Alzheimer's Disease Amyloid-β Binds Copper and Zinc to Generate an Allosterically Ordered Membrane-penetrating Structure Containing Superoxide Dismutase-like Subunits*

Amyloid β peptide (Aβ) is the major constituent of extracellular plaques and perivascular amyloid deposits, the pathognomonic neuropathological lesions of Alzheimer's disease. Cu2+ and Zn2+ bind Aβ, inducing aggregation and giving rise to reactive oxygen species. These reactions may play a deleterious role in the disease state, because high concentrations of iron, copper, and zinc have been located in amyloid in diseased brains. Here we show that coordination of metal ions to Aβ is the same in both aqueous solution and lipid environments, with His6, His13, and His14 all involved. At Cu2+/peptide molar ratios >0.3, Aβ coordinated a second Cu2+ atom in a highly cooperative manner. This effect was abolished if the histidine residues were methylated at Nε 2, indicating the presence of bridging histidine residues, as found in the active site of superoxide dismutase. Addition of Cu2+ or Zn2+ to Aβ in a negatively charged lipid environment caused a conformational change from β-sheet to α-helix, accompanied by peptide oligomerization and membrane penetration. These results suggest that metal binding to Aβ generated an allosterically ordered membrane-penetrating oligomer linked by superoxide dismutase-like bridging histidine residues.

The amyloid ␤ peptide (A␤) 1 is a normally soluble 4.3-kDa peptide found in all biological fluids, but it accumulates as the major constituent of the extracellular deposits that are the pathologic hallmarks of Alzheimer's disease (AD) (1). Genetic evidence from early onset cases of AD indicates that A␤ metabolism is linked to the disease (2). A␤ peptides are neurotoxic (3,4), but the mechanism of toxicity and the species of A␤ responsible have not been clearly defined.
There is mounting evidence that oxidative stress causing cellular damage is central to the neurodegeneration of AD (5,6). There is an increase in oxidation of proteins as well as nuclear and mitochondrial DNA in AD brains (7)(8)(9). A␤ has the ability to enhance the generation of reactive oxygen species in cells of neural origin as well as in cell-free media (10 -12). A␤ in vitro binds metal ions, including Zn 2ϩ , Cu 2ϩ , and Fe 3ϩ , inducing peptide aggregation that may be reversed by treatment with chelators such as EDTA (13,14). Furthermore, extensive redox chemical reactions take place when A␤ binds Cu 2ϩ and Fe 3ϩ , reducing the oxidation state of both metals and producing H 2 O 2 from O 2 in a catalytic manner (12). Because elevated levels of copper (400 M), zinc (1 mM), and iron (1 mM) are found in amyloid deposits in AD-affected brains (15,16), the oxidative stress observed in AD may be related to the production of reactive oxygen species by metal-bound forms of A␤. This hypothesis is supported by the recent observation that senile plaques and neurofibrillary tangles isolated from AD brains were capable of generating reactive oxygen species and that copper and iron were essential (17). Moreover, our studies (18) have shown that the solubilization of A␤ from post-mortem brain tissue of AD patients was increased in the presence of metal chelators such as N,N,NЈ,NЈ-tetrakis (2-pyridyl-methyl) ethylene diamine and bathocuproine. Recently, the dramatic inhibition of amyloid deposition in transgenic mice treated orally with a Cu 2ϩ /Zn 2ϩ -selective chelator has been reported (19).
The discovery that metal binding to A␤ may be responsible for some of the pathological effects of AD makes characterization of the metal-binding site of interest as a potential therapeutic target. The aim of the present study was to characterize the structural consequences of A␤ binding to Cu 2ϩ and Zn 2ϩ in solution and to identify amino acid residues involved in metal binding.

MATERIALS AND METHODS
Peptides were synthesized as described previously (20) or obtained from Auspep (Melbourne, Australia) and from the W. M. Keck Laboratory (Yale University, CT). 2 H 2 O, TFE-2 H 3 , NaO 2 H, SDS-2 H 25 , and 2 HCl were obtained from Cambridge Isotope Laboratories (Andover, MA). The spin-labeled zwitterionic phospholipid 1-palmitoyl-2-(16doxyl stearoyl) phosphatidyl choline was obtained from Avanti Polar Lipids Inc. (Pelham, AL). The acidic phospholipid spin label 1-palmitoyl-2-(16-doxyl stearoyl) phosphatidyl serine was synthesized according to Hubbell and McConnell (21). Both spin probes were checked for purity and to ensure that the number of spins/mol were Ͼ90% of theory (22). Synthetic palmitoyloleoyl phosphatidyl choline (POPC) was purchased from Sigma, and palmitoyloleoyl phosphatidyl serine (POPS) was purchased from Avanti Polar Lipids Inc. LUV were prepared by the method described by Mayer et al. (23). Peptides were added to the desired concentration to a suspension of LUV in PBS, and the mixture was vortexed under N 2 for 10 min at 305 K. Negatively charged LUV were made with 50% POPS and 50% POPC. Zwitterionic LUV were made with 100% POPC. Because the longer A␤ peptides are prone to aggregation in solution, all experiments were carried out with freshly prepared samples.
