Theoretical Sum Frequency Generation Spectroscopy of Peptides.

Vibrational sum frequency generation (SFG) has become a very promising technique for the study of proteins at interfaces, and it has been applied to important systems such as anti-microbial peptides, ion channel proteins, and human islet amyloid polypeptide. Moreover, so-called "chiral" SFG techniques, which rely on polarization combinations that generate strong signals primarily for chiral molecules, have proven to be particularly discriminatory of protein secondary structure. In this work, we present a theoretical strategy for calculating protein amide I SFG spectra by combining line-shape theory with molecular dynamics simulations. We then apply this method to three model peptides, demonstrating the existence of a significant chiral SFG signal for peptides with chiral centers, and providing a framework for interpreting the results on the basis of the dependence of the SFG signal on the peptide orientation. We also examine the importance of dynamical and coupling effects. Finally, we suggest a simple method for determining a chromophore's orientation relative to the surface using ratios of experimental heterodyne-detected signals with different polarizations, and test this method using theoretical spectra.

Introduction

The biological activity of many proteins is intimately linked with their interactions with interfaces, most notably cellular membranes. For instance, antimicrobial peptides (AMPs), used by many organisms to defend against infection, may function by permeabilizing bacterial membranes.[1,2] A number of other peptides, such as voltage-gated ion channels, possess specialized structures that allow them to span the membrane.[3] Moreover, some peptides are thought to malfunction via membrane interactions; for example, the membrane interactions of human islet amyloid polypeptide (hIAPP) may be important to its aggregation, a process speculated to damage the membrane.[4−6]

Membrane-active proteins often function through a series of fast structural changes. To study them, therefore, it is desirable to employ an experimental technique that is selectively sensitive to the structure of interfacial proteins and that operates on the fast time scale of protein conformational change. Here, we describe just a few common techniques. One powerful technique for determining the structures of interfacial proteins in realistic environments is nuclear magnetic resonance, which has been used to probe many systems, including AMPs and proton channels in micelles and lipid membranes.[1,7,8] Electron paramagnetic resonance with site-specific spin labeling, meanwhile, has allowed researchers to examine particular elements of protein structure in detail and to characterize their dynamics, and has provided detailed information on protein conformational change in membranes.[9,10] Finally, infrared (IR) spectroscopic techniques provide a powerful means of studying protein structure and dynamics due to the sensitivity of vibrational line shapes to environmental influences and (for time-domain studies) due to the sub-picosecond periods of IR pulses, which enable these techniques to distinguish transient protein conformations. For example, linear and two-dimensional (2D) IR methods have been used to study the conformations of AMPs on a bilayer surface[11] and the aggregation of hIAPP in the presence of lipid vesicles.[12,13] Attenuated total reflection Fourier transform IR spectroscopy provides additional interface selectivity due to the small penetration depth of the probe radiation, and it has been used to determine the orientation of membrane proteins.[14−19] Researchers using these techniques frequently employ detergent micelles or vesicles (instead of the native membrane) in order to create a soluble protein–lipid system, or remove excess solvent from a bilayer system.

Vibrational sum frequency generation (SFG) spectroscopy offers an attractive alternative to these techniques because of its intrinsic surface selectivity (described below), which obviates the need for vesicles or desolvated systems, and its fast dynamical time scale. Thus, it has become much-used for the study of interfacial proteins[18−38] and has enabled researchers to study these systems in real time and in situ.[19,24,27,31−33,39−44] In particular, many recent experiments have focused on the amide I (primarily CO-stretch) mode, which has also been extensively used for linear and 2D IR studies.[11,13,45−51] This mode is particularly attractive for proteins because it exhibits distinct spectral features for different secondary structures.[17,27,52]

In vibrational SFG spectroscopy, a resonant IR pulse and a non-resonant visible pulse are overlapped spatially and temporally on a sample surface, and the signal is detected at the sum frequency of the incident beams.[20,53,54] Typically, the pulses are controlled to propagate in a single plane (perpendicular to the sample surface), with polarizations either parallel (P) or perpendicular (S) to that plane, giving eight possible polarization combinations: PPP, PPS, PSP, SPP, PSS, SPS, SSP, and SSS (letters listed in decreasing frequency order—sum frequency, then visible, then IR). Within the dipole approximation, second-order nonlinear techniques such as SFG give no signal in centrosymmetric systems and are thus sensitive to interfaces, where centrosymmetry breaks. In systems of achiral molecules with random azimuthal orientation, the non-zero signals are SSP, SPS, PSS, and PPP;[55,56] these polarization combinations have been much-used in previous work on proteins.[19,27,29,30,35−37,53,56−62] In systems that additionally lack symmetry with respect to an arbitrary mirror plane perpendicular to the surface—e.g., randomly oriented chiral molecules—the SPP, PSP, and PPS signals can be non-zero as well.[63,64] SFG with these polarization combinations has thus been termed “chiral SFG”.[32,64,65]

Chiral SFG has been used to study interfacial proteins.[18,27,32−34,65,66] In particular, Yan and co-workers have found that different secondary structures can be distinguished on the basis of the existence of detectable PSP signals for the protein amide I and NH-stretch modes.[33] For example, a parallel β-sheet exhibited a distinct signal only for the amide I mode, an α-helix exhibited a signal only for the NH-stretch mode, and an antiparallel β-sheet exhibited signals for both modes. Yan and co-workers have applied this method to monitor in real time the aggregation of hIAPP at a lipid membrane.[32] Chiral SFG thus holds great promise as a means for interfacial secondary structure determination.

