Toward determining amyloid fibril structures using experimental constraints from Raman spectroscopy

Abstract

We present structural models for three different amyloid fibril polymorphs prepared from amylin_20–29 (sequence SNNFGAILSS) and amyloid-β_25–35 (Aβ_25–35) (sequence GSNKGAIIGLM) peptides. These models are based on the amide C=O bond and Ramachandran ψ-dihedral angle data from Raman spectroscopy, which were used as structural constraints to guide molecular dynamics (MD) simulations. The resulting structural models indicate that the basic structural motif of amylin_20–29 and Aβ_25–35 fibrils is extended β-strands. Our data indicate that amylin_20–29 forms both antiparallel and parallel β-sheet fibril polymorphs, while Aβ_25–35 forms a parallel β-sheet fibril structure. Overall, our work lays the foundation for using Raman spectroscopy in conjunction with MD simulations to determine detailed molecular-level structural models of amyloid fibrils in a manner that complements gold-standard techniques, such as solid-state nuclear magnetic resonance and cryogenic electron microscopy.

I. INTRODUCTION

Amyloid fibrils are filamentous, β-sheet-rich protein aggregates that are implicated in numerous diseases, including Alzheimer’s, Parkinson’s, Huntington’s, and type II diabetes.¹ Because of their insoluble and non-crystalline nature, conventional structural characterization methods, such as x-ray crystallography and solution-state nuclear magnetic resonance (NMR) spectroscopy, cannot be used to determine the molecular structure of fibrils.² Fortunately, advances in solid-state NMR (ssNMR) and cryogenic electron microscopy (cryo-EM) have enabled the determination of high-resolution structures of amyloid fibrils prepared in vitro,^3–6 as well as those harvested from the tissues of patients.^7–12 These studies indicate that fibrils share a common conformational motif known as a “cross-β” structure, in which extended β-sheets stack together with their strands aligned perpendicularly to the fibril’s long axis.^13,14 They also demonstrate that fibrils exhibit polymorphism, where a given peptide is capable of forming a variety of distinct fibril structures.^15,16

Despite the wealth of high-resolution structural information gleaned from ssNMR and cryo-EM, these techniques are not without their pitfalls. Solid-state NMR, for example, requires high sample loads ($>10$ mg),^6,17,18 expensive isotopic labeling schemes, and long spectral acquisition times that can take several days.⁶ To obtain high-resolution structural data, both ssNMR and cryo-EM require relatively extensive sample preparation to ensure that fibrils are both homogeneous and well-ordered. This can introduce bias in determining the molecular structure of amyloids by preferentially selecting for the most abundant fibril polymorphs. In addition, it can be difficult to resolve electron densities in cryo-EM or obtain good chemical shift dispersion in ssNMR for all but the most well-ordered regions of fibrils. Disordered, dynamic, or structurally heterogeneous regions, which could play important roles in initiating aggregation, sequestering other proteins into amyloid plaques, or aberrantly interacting with biological cells, can be difficult to study with these techniques.

In contrast, vibrational methods, such as Raman spectroscopy, do not suffer from these pitfalls. Raman spectroscopy can be used to interrogate a wide variety of samples with little preparation, including fibrils in solution,^19,20 gels,²¹ and fibril films.²² Data acquisition is relatively fast and robust. In addition, Raman spectroscopy can be used to quantify the distributions of the peptide bond and side chain dihedral angles in amyloid fibrils.^21,23–25 Similarly, polarized Raman measurements can be used to determine the relative orientation of chemical bonds and functional groups in fibrils.^22,26 Thus, the conformational sensitivity of Raman spectral features enables the facile differentiation of fibril polymorphs,^21,23,25 as well as the ability to robustly monitor the structural evolution of oligomeric precursors that aggregate into fibrils.²⁷

However, despite its versatility and structural sensitivity, Raman spectroscopy is generally considered to be a “low-resolution” characterization method in the broader amyloid community. One reason for this is because Raman spectroscopy has generally only been used to qualitatively evaluate fibril secondary structures rather than determining three-dimensional models, such as ssNMR. We believe, however, that structural parameters measured using Raman spectroscopy (vide supra) can, in fact, be harnessed to determine detailed molecular-level structural models of amyloid fibrils. We recognize that this can be accomplished by taking inspiration from ssNMR, in which experimentally measured distances and dihedral angles are used as constraints in energy minimization procedures performed on structural models of amyloid fibrils using molecular dynamics (MD) simulations.¹⁸

To test this idea, we investigated the molecular structures of three fibril polymorphs prepared from amylin_20–29 and amyloid-β_25–35 (Aβ_25–35) peptides. Amylin_20–29, which derives from residues 20–29 of the 37-amino acid amylin peptide, forms fibrils implicated in the pathology of type II diabetes.^28–32 Amyloid-β_25–35 (Aβ_25–35) derives from residues 25–35 of the 40–42 residue long Aβ peptide.^13,33 Fibrils formed from Aβ compose extracellular plaques that have been implicated in the pathology associated with Alzheimer’s disease^13,33 and cerebral amyloid angiopathy.^34,35

II. EXPERIMENTAL METHODS

A. Materials

Amylin_20–29 (sequence SNNFGAILSS) and Aβ_25–35 (sequence GSNKGAIIGLM) were purchased from GL Biochem (Shanghai, China) at 91% purity and used without further purification. Dimethylsulfoxide (DMSO) was purchased at $\ge 99.8\%$ purity from Supelco. Milli-Q grade water (18.2 MΩ cm) was obtained from a Milli-Q^® IQ 7000 ultrapure lab water system from Millipore Sigma. Phosphate buffer saline (PBS) tablets were purchased from Sigma-Aldrich. Phosphate buffer solution was made from sodium phosphate monobasic monohydrate and sodium phosphate dibasic anhydrous purchased from Fisher Scientific. Filters (0.45 µm pore-size) were purchased from Fisher Scientific. Mica (Product No. 50, V1 Grade) and silicon substrates (Product No. 21610-55) were purchased from Ted Pella Inc. An ultra-sharp commercial Atomic Force Microscopy (AFM) probe (160AC, OPUS by MikroMasch) and a standard AFM probe (PPP-NCHR, Nanosensors™) were both purchased from NanoAndMore USA Corporation.