The samples for NMR varied depending on the solution conditions used; in aqueous PBS with 10% 2 H 2 O added, samples containing A␤28 had a peptide concentration of 1 mM, whereas those containing A␤40 were run at 0.3 mM. When the peptide is made up fresh in metal-free conditions and the undissolved aggregates are spun out before use, the peptide aggregates only very slowly (days). Some studies were performed in SDS solution (50 mM phosphate, 200 mM SDS-2 H 25 , pH 5.3, 10% 2 H 2 O) where all peptide concentrations were 1.5 mM. NMR spectra were recorded on Bruker DRX-600 and AMX-500 spectrometers as described previously (24). Ultracentrifugation measurements were carried out on A␤28 and A␤40 in PBS at 1 and 0.3 mM, respectively, as described previously (24).
EPR Spectroscopy-X-band continuous wave EPR spectra of the Cu 2ϩ -peptide complexes were obtained using a Bruker EC106 spectrometer. Samples were loaded into hematocrit capillary tubes and inserted reproducibly into the cavity using a Kornberg holder (22). The sample temperature was maintained at 110 K using a flow-through cryostat. The microwave frequency was measured using a Bruker EIP 548B frequency counter, and the magnetic field was calibrated with an ␣,␣Ј-diphenyl-betapicryl hydrazyl sample. The exact peptide concentrations were determined by amino acid analysis of total copper by inductively coupled plasma mass spectrometry. EPR of peptide/spin-labeled LUV was carried out as above, except that the temperature was kept at 305 K, above the liquid crystal/gel transition temperature of the lipids. Analysis of EPR spectra of peptide/lipid mixtures was carried out using the spectral subtraction and addition methods described by Marsh (25). To check the validity of these procedures, the lipid spin label spectra were simulated using modified Bloch equations, as described in model I of Davoust and Devaux (26).
CD Spectroscopy-The CD spectra of peptides in the LUV were obtained using a CD spectropolarimeter model 62DS (AVIV, Lakewood, NJ). The peptide concentrations were determined using the molar extinction of the UV absorption from the tyrosine residue. CD was obtained for each solution both neat and after a 1:4 dilution with MilliQ® water. CD spectra were obtained using a 1-mm-pathlength quartz cell, acquired at 297 K in 0.5-nm steps over a 185-250-nm-wavelength range. The base line acquired in the absence of peptide was subtracted, and the resulting spectra were smoothed and analyzed for the percentages of ␣-helix, ␤-strand, and disordered structures using the K2d Kohonen neural network program (27,28).

RESULTS
NMR Spectroscopy-Sedimentation equilibrium measurements in aqueous solution at pH 6.9 (100 mM NaCl, 50 mM phosphate, PBS) indicated that under the conditions used for our NMR experiments, all peptides were monomeric (Table I). Tseng et al. (29) also report that synthetic A␤40 freshly dissolved in aqueous solution is monomeric. In aqueous solution, there was little chemical shift difference between the backbone amide and C ␣ H resonances of A␤28 (A␤1-28) and those of the corresponding residues of A␤40 (A␤1-40), suggesting that both peptides were in a similar conformation. Indeed, A␤28 and A␤40 backbone chemical shifts deviated little from random coil values. This, coupled with the lack of both inter-and intraresidue nuclear Overhauser enhancement connectivities in the nuclear Overhauser enhancement spectroscopy spectra, indi-cated that both peptides were undergoing significant conformational exchange in aqueous solution. NMR spectra were also recorded for A␤28 where the N ⑀2 nitrogens of the imidazole ring of the His residues 6, 13, and 14 were methylated, hereafter referred to as Me-A␤28. This peptide (purchased from Auspep) was prepared by incorporating histidine residues that were already methylated at the N ⑀2 nitrogens of the imidazole ring into the synthesis of A␤28, and its identity was verified by mass spectrometry and NMR. The spectra of Me-A␤28 were virtually identical to A␤28, the only significant differences being three strong singlets in the 1 H spectrum at 3.80, 3.82, and 3.83 ppm from the methyl groups attached to the His imidazole rings.