Commonly, vibrational SFG spectra are interpreted in terms of a summation over independent contributions from amide vibrations in fixed orientations relative to the surface. That is, the amide I vibrations are described in terms of a limited number of local or normal modes; the local modes are assigned a fixed molecular-frame transition dipole and transition polarizability; and the protein is assumed to be fixed in a single conformation relative to the surface (i.e., the orientational distribution of each amide group is assumed to be a delta function).[18−20,27,34,36,37,39,41,42,56,60,62,65−70] Such a model may fail for three main reasons. First, vibrational coupling may cause the difference between the SFG signals of the local and normal modes to become significant. Second, peptide dynamics can alter the spectrum through effects such as motional narrowing (a decrease in line widths due to frequency self-averaging). Finally, even if coupling and dynamical effects can be ignored, the orientational distribution of the local amide modes is never truly a delta function.

In order to determine the relative importance of each of these factors, it is necessary to develop a theoretical framework within which experimental results may be interpreted. Several researchers have made strides in this direction for peptides, applying theory to static structures[67,71] or to relatively short (2 ns or less) simulations of fairly large proteins, whose SFG spectra are difficult to interpret in detail.[72−74] (It should be noted, however, that such short simulation times may actually be more justified for large proteins as opposed to smaller peptides, since large proteins tend to have stable structures that give rise to quickly converging spectra.) In this work, we take a different approach, using long (200 ns–1.25 μs) molecular dynamics (MD) simulations to examine in detail both chiral and non-chiral SFG spectra for relatively small model systems.

In previous works, we have developed a theoretical strategy for the calculation of peptide amide I spectra from MD simulations and have applied this strategy to study, e.g., protein thermal unfolding and amyloid aggregation.[46−49,51,75] Here, we extend this method to the calculation of SFG spectra, providing a means to bridge SFG experiments and MD simulation, and to interpret experimental spectra in a detailed manner. We then apply this method to study the SSP and PSP spectra of three model systems, which are displayed pictorially in Figure 1. First, we examine a single-chromophore achiral model amide, N-methylacetamide (NMA), in order to validate our spectroscopic model and to demonstrate a zero PSP signal for an achiral system. Next, we study a single-chromophore chiral molecule, VG dipeptide (in its zwitterionic form), and illustrate how the presence of a single chiral center gives rise to a significant PSP signal. We then calculate spectra for gramicidin S (GS10), a cyclic decapeptide with the sequence (VKLYP)2, which possesses overall C2 symmetry and a highly stable anti-parallel β-sheet structure.[76−80] Analysis of the results provides insight into the influence of secondary structure, vibrational coupling, and dynamics on SFG spectra. We conclude by suggesting a simple method for using ratios of experimental heterodyne-detected signals for different polarizations to determine a chromophore’s orientation, and testing this method via our simulations.

Figure 1

Model systems examined in this study. Carbon atoms are shown in tan, oxygen in red, nitrogen in blue, and hydrogen in silver. Note that aliphatic −CH, −CH2, and −CH3 groups are treated as united atoms in the force field, so these hydrogens are not shown.

Methods

Polarization Effects in Vibrational SFG

In vibrational SFG, the sample interacts with a tunable IR beam with frequency ω and polarization K̂, and with a non-resonant visible beam with frequency ωvis and polarization Ĵ. The signal is emitted with frequency ωS = ω + ωvis and polarization Î, and is enhanced when the IR beam is resonant with a system vibration.[53,54,81] In Figure 2, we diagram the SFG setup. Typically, the scattering plane is perpendicular to the surface; we therefore define the Z-axis as the surface normal, and the X–Z plane as the scattering plane. The beams are polarized either in the X–Z plane (P polarization) or in the Y-direction (S polarization). The angles of the input and sum-frequency wavevectors with respect to the surface normal are β, βvis, and βS; by convention, we always consider the angle with respect to vectors pointing away from the surface, such that β is between 0 and 90° for all three beams. All these definitions apply to the laboratory frame, symbolized here using capital letters (X/Y/Z and I/J/K).

Figure 2

Diagram of the vibrational SFG setup. Î, Ĵ, and K̂ indicate the polarization directions for the sum frequency, visible, and IR beams, respectively (here drawn for PPP polarization). βS, βvis, and β indicate angles relative to the surface normal. Subscript “S” indicates sum frequency and “vis” indicates visible; no subscript indicates IR.

