Ligand-induced conformational changes in protein molecules detected by sum-frequency generation

Abstract

We present the first demonstration of ligand-induced conformational changes in a biological molecule, a protein, by sum-frequency generation (SFG). Constructs of KRas^G12D protein were prepared by selectively deuterating residues of a single amino acid type using isotope-labeled amino acids and cell-free protein synthesis. By attaching labeled protein to a supported bilayer membrane via a His-tag to Ni-NTA-bearing lipids, we ensured that single layers of ordered molecules were formed while preserving the protein’s native structure. Exceptionally large SFG amide I signals were produced in both labeled and unlabeled proteins, demonstrating a high degree of orientational order upon attachment to the bilayer. Deuterated protein also produced SFG signals in the CD_x spectral region, which were not present in the unlabeled protein. The CD_x signals were measured before and after binding a peptide inhibitor, KRpep-2d, revealing shifts in SFG intensity due to conformational changes at the labeled sites. In particular, peaks associated with CD_x stretching vibrations for alanine, valine, and glycine changed substantially in amplitude upon inhibitor binding. By inspection of the crystal structure, these three residues are uniquely colocated on the protein surface in and near the nucleotide binding site, which is in allosteric communication with the site of peptide inhibitor binding, suggesting an approach to identify a ligand’s binding site. The technique offers a highly sensitive, nonperturbative method of mapping ligand-induced conformational changes and allosteric networks in biological molecules for studies of the relationship between structure and function and mechanisms of action in drug discovery.

Significance

Protein conformational changes underlie all biological functions, but existing methods to study them often require static samples. We present a structural technique based on vibrational sum-frequency generation (SFG) to detect ligand-induced conformational changes in selectively deuterated protein molecules with only limited alignment imposed by attachment to a lipid bilayer surface. SFG signal strength depends quantitatively on the net, average orientation of resonant bonds with high sensitivity. Conformational changes were observed as amplitude shifts of the carbon-deuterium signals upon ligand binding. The technique requires nanograms per measurement and can be scaled to high throughput. Multiple spectra of a protein labeled with SFG-active moieties in different ways could yield protein-ligand structures in solution and enhance screening and mechanistic studies for drug discovery.

Introduction

Vibrational sum-frequency generation (SFG), introduced by Shen and coworkers in 1987, possesses unique properties as a nonlinear optical probe of molecular orientation and structure at interfaces (1,2,3). SFG requires a net, average orientation of molecules at an interface, which exhibit vibrational modes that are both infrared (IR) and Raman active. Since its introduction, SFG has been applied to study molecular orientation on surfaces in many systems and, in particular, the effects of interfacial adsorption on a molecule’s structure (4,5,6). These studies demonstrate the technique’s versatility in studying any ordered system with SFG-active vibrational modes. In recent years, SFG has been extended to study peptides and proteins at a variety of surfaces and interfaces (6,7,8,9,10,11,12). Some of these studies have used isotopic labels (13), notably in peptides and small proteins physisorbed to polystyrene surfaces (14,15,16,17). However, for studies of native proteins and their conformational changes upon ligand binding, it is critical that surface attachment does not denature the protein. To the best of our knowledge, ligand-induced conformational changes of biological molecules have not yet been investigated with SFG, and we demonstrate SFG’s potential for this application here for the first time.

We attached isotopically labeled protein molecules in a controlled way to a supported lipid bilayer membrane on SiO₂-coated CaF₂ hemispheres (Fig. 1), which produced a well-ordered monolayer of protein and exceedingly high signals in SFG. Supported lipid bilayers have previously been validated extensively as biomimetic membranes for preserving the native structure and function of biomolecules by a related nonlinear optical technique, second-harmonic generation (SHG). Ligand-induced conformational changes of proteins and other biomolecules representing a wide variety of classes including protein-protein complexes, membrane-associated proteins, disordered proteins, and oligonucleotides were studied by SHG (18,19,20,21,22).

Figure 1.

Schematic of the experimental setup. (A) A custom Teflon microfluidic cell is constructed to hold a hemispherical SiO₂-coated CaF₂ prism. The hemisphere is held in contact with a flow chamber of about 200 μL volume in the microfluidic cell with inlet and outlet ports at either side of the chamber and valves in the inlet and outlet lines for liquid injections, including a three-way valve for purging air bubbles. The vibrational SFG signal is generated at the solid/liquid interface at the flat side of the prism by total internal reflection at the overlap where the femtosecond infrared and visible (near IR) beams are focused. (B) A Jablonski diagram of the vibrational SFG process. (C) The cell is filled with buffer and sealed, enabling the preparation of supported lipid bilayer membranes on the SiO₂ surface by fusion of small unilamellar vesicles (SUVs) comprising 90% DOPC and 10% DOGS-NTA(Ni²⁺), to which the amino acid-selective isotopically labeled KRas^G12D protein constructs are attached via a His-tag. Specific attachment of the protein to the surface creates an ordered monolayer of functional protein and exceptionally high SFG signals.

