Curcumin Alters the Salt Bridge-containing Turn Region in Amyloid β(1–42) Aggregates*

Background: Curcumin reduces the risk of Alzheimer disease via an unknown mechanism. Results: Curcumin-incubated Aβ42 aggregates retain the hairpin architecture but have disruptions in the turn region (surprising similarity with Zn2+ incubation). Conclusion: Salt bridge-containing turn region is a major determinant of morphology and toxicity. Significance: Identification of crucial structural changes provides a checkpoint for developing effective AD therapeutics. Amyloid β (Aβ) fibrillar deposits in the brain are a hallmark of Alzheimer disease (AD). Curcumin, a common ingredient of Asian spices, is known to disrupt Aβ fibril formation and to reduce AD pathology in mouse models. Understanding the structural changes induced by curcumin can potentially lead to AD pharmaceutical agents with inherent bio-compatibility. Here, we use solid-state NMR spectroscopy to investigate the structural modifications of amyloid β(1–42) (Aβ42) aggregates induced by curcumin. We find that curcumin induces major structural changes in the Asp-23–Lys-28 salt bridge region and near the C terminus. Electron microscopy shows that the Aβ42 fibrils are disrupted by curcumin. Surprisingly, some of these alterations are similar to those reported for Zn2+ ions, another agent known to disrupt the fibrils and alter Aβ42 toxicity. Our results suggest the existence of a structurally related family of quasi-fibrillar conformers of Aβ42, which is stabilized both by curcumin and by Zn2+.

Curcumin, a phenolic compound from the Asian spice turmeric (Curcuma longa), is of great interest for its plausible role in countering neurodegenerative diseases like Parkinson disease (1)(2)(3) and Alzheimer disease (AD) 3 (4,5). In vitro studies have shown that curcumin can retard the process of amyloid ␤ (A␤) aggregation (5,6), which is supposed to be the initiator of AD. It disrupts the formation of the long straight A␤ fibrils (5,7) and reduces A␤ toxicity (8,9). Moreover, it can also disrupt the preformed A␤ fibrils (6,7). The reduction of toxicity is possibly related to these effects. Alternative modes of action have also been suggested; it has been shown to inhibit the process that produces A␤ from the amyloid precursor protein (10). In addition, curcumin in general can reduce the concentration of the reactive oxygen species (11), which plays a role in neurodegeneration. Whatever is its main mode of action, information about the conformational changes of A␤ induced by curcumin would be valuable in understanding its role in AD.
There are several modified curcumin analogues that have been developed that show even more potent anti-AD activity in animal models (12), and pharmaceutical development will benefit from this understanding. Curcumin is known to bind to ␤-sheet-rich A␤ species like protofibrils and fibrils and not to unstructured monomers (7). Theoretical studies have suggested that curcumin can interact with A␤ oligomers to disrupt the ␤-sheet content and possibly alter the conformation of several regions of the A␤ molecule (13). Experimental studies using curcumin analogues with different linker lengths have shown that there is an optimum linker length that interacts with A␤, suggesting that the interaction of A␤ with curcumin is site-specific (14,15). However, it is still not known whether curcumin induces specific structural changes in A␤ aggregates.
Like curcumin, Zn 2ϩ ions are also known to alter the toxicity of A␤ and disrupt the A␤ fibrils (16 -18). It is known that Zn 2ϩ ions preferentially precipitate the amyloid oligomers and can stoichiometrically bind to A␤ (19 -21). Zn 2ϩ is thought to bind to the N-terminal region, and this is most likely mediated by His-6, His-13, and His-14 (22)(23)(24)(25). Using ssNMR studies, we have previously shown that Zn 2ϩ binding disrupts the salt bridge between Asp-23 and Lys-28 (26). Although Zn 2ϩ and curcumin have a completely different chemical nature, there are interesting parallels between the effects induced by them. It is therefore worthwhile to compare their structural effects. If some of the structural changes are common, then those changes may be key to the disruption of the fibrils and possibly also to the modulation of toxicity.
