Site-specific Inhibitory Mechanism for Amyloid β42 Aggregation by Catechol-type Flavonoids Targeting the Lys Residues*

Background: The inhibitory mechanism of Aβ42 aggregation by flavonoid is fully unknown. Results: The oxidant enhanced the inhibitory activity of (+)-taxifolin against Aβ42 aggregation by forming Aβ42-taxifolin adducts between the Lys residues and oxidized (+)-taxifolin. Conclusion: The inhibitory activity of catechol-type flavonoids requires autoxidation to form an o-quinone to react with Lys. Significance: These may help design promising inhibitors against Aβ42 aggregation for Alzheimer disease therapy. The aggregation of the 42-residue amyloid β-protein (Aβ42) is involved in the pathogenesis of Alzheimer disease (AD). Numerous flavonoids exhibit inhibitory activity against Aβ42 aggregation, but their mechanism remains unclear in the molecular level. Here we propose the site-specific inhibitory mechanism of (+)-taxifolin, a catechol-type flavonoid, whose 3′,4′-dihydroxyl groups of the B-ring plays a critical role. Addition of sodium periodate, an oxidant, strengthened suppression of Aβ42 aggregation by (+)-taxifolin, whereas no inhibition was observed under anaerobic conditions, suggesting the inhibition to be associated with the oxidation to form o-quinone. Because formation of the Aβ42-taxifolin adduct was suggested by mass spectrometry, Aβ42 mutants substituted at Arg5, Lys16, and/or Lys28 with norleucine (Nle) were prepared to identify the residues involved in the conjugate formation. (+)-Taxifolin did not suppress the aggregation of Aβ42 mutants at Lys16 and/or Lys28 except for the mutant at Arg5. In addition, the aggregation of Aβ42 was inhibited by other catechol-type flavonoids, whereas that of K16Nle-Aβ42 was not. In contrast, some non-catechol-type flavonoids suppressed the aggregation of K16Nle-Aβ42 as well as Aβ42. Furthermore, interaction of (+)-taxifolin with the β-sheet region in Aβ42 was not observed using solid-state NMR unlike curcumin of the non-catechol-type. These results demonstrate that catechol-type flavonoids could specifically suppress Aβ42 aggregation by targeting Lys residues. Although the anti-AD activity of flavonoids has been ascribed to their antioxidative activity, the mechanism that the o-quinone reacts with Lys residues of Aβ42 might be more intrinsic. The Lys residues could be targets for Alzheimer disease therapy.

Alzheimer disease (AD) 3 is characterized by amyloid deposition in senile plaques mainly consisting of 40-and 42-mer amyloid ␤-proteins (A␤40, A␤42) (1,2). These proteins are generated from the amyloid precursor protein by ␤and ␥-secretases (amyloidogenic pathway). A␤ aggregates mainly through intermolecular ␤-sheet formation and shows neurotoxicity in vitro (3). A␤42 plays a more pivotal role in the pathogenesis of AD than A␤40 because of its higher aggregative ability and neurotoxicity (3). It has been well documented that soluble A␤ oligomeric assemblies rather than insoluble fibrils cause memory loss and neuronal death (4,5). Oxidative stress is one of the major contributing factors to neurodegenerative disease progression (6). A␤-induced toxicity has been correlated to oxidative damage through protein radicalization in vitro (7,8) and in vivo (9,10).
Researchers have reported protective effects of various polyphenols from green tea, turmeric, and red wine etc., against A␤ aggregation and neurotoxicity (11)(12)(13). Several compounds (e.g. (Ϫ)-epigallocatechin-3-gallate (EGCG), curcumin, and resveratorol) are in clinical or preclinical trials for AD treatment (14,15). However, the recent failures of some trials (16) motivated us to clarify the mechanism by which polyphenols inhibit the aggregation of A␤42 to develop promising leads for clinical use.
Concerning the molecular interaction of A␤ with flavonoids, a docking simulation by Keshet et al. (17) predicted the involvement of Lys 28 and the C-terminal region in the binding with myricetin. However, the precise mode of binding with flavonoids has scarcely been addressed, except for limited studies using NMR spectroscopy (curcumin (18), EGCG (19), and myricetin (20)), which suggested less-specific interaction with the ␤-sheet region in A␤.
