2D IR provides evidence for mobile water molecules in β-amyloid fibrils

Abstract

The motion of water molecules close to amide groups causes their vibrational frequencies to vary rapidly in time. These variations are uniquely sensed by 2-dimensional infrared spectroscopy (2D IR). Here, it is proposed from 2-dimensional experiments on fibrils of amyloid β (Aβ)40 that there are water molecules in the fibrils. The spatial locations of the water (D₂O) were inferred from the responses of 18 amide modes of Aβ40 labeled with ¹³C = ¹⁸O. Fast frequency variations were found for residues L17 and V18 and for the apposed residues L34 and V36, suggesting cavities or channels containing mobile water molecules can form between the 2 sheets. Spectroscopic analysis showed that there are 1.2 water molecules per strand in the fibrils. The ¹³C = ¹⁸O substitution of 1 residue per strand creates a linear array of isotopologs along the fibril axis that manifests clearly identifiable vibrational transitions. Here, it is shown from the distributions of amide-I′ vibrational frequencies that the regularity of these chains is strongly residue dependent and in most cases the distorted regions are also those associated with the putative mobile water molecules. It is proposed that Aβ40 fibrils contain structurally significant mobile water molecules within the intersheet region.

An atomic scale description of protein aggregation mechanisms is one of the great challenges for biophysics. One system that is essential to understand is the formation of amyloid plaques in Alzheimer's disease. They are composed of amyloid β (Aβ) proteins with 39–42 residues that have aggregated into fibrils. The aggregation process and the structure of amyloid fibrils have been the focus of many experimental and theoretical studies (1–14). The general consensus reached in these studies is that each protein in a fibril participates in 2 different parallel in-register β-sheets that pack against and parallel to each other. The presence of bulk water in fibrils has been discounted (4, 15), but the techniques applied to detect the presence of bulk water do not exclude the possibility that individual water molecules are buried in the fibril core. Individual water molecules, whether associated with equilibrium or nonequilibrium configurations, may have a significant role in the structure of amyloid fibrils and a significant influence on their kinetics of formation. If the water molecules are mobile, they can also have a large effect on the frequency relaxation of amide-I vibrations (16–21). In the present work, the frequency relaxation is examined by 2-dimensional infrared spectroscopy (2D IR) of amide modes of ¹³C = ¹⁸O isotopically labeled residues: The experiments measure the vibrational frequency fluctuations from the motions of nearby charges.

2D IR is a new spectroscopic approach that can expose the dynamic aspects of complex structures (22–28). It has already been applied to structural and dynamic properties of large biological assemblies (29–31) including macroscopic fibrils (8, 32). The structural sensitivity of 2D IR was convincingly demonstrated in previous studies of Aβ40 fibrils (8), which measured couplings that were consistent with the U-shaped protofilament unit and parallel in-register β-sheet structure established by solid-state NMR studies (1, 2, 4). These 2D IR investigations were made possible by the systematic isotope editing of the amide-I′ transitions of 5 residues of the Aβ40 strand with ¹³C = ¹⁸O. The isotope shift in the vibrational spectrum is large enough that the residues could be examined one-by-one. The isotopically edited strands were used to form macroscopic fibrils that clearly exhibited the vibrational excitons formed from the secondary structure associated with each residue.

In the present work, 2D IR is used to examine the amide-I′ modes of 18 isotopically edited residues of Aβ40 fibrils. The 2D IR experiments reveal the vibrational frequency dynamics of individual residues. A few of the residues undergo a fast relaxation that is consistent with the presence of water at specific locations in the fibril. The locations of the 18 ¹³C = ¹⁸O substituted residues in the Aβ40 strand are marked with red labels in Fig. 1A.

Fig. 1.

Descriptive diagrams of Aβ40 fibrils and the energy levels for linear excitons (8, 28). (A) A cross section of a U-shaped monomer unit of the Aβ40 fibril with the spatial region proposed to be occupied by D₂O molecules shown in blue. Residues 9–40 are shown. The ¹³C = ¹⁸O-labeled carbonyl groups are shown in red. (B) Energy level of a fibril formed from a ¹³C = ¹⁸O isotopolog of A. (Left) The fibril contains only a few strands having ¹³C = ¹⁸O. In this isotopically diluted limit the amide-I transition of the spatially isolated isotopolog is near 1600 cm⁻¹. (Right) All strands in the fibril are ¹³C = ¹⁸O isotopologs. In this limit, linear chain excitons having quasi-infinite length, N, and coupling β are formed. The details of the amide-I mode coupling (8) dictate that the transitions occur only to the lowest-energy group of exciton states.