In this study we examine the structural changes induced by curcumin on A␤(1-42) (A␤ 42 ) aggregates with ssNMR, probing two different samples that have two different sets of isotopically labeled amino acids (a total of 15 amino acids are labeled). We have previously observed the effects of Zn 2ϩ on the same sets of amino acids (26), which gives us an opportunity for a direct atom-by-atom comparison of the effects of these two agents. We also verify the effects of curcumin on the fibrillar morphology of the aggregates using transmission electron microscopy. Significant unexpected similarities emerge between the effects of Zn 2ϩ and curcumin, pointing toward a class of structural changes that may be the key to understanding how A␤ fibrillar architecture gets disrupted by small molecules, possibly altering its toxicity.

EXPERIMENTAL PROCEDURES
A␤ 42 Synthesis, Purification, and Sample Preparation-Synthesis and purification procedures of two different A␤ 42 peptide specimens with different 13 C and 15 N isotopic labeling schemes (P 1 and P 2 ) are described elsewhere (26). Purified A␤ 42 peptides were initially dissolved in pH 11.0 water (adjusted by NaOH) to prepare 2 mM stock solutions. A 4 mM stock solution of curcumin (Sigma) was also prepared by dissolving in pH 11.0 water. To grow A␤ 42 aggregates in the presence of curcumin, 1.5 ml of A␤ 42 stock (2 mM) and 150 l of freshly prepared curcumin stock (4 mM) were mixed in pH 11.0 water and immediately diluted with 5.85 ml of HEPES buffer (containing 20 mM HEPES, 146 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl 2 ⅐2H 2 O, and 0.8 mM MgSO 4 ⅐7H 2 O) at pH 7.4 such that the total volume was 7.5 ml. This final solution containing 400 M A␤ 42 and 80 M curcumin was incubated at room temperature (Ϸ24°C) for 4 days with mild rotation (10 rpm). This resulted in the formation of aggregates. The rest of the materials used in the experiments was same as described in Ref. 26.
Solid-state NMR-For ssNMR measurements, solutions containing peptide aggregates were subjected to centrifugation (16,000 ϫ g) for 1 h. The pellets thus collected were washed with de-ionized water twice by resuspending them in water and ultracentrifuging for 1 h each time. The final pellets so obtained were rapidly frozen using liquid nitrogen and then lyophilized. The powdered samples were rehydrated (33 weight %) by adding Ϸ0.5 l of de-ionized water per mg of dry peptide aggregates. The hydrated sample was then packed in a 2.5-mm magic angle spinning rotor such that it contains ϳ10 mg of sample.
All the ssNMR measurements were performed using a 700 MHz Bruker AVIII NMR spectrometer at a magic angle spinning frequency of 15 kHz using a 2.5-mm triple resonance magic angle spinning probe. Cross-polarization to 13 C and 15 N from 1 H was implemented using a linear ramped radio frequency field (27) centered around 65 kHz on the 1 H channel and with a 55-kHz field on the 13 C/ 15 N channel with a contact time between 2.5 and 4.0 ms. 1 H dipolar decoupling was accomplished using the swept frequency two-pulse phase modulation ( ϭ 10°) decoupling scheme (28) with a field strength of 85 kHz. Two-dimensional 13 C-13 C through-space NMR spectra were recorded using second-order dipolar recoupling schemes of PARIS-xy (m ϭ 1, n ϭ 2) (29) and PDSD (30) and mixing periods of 100 and 1000 ms, respectively. 1 H irradiation of 15 kHz was used while applying PARIS-xy (m ϭ 1, n ϭ 2). A total of 200 points was acquired in the indirect dimension with a dwell time of 12.5 s amounting to an acquisition time of 2.5 ms. The number of scans per free induction decay and interscan delay was fixed to 256 and 2 s, and 512 and 1.5 s in case of PARIS-xy (m ϭ 1, n ϭ 2) and PDSD, respectively. Frequencyselective 13 C-15 N rotational-echo double-resonance (REDOR) experiments were recorded using the pulse sequence designed by Jaroniec et al. (31). Rotor-synchronized trains of phase-alternated pulses following xy-8 scheme (32) were applied on the 15 N channel. 466.7 s long frequency-selective Gaussian pulses were applied at the 13 C␥ and 15 N frequencies of Asp-23 and Lys-28, respectively, in the middle of the REDOR-dephasing period ( mix ). Reference spectra (S 0 ) were recorded after removing the frequency-selective refocusing pulse on 15 N channel.