Our group recently found that silymarin, seed extracts of Silybum marianum, attenuated AD-like pathologic features, such as senile plaques, neuroinflammation, behavioral dysfunction, and A␤ oligomer formation using a well established AD mouse model (J20) (21). We also isolated (ϩ)-taxifolin (22), a flavanonol that has a catechol moiety on the B-ring (Fig. 1A), as a component of the extracts that prevents A␤42 aggregation (23). "Aggregation" used in this article means the process of A␤42 monomer to form fibrils by way of oligomer and/or protofibril. This article describes a comprehensive study on the ability of (ϩ)-taxifolin to prevent aggregation and ␤-sheet formation of A␤42, along with the effects of various flavanonols and flavonols on the aggregation of A␤42 mutants substituted at Arg 5 , Lys 16 , and/or Lys 28 with norleucine (Nle). These results together with the results of liquid chromatography-mass spectrometry (LC-MS) led us to propose a site-specific inhibitory mechanism for A␤42 aggregation by catechol-type flavonoids, where adduct formation at Lys residues in A␤42 with the o-quione derived from flavonoids could be involved in the suppression of A␤42 aggregation.
The effect of the addition of NaIO 4 on Met 35 oxidation was estimated by HPLC on a Develosil ODS UG-5 column (6.0 mm inner diameter ϫ 100 mm; Nomura Chemical, Seto, Japan) under a gradient of 10 -50% CH 3 CN containing 0.1% NH 4 OH for 40 min after the centrifugation of the A␤42 solution at 20,130 ϫ g at 4°C (MX-300; TOMY, Tokyo, Japan) for 10 min.
The seeds of A␤42 were also prepared basically according to the protocol developed by Naiki and Gejyo (30). Briefly, after incubation of A␤42 (25 M) in PBS (pH 7.4) for 24 h at 37°C, the pellet obtained by centrifugation at 20,130 ϫ g at 4°C for 1 h was suspended by pipetting in PBS (pH 7.4) at a concentration of 1 mg/ml. The resultant solution was sonicated for 1 h in an ultrasonic device (MUS-20; EYELA, Tokyo, Japan), followed by dilution with PBS at 10 g/ml before use. Th-T relative fluorescence was expressed as a percentage of wild-type A␤42 alone, whose maximum value was taken as 100%.
Transmission Electron Microscopy (TEM)-The aggregates of A␤42 after a 48-h incubation were examined under a H-7650 electron microscope (Hitachi, Ibaraki, Japan). The experimental procedure was described elsewhere (31).
UV-visible Spectrometry-Oxidation of (ϩ)-taxifolin was monitored by UV spectroscopy (UV-2200A; Shimadzu, Kyoto, Japan). (ϩ)-Taxifolin (50 M) was incubated with A␤42 (25 M) in PBS (50 mM sodium phosphate and 100 mM NaCl, pH 7.4) at 37°C. The solution was then loaded into a 1-cm path length quartz cell, and UV spectra were recorded at 200 -500 nm. The sample was diluted three times with PBS because of its strong absorbance.
Circular Dichroism (CD) Spectrometry-The secondary structure of A␤42 was estimated by CD spectrometry (J-805; JASCO, Tokyo, Japan) using a 0.1-mm quartz cell (121.027-QS, 10 mm; JASCO), as described elsewhere (32). A␤42 (25 M) was incubated with or without (ϩ)-taxifolin (50 M) in PBS (50 mM sodium phosphate and 100 mM NaCl, pH 7.4) at 37°C. An aliquot was loaded into the quartz cell, and CD spectra were recorded at 190 -250 nm. Experiments under an anaerobic condition were performed as described before. The spectra of A␤42 mutants are shown after subtraction of the spectrum for vehicle alone, and those in the presence of (ϩ)-taxifolin are shown after subtraction of the spectrum for (ϩ)-taxifolin alone.
LC-MS Analysis-A␤42 solution (25 M) was incubated with 50 M (ϩ)-taxifolin in PBS (50 mM sodium phosphate and 100 mM NaCl, pH 7.4) in the presence of 100 M NaIO 4 at 37°C. After a 4-h incubation, the mixture was desalted and condensed twice by ZipTip C18 (Millipore, Bedford, MA). Five microliters of the solution was subjected to a liquid chromatography mass spectrometry ion trap time-of-flight (LCMS-IT-TOF; Shimadzu) through a YMC-Pack ODS-AQ column (6.0 mm inner diameter ϫ 100 mm; YMC) at 25°C under a gradient of 20 -60% CH 3 CN containing 0.1% formic acid for 30 min.
The A␤42 mutants were synthesized in a stepwise fashion on 0.1 mmol of preloaded Fmoc-L-Ala-PEG-PS resin using a Pioneer TM Peptide Synthesizer (Applied Biosystems, Foster City, CA) as reported previously (35). After the chain elongation was completed, the peptide-resin was treated with a mixture containing trifluoroacetic acid, m-cresol, thioanisol, and 1,2-ethanedithiol for final deprotection and cleavage from the resin. The crude peptide was precipitated by diethyl ether and purified by HPLC under an alkaline condition as described previously (31).
