Time-Resolved Resonance Raman Spectroscopy and Application to Studies on Ultrafast Protein Dynamics

Abstract

Protein dynamics play a fundamental role in allosteric regulation, which is vital to the function of many proteins. In many proteins, rather than a direct interaction, mutual modulation of properties such as ligand affinity at spatially separated sites is achieved through a conformational change. Conformational changes of proteins are thermally activated processes that involve intramolecular and intermolecular energy exchanges. In this account, I review the work of my team on the development and applications of ultrafast time-resolved resonance Raman spectroscopy to observe functionally important protein dynamics. We gained insights into conformational dynamics upon external stimulus and energy flow with a spatial resolution of a single amino acid residue using time-resolved visible and ultraviolet resonance Raman spectroscopy. The results have contributed to a deeper understanding of the structural nature of protein motion and the relationship of dynamics to function. I discuss the protein dynamics and allosteric mechanism in terms of the nature of the high packing density of protein structures. In addition, I present a view of the future of molecular science on proteins.

1. Introduction

A fundamental challenge of biophysical chemistry is relating the dynamics of proteins to their function. Protein dynamics span an enormous range of temporal and spatial scales, including small atomic perturbations on the picosecond timescale, loop and domain motions on the nanosecond timescale, and larger conformational rearrangements on the microsecond and millisecond timescale.^1,2 A multidimensional energy landscape underlies dynamics continuously distributed in wide temporal and spatial ranges and relate to one another on different scales.

The connection between protein dynamics and function is based in allostery. The term “allostery,” derived from the Greek allos, meaning “other,” and stereos, meaning “solid (object),” implies that a change in a protein site modulates a property at another site. In allosteric processes, the activation event causes a local conformational change propagated through residue-to-residue contacts to the rest of the protein,³ which constitutes dynamics. Thus, allosteric communication is essentially dynamic in nature.^4–6

Allosteric regulation is used by proteins to control their activity.^6–8 However, achieving this type of sophisticated regulation in artificial molecules remains a huge challenge, despite some progress made in the development of artificial allosteric systems.⁹ Once the principle of allostery is well understood, it can be applied directly to the design and creation of highly functional molecules, which provides new frontiers in molecular science and extends molecular potential. Therefore, elucidating the principle of allostery is important, not only to gain a deeper understanding of proteins, but also to create highly functional molecules such as molecular machines. Although numerous allosteric proteins have been investigated, the principles that enable the communication between distant sites remain largely elusive.

Structural dynamics are linked to energetics. Protein molecules require energy to overcome activation barriers and undergo conformational changes. This energy is exchanged among the protein’s different internal nuclear degrees of freedom. The frequency that the reactive mode obtains energy sufficient for overcoming the energy barrier determines the pathway and rate of conformational changes. Thus, energy exchange dynamics among these different degrees of freedom determine the fluctuation and dissipation processes that govern protein conformational changes central to understanding how energy stored in the active site drives protein motion. Excess energy is generated immediately after reaction at a protein active site. For example, a primary step of photobiological reactions involves the deposit of excess energy in the chromophore immediately after photoexcitation, photoreaction, and nonradiative transition. As a result, primary structural changes occur along with dissipation of the excess energy. Thus, elucidation of the rate and mechanism of energy flow in the protein matrix and dissipation into the solvent is important.

Distribution of dynamics in a wide range is a striking feature of proteins. To understand the relationship between protein dynamics and function, the concatenation of experimental results obtained over many orders of magnitude of time is necessary to examine protein structure evolving from the earliest moments such as the picosecond regime toward time scales that are highly relevant to its function such as the microsecond or millisecond regime. Having a single experimental technique to observe the dynamics from picoseconds to milliseconds is beneficial. Vibrational spectroscopy has advantages as an experimental probe for studying dynamic processes over a wide range of timescales. The faster processes affecting protein motion involve relatively small alterations in overall structure. Therefore, the probes used to examine them must be sensitive to subtle protein changes. Thus, we decided to use a vibrational spectroscopic method, which is sufficiently structure sensitive at a chemical bond resolution to identify any ultrafast steps in conformational dynamics.

Resonance Raman (RR) spectroscopy provides vibrational spectra as a sensitive structure probe for proteins. One advantage of RR spectroscopy is enhancement of the scattering due to the resonance Raman effect. Resonance Raman spectra are obtained by irradiating samples with a monochromatic light source with a photon energy close to that of an electric-dipole-allowed electronic transition. This effect arises from electron-nuclear coupling of the electronic and vibrational transitions.¹⁰ Selective resonance enhancement is particularly important in studying proteins because sample concentrations can be as low as 10⁻⁶ M and scattering from the nonresonant portions of protein and buffer will not complicate the vibrational spectra. Vibrational modes that show enhancement are localized on the group of atoms that give rise to the electronic transition. Because of its unique sensitivity and selectivity, RR spectroscopy is particularly well suited for studies of protein structure. Using these advantages to characterize the structural changes in various protein sites advances the understanding of how allostery is achieved and how proteins function. Visible RR spectroscopy has helped elucidate molecular structures and reaction mechanisms of prosthetic groups such as heme and retinal, which have strong electronic transitions in the visible wavelength region.^11–14 Extension of the probe wavelength to ultraviolet (UV) and far UV regions have enabled us to monitor aromatic amino acid residues, such as tryptophan (Trp), tyrosine (Tyr), and phenylalanine (Phe), and peptide bonds, which have vibrational modes sensitive to the secondary structures of proteins.^15–19 Pulsed lasers, which extend the ability to investigate time domains, now allow time-resolved RR studies on protein dynamics. This time-resolved RR spectroscopy is capable of elucidating the temporal evolution of protein structures. The time resolution of vibrational spectroscopy is determined, in principle, by the vibrational dephasing time, that is ∼1 ps in condensed phases. Along with developments in ultrafast laser technology and optical detectors, recent methodological advancements have enabled the observation of protein dynamics in unprecedented detail, revealing intriguing aspects of allosteric mechanisms and energy flow in proteins.

This account is organized to review accomplishments of my research group, with an emphasis on developments of time-resolved RR methods that take advantage of the benefits of RR spectroscopy and new attempts that have been made possible only by utilizing advantages of RR spectroscopy. This provided opportunities for more detailed studies of the primary structural changes involved in allostery. The report begins with a description of time-resolved RR spectrometers we developed, followed by reviews of selected works on structural dynamics and energy flow in proteins from my research group. Then, protein dynamics, energetics, and architecture are discussed based on the resulting insights, and the final section presents a perspective of studies on protein dynamics.