In these studies, molecules were labeled with an SHG-active dye probe and attached to a supported bilayer. An extensive body of other work also indicates that biological molecules are functional when attached to supported bilayers (23,24,25,26).

In this work, we sought to detect ligand-induced conformational changes by SFG with an approach inspired by previous studies using supported lipid bilayers and SHG. Isotopic labels offer advantages compared with using SHG-active probes, as they are nonperturbative and can be placed at both surface and internal sites. SHG and SFG both share an extremely high sensitivity to orientational change of the dye label or the SFG-active bond, respectively, relative to the surface plane to which the protein is attached, and SHG has demonstrated this sensitivity to ligand-conformational changes in previous studies. Moreover, as with SHG, less than a monolayer of molecules is required for detection by SFG, far less than most other biophysical techniques, which typically require milligrams of sample per measurement. The supported bilayer provides an ideal architecture for tethering and ordering the molecules in a way that also preserves their native structure and function for conformational measurements. The bilayer’s composition can be tailored to include different lipid components with varying mole fractions. It serves an additional purpose, too, in passivating the underlying SiO₂ surface, minimizing nonspecific protein adsorption and the potential for artifacts. There is no limit on the protein size that can be studied—peptides, small proteins such as KRas, and antibodies and protein-protein complexes have been studied by SHG. Because less than a monolayer of protein is required per measurement, it is feasible to make a number of protein constructs labeled in different ways at very small scale by cell culture or cell-free protein synthesis (CFPS) for mapping conformational changes or allosteric networks. CFPS can be performed in volumes as small as 200 μL. One additional advantage of attaching the protein to supported bilayers is that they are formed at a solid-liquid interface so ligands, buffer, and other reagents can be flowed in without disturbing the sample under study.

Although previous studies had shown that the density of proteins attached to supported bilayers can be as high as 5 × 10¹² molecules/cm² (21), it was unclear a priori whether the nonlinear susceptibility of a protein monolayer attached to a bilayer via His-tags would be sufficient to detect CD_x vibrations from the labeled proteins by SFG, as the orientational distributions of the labeled residues in the protein were unknown. As demonstrated in this work, amide I SFG signals of exceptional magnitude were observed with striking signal/noise ratios (SNRs) of >100:1, indicating that the specific attachment of proteins to a lipid bilayer membrane results in a high degree of orientational order, which increased our confidence that CD_x signals could be observed.

We detected both the amide I band, due to C=O stretching of the peptide bond, and the CD_x vibrational modes in protein constructs selectively labeled with deuterated amino acids comprising both side-chain and backbone moieties. Each protein sample contained a single deuterated residue type (e.g., valine, alanine, etc.). Upon addition of a peptide inhibitor to the protein attached to the supported bilayer, conformational changes were observed as shifts in SFG peak amplitudes for some of the labeled protein constructs, whereas samples with other CD_x-labeled residues exhibited no signal change.

Materials and methods

Formation of SUVs and deposition of supported lipid bilayers

DOPC and DOGS-Ni-NTA (90 and 10% mole fraction, Avanti, Alabaster, AL, USA) in chloroform were dried down under N₂ and resuspended in HEPES buffer to 2 mg/mL, sonicated for 1 min in a bath sonicator, and then sonicated in a horn sonicator for 1 min at full power (Sonics, Newtown, CT, USA, VCX-500) to form small unilamellar vesicles (SUVs). The SUVs were then passed multiple times through an extruder with 100 nm pore membranes (Avanti) and injected into the sample cell that had been preassembled and charged with NiCl₂ (1 mM) before rinsing with HEPES buffer. Before assembly of the cell with HEPES buffer, the CaF₂ hemisphere with SiO₂ coating was cleaned in a PDC-002 plasma cleaner (Harrick, Ithaca, NY, USA) at medium power (11 W) under vacuum for 5 min. All water used was Milli-Q water. All fluidic components of the Teflon cell were sonicated for 10 min in a 5% aqueous Deconex solution followed by sonication for 10 min in water before copious rinsing with water. Syringes were sonicated for 10 min in neat ethanol followed by 10 min in water before copious rinsing with water. After cleaning, the components and syringes were kept immersed in water in beakers that were also cleaned following the above protocol. Assembly of the sample cell and injections of buffer and protein were carried out in a laminar flow hood (Thermo Fisher Scientific, Waltham, MA, USA, HERAsafe KS12).