B. Sample preparation

Two different fibril polymorphs were prepared from the amylin_20–29 peptide using a modified procedure based on that developed by Madine et al.³⁶ For both polymorphs, 2 mg of peptide were first disaggregated in 20 µl of DMSO after incubation at room temperature (22 °C) for 1 h. To prepare the antiparallel β-sheet polymorph (polymorph 1), 980 µl of water was slowly added, followed by an additional 1 ml of sodium phosphate buffer (20 mM, pH 7.13). The final solution was 2 ml with a final DMSO concentration of 1.41 × 10⁻⁴ mM, 10 mM of buffer, and a peptide concentration of 1.13 mM. To prepare the parallel β-sheet polymorph (polymorph 2), 1980 µl of sodium phosphate buffer (100 mM, pH 7.13) was slowly added to the sample. The final solution was 2 ml with a final DMSO concentration of 1.41 × 10⁻⁴ mM, 100 mM of buffer, and a peptide concentration of 1.01 mM. The final step for both polymorphs involved filtering solutions using a 0.45 µm pore-sized filter to remove any undissolved large aggregates. All starting solutions were visually clear, and both samples were incubated at room temperature (22 °C). Aggregates could be visually observed after 7 days of incubation for polymorph 1 and 9 days for polymorph 2.

Parallel β-sheet fibrils were also prepared from Aβ_25–35 peptides. This was accomplished by dissolving 2.12 mg of peptide in 1.06 ml of water. The resulting peptide solution appeared visually clear. After this, 1.06 ml of 2× phosphate buffer saline (PBS, pH 7.4) was slowly added to the sample for a final peptide concentration of 1 mM. The solution was gently mixed by carefully inverting the sample vial three times. Aggregates could be visually observed shortly after preparing the solution and were incubated at 37 °C for 3 days.

C. Sample alignment procedure using drop cast deposition

Following incubation, fibrils were harvested from each sample by centrifuging 500 µl of solutions for 1 h at 21 300 × g (Eppendorf Centrifuge 5425). The supernatant for each sample was carefully decanted so that the pellet remained. The pellet was washed twice to remove residual salt crystals by sequentially resuspending it in water and centrifuging. Following this, the pellet was resuspended in 250 µl of water. Aliquots of each sample were then diluted, deposited onto a silicon substrate, and allowed to dry in a dust free environment to create a coffee ring³⁷ for polarized Raman measurements.

D. Raman spectroscopy

Raman spectra were measured using a Horiba LabRAM HR Evolution Raman microscope (Horiba Scientific) using 633 nm excitation from a helium–neon laser. The laser light was focused on the sample using an infinity-corrected achromatic 100× objective (0.9 NA, MPLN100X, Olympus) with the average power ranging from 2.5 to 17.9 mW. Acquisition times ranged between 30 and 360 s per spectrum. Spectral acquisition parameters were carefully chosen to balance maximizing signal-to-noise ratio and mitigating photodegradation of samples. Under the illumination conditions used, we observed no visual signs or spectral signatures of sample degradation (Fig. S1). Amylin_20–29 polymorph 1 and Aβ_25–35 fibril samples exhibited substantial fluorescence backgrounds, which made it difficult to collect high signal-to-noise Raman spectra. The scattered light was collected by the focusing objective in a 180° backscattering geometry. The scattered light was focused into a spectrometer and dispersed using a 600 gr/mm grating. Spectra were imaged using a Synapse EM charge coupled device (CCD) camera (1600X200-FV, Horiba Scientific).

For polarization measurements, fibrils were aligned so that their long axis was parallel to the polarization of the laser light, which we defined as the laboratory coordinate’s Z direction. The lab frame’s Y coordinate was defined to be along the direction of the laser light’s propagation. A half-wave plate was used to rotate the polarization of the incident light along the X direction, but was removed for incident light polarization measurements involving the Z direction. As described in detail by Adar,³⁸ this was done to reduce artifacts introduced by the half-wave plate. A polarizer was used to select either the Z or X polarization component of the scattered light, which was then focused into a spectrometer and dispersed using a 600 gr/mm grating. An optical scrambler was installed before the spectrometer entrance to depolarize the scattered light to minimize the polarization bias of the grating. Polarized Raman measurements were made for the following four incident and scattered light configurations: ZZ, ZX, XZ, and XX (incident and scattered light, respectively).³⁸ The instrument was benchmarked with Raman polarization measurements of cyclohexane to ensure that our measurements were accurate. We found that the depolarization ratios measured for cyclohexane on our instrument were within experimental error of values reported previously in the literature.^39,40

E. Atomic force microscopy (AFM)

Samples were prepared for AFM imaging using a procedure that was modified from the work of Ostapchenko et al.⁴¹ Briefly, a mica disk was exfoliated to create an atomically flat and clean surface. A 10 µl aliquot (2.2 mg/ml) of fibril solution was then deposited onto the disk. The sample was incubated on the disk for 10 min in a dust-free environment before being washed 3× with water. The sample was then wicked dry with filter paper and dried overnight prior to imaging in a dust-free environment.

The AFM measurements were performed on an Asylum MFP-3D-BIO AFM instrument (Oxford Instruments) in the AC mode. Amylin_20–29 polymorph 1 [Fig. 1(a)] was imaged using an ultra-sharp commercial probe with a 26 N m⁻¹ force constant and 300 kHz resonance frequency. Amylin_20–29 polymorph 2 [Fig. 1(b)] and Aβ_25–35 fibrils [Fig. 1(c)] were imaged using a standard AFM probe with a 42 N m⁻¹ force constant and 330 kHz resonance frequency.

FIG. 1.

Representative AFM images of amylin_20–29 and Aβ_25–35 fibrils. (a) AFM image of amylin_20–29 polymorph 1, which is a 256 line image of 10 × 10 µm² size (4 µm scalebar). (b) AFM image of amylin_20–29 polymorph 2, which is a 512 line image of 10 × 10 µm² size (2 µm scalebar). (c) AFM image of Aβ_25–35 fibrils, which is a 256 line image of 10 × 10 µm² size (2 µm scalebar). All images were measured in the AC mode. Additional AFM images can be found in the supplementary material (Fig. S2).

The images were analyzed with Gwyddion software (open-source software for scanning probe microscopy data), and the height traces obtained from Gwyddion were visualized as graphs in Graphpad Prism 9. See the supplementary material for additional details and images (Fig. S2).

III. COMPUTATIONAL METHODS

A. Molecular dynamics (MD) simulations

For MD simulations, two 24-mer models of amylin_20–29 protofibrils and one 24-mer model of an Aβ_25–35 protofibril were constructed using Visual Molecular Dynamics (VMD).⁴² For each model, a single β-sheet containing 12 peptides was first built and then duplicated and stacked on the initial β-sheet. The putative fibril structures were constructed based on Raman data. For amylin_20–29, both an antiparallel and parallel β-sheet fibril model were constructed, while only a parallel β-sheet fibril was constructed for Aβ_25–35. The α-carbons of the two stacked β-sheets were initially separated by 10 Å for each model. Rather than solvate these models with an explicit solvent, we opted to use a generalized Born implicit solvent (GBIS) protocol with a dielectric constant of 3.23, mimicking that of the interior of proteins, and an ion concentration consistent with experimental sample preparation. Implicit solvent enabled us to better approximate the environment of an effectively infinite fibril. These models consist of 3336 and 3720 atoms for amylin_20–29 and Aβ_25–35 fibrils, respectively.