2. Time-Resolved Resonance Raman Spectrometers with Picosecond Time Resolution

In spontaneous Raman scattering, temporal pulse widths of the pump and probe pulses determine the time resolution. For optical pulses, the product of the temporal width and the spectral width cannot be smaller than a certain limiting value—the Fourier transform limit. Thus, spectral width increases as temporal width decreases. For a Gaussian pulse, the narrowest possible spectral width of a 1-ps pulse is 14.7 cm⁻¹.²⁰ The temporal width of 1 ps is the smallest practical limit for obtaining spontaneous Raman spectra with a reasonable spectral resolution.

When we started studying ultrafast protein dynamics in the mid-90s, picosecond time-resolved RR spectroscopy was not used as widely as nanosecond time-resolved RR spectroscopy because of practical limitations. The most crucial factor was a lack of a light source that fulfills the requirements for small timing jitter, appropriate repetition rates, and wavelength tunability of pulses applicable to picosecond time-resolved RR spectroscopy. Time-resolved RR measurements require pump and probe pulses. The wavelength of the pump pulse needs to fall within the electronic absorption band of the prosthetic group (such as retinal, flavins, and heme) of proteins to be photoexcited. At the same time, the wavelength of the probe pulse must be close to that of an electronic transition of the specific portion of the protein being investigated so that resonance enhancement of the Raman bands can occur. Furthermore, the spectral width of the probe pulse must be narrow enough to obtain well-resolved vibrational bands. To obtain time-resolved RR spectra of a wide variety of molecules with high signal-to-noise ratios within a reasonable measurement time, two independently tunable light sources with a high repetition rate for the pump and probe beams are needed. Repetition rates in the kilohertz range are desirable for practical purposes, but when we started the studies, no one had succeeded in observing picosecond time-resolved RR spectra with a widely tunable kilohertz pulse source. We developed spectrometers containing widely tunable light sources for picosecond time-resolved RR spectroscopy using a 1-kHz picosecond Ti:sapphire laser/regenerative amplifier system.^21–24 The amount of protein sample solution available for spectroscopic measurements is often limited. Therefore, the laser system must have greater stability and the detection system needs to possess higher sensitivity to obtain measurements for protein samples than for small molecules. We also constructed devices for sample circulation to obtain high signal-to-noise spectra for a limited amount of protein sample.²⁵

2.1 Time-Resolved Visible Resonance Raman Spectrometer

Figure 1A is a schematic of the time-resolved RR spectrometer using a visible probe pulse used in our laboratory. A picosecond mode-locked Ti:sapphire oscillator (Spectra-Physics, Tsunami 3950), pumped by an Ar⁺ ion laser (Spectra-Physics, BeamLok 2060) or a diode-pumped solid-state laser (Spectra-Physics, Millennia Vs) produced approximately 1.5-ps pulses with a repetition rate of 82 MHz and an average power of about 0.7 W. The seed pulse was amplified by a regenerative amplifier (Positive Light, Spitfire) operated at 1 kHz by pumping with the 527-nm output of an intracavity frequency-doubled Nd:YLF laser. The entire laser system was covered with a plastic sheet, equipped with dust cleaners containing high-efficiency particulate air (HEPA) filters to keep the laser system dust-free. As an example, typical experimental conditions, employed for time-resolved experiments of heme proteins (Section 3.1), used an amplification unit that provided 784-nm pulses with a spectral width of 6 cm⁻¹ in a nearly TEM₀₀ mode at 1 kHz. In the probe arm, a probe pulse of 442 nm was generated as the first-order Stokes stimulated Raman scattering from compressed methane gas in a Raman shifter excited by the second harmonic of the 784-nm output. Components other than the first-order Stokes scattering were removed spectrally with a glass filter and dichroic mirrors and spatially with a Pellin-Broca prism. Detailed characterization of the picosecond pulses of the stimulated Raman scattering from compressed methane or hydrogen gas was obtained.²⁴ The energy of the first Stokes stimulated Raman scattering was between 1.0 and 1.5 µJ and the bandwidth was 14 cm⁻¹. The first Stokes scattering was attenuated to 0.1 µJ using a Cr-coated quartz ND filter. In time-resolved RR experiments, using weak probe power is important. The probe power was selected empirically so that no photoproduct bands were observed in the probe-only spectrum. In the pump arm, a pump pulse of 530–600 nm was generated with a home-built OPG and OPA, which were pumped with the second harmonic of the output of the amplified laser. Properties of the picosecond pulses generated by the OPG/OPA system were characterized in detail.²³ The tunability of the pump pulse is shown in Figure 1D. The pulses were used for the pump beam for time-resolved visible RR measurements after attenuation to an appropriate magnitude using a Cr-coated quartz ND filter. The pump power was selected so that there was no saturation effect. The pump and probe beams were made to co-propagate using a dichroic mirror. Polarization of the pump beam was rotated by 55° relative to that of the probe beam to minimize the effects of molecular rotations on the observed kinetics. Both beams were continuously monitored with photodiodes (Hamamatsu Photonics, S2387-1010R) and were stable to within ±10%. Figure 1B shows a cross-correlation trace of the pump and probe pulses measured with a 1-mm BBO crystal, indicating a width of 2.3 ps. The time zero of the pump and probe delay (uncertainty <0.2 ps) was calibrated using sum-frequency mixing in the same crystal. The tunability of the probe and pump pulse are shown in Figure 1C and 1D, respectively.

The sample solution was placed in a 10-mmϕ NMR tube and spun with a device designed to minimize the off-center deviation during rotation.²⁶ The sample was spun at 3400 rpm in the spinning cell, which was configured for 135° backscattering illumination and collection. This configuration was important to minimize the effects of molecular rotations on the observed kinetics.²⁷ Spherical and cylindrical lenses were used to focus the pump and probe beams on the sample.

Raman scattering was collected by a doublet achromat (80-mm focal length (FL), f/2) and was imaged onto the 200-µm entrance slit of a single spectrometer (Spex, 500M) using a doublet achromat (200-mm FL, f/5). A dichroic short-pass filter was placed between the lenses to remove the scattered pump beam. A holographic notch filter (Kaiser Optical Systems, HSNF-441.6-1.0) was used to reject the unshifted scattering. A polarization scrambler was placed at the entrance slit to remove the effects of polarization on the spectrograph throughput. The spectrograph was equipped with a blazed holographic grating (2400 grooves/mm) that enables measurements of a spectrum as wide as about 1000 cm⁻¹ in the Soret region and with a spectral slit width of approximately 8 cm⁻¹. The dispersed light was detected by a liquid-nitrogen-cooled CCD detector (Princeton Instruments, CCD-1100PB). Raman shifts were calibrated with cyclohexane, benzene, or carbon tetrachloride. The peak positions of the Raman bands were accurate within ±2 cm⁻¹.