Results and Discussion

Vibrational Exciton Effects.

Previous 2D IR work (8) has shown that isotope replacement of 1 residue in each strand gives rise to the formation of an energetically isolated linear chain along the fibril axis of ¹³C = ¹⁸O-labeled residues. The vibrational frequencies of the chain at 1570–1600 cm⁻¹ are widely separated from the main bands at ≈1630 cm⁻¹. The interresidue coupling between amide-I modes along this chain gives rise to a 1-dimensional vibrational exciton band having allowed transitions only to the low-energy states of the band (8). These transitions are shifted to lower frequency by 2 β = 20 cm⁻¹ from the uncoupled ¹³C = ¹⁸O transition at ν_o, as illustrated in the energy diagram in Fig. 1B. The uncoupled transition is an amide-I vibrational frequency that is free from intermode couplings but influenced by other interactions that modify the force field, including electrostatic and dispersion interactions and hydrogen bonding. The value of ν_o is known from isotope dilution experiments where only a small fraction of the strands are ¹³C = ¹⁸O isotopologs (8). In this article, we deal with perturbed versions of these ideal linear chain spectra. Structural disruption of the chain will tend to localize the excitations and can cause multiple transitions to appear in the spectrum near the band edges denoted by k = 1 and k = N in Fig. 1B. The higher-energy transitions are buried in the main amide bands. Here, we report on the dynamics being undergone by these localized vibrations; however, a quantitative treatment of the dynamical disordering of the chain and localization of the excitation is not included.

Spectral Characteristics of Labeled Fibrils.

The ¹³C = ¹⁸O isotope-labeled regions of the amide-I′ 2D IR spectra of mature Aβ40 fibrils of 18 isotopologs are shown in Fig. 2. The band shape, frequency, and magnitude of the peaks in the isotope region vary significantly with residue. To compare 2D IR spectra with linear spectra, traces of the positive region of the 2D IR spectra in Fig. 2 along a diagonal line (see Materials and Methods) for 2 examples are illustrated in Fig. 2 S and T along with their corresponding linear IR spectra. The 2D IR signal is proportional to the square of the linear absorption cross section, which greatly increases its contrast in regions where the background signal has a significant absorbance but a low extinction coefficient. In addition, the diagonal trace of a 2D IR spectrum shows transitions having a narrower bandwidth than the corresponding linear IR spectrum (28). The spectrum in Fig. 2A of V12* is typical of a linear chain of residues that are well aligned along the fibril axis by having a well-separated sharp isotope transition highlighted by the thick red arrow in Fig. 1B. This is possible only when V12 is part of a parallel-in-register β-sheet (8). This same character is manifested by residues F20*, V24*, G29*, I30*, I31*, G33*, G37*, G38*, and V39*. However, the results in Figs. 2 and 3 indicate that isotopomers L17*, V18*, L34*, and V36* have much broader, more complex 2D IR spectra, which implies that these particular amide units are in environments that are significantly perturbed from the regular parallel-in-register β-sheet. The spectra of residues F19*, A21*, and I32* show a small broadening effect of a similar type. It is shown in the next section that this broadening is in fact associated with fast fluctuations of the amide vibrational frequency.

Fig. 2.

2D IR spectra, linear IR spectra, and diagonal traces of 2D IR spectra of isotopologs of Aβ40. (A–R) 2D IR spectra in the ¹³C = ¹⁸O region of 18 isotopologs as indicated. The residue symbol followed by an asterisk indicates the indicated isotopologs of the Aβ40 strand. Red contours are positive signals, and blue contours are negative. All of the 2D IR spectra are real absorptive spectra, were collected at zero waiting time (i.e., T = 0), and are plots of the coherence frequency ω_τ versus the detection frequency ω_t. (S and T) Linear IR spectra (blue) and diagonal traces (red) of 2D IR spectra for two isotopologs (L17* and G37*).

Fig. 3.

