Scanning Cysteine Mutagenesis Analysis of Aβ-(1-40) Amyloid Fibrils*

We describe here the use of cysteine substitution mutants in the Alzheimer disease amyloid plaque peptide Aβ-(1-40) to probe amyloid fibril structure and stabilization. In one approach, amyloid fibrils were grown from Cys mutant peptides under reducing conditions and then challenged with an alkylating agent to probe solvent accessibility of different residues in the fibril. In another approach, monomeric Cys mutants, either in the thiol form or modified with iodoacetic acid or methyl iodide, were grown into amyloid fibrils, and the equilibrium position at the end of the amyloid formation reaction was quantified by determining the concentration of monomeric Aβ. The ΔG values of fibril elongation obtained were then compared in order to provide information on the environment of each residue side chain in the fibril. In general, Cys residues in the N and C termini of Aβ-(1-40) were not only accessible to alkylation in the fibril state but also, when modified in the monomeric state, did not greatly impact fibril stability; these observations were consistent with previous indications that these portions of the peptide are not part of the amyloid core. In contrast, residues 16-19 and 31-34 were not only uniformly inaccessible to alkylation in the fibril state, but their modification with the negatively charged carboxymethyl group in monomeric Aβ also destabilized fibril elongation, confirming other data showing that these segments are likely packed into a hydrophobic amyloid core. Residues 20, 30, and 35, flanking these implicated β-sandwich regions, are accessible to alkylation in the fibril indicating a location in solvent exposed structure.

Amyloid fibrils and other non-native protein aggregates are now regarded as an important alternate universe of protein structures formed by normal proteins exposed to conditions of environmental or mutational stress. Fibrils, protofibrils, and other aggregates are associated with a variety of serious human diseases of the brain (1) and periphery (2). Aggregate formation has also long been recognized as a major technical problem in the industrial production and use of proteins (3)(4)(5)(6). The intrinsic tendency of polypeptide polymers to generate offpathway aggregates can be viewed as a design flaw in the structural biology of the cell. Since up to 50% of newly synthesized protein chains are aggregated and/or misfolded (7), it is not surprising that major biochemical pathways have evolved in cells and organisms to manage the misfolding and aggregation processes, as exemplified by the chaperone (8), ubiquitin-proteasome (9,10), aggresome (11), and autophagy (12) systems. In at least a few cases, nature has exploited the ability of polypeptide chains to form amyloid by evolving specific polypeptide sequences and the required cellular machinery for developing functional amyloids, for example as a means of cell attachment (13) or gene regulation (14).
Given the growing importance of non-native protein aggregates, it is desirable to gain an increased knowledge of their structures and how these structures might differ from those of typical globular proteins. Protein aggregates are not amenable to standard methods of structure determination, however, being too large and heterodisperse for solution phase NMR or x-ray crystallography. The inability of conventional methods to extract many details of aggregate structure has led to a search for alternative approaches. Some of these methods, such as those described here, prove to be derivatives of methods previously used to dissect the structures of globular proteins before structure determination by x-ray crystallography and solution phase NMR became routine.
The non-native aggregate most amenable to structural analysis is the amyloid fibril, and the most widely studied amyloid fibrils are those of the Alzheimer plaque peptide A␤, 2 in particular the 1-40 form of the peptide. Like other amyloid fibrils, the A␤-(1-40) fibril consists of a bundle of protofilaments (15). Like other amyloid fibrils, A␤-(1-40) fibrils exhibit a cross-␤ pattern in x-ray fiber diffraction (15), indicating a ␤-sheet-rich structure in which the ␤-extended chains are displayed perpendicular to the fibril/protofilament axis with the ␤-sheet H-bonds between these extended chains parallel to the fibril axis (16). Although some amyloid fibrils, including fibrils of some A␤ fragments, appear to consist of an anti-parallel ␤-sheet arrangement (17), fibrils of the 1-40 peptide are in parallel sheet, with the chains in-register (18 -20), as indicated in the structural models shown in Fig. 1. Studies by a variety of techniques have provided additional detail about A␤-(1-40) amyloid fibril/protofilament structure and energetics, focusing on the questions of how the A␤ peptide folds to engage that structure and which residues are involved in the H-bonded core. These include EPR of derivatized Cys mutants (20), solid state NMR (18,19), limited proteolysis (21), chemical accessibility of wild type residues (22), hydrogen-deuterium exchange (23)(24)(25)(26)(27)(28), and proline (26), alanine, 3 and disulfide (30) mutagenesis.