2.2 Time-Resolved Ultraviolet Resonance Raman Spectrometer

Ultraviolet resonance Raman (UVRR) spectroscopy is a versatile technique for studying protein structures because it enables the observation of Raman bands of aromatic amino acid residues and polypeptide backbones with high selectivity.^15–18 Several vibrational bands of aromatic residues can be used as structural markers for proteins; hence, time-resolved UVRR spectroscopy can provide site-specific information about protein dynamics. After the successful development of the time-resolved visible RR spectrometer, we extended the wavelength region of the probe pulse to the UV and far UV regions by constructing an apparatus consisting of a widely tunable light source in the UV and far UV regions using a 1-kHz picosecond Ti:sapphire laser/regenerative amplifier system.²² Figure 1E shows a schematic of the time-resolved UVRR measurement apparatus in our laboratory. A Ti:sapphire oscillator (Tsunami pumped by Millennia-Vs, Spectra-Physics) and amplifier (Spitfire pumped by Evolution-15, Spectra-Physics) system operating at 1 kHz provided 778–820 nm pulses, each with an energy of about 0.8 mJ and duration of 2.5 ps in a nearly TEM₀₀ mode under operation at 1 kHz. The entire laser system was covered with a plastic sheet, equipped with dust cleaners containing HEPA filters to keep the laser system free of dust. In the pump arm, a pump pulse of 530–600 nm was generated using a home-built optical parametric generator (OPG) and amplifier (OPA), which were pumped with the second harmonic of the amplified laser. To generate a pump pulse with a shorter wavelength (439–494 nm), stimulated Raman scattering in compressed methane or hydrogen gas was excited by the second harmonic beam of the laser output. In addition, the second harmonic (389–410 nm) could be used directly as a pump pulse. The tunability of the pump pulse is shown in the right panel of Figure 1H. In the probe arm, the second harmonic of the laser output was focused into a Raman shifter filled with methane or hydrogen gas to generate the first-order Stokes stimulated scattering in 439–494 nm. For example, a UV probe pulse at 225 nm was generated with a BBO crystal as the second harmonic of the 450-nm output. In this way, a UV pulse was generated in the wavelength range of 220–247 nm. Sum frequency generation between the second harmonic and the stimulated Raman scattering was generated to produce a UV probe pulse in the range of 206–218 nm. Tunability of the probe pulse is also shown in the right panel of Figure 1G. Light components other than the probe pulse were eliminated spatially with a Pellin-Broca prism and spectrally with dichroic mirrors.

After the pump and probe beams were made co-propagating using a dichroic mirror, they were focused with a spherical lens onto a flowing thin film of the sample solution formed by a wire-guided jet nozzle.²⁸ The focused spot sizes were 150 µm (fwhm) for the probe beam, and 250 µm (fwhm) for the pump beam. At the sample, the energies of the probe and pump pulses were attenuated to 0.5 and 5 µJ, respectively, using Cr-coated quartz ND filters. The two beams were configured for 135° backscattering illumination and collection. The pump power was selected to prevent a saturation effect. A cross-correlation trace of the pump and probe pulses measured by difference frequency generation with a thin BBO crystal indicated a pulse width of 3.0–3.7 ps. A representative example of the cross-correlation trace is shown in Figure 1F. The intensity of the pump and probe pulses was monitored using photodiodes (S2387-1010R, Hamamatsu Photonics) and were stable within ±10%. Raman scattered light was collected by an f/2 quartz doublet achromat (100-mm FL) and focused by an f/4 quartz doublet achromat (200-mm FL) onto the entrance slit of a Czerny-Turner configured Littrow prism prefilter²⁹ coupled to a 50-cm single spectrograph (500M, SPEX). The prism prefilter was constructed to eliminate Rayleigh scattering because no holographic notch filter was available for wavelengths shorter than 400 nm. The spectrograph was equipped with a 1200-grooves/mm, 500-nm blazed grating operating in the second order, or a 2400-grooves/mm, 250-nm blazed grating operating in the first order. Dispersed light in the spectrograph was detected with a liquid-nitrogen-cooled CCD detector (SPEC-10:400B/LN, Roper Scientific) with Unichrome UV-enhancing coating. Raman shifts were calibrated with cyclohexane to an accuracy of ±4 cm⁻¹.

2.3 Data Acquisition and Analysis

The time-resolved RR data acquisitions were conducted as follows. For visible RR experiments, in the forward scan of delay time, delay times were changed by increasing them. Measurements in the backward scan of delay time were done following the same procedure except for the direction of change in delay time. In UVRR experiments, the sequence of delay times in the time-resolved measurements were chosen to be random in each scan. At each delay time, Raman signals were collected for three 20-second exposures with both the pump and probe beams present in the sample. This was followed by equivalent exposures for pump-only, probe-only, and dark measurements. Time-resolved Raman data were obtained by averaging the data for the 10–100 measurement cycles described above depending on the signal-to-noise ratio and the amount of protein sample. This method avoided the errors caused by a slow drift in laser power and allowed quantitatively reproducible spectra from one day to the next, which was possible because of the excellent long-term stability of the laser system. The pump-only spectrum was subtracted directly from the pump-and-probe spectrum, yielding the “probe-with-photolysis” spectrum. The dark spectrum was subtracted directly from the probe-only spectrum, yielding the “probe-without-photolysis” spectrum, which was the spectrum of the photolyzed state. The probe-without-photolysis spectrum was subtracted from the probe-with-photolysis spectrum to yield the photoproduct spectrum.

