed to Gly 33 backbone atoms experiencing a change in chemical environment or even undergoing conformational changes to accommodate the NSAID. In theoretical studies, this glycine residue has been suggested to be involved in oxidation of A␤ 1-42 because of its close proximity to the side chain of Met 35 (73,74). Furthermore, Gly 33 has been reported as the key amino acid for A␤ toxicity and to be responsible for driving A␤ into neurotoxic conformations (75). Both residues, Gly 33 and Met 35 , have been hypothesized to stabilize reactive oxygen species (76,77). Sulindac sulfide might, therefore, act by binding to the hydrophobic pocket in the vicinity of Gly 33 , and preventing oxidation of the A␤ peptide. Oxidation of A␤ 1-42 reduces fibril assembly and aggregation because of the increased polarity introduced by the methionine sulfoxide (78). In accordance, we find that the A␤-sulindac sulfide complex exists in a stable fibrillar state.
CSPs for Lys 16 may be explained by a recent docking study involving sulindac sulfide and A␤ fibrils that suggested that sulindac sulfide may bind weakly to a shallow, solvent-exposed pocket in the vicinity of Lys 16 and Val 18 and does not interfere with fibrillation (30). This binding mode may, in addition, account for the large REDOR signal detected for the overlapping resonances Val 18 /Val 39 C␥ (Fig. 2a). We cannot exclude docking of sulindac sulfide to the fibril surface because our REDOR and CSP data are also in agreement with the blind docking model proposed by Yesuvadian et al. (30).
Sulindac sulfide has been reported to form colloidal aggregates above a critical micelle concentration in solution (28). This phenomenon is commonly observed for hydrophobic compounds (79,80). The size of small-molecule colloidal particles is typically on the order of 50 -600 nm (28). These colloids bind unfolded proteins in a promiscuous manner (79,80) and have been shown to lead to precipitation and inhibition of protein function (81)(82)(83). However, the effect of sulindac sulfide on A␤ fibril chemical shifts reported here indicates not a promiscuous but a specific interaction. Aggregates of the compound are observed in TEM images (Fig. 1b), implying the presence of colloids in solution. This is expected because the concentration used (250 M) lies above the critical micelle concentration of 50 -100 M (28). To account for the experimental single set of resonances, we must assume that individual NSAID molecules dissociate from the colloidal complexes and interact specifically with A␤ fibrils.
Binding of sulindac sulfide in different cavities for different A␤ peptides would result in splitting of the resonances and line-broadening. We observe, however, narrow lines, indicating that the small molecule must either bind simultaneously to different cavities or exchange between these cavities. Relaxation experiments will be carried out in the future to differentiate between these two scenarios.
In conclusion, we suggest that the NSAID sulindac sulfide is able to interact with A␤ fibrils in a rather specific manner. We find that several cavities can accommodate a sulindac sulfide molecule. This is supported by defined CSPs and 13 C-19 F REDOR contacts. Sulindac sulfide does not induce drastic architectural changes to the non-toxic fibrillar structure, as indicated by NMR and EM. In addition, the characteristic Asp 23 -Lys 28 salt bridge and the length and positioning of the ␤-strands are not affected. Molecular modeling suggests that sulindac sulfide intercalates between the two ␤-strands at, presumably, more than one position. The presented data contribute to the elucidation of the mechanism by which small molecules bind insoluble amyloids. This understanding is crucial for the design of pharmacologically relevant molecules that interfere with A␤ species and that might, in the future, be employed for the treatment of Alzheimer disease.
Author Contributions-E. P. expressed and purified isotopically labelled peptide samples, designed and performed NMR experiments, and carried out the data analysis. R. S. implemented the solidstate NMR experiments. J. M. L. d. A. recorded the initial experiments and assisted with data analysis. G. A. O. carried out the fluorine REDOR experiments. G. M. performed biochemical experiments. H. J. B. and P. W. H. performed and analyzed the molecular modeling and docking. All authors discussed the results. B. R. and E. P. conceived the project and wrote the paper with input from all authors.