We now focus our derivation on the lab-frame SSP and PSP signals, which consist of three types of factors: elements of the (second-rank) Fresnel tensor L, which describe effects caused by the different refractive indices of the media under study; factors arising from the angles of the input and sum-frequency beams; and elements of the complex third-rank tensor χ (individual elements are symbolized as χ), which describe the intrinsic response of the sample.[53,81−83] (χ is known as the second-order nonlinear susceptibility.) We symbolize the full lab-frame response (including all three effects) as χeff, the “effective” response. χeff is related to χ by the equation[56]To write the final expressions for the effective SSP and PSP signals, χSSPeff and χPSPeff, note that S-polarization corresponds to Ŷ-polarization in the lab frame, whereas P-polarization corresponds to polarization in the direction cos βX̂ + sin βẐ for the IR beam, cos βvisX̂ + sin βvisẐ for the visible beam, or −cos βSX̂ + sin βSẐ for the sum-frequency beam. Substituting these polarization vectors into eq 1 gives the results:In this work, we will assume β = βvis = 45°; because ωvis ≫ ω and ωS sin βS = ωvis sin βvis + ω sin β,[59] we can also set βS = 45°. Within this limit, the angular terms merely contribute constant factors to both χSSPeff and χPSPeff, and we will ignore these factors.

Equations 2 and 3 can be simplified for systems with azimuthal symmetry (with respect to an arbitrary rotation around the Z-axis), such as those considered here. We therefore consider a “surface-fixed” frame, symbolized using lower-case letters (x/y/z and i/j/k), for which Ẑ = ẑ, while the x- and y-axes have been rotated by an angle τ such that X̂ = cos τx̂ + sin τŷ and Ŷ = −sin τx̂ + cos τŷ. Azimuthal symmetry is imposed by averaging over all angles τ. Note that this causes all terms with an even number of Z-components to vanish, since these terms have an odd number of sin τ or cos τ factors when converted into the surface-fixed frame.[55] Thus, azimuthal averaging gives the following results:where the equality χ = χ, though not true in general, is proven for our model in the next section. Because the Fresnel factors only contribute constant factors to both χSSPeff and χPSPeff, we now drop them as well. Defining the resulting quantities as simply χSSP and χPSP, we have

It is worth considering the effect of replacing Ŷ with −Ŷ = sin τx̂ – cos τŷ in the above analysis. This amounts to performing a reflection across the X–Z plane. Because χSSPeff has two Y-polarized components, this reflection does not affect the SSP signal; χPSPeff, however, has only one such component, so this reflection will reverse the sign of the PSP signal. (The Fresnel and angular factors are not affected by this procedure.) Thus, for systems with reflection symmetry (i.e., achiral systems), χPSPeff = −χPSPeff = 0.

Note also that, due to the small time and length scales employed in simulated systems, azimuthal symmetry is not fully realized in our simulations. The experimental systems that we are attempting to model are azimuthally symmetric, however, and this ought to be reflected in the spectra. Accordingly, we identify the coordinate frame of the simulation not with the lab frame (X/Y/Z) but with the surface-fixed frame (x/y/z), for which we impose azimuthal symmetry. (Hereafter, we will refer to this frame as the simulation frame.) Practically, this means that we calculate spectra using eqs 6 and 7, rather than eqs 2 and 3.

For GS10, we also calculate PPP spectra. The derivation for χPPP is analogous to those for χSSP and χPSP and is presented in Appendix 1. Again ignoring the Fresnel and angular factors, the result isIt should be noted that the justification for ignoring the Fresnel coefficients is not as strong here as for χSSP and χPPP; see Appendix 1.

Mixed Quantum/Classical Approach for Vibrational SFG

In order to calculate SFG signals using eqs 6–8, we require a model for the second-order nonlinear susceptibility elements χ in the simulation frame. χ consists of both “resonant” (R) terms, which depend mostly on ω, and “non-resonant” (NR) terms, which depend mostly on ωvis.[53,81,83] These are represented, respectively, as χR(ω) and χNR(ωvis). In this work, we focus exclusively on the heterodyne-detected signal, which is given by the imaginary part of χR(ω) and is the most physically meaningful part of the response. χR(ω) can be expressed as the following quantum time-correlation function (TCF):[81]Here, the trace is over all nuclear quantum states, ρ is the equilibrium density operator for the nuclear Hamiltonian, α(t) is an element of the ground electronic state 1–0 transition polarizability tensor, and μ(0) is an element of the 1–0 transition dipole. This equation represents χR(ω) as a quantum equilibrium statistical mechanical average, which is extremely difficult to evaluate for proteins in the condensed phase.

To simplify the analysis, we divide the system into a quantum subspace consisting of the amide I vibrations and a classical bath consisting of low-frequency modes (translations, rotations, and torsions).[84−87] Other high-frequency modes are ignored. We have previously applied such a mixed quantum/classical approach successfully to linear and 2D IR spectroscopy of liquid water[88−93] and peptides in aqueous solution and at lipid membranes,[11,45−49,51,75,94] as well as to SFG spectra of water at the liquid/air interface.[81−83,95−97] Within this approach, the susceptibility for a multi-chromophore system is given by[83]Here, p and q index the amide I chromophores, and a(t) and m(t) are the time-dependent 1–0 transition polarizability and transition dipole elements for chromophores p and q, respectively. T1 is the amide I first excited-state lifetime and is chosen to be 600 fs.[98] Importantly, the brackets now indicate an average over configurations from a classical MD simulation, making the calculation feasible. F(t) describes the time propagation of the Hamiltonian:where κ(t) is the amide I Hamiltonian (divided by ℏ):Here, ω(t) is the time-dependent local mode frequency for chromophore p, and ω(t) is the time-dependent coupling between chromophores p and q. δ is the Kronecker delta.

