Investigating the Neuroprotective Effects of Turmeric Extract: Structural Interactions of β-Amyloid Peptide with Single Curcuminoids

A broad biophysical analysis was performed to investigate the molecular basis of the neuroprotective action of Curcuma longa extracts in Alzheimer’s disease. By combining circular dichroism and electron paramagnetic resonance experiments with molecular modeling calculations, the minor components of Curcuma longa extracts, such as demethoxycurcumin (2, DMC), bisdemethoxycurcumin (3, BDMC) and cyclocurcumin (4, CYC), were analyzed in a membrane environment mimicking the phospholipid bilayer. Our study provides the first evidence on the relative role of single curcuminoids interacting with Aβ-peptide. When the CYC and curcumin metabolite tetrahydrocurcumin (5, THC) were inserted into an anionic lipid solution, a significant modification of the Aβ CD curves was detected. These data were implemented by EPR experiments, demonstrating that CYC reaches the inner part of the bilayer, while the other curcuminoids are localized close to the membrane interface. Computational studies provided a model for the curcuminoid-Aβ interaction, highlighting the importance of a constrained “semi-folded” conformation to interact with Aβ analogously to the pattern observed in α-helical coiled-coil peptide structures. This combined approach led to a better understanding of the intriguing in vitro and in vivo activity of curcuminoids as anti-Alzheimer agents, paving a new path for the rational design of optimized druggable analogues.

Considering the implication of Aβ in oligomeric or fibrillar aggregates in the etiology and progress of Alzheimer's disease, thousands of molecules have been screened to identify new substances that can control the formation of β amyloid fibrils, either stabilizing the soluble peptide conformation or dissolving aggregates 27-30 . Indeed, among the substances tested for preventing or disaggregating β-amyloid fibrils, curcumin and several synthetic curcumin derivatives have been shown to bind Aβ, affecting the process of fibril formation [31][32][33][34][35][36] . In particular, the analysis of curcumin derivatives highlighted the importance of keto-enol tautomerism, proving that the derivatives existing predominantly in the enol-form have higher affinity for Aβ aggregates [37][38][39] . Among the curcumin derivatives, minor attention has been reserved for the remaining curcuminoids present in turmeric extract and their metabolites, although increasing evidence has led to the hypothesis that the activity of each of these compounds is determinant for the ultimate biological effect of Curcuma longa extract. Indeed, i) the turmeric extract containing optimized proportions of the single curcuminoids and integrated with a given proportion of the reduced metabolite THC exhibits enhanced biological activity 40 ; ii) the turmeric phytocomplex is therapeutic despite the unfavorable pharmacokinetic properties of curcumin alone; and iii) metabolites resulting from the metabolic transformation of CUR, such as THC, HHC and OHC (Fig. 1B), are themselves biologically active 20,21 .
Finally, we decided to undertake our biophysical study in the membrane-mimicking environment represented by a liposomal solution of different composition. This choice was dictated by the critical role played by the membrane surface in the β-amyloid toxicity, associated with the curcuminoids' aptitude for interaction with membranes, enhancing their biological effects 23,[49][50][51] .
Our study provides the first evidence regarding the relative role of the individual curcuminoid interacting with Aβ-peptide in membrane-mimicking environments. Among the examined curcuminoids, only CYC and THC can modify the Aβ conformation in the liposome system. Computational studies provide a molecular model for curcuminoid-Aβ interaction, suggesting that the constrained "semi-folded" conformation of CYC is suitable for interaction with Aβ, reproducing the pattern observed in α -helical coiled-coil peptide structures.

Analysis of curcuminoids.
We analyzed the curcuminoids present in Meriva ® using a different HPLC elution system (chromatograms obtained are shown in Supporting Information in Figure 1SI and 2SI). The chromatogram obtained under these new conditions showed the presence of CUR (retention time (rt) = 20.44 min), DMC (rt = 19.97 min), BDMC (rt = 19.50 min) and another component with the rt of 13.90 minutes. We therefore hypothesized that the new peak could correspond to cyclocurcumin. To validate our hypothesis, we embarked on the synthesis of cyclocurcumin to fully characterize the new entity via HRMS and NMR and then to evaluate its rt by HPLC analysis. The chromatogram obtained ( Figure 2SI) was compared with the previous one ( Figure 1SI), showing that the two had equal retention times. A quantitative analysis of curcuminoids present in the MIX, calculated as the percentage of the area under the curve, was performed, with the following results: 83% CUR, 11% DMC, 4% BDMC, and 2% CYC.
The three most abundant natural curcuminoids present in extracts 1, 2, and 3 were obtained by flash chromatography on silica gel using the eluent mixture DCM/MeOH (97:3). Each compound was subsequently characterized via HRMS and 1 H-13 C NMR. The purity of each isolated compound was assessed by HPLC analysis (see Supporting Information).

Synthesis of curcuminoids.