After lyophilization, we obtained the corresponding pure A␤42 peptide, the purity of which was confirmed by HPLC (Ͼ98%). The molecular weight of each A␤42 mutant was confirmed by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS, AXIMA-CFR; Shimadzu); R5Nle-A␤42 (m/z, calculated: 4472.11; found: 4472. 38  Solid-state NMR Analysis-A␤42 was labeled at Ala 2 ( 13 C 3 , 15 N), Ser 8 ( 13 C 3 , 15 N), Lys 16 ( 13 C 6 , 15 N 2 ), Val 18 ( 13 C 5 , 15 N), Phe 19 , and Phe 20 ( 13 C ␤ ). (ϩ)-Taxifolin was labeled with 13 C 6 on the B-ring as mentioned above (23,24). The labeled A␤42 (13 M) was incubated with 13 C 6 -(ϩ)-taxifolin (145 M) in PBS (50 mM sodium phosphate and 100 mM NaCl, pH 7.4) at 37°C. After 48 h of incubation, the solution was centrifuged (27,720 ϫ g, PRP-20 -2; Hitachi) for 15 min at 4°C, and then the precipitate was dried in vacuo to give the labeled A␤42 aggregate associated with 13 C 6 -(ϩ)-taxifolin (12 mg). The solid-state NMR experiments were performed at 14 T (600 MHz for 1 H) using a JEOL ECA-600 spectrometer and a custom-fabricated probe with a Chemagnetics 3.2-mm spinning system at a magic angle spinning frequency of 21 kHz at room temperature as reported previously (18). The 13 C chemical shifts were calibrated in ppm relative to tetramethylsilane by considering the 13 C chemical shift for methine 13 C of solid adamantine (29.5 ppm) as an external reference. The 13 C chemical shifts of labeled A␤42 and (ϩ)-taxifolin were assigned according to one-dimensional 13 C CP/magic angle spinning NMR spectra (supplemental Fig.  S2A). For the broadband 13 C-13 C correlation two-dimensional experiments, dipolar-assisted rotational resonance was used (36). Pulse sequence parameters of the NMR experiment were as follows: two pulse phase-modulated 1 H decoupling power ϭ 80 kHz, RAMP-CP contact time ϭ 1.2 ms, pulse delay ϭ 2 s, t 1 increment ϭ 23.7 s, t 1 points of two-dimensional ϭ 128 pt, and mixing time ( m ) ϭ 50 or 500 ms. The window function "HAMMING" was used in all two-dimensional FT spectra to minimize t 1 noise. As the two-dimensional FT dipolar-assisted rotational resonance spectra were difficult to analyze because of the t 1 noise (supplemental Fig. S2C), we applied covariance data processing to obtain a better representation of the twodimensional spectrum (supplemental Fig. S2B). After Fourier transformation along the t 2 dimension and phase correction, the resulting data matrix was used for covariance processing as previously reported (18,37). The covariance processing step was accelerated by singular value decomposition (38).
Statistical Analyses-All data are presented as the mean Ϯ S.E. and the differences were analyzed with an one-way analysis of variance followed by Bonferroni's test or unpaired Student's t test. These tests were implemented within GraphPad Prism software (version 5.0d). p values Ͻ 0.05 were considered significant.

Effects of Autoxidation of (ϩ)-Taxifolin on Its
Ability to Prevent the Aggregation of A␤42-We recently revealed that a catechol moiety on the B-ring of (ϩ)-taxifolin (Fig. 1A) played an important role on the inhibitory activity against A␤42 aggregation (23). A catechol moiety is easily oxidized to form an o-quinone (39). To investigate the contribution of autoxidation to the inhibitory ability, we examined the aggregative ability of A␤42 in the presence of (ϩ)-taxifolin treated with sodium periodate (NaIO 4 ), which is known as an oxidant of catechol (40). As shown in Fig. 1B, NaIO 4 extensively promoted the suppressive ability of (ϩ)-taxifolin compared with (ϩ)-taxifolin alone. These observations were also confirmed by the TEM experiment ( Fig. 2A). NaIO 4 treatment in the presence of (ϩ)taxifolin formed only shorter and thinner fibrils compared with (ϩ)-taxifolin alone. A␤42 formed the typical fibrils even in the presence of NaIO 4 alone, and almost no differences (e.g. length, thickness) were observed between morphology in the presence and absence of NaIO 4 ( Fig. 2A). NaIO 4 alone slightly affected the Th-T fluorescence of A␤42 aggregates (Fig. 1B) possibly because NaIO 4 can oxidize Met 35 in A␤42 to its sulfoxide, the formation of which was confirmed by HPLC (Fig. 1D) and MALDI-TOF-MS (A␤42-M35 ox ; m/z: calculated: 4531.14; found: 4531.55 [M ϩ H] ϩ ). This is in good agreement with a report that oxidation using hydrogen peroxide, a strong oxidant, reduced A␤42 aggregation (41). However, in the presence of both (ϩ)-taxifolin (50 M) and NaIO 4 (100 M), Met 35 was not oxidized by NaIO 4 ; this was confirmed by HPLC (Fig. 1D) and MALDI-TOF-MS (A␤42-M35 red ; m/z: calculated: 4515.14; found: 4516.26 [M ϩ H] ϩ ). This indicates that NaIO 4 preferred to oxidize (ϩ)-taxifolin more than the sulfur atom of the Met 35 of A␤42.