The results shown in this study for amyloid aggregates of only A␤ 42 and those grown in the presence of 400 M Zn 2ϩ ions (A␤ 42 -Zn) are based on higher quality ssNMR spectra compared with those reported in our previous article (26). This is a result of better purification and degree of hydration (50 weight %), but they do not change any of the conclusions of the previous study.
NMR Data Analysis-All one-dimensional spectra were processed and analyzed using TopSpin 3.1. All two-dimensional spectra were processed with TopSpin 3.1 and analyzed using CcpNmr Analysis 2.2.2. The data were zero-filled in the t 1 and t 2 dimensions to 512 and 4096 points, respectively. A mixed sine/cosine ( ϭ /3 at t ϭ 0) apodization function was used in each dimension. All the spectra were externally referenced to tetramethylsilane in methanol (33). TALOSϩ (34) was used to obtain predicted and backbone torsion angles, based on 13 C chemical shifts of ␣, ␤, and carbonyl carbons.
Calculating Average Chemical Shift Change (⌬␦ )-13 C chemical shifts of isotopically labeled amino acids in A␤ 42 fibrils (26) were subtracted from that of A␤ 42 -Zn (26) and A␤ 42 -Cur aggregates. These chemical shift differences for ␣, ␤, and carbonyl carbons (the backbone) and the remaining carbons (the side chain) were then averaged over the number of carbons constituting the backbone and side chain, respectively, to obtain an average chemical shift change (⌬␦).
Electron Microscopy-10-l solutions of A␤ 42 (400 M) and A␤ 42 -Cur (400 M A␤ 42 ϩ 80 M curcumin), aggregated for 4 days, were placed on carbon-coated 100-mesh copper grids for 2 min, followed by blotting by a filter paper. The extra salt on the grids was removed by four cycles of mild washing with double-distilled water. Then a drop of 0.1% uranyl acetate was added to each grid and left for 5 min for staining. After removing the extra uranyl acetate solution by a filter paper, the grids were dried under an infrared lamp. The samples were examined with a transmission electron microscope (LIBRA 120, EFTEM, Carl Zeiss, Germany). The width analysis of fibrils was performed with ImageJ (open source software, rsbweb.nih.gov).
Cell Culture-Primary cortical neurons were cultured from pregnant female Wistar rats obtained from the Institute Animal Facility. All animal handling procedures were approved by the animal ethics committee of the Institute. Neuronal cultures were obtained from the cortex of 17-day-old embryos. Cells were grown in Neurobasal media supplemented with 2% B-27 supplement, 0.5% penicillin/streptomycin, and 0.25% L-glutamine. Cell culture media and chemicals were obtained from Invitrogen.
Cell Viability Assay-Primary cortical neurons grown in 96-well plates were treated with A␤ 42 (40 or 400 M) on day 5. After 48 h, the cells were assessed for viability. The cells were treated with 0.01 mg/ml propidium iodide (Molecular Probes) for 10 min, washed with phosphate-buffered saline, and imaged with an epifluorescence microscope (Zeiss Axiovert 200, Germany) using a 40ϫ objective. Propidium iodide binds to DNA but is cell-impermeable and hence can penetrate cells with damaged membrane. This gives the dead cell count. The number of live cells (propidium iodide-positive cells were subtracted from total cells counted from transmission images) expressed as a percentage of total cell count gives the percentage cell viability. Cell viability of a control set of cells treated with vehicle is normalized to 100%, and all the results are expressed with respect to that. To investigate the effectiveness of curcumin in reducing the A␤-induced toxicity, curcumin was added at 1:5 molar ratio (i.e. 8 M curcumin to 40 M A␤ 42 and 80 M curcumin to 400 M A␤ 42 ). Each experiment was performed on six different wells.