Since the quantitative analysis showed 2% cyclocurcumin in the mixture, we opted to synthesize this curcuminoid. The general reaction scheme is depicted in Fig. 2A.  (6) and OHC (7); reagents and conditions: c) H 2 (6.8 atm), Pt/C, MeOH, r.t., 13 h; 6, HHC (37%); 7, OHC (55%). Several attempts were made to obtain CYC (see Table SI1 in Supporting Information). We first reproduced the acid-catalyzed cyclization of curcumin as reported by Kiuchi and coworkers 52 using benzene and TFA (trifluoroacetic acid) for 65 h in the dark. The reported reaction yield is 20%; nevertheless, in our hands, the same reaction never provided a yield higher than 5%, probably because of the difficulties shown in the purification process. We attempted to modify the acidity of the reaction medium by replacing the TFA with TfOH (trifluoromethanesulfonic acid) (Table 1SI); however, the yield did not improve. A reduction of both the reaction time and the amount of by-products was achieved using microwave-assisted synthesis in benzene. The best result was obtained through a solvent-free reaction, performing just one 4 min MW run using TFA at 100 °C. Under these conditions, the reaction yield was 10%, barely sufficient to accumulate the desired quantities despite being completed in only a few minutes.
Among the metabolites formed after the administration of curcumin, the most interesting are THC, HHC and OHC, which are obtained after the reduction in vivo of curcumin. In all these compounds, the extended conjugation is lost, and in the case of HHC and OHC, the keto-enol tautomerization is lost as well. This behavior was very interesting as a way to assess the actual importance of tautomerism in the pharmacodynamics of curcumin.
The synthesis of these three metabolites was conducted through catalytic hydrogenation, suitably varying the reaction time and H 2 pressure 53 . THC was obtained in 77% yield using H 2 (1 atm), Pt/C, and MeOH for 1 h at room temperature (Fig. 2B).
HHC and OHC were obtained in a 40:60 mixture using higher hydrogen pressures and longer reaction times in a Parr ® apparatus (6.8 atm H 2 , Pt/C, MeOH at room temperature for 13 h (Fig. 2C). HHC and OHC were obtained with yields of 37% and 55%, respectively.

Circular dichroism analysis. Biological and biophysical experiments show that curcumin can interact with
Aβ-peptide in soluble or fibrillar form [31][32][33][34][35][36] . The correlation between the pathological role of Aβ-peptide and its interaction with the neuronal membrane is well established 23,[49][50][51]54 . The phytosomal formulation of curcuminoids (Meriva ® ) is proven to enhance the biological action of curcumin in terms of plasmatic availability 24,26 and penetration in tissues characterized by amyloid plaque deposition 55 .
To test the hypothesis that all these actions might share a common biophysical mechanism related to the interaction of Aβ with curcumin at the biomembrane, we analyzed Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) in liposomal solution by CD spectroscopy (Fig. 3A,B), focusing on the variation of its conformation in the presence of the individual curcuminoids shown in Fig. 1. We chose as the liposome systems small unilamellar vesicles (SUV) composed of 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) and/or 1,2-dioleoyl-sn-glycero-3-phospho-rac-(1-glycerol) (DOPG). Varying the lipid constituents can suitably modulate the liposome composition. The constituents can differ in the length and number of the methylene chains, as well as the polarity and charge of the polar heads. Given the importance of the superficial charge in driving the behavior of the molecules on the membrane surface, we decided to record CD spectra in three different liposome solutions distinguished by different superficial charges. DOPC is characterized by a zwitterionic membrane surface, DOPG by a negatively charged surface, and DOPC/DOPG (90/10 molar ratio) by a zwitterionic surface doped with a small amount of negative charge.

EPR experiments.
The CD experiments suggest that a significant perturbation of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) conformational preferences could be induced by the peptide interaction with CYC and THC. These interactions are modulated by the presence of lipid membranes in the case of THC and MIX, whereas CYC can modify the behavior of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) directly. To investigate this effect more deeply, we performed EPR analysis using experimental conditions consistent with the ones adopted for CD analysis. EPR spectroscopy requires the presence of unpaired electrons in the system to be investigated. To study microstructured systems, specific components can be selectively labeled with radical moieties, thus making it possible to obtain detailed information on the specific site where the label is positioned. In this work, we have used phospholipids specifically spin-labeled at different positions along the acyl chain. 5-PCSL is labeled very close to the hydrophilic head-group, thus allowing observation of the behavior of the bilayer region just below the interface. In contrast, 14-PCSL allows analysis of the properties of Scientific RepoRts | 6:38846 | DOI: 10.1038/srep38846 the inner part of the bilayer hydrophobic core 58 . Using these two spin-labels, we focused on two different aspects of the systems under consideration in this work. First, we studied the interaction of curcumin and its analogues with the lipid membranes formed by DOPC, DOPG and DOPC/DOPG (90/10 molar ratio); second, we analyzed the effects of these ligands on the interaction of the Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) peptide with the lipid bilayers. The 5-PCSL spectrum in DOPC bilayers (Fig. 4A, solid line) shows a clearly defined axially anisotropic behavior, as detectable by the splitting of the low-and high-field lines, indicating an ordered organization of the outer segments of the lipid acyl chains. The 14-PCSL spectrum, registered in the same system, shows an almost isotropic three-line lineshape (Fig. 4B, solid line).