The problem of structural analysis of amyloid fibrils has been made even more complicated by the revelations of recent years that single amino acid sequences can form multiple conformational forms of amyloid fibrils (31)(32)(33). The structural bases of these conformational differences remain to be elaborated in full. Interestingly, perhaps the best characterized example of conformational variation is in amyloid fibrils of A␤- . Fibrils grown under conditions of stirring ("agitated") differ from fibrils grown without stirring ("quiescent"), both in their intramolecular architecture and in their EM morphology (33). It is * This work was supported by Grant AG 18416 from the National Institutes of Health. The therefore likely that some differences in published structural data on A␤-(1-40) fibrils, such as among the studies cited above, can be attributed to real structural differences in the fibrils due to variations in fibril growth conditions. We describe here several uses of cysteine mutants of the A␤-(1-40) sequence, on the one hand exploiting the unique chemical reactivity of the Cys sulfhydryl group to assess side chain accessibility in the fibril and on the other hand taking advantage of Cys reactivity to create multiple chemical mutants of A␤-(1-40) that can be assessed for their relative abilities to make fibrils. The data provided new, unique information about the solvent accessibility of various residue positions in A␤-  amyloid fibril grown under the physiological conditions of PBS buffer, 37°C, and the absence of stirring or shaking. The data also provided additional information on the chemical environments within the fibril for the amino acid side chains at most residue positions. The data allowed us to further refine our understanding of amyloid structure in general and of one important conformer type of the A␤-(1-40) fibril in particular.

MATERIALS AND METHODS
All the peptides used in this study were purchased from Keck Biotechnology Center at Yale University. WT A␤-(1-40) (for amino acid sequence, see Fig. 1) was obtained as a pure peptide. Single cysteine replacement mutants of A␤-(1-40) (for representation of which mutant peptides were obtained, see Fig. 1) were obtained unpurified. The mutant peptides were purified on Bio-Rad RP-HPLC using a semipreparative C3 column and a water/acetonitrile gradient with 0.05% trifluoroacetic acid. The identity and purity of the final peptides were confirmed by mass spectrometry and analytical RP-HPLC on an Agilent 1100 series mass spectrometry detector. Bond breaker TCEP solution (Tris(2-carboxyethyl)phosphine) was purchased from Pierce. Iodoacetic acid, iodoacetamide, and iodomethane were purchased from Sigma.
Covalent Modification of Cys Sulfhydryl Group at the Monomer Level-The side chain sulfhydryl group of Cys in each of the mutant peptides was modified by treating the monomeric peptide with two different alkylating agents. For modification of the side chain, each purified mutant peptide at a concentration of 1-2 mg/ml was incubated in argon-purged 8 M urea buffer at 37°C with either iodoacetic acid (10 mM; 20 min) or iodomethane (10 mM; 60 min) at pH 8.0. Reaction mixtures were immediately injected onto RP-HPLC and purified. Eluate fractions were pooled based on mass spectrometric analysis. The final purity of each pooled modified peptide was confirmed by analytical HPLC and covalent modification confirmed by electrospray ionization mass spectrometry.
Amyloid Fibril Assembly-Fibril formation was carried out on both side chain-modified and -unmodified cysteine mutants. For the side chain-modified mutants, fibril growth was carried out using the standard protocol described (21,34). Briefly, freshly disaggregated (34,35) peptide was incubated at 37°C in PBS (pH 7.5) containing 0.5% sodium azide (PBSA) together with a seed consisting of 0.1 weight percentage of sonicated WT A␤-(1-40) fibrils. For the unmodified cysteine mutants, the protocol was similar except that the PBSA buffer contained 10 mM TCEP plus 1 mM EDTA and was bubbled with argon gas prior to dissolving the freshly disaggregated peptide, to keep the sulfhydryl group of cysteine in the reduced state during fibril assembly. Fibril formation reactions for all mutants were seeded initially with 0.1 weight percentage of sonicated WT A␤-(1-40) fibrils. (For some mutants, which exhibited a lag time of more than 15 days, we used 1% seed.) For most of the mutants, assembly reactions were carried out at starting concentrations of monomeric peptide of about 50 M. For those peptides that did not aggregate at 50 M concentration, a new fibril assembly reaction was initiated at a higher concentration, in the range of 100 -200 M. Progress of the reaction was monitored by ThT measurement of fibrils (36). When the ThT signal reached a plateau value, reaction progress was further monitored by HPLC, as described previously (21,34), to determine the concentration of A␤-  at equilibrium, the critical concentration (C r ) (34,37). Equilibrium constants for fibril elongation derived from C r values were used to calculate free energies using the relationship ⌬G ϭ ϪRTlnK eq .