Protein production and initial confirmation of binding to the supported lipid bilayer

KRas^G12D constructs were produced by CFPS as described previously (27) in a 2 mL reaction volume. Isotope-labeled amino acids at 5 mM and GTP at 0.8 mM were included in the translation mixture. A His₈-tag was introduced at the C-terminus of the KRas^G12D constructs for binding to the DOGS-NTA(Ni²⁺) lipids in the supported bilayers that were charged with Ni²⁺ in a HEPES buffer. A protein concentration of 1–3 μM and >30 min of incubation time were used in the broadband SFG spectroscopy experiments and excess protein was flushed out with buffer (40 mM HEPES in 150 mM NaCl [pH 7.4]) before each data acquisition.

Broadband SFG spectrometer and vibrational measurements

SFG spectra were measured in the ssp polarization with a femtosecond broadband system described previously (28). In the first experiments we performed, the laser source comprised a Ti-Light 200 oscillator and an Integra-HE-7-FS amplifier (Quantronix [acquired], formerly of Smithtown, NY, USA). This system was upgraded in later experiments with a one-box Astrella integrated amplifier (Coherent, Saxonburg PA, USA). In both cases, a portion of the laser power was used as the visible excitation beam, while most of the power was used to pump a TOPAS-Prime with a DFG2 extension (Light Conversion, Vilnius, Lithuania) to generate the tunable IR beam. At the sample, the angle of incidence and power for the IR beam were 55° and 3–4 mW at 1650 cm⁻¹ with a diameter of about 150 μm, while the 800 nm visible beam angle of incidence was set to 70° with a power of 10 mW and a diameter of about 350 μm. Spectra taken across the full range were measured in steps of 100 cm⁻¹ from 1300 to 2300 cm⁻¹ with 3 min averages per step and then stitched together to yield a continuous spectrum. For the CD_x region, the spectrum was measured in 5 steps from 2000 to 2300 cm⁻¹ for a total of 15 min of acquisition time, including a darkfield measurement. Before detection, the SFG signals were filtered spatially and optically using a slit between two positive lenses and a short-pass filter (RazorEdge 785 nm, Semrock, Rochester, NY, USA) before being passed through a tunable half-wave plate and a MgF₂ Rochon polarizer. The detection system, comprising a spectrograph (Kymera328) and an EMCCD camera (Andor, Belfast, Northern Ireland, UK), was calibrated each time a spectrum was taken by measuring characteristic vibrational peaks from a reference polystyrene sample. Nonresonant background signals were measured for each spectral step from substrates coated with Au and were used to compensate for the spectral profile of the IR beam. After collecting data for each KRas^G12D construct attached to the bilayer, the KRpep-2d peptide inhibitor with protein was injected into the sample cell and SFG spectra were collected again after a 20 min incubation.

SFG data analysis

The SFG spectra in the CD_x stretching region, 2000–2300 cm⁻¹, were fitted with Eqs. 1 and 2 using OriginPro 2023. Assuming the main contributions to the background are off-resonance signals that arise from the substrate, ordered water molecules, lipid bilayer asymmetry, and the attached proteins, we followed the strategy in previous work (29) and modeled the nonresonant background collectively as broad peaks (>100 cm⁻¹ width) in the OH stretching and fingerprint regions, located approximately 1000 cm⁻¹ above and below the fitted CD_x stretching region. The absolute values of the resonant CD_x peak amplitudes were extracted for each peak before (A_i) and after (A_f) addition of the KRpep-2d inhibitor. The difference between the final and initial amplitudes, normalized by their average, was evaluated for each observed peak.