Very similar results were obtained for the DOPG and DOPC/DOPG (90/10 molar ratio) bilayers (spectra not shown). This evidence indicates, for all the considered lipid systems, an increase flexibility in terms of the segmental chain mobility moving inward from the polar head-groups to the inner hydrophobic core, as expected for membranes in the liquid-disordered (L d ) crystalline state 59 . To quantitatively analyze the spectra, the values of the outer hyperfine splitting, 2A max , were determined by measuring the difference between the low-field maximum and the high-field minimum, using a home-made MATLAB-based software routine. In general, 2A max is dependent on both the amplitude (i.e., order) and the rate of chain rotational motion and is therefore a useful parameter for characterizing chain dynamics, as determined by local ordering, in phospholipid membranes 60 . It is a sensitive means for detecting and quantifying lipid bilayer interactions with guest molecules, as previously demonstrated 61 . The 2A max value is much higher for 5-PCSL than for 14-PCSL (approximately 51 G vs. 32 G) and is quite insensitive to the lipid charge (for 2A max values for 5-PCSL and 14-PCSL in DOPC, DOPG and DOPC/DOPG (90/10 molar ratio) bilayers (Table 2SI in Supporting Information). In the presence of curcumin, the anisotropy of the 5-PCSL spectrum in DOPC increases slightly but significantly, as shown by Fig. 4A, dashed line. The anisotropy increase indicates a motional restriction of the spin-labeled chain and is usually related to the presence of guest molecules 61 . In contrast, the 14-PCSL spectrum remains almost unaffected (Fig. 4B). These results indicate that curcumin interacts with the bilayer, positioning close to the head-group region with a limited but still detectable intercalation among the outer segments of the lipid acyl chains. This conclusion is in reasonable agreement with the "surface association" of curcumin proposed by Sun 62 , while confirming a certain degree of internalization as found by Barry 23,54 .
Qualitatively similar effects were observed for DOPG and DOPC/DOPG (90/10 molar ratio) bilayers, perturbation being slightly higher for the anionic lipid system. Consistently, the 2A max values reported in Table 2SI show an increase for 5-PCSL, while remaining almost constant in the case of 14-PCSL. The analogues of curcumin  Table 2SI).
To easily compare the effects of the various ligands on the bilayer structure, Fig. 4 (panels C-E) reports the 2A max variation upon ligand inclusion in all the considered systems for both spin-labels. In the case of 5-PCSL, the variation is positive in all cases and higher for the DOPG bilayers.
In the case of 14-PCSL, the 2A max variations are almost negligible. Thus, it can be concluded that all CUR derivatives present a marked tendency to interact with the bilayers, shallowly penetrating between the lipid acyl chains. The differences between the various ligands are not dramatic, even though CYC seems more effective in perturbing the membrane inner layer. Indeed, for this ligand, even the 14-PCSL 2A max in anionic membranes is slightly but significantly perturbed. Then, we turn our attention to the interaction of the Aβ peptide with lipid membranes, analyzing the extent to which it is affected by the presence of curcumin and its analogues.
Inspection of the figure indicates that the presence of ligands at the bilayer interface hinders any membrane penetration of the peptide. We propose that the accumulation of a layer of bound curcumin (or its analogues) at the lipid membrane interface reduces the accessibility and number of binding sites for Aβ on the membrane. Moreover, curcumin on the membrane surface can reduce Aβ-membrane interaction by altering the lipid head-group charge and reducing the favorable electrostatic interactions between Aβ and lipid head-groups 64 . The inspection of Fig. 3 clearly shows that all the ligands behave similarly, with the possible exception of CYC in DOPG membrane: in this case, the further 2A max increase due to Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) suggests that the peptide preserves a detectable ability to interact with the bilayer even in the presence of the ligand. This evidence supports the hypothesis that DOPG membranes create a perfect environment to enhance Aβ(25-35) interaction with CYC.

Molecular Modeling Studies.
To relate the obtained CD and EPR results to the different structural features of the tested curcuminoids, a molecular modeling study was conducted. In particular, the structural features responsible for the interaction with Aβ(25-35) were investigated.
First, a comprehensive conformational analysis was performed on all possible tautomers of i) CUR, which is unable to affect the Aβ(25-35) conformational equilibrium either in a liposomal environment or in HFIP/water solution; ii) THC, which affects the Aβ(25-35) conformational equilibrium only in a liposomal environment; and iii) CYC, which perturbs the conformational behavior of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35), both, in DOPG and HFIP/water solution. The conformational space was sampled using a protocol including molecular dynamics (MD) simulations followed by molecular mechanic (MM) optimization, mimicking the polarity of a water medium (ε = 80*r). The identified minima were then subjected to full geometry optimization using the semi-empirical PM7 (MOPAC) method (ε = 1; for details see Experimental Section), thus obtaining a new set of conformers using a dielectric constant similar to the one for the highly hydrophobic membrane environment 65 . In this regard, it must be emphasized that only qualitative considerations can be deduced by comparing the results obtained from MM (ε = 80*r) and PM7 (ε = 1) calculations.