In addition, we tested whether treatment of NaIO 4 leads to the oxidation of Met 35 in the preformed A␤42 fibrils. The fibrils (about 28 g) treated with NaIO 4 for 4 h were dissolved in formic acid (10 l), and were sonicated for 1 h. After volatilization, the resultant pellets were resolved in 50% acetonitrile containing 0.1% trifluoroacetic acid, followed by subjection to MALDI-TOF-MS analysis. NaIO 4 did not oxidize Met 35 in the preformed A␤42 fibril (A␤42-M35 red ; m/z: calculated: 4515.14; found: 4515.12 [M ϩ H] ϩ ). Also in Th-T assay, A␤42 fibril was not disassembled by NaIO 4 (data not shown). This means that NaIO 4 could partially oxidize Met 35 in the monomeric A␤42, but not the fibrils.
To investigate the role of oxygen, suppression of A␤42 aggregation by (ϩ)-taxifolin was tested in vacuo. Notably, (ϩ)-taxifolin little suppressed the aggregation of A␤42 under an anaerobic condition (Fig. 1C). In TEM images, typical fibril formation was observed even in the presence of (ϩ)-taxifolin under the anaerobic condition ( Fig. 2A). Furthermore, A␤42 aggregated in the presence of (ϩ)-taxifolin and TCEP, a reductant (Fig. 1E). These results suggest the autoxidation of (ϩ)taxifolin is required for the inhibitory activity against A␤42 aggregation.  (n ϭ 8). Th-T relative fluorescence was expressed as a percentage of the fluorescence for wild-type A␤42 alone, whose maximum value was taken as 100%.
The mechanism of A␤42 fibril formation is well explained by a nucleation-dependent polymerization model mainly consisting of nucleation and extension phases (42,43). To determine which stage (nucleation phase or extension phase) was affected by (ϩ)-taxifolin, we examined the effect of (ϩ)-taxifolin on A␤42 aggregation in the presence of the fibril seed as a template, according to the protocol developed by Naiki and Gejyo (30). As shown in Fig. 1F, there was a nucleation phase (ϳ1 h) when A␤42 was incubated alone, whereas addition of the seeds skipped the nucleation phase, resulting in rapid formation of A␤42 fibrils. In the case of co-incubation of (ϩ)-taxifolin with the seeds, the nucleation phase of A␤42 did not drastically change, but fluorescence gradually decreased after incubation for 4 h, suggesting that (ϩ)-taxifolin could prevent the elongation phase (ϳ2 or 4 h) in A␤42 aggregation, rather than the nucleation phase (ϳ1 h) (Fig. 1B, F). Although the slight difference of the length of the elongation phase between Fig. 1, B and F, might be deduced from several factors, for example, outside temperature or batch (lot) of A␤42, an averaged time of 2-4 h for the elongation phase was observed in another independent experiment. Moreover, we have recently reported the ability of (ϩ)-taxifolin to destabilize the preformed A␤42 fibril (23). The disappearance of the nucleation phase in the presence of seed and NaIO 4 (Fig. 1F) implied the ability of oxidized taxilfolin to disassemble even the seed. Indeed, the fluorescence of preformed A␤42 fibrils immediately disappeared after addition of the (ϩ)-taxifolin treated with NaIO 4 (data not shown).
Next, we measured UV spectra of (ϩ)-taxifolin incubated with A␤42 to evaluate the effects of NaIO 4 or the anaerobic condition on the autoxidation of (ϩ)-taxifolin. When A␤42 was incubated with (ϩ)-taxifolin under air, the intensity of the peak at 260 and 400 nm gradually increased, and that of the peak at 320 nm decreased during 48 h of incubation (Fig. 2B). These spectral changes are characteristic of the oxidation of catecholtype flavonoids to form the o-quinone structure (44). The addition of NaIO 4 accelerated these UV changes (Fig. 2B). In contrast, there was almost no change in the UV spectra when (ϩ)-taxifolin and A␤42 were co-incubated in vacuo or with TCEP (Fig. 2B). These results indicate that the o-quinone formation in (ϩ)-taxifolin through autoxidation plays a critical role in the inhibition of A␤42 aggregation. The UV spectra of A␤42 alone remained almost constant during the incubation (data not shown), meaning that the spectra of A␤42 itself did not affect those of (ϩ)-taxifolin. Conversion to the o-quinone from (Ϯ)-taxifolin in the presence of NaIO 4 was also verified by reacting with o-phenylenediamine to yield phenazine (supplemental Scheme S1B), whose structure was confirmed by 1

H NMR and high resolution EI-MS.