Metal Binding-When Zn 2ϩ was added to the solutions of A␤28 or A␤40 in PBS, a precipitate formed. NMR spectra of the supernatant of A␤28 treated with Zn 2ϩ showed that peaks assigned to C2H and C4H of His 6 , His 13 , and His 14 of A␤28 had broadened significantly. However, there was little or no change in the rest of the spectrum compared with A␤28 prior to the addition of Zn 2ϩ (Fig. 1). This broadening of the NMR peaks caused by histidine residues is the result of the interaction of these residues with Zn 2ϩ . The histidyl side chain is a well established ligand of zinc in proteins and peptides (30), and this result suggested that three of the ligands bound to Zn 2ϩ were most likely the imidazole rings of the histidine residues. The broadening of these peaks is the result of chemical exchange between free and metal-bound states or among different metal-bound states. The broadening of peaks is indicative of intermediate exchange that on the NMR time scale suggests that the metal binding affinity is in the micromolar range, in agreement with the low affinity site described by Bush et al. (31). The absence of any change in the rest of the spectrum suggests that the metal-bound form of the peptide is monomeric and that there is little or no significant amount of soluble oligomer in solution, because higher order aggregates would result in significantly broadened resonances. When Cu 2ϩ or Fe 3ϩ was titrated into an aqueous solution of A␤28, similar changes were observed in the 1 H spectrum, with the peaks assigned to the C2H and C4H of His 6 , His 13 , and His 14 disappearing from the spectrum. A slight broadening of all peaks in the spectrum (associated with the paramagnetism of Cu 2ϩ and Fe 3ϩ ) was also observed, but there were no other major changes following the addition of Cu 2ϩ or Fe 3ϩ . The metal-induced precipitation prevented the addition of enough metal to saturate the metal-binding site. Addition of Zn 2ϩ or Cu 2ϩ to an aqueous solution of A␤40 (0.3 mM) at pH 6.9 caused the formation of large amounts of precipitate, as previously observed (13,14,31). The precipitate made the observation of NMR spectra problematic, and few conclusions could be drawn from spectra of the peptide that remained in solution. When Cu 2ϩ was added to an aqueous solution of Me-A␤28, the changes observed in the spectrum were identical to those observed for Cu 2ϩ added to A␤28, but there was no visible precipitate.
Copper ions also induced aggregation in the rat A␤28 peptide in aqueous solution. Rat A␤28 differs from human A␤ by three substitutions, with Arg 5 , Tyr 10 , and His 13 of human A␤ becoming Gly 5 , Phe 10 , and Arg 13 (32). In vitro it has been shown that, compared with human A␤, rat A␤ binds Zn 2ϩ and Cu 2ϩ less avidly (14,33), that the coordination of Cu 2ϩ or Fe 3ϩ does not induce redox chemical reactions, and that limited reactive oxygen species are generated (11). 1 H NMR spectra of the supernatant showed that peaks from His 6 and His 14 had broadened beyond detection, indicating that these residues were involved in copper binding. Apart from some general broadening associated with paramagnetic Cu 2ϩ , no other significant changes were observed in the spectra.
We next studied metal binding to A␤28 and A␤40 in SDSmicelle solution to investigate the metal binding properties of A␤ in membrane-mimicking environments. Chemical shift differences between A␤28 and A␤40 were very small, again suggesting that both peptides adopt similar conformations in solution. In SDS-micelle solution we determined that A␤ adopted a well defined ␣-helical structure from residue 15 to the C terminus. For A␤40 the helix was kinked near residue 31. These results are similar to those recently described (34 -36).
At pH Ͻ6.5 no interaction was observed between Zn 2ϩ and A␤ in SDS-micelles. The addition of Zn 2ϩ to A␤40 in SDSmicelles at pH 6.5 broadened resonances caused by C2H and C4H of all three histidine residues such that they were not observed, indicating that the zinc was in exchange with these residues, but precipitation was not observed. Addition of more Zn 2ϩ (ϳ10-fold) in an attempt to saturate binding did not make the resonances observable. Raising the pH to 7 sharpened the resonances slightly, although they were still broad, suggesting that Zn 2ϩ binding was slightly stronger at this pH. Raising the pH further did not measurably increase the affinity for Zn 2ϩ by the peptide. Further addition of large quantities of Zn 2ϩ (up to 200 equivalents) failed to produce sharper resonances attributable to the histidine residues, suggesting that binding was not saturated and is weak under these conditions. When Cu 2ϩ was added to A␤40 in SDS-micelles at pH 5.5, resonances caused by residues in the 6 -14 region were broadened as a result of their proximity to bound paramagnetic copper. This region of the peptide contains the three histidine residues previously implicated in copper binding, and peaks attributable to the side chains of these residues disappear completely from the spectrum. This region also contains residues Asp 7 , Tyr 10 , and Glu 11 that could act as ligands for Cu 2ϩ , and peaks caused by the side chains of these residues were also broadened beyond detection. The rest of the spectrum was largely unaffected, suggesting that there is no major conformational change by the peptide in SDS-micelles upon copper binding and that the helical C terminus is unaffected. Precipitation was not observed when Cu 2ϩ bound to A␤40 in SDS-micelle solution.