The MM and PM7 conformers obtained were ranked by their potential energy values (i.e., Δ E from the global energy minimum) and classified into different families based on the values of their torsional angles. Then, the distance between the centroids of the two aromatic rings was calculated (d1 ; Figures 3SI, 5SI-8SI; Tables 3SI-6SI). According to this latter parameter, the conformers were further classified as i) "folded" when d1 was ≤ 5 Å, ii) "semi-folded" when d1 was > 5 Å and < 9 Å, and iii) "extended" when d1 was ≥ 9 Å. Finally, the occurrence rate of each conformational family was calculated (Table 1 and Tables 3SI-6SI).
The results showed peculiar conformational features for each of CUR, CYC, and THC. The conformational behavior of CUR differed in the two tautomeric forms. Indeed, the conformers of the CUR-diketone form are distributed in two main conformational families: the "extended" and the "folded" (Tables 1 and 3SI; Figures 3SI-4SI), with the latter including the global minimum (GM) conformer. In particular, comparing PM7 with MM shows an increase in the number of "folded" conformers, i.e., presenting the two aromatic rings interacting with each other, and a concomitant decrease in the "extended" conformers can be observed. This result is due to the higher degree of accuracy of semi-empirical vs. MM methods in calculating electronic parameters, as well as to the fact that electrostatic interactions are expected to increase in non-polar (low dielectric) solvent.
In contrast, in the keto-enol form of CUR, the conjugation of the double bonds extending from one aromatic ring to the other and the formation of a hydrogen bond between the ketone oxygen and the hydrogen of the enol function ( Fig. 5A and B) strongly constrain the conformational freedom of the structure. Consequently, all CUR keto-enol conformers belong to the "extended" conformational family, resulting from both the MM and PM7 calculations (Tables 1 and 4SI; Figure 5SI). In detail, the PM7 low-energy conformers (within 5 kcal/mol from the GM) of the keto-enol form of CUR showed a specific preference for two sets of conformations, differing only for the value of the torsional angle τ 1, herein named conf I (τ 1 = ∼ 180°) and conf II (τ 1 = ∼ 0°) ( Fig. 5A and B; Table 4SI). These results are consistent with the ones obtained from the analysis of the experimentally determined structures of CUR (all in the keto-enol form) deposited in the Cambridge Structural Database (CSD) (Table 7SI). Indeed, CUR can assume either conf I (CSD codes: AXOGIE, AXOGOK, BINMEQ06, BINMEQ07, and BINMEQ08) or conf II (CSD codes: BINMEQ, BINMEQ01, BINMEQ02, BINMEQ03, BINMEQ04, and BINMEQ05), as shown by our conformational analysis.
The presence of the dihydropyran-4-one ring in CYC determined a different conformational behavior from the observation for CUR tautomers. Indeed, all CYC conformers, both MM and PM7, fall into the "semi-folded" conformational family (Table 1 and Figure 6SI). Overall, CYC showed four possible subfamilies of conformers, differing for the values of the torsional angles τ 1, τ 2, and τ 3 and characterized by d1 ranging from 7.00 to 9.66 Å (conf I-IV in Fig. 5C-F and Table 5SI).
THC shows yet another conformational preference compared either to CUR or CYC. According to the results of our analysis, THC conformers fall into all three conformational families regardless of the tautomeric form or the computational method used (Table 1; Figure 9SI; Table 6SI). However, either in the diketone or in the keto-enol form, the "folded" conformation is strongly stabilized by the formation of a hydrogen bond between the hydroxyl groups of the two aromatic rings ( Figure 9SI). This intra-molecular H-bond may compete with the formation of H-bond interactions with solvent molecules, which are not treated explicitly in this simulation. Thus, it is likely that the actual conformational behavior of THC strongly depends on the chemical environment.
Furthermore, to investigate the experimentally observed interaction between Aβ(25-35) and CYC in HFIP/ water 80/20 solution, we performed a dynamic molecular docking simulation on the CYC-Aβ(25-35) complex, mimicking the polarity of the media with a dielectric constant of 30*r. To fully explore all possible binding sites/ modes during the dynamic docking procedure, the binding site was defined as the whole peptide, and all rotatable bonds of CYC and Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) were left fully free to move. A force constant of 10 kcal/mol/Å 2 was only applied on the backbone hydrogen bonds of (Asn27-Leu34) to avoid unrealistic results during the annealing simulation (for details, see the Experimental Section). First, a Monte Carlo/minimization approach for the random generation of a maximum of 20 acceptable complexes was used. To ensure a wide variance of the input structures to be minimized afterward, an energy tolerance value of 10 6 kcal/mol from the previous structure was used. After the energy minimization step, the energy test, with an energy range of 50 kcal/mol, and a structure similarity check (rms tolerance = 0.3 kcal/A) was applied to select the 20 acceptable structures. To test the thermodynamic stability of the resulting docked complexes, these latter were then subjected to simulated annealing (SA) calculations. In SA the temperature is altered in time increments from an initial (500 K) to final (300 K) temperature. The temperature of 500 K was applied with the aim of surmounting torsional barriers, thus allowing a full rearrangement of the ligand and the peptide (see Experimental section for details).