To extract the spectral components of the photoproducts, we calculated difference spectra between the probe-with-photolysis spectra and probe-without-photolysis spectra. In visible RR spectra, the difference spectra were calculated so that the contributions of the unreacted species were subtracted. First, scattering intensity for the change in the optical absorption of the sample at each time point was corrected by normalizing the data to the intensity changes of the 982-cm⁻¹ band of sulfate ions dissolved with the sample. After normalizing the band intensities in all of the spectra, the probe-without-photolysis spectrum was subtracted from the probe-with-photolysis spectrum to generate the difference spectrum. The subtraction parameter was determined by subtracting the probe-without-photolysis spectrum from the probe-with-photolysis spectrum until negative features were seen in the location of prominent bands of the reactant. The subtraction parameter then was reduced until these negative peaks were just eliminated, thereby accounting for the depletion of reactant caused by the pump pulse. In the data analysis of time-resolved Stokes UVRR spectra, the scattering intensity for the change in the optical absorption of the sample at each time point was corrected by normalizing the data to the intensity changes of the OH stretching band of water at ∼3400 cm⁻¹ of the sample solution because the spectrograph enabled measurements of a spectrum as wide as about 3500 cm⁻¹ (500–4000 cm⁻¹).²² For Stokes spectra of amino acid residues, one-to-one difference spectra were calculated by subtracting the probe-without-photolysis spectra from the probe-with-photolysis spectra without multiplying the subtraction parameter, because vibrational frequencies of the aromatic amino acid residues between reactant and products were similar enough that product RR bands were not resolved from those of reactant, and thus it can be difficult to uniquely determine the magnitude of the subtraction parameter. In the data analysis of time-resolved anti-Stokes UVRR spectra, the scattering intensity for the change in the optical absorption of the sample at each time point was corrected by normalizing the data to the intensity changes of the 982-cm⁻¹ band of sulfate ions dissolved with the sample because the anti-Stokes band of the OH stretching mode was too weak. After normalizing the band intensities in all of the spectra, the probe-without-photolysis spectrum was subtracted from the probe-with-photolysis spectrum to generate the pump-induced difference spectrum.

3. Primary Structural Dynamics of Allosteric Proteins

Using the time-resolved RR instruments developed, we pursued detailed studies for paradigmatic allosteric proteins that undergo structural changes crucial to their functions. I will review the results for gas-binding proteins and photoreceptor proteins. Structural changes could be initiated by photoreactions in prosthetic groups for both protein classes. Structural dynamics following photodissociation were examined using a combination of visible and ultraviolet RR spectroscopy.

3.1 Dynamics of Gas-Binding Proteins (Protein Dynamics Induced by Ligand Photodissociation)

The molecular mechanism of cooperativity in oxygen binding of hemoglobin (Hb) is a classic example of protein structural changes accompanying a reaction at a specific site that must spatially extend to the mesoscopic dimensions of the protein to achieve the biological function.³⁰ In Hb allostery, large-amplitude motions at the quaternary level, which are driven by structural changes at the subunit interface, are triggered by changes in the tertiary structure upon ligation. Myoglobin (Mb) can serve as a model system for the tertiary relaxation processes. Structurally very similar to a subunit of Hb, Mb is involved in oxygen (O₂) storage in muscles. Twenty years after a pioneering microsecond flash photolysis study by Gibson,³¹ the first ultrafast optical spectra of heme proteins were reported.³² Ultrafast optical absorption studies^32,33 revealed that the excited-state surface of the ligated heme is dissociative, allowing photoinitiation of structural changes from the ligated form to the deligated form within 50 fs with high quantum yield.³⁴ Using carbon monoxide (CO) as the ligand rather than O₂, the geminate recombination processes complicating the reaction dynamics in the picosecond regime could be avoided.³⁵ Thus, the carbonmonoxy myoglobin (MbCO) system is particularly interesting for protein dynamics because it provides access to the full range of protein relaxation processes in a single perturbation, namely ligand dissociation without complicated side reactions and associated population dynamics. Thus, an enormous body of work has accumulated about Mb relaxation processes, which have helped to establish general principles related to protein dynamics.

3.1.1 Myoglobin

The three-dimensional structure of the equilibrium deligated form of Mb (deoxyMb)³⁶ is shown schematically in Figure 2A, in which the arrow points to the heme prosthetic group. Heme is an iron-protoporphyrin IX (Fe²⁺PPIX), in which the Fe²⁺ ion is bound to the proximal histidine (His) in Mb. The heme iron binds diatomic molecules, such as nitric oxide (NO), CO, and O₂, at the opposite side of the proximal His. X-ray crystallography provided extensive information about the end-point equilibrium structures, as shown in Figure 2B. The magnitude of atomic displacements at the heme site that must occur upon ligand binding and release can be estimated by examining the structures containing the heme with and without bound ligands. The ligated heme has a planar structure in which the low-spin iron atom is in the porphyrin plane.³⁶ Upon ligand dissociation, the iron atom is converted from low to high spin and moves out of the porphyrin plane by approximately 0.3 Å.

3.1.1.1 Structural Dynamics of Heme;

First, we studied structural changes that occur in the heme group because heme is a ligand binding site and, hence, a starting site of allosteric dynamics.³⁷ Figure 2C and 2D show the time-resolved RR difference spectra of photodissociated Mb in the 150 to 850 cm⁻¹ and the 1080 to 1680 cm⁻¹ regions, respectively, for different delay times of the probe pulse with respect to the pump pulse. In these spectra, the contribution of unreacted species has been subtracted. The fraction of photolyzed MbCO was estimated to be 6% based on the intensity loss of the Raman bands of MbCO. At a −5 ps delay, no difference features were observed; this result corroborates the cross-correlation measurement of 2.3 ps. The time-resolved RR spectrum in the 1300 to 1650 cm⁻¹ region for a 1-ps delay contained only the bands arising from the in-plane vibrations of heme at 1352 (ν₄), 1560 (ν₂), and 1617 (ν₁₀) cm⁻¹.^38–40 The characters of the vibrational modes discussed in this account are summarized in Table 1. The ν₄, ν₂, and ν₁₀ bands exhibited an appreciable narrowing and a frequency upshift in the first few picoseconds, due to vibrational energy relaxation (discussed in Section 4.1), but no further changes occurred. The time-resolved RR spectrum for a 10-ps delay closely resembles the spectrum of deoxyMb, indicating that the photodissociated heme has relaxed to the equilibrium structure within the instrument response time (∼2 ps). Most interestingly, the ν₂ band, whose frequency is sensitive to the core size of the porphyrin ring,^41–44 appeared at a position close to that of the ν₂ band of deoxyMb. This demonstrates that initial expansion of the core size occurred simultaneous with CO photodissociation. Thus, the heme in Mb is led to the ground electronic state of the dissociated form, in which the iron atom is converted from low- to high-spin states. This spin-state transition transforms the structure of the heme from planar to domed within the instrumental response time, that was 2.3 ps.