RESULTS AND DISCUSSION
We tested the effects of curcumin on A␤ 42 -induced toxicity on rat primary cortical neuronal cultures. Fig. 1, A-D, shows the superimposed transmission and propidium iodide-stained (marks dead cells, green) images of neurons for different treatments. The results of this study are summarized in Fig. 1E. The cells exposed to 40 M A␤ 42 for 48 h showed a reduced viability of 62 Ϯ 4% compared with vehicle-treated control cells. However, co-incubation with 8 M curcumin led to an improved viability of 85 Ϯ 2%. This shows that a sub-stoichiometric amount of curcumin (at 1:5 molar ratio to A␤ 42 ) is able to significantly ameliorate the toxic effects A␤ 42 . This neuro-protective role of curcumin was more evident at a higher concentration of A␤ 42 . Although the cells treated with only 400 M A␤ 42 for 48 h showed very low viability (4 Ϯ 1%), the addition of 80 M curcumin improved the cell viability to 74 Ϯ 3%, manifesting a very pronounced effect. Based on these results, we decided to probe the structural alterations of A␤ 42 aggregates caused by curcumin, starting from a solution containing 80 M curcumin and 400 M A␤ 42 .
A␤ 42 fibrils are characterized by a cross-␤ architecture in which each A␤ 42 monomer adopts a hairpin structure (although not a ␤-hairpin). The two largely hydrophobic ␤-strands are connected by a loop, and there is an overhanging unstructured N terminus (35,36). These ␤-strands polymerize in a parallel in-register orientation forming intermolecular ␤-sheets running perpendicular to the direction of the fibril axis. The presence of an intra and intermolecular contact between the side chains of amino acids Phe-19 -Leu-34 and Gln-15-Gly-37 (36) and an intermolecular salt bridge between the COO Ϫ and NH 3 ϩ groups present in the side chains of Asp-23 and Lys-28 (35) are some other prominent features in the available structural models of A␤ 42 aggregates. In our previous study (26), we investigated the effect of Zn 2ϩ on the molecular structure of A␤ 42 fibrils with ssNMR using two A␤ 42 peptides, namely P 1 and P 2 , containing 13 C-and 15 N-labeled amino acids at specific positions along the peptide sequence (26). P 1 has uniformly 13 C-and 15 N-labeled Gln-15, Phe-19, Ala-30, Leu-34, Val-36, and Gly-38 and uniformly 15 N-labeled His-13, whereas P 2 has uniformly 13 C-and 15 N-labeled Ser-8, Val-12, Phe-20, Asp-23, Lys-28, Met-35, and Ile-41 and uniformly 15 N-labeled His-14. We use the same peptides in this work, which allow a direct comparison.