To apply these equations, it is necessary to calculate ω(t), ω(t), m(t), and a(t) from an MD simulation. The calculation of the first three quantities has been described in previous works.[48,99,100] For proline frequencies and for modeling nearest-neighbor frequency shifts and couplings including proline, however, methods by Roy et al. are used.[101] To calculate a(t), we follow others[60,72−74] in employing the tensor determined by Tsuboi et al. for aspartame, a model amide.[102] In previous works, this tensor has been presented in its diagonal form; here, however, we present it in a molecular frame (symbolized using x′/y′/z′) in which one of the axes is aligned with the transition dipole. (We do this so that the geometry of the chromophore in the simulation frame can be easily described using Euler angles of the transition dipole in the next section.) The transition dipole—which is modeled as a point dipole—has a location given by r⃗C + (0.665 Å)n̂CO + (0.258 Å)n̂CN, where r⃗C is the position of the amide C, and n̂CO and n̂CN are unit vectors pointing from the amide C to the amide O and N, respectively. The dipole lies in the CON plane and forms an angle of 10° with the OC vector.[99] The molecular frame, then, is defined as follows: ẑ′ is aligned with the transition dipole; x̂′ lies perpendicular to ẑ′ and in the CON plane such that the amide N has a positive x̂′-coordinate; and ŷ′ lies perpendicular to the CON plane so as to form a right-handed coordinate system. In this reference frame, depicted in Figure 3, the transition polarizability tensor is given byThe equality χ = χ, stated in the previous section, is a consequence of the symmetry of the transition polarizability tensor, specifically the fact that α = α. (Note that any rotation of this tensor, which can be represented as a unitary transformation, will preserve the symmetry of the tensor. Thus, the tensor is symmetric not only in the molecule-fixed frame, but in the simulation frame as well.)

Figure 3

Diagram of the molecular frame Cartesian axes. The red dot signifies the location of the transition dipole. The z′-axis forms an angle of 10° with the OC vector, and the x′ and z′ axes are defined to lie in the CON plane. The y′-axis (not pictured) lies orthogonal to the CON plane so as to form a right-handed coordinate system.

Effect of Orientation on SFG Amplitudes

For later analysis, it is useful to consider the SFG spectrum of an isolated chromophore in the inhomogeneous limit (i.e., ignoring both coupling and dynamical effects). The isolated-chromophore version of eq 10 is[81]In the inhomogeneous limit, the time dependence of the variables in the above expression is negligible, and we can replace a(t) and ω(τ) with a(0) and ω(0), respectively. If we ignore the lifetime decay (effectively letting T1 → ∞), we can carry out the integration over t. We focus on the imaginary part of the result because we have chosen to examine the heterodyne-detected signal. This gives us the SFG spectral density:

For our model, the quantity a(0)m(0) has no explicit frequency dependence, and it can be determined entirely from the geometry. Following an approach similar to that of Laaser and Zanni (among others),[103] we show in Appendix 2 that the value of a(0)m(0) can be expressed in terms of two Euler angles: a tilt angle θ that describes the angle between the transition dipole vector and the surface normal, and a twist angle ψ that describes a subsequent rotation around the transition dipole axis. (That a third angle is not required is a consequence of the azimuthal symmetry of the system.) We may therefore writewhere A(θ,ψ) = a(θ,ψ)m(θ,ψ).

A(θ,ψ) is independent of the configurational average denoted by the angled brackets and can therefore be moved outside the average. We may also take the integrals outside the average, with the result:where P(ω,θ,ψ) = ⟨δ(ω – ω(0))δ(θ – θ(0))δ(ψ – ψ(0))⟩ is the joint probability distribution describing the chromophore’s frequency and orientation. Note that in the bulk, the orientational part of P(ω,θ,ψ) is isotropic and independent from the frequency distribution, such that P(ω,θ,ψ) = P(ω)P(θ,ψ) = P(ω) sin θ/4π.

Equation 17 provides a means of determining the effect of local-mode orientation on the SFG amplitude, independent of dynamic or coupling effects. To do this, we determine P(ω,θ,ψ) from simulation and use the following theoretical results (derived in Appendix 2) for A(θ,ψ):Note that the numerical prefactors in the last two equations result from the substitution of values specific to our models of the amide I local mode transition dipole and polarizability. Thus, these equations apply only to amide I local modes treated using these models.

In Figure 4, we depict these results graphically. Note that whereas the SSP and PPP amplitudes depend mostly on the tilt angle θ, the PSP amplitude depends more strongly on the twist angle ψ. Also, note that all three amplitudes are anti-symmetric on reflection over the line θ = 90°, followed by a 180° shift in ψ. This operation transforms x̂′ to −x̂′, ŷ′ to −ŷ′, and ẑ′ to −ẑ′; i.e., it inverts the molecular orientation. Equivalently, we could replace A with A–, which always reverses the sign because there are three Cartesian components.

