Jiří
Kessler
ab,
Shigeki
Yamamoto
c and
Petr
Bouř
*a
aInstitute of Organic Chemistry and Biochemistry, Academy of Sciences, Flemingovo náměstí 2, 16610 Prague, Czech Republic. E-mail: bour@uochb.cas.cz
bDepartment of Physical and Macromolecular Chemistry, Faculty of Science, Charles University, Hlavova 8, 12840 Prague, Czech Republic
cDepartment of Chemistry, Graduate School of Science, Osaka University, Osaka 560-0043, Japan
First published on 11th May 2017
Folding of proteins into insoluble amyloidal fibrils is implicated in a number of biological processes. Optical spectroscopy represents a convenient tool to monitor such structural variations. Recently, characteristic changes in Raman optical activity (ROA) spectra of insulin during a pre-fibrillar stage were reported but not supported by a theoretical model. In the present study, molecular dynamics and the density functional theory are used to simulate the spectra and understand the connection between the structure, and ROA and Raman spectral intensities. Theoretical results are consistent with the observations and only confirm exceptional ROA sensitivity to the protein tertiary structure. Surprisingly, this sensitivity reflects local conformational changes in the peptide main and side chains, rather than a direct through-space interaction of the protein components. Side chains providing strong ROA signals, such as tyrosine, can additionally report on local conformational features. Theoretical modeling helps in explaining the observed spectral changes and is likely to enable future applications of ROA spectroscopy in protein structural studies.
Insoluble protein precipitates are difficult to study by standard high-resolution techniques; the structure is not regular enough for standard X-ray diffraction and provides a mediocre signal for nuclear magnetic resonance spectroscopy.4 Vibrational optical activity (VOA), including spectroscopies of vibrational circular dichroism (VCD) and Raman optical activity (ROA), has been suggested as an alternative technique that is extremely sensitive to protein structural variations.5,6 Indeed, VCD spectroscopy could detect not only the formation of insulin fibrils, but also subtle changes in their macroscopic helical twist caused by pH variation.7
The ROA technique, although in principle capable of capturing a wider range of molecular vibrations than VCD, has not been extensively used yet for fibrils due to experimental artifacts8 inherent to inhomogeneous samples. The problem was recently overcome for a pre-fibrillar state of insulin, where ROA spectra could be recorded on an incident circularly polarized (ICP) ROA spectrometer.9 Amyloidal clear or milky gels were incubated from a solution of bovine insulin and 0.1 M hydrochloric acid at 82 °C. After several hours at 22 °C the amyloid/fibrils refolded to their native form, and the whole process could be monitored spectroscopically.10 In the future, ROA spectroscopy is thus expected to provide a more detailed insight into the fibrils' world. However, rather minute changes in the spectra have been interpreted predominantly empirically or only on simplified systems.
In the present study we develop a theoretical basis to interpret these experimental results, by comparing them to density functional theory simulations of Raman and ROA intensities. This is a challenging task because the amyloid insulin structure is unknown at atomic resolution. In the past, the Cartesian coordinate transfer (CCT)11,12 technique enabled us to model spectral intensities of the native insulin form13 or even of larger globular proteins14 with unprecedented precision. In these cases, however, the geometries in solutions were supposed to be rather rigid and close to the published X-ray structures.
Fortunately, earlier studies indicate the most likely insulin secondary and tertiary structure in the fibrils as well. On a simplified insulin peptide sequence, a parallel β-sheet structure has been identified by X-ray as the basis of fibril conformation.15 The parallel β-sheets are also formed by proteins of similar size and sequences to insulin.16,17 Finally, ROA and Raman spectra of fibrous insulin resemble those of other proteins forming parallel β-sheets.10 In the present study, β-sheet structures derived from X-ray geometries of other proteins and molecular dynamics (MD) simulations are jointly used to model the insulin structure.