Chemical shifts of 13 C atoms in isotopically enriched amino acids were obtained from two-dimensional 13 C-13 C correlation spectra of A␤ 42 aggregates grown in the presence of curcumin (termed as A␤ 42 -Cur) recorded using PARIS-xy (m ϭ 1) (n ϭ 2) (29) recoupling scheme with a mixing time of 100 ms. Selected regions of these spectra containing aliphatic-carbonyl and aliphatic-aliphatic cross-peaks are shown in Fig. 2, A and B. The corresponding spectra for A␤ 42 are shown in Fig. 2, C and D. The chemical shift values are listed in Table 1 along with the values obtained for A␤ 42 -Zn aggregates. For A␤ 42 -Cur aggregates, multiple sets of chemical shifts were observed for amino acids Val-12 (two), Asp-23 (three), Lys-28 (two), Ala-30 (three), Val-36 (three), and Gly-38 (two), where each set represents a distinct structural conformation. The rest of the amino acids yield unique sets and hence adopt a unique structural conformation in A␤ 42 -Cur aggregates. Similar structural heterogeneity prevails in the molecular structure of A␤ 42 aggregates (grown in absence of curcumin) with these same amino acids exhibiting multiple structural conformations. Even the number of multiple conformations is same in both cases, except for Val-12 and Lys-28, both of which exhibit an extra conformation in case of A␤ 42 aggregates. Fig. 3, A and B, shows the average chemical shift changes (⌬ ␦ ) incurred by the backbone and sidechain carbons of isotopically labeled amino acids in A␤ 42 when aggregated in presence of Zn 2ϩ ions and curcumin, respectively (see under "Experimental Procedures" for calculation of ⌬ ␦ values). Only those ⌬ ␦ values that are Ն0.5 ppm and at least three times larger than the error associated with them are considered as significant, i.e. represent a significant structural change. These are highlighted in gray in Fig. 3, A and B. These changes are generally associated with local structural changes, which can be caused by direct or indirect interaction of the external ligand. These values indicate that both Zn 2ϩ ions and curcumin cause structural perturbations of the Val-12 backbone. In addition to these, curcumin also causes chemical shift changes in the Asp-23 and Leu-34 side chains (Fig. 3B).
More differences are observed when other types of spectral features, like changes in the peak intensities in both one-dimensional and two-dimensional spectra, are considered. These changes are shown in Fig. 3, C-G, which shows selective regions of one-dimensional and two-dimensional spectra of A␤ 42 (black), A␤ 42 -Zn (blue), and A␤ 42 -Cur (orange) aggregates. Amino acids showing other type of spectral changes are Ser-8, His-13, and His-14 in the case of Zn 2ϩ ions and Val-36 and Gly-38 in the case of curcumin. Our previous study has shown that the presence of Zn 2ϩ ions imparts increased structural order to the otherwise less ordered side chains of His-13 and His-14 (26). This observation was based on observation of broad peaks around 174 and 208 ppm in 15 N one-dimensional spectrum of A␤ 42 -Zn aggregates (Fig. 3, C and D, blue). These broad peaks arise from the imidazole ring nitrogens, ⑀ 2 and ␦ 1 , present in the side chain of histidine. The absence of signal in the case of A␤ 42 aggregates grown in the absence of Zn 2ϩ ions ( Fig. 3, C and D, black) is attributed to the structural heterogeneity associated with these side chains, which become structurally more ordered due to Zn 2ϩ binding. However, the presence of curcumin does not impart any structural order to these flexible His side chains as is evident from the absence of NMR peaks in the region of interest (Fig. 3, C and D, orange). Fig. 3E highlights the C ␤ -CЈ correlations in Ser-8 as observed in the PARIS-xy (m ϭ 1) (n ϭ 2) spectrum recorded with a mixing time of 100 ms. Clearly, a much stronger correlation is observed in the case of A␤ 42 -Zn aggregates (Fig. 3E, blue) compared with the A␤ 42 (black) and A␤ 42 -Cur (orange) aggregates. The intensity of such throughspace 13 C-13 C correlations produced by second-order recoupling schemes like PARIS-xy (m ϭ 1) (n ϭ 2) depends strongly on the molecular properties of the system as well as the experimental parameters used (37). Because all spectra were recorded under similar experimental conditions, this change in intensity of C ␤ -CЈ correlations in Ser-8 must have its origin in changes occurring at the molecular level. It most likely reflects that the Ser-8 backbone acquires more structural order when A␤ 42 is aggregated in presence of Zn 2ϩ ions but not curcumin. Fig. 3F highlights the C ␣ -C ␤ cross-peaks observed for three conformers of Val-36 as observed in PARIS-xy (m ϭ 1) (n ϭ 2) spectrum using a mixing time of 100 ms. In the case of A␤ 42 (Fig. 3F, black) and A␤ 42 -Zn (blue) aggregates, the Val-36 conformer yields the most intense cross-peak (cross signs) followed by Val-36Ј (plus signs) and Val-36Љ (circled cross signs), and hence it is the most populated conformer among the three. However, in the case of A␤ 42 -Cur aggregates (Fig. 3F, orange), cross-peaks of both Val-36 and Val-36Ј conformers are nearly equally intense and that of Val-36Љ was not observed at all (only observed in PDSD spectrum recorded with long mixing time of 1 s). It is thus clear that curcumin disturbs the population equilibrium of various structural conformers of Val-36. Similar observations exist in the case of Gly-38, where population distribution between the two conformers of Gly-38 (Fig. 3G), Gly-38 (cross signs) and Gly-38Ј (plus signs), gets perturbed only in the presence of curcumin but not in the presence of Zn 2ϩ ions.