Figure 4

Theoretical SSP (top), PSP (middle), and PPP (bottom) amplitudes as a function of transition dipole orientation for the amide I model used in this study. Blue areas are positive, and red areas are negative. SSP contours run from a relative amplitude of −0.8 to 0.8 in increments of 0.16, PSP contours run from −0.25 to 0.25 in increments of 0.05, and PPP contours run from −1.64 to 1.64 in increments of 0.328.

A final important point concerns the fact that the SSP and PPP amplitudes are symmetric, and the PSP amplitude anti-symmetric, on reflection over the line ψ = 90° or ψ = −90°. The location of these lines of symmetry is a consequence of the fact that we chose the y′-axis to point out of the amide plane, such that for our model, α = α = 0, while α is non-zero. Because of this, a 90° rotation in ψ is required to align the amide plane (which for achiral molecules is the symmetry plane of the molecule) with the surface normal for arbitrary θ. If we had instead chosen the x′-axis to point out of the amide plane, then α would have been the non-zero element, and the lines of symmetry would have been located at ψ = 0° and ψ = 180°.

Calculation of Raman Spectra

In this work, we calculate the Raman spectrum of dilute NMA in bulk water in order to verify the transferability of the transition polarizability tensor. Within the mixed quantum/classical approximation, the polarization-dependent Raman line shape for a single chromophore is given by[92,104,105]Typically, Raman spectra are recorded either in VV mode (input and signal polarizations parallel) or in VH mode (input and signal polarizations perpendicular). Due to the isotropy of the bulk system, the VV and VH spectra can be represented as spherical averages over the constituent I’s:[104]The depolarization ratio is defined as[106]where

Simulation Details

MD simulations are performed according to the following procedure. First, the initial (cubic) box is constructed using the genbox utility of GROMACS4.5.3.[107] The box contains a single (N-deuterated) peptide of interest (NMA, VG, or GS10), sufficient water (D2O) molecules to fill a box with edge lengths of at least 4.05 nm (3004 for NMA, 2217 for VG, and 2354 for GS10), and, for GS10, two Cl– counterions. This box is equilibrated in the NPT ensemble for at least 4 ns. Next, a continuation run is performed in the NPT ensemble, and configurations are output every 100 ps so as to create an ensemble of 20 initial configurations for NMA, and 50 initial configurations for VG and GS10. For each of these configurations, a pair of surfaces is created by extending the z-dimension of the box to at least three times its initial size. Equilibration runs are then performed in the NVT ensemble for 5 ns for NMA and VG, and for 15 ns for GS10. Finally, NVT production runs are performed. For NMA, production runs are 10 ns in length, for a total simulation time of 200 ns; for VG, production runs are 25 ns, for a total of 1.25 μs; and for GS10, production runs are 20 ns, for a total of 1 μs. Configurations for production runs are output every 10 fs.

For all runs, a 2 fs time step was employed, particle-mesh Ewald was used for electrostatics, and a simple cutoff with a long-range dispersion correction was applied for Lennard-Jones forces. Force-field parameters were taken from the GROMOS96 53a6 parameter set,[108] which employs the SPC water model.[109] For equilibration runs, the Berendsen thermostat and barostat[110] were used; for production runs, the Nosé–Hoover thermostat[111,112] was used.

Although GS10 remains at a single interface in all simulations, both NMA and VG cross the box and find the other interface on occasion; to address this, we apply a switching function to the transition dipole elements, as in previous work for water.[96] (The dipole elements are scaled based on the dipole position, as parametrized by Torii and Tasumi.[99])

All reported spectra are averaged over the full length of the production runs, with starting points for the calculation of TCFs taken every 50 fs. All reported SFG spectra are normalized by the number of chromophores and the number of starting points used, such that the amplitudes of each are directly comparable.

Results and Discussion

Raman and SFG of NMA

To investigate the transferability of the amide I transition polarizability tensor obtained for aspartame,[102] in Figure 5 we show VV and VH Raman spectra for bulk NMA in water, as well as experimental spectra from Chen et al.[106] The theoretical and experimental spectra compare favorably, and the theoretical depolarization ratio of 0.21 compares reasonably to the experimental value of 0.16 (see Table 4 from the experimental work[106]), considering that the theoretical tensor was taken from a different peptide.

Figure 5

Experimental (solid) and theoretical (dashed) Raman VV (black) and VH (red) line shapes for bulk, aqueous N-deuterated NMA. Experimental curves are taken from Chen et al.[106] and are scaled to match the reported depolarization ratio of 0.16. Experimental and theoretical curves are also normalized to the maximum value for the VV line shape.