Plots of semibandwidth (SB) and inverse slope (IS) for indicated ¹³C = ¹⁸O-edited residues. A more detailed technical description of how the parameters were obtained appears in the text and SI Text.

2D IR Spectra Suggest the Presence of Mobile Water (D₂O) Molecules.

As a means to quantify the dynamic characteristics of each residue, the semibandwidth (SB) and inverse slope (IS) were measured (See SI Text and Figs. S1 and S2). These parameters are plotted against residue number in Fig. 3, which summarizes the main results of this article. These 2 parameters are expected to be sensitive to the initial width of the frequency distribution of the amide-I′ modes of the indicated isotopologs.

The SB is the diameter of a circle having the area in the 2D frequency space of a cut through the signal at 25% of the peak volume of the v = 0 → 1 transition in the isotope region at waiting time T = 0. The IS refers to the slope of the nodal line in the region of the 2D IR spectra where the positive and negative signals (see Fig. 2) overlap. For an isolated anharmonic oscillator it is a measure of the ratio of inhomogeneous to the homogeneous broadening contributions to the spectral width: This knowledge allows spectral diffusion to be represented quantitatively (33, 34). The waiting-time (see Materials and Methods) dependence of the IS is then directly proportional to the vibrational frequency autocorrelation function for the amide-I transition of a ¹³C = ¹⁸O substituted residue. For a vibrational transition, the time dependence of the spectrum causes the initially inhomogeneous band to appear more homogeneous as the waiting time increases, resulting in a decrease of IS.

The residues L17*, V18*, L34*, and V36* having the large SB and large IS have a major portion of their broad transitions at relatively higher frequencies than those of other isotopologs (28) as can be gleaned from Fig. 2. This deviation must be attributed to a more significant structural perturbation of the linear chain along the fibril axis for these residues. A likely explanation for the anomalous frequencies, which also rationalizes the peaks in SB and IS, is that the regularity of the linear chain is perturbed by water molecules. Therefore, the linear chains, and consequently the parallel-in-register sheets from which they are derived, are postulated to be disordered and some of the distorted regions are hydrogen bonding to water.

The 2 dynamic parameters SB and IS show peaks at 2 regions, 1 near L17 and the other at L34. The correlation coefficient between SB and IS, incorporating all residues in the plot of Fig. 3, is 0.78, so they are most probably signaling the same properties of the frequency distributions. Among the 18 isotopomers, L17*, V18*, L34*, V36*, and G37* show unusually large values of both the SB and the IS. The enlarged contributions to these line widths are attributed to residue-dependent distributions of vibrational frequencies that are shown in the next section to have a picosecond time evolution. The fluctuations by residue of SB are dominated by changes in the diagonal width (Fig. S3).

Waiting-Time Evolution of 2D IR.

The existence of mobile polar groups, consistent with water (D₂O) molecules or small aggregates of D₂O, near specific residues of Aβ40 fibrils is evidenced by 2D IR spectra at finite waiting times (28), where spectral diffusion caused by picosecond-timescale field fluctuations is clearly revealed. The T dependence of the IS is exemplified by the 4 typical cases in Fig. 4. The IS is not influenced by the population decay kinetics, but it shows a prominent ≈1- to 2-ps decay to a plateau in the case of L17*, V18*, L34*, and L36*. None of the residues having IS less ≤0.5 on the scale of Fig. 3 showed significant evolution of their IS during the waiting period. These results strongly suggest that it is the vibrational frequency dynamics being tracked by the data in Fig. 3: Large IS or SB corresponds to fast relaxation of the vibrational frequency distributions. Because most of the isotopologs show no fast relaxation component (as for F20 and A30 in Fig. 4), the fluctuations of side chain charges are inferred to cause relaxation on a much slower time scale than is observed. Relaxations on the picosecond timescale are typical of amide modes in direct association with water (17, 35–37) and therefore the present results suggest that there could be water molecules trapped in the fibrils. These water molecules would have to be sufficiently free to undergo rotational or librational motions on the ≈1- to 2-ps timescale so that fluctuating fields of sufficient magnitude are generated. The T dependence of L17* shows 3 closely spaced diagonal peaks that become more clearly resolved at larger values of T (28). These 3 peaks are visualizable from the L17* panel in Fig. 2. The lowest-frequency diagonal peak is proposed to be generated by delocalization of the vibrational excitation along a linear chain, whereas the other 2 peaks originate from vibrational excitations that are more spatially localized on residues where they must reside for times that are long enough to permit the observed dynamics to occur. This example and also that of L34* shown in Fig. 5 illustrate that chains of well-aligned identical amide-I′ groups, typified by a single sharp vibrational transition, are structurally perturbed. Confirmation of the presence of water in the fibrils is provided by FTIR measurements as now described.