EPR Spectroscopy: Cu 2ϩ Studies-The EPR spectra of Cu 2ϩ complexed with A␤28 peptide in PBS over a metal/peptide molar fraction range of 0.2-1.0 are shown in Fig. 2. By the criteria of Peisach and Blumberg (37), the Aʈ (15.9 millikaisers) and gʈ (2.295) values for spectrum 2A suggest a square planar configuration for Cu 2ϩ with, most probably, a 3N1O coordination sphere about the metal. Spectra of Cu 2ϩ complexed at a 0.2 molar fraction with A␤42 peptide in 35 mM SDS in PBS were identical for those of A␤28 in PBS at the same Cu 2ϩ molar fraction. On the other hand, the Aʈ (15.8 millikaisers) and gʈ (2.341) for the rat A␤28 peptide in PBS, which has one histidine less, fell clearly within the criteria for a 2N2O coordination sphere. The concentration of Cu 2ϩ in each sample was determined by double integration of the spectra of the complexes and of a pure standard solution of CuCl 2 . The concentration of Cu 2ϩ determined in this way in each sample was within 95% of the total copper determined by inductively coupled plasma mass spectrometry, suggesting that, within the limits of experimental error, reduction of Cu 2ϩ to the EPR-silent Cu ϩ had not occurred in the presence of A␤28 or rat A␤28, in agreement with the findings of Huang et al. (11), nor did reduction occur for A␤42 bound to Cu 2ϩ in SDS-micelles.
When the Cu 2ϩ /peptide molar fraction was increased above 0.2, line broadening was observed with A␤28 that was attributed to Heisenberg exchange effects brought about by the metal beginning to occupy sites adjacent to those initially occupied at lower molar ratios. As can be seen from Fig. 2, the broadening was more pronounced with the increasing molar fraction until the spectrum was apparently completely broadened at 1.0 mole equivalent of Cu 2ϩ , giving a line identical to that found by Ohtsu et al. (38) for the exchange broadened spectrum in Cu 2ϩ -bridged imidazolate complexes. The amount of broadened line in each spectrum was determined by subtracting from it the spectrum obtained at the 0.1 molar fraction until the remaining line was identical to that of spectrum I in Peaks caused by the C2H (marked with an asterisk) and C4H (marked with #) of histidines 6, 13, and 14 have been broadened because of binding to the zinc (approximately micromolar affinity). The rest of the spectrum was unaffected by zinc binding. Similar spectra were observed when Cu 2ϩ or Fe 3ϩ was added to A␤28.

FIG. 2. X-band EPR spectra of increasing concentrations of
Cu 2؉ added to A␤28 in PBS at pH 6.9. 0.2 mole fraction Cu 2ϩ (Spectrum A) was increased by 0.1 mole fraction steps (spectra B-I) to 1 mole fraction. Frequency was 9.485 GHz, modulation was 50 KHz, and power 2mW. Temperature was 110 K. Fig. 2. The integral of each broadened Cu 2ϩ line was plotted against added Cu 2ϩ (Fig. 3, curve A) to give a nonlinear curve, suggesting that metal binding to the exchange-broadened sites was cooperative in nature. A Hill coefficient of 5.9 suggested that copper binding to A␤ exhibits strong positive cooperativity. When the same experiment was repeated in the presence of excess Zn 2ϩ (molar fraction 4.0), the exchange-broadened curve shifted to the right (Fig. 3, curve B) and concomitantly decreased the amount of maximal exchanged Cu 2ϩ by 50%. This indicates that the Zn 2ϩ selectively competed for the Cu 2ϩbinding sites but that the Cu 2ϩ binding remained cooperative. When Cu 2ϩ was added to Me-A␤28 peptide, no Heisenberg exchange broadening was observed in the EPR spectra up to a 0.9 molar fraction, indicating that the peptide was not forming multimers. It appears that metal ions have to coordinate both nitrogens on the imidazole ring of the His residues before aggregation can occur.