During minimization and equilibration simulations, structural constraints obtained from the Raman experimental data were applied to the 24-mer fibril models using NAnoscale Molecular Dynamics’ (NAMD’s) collective variable functionality. In particular, the ψ-angles of the peptides were harmonically restrained about the ψ-angle peaks we observe experimentally (150° amylin_20–29 polymorph 1 and 139° for amylin_20–29 polymorph 2 and Aβ_25–35). The angle between peptide carbonyl bonds and the fibril long axis was harmonically constrained to the values obtained experimentally. Additionally, the constraint force constants were tuned to reproduce the experimentally obtained ψ-angle and C=O bond angle distributions. The fibril axes were approximated in these models by using VMD to calculate the inertial tensor of the α-carbons and extracting the principal components. The fibril long axes were taken as the resulting principal component perpendicular to the peptides. They were calculated separately for every frame of the simulation trajectories, as atomic fluctuations in the fibrils cause small changes to the axis vectors.

The NAMD MD package was used to energy minimize and simulate the fibril models. The all-atom CHARMM36m force field was used to calculate potential energies and forces due to its improved treatment of the secondary structure compared to CHARMM36.^43,44 All simulations were performed under a constant temperature and pressure of 298 K and 1 atm, respectively. The Verlet velocity integration algorithm was used with a time step of 1 fs, and the SHAKE algorithm was employed to constrain heavy atom-hydrogen covalent bonds. Non-bonded interactions were calculated for atom pairs using a cutoff of 12 Å, and a switching function was used at distances greater than 10 Å to truncate the potential. The particle mesh Ewald method⁴⁵ was used to calculate long range electrostatics. Visual and quantitative analysis of MD simulations was performed with Amber’s cpptraj tool^46,47 and VMD.⁴²

The 24-mer fibril models were first energy minimized for 5000 steps with rigid backbone atoms using the conjugate gradient minimization scheme. Following this, we released the rigid backbone atoms and equilibrated the three fibril models for 2 ns with the aforementioned ψ- and C=O bond angle constraints for each polymorph taken from the Raman data. Following these constrained equilibration simulations, we released all the constraints and ran another 2 ns to observe the thermodynamic stability of the 24-mer fibrils. We collected ψ- and C=O bond angles, as well as inter-strand and inter-sheet distances from the constrained and unconstrained simulations.

IV. RESULTS AND DISCUSSION

A. Choice of peptide systems to study

We chose to study fibrils prepared from the amylin_20–29 and Aβ_25–35 peptides because of their small, tractable nature, which makes them ideal for quantitatively analyzing Raman spectra without the need for complex isotopic labeling schemes and performing detailed MD simulations. Studies suggest that residues 20–29 form the amyloidogenic core of amylin.^48–50 Fibrils prepared from amylin_20–29 can be poised to adopt well-defined morphologies,^51–54 but there have been conflicting reports on whether it forms parallel, antiparallel, or mixed β-sheet fibril structures.^36,52–54 Wildtype Aβ consists of two predominate isoforms, the 40-residue Aβ_1–40 and the 42-residue Aβ_1–42. Although it is a less abundant isoform, Aβ_25–35 has also been found in the brain⁵⁵ and has been shown to aggregate into β-sheet structures.^56–59 Despite this, we are unaware of any known structures reported for Aβ_25–35 fibrils.

B. AFM imaging

AFM imaging (Fig. 1) reveals that the aggregated samples prepared from amylin_20–29 and Aβ_25–35 peptides resulted in the formation of amyloid fibrils. The two polymorphs produced from amylin_20–29 [Figs. 1(a) and 1(b)] exhibit similar morphologies, consisting of unbranched fibrils that are several micrometers in length. In contrast, the fibrils prepared from Aβ_25–35 are shorter, being only several hundred nanometers to a few micrometers in length.

C. Raman spectroscopy analysis

To further characterize the fibrils, we used Raman spectroscopy to investigate their molecular structures. As shown in Fig. 2, each polymorph exhibits unique spectral fingerprints in the region between 1100 and 1700 cm⁻¹. The most structurally informative Raman bands in the spectra occur in the amide I–III regions, located between 1600–1700, 1500–1600, and 1200–1350 cm⁻¹, respectively. We examined these regions in detail to obtain structural information about the fibrils by performing spectral deconvolution analysis (see the supplementary material for details, Figs. S4 and S5). The results of this analysis are summarized in Table I, which lists the Raman band assignments for the three polymorphs.

FIG. 2.

Raman spectra of amylin_20–29 and Aβ_25–35 fibrils. (a) Amylin_20–29 polymorph 1 fibrils. (b) Amylin_20–29 polymorph 2 fibrils. (c) Aβ_25–35 fibrils. All spectra were smoothed with a Savitzky–Golay filter⁶⁶ using a fourth-order polynomial over an 11-point window for visual clarity. The original spectra can be found in Fig. S3.

TABLE I.

Band frequencies and assignments for the amide I–III regions of amylin_20–29 and Aβ_25–35 fibrils.

Amylin20–29 (cm−1)a
Polymorph 1	Polymorph 2	Aβ25–35 (cm−1)a	Band assignmentb	References
1207	1207	⋯	Phe phenyl-C str	Asher et al.60
1220	⋯	1213	CH b, heavily mixed with NH ipb, CN str	Moore and Krimm61
1236	1225	1225	Amide III3 of fibrils	Mikhonin et al.62 and Punihaole et al.21
⋯	1241	1253	Amide III3 of non-fibrillar species	Mikhonin et al.62
1282c	1282	1279	CH b, heavily mixed with NH ipb, CN str	Moore and Krimm61
⋯	1555	1557c	Amide II	Barth63
⋯	1586	⋯	Phe CC ring str	Barth64
1605	1606	⋯	Phe ip ring str/Phe CC ring str	Asher et al.60 and Barth64
1633c	1630	1623	Amide I B of fibrils	Moore and Krimm61,65
⋯	1653	⋯	Amide I of non-fibrillar aggregates	Maiti et al.20 and Apetri et al.27
1671	1670	1671	Amide I A of fibrils	Moore and Krimm61,65
⋯	1677	⋯	Amide I of non-fibrillar aggregates	Maiti et al.20

^a Unless otherwise noted, the uncertainty is <±1 cm⁻¹.