Fig. 4.

Plots of inverse slope (IS) against the waiting time (T in picoseconds) and the fitted responses for L17, V18, F20, and A30. The fitted curves are IS = 0.496·exp(−T/1.026) + 0.275 for L17, IS = 0.402·exp(−T/0.961) + 0.3 for V18, IS = 0.00305·T + 0.293 for F20, and IS = 0.0129·T + 0.362 for A30. The points and fitted curves for each residue are represented in the same color.

Fig. 5.

2D IR spectra of L34* at indicated waiting times. Left (A, C, E, and G): 2D IR spectra at the waiting times of T = 0, 0.5 ps, 1 ps, and 2 ps. Right (B, D, F, and H): Scaled up by 10×, views of the boxed areas in their corresponding 2D IR spectra on the Left. The time evolution of the signal in the cross-peak region S(ω_τ = 1573 cm⁻¹, ω_t = 1584 cm⁻¹; T) is clearly demonstrated on the Right.

Hydration of Fibrils Is Evident from FTIR.

The presence of water inside the fibrils was confirmed by FTIR experiments. Fig. 6 shows the average FTIR spectra of the 18 isotopomers in the amide-A′ spectral region. The doublet at 2400–2500 cm⁻¹ is the N-D stretch and an associated Fermi resonance from amide II (38, 39). The integrated area in the amide-A′ spectral region decreased roughly by half after long exposure to air because of the H/D exchange of the N-D′s that are most accessible to atmospheric H₂O molecules. The remaining N-D′s did not exchange over months of exposure, suggesting they were located deep inside the fibrils. There is a noticeable peak at near 2,565 cm⁻¹ in all of the isotopomer fibril spectra that is assigned to the O-D stretch of residual D₂O located inside the fibrils. This peak has not been reported in studies of the amide-A′ band (39). It is up-shifted from bulk D₂O by 60 cm⁻¹ and the peak is considerably narrower. Assuming that the O-D stretch of the residual D₂O has the same transition dipole strength as in bulk D₂O, the average number of molecules was estimated from the integrated intensity of the 2565 cm⁻¹ band to be 1.2 per Aβ40 strand. The similarity of the IR spectra of fibrils formed from the 18 different isotopomers indicated that a similar number of water molecules is present in every case when the same preparation method was used (200 μM in D₂O, pD 2.0, 10 weeks of maturation at 37 °C, and no agitation). The fact that partial H/D exchange readily occurs was confirmed by studies of the linear spectra of isotopomers with deliberate wetting and drying cycles. Such partial exchange has been widely used for structure determination by NMR (40).

Fig. 6.

The averages of the linear IR spectra of Aβ40 fibrils of 18 isotopomers at two different degrees of exposure to air (containing H₂O vapor) and resolved-curve fits of the average of the linear IR spectra of more air-exposed samples. Blue curve: The average of the IR spectra of the fibrils generated from D₂O solution without being exposed to air. Red curve: The average of the IR spectra of the fibrils generated from D₂O solution and exposed to air for 10 months. Cyan and magenta curves: The resolved curves of the red spectrum, where the magenta curve represents the peak being centered at 2565 cm⁻¹. Green curve: The sum of the resolved components. The green curve is almost overlapping the red curve. Seven Voigt profiles were used to fit the red spectrum and the frequency range used for the fitting was 2160–2740 cm⁻¹. The dotted circle highlights the O-D stretch spectral region in the linear IR spectra used to determine the number of water molecules per strand.

Individual Fibrils Show Heterogeneity.

The 2D IR spectra of I31* (Fig. 2) also show microscopic heterogeneity of the individual fibrils. Its 2D IR spectra show multiple diagonal peaks located at ω_τ = 1580 and 1568 cm⁻¹ together with obvious cross peaks, which indicate coupling, that are expected only when the diagonal transitions originate from the same fibril. In this example the cross peaks are strong enough to seriously distort the v = 1 → 2 part of the diagonal peak at ω_τ = 1,580 cm⁻¹ as seen in the I31* panel of Fig. 2.