3.1.1.2 Structural Dynamics of Heme-Polypeptide Chain Linkage;

Next, the focus was shifted to the iron-histidine (Fe–His) bond between the heme and proximal His.³⁷ The Fe–His bond is the sole covalent bond between the heme group and polypeptide chain, and it functions as the focal point for the forces driving the structural changes. Thus, time-resolved RR studies of the proximal His are key to elucidating protein dynamics. The assignment of the stretching mode of the Fe–His bond [ν(Fe–His)] by Kitagawa and co-workers prompted a variety of investigations because this bond serves as a good measure of the tertiary and quaternary structures of globin.^12,13 To study protein relaxation upon ligand dissociation, the temporal behavior of the ν(Fe–His) band was examined and compared with those of other heme Raman bands.

Figure 2C shows time-resolved RR spectra in the 150 to 850 cm⁻¹ region and delay of between −7 and 1000 ps. The bands at 220, 301, 341, 369, and 671 cm⁻¹ in the spectrum at a 1000-ps delay were assigned to heme vibrations: ν(Fe–His),⁴⁵ out-of-plane methine wagging (γ₇),^46,47 in-plane iron–pyrrole nitrogen stretching (ν₈),^38–40 propionate deformation [δ(C_βC_cC_d)],⁴⁶ and pyrrole ring deformation modes (ν₇),^38–40 respectively. Plots of frequencies of the ν(Fe–His) and γ₇ bands of Mb against time are shown in Figure 2E. For Mb, the ν(Fe–His) band showed a downshift in a 100-ps time range, while the γ₇ band showed no shift in this range. The same measurements were carried out for a model system without a protein matrix, involving a 2-methylimidazole (2MeIm) complex of iron protoporphyrin-IX, (Fe²⁺PPIX)-2MeIm, in a 0.1% aqueous CTAB (hexadecyltrimethylammonium bromide) solution. The band of the Fe–2MeIm stretching mode [ν(Fe–Im)] for the (Fe²⁺PPIX)-2MeIm complex showed no frequency shift with an increase in delay time (Figure 2E). Therefore, the downshift of the ν(Fe–His) band of Mb was attributed to some change in the ligation state of the proximal His caused by a structural change of the protein moiety. The downshift of the ν(Fe–His) band can be described by single-exponential kinetics, which provided a time constant of 106 ± 14 ps, although the kinetics would involve several relaxation steps. Relaxation of the Mb protein upon CO dissociation previously was thought to be complete within 30 ps because the previous time-resolved RR measurements with a 30-ps wide pulse could not detect any frequency difference between the initial 30-ps spectrum and the equilibrium spectra.⁴⁸ However, this high-precision study revealed the presence of a relaxation process of Mb in a 100-ps scale.³⁷

The most probable origin of the ν(Fe–His) downshift of the photodissociated Mb is a change in the hydrogen bond of the proximal His (His93) to the nearby residues. The change in the hydrogen bond causes a change in the mixing and delocalization of electron density from the σ* orbital of the Fe–His bond to the π* orbital of the porphyrin ring, which affects the strength of the Fe–His bond.⁴⁹ Changes in the basicity of an imidazole ring of a His residue are known to affect the frequency of the ν(Fe–His) mode.¹² Crystallographic data for Mb showed that the N_ε proton of His93 is hydrogen-bonded to the backbone carbonyl of Leu89 and to the O_γ atom of Ser92.³⁶ Accordingly, the 100-ps relaxation process involves a structural change in the heme pocket, including a change in the hydrogen bond between His93 and the nearby residues.

3.1.1.3 Structural Dynamics of the Protein Moiety;

Following the visible RR observations on heme and the Fe–His bond, we focused on the protein moiety to investigate the propagation of structural changes of Mb using time-resolved UVRR spectroscopy.^22,50 Figure 3A shows time-resolved UVRR difference spectra of Mb upon CO dissociation at delay times from −5 ps to 1000 ps. The UVRR Trp and Tyr residue bands in the spectra were identified by comparison with UVRR spectra of aqueous amino acid solutions. The bands labeled W and Y in Figure 3A have arisen from Trp and Tyr residues, respectively. Mode assignments made by Harada and coworkers were adopted.^15,51,52 Figure 3B and 3C show temporal changes in band intensities observed in the time-resolved difference UVRR spectra. After dissociation of CO from the heme group, UVRR bands of Tyr decreased in intensity with a time constant of 2 ps. The intensity decrease was followed by intensity recovery with a time constant of 8 ps. The UVRR bands for the Trp residues underwent a decrease in intensity that was completed within the instrument response time. This intensity decrease was followed by an intensity recovery with a time constant of about 50 ps and which lasted up to 1 ns.

Assignments of amino acid residues that exhibit these spectral changes help to clarify primary structural changes in Mb. Sperm whale Mb possesses two Trp residues, Trp7 and Trp14. Spectral changes in the Trp bands were attributed to a decrease in local polarizability around Trp14, based on the comparison of time-resolved UVRR spectra between wild-type Mb and Mb mutants.⁵⁰ The crystal structures of deoxyMb and MbCO showed significant differences in the positions of the E and F helices,³⁶ which hold the heme group, as shown in Figure 3E. In the deoxyMb, backbone segments near His64 in the E helix and His93 in the F helix shifted in the same direction by 0.3 Å compared to that in MbCO. The Trp14 interacts with hydrophobic residues in the E helix. Thus, we concluded that the decrease in local polarizability reflects E helix movements toward the heme that occurred within the instrument response function. The subsequent intensity recovery indicates that the A helix approaches the E helix. Spectral changes in the Tyr bands were attributed to Tyr146, which forms hydrogen bonds with the backbone carbonyl of His93 located in the F helix and with that of Ile99 located in the FG corner (Figure 3D). The out-of-plane motion of iron following CO dissociation, which was observed by time-resolved RR spectroscopy,^37,53 pushes on His93, subsequently inducing protein motion, such as displacement of the F helix. This F helix displacement can affect the strength of hydrogen bonding and hydrophobic interactions between Tyr146 and neighboring residues in the F helix. Therefore, a decrease in intensity of the Y8a band in 2 ps reflects rapid F helix displacement induced by the iron out-of-plane motion. These observations demonstrate that the EF helical section responds very fast in Mb, which is consistent with recent femtosecond X-ray solution scattering studies.^54,55 The recovery of Y8a band intensity subsequent to an intensity decrease indicates that environmental changes near Tyr146 occur through further structural rearrangement in the F helix and/or the FG loop. The ν(Fe–His) frequency and the transient circular dichroism signal of the N band of photodissociated MbCO⁵⁶ showed structural changes in the proximal heme pocket in ∼100 ps. These data demonstrate that the proximal heme pocket undergoes structural rearrangements in 10–100 ps. Recently, a consistent picture was obtained by studies using time-resolved serial femtosecond crystallography⁵⁷ and femtosecond X-ray absorption spectroscopy.⁵⁸