In Figure 6, we show SSP and PSP line shapes for NMA (from the surface simulations). As significant statistical error is still present in our result even after 200 ns of MD, we also provide a 95% confidence interval for the results (thin lines), calculated by comparing the values for each frequency across the 20 independent runs. (This uncertainty, also present for VG dipeptide, seems to result mainly from the fact that NMA does not sit stably at one interface but ventures into the bulk and even crosses the box to find the other interface at several points in the simulations. Thus, portions of the simulation are spent not describing the interfacial configurations, but rather averaging over bulk configurations.) Note that the PSP signal for this achiral molecule is 0 (within the error), as expected. The interpretation of these results is fairly straightforward. NMA’s frequency-dependent orientational distribution P(ω,θ,ψ) (determined from the MD simulations) is symmetrical on reflection over the line ψ = 90° at all frequencies, such that its PSP signal vanishes (see Figure 4). The distribution is peaked near θ = 90°, but angles less than 90° are weakly favored at low frequency, while angles greater than 90° are favored somewhat more strongly at high frequency. Thus, although the peak in the frequency distribution P(ω) is close to 1625 cm–1 (as in Figure 5), the SFG intensity is relatively weak at this frequency because the average θ is very close to 90°. Instead, the SFG intensity peaks at ∼1645 cm–1, where P(ω) is smaller but a larger proportion of the transition dipoles point in the same direction (down).

Figure 6

Theoretical SSP (black) and PSP (red) SFG spectra (imaginary part) for NMA. Thin lines mark the 95% confidence interval (±2 standard errors of the mean).

To understand the frequency-dependent sign of the SSP signal, recall that in our model (taken from Torii and Tasumi[99]), the transition dipole points nearly opposite the CO bond vector. At low frequency, therefore, there is a slight tendency for the CO to point down. This favors the formation of hydrogen bonds between water and the carbonyl oxygen and lowers the amide I frequency. At high frequency, meanwhile, the CO tends to point up, decreasing the number or strength of hydrogen bonds to water, and increasing the frequency.

SFG of VG Dipeptide

In Figure 7, we show SSP and PSP line shapes for VG dipeptide, again with 95% confidence intervals. The SSP signal can be explained along similar lines as for NMA. VG differs in that it has a taller positive peak and a shallower negative peak, which may indicate that configurations with upward-pointing transition dipoles are overall more favorable in this case. Unlike NMA, the chiral VG molecule generates a significant PSP signal, which is roughly as intense as the SSP signal over the entire frequency range (in the limit that the Fresnel and angular factors are ignored). The presence and sign of the PSP signal can be understood from the fact that the transition dipole orientational distribution P(θ,ψ) (again calculated from the simulation) is peaked near θ = 95°, ψ = 195°, in a region which Figure 4 shows to correspond to a positive PSP signal. Visualization of this geometry shows that it allows the carbonyl oxygen to point down, toward water, while allowing the valine side chain to point up, toward the vapor. Because of the chirality of the valine Cα, the adoption of such a geometry requires specific tilt and twist angles. The PSP signal can therefore be readily understood from the geometric constraints imposed by the chirality of the molecule and the varying hydrophilicity of its parts.

Figure 7

Theoretical SSP (black) and PSP (red) SFG spectra (imaginary part) for VG dipeptide. Thin lines mark the 95% confidence interval (±2 standard errors of the mean).

SFG of GS10

Before showing spectra for GS10, it is worth describing its important structural features, as well as its typical orientation at the interface. GS10, a cyclic peptide with the sequence (VKLYP)2, effectively possesses C2 symmetry because the two VKLYP chains are structurally equivalent. (We say “effectively” because any given simulation snapshot is unlikely to show C2 symmetry; only the time-averaged structure does.) GS10 has a primarily anti-parallel β-sheet structure, with tight turns at the Pro side chains. The eight amide groups whose N is not donated by Pro (i.e., all except the amides arising from the Tyr–Pro peptide bond) form four cross-strand hydrogen bonds. At the interface, GS10 orients such that the Val and Leu side chains point toward the vapor, while the Lys and Tyr side chains point toward water. This interfacial configuration is quite stable, and the SFG spectra of GS10 are well-converged after 1 μs of simulation. Consequently, no confidence interval is shown for the SFG spectra of GS10.

In Figure 8, the SSP and PSP spectra for GS10 are shown (black), along with two limiting cases. In the first (red), the spectrum is calculated with all couplings artificially set to zero. In the second (blue), the spectral density is calculated using eq 17 (summing over the individual results for each chromophore); this result is insensitive to both coupling and dynamical effects. Comparing the spectra with and without coupling (black and red), we see that coupling influences the spectra moderately: the peak positions shift by 10–15 cm–1, and the peak shape changes somewhat for the PSP spectrum. However, the influence of the coupling is small enough that the line shapes can be qualitatively explained without considering coupling effects. Meanwhile, comparing the line shape without coupling (red) to the spectral density (blue) reveals that dynamical effects such as motional narrowing have very little effect on the spectrum. Thus, to a very reasonable approximation, we can understand the SFG spectra simply in terms of the orientations of the individual amide I modes. For a significant signal to be observed, the amides must adopt orientations that not only give rise to large SFG amplitudes, but also do not significantly cancel each other. In GS10, for example, the SSP and PSP signals are strongly negative for only 2 of the 10 amides, causing the full-system SFG spectra to be mostly positive. The most important influence of secondary structure, then, appears not to be its influence on the couplings between amides, but rather its influence on the relative orientation of the amides.