Results

KRas^G12D protein constructs were made by CFPS, as described previously (27). Successful protein expression was verified with SDS-PAGE (Fig. S1). Each construct had an 8-His-tag at the C-terminus which allowed attachment to DOGS-NTA lipids charged with Ni²⁺ in a supported lipid bilayer on SiO₂-coated CaF₂. Each batch of KRas^G12D was produced with protein-wide isotopic labeling of a single residue type, achieved by substituting that particular amino acid with a deuterated version in the amino acid mixture provided for CFPS. Five different constructs were labeled with Ala-d₄, Asp-d₃, Gly-d₅, Leu-d₃, and Val-d₈, and we refer to them as KRas-A, KRas-D, Kras-G, KRas-L, and KRas-V, respectively. Mass spectrometry of the constructs confirmed that the isotopic labeling was successful as the proteins were heavier than unlabeled proteins with weights typically within a few Da from the theoretically calculated mass (Fig. S2). However, KRas-D was lighter than the expected value by 18 Da, indicating a less than complete replacement of the native aspartate residues.

A custom Teflon microfluidic cell was built to hold the SiO₂-coated CaF₂ hemisphere (10 mm diameter) in place, seated on an o-ring. The flat surface of the hemisphere was in contact with a circular chamber of 200 μL volume with inlet and outlet ports at either side for liquid injection. A three-way valve was installed in the line so air bubbles in the samples could be purged before injection of liquid into the sample chamber. This was necessary to prevent destroying the supported lipid bilayer on the SiO₂ surface. SUVs were first injected into the cell via a Luer-lock syringe followed by a buffer wash. The KRas^G12D protein constructs were then injected into the cell and attached to the lipid bilayer. After washing excess protein out with a buffer solution, the KRas-2d peptide inhibitor was introduced. During each step in the sample preparation, the SFG signal was monitored and spectra were captured before and after injection of the inhibitor. A schematic of the experimental setup and an energy diagram for SFG is shown in Fig. 1.

As the KRas^G12D constructs at 1 μM were allowed to attach to the supported bilayer via His-tag binding to the Ni-NTA lipids, the amide I SFG signal was monitored over time in ppp polarization (p-polarized SFG, visible, and IR beams) at 1650 cm⁻¹. The protein rapidly attached to the supported bilayer and the signal plateaued within 5 min. The signals were exceedingly strong, indicating a well-ordered layer of protein. Addition of EDTA or imidazole (200 mM) released the His-tagged protein from the bilayer and completely abolished the SFG signal within seconds, demonstrating specific and reversible attachment to the Ni-NTA lipids (protein could also be reattached after flushing out the EDTA or imidazole). With the protein molecules attached, broadband SFG spectra were measured across the spectral range 1300–2300 cm⁻¹ in ssp polarization (s-polarized SFG and visible beams, p-polarized IR beam) for the various KRas^G12D constructs with 1 μM protein in the chamber. The CD_x signals were initially larger than they were shortly after injection. To better understand this phenomenon, we injected a second aliquot of protein at 3 μM (after monitoring SFG signals during an hour-long incubation with 1 μM) and transiently observed larger CD_x signals, which again diminished within minutes to a much smaller level. For subsequent experiments investigating the effect of inhibitor binding, the protein concentration was 3 μM. After acquisition of a CD_x spectrum of protein alone, a mixture with the KRas^G12D protein construct at 3 μM and the inhibitor at 10 μM was injected into the cell and a spectrum was measured again across the same range.

A peak at ∼1415 cm⁻¹ was observed before addition of protein and assigned to the lipid molecules, indicating an asymmetry in the orientational order of the lipids between the two bilayer leaflets. Amide I signals were observed in the range 1620–1690 cm⁻¹, but no peaks were observed between 1800 and 2300 cm⁻¹ for unlabeled KRas^G12D (Fig. 2 A). Upon repeating the experiment with selectively deuterated protein, SFG peaks from CD_x stretching vibrations were observed in the range 2000–2300 cm⁻¹ in some of the KRas^G12D constructs (Fig. 2 B). The CD_x signals were initially stronger (Fig. 2), but decreased over a period of minutes with exposure to the laser beams (Fig. 3). This may be due to heating from the IR beam as discussed below. We detected CD_x stretching signals for KRas-A, KRas-V, and KRas-G, and the signals were strongest for KRas-V. KRas-D and KRas-L did not exhibit detectable CD_x signals with or without bound inhibitor. As the IR and visible beams overlap in an area with a diameter of about 150 μm, an upper limit for the number of molecules producing the CD_x signals is 10⁹ molecules assuming a maximum protein density of about 5 × 10¹² molecules/cm² on the surface, as measured for other proteins previously (21). This compares with about 3 × 10¹³ molecule/cm² at close-packed density for the KRas protein given a diameter of 2 nm. The density of Ni-NTA-bearing lipids in the supported lipid bilayers is expected to be about 2 × 10¹³ molecules/cm² given a mole fraction of 10% and we expect most Ni-NTA sites to be occupied. The Ni-NTA density and SNR could be increased by increasing the mole fraction.