^b ip: in-plane; b: bending; str: stretching.

^c Uncertainty is <±3 cm⁻¹.

The most intense features in the spectra shown in Fig. 2 occur in the amide I region. The amide I vibration mainly consists of amide C=O stretching.⁶³ Its structural sensitivity derives from transition dipole coupling between neighboring C=O oscillators that produce a delocalized amide I normal mode.^{61,65,67–70} Coupling also results in characteristic “excitonic splitting” patterns in the amide I band that are diagnostic of different protein secondary structure elements. For example, the amide I band for canonical (infinitely long) parallel β-sheets is predicted to split into two subbands, a high frequency A(0, 0) mode and a low frequency B(π, 0) mode that are both Raman and IR-active.^71–73 In contrast, the amide I band is predicted to split into four subbands for canonical (infinitely long) antiparallel β-sheet structures: the A(0, 0) mode, which is Raman-active and forbidden in the IR; the B₁(0, π) mode, which is relatively weak in both Raman and IR; the B₂(π, 0) mode, which is very strong in IR spectra; and the B₃(π, π) mode, which is essentially forbidden in IR and Raman spectra.^71–73 In practice, however, it can be difficult to differentiate between parallel and antiparallel β-sheet structures using the amide I band alone.⁷⁴ This is because the amide I band depends on additional factors, such as hydrogen bonding patterns of the peptide bond C=O groups, the registry of β-strands, and the twist of the β-sheets.^63,75–77

The intense band at ∼1670 cm⁻¹ can be assigned to the amide I A(0, 0) for all polymorphs. The B(π, 0) mode occurs near 1630 cm⁻¹ for both amylin_20–29 polymorphs and is downshifted to ∼1620 cm⁻¹ in Aβ_25–35 fibrils. We do not observe Raman bands in any of the fibril spectra that can be assigned to the B₁(0, π) mode characteristic of canonical antiparallel β-sheet structures. However, for amylin_20–29 polymorph 1, we observe a band at 1695 cm⁻¹ in the IR spectrum (Fig. S6), which we assign to the B₁(0, π) mode. The presence of this band suggests that these fibrils are composed of antiparallel β-sheet structures (vide infra). We attribute the absence of the B₁(0, π) mode in the Raman spectrum to the fact that there is likely some local disordering of the β-strand registries in the amylin_20–29 polymorph 1 fibrils.

Spectral deconvolution analysis of the amide I region of amylin_20–29 fibrils [Figs. S4(a) and S4(b)] suggests the presence of additional structures for polymorph 2 [Fig. S4(b)]. As discussed in detail in the supplementary material, modeling the amide I region between ∼1600–1700 cm⁻¹ of polymorph 2 with just two bands (the amide I A and B modes) yields a poor fit [Fig. S4(c)]. In addition, adding only the 1677 cm⁻¹ band improves the fit, but the amide I B bandwidth becomes unphysically large [Fig. S4(d)]. The most parsimonious, satisfactory, and physically reasonable fit that we obtain is by incorporating two additional bands at 1653 and 1677 cm⁻¹ into the fit. As discussed in detail below, we assign these bands to amide I modes deriving from α-helical (1653 cm⁻¹)^20,27 and PPII-like structures (1677 cm⁻¹) from non-fibrillar aggregates present in the sample.²⁰

The amylin_20–29 fibrils contain additional bands in the ∼1600–1700 cm⁻¹ region. Both amylin_20–29 polymorphs exhibit bands at ∼1605 cm⁻¹, while polymorph 2 exhibits an additional band at 1586 cm⁻¹. These bands can be assigned to the in-plane ring stretching modes of phenylalanine (Phe).^60,64 For amylin_20–29 polymorph 2, these bands exhibit narrow linewidths, which we attribute to the Phe rings adopting more well-defined conformations compared to polymorph 1 (vide infra). Amylin_20–29 fibrils are also expected to contain spectral contributions from asparagine’s (Asn) side chain C=O stretching mode.^24,78,79 We see no evidence that the Asn residues of amylin_20–29 fibrils contribute significant spectral contributions to the 1600–1700 cm⁻¹ region. However, to confirm that our assignment of ∼1670 cm⁻¹ bands in amylin_20–29 fibrils is correct, we measured the Raman spectrum of a lyophilized powder of Asn. The Raman spectrum of Asn (Fig. S7) exhibits a strong band at 1640 cm⁻¹, which can be assigned to the C=O stretching mode of Asn side chains. This suggests that the spectral contributions of Asn side chains are negligible for the ∼1670 cm⁻¹ bands that we assign to amide I A(0,0) of the peptide backbone amide groups.

The amide II mode gives rise to a band located in the 1500–1600 cm⁻¹ region. It consists of an out-of-phase combination of NH in-plane bending and CN stretching. The amide II is typically strong in IR spectra, but weak in Raman spectra excited with visible wavelengths. Interestingly, Lee and co-workers¹⁹ suggested that amide II is enhanced in the Raman spectra of parallel β-sheet structures. As seen in Fig. 2, amide II is suppressed in the spectrum of amylin_20–29 polymorph 1 [Fig. 2(a)], but appears in the spectra of the other two fibril polymorphs [Figs. 2(b) and 2(c)]. This suggests that amylin_20–29 polymorph 2 and Aβ_25–35 fibrils consist of parallel β-sheet structures.

The most structurally sensitive bands in the Raman spectra shown in Fig. 2 occur between ∼1200–1350 cm⁻¹ in the amide III region. The canonical amide III band, as characterized in N-methylacetamide, occurs at ∼1315 cm⁻¹ and originates from a vibration consisting of an in-phase combination of NH in-plane bending and CN stretching.⁸⁰ In peptides and proteins, however, the amide III region is considerably more complex, consisting of several bands that derive from vibrations containing significant contributions of CN stretching, NH bending, and/or C_αH bending motions. Asher and co-workers⁸¹ assigned the amide III region of peptides and proteins in detail, identifying three subbands called amide III₁, amide III₂, and amide III₃. They have shown that amide III₃ is the most conformationally sensitive,⁶² as its frequency can be correlated to the Ramachandran ψ-dihedral angles of peptide bonds.^62,82

For β-sheet structures, the amide III₃ band occurs between ∼1220–1240 cm⁻¹, and it is easily identified due to its relatively strong intensity (even with visible Raman excitation) compared to other bands in the region.⁸³ Based on peak intensities from our spectral deconvolution analysis (Figs. S4 and S5), we assign the 1236 cm⁻¹ band for amylin_20–29 polymorph 1, the 1225 cm⁻¹ band for amylin_20–29 polymorph 2, and the 1225 cm⁻¹ band for Aβ_25–35 fibrils to amide III₃ modes that are diagnostic of amyloid fibril β-sheet structures.^21,23,25,84