Effects of Autoxidation of (ϩ)-Taxifolin on Its Ability to Inhibit Transformation of a Random Structure into a ␤-Sheet in A␤42-We investigated the effects of autoxidation of (ϩ)taxifolin on the secondary structure of A␤42 by using CD spectroscopy. Shown in Fig. 3A is the data for A␤42; the positive peak at 195 nm and negative peak at 215 nm drastically increased even after 4 h of incubation, and remained until 48 h of incubation, suggesting that a random structure transformed into a ␤-sheet in A␤42. On the other hand, (ϩ)-taxifolin strongly delayed the transformation of A␤42 (Fig. 3B). Furthermore, addition of NaIO 4 decelerated the transformation process during 0 -8 h (Fig. 3C).
We also measured the CD spectra under an anaerobic condition (Fig. 3, D and E). A spectrum related to ␤-sheet formation was found only after 24 h of incubation, but its peak intensity was weaker than that of A␤42 under air in Fig. 3A. Because radicalization of A␤42 induced by the reactive oxygen species is indispensable to its aggregation (8), these results seem to be reasonable. The transformation into a ␤-sheet was not suppressed either by (ϩ)-taxifolin in vacuo. The findings suggest that the effects of autoxidation of (ϩ)-taxifolin on its ability to inhibit A␤42 aggregation are closely associated with prevention of the transformation into a ␤-sheet.

LC-MS Analysis of A␤42 Treated with Oxidized Taxifolin-
The o-quinone of flavonoids can form covalent bonds with nucleophilic residues in proteins (e.g. Cys, Arg, and Lys) to modulate their activity (39,45). Because A␤42 has three basic amino acid residues (Arg 5 , Lys 16 , and Lys 28 ), we asked if these residues bound to oxidized taxifolin covalently. The o-quinone derived from (ϩ)-taxifolin can react with Lys or Arg residues in A␤42 through a Michael addition or Schiff base formation (Fig.  4A). We analyzed an A␤42 solution incubated with (ϩ)-taxifolin and NaIO 4 for 4 h using a highly sensitive ion trap-type LC-MS equipped with a TOF mass analyzer (LCMS-IT-TOF). As shown in Fig. 4B, LC-MS measurements gave the mass envelop at ϩ7, ϩ6, and ϩ5 charge distribution (deconvoluted mass: 4817.12, calculated: 4816.38), corresponding to the A␤42-oxidized taxifolin adduct resulted from Michael addition. These results imply that the basic amino acid residues of A␤42 might be involved in the covalent bonding with the oxidized taxifolin.
Inhibitory Effect of (ϩ)-Taxifolin on Aggregation of A␤42 Mutants Substituted at Arg 5 , Lys 16 , and/or Lys 28 -Although formation of Michael adducts between the o-quinone of (ϩ)taxifolin and the Lys residues of A␤42 was suggested in LC-MS (Fig. 4B) together with the verification of the o-quinone formation (supplemental Scheme S1B), an attempt to determine the Lys residues involved in adduct formation by LC-MS-MS analysis was disappointing, possibly because of the extremely low amount and/or instability of the adduct. To obtain further insight into the mechanism by which (ϩ)-taxifolin inhibits the aggregation of A␤42, we prepared five A␤42 mutants (R5Nle-, K16Nle-, K28Nle-, K16,K28(Nle) 2 -, and R5,K16,K28(Nle) 3 -A␤42), where the basic amino acid residues of A␤42 were substituted with Nle. The aggregative ability in the presence or absence of (ϩ)-taxifolin was also estimated (Fig. 5, A-E). These mutants retained substantial aggregative abilities to form fibrils (70 -80%) compared with wild-type A␤42 in Th-T test (Fig.  5F). (ϩ)-Taxifolin did not suppress the aggregative ability of K16Nle-A␤42 (Fig. 5B). K28Nle-A␤42 also aggregated in the presence of (ϩ)-taxifolin, although the intensity of the Th-T fluorescence was slightly decreased than for the K28Nle-A␤42 alone (Fig. 5C). Moreover, (ϩ)-taxifolin did not prevent the aggregation of K16,K28(Nle) 2 -A␤42 and R5,K16,K28(Nle) 3 -    (n ϭ 8). Th-T relative fluorescence was expressed as a percentage of the fluorescence for the A␤42 mutant alone, whose maximum value was taken as 100%. *, p Ͻ 0.05 compared with A␤42 mutant alone. The time points without asterisks means no significant difference between A␤42 mutant treated and untreated with (ϩ)-taxifolin. F, the comparison of aggregative ability of A␤42 mutants. Th-T relative fluorescence of each mutant after incubation for 24 h was expressed as a percentage of the fluorescence for wild-type A␤42 alone, whose maximum value was taken as 100%. The data are presented as the mean Ϯ S.E. (n ϭ 8). A␤42 (Fig. 5, D and E). On the other hand, (ϩ)-taxifolin largely suppressed the aggregation of R5NleL-A␤42 (Fig. 5A). These results indicate that lysine residues at positions 16 and 28 could be targets for the oxidized taxifolin to prevent aggregation of A␤42. More correctly, because the aggregative ability of K28Nle-A␤42 was slightly suppressed by (ϩ)-taxifolin compared with that of K16Nle-A␤42 (Fig. 5, B and C), Lys 16 would be a more specific target than Lys 28 in inhibition of A␤42 aggregation.