In our previous study, it was shown that Zn 2ϩ ions disrupt the Asp-23-Lys-28 salt bridge in A␤ 42 aggregates by stabilizing only the non-salt bridge-forming conformations of Asp-23 and

C NMR chemical shifts of uniformly 13 C-and 15 N-labeled amino acids in A␤ 42 -Cur, A␤ 42 , and A␤ 42 -Zn aggregates
All chemical shifts are relative to tetramethylsilane with uncertainties of approximately Ϯ0.1 ppm. Values followed by ( †) have an uncertainty Ն Ϯ0.3 ppm associated with them. The values in the parentheses and braces parenthesis are chemical shift values obtained for A␤ 42 and A␤ 42 -Zn aggregates, respectively. TALOSϩ was used to predict backbone torsional angle for each structural conformer in A␤ 42 -Cur, A␤ 42 , and A␤ 42 -Zn aggregates. Only "GOOD" predictions (as explained in Ref. 34) are reported here.  (26). Interestingly, presence of curcumin also causes major population redistribution between different conformers of Asp-23 and Lys-28, leading to the near disappearance of the Asp-23 and Lys-28Љ conformers as assigned in A␤ 42 aggregates (see Table 1). This effect is clearly shown in Fig. 4, A and B, which shows an overlay of selected regions of the one-dimensional 13 C and 15 N spectra of A␤ 42 (black) and A␤ 42 -Cur (orange) aggregates, respectively. It is further exemplified in Fig. 4, C and D, which shows the connectivity between the aliphatic carbons (␤, ␥, ␦, and ⑀) of the three conformers of Lys-28 in A␤ 42 , and the only two conformers of Lys-28 in A␤ 42 -Cur aggregates, respectively. These are two-dimensional 13 C-13 C PARIS-xy spectra recorded with a mixing time of 100 ms. However, because of the extremely fast decaying nature of C ␥ resonance associated with the curcumin-stabilized conformers in A␤ 42 -Cur aggregates (see Fig. 5B, orange), a REDOR dephasing curve lasting long enough to comment on the presence or absence of the salt bridge could not be derived in this case. A comparative REDOR dephasing signal for A␤ 42 aggregates is shown in Fig. 5A. All the observations made so far are summarized in Fig. 4, E and F, where amino acids having significant chemical shift changes are highlighted using dark cyan filled circles and those exhibiting other type of spectral changes are highlighted in circles with red outline. The effect of curcumin and Zn 2ϩ on the molecular structure of the ␤-sheet regions of A␤ 42 is therefore somewhat complementary, with curcumin targeting amino acids mainly in the C terminus and Zn 2ϩ ions mainly in the N terminus. However, both of them affect the salt bridge-forming amino acids Asp-23 and Lys-28 present in the loop region connecting the two ␤-sheets.