Modification of Sulfhydryl Group of Cys at the Fibril Level-Unmodified Cys mutant fibrils grown under reducing conditions were centrifuged at 85,000 rpm (315,000 ϫ g) for 30 min to remove any soluble peptide. The pellet was then resuspended and incubated in the dark for 1 h with 10 mM iodoacetamide in 50 mM Tris buffer at pH 8.0 at room temperature. Fibril products were collected by centrifugation (as above) and washed twice with PBS buffer to remove any dissociated monomer and unreacted alkylating agent. After the last wash, the centrifugation pellet was dissolved in 20% aqueous formic acid, injected on a Zorbax SB-C3 analytical HPLC column linked to an Agilent 1100 electrospray mass spectrometer, and eluted with an acetonitrile-aqueous 0.1% formic acid gradient (formic acid is used instead of the more standard trifluoroacetic acid to improve MS sensitivity). Eluent was passed through a UV detector set at 215 nm and an Agilent 1100 electrospray mass spectrometer. Using A 215 peak area(s) for quantitation, A␤-(1-40) recovery was typically 80% or better based on the amount of A␤ expected for the mass of fibrils used in the experiment (38).
For most analyses, the amount of Cys modification was quantified using the MS data. Using the Agilent software, the mass distribution data across the HPLC elution peak(s) corresponding to unmodified, singly modified, and multiply modified A␤ molecules were deconvoluted to generate the relative abundance of each species underneath the peak envelope. These relative abundances were used to calculate percentage of modification. In some cases, the unmodified and modified A 215 peaks were sufficiently separated in the HPLC that the relative A 215 peak areas could be used to calculate the percentage of modification. Since, however, for many A␤ mutants, the modified and unmodified peaks were not fully separated, it was necessary in these cases to use the deconvoluted MS data to determine the percentage of modification. Although this requires the assumption of equal MS sensitivity for the modified and unmodified peptides, this appears to be a valid assumption since there is good agreement with A 215 peak areas in those cases in which analysis by both means is feasible (38).
Electron Microscopy-Fibrils grown in PBS were adsorbed onto carbon-and Formvar-coated grids and negatively stained with 0.5% (w/v) uranyl acetate solution. The stained grids after washing of the excess uranyl acetate were examined and photographed on a Hitachi-600 EM. Fig. 1 shows the A␤-(1-40) sequence (a) and two published models of the A␤-(1-40) amyloid fibril. The model in b was derived from a combination of scanning proline mutagenesis experiments (26) and a threading approach applied to parallel ␤-helical domains of globular proteins (39). The model in c was derived from solid state NMR measurements (40). Both models posit a basic unit of fibril structure in which the A␤ molecule folds into a loop composed of 2-3 elements of extended chain oriented in such a way that the amino acid side chains in these segments project either into the interior of the loop or out of the loop. In both models, the inward-projecting side chains form a densely packed interior (although perhaps not evident from the depiction in Fig.  1b, the interior of the model, like the interiors of all parallel ␤-helical domains (41,42), is densely filled with side chain mass). Both models also presume that some of the outward-projecting side chains will also be packed into similar, adjacent elements to build up the lateral packing required to form fibrils from protofilaments. Importantly, neither published model addresses the orientation of these extended chains, that is, whether particular amino acid side chains in these segments are inward-or outward-projecting. Our Cys accessibility studies were designed to address this issue, as well as the environments of other A␤ amino acid side chains in the fibril.