Figure 2.

(A) The vibrational SFG spectrum of unlabeled KRas^G12D exhibits an amide I band at 1620–1690 cm⁻¹. A peak around 1420 cm⁻¹ is also observed, which arises due to asymmetry in the lipid bilayer membrane. No discernable features are observed between 1800 and 2300 cm⁻¹. (B) Amino acid-selective isotopically labeled KRas^G12D protein CD_x construct KRas-V (valine-d8) exhibits similar spectral features in the amide I and fingerprint region as the unlabeled proteins. Additional peaks in the spectral region 2000–2300 cm⁻¹ arise from the deuterated amino acids.

Figure 3.

Evaluation of SFG spectra for three different KRas^G12D constructs isotopically labeled with deuterated alanine-d4 (KRas-A), valine-d8 (KRas-V) and glycine-d5 (KRas-G). Peak changes were evaluated as the ratio of the difference (A_f –A_i) divided by the average (½^.A_f + ½^.A_i) of the peak amplitudes before (A_i) and after (A_f) addition of the KRpep-2d peptide inhibitor. (A) KRas-A exhibited four CD_x peaks due to CD₃ and CD vibrations, with minor (2146 and 2245 cm⁻¹) to moderate (2086 and 2214 cm⁻¹) changes in peak amplitudes upon inhibitor binding. (B) KRas-V initially exhibited weak CD_x signals due to CD₃ and CD vibrations that increased upon inhibitor binding. The relative amplitude changes were moderate (2207 cm⁻¹) to strong (2065, 2128, and 2260 cm⁻¹). (C) KRas-G had weaker CD₂ signals initially that also increased upon inhibitor binding, with strong relative amplitude changes for both peaks observed (2107 and 2165 cm⁻¹).

For the constructs that exhibited signals in the CD_x stretching region, we analyzed the data by peak fitting, using Eqs. 1 and 2 below,

$$I_{SFG}={|{\chi }^{\left(2\right)}\times E_{IR}{E}_{Vis}|}^2$$ (1)

$${\chi }^{\left(2\right)}={\chi }_{NR}^{\left(2\right)}+{\chi }_R^{\left(2\right)}={\chi }_{NR}^{\left(2\right)}+\sum _q\frac{A_q}{{\omega }_q-\omega +{i\mathrm{\Gamma }}_q}$$ (2)

where I_SFG is the intensity of the SFG signal, E_IR and E_Vis are the electric fields for the IR and visible excitation beams, and χ⁽²⁾ is the second-order susceptibility for SFG, which is modeled at each wavenumber, ω, with a nonresonant contribution ${\chi }_{NR}^{\left(2\right)}$ and a sum over resonant contributions ${\chi }_R^{\left(2\right)}$, where A_q, ω_q, and Γ_q are the amplitude, frequency and width for the q^th resonant peak. The spectra and peak amplitudes from the data fitting (Fig. 3) allowed us to analyze the KRpep-2d-induced conformational changes upon binding to the the KRas^G12D constructs. All three constructs with detectable signals in the CD_x stretching regions exhibited signal increases upon inhibitor binding, suggesting an increase in orientational order in the bound complex relative to the apo protein. KRas-A exhibits four vibrational peaks with relatively weak to moderate signal changes (Fig. 3 A). The largest relative signal increases were observed for KRas-V, which also has four peaks (Fig. 3 B). KRas-G has two peaks that exhibit a moderately strong change in amplitude upon inhibitor binding (Fig. 3 C). The number of peaks included in the fits and the vibrational mode assignments were chosen based on previous publications (30,31,32). All the fits converged and the fitting parameters are presented in Table 1. The number of these peaks is consistent with the isotopic labeling of each amino acid: glycine is only labeled at CD₂, whereas the alanine and valine contain both CD₃ and CD labels. In summary, KRas-G exhibits only two peaks, related to symmetric and asymmetric CD₂ stretches. KRas-A and KRas-V both exhibit two peaks due to symmetric and asymmetric CD₃ stretches, with a third peak related to CD stretches as well as a fourth peak, assigned as a Fermi resonance.

Table 1.