The resulting complexes were analyzed and ranked by their conformational and interaction energy values (Table 8SI). On twenty Monte Carlo generated structures, only twelve ( Figure 10SI) still present CYC bound to Aβ(25-35) peptide after SA calculations. Interestingly, just one of these twelve SA complexes preserves the same binding mode resulting from Monte Carlo docking procedure, showing thermodynamic stability (Complex 1, Figure 10SI). Importantly, this complex also resulted to be the lowest energy complex and the one characterized by the most favorable interaction energy (Table 8SI). Such results strongly suggest that Complex 1 is representative for a stable and rather specific binding mode of CYC to Aβ (25-35) peptide ( Fig. 6A and B).
Interesting data provided by our docking results showed that the calculated binding mode of CYC to Aβ(25-35) resembles the canonical packing of the α -helical coiled-coil dimerization motif, where the side-chains of the "a" and "d" residues of two facing heptad repeat motifs interact with each other (Fig. 6C) 66 . In fact, in the consensus sequence of the heptad repeat motif, the "a" and "d" positions are occupied by hydrophobic residues (one of which can be replaced by an asparagine), favoring the formation of coiled-coil structures, where the i, i + 4, and i + 7 residues of a helix interact with the equivalent residues of another helix. As evidenced by the superimposition reported in Fig. 6C, CYC can mimic the interactions performed by such residues. In particular, the two substituted phenyl rings (d1 = 8.85 Å) can reproduce the orientation of the hydrophobic side chains of the i and i + 7 residues (distance from the side chain centroids = 8.20 Å), and at the same time, the dihydropyran-4-one ring and the ethylene bridge overlap with the central i + 4 residue (Fig. 6C).
Taken together our computational results support the hypothesis that CYC targets heptad repeat motifs like the one found in Aβ(25-35) (Fig. 6C), whose structure is preserved in full length Aβ(1-42) and which is part of

Discussion
A great deal of evidence shows that curcumin and several synthetic curcumin derivatives bind beta amyloid peptides, affecting the process of fibril formation [31][32][33][34][35][36] . In light of the well-documented pharmacological effects, including the neuroprotective effect exerted by the less abundant curcuminoid molecules in the Curcuma Longa extract, we studied these molecules for their ability to affect the conformational arrangement of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35), providing a molecular basis for the observed biological effect. In particular, our analyses were intended to understand whether the neuroprotective effect of the phytocomplex derived from Curcuma Longa can be credited to the ability of the curcuminoids to lessen lipid membrane-Aβ interactions and, by consequence, the Aβ-induced membrane-disruptive perturbations [49][50][51] . Accordingly, we investigated how curcuminoids, taken as single purified molecules, affect Aβ(25-35)− membrane interactions through CD and EPR experiments, using DOPC, DOPG and DOPC/DOPG liposomal solutions to mimic lipid bilayer membrane systems. The CD spectra of monomeric Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) in the different superficially charged liposome environments were recorded after the addition of the compounds CUR, DMC, BDMC, CYC, THC, HHC, OHC and MIX. These experiments were conducted to evaluate the effect of curcuminoids in solution on Aβ(25-35)-membrane interactions. In the presence of these curcuminoids, used pure or mixed, no significant modification of the Aβ(25-35) conformational equilibrium was observed in DOPC and DOPC/DOPG liposomal solution. In contrast, when CYC and THC were inserted into anionic DOPG lipid solution, a significant modification of the Aβ(25-35) CD curves was detected, corresponding to a perceptible increase in the soluble form of the α-helix vs. β-strand and random coil conformations. These data were obtained by EPR experiments that revealed, for both Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) and curcuminoids, a shallow insertion in the DOPC, DOPC/DOPG, and DOPG lipid layers, just below the outer interface (5-PCLS 2A max shift). Only in the case of CYC in DOPG liposomal solution was a slight but significant perturbation of 14-PCSL 2A max observed, demonstrating that the molecule reaches the inner part of the bilayer. The inclusion of the curcuminoid ligands on the bilayers prevents further interaction of the membrane with the Aβ peptide, causing only minimal variation in the ligand-bilayer systems, with the sole exception of CYC in DOPG. These results are consistent with other literature data, based on TEM (transmission electron microscopy) images and CD spectra 64 , which also evidenced that the presence of CUR does not inhibit Aβ40 aggregation or significantly alter the morphologies of Aβ monomers, prefibrillar or fibrillar aggregates, while it reduces the extent of cell membrane permeabilization induced by Aβ aggregates.