Although experimental information on the ultrafast motion of Mb is available, the specific motions of protein portion that contributed to the early time response were not known, and only qualitative statements about the possible motions could be derived from stationary structures. Our study identified experimentally for the first time the specific motions of the portion of the protein that contribute to the primary protein response. The propagation pathway of structural changes from the heme to the entire protein were discussed in detail in our original paper.⁵⁰ These data on ultrafast protein response were used directly to verify the linear response theory of protein dynamics.⁵⁹ Yang et al. formulated a general linear response theory to address the physical nature of protein conformational changes.⁶⁰ Inspired by our UVRR study, they applied the theory to calculations of the Mb protein response to CO dissociation. Their calculations demonstrated that the theory agreed well with the primary protein changes revealed by our measurements.⁶⁰

3.1.1.4 Coupling of Motions in Heme and the Protein Moiety;

The combination of time-resolved visible and ultraviolet RR studies indicated that reaction forces at the heme become channeled very efficiently into the spatially extended collective motions of the globin. The structural change reflected by the ν(Fe–His) frequency shift is not localized in the heme pocket but accompanies the structural rearrangement at the protein surface. The rate of the frequency shift was decelerated in viscous buffer; the rate was proportional to the power of about −0.3 of the buffer viscosity, as shown in Figure 2F. This behavior can be described by a modified Kramers relation.⁶¹ The dependence of the rate on viscosity suggests the structural rearrangement in ∼100 ps involves not only atomic displacements of the heme pocket but also those at the protein surface, which is slaved to solvent motion.⁶²

We investigated the protein dynamics of an Mb mutant without a covalent linkage between the heme and protein matrix.⁶³ In the mutant used, the proximal His was replaced by glycine, which eliminates the sole covalent connection between the protein moiety and heme and creates a cavity that can be occupied by exogenous ligands such as imidazole. We performed time-resolved RR spectroscopy of the Mb mutant, including imidazole, to see if the frequency shift of the ν(Fe–His) mode was affected by the loss of the covalent connection. Interestingly, even though the covalent connection was abolished, the Fe–Imidazole stretching mode underwent a frequency shift, as shown in Figure 2E; the time constant of the shift (121 ± 20 ps) was close to that of the wild-type Mb, which suggested that the protein response upon ligand dissociation was induced, even though no covalent linkage existed between the heme and protein matrix. This mutant can be viewed as a heme-imidazole complex “dissolved” in the protein matrix; the heme-imidazole complex did not show a frequency shift in the micelle solution, but did show the frequency shift when the complex was surrounded by the protein matrix. Therefore, the covalent bond is not needed to induce structural rearrangements in Mb, although atomic packing around the complex in the protein matrix is essential.

3.1.2 Hemoglobin

Figure 4A shows a three-dimensional structure of human adult Hb, which is a tetrameric heme protein composed of two α- and two β-subunits. Each subunit contains a heme group, which is bound to a His residue (Figure 4B). The quaternary structure of Hb undergoes a reversible transition between the low-affinity (T, or tense) state and the high-affinity (R, or relaxed) state upon partial ligand association/dissociation of the four heme groups. This constitutes the structural basis of the cooperativity and allostery of Hb.³ The completely unligated (deoxy) structure typically adopts the T state, while the fully ligated form adopts the R state. The association/dissociation of the diatomic ligands to/from heme in Hb, which is a highly localized perturbation, initiates a sequence of propagating structural changes in the subunits and intersubunit contacts. Thus, the protein dynamics of the R-T allosteric transition must be elucidated to fully understand the allosteric mechanism of Hb.

3.1.2.1 Differences between Spectral Changes of Hemoglobin and Myoglobin;

The structural dynamics of Hb were compared to those Mb. Figure 4C and 4D show time-resolved RR spectra of photodissociated Hb in the 180 to 850-cm⁻¹ and 1030 to 1700-cm⁻¹ regions, respectively, for delay times of 4 and 1000 ps of the probe pulse with respect to the pump pulse.⁶⁴ The time-resolved RR spectra at 4 ps resembles the spectrum of equilibrium unligated Hb (deoxyHb), indicating that the photodissociated heme has mostly relaxed to the domed structure within the instrument response time (∼2 ps). However, a close comparison of the spectra revealed that the ν₂, ν₃, and ν₄ bands at 1563, 1470, and 1354 cm⁻¹, respectively, of the transient species at 1000 ps were 2–3 cm⁻¹ lower in frequency than those of the deoxyHb (Figure 4C), which is consistent with the results reported by Friedman et al.⁶⁵ and Dasgupta and Spiro⁶⁶ using a nanosecond pulse. These features contrast to those of Mb, in which a barely detectable shift was found between the corresponding species. Differences from Mb also were observed for the ν(Fe–His), γ₇, and ν₈ bands in the time-resolved RR spectra of Hb (Figure 4D). First, the ν₈ band at 341 cm⁻¹ in deoxyHb was absent in the photoproduct. Second, the band position of the ν(Fe–His) and γ₇ modes at 231 and 305 cm⁻¹, respectively, of the photoproduct at 1000 ps differed from that of the deoxy form. The difference between the spectra at 1000 ps and that of deoxyHb showed that Hb adopts a metastable structure within the instrument response time and remains nearly unchanged in the subnanosecond to nanosecond time region. The observed differences in the ν(Fe–His), γ₇ and ν₈ bands between the transient species and the deoxy form provided insight into the structure of the metastable form in the picosecond time region. Heme modes in low frequency regions suggested that the primary metastable form of Hb has a more disordered orientation of propionates and a less strained environment compared to the deoxy form.⁶⁴ The latter discovery was consistent with the experimental observation that the ν(Fe–His) frequency of the metastable form was higher than the deoxy form.

Figure 4E shows temporal changes in the ν(Fe–His) frequency of Hb and Mb following CO dissociation. The ν(Fe–His) mode of Hb exhibited a 2-cm⁻¹ downshift with a time constant of about 284 ± 38 ps, suggesting a structural change in the heme pocket following ligand dissociation. Although the size of the frequency shifts of the ν(Fe–His) band was similar for Hb and Mb, the time constants of the frequency shift of Hb was about three-fold slower than that of Mb. The ∼300-ps process for the Fe–His linkage is the fastest structural response reported for the Fe–His linkage of the Hb photoproduct.