A recent ssNMR study by Masuda et al. (38) has shown that curcumin interacts with the residues from 17 to 21, as well as with residue 12 present in the N-terminal ␤-sheet. Masuda et al. (38) postulate that curcumin targets the C-terminal ␤-sheet in a nonspecific fashion. Our results substantiate this speculation with three amino acids in this region, namely Leu-34, Val-36, and Gly-38, undergoing either structural reorganization or population redistribution between different conformers in the presence of curcumin. Regarding the N terminus, our results are partially in agreement with their findings. We also found that Val-12 was affected by curcumin. However, Phe-19 and Phe-20, the only isotopically labeled amino acids in the 17-21 region, remain unaffected by the presence of curcumin. This discrepancy most likely has its origin in different way these studies have been performed. Masuda et al. (38) incubated curcumin with preformed fibrils, whereas we precipitated A␤ 42 from the solution in the presence of curcumin. In a nut shell, curcumin does target the ␤-sheet-rich regions of A␤ 42 , but these are not the only regions affected by it, as shown by our results.
Secondary chemical shifts (⌬␦) of A␤ 42 and A␤ 42 -Cur aggregates, defined as the difference between the observed and random coil chemical shifts, were calculated for C ␣ , C ␤ , and carbonyl carbon atoms and are shown in Fig. 6, A and B, respectively. A positive value for C ␤ carbon and negative values for C ␣ and carbonyl carbon atoms is indicative of the amino acid being part of a ␤-sheet region. Most of the amino acids follow these criteria, thus indicating dominance of the ␤-sheet structural elements in both A␤ 42 and A␤ 42 -Cur aggregates. With the current isotopic labeling schemes, it seems that these regions run roughly between amino acids Val-12 and Phe-20 (N-terminal ␤-sheet) and Ala-30 and Ile-41 (C-terminal ␤-sheet). Even though multiple conformations are observed for some amino acids in this region, all of them seem to adopt ␤-sheet secondary structure except Gly-38. The presence of curcumin increases the population of the non-␤-sheet conformer (Gly-38Ј) of Gly-38. Among the remaining amino acids, Ser-8 clearly seems to be part of the unstructured N terminus in both A␤ 42 and A␤ 42 -Cur aggregates. The salt bridge-forming conformer of Asp-23, denoted D23 in figures, adopts a totally random structure. Its other two conformers in A␤ 42 aggregates also adopt non-␤-sheet conformation. However, in case of A␤ 42 -Cur aggregates, two out of its three conformers (Asp-23Ј and Asp-23Љ) seem to adopt the ␤-sheet structure. Similarly, in the case of Lys-28, the disappearing conformer, denoted K28Љ in figures, adopts a structural conformation that is significantly different from the remaining two conformers (K28 and K28Ј). The structural conformation of the other two conformers of Lys-28 is not completely like ␤-sheet, but this conformation is more or less retained in the presence of curcumin. It seems that in A␤ 42 aggregates, both Asp-23 and Lys-28 adopt a secondary structure that is partially ␤-sheet in nature, and hence, they are most likely part of the turn region connecting the N-and C-terminal ␤-sheets. The presence of curcumin results is an increase in the ␤-sheet content of both Asp-23 and Lys-28 in A␤ 42 -Cur aggregates. However, as both of them still contain conformers that are not strictly ␤-sheet-like, they might still be part of the turn region. To summarize, even though presence of curcumin causes local disturbances in the molecular structure, A␤ 42 -Cur aggregates still retain the basic secondary structural motifs of A␤ 42 aggregates.