RESULTS
We utilized Cys mutants in two ways. First, by growing fibrils from single Cys mutants under reducing conditions, we could use chemical reactivity of the free Cys sulfhydryl group to probe the solvent accessibility of that side chain in the fibril. Second, by modifying peptides at the monomer level, then growing fibrils and assessing the equilibrium positions of the fibril formation reaction, we could determine the compatibility of the modified Cys side chain with the chemical environment within the fibril.
Fibril Growth and Characterization-The ability to reliably grow amyloid fibrils from mutant A␤-(1-40) peptides is a requirement for all the experiments described in this report. All peptides were purified, disaggregated, and subjected to amyloid growth conditions, including seeding with wild type A␤-(1-40) fibrils, as described under "Materials and Methods." Typical kinetics results for free Cys mutants, grown in the presence of the non-alkylatable reducing agent TCEP, are displayed in Fig. 2a, which shows a variety of lag times and final ThT levels.
Mutants with Cys at positions 20, 21, 25-30, and 38 exhibited fibril formation kinetics similar to wild type. Mutants with Cys at positions 24, 31, 32, and 36 showed longer lag times, whereas those at positions 17, 18, 34, and 35 exhibited unusually low ThT signals. The D23C mutant peptide proved to be highly insoluble and therefore could not be used to  grow fibrils. As discussed below, low ThT levels do not necessarily indicate the absence of fibril formation since there are large mutational effects on ThT signals of amyloid fibrils on a weight-normalized basis. The final equilibrium position of each fibril formation reaction was confirmed and quantified by HPLC analysis of the centrifugation supernatant of a reaction aliquot ("Materials and Methods"). The reduced state was confirmed by subsequent HPLC analysis and/or by reactivity with the probe alkylating agent, iodoacetamide.
Representative electron micrographs of aggregation products are shown in Fig. 3. In the work described in this report, we set out to make fibrils from 75 different peptides (25 Cys mutants, each in three chemical states at the Cys residue). All except the Cys-23 mutant could be grown into an aggregate from soluble monomeric peptide. The vast majority (88%) of these aggregated products resembled WT amyloid fibrils, with 64% being indistinguishable from WT fibrils and another 24% being very similar. In addition, three of these peptides (L17C, V18C, and K28C) made filamentous aggregates that are particularly short, and another two peptides (G29C and the carboxymethyl-Cys derivative of A21C) produced aggregates that appeared to be a mixture of fibrils and other structures. Representative images of these different classes of structure are shown in Fig. 3.
Because of their relatively low ThT responses (see below), we were particularly concerned whether the three short filamentous aggregates (L17C, V18C, and K28C) were actually fibrils or perhaps were protofibrils. The word protofibril was introduced to apply to distinctive intermediate structures that appear transiently in the early stages of A␤ amyloid fibril formation under native conditions (43,44). Such A␤ aggregates appear in the EM as short, curvilinear filaments ϳ5 nm in diameter and of variable length, normally less than 200 nm (45)(46)(47)(48), and exhibit low to negligible ThT fluorescence (28,45) and substantially fewer backbone amide protons protected from hydrogen-deuterium exchange as compared with mature fibrils (25,28). Given the short lengths and low ThT responses, we were concerned that some of the aggregates formed by Cys mutants might be protofibrils with arrested development; this would not be surprising since it is known that certain mutations can extend the stability of A␤ protofibrils (49). We applied two tests of these short aggregates to determine whether they were fibrils or protofibrils. First, we measured fibril diameters in the EM fields. Amyloid fibrils tend to have diameters in the range of 8 -12 nm (16), whereas A␤ protofibrils tend to be in the 5-nm diameter range (45)(46)(47)(48). Width ranges (nm) for aggregates of various sequence forms of A␤-(1-40) were: WT, 7.5-14.3; L17C, 8.5-12.9; V18C, 7.5-11.5; K28C, 6.5-9.5. Thus, all of these short aggregates have diameters consistent with a fibril structure.