EPR Studies of the Interaction of A␤ Peptide with Spinlabeled LUV- Fig. 4 shows spectra of 1-palmitoyl-2-(16-doxyl stearoyl) phosphatidyl serine in the negatively charged LUV in the absence (spectrum A) and the presence (spectrum B) of a 0.05 mole fraction of A␤42, coordinated by 0.3 mole equivalents of Cu 2ϩ . Spectrum C was obtained by subtracting the broad Cu 2ϩ line from spectrum B and shows that there is a relatively immobilized component in the spectra. Increasing amounts of spectrum A of the control LUV preparation were subtracted from spectrum C until a spectrum with a clear end point was obtained (Fig. 4, spectrum D). Attempts were made to determine whether the presence of the peptide extensively perturbed the bulk lipid, as suggested by McIntyre et al. (39), by subtracting the spectrum of the control LUV recorded at 300 K. This assumes that the lower temperature spectrum would simulate that of bulk lipid, the motion of which had been restricted by long range effects arising from the presence of the peptide. Using the lower temperature spectrum did not lead to a clear end point, and it was impossible to obtain a well behaved first integral with no negative excursions below the base line (40). It was concluded, therefore, that long range effects of the type described by McIntyre et al. (39) were absent in our system. It was also possible to simulate the spectrum as described by Davoust and Devaux (26). The proportion of the slow motion component in spectrum C was therefore calculated by double integration of spectrum D. Spectra were then run over a range of Cu 2ϩ A␤/lipid ratios, from 0.025 to 0.15 mole fraction, and the proportion of slow motion component at each mole fraction is plotted in Fig. 4 (bottom panel). The relationship between the mole fraction and proportion of slow component was linear, suggesting that even at a fraction of 15%, all of the peptide was associated with the lipid and that the aggregation of the peptide did not increase with the increasing mole fraction. The fact that this ratio did not change when peptide:lipid ratios or peptide:metal ratios changed suggested that the structure that penetrated the lipid membrane was well defined. The lipid: peptide ratio of ϳ4:1 is much lower than the value of 10:1 usually associated with a single ␣-helix penetrating the hydrophobic region of the membrane yet higher than the expected ratio for ␤-sheet conformation, which is 1-2 lipids/strand, depending on the tilt of the structure in the bilayer (41).
The Cu 2ϩ A␤42 peptide/negatively charged phospholipid mixtures were examined at 110 K to determine the nature of the coordination sphere about the copper bound to A␤. They gave spectra characteristic of those illustrated in Fig. 2, except that a small amount of broadening was evident at the high field end of the line caused by the contribution of frozen phospholipid spin label. The copper spectra taken at a 0.3 mole fraction of Cu 2ϩ showed no trace of exchange broadening, but at Cu 2ϩ mole fractions above 0.5 there was exchange broadening characteristic of the spectra shown in Fig. 2. Measurement of the parameters of the copper EPR signal at a 0.3 mole fraction showed that the coordination sphere about the copper ions in each case was 3N1O.
A similar relatively immobilized component was found when

FIG. 4. X-band EPR spectra of negatively charged LUV made from 50% POPS and 50% POPC containing the negatively charged spin probe 1-palmitoyl-2-(16-doxyl stearoyl) phosphatidyl serine (probe/lipid 1/300) (spectrum A) and X-band spectra of system A at the same temperature after the addition of Cu 2؉ A␤42 (peptide/lipid 1/50) (spectrum B).
The curved base line of spectrum B is due to the broad copper resonance at high temperature (296 K). Spectrum C, spectrum B with the broad Cu 2ϩ line subtracted, showing a shoulder to the left of the low field line. This is typical of peptide penetration into the bilayer core. Spectrum D, difference spectrum obtained when spectrum A is subtracted from spectrum C. This spectrum represents the motionally restricted lipid in the boundary layer of the peptide. All spectra were recorded at 305 K. The bottom panel is plot of integrals of spectrum C, giving the percentage of boundary lipid obtained at increasing concentrations of added Cu 2ϩ bound A␤42.
Zn 2ϩ coordinated to A␤42 was added to the negatively charged LUV, except that much higher concentrations of metal (up to a molar ratio of 4) were required. This reflects the lower affinity of zinc for the metal-binding sites of A␤ (31,42,43), suggesting that the metal is playing a structural role and that redox chemistry is not involved in this process.
These experiments were repeated in the presence of the chelating agent diethylenetriaminepentaacetic acid at a 2:1 molar ratio relative to Cu 2ϩ . No sign of peptide penetration of the bilayer was then evident in the EPR spectra, establishing that A␤42 penetration of the membrane was a consequence of metal binding. No bilayer penetration was observed over a range of mole fractions from 0.025 to 0.1 for Cu 2ϩ A␤28. Similarly, no penetration over this mole fraction range was observed with the zwitterionic phospholipid DOPC LUV and Cu 2ϩ A␤42 probed with 1-palmitoyl-2-(16-doxyl stearoyl) phosphatidyl choline, indicating that A␤ penetration only occurs with negatively charged membranes.