Inhibition of A␤42 Aggregation by Non-catechol-type Flavonoids-Myricetin, quercetin, morin, and kaempferol, which were previously reported to inhibit A␤42 aggregation, belong to the flavonols (43). Flavonols contain a double bond between C2 and C3 on the C-ring, whereas flavanonols like (ϩ)-taxifolin do not (Fig. 6A). We calculated IC 50 (Fig. 6A). Regarding the relevance of autoxidation to the inhibition of A␤42 aggregation, we measured the Th-T fluorescence of A␤42 treated with these three non-catechol-type flavonols under an anaerobic condition or in the presence of TCEP. These flavonols suppressed the aggregation of A␤42 even in vacuo (Fig. 6B). Moreover, addition of excess TCEP (A␤42:flavonols: TCEP ϭ 25:50:200 M) did not affect the suppressive ability of these flavonols (data not shown), indicating that the inhibition of non-catechol-type flavonols could not be ascribed to their autoxidation. To gain further insight into the mechanism by which flavonoids inhibit A␤42 aggregation by targeting the Lys residues, aggregation tests were carried out in the presence of catecholtype flavonoids (dihydromyricetin, (ϩ)-taxifolin, myricetin, or quercetin), or non-catechol-type flavonols (morin, kaempferol, or datiscetin) using K16Nle-and K16,K28(Nle) 2 -A␤42. We compared the aggregative ability of the A␤42 mutant (25 M) incubated with each flavonoid (50 M), the concentration of which was the maximal value to suppress the fluorescence of A␤42 under 50% by (ϩ)-taxifolin (data not shown). The catechol-type (ϩ)-taxifolin and quercetin did not suppress the aggregation of these A␤42 mutants. Although dihydromyricetin and myricetin with contiguous trihydroxyl groups significantly prevented the aggregation of K16Nle-A␤42, these flavonoids did not change the aggregative potency of K16,K28(Nle) 2 -A␤42 (Fig. 7A), implying that they could react with Lys 28 as well as Lys 16 because contiguous trihydroxyl groups might facilitate the autoxidation of the B-ring compared with (ϩ)-taxifolin and quercetin containing vicinal hydroxyl groups. Notably, in the case of non-catechol-type flavonols (morin, kaempferol, and datiscetin), there was little difference in the inhibitory activity among the wild-type, K16Nle-A␤42, and K16,K28(Nle) 2 -A␤42 (Fig. 7B). These results suggest the existence of another inhibitory mechanism for A␤42 aggregation by non-catechol-type flavonols other than the autoxidation followed by the Michael addition of Lys residues, as observed for (ϩ)-taxifolin.