Figure 8

Theoretical SFG spectra (imaginary part, SSP and PSP) for GS10 peptide. Black line is the full calculation; red line is the calculation without coupling effects; blue line is the spectral density calculated using eq 17. Blue line is scaled to match the peak height of the red line.

It is notable that the SFG spectra of GS10 lack the characteristic a– and a+ peaks seen in the IR spectra of many β-peptides. For instance, the IR spectrum of the 12-residue β-hairpin Trpzip2 shows a low-frequency peak near 1636 cm–1 and a high-frequency shoulder near 1673 cm–1 in both experiment[78] and theory.[48] GS10’s SFG spectra, however, lack a high-frequency shoulder. There are a few possible reasons for this. First, GS10 is even smaller than Trpzip2, and two of the four cross-strand hydogen bonds (the ones nearest the turns) are somewhat unstable. Both of these factors will cause GS10’s spectra to deviate from that of an idealized β-sheet. Second, the SFG spectrum includes contributions from both the transition dipole and the transition polarizability. Density functional theory calculations on a large β-sheet complex by Welch et al. indicate that the a+ mode of β-sheets is significantly less active in Raman spectroscopy than in IR spectroscopy,[113] which suggests that the transition polarizability associated with this mode is small.

Determination of Individual Peptide Orientations Using SFG Spectra: Application to GS10

In the previous sections, we have presented SFG spectra for model peptides and explained how their interfacial orientations—determined using simulation—give rise to the observed spectral features. It is also desirable to be able to determine peptides’ orientations based simply on experimental data. To this end, it is common to employ a simplified model for the angular distribution, such that it is proportional to a two-dimensional delta function, δ(θ–θ′)δ(ψ–ψ′). In this way, the problem of determining the angular distribution is reduced to the determination of two angles, θ′ and ψ′, a problem which can be solved using constraints from SFG or other techniques, such as linear dichroism.[18−20,27,34,36,37,39,41,42,56,60,62,65−70]

In the context of our formalism, within this approximation the triple joint distribution function isIf one is in the inhomogeneous limit, then from eq 17,The important point is that if one forms the ratio of Im χ for two different polarizations, then for a given frequency, P̃ cancels and one obtains simply the ratio of the two A factors. For a given ratio of signals from two different polarizations, one can therefore use the expressions in eqs 18–20 to determine θ′,ψ′ pairs that are consistent with this ratio. This generates iso-ratio curves in θ–ψ space. Because the SFG amplitudes depend on the frequency, data at different frequencies may produce different curves. Furthermore, various choices of the two polarizations lead to different functional forms for the ratio, generating different sets of curves. If the angular distribution is truly a delta function, then the curves for different polarization ratios will cross at certain points, which represent values of θ and ψ consistent with the signal ratios. If the angular distribution is not a delta function (it never truly is) but is nevertheless sharply peaked, or if the spectral density expression of eq 17 is not an excellent approximation (because of motional narrowing), then the curves may not cross at the same point, but they should still approach each other.

In this section, we explore this idea by using the relative amplitudes of SSP, PSP, and PPP spectra of individual chromophore pairs in GS10 to determine constraints on those peptides’ interfacial orientations consistent with a delta distribution. We examine chromophore pairs because the C2 symmetry of GS10 dictates that each peptide in GS10 has a counterpart whose environment is identical, on average. (E.g., the two chromophores derived from Val–Lys peptide bonds are structurally equivalent.) Moreover, analysis of the MD simulations reveals that the orientational distributions P(θ,ψ) of each chromophore possess a single dominant peak, indicating that for the most part, the two amides sample similar orientations at any given time. This allows us to interpret the SFG spectra of each amide pair in terms of a single orientation.

In Figure 9, we present SSP, PSP, and PPP spectra (without coupling) for the amide pairs arising from the Val–Lys and Tyr–Pro peptide bonds. To analyze these spectra, we determine the amplitude ratios for PSP/SSP and PPP/SSP over a range of frequencies including the peak: 1650–1680 cm–1 for Val–Lys (SSP peak at 1667 cm–1), and 1620–1650 cm–1 for Tyr–Pro (SSP peak at 1638 cm–1). For Val–Lys, the PSP/SSP ratio varies from −0.787 to −0.597, and the PPP/SSP ratio varies from 0.887 to 1.353. For Tyr–Pro, the PSP/SSP ratio varies from −0.196 to 0.059, and the PPP/SSP ratio varies from 3.105 to 3.399. As discussed above, each amplitude ratio is consistent with a line of constraint in θ and ψ; in Figure 10, these lines of constraint are plotted. Solid lines show orientations consistent with the PSP/SSP ratios, while dashed lines show orientations consistent with the PPP/SSP ratios. Differently colored lines correspond to ratios determined at different frequencies.

Figure 9

Theoretical SSP (black), PSP (red), and PPP (blue) spectra (imaginary part) including only chromophores from Val–Lys (top) or Tyr–Pro (bottom) peptide bonds. Coupling between the amide groups is not included.