We capitalized on the structural sensitivity of the amide III₃ band to determine the distribution of Ramachandran ψ-angles for the amylin_20–29 and Aβ_25–35 fibril peptide bonds. To do this, we utilized the methodology of Asher and co-workers,⁸⁵ which correlates the frequencies of the amide III₃ band envelope to different ψ-angles (see the supplementary material for details). As shown in Fig. 3, the ψ-angles for all the fibril polymorphs occur between ∼120° and 160°. The distribution for amylin_20–29 polymorph 1 [Fig. 3(a)] is centered at 151°, well within the range of canonical antiparallel β-sheet structures.⁸⁶ In contrast, the distributions for the other polymorphs are centered at ∼140° [Figs. 3(b) and 3(c)], indicating that they are composed of parallel β-sheet structures.⁸⁶ Based on its narrower ψ-angle distribution width, amylin_20–29 polymorph 2 fibrils exhibit a more well-defined conformation than amylin_20–29 polymorph 1 or Aβ_25–35 fibrils. Additionally, we observe no ψ-angle populations in the fibrils that suggest there are β-turn structures. This indicates that all three polymorphs consist of extended β-strands that assemble into β-sheets.

FIG. 3.

Ramachandran ψ-angle distribution determined from the amide III₃ mode for amylin_20–29 and Aβ_25–35 fibrils. (a) Amylin_20–29 polymorph 1 Ramachandran ψ-angle distribution with a mean ψ-angle of 151°. (b) Amylin_20–29 polymorph 2 Ramachandran ψ-angle distribution with a mean ψ-angle of 139°. (c) Aβ_25–35 fibrils Ramachandran ψ-angle distribution with a mean ψ-angle of 139°. The distributions indicate that the cross-β core structure of (a) is composed of extended β-strands oriented in an antiparallel β-sheet, while (b) and (c) are composed of extended β-strands oriented in a parallel β-sheet.

We observe additional amide III₃ bands between ∼1240–1250 cm⁻¹ for amylin_20–29 polymorph 2 and Aβ_25–35 fibrils. The frequencies of these bands are upshifted beyond the range that is typical of amyloid β-sheet structures and likely derive from non-fibrillar aggregates that we observe in some of the AFM images for these two polymorphs [Figs. S2(c)–S2(f)]. These bands correspond to ψ-angle distributions centered at either 156° (168°) or −28° (−40°) for amylin_20–29 polymorph 2 (Aβ_25–35) fibrils (Fig. S8). Based on our assignments of the amide I bands at 1653 and 1677 cm⁻¹ (vide supra), the ψ-angle distributions for amylin_20–29 polymorph 2 centered at 156° and −28° likely derive from PPII-like/disordered and α-helical structures, respectively. In contrast, our analysis of the amide I region for Aβ_25–35 fibrils suggests that there are no α-helical or disordered structures present (Fig. S5). Thus, the amide III₃ band located at 1253 cm⁻¹ likely corresponds to non-fibrillar aggregates that are rich in β-strand-like structures with ψ-angles centered at −168°.

D. Polarized Raman measurements

We used polarized Raman spectroscopy to gain additional structural insights on amylin_20–29 and Aβ_25–35 fibrils. Raman anisotropy measurements on aligned samples can be employed to determine the relative orientation of chemical bonds and functional groups in fibrils, as first demonstrated by Lednev and co-workers.²² We prepared anisotropic samples of aligned fibrils for each polymorph using a drop coat deposition method.^22,37 This method creates a “coffee ring” peptide film that forms because the evaporation of the solvent produces shear forces that move fibrils in the center of the droplet toward the perimeter.³⁷ As the fibrils approach the perimeter of the droplet, they align themselves parallel to its edge with a high degree of ordering.⁸⁷

We aligned the incident laser light to the edge of the coffee ring for the polarized Raman measurements. We designated a laboratory coordinate system (XYZ) where the Z-direction corresponds to the long axis of the fibrils (tangent to the coffee ring edge) and the Y axis to the propagation direction of the incident laser light (Fig. 4, inset). Given the uniaxial symmetry of the fibrils, we only needed to acquire Raman spectra using four different combinations of incident and scattered light polarization configurations to obtain orientation information of chemical functional groups: ZZ, ZX, XZ, and XX (where the first and second letter indicate the direction of the incident and scattered light, respectively, along the laboratory coordinate system).³⁸

FIG. 4.

Polarized Raman spectra of amylin_20–29 and Aβ_25–35 fibrils. (a) Amylin_20–29 polymorph 1. [(a), inset)] Schematic of polarization configurations. (b) Amylin_20–29 polymorph 2. (c) Aβ_25–35 fibrils. The spectra shown here have not been smoothed.

The polarized Raman spectra of the three different fibril polymorphs are shown in Fig. 4. As expected, the cross-polarized spectra (ZX and XZ, shown in blue and purple, respectively) overlap almost perfectly, indicating that there is no photodamage of the samples, no artifacts introduced by the different configurations of the polarization optics used (see Sec. II for more details), and that no displacements/rotations occurred in the samples during the measurement. The ZZ (black) and XX (red) spectra show the largest intensity variations. The most striking difference occurs with the amide I A(0, 0) band, which is most intense in the ZZ spectrum and very weak in the XX spectrum.

Quantitatively interpreting the polarized spectra to extract orientation information requires knowing the Raman tensors for particular vibrational normal modes. Fortunately, the Raman tensor for the (delocalized) amide I A(0, 0) mode has been determined. It is oriented parallel to the C=O axis of peptide bonds in β-sheet structures.⁸⁸ Thus, the amide I A(0, 0) band is an ideal spectroscopic marker to determine the relative orientation of C=O bonds in amylin_20–29 and Aβ_25–35 fibrils.