In the case of DOPG membrane containing CYC or THC, the presence of the guest molecules induces a significant alteration of the Aβ(25-35) conformational equilibrium, as observed in the CD spectra, and a marked perturbation of the membrane outer layer, as observed in the EPR experiments. This result may indicate a direct interaction of CYC and THC with Aβ(25-35), but it still cannot be excluded that the observed effects are mediated by the lipid membrane. To investigate this issue, CD spectra were recorded in an HFIP/water (80/20 v:v) mixture in the absence and in the presence of CYC and THC. Under these conditions, CYC was still able to modify the conformational equilibrium of Aβ(25-35) (Fig. 3C,D), whereas THC induced only minimal modification of the Aβ(25-35) secondary structure. Thus, THC does not directly interact with monomeric Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) in solution; nonetheless, when integrated into the DOPG system, it induces a modification in the Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) conformational equilibrium mediated by the lipid membrane. By contrast, CYC can modify the conformation of Aβ(25-35) either in HFIP/water solution or in anionic DOPG lipid solution, indicating that its constrained structure -naturally derived from an intramolecular cyclization -is suitable for direct interaction with Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) in both environments. Computational studies provided a molecular model for this interaction and a possible explanation for the peculiar ability of CYC, unlike CUR and THC, to interact directly with Aβ(25-35), either in solution or in the liposomal environment. Indeed, according to the results of our molecular modeling studies, this result is due to the constrained "semi-folded" conformation of CYC, which allows it to mimic the orientation of the hydrophobic side chains of short peptide motifs, similarly to our previous observations for other small ligands 68,69 . Thus, by mimicking the orientation of the i, i + 4, and i + 7 residues of an α -helix, CYC could interact with Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35), reproducing the pattern observed in α -helical coiled-coil structures, the latter being characterized by the presence of one of the principal dimerization motifs in proteins 70 . Coiled-coil peptide structures are known to anchor to cellular membranes 71 and are formed by heptad repeats containing hydrophobic residues at the "a" and "d" positions. Importantly, the formation of helical coiled-coil structures, through the reciprocal interaction of several heptad repeat motifs, was also observed for APP in micelles, leading to its dimerization 67 . In line with these results, our previous NMR studies evidenced a strong similarity between the structure of Aβ(1-42) peptide in an apolar microenvironment and the structure of the fusion domain of influenza hemagglutinin in detergent micelles (also characterized by the ability to form coiled-coil structures) 72 . A heptad repeat motif is still contained in the Aβ(25-35) fragment, and, similarly to previous observations, could lead to the formation of α -helical coiled-coil structures on the membrane lipid surface that are amenable to interaction with CYC. On the other hand, monomers and soluble aggregates of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) in the form of α -helix are also likely to occur in HFIP/water solution, since, in this environment, the helical structure of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) is strongly stabilized 73 . Accordingly, CYC also perturbs the conformational equilibrium of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) in this environment.
The "semi-folded" conformation of CYC is poorly present in CUR (not present at all in its predominant keto-enol form) but still significantly present in the case of THC. However, due to the higher conformational flexibility of THC with respect to CYC, it represents just a fraction of the whole conformational space, with its relative weight depending on the polarity and the H-bond donor/acceptor properties of the solvent. Accordingly, CUR and other analogues characterized by a conjugated and mostly planar structure (DMC, BDMC) behave only as membrane stabilizers, forming ideal mixtures with DOPG lipids, reducing lipid flexibility, and preventing Aβ(25-35)-peptide insertion. On the contrary, CYC, which is structurally constrained in the "semi-folded" conformation, is able to bind/stabilize the monomeric helical form of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35). In this way, CYC may prevent not only the formation of toxic β -sheet aggregate, but also the formation of membrane disrupting aggregates formed by α -helical coiled-coil structures 74,75 . Finally, the more flexible THC shows a different behavior in DOPG bilayer or in the HFIP/water mixture; indeed, THC shifts the conformational equilibrium of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) to the helical structure in DOPG, but this ability was not preserved in the HFIP/water solution.

Conclusions
The results of this work show that the well-documented biological activity of turmeric extract is based on a molecular mechanism in which CUR and its analogues reduce the Aβ-membrane interactions due to their marked tendency to interact with the lipid bilayer. However, this effect is differently mediated among the examined curcuminoids since i) only CYC and THC can modify the conformational behavior of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) in the liposome environment, stabilizing the α -helix conformation and ii) only CYC can still modify the conformational behavior of Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) in the HFIP/water mixture. While the enol form of CUR ("extended" conformation) was found to be responsible for the inhibition of Aβ fibril formation, our molecular modeling studies suggest that, to interact with the helical form of Aβ and prevent the formation of membrane-disrupting aggregates (possibly consisting of α -helical coiled-coil structures), it is necessary for CUR and its derivatives to assume the partially folded conformation in which CYC is naturally constrained.
This combined approach leads to a better understanding of the intriguing in vitro and in vivo activity of curcuminoids as anti-Alzheimer agents, paving a new path for the rational design of optimized druggable analogues.

Separation of curcuminoids. Liquid chromatography was performed on a Waters system
Subsequently, the peptide was purified by reversed phase chromatography (HPLC) using a Phenomenex column C18 (30 cm, 4 cm, 300 Å, 15 mm spherical particle size column). The peptide was characterized on a Finningan LCQ Deca ion trap instrument equipped with an electrospray source (LCQ Deca Finnigan, San José. CA, USA). The samples were directly infused in the ESI source using a syringe pump at a flow rate of 5.0 mL/min. The data were analyzed using the Xcalibur software. The sample purity was > 98%.