CD Spectroscopy-CD spectra were obtained on freshly prepared solutions of A␤42 with one molar equivalent of CuCl 2 , in the presence of negatively charged LUV so that the peptide: LUV mole fraction was 0.15. Spectra were also obtained of 0.15 mole fraction A␤42 in 1:1 DOPC:DPPS LUV and 1 mM DTPA. In the absence of Cu 2ϩ in a membrane environment (Fig. 5, curve A), A␤42 was mainly ␤-sheet, as indicated by the minimum at ϳ220 nm. There is a small inflection in the spectrum at 228 nm. This has been observed previously in CD spectra of lipid-bound peptides and was believed an artifact caused by anomalous light scattering. The structures of these peptides were confirmed as ␤-sheets by Fourier transform infrared spectroscopy in the same lipid (44).
A␤42 in the presence of Cu 2ϩ was significantly more ␣-helical, as indicated by a double minimum at 208 and 222 nm (Fig.  5, curve B). The helix content was calculated to be 57% for the Cu 2ϩ -containing system and 8% in the absence of the metal, indicating that it is involved in converting A␤42 from predominantly ␤-sheet to predominantly ␣-helix in this membranemimetic environment. This is consistent with Cu 2ϩ promoting A␤42 membrane insertion as helical multimers.
Reduction of Cu 2ϩ in the Presence of Phospholipids-Double integration of the copper EPR spectrum of aliquots taken at 30-s intervals of the Cu 2ϩ A␤1-42 peptide/acidic phospholipid mixtures at 110 K determined the amount of the Cu 2ϩ in each sample. We assumed that the diminution in the EPR signal was due to Cu 2ϩ reduction to Cu ϩ and not due to the formation of antiferromagnetically coupled (S ϭ 0) dicopper because it has already been shown by other assays that A␤42 reduces Cu 2ϩ (11,12). As shown in Fig. 6A, there was a 25% reduction of Cu 2ϩ to Cu ϩ by A␤42 in the presence of acidic phospholipid, as compared with Ͻ10% for the peptide in SDS and almost 100% reduction in aqueous buffer. In neutral lipids there was almost 90% reduction of Cu 2ϩ .
Role of Methionine in the Reduction of Cu 2ϩ by A␤-The decrease in the reduction of Cu 2ϩ to Cu ϩ by A␤42 in the presence of acidic phospholipids and SDS might have been due to the seclusion of the hydrophobic Met 35 residue in the hydrocarbon core lipid bilayer of the LUV or the interior of the SDS-micelles. Therefore, we studied the effect of adding soluble methionine to the metal-peptide mixtures. Cu 2ϩ A␤28 was also studied, because it lacks the hydrophobic region containing the Met residue. As shown in Fig. 6B, adding methionine (mole fraction 2.0) to the metal-peptides dramatically increased the rate of reduction of Cu 2ϩ in each case, supporting a critical role for Met 35 in the redox activity of A␤.

DISCUSSION
The Structure of the Zn 2ϩ and Cu 2ϩ -binding Sites on A␤ Resembles Superoxide Dismutase 1-NMR and EPR evidence (Figs. 1 and 2) revealed that the Zn 2ϩ and Cu 2ϩ coordinate to the three His residues of A␤28. Zinc and copper are normally tetravalent, but the NMR spectra give no definitive indication as to what the fourth ligand might be. The EPR spectra of human A␤28 and A␤42 plus Cu 2ϩ suggest an oxygen ligand that could come from one of the carboxyl or hydroxyl side chains, water/hydroxo, or phosphate from the buffer. The evidence of Miura et al. (45) suggests that for copper the oxygen ligand is the hydroxyl group from the side chain of Tyr 10 . The fourth ligand for Zn 2ϩ has not been identified, but the lack of changes observed in the rest of the NMR spectrum upon the addition of Zn 2ϩ favor the water/hydroxo or phosphate options. Fig. 7A shows a model of the initial and Cu 2ϩ -binding sites of A␤ incorporating His 6 , His 13 , His 14 , and Tyr 10 .