It was seen that despite some major structural changes, A␤ 42 aggregates grown in the presence of Zn 2ϩ ions also retain both the secondary structure as well as the hairpin structure of A␤ 42 fibrils (26). In the hairpin structure of A␤ (both 40 and 42) fibrils, the side chains of amino acids present in the two ␤-sheets interdigitate to form a hydrophobic steric zipper (35, 36, 39 -43). This interdigitating pattern differs from one structural model to another (35,36,39,40,42,43). In other words, even though each model proposes the same hairpin structure for A␤, they differ in terms of inter-residue contacts proposed between the two ␤-strands. However, intramolecular Phe-19 -Leu-34 side-chain contacts have been proposed in most of the A␤ 42 and A␤ 40 models (36, 40, 42, 43), and they have even been  The topmost spectrum (one-dimensional) in each case is recorded by simply acquiring the signal after cross-polarization from protons. Other spectra are recorded using frequency selective REDOR pulse sequence, without applying the frequency selective pulse on 15 N channel (reference spectra, S 0 ) and spin-echo periods as indicated.
used to verify the hairpin structure of A␤ (26,40). We also looked for this contact in A␤ 42 -Cur aggregates. Because it is a long range contact, a two-dimensional 13 C-13 C through-space correlation spectrum was recorded using the PDSD recoupling scheme (30) with a long mixing time of 1 s. A selected region of this spectrum is shown in Fig. 6C. The presence of the unambiguous cross-correlation between the aromatic (␦/⑀) carbons of Phe-19 and ␥ carbons of Leu-34 is indicative of their side chains being spatially proximal. Hence, just like the secondary structure, the tertiary structure also seems to be conserved in A␤ 42 -Cur aggregates.
We note that the Phe-19 -Leu-34 contact by itself does not prove that the peptide is bent into a hairpin shape. This contact constrains the possibilities to either a hairpin or to an open state with the monomeric strands in antiparallel arrangement, with Phe-19 and Leu-34 in register. This alternative possibility would imply a fibrillar width that is much larger. We examined the width of the A␤ 42 and A␤ 42 -Cur aggregates by transmission electron microscopy. The A␤ 42 aggregates show a fibrillar morphology with the dominant form showing long, straight, unbranched fibrils that extend frequently beyond 1 m in length (Fig. 7A). The A␤ 42 -Cur aggregates, however, show only short fibrils or quasi-fibrillar structures (Fig. 7B). An analysis of the width of the fibrillar A␤ 42 and quasi-fibrillar A␤ 42 -Cur aggregates (a few representative positions are marked in red in Fig. 7, A and B) yields similar values (7.9 Ϯ 0.9 and 8.1 Ϯ 1.1 nm, respectively). This implies that the A␤ 42 -Cur aggregates have a similar hairpin architecture as the A␤ 42 fibrils. The disruption of the mesoscopic structures observed in the presence of curcumin is consistent with those reported earlier (5,7) and also as reported for Zn 2ϩ (26). It is possible that both of these agents precipitate the aggregates of A␤ 42 at an early stage and reduce their toxicity by locking them into a nontoxic conformation, and/or by simply removing them from the solution. Evidence for such a mode of action has indeed been seen for A␤ 40 -Zn 2ϩ (17,21) and also for A␤

CONCLUSION
Significant changes in the Asp-23 and Lys-28 conformation is a major feature of the curcumin-induced disruption of the A␤ 42 conformation. We note that the Phe-19 -Leu-34 contacts in A␤ 42 are thought to be intramolecular, whereas the salt bridge connecting Lys-28 and Asp-23 is thought to be between neighboring strands (35). Thus curcumin (like Zn 2ϩ ) seems to 4 B. Sahoo, unpublished data.  preserve the gross intramolecular conformation, while disrupting the intermolecular arrangements. This is consistent with its ability to disrupt the fibrillar architecture (as observed by transmission electron microscopy in Fig. 7) (5, 7). There are also specific changes in the intramolecular structure, with the C-terminal residues mostly affected by curcumin, whereas the N-terminal residues are mostly affected by Zn 2ϩ ions. Both of these therefore appear to stabilize a nonfibrillar or partly fibrillar family of aggregate structures whose morphology and toxic properties are very different from that of the regular fibrillar aggregates. These structural differences highlight potential target regions of the peptide for designing therapeutics for Alzheimer disease.