Analysis of the Interaction of A␤42 Aggregates with (ϩ)-Taxifolin Using Solid-state NMR-Our recent study using a solidstate NMR showed that curcumin with an ␣,␤-unsaturated ketone interacted with the aromatic hydrophobic core (A␤17-21) due to its inherent hydrophobicity and planarity, resulting in the inhibition of A␤42 aggregation via intercalation (18). Curcumin was reported to interact with A␤40 fibrils through the planarity of the enol form of curcumin (46). Also given are the preferable detection of monomeric A␤42 in LC-MS analysis, similar analysis was performed to clarify the interaction between A␤42 and (ϩ)-taxifolin. (ϩ)-Taxifolin was labeled with 13 C 6 on the B-ring based on previous structure-activity relationship studies, in which the catechol moiety on the B-ring is critical in the inhibitory potential, and the methylation of the hydroxyl group at position 7 on the A-ring did not influence the aggregation of A␤42 (23). A␤42 was also 13 C-labeled site-specifically at Ala 2 , Ser 8 , Lys 16 , Val 18 , Phe 19 , and Phe 20 , in which only C ␤ was labeled in Phe 19 and Phe 20 to avoid overlapping the A␤42 and (ϩ)-taxifolin signals. For the broad band 13 C-13 C correlation two-dimensional experiments, dipolar-assisted rotational resonance was employed (34). As shown in supplemental Fig. S2B, the interaction peaks between A␤42 and (ϩ)taxifolin were as weak as noise signals despite the use of a 10-fold excess of (ϩ)-taxifolin (A␤42:(ϩ)-taxifolin ϭ 13:145 M), whereas a 5-fold excess was employed for curcumin (A␤42:curcumin ϭ 10:50 M) (18). Remarkably, the 13 C-13 C cross-peaks between Lys residues and the B-ring of (ϩ)-taxifolin was not observed significantly (supplemental Fig. S2B). More specifically, the interaction of (ϩ)-taxifolin with the aromatic hydrophobic core (A␤17-21), which was previously found in the curcumin case due to its inherent hydrophobicity and planarity (18) was not observed. These indicate that the inhibitory mechanism of A␤42 aggregation by (ϩ)-taxifolin (via covalent bonding) could be different from that of curcumin (via intercalation).
Because lack of the double bond at positions 2 and 3 of (ϩ)taxifolin could decrease its planarity, it might not be able to insert into the ␤-sheet region of the A␤42 aggregate (Lys 16

DISCUSSION
Thus far, the anti-AD activity of flavonoids has been believed to originate from their antioxidative activity and/or ␤-sheet recognition due to their hydrophobicity and planarity. However, these parameters are not necessarily accompanied by the ability to inhibit the aggregation of A␤42 and other amyloidogenic proteins (11). This background led us to reconsider whether the inhibitory activity can be simply explained by these "less specific" properties (antioxidation, hydrophobicity, and planarity) or not.
On the basis of the present results, we propose a site-specific mechanism whereby catechol-type flavonoids inhibit the aggregation of A␤42, in which a catechol structure could be autoxidized to form the o-quinone on the B-ring, followed by the formation of the o-quinone-A␤42 adduct targeting Lys residues at positions 16 and 28 of A␤42, but not be originated from the antioxidative activity (Fig. 8). This could provide unique opportunities to design potent inhibitors of A␤42 aggregation. On the other hand, the inhibitory ability of non-catechol-type flavonols containing a double bond between C2 and C3 on the C-ring (Fig. 6A) does not require the autoxidation. The data in Figs. 6 and 7 indicate that the interaction of non-catechol-type flavonols with A␤42 might be less effective than the conjugate addition of the Lys residues to the o-quinone moiety derived from autoxidation. These findings might explain in part the difference in the inhibitory ability between flavanonols and flavonols.
Our previous investigation using solid-state NMR together with systematic proline replacement identified a toxic conformer with a turn at positions 22 and 23 in A␤42 (47), and proposed that the residues at positions 15-21 and 24 -32 containing Lys 16 and Lys 28 are involved in the intermolecular ␤-sheet region, whereas the N-terminal 13 residues are not (48). A monoclonal antibody against the toxic turn at positions 22 and 23 detected A␤ oligomers in human AD brain (49) and induced pluripotent stem cells (50) as well as in AD mice (51,52). As mentioned above, an attempt to determine the Lys residues involved in adduct formation by LC-MS-MS analysis gave disappointing results, possibly because of the extremely low amount and/or instability of the adducts. Because the targeted Lys residues (Lys 16 and Lys 28 ), not the Arg residue (Arg 5 ), are incorporated in the intermolecular ␤-sheet region (Fig. 8), even a small amount of covalently bonded adducts at the Lys residues of A␤42 oligomers and/or protofibrils could inhibit the formation of A␤42 aggregates (fibrils) detected by the Th-T fluorescence.
Recently, Bitan's group reported that the Lys-specific synthetic compound (molecular tweezer, CLR01) prevented cytotoxicity and oligomerization of A␤42 through non-covalent interaction in vitro (53) and in vivo (54). Their subsequent study using A␤42 mutants substituted at Lys 16 or Lys 28 with Ala revealed a key role for Lys residues in A␤42-induced neurotoxicity rather than aggregation (55). These reports did not contradict the results that the aggregates of double and triple mutants (K16,K28(Nle) 2 -A␤42 and R5,K16,K28(Nle) 3 -A␤42) after 24 h of incubation were slightly less than that of wild-type A␤42 in this study (Fig. 5F).