The NMR evidence indicates that initial metal binding does not cause a significant structural change in the peptide, indicating that metal-induced aggregation is not mediated by metal-induced conformational changes. One possibility is that the metal could bridge between histidine residues of different peptides as proposed by Miura et al. (45). Another possibility involves histidine ligands acting as bridges between metal centers on different peptides. When a metal ion binds to the N ␦1 of a histidine residue, it reduces the pK a of N ⑀2 NH, making this nitrogen more available for metal coordination (46). This results in a histidine residue that is capable of bridging metal ions, the best known example being His 63 at the active site of superoxide dismutase (SOD) (47). Similar bridging histidine residues have been proposed within the octarepeat region of the prion protein when this protein binds Cu 2ϩ (48). Miura et al. (45) reported that complexes containing bridging histidine residues were formed between Zn 2ϩ and A␤ but were not observed for Cu 2ϩ and A␤. It is important to note that the Heisenberg exchange broadening observed at Cu 2ϩ mole fractions Ͼ0.3 with A␤ was eliminated when the N ⑀2 of three imidazole rings of the His residues were methylated, thereby preventing the formation of histidine-bridged complexes. This evidence provides definitive proof that the imidazole rings of the His residues of A␤ are able to bridge between metal centers under the conditions studied here. Fig. 7B shows how the histidine could act as a bridge between metal centers; the distance between copper ions in this model is 6 Å, and exchange phenomena would be evident in the EPR of such a complex, matching those observed (Fig. 2). It is the bridging histidine that is probably responsible for the reversible metal-induced aggregation that is observed when A␤ is metallated with Cu 2ϩ and Zn 2ϩ . Bridging histidine residues would also explain the multiple metal-binding sites observed for each peptide (43) and the high degree of cooperativity evident for subsequent metal binding (Fig. 3). With three histidines bound to the metal center there are several potential sites for further coordination of metal ions such that a large scope exists for metal-mediated cross-linking of the peptides leading to aggregation, which will be reversible when the metal is removed by chelation. This type of metal binding with bridging histidine residues would result in complexes very similar to the active site of SOD. The occurrence of exchange broadening in the Cu 2ϩ EPR spectra of A␤28 and its modulation by Zn 2ϩ , which has also been observed in model SOD imidazolate-bridged dinuclear complexes (38), further suggests the occurrence of structured complexes that are not merely random aggregates of peptide.
Cu 2ϩ and Zn 2ϩ Induce A␤ to Form Allosterically Ordered Multimers That Penetrate Lipid Membranes-Zn 2ϩ and Cu 2ϩ bound to the same site on A␤ in SDS-micelles as in aqueous solution, although there were significant differences in the effects of metal binding. Metal-induced aggregation and reduction of Cu 2ϩ to Cu ϩ , as observed for A␤ in aqueous solution (11,12,14), were not observed when copper bound to A␤40/42 in SDS-micelles or LUVs. One of the effects of SDS is to drive the C-terminal part of the peptide (residues 15-42) into an ␣-helical conformation, preventing the formation of aggregated ␤-sheet structures. The ␣-helical region of the peptide starts just after the metal-binding site. This result suggests that the peptide may need to be in a ␤-sheet form before it becomes redox-active, which would be consistent with previous observations that the neurotoxicity associated with A␤ also requires the presence of aggregated ␤-sheet structures (49 -51). However, results with the addition of exogenous methionine that are discussed below show that ␤-sheet/aggregation are not necessary for Cu/A␤ to produce redox-active species.
We found that as a consequence of Zn 2ϩ and Cu 2ϩ binding, A␤42 forms allosterically ordered, ␣-helical structures that penetrate negatively charged membranes. These results suggest that the A␤42-and A␤43-induced decrease in fluidity in the hydrocarbon core of human cortex membranes as measured by fluorescence polarization (52) was most likely due to association of the membrane lipids with the hydrophobic face of the rigid peptide. There have also been reports of cation channel formation (53,54) by A␤40 and A␤42 that indicate penetration of the bilayer. Many previous studies have indicated that A␤ peptides form ␤-sheet structures in lipid environments (55)(56)(57), although the structure appears to be dependent on the composition of the lipid. For example, it was shown that A␤1-40 was A, the initial coordination site of Cu 2ϩ on A␤ in solution (His 6 , His 13 , His 14 , and Tyr 10 ) as determined by NMR and EPR spectroscopy. B, a proposed model explaining the aggregation, cooperative binding, and redox properties of metal bound A␤ peptides. The imidazole ring of His 6 is shown forming a bridge between copper atoms to form a dimeric species; this residue is used as an example, and other histidine residues could form similar bridges and therefore lead to aggregation. The coordination sphere about the metal ions is similar to that observed in the active site of SOD. The models were generated by binding Cu 2ϩ to the appropriate residues of A␤28 with the coordination sphere about the copper atoms constrained to a square planar configuration; the rest of the molecule was not constrained. Structures were then energy minimized in vacuo with a distance-dependent dielectric using the esff force field within the Discover module of Insight98.
␣-helical in phosphatidyl glycerol vesicles but adopted a ␤-sheet conformation when GM1 ganglioside was added (58). Further, the physical state of the lipid systems used varied widely, with small unilamellar vesicles being the most common. In many cases, the temperature at which the measurements were made was not reported, leaving some doubt as to whether the lipids were in the liquid crystal or gel state. It is highly probable that the A␤ peptides are pleiomorphic, able to adopt different conformations in different lipid environments. Because the cell membrane is a mosaic of different lipid environments, it is even possible that the peptides will exhibit different structures with different properties in different parts of the mosaic.