The 2-cm⁻¹ downshift of the ν(Fe–His) band in the picosecond region appears to be characteristic of a globin-folded protein with penta-coordinated structure.^67,68 The frequency shift of the ν(Fe–His) band has been studied for several proteins that exhibit a globin fold. Interestingly, this trend correlates with the coordination structure of the heme in the deoxygenated ferrous form. A frequency shift was observed for proteins that have penta-coordinated structures in the deligated ferrous forms,^37,64,67 while no frequency shifts were observed for proteins that have hexa-coordinated structures in the deoxygenated ferrous forms.⁶⁸

Distinct differences were observed in time-resolved RR spectra of the photolyzed species of Hb and Mb. The transient spectra of Hb showed differences in the ν(Fe–His), γ₇, and ν₈ and bands of the deoxy spectra, while those of Mb did not. To determine whether the faster response in Mb was related to it being monomeric and/or noncooperative, we obtained a similar set of spectra from the isolated subunits of Hb (Figure 4E).⁶⁴ The RR bands of the isolated subunits exhibited similar temporal behavior to those of Hb, but were completely different from those of Mb.⁶⁴ In solution, isolated α and β chains are supposed to form monomers/dimers and tetramers, respectively, and do not have cooperative oxygen-binding interactions. Thus, the difference in the dynamics between Hb and Mb was not due to intersubunit interactions, such as tetramer formation or cooperativity, but to the inherent character of the Hb subunits. Such a phase of tertiary structural change in the nanosecond region may be important for allostery of Hb to bridge the structural changes on a sub-nanosecond scale and the quaternary structural changes in the microsecond region.

A frequency difference in the ν(Fe–His) mode as large as 16 cm⁻¹ was observed between the metastable state at 1000 ps and deoxyHb, suggesting a subsequent shift that occurred later (>nanoseconds). Three subsequent changes in ν(Fe–His) frequency occurred step-wise in the nanosecond to microsecond region.^69–71 These changes were associated with the tertiary and quaternary structure changes, which also were investigated by nanosecond time-resolved UVRR spectroscopy.⁷² Time-resolved UVRR spectroscopy showed that, upon CO dissociation, the subunit interface underwent structural changes in two well-separated steps with time constants of 3 µs and 20 µs.⁷² A time-resolved wide-angle X-ray scattering study on Hb showed that the relative rotation of one αβ dimer with respect to the other occurred at 1–3 µs.^73,74 These studies showed that the changes occurred at the subunit interface, indicating quaternary structural changes, in Hb on a microsecond time scale. However, how the heme of one subunit responds to ligand dissociation of another heme on a neighboring subunit has not been fully elucidated, despite the importance of understanding this inter-subunit communication to clarify the molecular mechanism of Hb cooperativity.

Discriminating between the α and β subunits spectroscopically is key because both subunits have identical heme structures. Our strategy to investigate inter-subunit communication through the α-β subunit interface involves observation of changes in the RR spectra of the Ni-heme of the β subunit upon CO dissociation in the α subunit using carbonmonoxy α₂(Fe)β₂(Ni) [α₂(Fe–CO)β₂(Ni)].⁷⁰ Hybrid Hb α₂(Fe–CO)β₂(Ni) complexes possess two characteristics that make them extremely useful for investigating individual subunit properties in time-resolved experiments. First, resonance enhancement enables selective observation of the Fe-heme and Ni-heme spectra in the hybrid Hb because the peak wavelengths of the Soret band are different for the deoxy Hb and the Ni-substituted Hb. Second, no photochemical reaction occurs in the Ni-heme upon irradiation because it does not bind a ligand. Therefore, the RR spectrum of the Ni-heme is a good spectroscopic probe for investigating inter-subunit communication. We examined the changes in the RR spectra of the Ni-heme of the β subunits upon CO dissociation in the α subunits using α₂(Fe–CO)β₂(Ni).⁷⁰ Concerted changes in the ν(Fe–His) and ν(Ni–His) frequencies of the α and β subunits of ∼20 µs indicated that the proximal tension imposed on the bond between the heme and proximal His strengthened after the quaternary changes. This is first direct observation for the Perutz mechanism for allosteric control of oxygen binding in Hb.⁷⁵ Our time-resolved RR study clarified not only the initial steps but also the very last allosteric transition of Hb upon CO dissociation.

3.1.2.2 Differences between Protein Dynamics upon Dissociation of O₂ and CO;

Thus far, the time-resolved RR studies of Hb and Mb were discussed upon CO dissociation. Along with the studies of Gibson,⁷⁶ numerous kinetic studies of heme proteins by ligand photolysis techniques have been reported.^{32,33,77–79} However, time-resolved RR studies on Hb were conducted mainly for dynamic investigations related to the CO dissociation. Very few studies on structural dynamics upon O₂ dissociation have been reported,^80,81 even though the physiological ligand of Hb is O₂. Time-resolved measurements following O₂ dissociation are rarely reported for several reasons. First, the quantum yield of O₂ photolysis in heme proteins is low.³⁴ Second, in the presence of O₂, the heme iron is gradually autoxidized. Thus, a fraction of the sample becomes inactive for ligand binding during measurement, which makes acquiring data over a long period and getting a good signal-to-noise ratio with a limited amount of the protein sample difficult. Third, CO usually is assumed to be a perfect analog for O₂. However, the validity of this assumption has not been examined. Studies of protein dynamics related to O₂ dissociation are essential for understanding the allostery of Hb. The high stability and sensitivity of our time-resolved RR spectrometers prompted time-resolved measurements of protein dynamics of Hb upon O₂ dissociation. Striking differences were found between the protein dynamics of Hb upon O₂ and CO dissociation, providing evidence that CO is not a perfect analog for O₂ in Hb.⁸² The time-resolved RR spectra of both photoproducts at 1-ns delay differed from that of the deoxy form at frequencies of the ν(Fe–His) and γ₇ modes, and at the intensity of the ν₈ modes. Spectral changes of the O₂ photoproduct in the submicrosecond region were faster than those of the CO photoproduct, indicating that the structural dynamics of the submicrosecond metastable state is ligand-dependent for Hb. In contrast, no ligand dependence of the dynamics was observed for Mb. As discussed in Section 3.1.1.1, the structure of the heme in Mb changes to one closely resembling the deoxy form within a few picoseconds.³⁷ These experimental data indicate that the structural dynamics of the heme occurring later than a nanosecond are specific to Hb, are ligand-dependent, and are associated with allostery of Hb. A recent finding showed that the structural dynamics of the heme later than a nanosecond also are dependent on the initial quaternary structure.⁸³ These results strongly suggest that the structural changes of the metastable state are essential to the Hb allostery.