Quantum-mechanical simulations of ROA spectral intensities stimulated this kind of spectroscopy, making it a reliable tool to determine the absolute configuration18,19 and conformation20 of biomolecules. In particular, spectral computations within the density functional theory (DFT) provide sufficient precision at acceptable computational cost.21,22 They are based on a perturbation treatment of the interactions of molecules with circularly polarized light;23,24 the origin-independence of results obtained at approximate computational levels is typically ensured by field-dependent atomic orbitals.25 Normally, a harmonic approximation26,27 is used for vibrational frequencies, although anharmonic extensions are possible as well.28
Nevertheless, in spite of efficient implementations22,29 the direct “ab initio” simulations (including DFT) become very lengthy indeed for larger molecules.30 Therefore, various simplifications were suggested in the past, such as the intensity-carrying normal mode algorithm31 and the molecules-in-molecules (MIM) fragment-based approach.32 We use the CCT methodology11,12 allowing one to efficiently simulate large-molecule spectra14 as well as to average a large number of MD snapshots with a minimal loss of DFT accuracy.13
As we show in the present study, the method enabled us to link the observed spectral features to the protein structure and the native → fibril conformational change. By several computational experiments we could also test the sensitivity of ROA spectra to finer structural variations, investigate the role of contact (non-covalent) interactions of peptide chains, identify aromatic marker bands sampling the local geometry, and test other details of the computational methodology. Although primarily aimed at the insulin fibril experiment, the results confirm the potential of ROA spectroscopy to sense protein structural variations, including subtle conformational changes.
Other MD runs were performed within the Tinker program36 modified to simulate the macroscopic twist.37 The X-ray derived insulin geometry was placed in a rectangular box (12 Å × 110 Å × 110 Å) filled with water molecules. Helical periodic boundary conditions37 were applied to allow for a minor (0°–6°) twist between neighboring insulin units. The Amber99 force field38 was used in the NVT run with 1 fs integration time. During a 0.1 ns production run 100 snapshots were selected each ps and the spectra were generated as in the previous case.
The fragments were capped by methyl groups and subjected to partial optimization in vibrational normal mode coordinates,41 fixing the limiting normal frequency42 to 100 cm−1. This lower value allowed for a more extensive relaxation of the geometries and provided slightly better spectra (Fig. S4, ESI†) than the limit of 300 cm−1 used as default in the past.13 Program Qgrad41 was used for the normal mode optimization; the program is interfaced with the Gaussian43 software. All quantum chemical computations on the fragments were carried out using the B3PW9144 functional providing excellent results for protein ROA,45 the standard 6-31++G** basis set and the conductor-like polarizable continuum solvent model (CPCM)46,47 with water parameters accounting for both protein and aqueous environments.
For the optimized fragments, the harmonic force field (second energy derivatives) and derivatives of the α, G′ and A polarizability tensors23,24 were calculated using the Gaussian and transferred to the insulin molecule, and then back-scattered Raman and ICP24 ROA intensities were generated as usual.24,39 From the intensities, spectral curves were generated by a convolution with Lorentzian bands; full width at half maximum was set to 10 cm−1.
The simulation (lower part of Fig. 2) reproduces many of these observations, sometimes in surprising details. The CO stretching frequencies of amide I and carboxyl (experimentally within 1659–1728 cm−1) are calculated to be very high, which is common for modeling at this level and has been identified as an error inherent to the dielectric solvent model.51,52 In the Raman spectra, the observed shift in an average amide I frequency (1659 → 1674 cm−1 for native → fibril transition) is reproduced reasonably well, as 1730 → 1742 cm−1. In ROA, the fibril form exhibits a sharp 1660(−)/1674(+) “couplet” in the experiment, which is well reproduced by the theory at 1728/1740 cm−1. This seems to be a typical feature of β-sheet structures also observed in globular proteins rich in β-sheets (e.g. concanavalin A),53 although model simulations indicate a significant dependence on detailed geometry, such as the β-sheet twist.20 In the native form, the amide I ROA signal is broader and mostly positive. The experimental Raman 1617 cm−1 band (calculated at 1669 cm−1, cf.Table 1) belongs to C
C stretching vibrations in aromatic tyrosine residues, and is accompanied by a close band (“shoulder”, experimentally at ∼1608 cm−1) due to analogous phenylalanine vibrations. These “aromatic” vibrations generate a fairly strong negative ROA signal at 1620 cm−1 for the amyloid, reproduced by the simulation at 1657 cm−1.