The other test we applied was hydrogen-deuterium exchange by mass spectrometry (23, 24, 26). Hydrogen-deuterium exchange is a particu- larly sensitive test for the nature of A␤ aggregates and can even quantify the relative levels of fibrils and protofibrils in mixtures of the two. 4 Previously reported values for WT A␤-(1-40) amyloid fibrils exposed to D 2 O for 15-24 h, corrected for rapidly exchanging side chain protons but not for backbone amide protons, are in the range of 11 deuteriums (23,25). In contrast, protofibrils have fewer protected main chain amide hydrogens and thus give higher exchange, in the range of 18 deuteriums (25,28). The stable aggregates described here gave exchanges of: 10.6 Ϯ 0.6 (WT), 9.1 Ϯ 0.8 (L17C), 11.4 Ϯ 0.7 (V18C), and 11.6 Ϯ 0.4 (K28C). Thus, all three short aggregates tested gave results very similar to WT fibrils and very different from WT protofibrils.
Altogether, we think it is clear that, despite the different morphological appearance of these three mutant A␤ aggregates, they are structurally amyloid fibrils and not protofibrils. The reasons for differences in fibril morphology are not clear. In the case of uniform distributions of short fibrils, the explanation could be as simple as a more efficient nucleation mechanism, leading to rapid formation of a greater number of independent growth sites, and consequently, a smaller average size of the fibril product at the end of the reaction.
Solvent Accessibility-To assess the solvent accessibility of the Cys side chain in the amyloid state, Cys mutant fibrils were collected by centrifugation and resuspended in alkylation buffer and then challenged with iodoacetamide. To analyze the product, fibrils were again collected by centrifugation, washed, dissolved, and analyzed by HPLC for recovery of A␤-(1-40) peptide and for the relative amount of peptide in the reduced and alkylated states. The results are summarized in Fig. 4, shown as percentage of the recovered peptide that is alkylated. Fig. 4 shows that, as expected, fibrils comprised of an F4C mutant yield 100% of the recovered A␤ in the alkylated state, whereas Cys at residues 6 and 13 are also significantly modified, consistent with the demonstrated flexibility and solvent accessibility of the N terminus of A␤ in the fibril (20,21). Likewise, consistent with most models of A␤-(1-40) amyloid fibrils, including those shown in Fig. 1, b and c, most of the residues in the segments 16 -21 and 31-36 are completely inaccessible to alkylation in the fibril, suggesting they are buried in packed structure as one would expect to find in the ␤-sheet elements of amyloid. Two residues in these segments, residues 20 and 35, stand out as being accessible in the fibril. The implications of these nuances of the data will be addressed under "Discussion." Other data in Fig. 4 address aspects of different models of A␤-(1-40) fibrils. Residues 24 -27 are all accessible to alkylation within the fibril. This is inconsistent with their being in a ␤-sheet, where one or both faces are inaccessible, and is more consistent with their being in some kind of irregular structure. Residue 38 is also alkylatable, which is inconsistent with it being in an extension of the 31-36 ␤-sheet since it would be expected to be on the buried face of the extended chain but which is consistent with it being in a less ordered, non-H-bonded structure. Residues 29 and 30 are accessible, consistent with their being in a turn. Residue 22 is buried, which is inconsistent with being located in a simple turn. It is interesting that in these amyloid fibrils grown under unstirred conditions, a Cys residue at position 28 is inaccessible to alkylation. This would be expected if this residue were buried in an internal salt bridge. Such an interaction has been characterized within A␤-(1-40) fibrils grown under agitated conditions but apparently is not present in a clearly defined way in quiescent fibrils (33) grown under conditions similar to those used here.
The explanation for partial modification of some side chains is not clear. These partial modifications are not due to insufficient reaction times. In cases in which partial modification was observed, reaction times were extended and later time points were analyzed, but the percentage of modification of the Cys residue changed minimally or not at all. Longer extended reaction times could not be done routinely since incubation for times longer than 1 h typically led to multiple modification of A␤; this is not surprising since iodoacetamide can also react with His, Lys, and Met residues, as well as ␣-amino groups (50). For example, mutants H6C, S26C, N27C, V24C, and I32C, which exhibit a range from 0 to 51% modification after 1 h, were exposed to further reaction time up to 4 h. This led to observation of minimal change in the singly modified peak, as compared with the 1-h result, and formation of an additional mass ion corresponding to doubly modified material at 12, 10, 17, 0, and 0%, respectively, of the total A␤ recovered. Assuming that the doubly labeled A␤, like the singly labeled A␤, is modified at Cys, then the change in total percentage of molecules modified at Cys, by extending reaction time from 1 to 4 h, was: H6C, from 51 to 52%; S26C, from 28 to 32%; N27C, from 38 to 45%; V24C, from 27 to 27%; I32C, from 0 to 0%. Thus, partial modification appeared to be due to multiple environments. Whether these consist of subtle microenvironment differences in a largely uniform fibril preparation, or more major differences indicative of multiple conformers of fibrils, is not clear. With respect to other chemical probes, some fibril preparations previously have been shown to exhibit protein environments that behave uniformly in some parts of the structure and variably in other parts (21,22,51).