SFG fitting parameters for the deuterated residues in the KRas^G12D constructs

	ωq (cm−1)	Γq (cm−1)	Aq,i (a.u.)	Aq,f (a.u.)	Vibration
KRas-A	2086	10.8	0.055	0.052	vs(CD3)
	2146	8.4	0.048	0.052	vFR(CD3)
	2214	16.7	0.013	0.022	vas(CD3)
	2245	17.1	0.101	0.039	v(CD)
KRas-V	2065	15.1	0.042	0.141	vs(CD3)
	2128	20.8	0.037	0.176	vFR(CD3)
	2207	17.8	0.015	0.415	vas(CD3)
	2260	16.2	0.268	0.093	v(CD)
KRas-G	2107	18.1	0.014	0.124	vs(CD2)
	2165	5.8	0.002	0.006	vas(CD2)

Amplitudes before and after inhibitor binding (A_q,i and A_q,f) were extracted from the fits, all of which converged.

We evaluated the locations of the various residues that exhibit a change in CD_x stretching signals upon binding of the KRpep-2d inhibitor by inspection of the KRas^G12D-peptide crystal structure (PDB: 5XCO) (33). Individual deuterated amino acids are located in and around the binding pocket for nucleotide ligands (Fig. 4). The nucleotide binding site is the only locus on the protein’s surface at which all three residues are found in proximity to each other (Fig. S8). Valine is also located in the binding site for KRpep-2d. The nucleotide and peptide inhibitor binding sites are known to be in an allosteric network with each other, as the peptide does not bind to the protein in the absence of nucleotide (34). As the two binding sites must be connected allosterically, binding at one site is expected to change the conformation at the other, which we observed in our experiments.

Figure 4.

A surface view of the KRas^G12D protein with all residues colored for alanine (cyan), valine (magenta), and glycine (yellow) (figure prepared with ChimeraX). The KRpep-2d inhibitor (green) and nucleotide (white) are shown as ribbon and stick representations, respectively. KRas-V exhibited the largest peak amplitude change upon inhibitor binding, and valines are found at both the inhibitor and nucleotide binding sites. The crystal structure (PDB: 5XCO) shows that individual alanine, valine, and glycine residues are in close proximity only in or near the nucleotide binding site on the protein surface, which is in an allosteric network with the binding site of KRpep-2d peptide inhibitor.

Discussion

SFG spectroscopy is highly dependent on the orientation (tilt angle relative to the surface normal) and the orientational distribution width of the bond vectors probed in a protein molecule. The second-order susceptibility ${\chi }^{\left(2\right)}$ in Eq. 2 is a tensor with 27 elements. All nonzero elements depend on the number density of SFG-active molecules, the molecular hyperpolarizability, β, and the orientational averages of the vibrating bonds that contribute to each resonant peak. Chiral tensor elements, representing interactions between orthogonal E-field components, allow nonzero signals for isotropic samples; however, achiral tensor elements can only be nonzero for systems that exhibit long-range ordering of the probed molecules. Each polarization combination of the SFG, visible, and IR beams interrogates a subset of the ${\chi }_{ijk}^{\left(2\right)}$ tensor elements, which in turn are described by analytic functions that include the nonzero molecular hyperpolarizability tensor elements, ${\beta }_{lmn}$, weighted by factors from rotation matrices that transform the molecular coordinates (a, b, and c) into the lab frame coordinates (x, y, and z) (35). The only nonzero ${\chi }_{ijk}^{\left(2\right)}$ element probed by the ssp polarization for protein molecules attached to a surface with C_∞ symmetry, as is the case for the experiments in this study, is ${\chi }_{yyz}^{\left(2\right)}$. When probing, for example, methyl symmetric stretching vibrations, v_s(CD₃), the only nonvanishing hyperpolarizability tensor elements are ${\beta }_{aac}={\beta }_{bbc}$ and ${\beta }_{ccc}$ (36). This makes it possible to derive an expression for the probed tensor element, which depends on ensemble orientational averages of trigonometric functions integrated over the bond vector tilt angles and weighted by a distribution function. As SFG intensity scales quadratically with χ⁽²⁾, the SFG signal dependence on the average tilt angle is striking, with a sensitivity to conformational change similar to that of SHG (37,38). Accordingly, SFG is sensitive to subtle changes in the average orientation of a probed vibration where just a few degrees (or subangstrom shifts of a bond vector direction), due to ligand-induced conformational change, can be detectable. An evaluation of this kind allows us to conclude from the data collected for the KRas-V construct that either the average tilt angle, θ₀, the tilt angle distribution width, σ, or both, relative to the surface normal, shifts significantly to lower values upon ligand binding. Similar evaluations can be made for the other vibrational modes and residues. Although qualitative conclusions from such theoretical models are reliable, quantitative evaluation requires additional work beyond the scope of this contribution. This includes forming ratios of polarization combinations (e.g., ssp versus sps), determination of the corresponding interfacial Fresnel coefficients, and formalizing a custom angular distribution function to account for the specific orientations of the local, probed residues relative to the global protein orientation, expressed in order parameters.