Calculated | Experimental | Assignmenta | |
---|---|---|---|
Native | Fibril | ||
a ν and δ denote stretching and bending modes, respectively, “s” in the band frequency denotes a shoulder, oop – out of plane bending. b Approximate local symmetry with respect to the benzene (D6h) ring. | |||
1790 | 1728s | 1728s |
ν(C![]() |
1742 | 1659 | 1674 |
ν(C![]() |
1669 | 1617 | 1617 | Tyr ν(C![]() |
1662s | 1608 | 1609s | Phe ν(C![]() |
1557 | 1537 | — |
ν(N–CO) + δ(N–H) (amide II), ν(C![]() |
1522 | 1460s | 1460s | His def. |
1477 | 1446 | 1446 | δ(C–H), sciss. |
1408 | — | 1395 | Amide III, δ(C–H) |
1352 | 1345 | 1345 | Tyr/Phe δ(C–H) |
1336 | 1345 | 1345 | δ(αC–H) |
1306 | — | 1271 | δ(αC–H) |
1290 | 1264 | 1247 | Tyr δ(C–H), ν(C–O) |
1239 | 1240 | 1233 | Amide III |
1230 | 1205 | 1205 | Tyr δ(C–H) |
1140 | 1123 | 1123 | δ(αC–N) |
1052 | 1032 | 1032 | Phe def. (E1ub) |
1012 | 1002 | 1002 | Phe def. (B2ub) |
995 | 993 | — | ν(C–C) |
866 | 852 | 855 | Tyr breathing def. |
846 | 830 | 828 | Tyr, Phe, oop(C–H) |
757 | 741 | 741 | oop(C![]() |
670 | 640 | 640 | oop(C![]() |
503 | 508 | 508 | ν(S–S) |
263 | 282 | — | Backbone |
The amide II band (C–N stretching and NH bending, around 1540 cm−1) is weak, which is usual in non-resonance Raman peptide spectroscopy. Only in resonance does its Raman intensity become comparable to that of amide I.54 In native insulin, however, there is a weak negative ROA signal (1537 cm−1, computed at 1557 cm−1), disappearing for the amyloid, both in theory and experiment.
The histidine C–H bending vibration (calculated 1522 cm−1, experimentally 1446 cm−1) gives strong Raman bands but generates a very weak ROA signal, at spectrometer detection limits. In the extended “amide III” region (∼1200–1400 cm−1) the simulation confirms that ROA is relatively strong and changes with the conformation. The 1313/1339 cm−1 (exp./calc.) ROA positive band of native insulin loses intensity and a new 1271/1306 cm−1 positive band appears in the fibril. The simulated spectrum changes even more, and predicts a 1343 cm−1 negative ROA band for the amyloid, undetected experimentally. Previously, vibrational modes in this region were identified as vibration of the main peptide chain coupled with αC–H bending and side chain vibrations,9,14,48–50 lending them exceptional sensitivity to fine geometry changes.
The negative 1240 cm−1 (exp.) ROA band of native insulin slightly shifts to 1233 cm−1 for the fibril, in accord with the simulations (1239 → 1213 cm−1), although here the observed intensity also changes less than that in theory. Partially, this can be explained by the incomplete conversion of the native state into the fibrillar one in the experiment.10 The negative 993 cm−1 experimental ROA band of native insulin is not well predicted theoretically either.
What is truly remarkable is the intense ROA signal in the lowest-wavenumber region (200–300 cm−1), comparable with the strongest bands of the extended amide III region. This region was largely ignored in previous protein studies, mostly for experimental reasons (unavailability of narrow filters) and due to difficult interpretation. The +/− native insulin bands at 224/282 cm−1 transform into a positive signal for the fibril, which seems qualitatively reproduced by the theory, although the simulated intensity is lower. Visualization of the dynamic displacement of vibrational normal modes revealed that the negative (282 cm−1) ROA bands largely arise from α-helical like segments of insulin. Previously, a similar +/− pattern was observed at 229/302 cm−1 for highly-helical human serum albumin,14 thus confirming the potential of the low-energy vibrations for probing the protein structure.
To understand how the Phe and Tyr residues may contribute to insulin ROA spectra, we simulated Raman and ROA spectra of NH2-Phe-COH and NH2-Tyr-COH model aldehydes (Fig. S5, ESI†). The simulations revealed significant changes in spectral shapes of aromatic bands at 1005, 1330 and 1650 cm−1 due to the conformation of the aromatic side chain. For NH2-Phe-COH, the conformers were averaged using MD populations; Raman and ROA spectra were generated for the native and fibrillar insulin and are plotted in Fig. 3 (spectra of individual conformers are shown in Fig. S6, ESI†). For the fibril, the CC stretching band (number III in Fig. 3) generates a relatively strong negative ROA signal, corresponding to the observed band at 1620 cm−1 (Fig. 2). A similar intensity change occurs for the aromatic hydrogen bending (∼1350 cm−1, number II) and out of plane motion (∼1000 cm−1, number I), where, however, correspondence to the experiment is not so obvious due to overlap with other vibrations.