Thermodynamic Experiments-The alkylation of the Cys sulfhydryl group is one of the most efficient and highly selective chemical modifications that can be done on polypeptides. A variety of alkylating agents of various chemical structures are available. We chose two of these reagents, iodoacetic acid and methyl iodide, to modify the sulfhydryl groups of the Cys mutant peptides to produce additional A␤ analogs with different chemical characteristics in the side chains. The reaction conditions are mild, and the products are easily purified from unreacted starting material and trace side products by HPLC ("Materials and Methods"). We used these chemically derived mutant peptides, plus the unmodified Cys mutants, to probe the effect of side chain chemistry at various A␤-(1-40) sequence positions on the thermodynamics of fibril formation. Fig. 2, b and c, show the fibril formation reactions of two sets of Cys derivatives, at residues 21 and 34, under standard conditions, as monitored by ThT fluorescence (36). As with most other fibril formation reactions in this work, these were seeded with 0.1% weight percentage  Bars reflect the percentage of A␤ peptide recovered in analytical HPLC as carboxamidomethyl-Cys. All listed residue numbers were tested, except residue 23 (the asterisk indicates that it could not be dissolved in native buffer), so the lack of detectable bar signifies essentially 0% modification and thus no solvent accessibility.
wild type A␤-(1-40) fibrils, to bias growth into the same fibril conformation as the (quiescent) wild type fibrils that are the subjects of other investigations in our laboratory (21,23,26,27,30,(52)(53)(54). Although these reactions differ in both lag times for fibril formation onset and the ThT amplitude at equilibrium, all of these reactions, even those that exhibit negligible ThT signals, involved significant amyloid fibril formation that reaches an equilibrium position giving a critical concentration measurable by quantitative HPLC analysis. The products of these aggregation reactions, even those exhibiting low ThT values, show typical fibril morphologies in the EM (Fig. 3), suggesting that the low ThT values are due to intrinsically low fluorescence yields or low ThT binding to fibrils.
Because at the end point of a fibril formation reaction, the molar concentration of fibrils does not change as monomer is consumed in the elongation reaction, the equilibrium constant for fibril elongation reduces to the inverse of the molar concentration of monomer at equilibrium, or 1/C r (26,37). This allows determination of a free energy describing the elongation equilibrium. Thus, the calculated ⌬G for wild type A␤-(1-40), Ϫ8.6 kcal/mol, is in the Ϫ10 to Ϫ15 kcal/mol range typically found for ⌬G folding values for small globular proteins (29). By determining the corresponding equilibrium positions for mutant A␤ peptides, ⌬⌬G values comparing mutant and wild type have been determined for proline (26), alanine, 3 and disulfide (30) mutants of A␤-(1-40) that provide insights into amyloid structure and dynamics. In the same way, ⌬G values for various Cys mutants of A␤-(1-40) should provide additional structural information. Fig. 5a summarizes the free energies of fibril formation, calculated from C r values as described above, for all the derivatives described in this report. Since they are calculated and plotted with respect to WT A␤ fibrils, these are ⌬⌬G values comparing mutant with WT in each case, with WT being equal to zero. Consistent with a fibril core held together largely by hydrogen bonds, which are not necessarily affected by side chain substitutions, and hydrophobic interactions, which can be, it is not surprising that the iodoacetic acid modification of Cys residues in general seems more destabilizing than the methylation of Cys. Two regions stand out as being especially insensitive to replacement by Cys derivatives: the free N terminus and positions 22 and 23. Replacement of residue 38 with a Cys derivative is only modestly destabilizing, and other residues in the C terminus were not tested, so there are limited data from the Cys mutagenesis to probe the involvement of the C terminus in packed structure. Except for these 2-3 segments, placement of a negatively charged side chain at other residues is uniformly destabilizing to fibril structure.