The theory to obtain orientation information from polarized Raman spectra has been described in extensive detail elsewhere.^89–92 The key step is to determine the respective most probable orientation distribution functions, N_mp(θ), for the amide I A(0, 0) tensor of each polymorph. The method of finding N_mp(θ) is described in the section Basic Approach to Determine the Most Probable Orientation Distribution Function of the supplementary material. Briefly, for systems such as fibrils with uniaxial symmetry, N_mp(θ) can be estimated with reasonable accuracy by experimentally measuring two order parameters called ⟨P₂⟩ and ⟨P₄⟩.⁹³ As discussed in detail in the supplementary material, these order parameters can be determined from the intensity ratios R₁ = I_ZX/I_ZZ and R₂ = I_XZ/I_XX for the amide I A(0, 0) mode (see Tables S1 and S2 and Figs. S9–S17).⁸⁹

Figures 5(a)–5(c) show N_mp(θ) for the amide I A(0, 0) tensors. The distributions are normalized such that multiplying N_mp(θ) by sin θ yields the preferred orientation distribution of the tensors. As shown in Figs. 5(d)–5(f), the preferred orientation distributions of the amide I A(0, 0) tensors for all the polymorphs are bimodal. The maximum probability for the distributions corresponding to the amide I A(0, 0) tensors occur at ∼±10° and ±15° for amylin_20–29 polymorphs 1 and 2, respectively, and ±11° for Aβ_25–35 fibrils (Table II). These values indicate that the β-strands of the polymorphs are oriented approximately perpendicularly to the fibril long axis, which is the hallmark of the cross-β architecture observed in amyloids. The smaller peaks in the C=O bond angle distributions that occur at 90° can be attributed to local disordering of the cross-β structure or because of misalignment of the fibrils filaments in the coffee rings that we prepared. Based on the relative intensities of the peaks, disordering and misalignment account for roughly 19% for both amylin_20–29 polymorphs and 33% for Aβ_25–35 fibrils of the total probability for the C=O bond angle distributions.

FIG. 5.

The most probable orientation distribution function (a)–(c), N_mp(θ), and the preferred orientation distribution (d)–(f), N_mp(θ) sin(θ), for the amide I A(0, 0) Raman tensors for amylin_20–29 and Aβ_25–35 fibril polymorphs. θ defines the angle of the tensor or bond with respect to the long axis of the fibril. (a) and (d) Amylin_20–29 polymorph 1. (b) and (e) Amylin_20–29 polymorph 2. (c) and (f) Aβ_25–35 fibrils.

TABLE II.

Summary of Raman-derived structural constraints for amylin_20–29 and Aβ_25–35 fibrils.

	β-sheet	ψ angle (deg)	κ for ψ [(J/mol)/deg]	C=O bond angle (deg)	κ for C=O [(J/mol)/deg]
Amylin20–29 polymorph 1	Antiparallel	151	40.54	±10	43.07
Amylin20–29 polymorph 2	Parallel	139	258.6	±15	18.63
Aβ25–35 fibrils	Parallel	139	52.67	±11	32.72

E. Determining force constants for harmonic constraints

The widths of the experimentally measured C=O and ψ-angle distributions can be used to determine harmonic constraints that can be applied in MD simulations. For example, the probabilities of the experimentally measured ψ-angles can be modeled using the Boltzmann distribution,

$$\frac{p({\psi }_i)}{p({\psi }_{eq})}=e^{-\mathrm{\Delta }G({\psi }_i)/RT},$$ (1)

where p(ψ_i)/p(ψ_eq) is the ratio of peptide bonds with ψ-angles angles, ψ_i and ψ_eq, respectively. The angle, ψ_eq, is the “equilibrium” or minimum energy ψ-angle angle for reach polymorph, while R is the molar gas constant, T is the temperature, and ΔG(ψ_i) = G(ψ_i) − G(ψ_eq) is the apparent Gibbs free energy difference between ψ_i and ψ_eq. From Eq. (1), ΔG(ψ_i), can be determined,

$$\mathrm{\Delta }G({\psi }_i)=-RT\ln \left[\frac{p({\psi }_i)}{p({\psi }_{eq})}\right].$$ (2)

Equation (2) can be used to determine an apparent free energy landscape for the fibril peptide bonds along the ψ-angle structure coordinate. As shown in Fig. 6, the apparent energy landscapes behave harmonically around ψ_eq and can be modeled in terms of a simple torsional spring using Hooke’s law,²⁴

$$\mathrm{\Delta }G({\psi }_i)=\frac{1}{2}\kappa {({\psi }_i-{\psi }_{eq})}^2,$$ (3)

where κ is the torsional spring force constant, which reflects the curvature of the harmonic potential wells. A similar approach can also be used to determine the torsional spring constants for the C=O bond angles. These distributions are more complex, and, as shown in Fig. 7, the potential energy landscapes are anharmonic. To calculate the force constants, however, we assume that the potential wells behave harmonically in the region between 5° and 20°, near the equilibrium bond angles.

FIG. 6.

Apparent Gibbs free energy landscapes (shown in black circles) for amylin_20–29 and Aβ_25–35 fibrils Ramachandran ψ-angle. The energy wells can be modeled by assuming a harmonic oscillator model torsional spring near the equilibrium angles (shown in red). (a) Amylin_20–29 polymorph 1. (b) Amylin_20–29 polymorph 2. (c) Aβ_25–35 fibrils.

FIG. 7.

Apparent Gibbs free energy landscapes (shown in black circles) for amylin_20–29 and Aβ_25–35 polymorph C=O bond angles. The energy wells can be modeled by assuming a harmonic oscillator model torsional spring near the equilibrium angles (shown in red). (a) Amylin_20–29 polymorph 1. (b) Amylin_20–29 polymorph 2. (c) Aβ_25–35 fibrils.

Table II summarizes the force constants obtained from the experimentally measured ψ- and C=O angle distributions. It is important to note that fibril misalignment introduces additional uncertainty in determining the force constants for the C=O bond angles, but not the Ramachandran ψ-angles. The effect of this greater uncertainty is that the apparent force constants determined experimentally for the C=O bond angles are lower than in actuality. This means that the structural constraints applied in MD simulations are relaxed and allow for a greater ensemble of fibril structures to be sampled in the trajectories, effectively reducing the apparent resolution of our structural models.

F. Guided MD simulations using experimental constraints from Raman spectroscopy

We recognized that the C=O and ψ-angle distributions obtained from Raman measurements provide structural constraints that can be used to determine molecular structural models of the amylin_20–29 and Aβ_25–35 fibril polymorphs. To accomplish this, we first manually constructed putative structures of the three different fibril polymorphs for MD simulations. These initial models were constructed by arranging 24 peptides in a two β-sheets comprised of 12 peptides per sheet. Using the Raman data as a guide, we constructed β-sheets comprised of only extended β-strands that were arranged in an antiparallel configuration for amylin_20–29 polymorph 1 and parallel configurations for amylin_20–29 polymorph 2 and Aβ_25–35 fibrils. When initially constructing these models, we only considered β-sheet structures wherein the strands were in-register and explicitly solvated these structural models in TIP3P⁹⁴ before performing energy minimization and equilibration runs. Upon performing the equilibration runs, we found that the Aβ_25–35 model was structurally unstable since water molecules penetrated the interior of the β-sheet and formed strong dipole-charge interactions with the lysine (Lys) side chains.