We found that approximately four lipids are associated with each membrane-penetrating peptide subunit (Fig. 4, bottom  panel), which is a low ratio for lipid association with a single ␣-helix and suggests the presence of oligomers penetrating the lipid. Preliminary modeling showed that 24 lipid molecules would fit around an oligomer of six helices. Taken with the value of 5.9 for the Hill coefficient calculated from the data in Fig. 3 (curve A), the modeling suggests that A␤42 forms an allosterically induced hexamer in the presence of metal ions. The metal ions do not penetrate the membrane but form structures very similar to the active site of SOD on the surface of the membrane.
Earlier studies of A␤ peptide-membrane interaction were not controlled for the presence of metal ions, ubiquitous trace contaminants, the concentration of which was minimized in our controls by the use of excess chelator. It is possible, therefore, that the other reported results of A␤ penetration into the membrane, measured by sensitive methods, might have been initiated by the peptide binding trace Cu 2ϩ or Zn 2ϩ .
The failure of Cu 2ϩ :A␤28 to perturb the vesicle bilayer shows that, as is the case with the redox activity of the peptide (11,12), the hydrophobic C terminus is essential for the interaction. The absence of line broadening, which is characteristic of Heisenberg exchange between free radicals and transition metals (59), in the phospholipid nitroxide label spectrum in the LUV interacting with the Cu 2ϩ /A␤42 also indicates that the Cu 2ϩ -binding site is outside the bilayer. The role of Cu 2ϩ in the interaction may indicate that the peptide must be oligomerized for it to penetrate the bilayer. It has been suggested recently (60) that aggregation of A␤ peptides might be a prerequisite for penetration of the lipid bilayer.
The Role of Methionine in A␤ Redox Reactions-It has been reported that Met 35 is essential for the toxicity and induced oxidative stress of A␤ (61,62). The lack of redox activity associated with the copper interactions with A␤28 suggested that the Met of A␤42 incubated in SDS or negatively charged LUV was buried in the hydrophobic core of the micelles and vesicle bilayers and therefore unavailable as a cofactor in the metal reduction. Addition of exogenous methionine restored redox activity, albeit with a much slower rate of reaction. These results suggest that Met 35 participates in the reduction of copper by A␤. Although A␤28 and A␤42 in LUV with bound copper have multiple copper-binding sites resulting in peptide aggregation, A␤42 in SDS is monomeric in the presence of bound Cu 2ϩ with a single metal-binding site as determined by the lack of Heisenberg interactions in the EPR spectra. The coordination sphere of Cu 2ϩ under these conditions includes the three histidine residues and an oxygen ligand, probably Tyr 10 , and this is the initial copper-binding site of A␤. The effects of the addition of methionine (Fig. 6) showed that this is potentially a redox-active site.
Pathological versus Functional Metal Binding to A␤: a Model for Alzheimer's Disease-We have shown previously that Cu 2ϩ and Zn 2ϩ binding to A␤ modulate the toxicity of the peptide through the generation of H 2 O 2 by electron transfer to O 2 (9,12). Cu 2ϩ and Zn 2ϩ binding to A␤ also induces the precipitation of the peptide (13,14,31,42). Increased binding of these metals to A␤ is evident in AD (18), and we have found recently that the metal-mediated redox activity and aggregation of A␤, as well as amyloid deposition in APP2576 transgenic mice, are inhibited by treatment with a bioavailable zinc/copper-selective chelator, clioquinol (19). Therefore, the structural characterization of the Cu 2ϩ -and Zn 2ϩ -binding sites on A␤ may be essential for elucidating the pathogenesis of AD, as well as for developing new therapeutics.
In light of our current findings that Cu 2ϩ and Zn 2ϩ binding induce monomeric A␤ to form allosterically ordered SOD-like metallopeptide complexes that insert into negatively charged membranes, it is interesting to speculate on the likelihood that this redox-inert A␤ assembly may be biologically relevant and subsume some function. We have recently reported that Cu 2ϩ / Zn 2ϩ -bound A␤ possesses significant SOD catalytic activity (69). Our current structural data support the possibility of such SOD-like activity. Physiologically, the combination of A␤ with a lipid vesicle occurs in high density lipoproteins (HDLs), which are found in plasma and cerebral spinal fluid (63,64). HDLs possess antioxidant properties (65)(66)(67), and we hypothesize that copper/zinc-A␤ complexes inserted into HDL membranes may play a role in superoxide clearance by HDLs. Supporting a role for HDLs in nullifying the adverse redox activity of soluble A␤, HDLs are able to decrease A␤ toxicity in cortical cell culture (68). In the heuristic model that we contemplate, the perturbation of Zn 2ϩ or Cu 2ϩ homeostasis that is associated with AD (15) may interfere with A␤ insertion into membranes, liberating increased amounts of neurotoxic, redox-active A␤.