Figure 10

Transition dipole orientational distribution P(θ,ψ) for Val–Lys (top) and Tyr–Pro (bottom) amides of GS10 at an air/water interface, along with lines of constraint derived by assuming a delta-function distribution consistent with the PSP/SSP (solid) and PPP/SSP (dashed) ratios at several frequencies. For orientational distributions, blue indicates low probability, and red indicates high probability. Val–Lys lines of constraint are determined at 1650 (magenta), 1660 (yellow), 1670 (cyan), and 1680 (gray); Tyr–Pro lines of constraint are determined at 1620 (magenta), 1630 (yellow), 1640 (cyan), and 1650 (gray).

For Val–Lys (top panel) one sees that there two regions in the θ–ψ plane where all the curves approach each other. Thus, this analysis yields two possible orientations consistent with the calculated spectra. However, we can rule out the solution near θ = 65°,ψ = 125° because it corresponds to incorrect signs for the different signals. This leaves a single possible solution near θ = 115°, ψ = 235°. We can now compare this predicted orientation with the actual orientational distribution, as determined by the simulation. This distribution is also shown in the top panel, and its dominant peak very nicely overlaps the region of the solution determined using SFG. A similar situation obtains for the Tyr–Pro chromophore pair (bottom panel). Again, although multiple solutions for the peptide orientation are possible, only one (the correct one, near θ = 42°, ψ = 98°) is consistent with the correct sign for each signal and with the data at each frequency.

Therefore, we conclude that for a single chromophore with a well-defined orientation, this method could be used to determine one or a few possible orientations from experimental heterodyne-detected signals using at least three different polarizations (giving at least two amplitude ratios). In certain pathological situations, the method does not work well. For instance, the Leu–Tyr chromophore has an average tilt angle θ very close to 90°, such that its SSP and PPP signals are small, and the SSP/PPP ratio does not offer a reliable constraint. Alternatively, if motional narrowing is important, or if the chromophore is not isolated (i.e., it has significant couplings to other chromophores), the results will worsen. Finally, there is the possibility that the angular distribution is not sharply peaked, in which case the method also will not work. Nonetheless, there would seem to be many instances in which the method will work well, especially if isotope-labeling is used to isolate single chromophores.

These results also emphasize the utility of examining the imaginary part of χ (using heterodyne detection) rather than the magnitude of χ (using homodyne detection). Because homodyned results lack sign information, they are not as useful for discriminating between multiple solutions that are consistent with the PSP/SSP and PPP/SSP ratios, making it necessary to use physical intuition or theoretical modeling to decide which result is correct.[34,36,37,66,70] Although such results are useful in many instances, it is of course desirable to use the most detailed information possible when interpreting experiments.

Conclusions

In this study, we have presented a theoretical protocol for the calculation of peptide amide I SFG spectra (including “chiral” spectra) from MD simulations and applied it to study both achiral and chiral single-chromophore molecules, as well as a 10-amino-acid chiral peptide with an anti-parallel β-sheet secondary structure. We have shown that the SSP and PSP spectra of these model systems can be largely understood based simply on the orientational distributions of the amide I modes involved, with only minimal effects due to coupling or dynamics; indeed, such an approach has already been applied to static structures in previous studies (e.g., for the second harmonic generation spectra of bacteriorhodopsin[114]). This implies that the main roles of secondary structure in determining SFG amplitudes are simply to lock amide chromophores into tight orientational distributions and to determine whether the spectral contributions of different amide groups tend to reinforce or cancel each other.

For GS10, we made use of the PSP/SSP and PPP/SSP amplitude ratios to attempt to predict the amide orientations from the spectra by assuming a delta-function distribution for the orientation of the amides at the interface. This method typically yields tilt and twist angles consistent with the actual peak in the distribution. This suggests that as long as the orientational distribution is dominated by a single, relatively uniform peak, SFG amplitude ratios offer a reasonable means of determining the interfacial orientations of amide chromophores.

Despite some challenges, the development of experimental techniques for heterodyne detection and isotope labeling holds great promise for future studies in protein SFG spectroscopy. At present, most experiments employ homodyne detection, which, while simple and reasonably robust, does not permit the reliable extraction of the imaginary part of the resonant SFG signal; these experiments are thus difficult to compare to theoretical results, and do not provide a particularly strong basis for judging the accuracy of MD simulations. Heterodyne detection, meanwhile, directly probes the imaginary part of the signal and thus allows for the extraction of detailed information related to the orientation of interfacial chromophores.[115−119] Meanwhile, isotope labeling provides a means of simplifying the SFG spectra of complex systems by shifting the frequencies of labeled chromophores away from the main amide I band.[11,94,98,120−131]13C18O labeling, for instance, typically lowers the amide I frequency by ∼70 cm–1. By isolating the spectral contributions of individual chromophores, their interfacial orientations can be probed even more directly by SFG.[70] Through the combination of these techniques with computational methods such as those outlined here, it becomes possible to obtain a detailed understanding of protein interactions with interfaces and to shed light on important processes such as the function of AMPs and ion channels or the aggregation of hIAPP.