Sample preparation for CD analysis. All CD experiments were recorded in small unilamellar liposome vesicles (SUVs) consisting of pure DOPC lipid, pure DOPG lipid and DOPC/DOPG mixtures at 90/10 M/M (molar ratio). Samples of DOPC, DOPG and DOPC/DOPG (90/10) small unilamellar liposome vesicles (SUVs) were prepared as follows: 200 μ M of each phospholipid were dissolved in a CH 2 Cl 2 /MeOH solution, then kept in a round-bottom test tube, and a thin lipid film was produced by evaporating the solvent with dry nitrogen gas. The final traces of solvent were removed by subjecting the sample to vacuum desiccation for one night. The samples were then hydrated with an appropriate volume of 10 mM phosphate buffer at pH = 7.4 and repeatedly vortexed, obtaining a suspension. The MLV suspension thus obtained was sonicated for 5-10 minutes to obtain SUVs.
To remove aggregate states, which can be present in Aβ samples, the dry peptide was pretreated with TFA for 3 hours, followed by dilution with MilliQ water and lyophilization. The TFA pretreatment gives the Aβ peptide the properties of monomeric, random coil structures and eliminates preaggregated material 77 . This procedure was adopted for all CD samples, immediately before dissolution in the appropriate solvent. This procedure was also adopted before preparing Aβ(25-35) amyloid fibrils 78 .
Scientific RepoRts | 6:38846 | DOI: 10.1038/srep38846 EPR spectroscopy. For EPR experiments, multi-lamellar vesicles (MLVs) of DOPC, DOPG and DOPC/ DOPG (90/10 molar ratio) were prepared by mixing appropriate amounts of lipids, dissolved in a CH 2 Cl 2 /MeOH mixture (2:1 v/v, 10 mg/mL concentration), in a round-bottom test tube. Spin-labeled phosphatidylcholines (1-palmitoyl-2-stearoyl-(n-doxyl)-sn-glycero-3-phosphocholine, n-PCSL with n = 5, 14) were added to the lipid mixture (1% by weight of the total lipid) by mixing appropriate amounts of a spin-label solution in ethanol (1 mg/ mL) with the lipid organic mixture. A thin lipid film was produced by evaporating the solvents with dry nitrogen gas. The total weight of the lipids for each sample was 0.2 mg. The final traces of solvents were removed by subjecting the sample to vacuum desiccation for at least 3 h. The samples were then hydrated with 20 μ L of 10 mM phosphate buffer at pH = 7.4, gently warmed and repeatedly vortexed to obtain a MLV suspension.
Samples containing curcumin or one of its analogues were prepared by the same procedure, adding appropriate amounts of the flavonoid dissolved in DMSO (10 mg/mL) to the buffer during the re-hydration step to obtain a 0.5 molar percentage with respect to the total lipids. The addition of DMSO was preventively checked to result in no perturbation of the vesicular systems.
EPR spectra were recorded with a 9 GHz Bruker Elexys E-500 spectrometer (Bruker, Rheinstetten, Germany). Samples were placed in 25 μ L glass capillaries and flame sealed. The capillaries were placed in a standard 4 mm quartz sample tube containing light silicone oil for thermal stability. All measurements were performed at 25 °C. The spectra were recorded using the following instrumental settings: sweep width, 100 G; resolution, 1024 points; time constant, 20.48 ms; modulation frequency, 100 kHz; modulation amplitude, 1.0 G; incident power, 6.37 mW. Several scans, typically 16, were accumulated to improve the signal-to-noise ratio.

Molecular modeling. Molecular modeling calculations were performed on an SGI Origin 200 8XR12000
and an E4 Server Twin 2 x Dual Xeon 5520, equipped with two nodes. Each node was 2 x Intel Xeon QuadCore E5520, 2.26 GHz, 36 GB RAM. The molecular modeling graphics were implemented on SGI Octane 2 workstations.
The apparent pKa values of CUR and its derivatives in their tautomeric forms were estimated using the ACD/ Percepta software 79 . Accordingly, the percentages of neutral/ionized forms were computed at pH 7.2 (cytoplasm) using the Henderson-Hasselbach equation.
Conformational analysis. All compounds were built considering the prevalent ionic forms of each tautomer using the Insight 2005 Builder module (Accelrys Software, Inc., San Diego).
Atomic potentials and partial charges were assigned using the CVFF force field 80 . The conformational space of the compounds was sampled through 200 cycles of simulated annealing (ε = 80*r). To avoid unrealistic results, the torsional angle of the double bonds was constrained within 180° using a force constant of 100 (kcal/mol)/Å. In simulated annealing, the temperature is altered in time increments from an initial temperature to a final temperature by adjusting the kinetic energy of the structure (by rescaling the velocities of the atoms). The following protocol was applied: the system was heated to 1000 K over 2000 fs (time step of 3.0 fs); the temperature of 1000 K was applied to the system for 2000 fs (time step of 3.0 fs) to surmount torsional barriers; then, the temperature was linearly reduced to 300 K in 1000 fs (time step of 1.0 fs). The resulting conformations were then subjected to Molecular Mechanics (MM) energy minimization within the Insight 2005 Discover 3 module (CVFF force field; ε = 80*r) until the maximum RMS derivative was less than 0.001 kcal/Å, using the conjugate gradient 81 as the minimization algorithm. The resulting MM conformers were ranked by (i) their potential energy values (i.e., Δ E from the global energy minimum), (ii) intra-molecular hydrogen bonds, (iii) torsional angles, and (iv) the interatomic distance between the centroids of the two aromatic rings. The centroids were determined considering the heavy atoms of the aromatic rings (Pseudo_Atom Define command, Biopolymer Module, Insight 2005). Then, the occurrence rates were calculated.