Another example where there are multiple transitions is L34*. After the fast spectral diffusion has occurred, significant cross peaks between the diagonal peaks at 1584 cm⁻¹ and 1573 cm⁻¹ (see Fig. 5) become evident. This cross-peak time evolution indicates that picosecond chemical exchange of water molecules or energy transfer between amide-I′ modes, or both (20, 28), is occurring.

These examples evidence a significant intrafibril heterogeneity that for some residues seriously disturbs the ideal linear exciton chains of amide units whose spectra should be dominated by a single sharp transition. The spectra of disordered linear chains are known to have transitions that are located between the infinite chain (see model in Fig. 1B) and the zero-order transition at ν_o: These transitions correspond to finite clusters created by the disordering (41). For the case of L34* we have proposed that the clustering might be caused in part by the disruption of the chain structure by trapped water molecules that are directly sensed through the spectral diffusion of the trapped excitation. On the other hand, for G37* and I31* if water molecules are responsible for the structure disruption, they must be immobilized for longer than ≈5 ps or be further from the amide unit than in the residue L17*, V18*, L34*, and V36* cases, because these amide-I′ modes show no significant spectral diffusion. Therefore, both mobile and immobilized D₂O may be present in the fibrils.

One of the 5 Aβ40 isotopologs that shows large SB and IS is G37*. In contrast to the other 4 cases, G37* shows 2 sharp and circular peaks separated by 8 cm⁻¹ in the v = 0 → 1 transition region (see Fig. 2) at all waiting times and no significant time evolution. The presence of a pair of transitions distorts the nodal line from which the slope is determined. These spectra make it clear that the in-register property of the β-sheet is significantly disturbed in the region of this residue.

Structural Characteristics of Fibrils.

The experiments suggest that residues L17 and V18 are apposed to residues L34 and V36 because both pairs sense the putative water (D₂O) motions. In a U-shaped model of an Aβ40 fibril (Fig. 1A), the residues L17 and V18 are indeed apposed to L34 and V36, supporting the notion that the water is present mainly in the regions outlined qualitatively by the blue shadings.

The amyloid fibrils isolated from Alzheimer's disease brain are polymorphic by electron microscopy, and several of the morphologies observed in vivo have been reproduced in vitro by varying the conditions under which fibrils are formed (42). The fibrils prepared for this study were morphologically uniform with a 20- to 50-nm periodicity (see the electron micrograph provided in Fig. S4), and they resemble some of the fibrils isolated from human brain tissue in this respect (43, 44). However, they differ in morphology from the fibrils for which the most structural information is available from NMR and for which detailed molecular models have been prepared (9). In addition to being based on fibrils with a different morphology, these models do not suggest how a water molecule might situate between the β-sheets and interact with the backbone peptide groups of residues L17, V18, L34, and V36. Instead, they depict the intersheet space as bulging to accommodate the large side chains of these residues. These models also do not account for the close proximity of the F19 and I31 side chains, detected in fibrils seeded with brain-derived fibrils (45). An alternative model in which the side chains of L17 and F19 are both packed within the intersheet space has been previously proposed (1) and minor adjustments to this model produce one in which these side chains interact with that of M35 and in which all 3 side chains and a water molecule can be accommodated by the lack of inward-directed side chains on adjacent residues G33 and G37 (Fig. 1A and Fig. S5). This alternative model also provides space for a water molecule in the intersheet space between residues A21 and I32. Details of model generation are provided as SI Text.

Fig. 7A illustrates how in principle the interactions with 1.2 water molecules per strand could cause clustering of amide units in the fibril axis direction and thereby result in significant changes in the IR spectra of the linear chains. Fig. 7B shows a numerical computation of a water molecule interacting with 2 tripeptides in a configuration resembling that of a parallel β-sheet. The C = O axis directions of the 2 amide groups that are hydrogen bonded to the water molecule are significantly different from each other and the distance between the 2 amide groups is increased. Interactions of this type and the concomitant variation in the frequency of the amide-I modes by H-bonding to water would lead to a significant perturbation of the amide-I linear-chain exciton spectrum and localization of the vibrational excitation.