Because of this, we employed a generalized-Born implicit solvent with a dielectric constant set to 3.23 (appropriate for the interiors of proteins)⁹⁵ for all three structural models. We reasoned that the use of implicit solvent was physically reasonable for our simulations since it better mimics the bulk properties of real fibril systems that were examined experimentally compared to the small, well-solvated β-sheet oligomers in the explicit solvent simulations. However, following the equilibration of these GBIS solvated models, we found that the Aβ_25–35 structure was still not thermodynamically stable. Upon analysis of this structure, we observed that the in-register arrangement of the β-strands resulted in repulsive electrostatic interactions of Lys side chains, which caused the β-sheets to dissociate. To alleviate this problem, we offset the β-strands in the sheets by one residue. We found that this resulted in thermodynamically stable structures, as it allowed the Lys residues to form favorable hydrogen bonds with the nearby Asn and serine (Ser) side chains (vide infra), rather than the repulsive interactions that we previously observed.

Using an analogous approach to ssNMR, we restrained the C=O and ψ-angles of the fibril models during the energy minimization and equilibration simulation runs using the equilibrium angles and torsional force constants determined experimentally from our Raman measurements. Figure 8 represents representative snapshots from the resulting ensemble of fibril structures for the three polymorphs from the MD simulations that are consistent with the experimental data from the Raman measurements. All three polymorphs are composed of extended β-strands that assemble into sheets. The strands for amylin_20–29 polymorph 1 assemble into antiparallel β-sheets, while those of polymorph 2 and Aβ_25–35 form parallel β-sheet structures. For amylin_20–29 polymorph 1, hydrophobic contacts occur between the side chains of Phe, alanine (Ala), and isoleucine (Ile) residues within the same β-sheet, while the side chains of leucine (Leu), Ala, and Ile form hydrophobic zippers between β-sheets [Fig. S18(a)].

FIG. 8.

Representative structural snapshots of fibrils obtained from MD simulations using Raman experimental constraints. (a) Amylin_20–29 polymorph 1. (b) Amylin_20–29 polymorph 2. (c) Aβ_25–35 fibrils.

In contrast, the side chains of Ala, Ile, and Leu residues are aligned down the length of the β-sheets for polymorph 2, forming a string of hydrophobic interactions between neighboring residues and hydrophobic zippers between opposite β-sheets, similar to polymorph 1 [Figs. S18(b) and S18(c)]. In addition, the side chains of Phe residues between neighboring β-strands in polymorph 2 form π-stacking interactions with each other. This constrains the Phe rings so that they adopt well-defined conformations, which is confirmed by the narrow bands observed for the Phe rings in the Raman spectrum shown in Fig. 2(b).

For Aβ_25–35, hydrophobic interactions occur between the side chains of Ala an Ile residues within the same β-sheet, as well as Leu and Ile residues from opposite β-sheets [Fig. S19(a)]. In addition, hydrogen bonding interactions are observed between Ser, Asn, and Lys residues between neighboring β-strands [Fig. S19(b)].

All three polymorphs obtained from the MD simulation exhibit structural features that are physically reasonable for amyloid fibrils. The average inter-sheet distances for the three structures are between 9.4 and 10.4 Å, and the average inter-strand spacings are between 4.9 and 5.1 Å (Table S3). These values are in good agreement with the equatorial and meridional spacings observed in experimental fiber diffraction data of amyloid fibrils.⁹⁶ Similarly, no unrealistic bond angle distortions or steric clashes are observed, and the (ψ, φ, ω) Ramachandran angles are all within allowed values (Tables S4–S6). As a result, we have greater confidence that the structural models shown in Fig. 8 are physically reasonable.

To further validate the models, we examined whether the simulated fibrils were structurally stable. We tested this by removing the structural constraints on the fibril structures and then running additional simulations. The unconstrained structures held together during the simulation production runs and did not dissociate. The C=O bond distributions (Fig. S20) of the unconstrained structures are bimodal, exhibiting peaks around ±10°–15° from the fibril axes, in good agreement with our experimental data. Similarly, the median values for the ψ-angle distributions (Fig. S21) are close to our experimental measurements. The distribution of amylin_20–29 polymorph 1 is unimodal and peaked near 147°. In contrast, the distributions for amylin_20–29 polymorph 2 and Aβ_25–35 appear bimodal, exhibiting a main peak around 139° (close to experimental measurements) and a smaller peak that is downshifted to ∼120°. This indicates that the fibril structures corresponding to amylin_20–29 polymorph 2 and Aβ_25–35 have both slightly evolved compared to the original constrained structures. The reason for these discrepancies is likely due to the limitations of our simulations since we could only model a single protofilament segment. We hypothesize that the structures we simulated are more conformationally flexible (thus allowing the ψ-angle distributions to evolve) since they are not subjected to lattice packing forces of multiple protofilaments that stabilize real amyloid fibrils.

G. Comparison with other fibril models

Early studies utilizing ssNMR and IR spectroscopy by Landsbury and Griffin^52,53 suggested that amylin_20–29 fibrils are composed of antiparallel β-sheet structures. These finding were later corroborated by Nielsen and co-workers⁵⁴ who also employed ssNMR to investigate the structure of amylin_20–29 fibrils. The Landsbury–Griffin–Nielsen studies, however, have been contradicted by others, who suggest that amylin_20–29 fibrils can also form parallel β-sheet structures. Middleton and co-workers³⁶ employed ssNMR and x-ray fiber diffraction to investigate amylin_20–29 fibrils. Their analysis of ¹³C cross-polarization magic angle spinning (MAS) and RR ssNMR data revealed that amylin_20–29 can form both parallel and antiparallel β-sheet fibril polymorphs, despite electron micrographs of their fibril samples appearing morphologically homogeneous. Similarly, Song et al.⁹⁷ published a recent cryo-EM study, which also suggests that amylin_20–29 fibrils can adopt parallel β-sheet structures.

Our work corroborates the findings of Middleton and co-workers³⁶ that amylin_20–29 forms both parallel and antiparallel β-sheet fibril polymorphs. Furthermore, the structures that we determine for both polymorphs exhibit hydrophobic zippers between the side chains of Leu, Ala, and Ile that are consistent with ssNMR data obtained by both the Middleton and Nielsen groups.^36,54 In addition, the β-sheets of both polymorphs also exhibit a slight twist around their fibril axes, in agreement with the antiparallel β-sheet structure reported by Nielsen and co-workers.⁵⁴ This agreement between our structural models and those derived from ssNMR data highlights the utility and complementary nature of using Raman spectroscopy to quantitatively assess the molecular structures of different amyloid fibrils.