R5Nle-A␤42 (ϳ4 h) and K28Nle-A␤42 (ϳ2 h) had the nucleation phase (Fig. 5, A and C), whereas three A␤42 mutants (K16Nle-A␤42, K16,K28(Nle) 2 -A␤42, and R5,K16,K28(Nle) 3 -A␤42) including the substitution of Lys 16 with Nle did not (Fig.  5, B, D, and E). These mean that the aggregative ability of K16Nle-A␤42 seems to be more potent than that of K28Nle-A␤42. Because Lys 16 is located at a hydrophobic cavity in the ␤-sheet region (56), the substitution with norleucine without an amino group could enhance the hydrophobic interaction in A␤42 aggregates, leading to passing the nucleation phase. In contrast, Lys 28 was involved in the formation of the salt bridge between Asp 23 and Lys 28 for A␤42 aggregation (56). These findings imply the different role of lysine resides at positions 16 and 28 in A␤42 aggregation, which might explain the difference of aggregative ability between K16Nle-and K28Nle-A␤42.
Notably, the nucleation phase of R5Nle-A␤42 (ϳ4 h) was partially longer than that of wild-type A␤42 (ϳ1 h) (Fig. 1, A, B, and F). Because flexibility of the N-terminal region has been thought to be essential for aggregation of A␤42 (56), the replacement of Arg 5 with norleucine might retard the nucleation phase in wild-type A␤42 aggregation by increasing the hydrophobic interaction. Moreover, given no nucleation phase of the three A␤42 mutants of Lys 16 , it is not surprising that (ϩ)-taxifolin did not largely alter the aggregation properties of these mutants, because (ϩ)-taxifolin could specifically target the elongation phase in wild-type A␤42 aggregation, rather than the nucleation phase.
LeVine et al. (57) suggested preventive effects on A␤42 aggregation by several dihydroxybenzoic acid isomers, in which 2,3-and 3,4-dihydroxy benzoic acids delayed the velocity of oligomer formation. Fisetin, a quercetin analog without the 5-OH group on the A-ring, also inhibited the aggregation of A␤42 (58). Ushikubo et al. (59) reported that the 7-OH group on the A-ring is not involved in the anti-aggregative ability of flavonols. These are consistent with our previous study on the structure-activity relationship of (ϩ)-taxifolin (23). On the other hand, lacmoid without a catechol moiety bound less specifically to A␤42 (60). These findings together with the present results strongly support that flavonoids with vicinal hydroxyl groups on the B-ring could be indispensable to bind covalently with A␤42 to suppress its aggregation. It is also reasonable that 3,4,5-trihydroxybenzoic acid, gallic acid (57), and 3Ј,4Ј,5Ј-trihydroxyflavone (59) as well as dihydromyricetin (Fig. 6A) suppressed the aggregation of A␤42. In particular, the inhibitory activities of dihydromyricetin, (ϩ)-taxifolin, myricetin, and quercetin were higher than those of morin, kampferol, and datiscetin (Fig. 6A), suggesting that nucleophilic addition to the o-quinone moiety by the Lys residues of A␤42 could contribute more significantly to the inhibition of A␤42 aggregation than the hydrophobic interaction. However, an additional role of the A-and C-rings of catechol-type flavonoids in the suppression of A␤42 aggregation is not negligible, because the inhibitory activity of catechol itself was low (61).
Fink and colleagues (45) previously proposed the contribution of interaction of Lys residues with the baicalein o-quinone to the inhibition of ␣-synuclein responsible for Parkinson disease. However, the underlying molecular mechanism cannot be fully explained by the oxidized baicalein because the aggregation of ␣-synuclein was inhibited by baicalein even under an anaerobic condition (45). Quite recently, the oxidation product of EGCG was in part involved in remodeling the preformed fibrils of A␤40 by EGCG (62).
To the best of our knowledge, this is a first report that dihydromyricetin and datiscetin as well as (ϩ)-taxifolin have antiaggregative activity against A␤42. We also demonstrated that (ϩ)-taxifolin could suppress the elongation phase in the aggregation of wild-type A␤42, rather than the nucleation phase. Seed extracts of S. marianum, known as silymarin, have long been used as an anti-hepatotoxic medicine without notable adverse effects (63), and in particular, are efficacious against damage induced by alcohol and disturbances in the function of the gastrointestinal tract (64). Booth and Deeds (65) showed that (ϩ)-taxifolin was not toxic when given long term to albino rats. Therefore, (ϩ)-taxifolin may be a worthy candidate for AD therapeutics. Although some polyphenols (naringenin (66) and curcumin (67)) were reported to pass through the bloodbrain barrier after oral administration, caution should be used because of the differences between animal and clinical conditions.