![]() | ||
Fig. 4 ROA and Raman spectra of an insulin fibril simulated for three values of the twist (τ) between neighboring insulin units (average from 100 MD snapshots). |
While the Raman spectra (bottom of Fig. 4) are more or less insensitive to the twist, the opposite is the case for ROA shapes (top). The ROA intensity variations are less pronounced above 1500 cm−1, e.g., the amide “W” shape is conserved for the three twist values. On the other hand, the extended amide III region 1200–1400 cm−1 undergoes more profound changes. All intensity variations are not monotonic, e.g., for the −3° twist a positive ROA signal appears around 1275 cm−1, with the intensity being smaller both for 0° and −6°. In the lower-wavenumber region (below ∼1000 cm−1) the spectra are less sensitive to the twist and depend on it in a more monotonic way. At the current level of experimental noise and computational precision it would be too speculative to deduce the twist of insulin fibrils merely by comparing the theoretical and experimental spectra; this “computational experiment” nevertheless documents the potential of ROA spectroscopy for future fibril structural studies.
Because the spectra for different fibril twist were generated from a limited number of MD snapshots, we had to ensure that the error of averaging is smaller than the effect of the twist itself. This is documented in Fig. S7 (ESI†): the small differences in Raman and ROA intensities obtained for the 50 and 100 snapshot simulations indicate that the twist-induced changes in the spectra are realistic indeed, i.e. they are mostly caused by structural changes, such as variation of the distribution of torsional angles, rather than by incomplete MD averaging. Examples of the angular distributions are given in Fig. 5, for the main chain φ and ψ torsion angles, and histidine and tyrosine χ1 side chain angle. Note, for example, that the (φ, ψ) values are close to (−145°, 150°), i.e., corresponding to the canonical β-sheet conformation,57 and that some twist-induced changes are not monotonic in the row 0 → −3° → −6°, as is the case for the ROA spectra. As may be expected, the individual side chain angle χ1 distributions react more sensitively to the twist than the averaged φ and ψ ones.
![]() | ||
Fig. 5 Distribution of selected torsion angles for the three twist values (cf. τFig. 4): average φ and ψ main chain angles, and the χ1 angle of the histidine and tyrosine side chains. |
Based on the results shown above, we can also conclude that the formation of the fibrils causes specific changes in ROA spectra. They are less spectacular than those for VCD; in particular insulin fibrils provided 10–100 fold VCD enhancement if compared to the native form.7,60 The enhancement seems to stem from a long-range synchronization of amide I (CO stretching) vibrations.61 Such an enhancement is to some extent possible also for ROA,62 but has not been observed for the insulin fibrils. Here, the amide I signal “only” changes the shape (Fig. 2). As indicated by the simulations incorporating the macroscopic twist (Fig. 4), the longer-range order is reflected in the ROA spectra, too, but it is mediated by local conformational changes.
Finally, as a big advantage of ROA and Raman techniques in general, a wider range of wavenumbers is accessible than for VCD. For the insulin fibrils, secondary structure change visible in the amide I region could be confirmed by the αC–H bending (around 1310 cm−1) and low-frequency modes (<300 cm−1). Obviously, a combination of multiple techniques, such as ROA and VCD, is desirable for obtaining complete information on protein conformational behavior.
The aromatic residues not only had a strong Raman signal, but also an ROA signal that may locally probe the structure. One has also to admit that many spectral features potentially useful for structure determination are obscured due to limited accuracy of the simulations, band overlaps and experimental noise.
Rather surprising and potentially useful for future fibril studies was the sensitively responding amide III ROA signal to the fine changes of the conformation simulated for different macroscopic twists. The modeling showed that this does not reflect any direct non-covalent interaction of peptide chains, but is mediated by subtle changes in the local protein conformation. All the results confirm that the ROA technique is indeed very sensitive to protein conformational changes including the fibrillation, and that the complex simulations help extract additional details about the molecular structure and dynamics.
Footnote |
† Electronic supplementary information (ESI) available: Computational tests and details of spectra simulation. See DOI: 10.1039/c7cp01556a |
This journal is © the Owner Societies 2017 |