To present the data in a different way, we recalculated the data shown in Fig. 5a as ⌬⌬G values comparing the stability of the fibrils from an alkylated mutant to the stability of the corresponding free Cys derivative at each residue position. The data (Fig. 5b) show how individual environments within the amyloid fibril tolerate a negatively charged group, as compared with a hydrophobic group, added to the Cys side chain sulfhydryl. In this formalism, as expected for a hydrophobic packed core, negatively charged substitutions at most positions are destabilizing, whereas hydrophobic replacements at the same positions are most often somewhat stabilizing. The most dramatic effect in Fig. 5b is the ability of both Cys modifications at position 22, relative to the free Cys peptide, to enhance the stability of fibrils. Position 23 is not represented in this figure because it was not possible to grow fibrils from the free Cys peptide; however, the ⌬⌬G values for the Cys derivatives at position 23 are similar to the position 22 ones in having essentially no effect on stability relative to WT. These data are consistent with previous assignments of a turn position to these positions based on proline mutagenesis (26,39).
One difficulty with interpreting the data from the modified Cys peptides is that the substitutions are inevitably being tested, not only for the compatibility of their hydrophobic/hydrophilic character but also for their bulk. Although the densely packed regions of the amyloid fibril may have somewhat more flexibility to accommodate additional side chain bulk, as compared with typical globular proteins, there are also presumably limits to fibril plasticity. This might explain situations such as position 33, where both alkylated forms of Cys destabilize the fibril relative to the free Cys derivative. In a sense, the relevant difference at position 33 (which is a Gly in the WT) is the relative ease with which it accommodates the Cys replacement, as seen in Fig. 5a. Besides in the N and C termini, those mutations for which the substitution of the relatively bulky, hydrophobic Cys destabilizes fibrils as compared with WT by less than 0.5 kcal/mol are K16C, F20C, G26C, K28C, A30C, and G33C (Fig. 5a). Of these, the modest effects at positions 20, 26, and 30 are understandable simply in terms of their surface exposure (Fig. 4) and presumed ability to accommodate some additional bulk.
Weight-normalized Thioflavin T Signals-ThT is often used to follow amyloid formation reactions (36). For any amyloidogenic protein, the time-dependent ThT increase is a good measure of the time required to reach the reaction end point. ThT cannot be used to accurately determine the position of the fibril-monomer equilibrium, however, since it cannot sense the amount of unpolymerized monomer remaining at equilibrium. In addition, ThT cannot be used to compare the degree of completeness of fibril formation for different amyloidogenic peptides, even peptides that are highly related structurally. This is because different amyloid fibrils can generate significantly different fluorescence yields in response to ThT. This is illustrated in Fig. 6. ThT yields were measured, and the fluorescence values were normalized based on independent determination of the amount of peptide in the reaction that had converted to amyloid (that is, the fibril component in the C r determina- tion). It can be seen that weight-normalized ThT responses vary by over a factor of 20 between the least responsive and most responsive fibrils. Similar variation has been observed for weight-normalized ThT values for amyloid fibrils from Pro and Ala mutants of A␤-(1-40). 3 Interestingly, fibrils of Cys derivatives in the N and C termini of A␤-(1-40) exhibit only small differences in ThT fluorescence yields as compared with WT fibrils (Fig. 6). Derivatives at other positions exhibit some stark differences, however. Particularly striking are the ThT yields for fibrils containing carboxymethyl-Cys at positions 29, 30, and 31, which are higher than WT fibrils by 2-4-fold. In both models for A␤-(1-40) fibril structure shown in Fig. 1, this segment is on the border of the predicted ␤-sheet starting around residue 31. If this site is normally a site for ThT binding in the fibril, the addition of negative charge in this region might increase binding of the positively charged ThT molecule. An equally striking effect is the diminution of ThT signal in fibrils from methyl-Cys replacement at residues 24 -28 (Fig. 6).