Other techniques such as ATR-FTIR have been used to probe protein structure and conformational changes; however, they are often used to study secondary rather than tertiary structure, as we demonstrate here. SFG’s throughput and sensitivity to conformational changes, due to the nonlinear dependence on bond vector orientation, are also higher (39,40,41,42).

By selectively labeling different residues in the protein and measuring SFG signals before and after ligand binding, it is possible to extensively map conformational changes and allosteric networks, as suggested in an earlier study of SHG with proteins (43). Given that the largest conformational changes upon ligand binding are likely to be in and near the ligand binding site, as well as distal residues connected allosterically, it will arguably be possible to map the sites of ligand binding and of allosteric networks by SFG. We have demonstrated the feasibility of doing so here with KRas at the unique intersection of alanine, valine, and glycine around the nucleotide binding site, which shares an allosteric network with the peptide inhibitor binding site. In this study, we have made different constructs of KRas^G12D that are deuterated in one of five different amino acids. KRas-A, KRas-V, and KRas-G exhibit detectable SFG signal changes in ssp polarization upon binding of the KRpep-2d inhibitor. We observed CD_x signals that were stronger initially, but quickly decreased upon exposure to the IR and visible beams, which we attribute to heating artifacts (44). The samples also exhibit a tendency to higher CD_x signals upon inhibitor binding, suggesting a higher degree of orientational order (reduced σ) in the CD_x bonds due to decreased flexibility in the protein-ligand complex, or shifts in the average bond tilt angles (reduced θ₀) for the probed vibrations. KRas-L exhibited a very small SFG signal at around 2230 cm⁻¹ in ssp polarization upon inhibitor binding (Fig. S9), but the signal magnitude was not strong enough to allow reliable analysis by peak fitting. The small signal may be due to an average orientation of the leucine residues lying closer to the membrane plane. KRas-D did not exhibit CD_x signals in ssp polarization before or after addition of inhibitor, which may be due to incomplete isotopic labeling of this residue, as indicated by the mass spectrometry results (Fig. S2).

The relatively low SNRs in the CD_x region of the spectrum compared with the amide I band is due to the following combination of factors: 1) the deuterated residues constitute only a small fraction of the total number of residues in the protein, all of which contain C=O, 2) most of the CD_x bonds are in amino acid side chains that are more mobile and less ordered than the carbonyl C=O bonds of the backbone, and 3) the CD_x groups possess potentially lower molecular hyperpolarizabilities (β) than the C=O bonds. KRas-V exhibited the largest CD_x initial signals and signal changes among the valine-, lysine-, and glycine-labeled constructs. Valine is the more abundant amino acid in the KRas construct (15 out of 178 residues vs. 10 glycines and 9 alanines). The SNR of the CD_x signals could be substantially improved, for example, by translating the sample (45) during acquisition or by better temperature control and circulation of the buffer. In addition, the SNR could be improved by longer integration times, both for the protein samples and the Au-coated hemispheres used to normalize the spectra, for example, by increasing the signal averaging time to 30 min per step. Increasing the protein surface coverage by using a higher mole fraction of DOGS-NTA(Ni²⁺) in the bilayer, would also substantially improve the SNR as the SFG signal scales quadratically with the molecular number density at the surface. Moreover, the protein may be at least partly polyvalently bound to the Ni-NTA-bearing lipids in the experiments described here (46), further reducing the surface density. By using a longer His-tag, for example, we could increase the monovalent affinity of a protein for the Ni-NTA, shifting the equilibrium closer to a 1:1 stoichiometry for bound protein and Ni-NTA lipids. Finally, machine learning strategies have previously been employed in nonlinear optical spectroscopy to subtract nonresonant background signals and improve the SNR severalfold (47,48).

The exceedingly high SNR of the amide I band with our surface attachment approach further suggests the technique’s use as a method to obtain vibrational spectra of proteins without the complication of a water background, which is excluded here as water molecules have a low Raman cross section and are predominantly isotropically oriented in the excitation volume.