Regarding Aβ_25–35, we are unaware of any detailed fibril structures being reported in the literature. In the case of the full-length (1–40 or 1–42) wildtype Aβ peptide, residues 25–29 adopt a bend structure that brings the two β-sheets of the fibril protofilaments together.¹⁸ In contrast, Aβ_25–35 is too small to adopt bend or turn structures. Its cross-β core structure resembles that of amylin_20–29 polymorph 2. Interestingly, Aβ_25–35 fibrils appear to be more structurally heterogeneous than amylin_20–29 polymorph 2 fibrils, as evidenced by the former’s broader ψ-angle distribution [cf. Figs. 3(b) and 3(c)]. This may be due to the difference in side chain composition of the two peptides. Amylin_20–29 polymorph 2 adopts parallel β-sheets that allow Phe, Ala, and Leu side chains to be aligned in-register between neighboring β-strands, thereby allowing hydrophobic and π-stacking interactions that make the fibril structures more rigid. The MD simulations suggest that this does not occur for Aβ_25–35 fibrils since the side chains hydrophobic residues in the interior between the two β-sheets are more disordered (Fig. S19). This side chain disordering could give rise to the greater conformational heterogeneity observed in the Raman spectra for Aβ_25–35 fibrils compared to amylin_20–29 polymorph 2.

V. CONCLUSION

In this study, we present detailed molecular-level structural models of amylin_20–29 and Aβ_25–35 fibrils determined using MD simulations that are based on C=O bond and ψ-dihedral angle constraints measured by Raman spectroscopy. Agreement between our fibril models of amylin_20–29 polymorphs 1 and 2 with those reported by the Landsbury,⁵² Griffin,⁵³ Middleton,³⁶ and Nielsen⁵⁴ groups highlights the potential to develop quantitative molecular structural models of fibrils using Raman spectroscopy. It also highlights the fact that our approach provides structural information that complements gold-standard techniques, such as ssNMR and cryo-EM.

We believe that using experimental parameters measured by Raman spectroscopy to guide MD is more powerful and synergistic than using either technique alone. Namely, the bond and dihedral angle parameters measured by Raman spectroscopy provide structural constraints that help guide the construction of starting models and narrow the conformational phase space sampled over the course of the simulation production runs. In addition, MD simulations provide a powerful method to visualize the three-dimensional structure of amyloid fibrils in a way that cannot be appreciated by only inspecting the bond and dihedral angle distributions measured by Raman spectroscopy. Finally, our work shows that Raman spectroscopy can be used to quantitatively discriminate between different fibril polymorphs.

We believe that our approach could be useful in refining fibril structures determined by ssNMR or cryo-EM. This is particularly true for refining disordered or dynamic regions of fibrils, which are difficult to probe using ssNMR and cryo-EM, that could play important roles in initiating aggregation or aberrantly interacting with biological cells. For ssNMR, the final ensemble of structures is visualized by selecting the top 10 or 20 most energetically favorable protein conformers consistent with the experimental constraints applied in MD simulations. In cryo-EM, disordered, dynamic, or heterogeneous regions could result in poorly resolved electron densities that make it difficult to obtain any structural information at all. In contrast, the widths of dihedral and bond angle distributions measured by Raman spectroscopy naturally give insights into the structural dynamics and conformational heterogeneity of fibrils. This information can be directly incorporated into simulations by parameterizing MD force fields with experimentally measured harmonic force constants, as described above.

The Raman structural constraints used to parameterize the MD simulations can theoretically be determined using either non-polarized or polarized measurements. Although measuring the C=O bond angle distributions require polarization measurements on aligned samples, determining parameters such as ψ-dihedral angles with Raman spectroscopy does not. Thus, in theory, our approach does not inherently rely on aligned fibril samples or polarization measurements. However, for this to be feasible, more structurally sensitive Raman spectroscopic markers need to be discovered to enable the determination of higher-resolution fibril models. There are several reports describing Raman spectroscopic markers that can be used to measure amino acid side chain dihedral angles. However, these apply to only a handful of amino acids, such as glutamine, asparagine, and tryptophan.^{21,24,25,98,99} Discovering a Ramachandran φ-dihedral angle spectroscopic marker would be especially powerful for constraining the structure of the peptide backbone in fibrils. Work by Schweitzer-Stenner and co-workers^100,101 describes an interesting approach to analyze the amide I band to determine both Ramachandran ψ- and φ-angles. However, it remains unclear whether their methodology can be applied to systems beyond tetrapeptides.

Obtaining site-specific structural information would also be useful in developing our technique. Residue-specific structural information could conceivably be obtained through site-specific labeling schemes of amide nitrogen and carbon atoms. Since the pathophysiologically relevant lengths of many amyloidogenic proteins, including amylin, are generally less than 50 amino acids, isotopic editing of peptides could be achieved using solid-phase synthesis.^70,102,103 Isotopic labeling of the peptide backbone carbon and nitrogen atoms would decouple the NH bending, CN stretching, and C=O stretching motions in amide vibrations.⁸¹ Isolating the corresponding amide I and III modes for individual residues could therefore be achieved by determining the difference spectra between the isotopically labeled and unlabeled fibril species.⁸¹ Thus, with these current limitations, it is clear that follow-up studies are needed to increase the utility of our methodology. Despite this, however, we believe that our approach lays a foundation toward potentially using Raman spectroscopy with MD to visualize the three-dimensional structures of amyloid fibrils and other biological macromolecules.

SUPPLEMENTARY MATERIAL

Tables S1–S6 and Figs. S1–S21, the description of additional Raman spectroscopy methods, spectral processing and peaking fitting methods, AFM methods, and FTIR methods, and the determination of Ramachandran ψ-angle distribution and order parameters.

AUTHOR DECLARATIONS

Conflict of Interest

The authors have no conflicts to disclose.

Author Contributions

Madeline Harper: Conceptualization (supporting); Formal analysis (equal); Investigation (lead); Methodology (equal); Software (equal); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). Uma Nudurupati: Formal analysis (supporting); Investigation (supporting); Methodology (supporting); Visualization (supporting); Writing – review & editing (supporting). Riley J. Workman: Formal analysis (supporting); Investigation (supporting); Methodology (supporting); Writing – review & editing (supporting). Taras I. Lakoba: Formal analysis (supporting); Software (equal); Writing – review & editing (supporting). Nicholas Perez: Investigation (supporting). Delaney Nelson: Investigation (supporting). Yangguang Ou: Funding acquisition (equal); Resources (supporting); Supervision (supporting); Writing – review & editing (equal). David Punihaole: Conceptualization (lead); Formal analysis (equal); Funding acquisition (equal); Methodology (equal); Project administration (lead); Resources (lead); Software (lead); Supervision (lead); Visualization (equal); Writing – review & editing (lead).

DATA AVAILABILITY

The data that support the findings of this study are available from the corresponding author upon reasonable request.