We think it is likely that these fluctuations in ThT fluorescence are due to differences in the amount of ThT "bound" in each case (37), but it is also possible that there are changes in fluorescence yield due to differences in the binding sites. At present, the main point in these and similar data is that ThT cannot serve as a reliable means to compare fibril formation reactions of different peptides.

DISCUSSION
Perhaps the most immediately interpretable data described in this report are the solvent accessibility data. Previously, Cys mutants were modified at the monomer level with EPR probes and the product fibrils, which were grown with agitation, were assessed for the degree of mobility in each side chain (20). These data are in rough agreement with the models shown in Fig. 1, especially in indicating flexible structure in the N-terminal segment and relatively rigid structure in the 16 -21 and 31-36 segments. These EPR data were not used to address solvent accessibility, however.
We interpret the accessibility of residues 20 and 35 to indicate the orientations of the ␤-extended chains in segments 16 -21 and 31-36. Our reasoning is that side chains projecting into the solvent-shielded interior of the A␤ loop will be uniformly buried and inaccessible, whereas side chains in the ␤-extended chain segments pointing outward will be either buried or accessible, depending on the involvement of these side chains into packing between filaments. In principle, the ␤-sandwiches formed by outwardly projecting side chains will have a central, strongly buried segment and edges with some potential for exposure to solvent. This idea is captured in Fig. 1d, which shows a schematic view of the 16 -21 and 31-36 segments within a single A␤ molecule incorporated into the amyloid fold. This model posits that residues 20 and 35 are outward projecting and on the edge of the packing interfaces involving the adjacent outward projecting side chains.
Although this argument is admittedly speculative, the basic arrangement arising from this logic was subsequently confirmed using further cysteine mutagenesis (30). Postulating that residue 20 is outward-facing leads directly to the corollary that residues 17 and 19 must be inwardfacing. Based on this hypothesis, residue 17 was chosen as an anchor for a series of double Cys mutant cross-linking experiments. Thus, double Cys mutants were constructed in which each mutant contains Cys-17 plus 1 additional Cys within the 34 -36 segment. Each double Cys mutant was used in a variety of experiments, which uniformly show that residues 17 and 34 within the same A␤ peptide molecule are packed together within A␤-(1-40) fibrils grown in PBS at 37°C under quiescent conditions (30). Since some degree of interaction between the 17 and 36 side chains was also indicated in these experiments, Fig. 1d is drawn to suggest a possible staggered orientation such that 17 resides between 34 and 36 in the filament interior. Later experiments further confirm the schematic model of Fig. 1d, in showing that side chains 19 and 32 from the same peptide molecule are also within disulfide crosslinking distance in the packed interior of these quiescent fibrils. 5 Of the other accessibility data, some were expected, and some were puzzling and surprising. It is surprising that residue 30 is so accessible to alkylation; no other residue, except Cys-4 in the presumably completely flexible N terminus, is completely alkylated in these experiments. The accessibility of Cys-29 and Cys-30 is consistent with these residues being in a turn or loop, as suggested by the models in Fig. 1. The accessibility of residue 38 is consistent with data showing that the C terminus, like the N terminus, is relatively unstructured (20,26,27). That residues 24 -27 are all partially accessible to alkylation is less consistent with their being involved in a third ␤-extended chain segment in the quiescent A␤-(1-40) amyloid fibril (26,39) and more consistent with their being in more irregular structure (27), 3 as suggested previously for A␤-(1-40) fibrils grown under agitated conditions (40).
Residues 22 and 23 remain structurally enigmatic. These were predicted to be in a turn based on Pro mutagenesis (26), and this assignment is also consistent with their lack of response to modified Cys mutations (Fig. 5) and Ala mutations. 3 However, the backbone amide hydrogens of these residues are strongly protected from exchange, and the side chain of Cys-22 is completely protected from alkylation. (Cys-23 cannot be tested because it cannot grow amyloid fibrils.) The free energy effects of Cys derivatization provide a wealth of detail on the environments within the folded amyloid fibril that must accommodate these side chains. Overall, there are some trends in these data that are understandable in the context of the current states of the models shown in Fig. 1. Other effects are not so easily understood. These data, and additional data from similar experiments, are expected to be of 5 S. Shivaprasad and R. Wetzel, unpublished observations. increasing importance in the future to direct and refine further model building.