In future work, SFG spectra measured in multiple polarizations (e.g., ssp, sps, and ppp) could be used to determine the structure of a protein-ligand complex in aqueous environments as compared with the apo structure. For example, as the ssp, sps, and ppp polarizations probe different elements of the susceptibility tensor elements, they could be used to determine the average angular changes of CD_x bond vectors and distribution widths that occur for a particular ligand-induced conformational change. Similar approaches have been suggested and achieved for biomolecules by SHG (49), and previous studies indicate that corresponding investigations by SFG is feasible by probing vibrations inherent in proteins (6,50,51). It is worth noting that the standard way to analyze such data has evolved over the years and often includes simulations of SFG spectra for the whole protein, based on assumed orientations from the crystal structure or on results from molecular dynamic simulations (6,13,50,52,53). While promising, such evaluations may present an underdetermined problem, as the orientational distribution of any vibration that occurs naturally in a protein will be broad, making it a challenge to assign a unique structural change to observed shifts in SFG signals. Our results in this work demonstrate a potential solution to this problem: to selectively label different parts of the protein and analyze them separately, isolating the origin of signal shifts during ligand-induced conformational changes. Given the tiny amount of sample required for each measurement, making hundreds or even thousands of unique protein constructs with different combinations of isotopically labeled residues is possible and could yield sufficient independent spectra to determine the structure of a protein-ligand complex and its conformational landscape, particularly if the structure of the apo form and the corresponding SFG spectra are known. The technique could also be used as a means to identify a ligand’s binding site, by triangulating the residues that exhibit conformational changes upon ligand binding, or map allosteric networks, as we have established feasibility for here. This is analogous to common NMR strategies, but with a much smaller sample requirement, as proteins and complexes of any size can be studied by SFG.

To obtain additional spectra to enhance the precision in the analysis, proteins could be attached to the bilayer in different orientations, for example, via a His-tag at the C- vs N-terminus. Mutational scanning could also be used (54,55,56) to dissect the contribution of each residue to the observed spectra and provide additional information for more accurate structural modeling. Methods such as native chemical ligation and expressed protein ligation could be employed to obtain site-specific isotopically labeled protein constructs (57,58,59,60). To rapidly obtain SFG spectra for many different isotope-labeled protein constructs or variants, we envision scaling SFG to high throughput to obtain a large body of independent spectra for structural modeling and machine learning, which could ultimately support predictions of ligand-bound structures in an analogous way to AlphaFold and related methods for the (mainly apo) structure data set in the Protein Databank (61). In all envisioned studies, the use of the lipid bilayer or an equivalent surface that preserves protein native structure and function, including ligand-induced conformational changes, seems critical, and has often been overlooked. Supported bilayers with transmembrane proteins, too, could be prepared in a variety of ways including lipid-detergent exchange on a surface (62). Finally, with SFG’s virtually instantaneous time resolution, it may be possible to measure conformational changes in real time and space by modeling many independent spectra obtained at different time points after ligand binding.

In conclusion, our work demonstrates a surface-mediated approach for mapping conformational changes and allosteric networks in proteins based on SFG, enabled by creating a highly ordered and functional protein monolayer on a supported lipid bilayer with amino acid-selective isotopic labeling. Ligand-induced conformational changes are evaluated by measuring SFG signal amplitudes before and after ligand binding. We anticipate the technique providing new ways of studying conformational changes in solution and identifying therapeutics that bind to specific conformations in protein targets or complexes. Sufficient independent SFG spectra of a protein isotopically labeled in different ways could enable the determination of a protein-ligand structure, particularly given a known apo structure.

Acknowledgments

This material is based upon work supported by the National Science Foundation (under NSF grant no. 2111821). Part of this work was conducted at the Washington Nanofabrication Facility/Molecular Analysis Faculity, a National Nanotechnology Coordinated Infrastructure (NNCI) site at the University of Washington with partial support from the National Science Foundation via awards NNCI-1542101 and NNCI-2025489. J.S. gratefully acknowledges many stimulating and insightful discussions with Eric Tyrode and his generous assistance with the experiments and the sample cell design. Financial support by the Australian Research Council Centre of Excellence for Innovations in Peptide and Protein Science to E.A. and G.O. (grant no. CE200100012) is gratefully acknowledged.

Author contributions

E.A. prepared the samples. G.O. supervised their preparation. P.K.J. prepared and characterized the substrates and analyzed data. J.S. conceived the technique, performed the experiments, and supervised the project. P.K.J. and J.S. wrote the manuscript.

Declaration of interests

J.S. is the founder of Skylight Discovery, Inc. P.K.J. contributed to this work as a paid consultant.

Editor: Ronald Koder.