All MM conformers were subjected to a full geometric optimization by semi-empirical calculations, using the quantum mechanical method PM7 82 in the MOPAC2012 package 82 and EF 83 (eigenvector following routine) as the geometric optimization algorithm. The GNORM value was set to 0.01. To achieve full geometric optimization, the criterion for terminating all optimizations was increased by a factor of 100, using the keyword PRECISE. The resulting PM7 conformers were ranked as MM conformers.
The docking protocol included a Monte Carlo-based conformational search of the ligand (CYC) within the Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) peptide. The binding domain area was defined as a subset including all residues of the Aβ(25-35) peptide. All atoms included in the binding domain area were left free to move throughout the course of the docking calculations, whereas to avoid unrealistic results, a tethering restraint was applied to the SCRs of the protein. In particular, for the α -helix Asn27-Leu34, the distance between backbone hydrogen bond donors and acceptors was restrained within 2.5 Å using a force constant of 10 kcal/mol/Å 2 .
A Monte Carlo/minimization approach for the random generation of a maximum of 20 acceptable complexes was used. During the first step, starting from the previously obtained roughly docked structures, the ligand was moved by a random combination of translation, rotation, and torsional changes to sample both the conformational space of the ligand and its orientation with respect to the protein (MxRChange = 3 Å; MxAngChange = 180°). During this step, the van der Waals (vdW) and Columbic terms were scaled to a factor of 0.1 to avoid very severe divergences in the vdW and Columbic energies. If the energy of a complex structure resulting from random moves of the ligand was higher by the energy tolerance parameter than the energy of the last accepted structure, it was not accepted for minimization. To ensure a wide variance in the input structures to be successively minimized, an energy tolerance value of 10 6 kcal/mol from the previous structure was used. After the energy minimization step (conjugate gradient; 10000 iterations; ε = 30*r), the energy test, using an energy range of 50 kcal/mol, and a structure similarity check (rms tolerance = 0.3 kcal/Å) were applied to select the 20 acceptable structures. Each subsequent structure was generated from the last accepted structure. Following this procedure, the resulting docked structures were ranked by their conformational energy and analyzed. Finally, to test the thermodynamic stability of the resulting docked complexes; these latter were subjected also to a molecular dynamics simulated annealing protocol (ε = 30*r). A tethering restraint was applied to the SCRs of the complex. The same set of structural restraints used for previous docking calculations was applied. The protocol included 5 ps of a dynamic run divided into 50 stages (100 fs each), during which the temperature of the system was linearly decreased from 500 to 300 K (Verlet velocity integrator; time step = 1.0 fs). In simulated annealing, the temperature is altered in time increments from an initial temperature to a final temperature. The temperature is changed by adjusting the kinetic energy of the structure (by rescaling the velocities of the atoms). Molecular dynamics calculations were performed using a constant temperature and constant volume (NVT) statistical ensemble, with direct velocity scaling as the temperature control method (temp window = 10 K). In the first stage, the initial velocities were randomly generated from the Boltzmann distribution, according to the desired temperature, while during the subsequent stages, the initial velocities were generated from the dynamics restart data. A temperature of 500 K was applied to surmount torsional barriers, thus allowing an unconstrained rearrangement of the "ligand" and the "protein" active site (initial vdW and Columbic scale factors = 0.1). The temperature was then linearly reduced to 300 K in 5 ps, and, concurrently, the vdW and Columbic scale factors were similarly increased from their initial values (0.1) to their final values (1.0). A final round of 10 5 minimization steps (ε = 30*r) followed the last dynamics steps, and the minimized structures were saved in a trajectory file.
The resulting complexes were ranked by their conformational energy and analyzed by considering the non-bond interaction energies between the ligand and the peptide (vdW and electrostatic energy contribution; Group Based method; CUT_OFF = 100; ε = 2*r; Discover_3 Module of Insight2005). The complex characterized by the lowest conformational and interaction energy was chosen as the most representative. To allow the relaxation of the whole system, this complex was further minimized with no tethering restraint (CVFF force field; Cell Multipole method for non-bond interaction) using the Conjugate Gradient algorithm (maximum rms derivative less than 0.01 kcal/Å). Finally, the selected complex was compared with the experimentally determined structures of i) HDM2 protein in complex with p53 (PDB ID: 1YCR), ii) estrogen receptor β in complex with the nuclear receptor coactivator 5 (Ncoa5; PDB ID: 2J7X), iii) the coiled-coil dimerization motif of Geminin (PDB ID: 1T6F), and iv) the coiled-coil dimerization motif of the human ROCK I protein (PDB ID: 3O0Z) using the Biopolymer and Homology module of Insight 2005 (Accelrys, San Diego).