Fig. 7.

Pictorial representation of possible water distribution inside Aβ40 fibrils. (A) A top-view cross section of the structure in Fig. 1A showing the β-sheet region of G29–V40, where water molecules are shown as blue ellipses. Water-free regions are highlighted with yellow boxes and the distances between adjacent strands in the water-free regions are depicted as being smaller than those in other regions. The manner of clustering of water molecules along the fibril axis is hypothetical. (B) Enlarged views of the region destabilized by a water molecule and highlighted with a rectangle in A. A partially optimized structure using Gaussian03 at the HF/6–31G(d,p) level with three backbone dihedral angles (one at the upper strand and two at the lower strand) fixed to be 180° is shown. The dotted lines represent hydrogen bonds.

Computations have predicted that water channels will form near the turn region between residues D23 and K28 in amyloid fibrils (3, 6, 46) but no water was predicted to be in the region of residues L17, V18, L34, and V36 as observed in our work.

Water Trapping and Kinetics.

The water distribution inside the fibrils is likely to be a characteristic of the aggregation kinetics. The aggregation of Aβ proteins has been suggested to be driven mainly by hydrophobic cooperativity (6, 11, 47). Our experiments provide evidence for water molecules near the amide groups of the hydrophobic residues L17 and L34 and we speculate that they become trapped as a nonequilibrium distribution at a stage in the fibril formation where there is a rapid expulsion of water (48) to appose the adjacent β-sheets. Accordingly, investigations of these trapped water molecules in kinetic experiments will be expected to lead us to better understand the aggregation process of Aβ proteins and the structure of the fibrils.

Conclusions

Through 2D IR experimental studies on the fibrils of 18 ¹³C = ¹⁸O-labeled amide mode isotopomers of Aβ40, we have observed time-dependent frequencies of these modes that suggest the presence of water molecules nearby to trapped (localized) vibrational excitations. These water molecules would require sufficient mobility to dephase the amide vibrational transitions on a picosecond timescale. The 2D IR analysis of the ¹³C = ¹⁸O bands is based on the SB of the v = 0 → 1 transition, the IS of the nodal line due to the overlap of the v = 0 → 1 and v = 1 → 2 transitions, and the time dependence of the vibrational frequency. The analysis revealed fast spectral diffusion that is typical for amide units of peptides nearby and perhaps hydrogen bonded to water molecules. The L17 and V18 as well as L34 and V36 residues exhibit the largest effects, suggesting that the water is in the neck region of strands. On average, 1.2 water molecules were estimated to exist per Aβ40 monomer unit inside the fibrils from the investigation of the linear IR spectra of 18 isotopomers. The 2D IR spectra manifest significant disruption of the in-register β-sheet structure of the fibrils. The picosecond spectral diffusion suggests that a nearby D₂O molecule undergoing hindered rotation could generate appropriate field fluctuations at the amide CONH atoms. As yet we have found no precedent for such behavior from the simulations of Alzheimer's β-amyloid proteins (3, 49). The history and the function of the putative captive water molecules in fibril formation and structure require further study, as does the additional effect of fluctuating side chain charges.

Materials and Methods

2D IR Spectral Measurements.

The 2D IR experiments were carried out as reported previously (8, 50) by means of a sequence of 3 infrared pulses. The processing of the heterodyned signal leads to a 2-dimensional spectrum S(ω_τ, ω_t; T) for each waiting time (T) during which the intermediate population is allowed to evolve. The coherence and detection frequencies are ω_τ and ω_t, which loosely correspond to the pumped and probed vibrational frequencies. The specific forms of the functions that describe the response of a vibrator to the pulse sequence are defined in a number of places (24, 51–53) and so need not be repeated here. The main point is that both the v = 0 → 1 and the v = 1 → 2 transitions appear in the spectrum with different signs and they are displaced along ω_t by the anharmonicity. The elongation of individual peaks along the diagonal exposes the correlation of the pumped and probed frequencies from the inhomogeneous distribution. This correlation is lost as T progresses if spectral diffusion occurs, and the spectra become more circular. Each transition is represented by a diagonal peak and 2 transitions are coupled if there are cross peaks linking these diagonal transitions.

Model Generation and Electron Microscopy.

See SI Text.

Fibril Preparation and Morphology.

See SI Text.