Previous time-resolved absorption⁸⁴ and RR^64,72,85 spectroscopic studies on Hb have established a series of intermediates following CO dissociation. Figure 5 shows structural dynamics of Hb following ligand dissociation as well as that of Mb. In the primary intermediate, the heme structure changes to a metastable form within 2 ps. The heme pocket undergoes rearrangement to generate the B intermediate. Protein relaxation leads to the subsequent intermediate, R_deoxy. In the transition from B to R_deoxy, the distal E helix displaces toward the heme plane, probably driven by motion of the proximal F helix in response to Fe–His bond relaxation.^72,85,86 Therefore, the spectral changes of the γ₇ and ν₈ bands in nanoseconds are likely associated with EF helical motion and this motion is ligand-dependent in Hb. A remarkable difference in the interactions of the heme-bound ligand and the E helix was found between the O₂ and CO-bound forms of Hb. In the oxy form, the bound O₂ is hydrogen-bonded to the distal histidine, which is in the E helix. In contrast, no such interaction in the CO-bound form exists between the bound CO and the E helix. The difference in the interaction would cause a change in the structure of the EF helical section. This explains the ligand dependence of the protein dynamics of Hb.

Both CO and O₂ undergo geminate recombination at room temperature; a fraction of the docked ligand diffuses into the heme pocket and rebinds. The time constants of geminate rebinding for O₂ and CO reported for Hb are 1.5 ns⁸⁷ and 70 ns,⁸⁸ respectively. Interestingly, these rebinding rates and the rates of the B-R_deoxy transition observed in the visible RR spectra are of the same order of magnitude. This similarity can be explained as follows. If intermediates B and R_deoxy have slow and fast intrinsic rebinding rates, respectively, the apparent geminate rebinding rate is determined by the rate of the transition from B to R_deoxy. In R_deoxy, the E-helix is displaced toward the heme plane. The displacement reduces the volume accessible for the ligand to diffuse and hence accelerates geminate rebinding in R_deoxy. Therefore, it is likely that the structural changes in the B-R_deoxy transition are associated with the ligand rebinding process.

Based on these observations, the following hypothesis explains the functional role of the observed structural dynamics. The R structure, which is the high-oxygen-affinity structure, dominates in the fully ligated state, while the T structure, the low-oxygen-affinity structure, dominates in the deoxy state. The R-T transition occurs in doubly or singly ligated states. To change the quaternary structure from R to T, Hb needs to release at least two O₂ molecules. In red blood cells, ligands dissociate from Hb through thermal reactions. After the first ligand dissociates, the second ligand needs to dissociate before the dissociated ligand rebinds to undergo the R-T transition. The presence of the slow-binding intermediate allows the second ligand to dissociate before the dissociated ligand rebinds. Therefore, the submicrosecond phase specific to Hb may be important for ensuring that Hb has a high R-T transition efficiency.

Our data indicated that the structural evolution of the O₂-dissociated species was not same as that of the CO-dissociated species. Even though CO is not a physiological ligand, the studies using CO dissociation have significantly contributed to the understanding of the allostery of Hb. However, our results demonstrated that time-resolved studies on O₂ dissociation are indispensable to understand the R-T allosteric transition even though they are experimentally difficult.

3.1.3 Heme-Based Oxygen Gas Sensor Proteins

A comparison between structural changes that occur upon O₂ and CO dissociation is essential to elucidate the sensing mechanism of heme-based oxygen gas sensors as well as the allosteric mechanism of Hb. Oxygen gas sensors are thought to discriminate O₂ from other gas ligands for O₂ detection.⁸⁹ Therefore, the structural dynamics for O₂ must be different from those for CO dissociation. Structural changes upon CO dissociation are observed because structures of the CO-bound and deligated protein are different. However, such structural changes are not necessarily associated with the sensing mechanism, i.e., all that glitters is not gold. Encouraged by the success of the time-resolved RR study on Hb dynamics, we tried to obtain time-resolved RR spectra of the heme-based gas sensors upon O₂ dissociation. We succeeded in detecting ligand-dependent protein dynamics, which allowed discussion of signal transduction and ligand discrimination. We identified the structural changes essential to the signal transaction and ligand discrimination for SmFixL,^90–92 an expression regulator of genes associated with nitrogen fixation of Sinorhizobium meliloti, and HemAT-Bs,⁹³ a chemotactic signal transducer of Bacillus subtilis, by comparing the structural changes due to the dissociation of O₂ and CO. These studies clearly showed that CO is not a perfect analog for O₂ in studies of the heme-based oxygen gas sensors.

3.2 Dynamics of Photoreceptor Proteins (Protein Dynamics Induced by Photoisomerization)

Photoreceptor proteins are particularly suitable for time-resolved studies on protein dynamics because their reactions can be triggered by an optical pulse, allowing study of functionally important dynamics over a wide span of timescales. Their functions include light-driven ion pumping, light sensing, and others. Large conformational changes are expected to be involved in their reactions. Although photointermediates of photoreceptor proteins have been identified by transient absorption spectroscopy, conformational changes in the protein moiety have been less understood. We elucidated the primary protein response to the photoreaction of a protein chromophore using time-resolved UVRR spectroscopy.

3.2.1 Rhodopsins

Microbial rhodopsins are representative photoreceptor proteins. Figure 6A shows the crystallographic structure of bacteriorhodopsin (BR), which is the most widely studied microbial rhodopsin. Microbial rhodopsins consist of seven transmembrane α helices. A retinal chromophore is covalently bound to a lysine residue through a protonated Schiff base linkage. The all-trans configuration of the retinal chromophore is the most thermodynamically stable. Absorption of a photon by the chromophore causes isomerization from the all-trans to the 13-cis configuration, as shown in Figure 6B. The photoisomerization of the chromophore takes place in subpicoseconds,^94–97 which induces sequential changes in the protein structure that are important for protein functions. To examine protein responses to the chromophore isomerization, time-resolved UVRR spectra of five types of microbial rhodopsins, BR,⁹⁸ halorhodopsin (HR),⁹⁹ sensory rhodopsin I (SRI),¹⁰⁰ sensory rhodopsin II (SRII),¹⁰¹ and Anabaena sensory rhodopsin (ASR),¹⁰² were obtained. In this subsection, I summarize our recent studies on the primary protein response of the microbial rhodopsins associated with photoreactions of the chromophore.

Bacteriorhodopsin, found in the purple membrane of halobacteria, functions as a light-driven proton pump. The UVRR spectrum of BR in the unphotolyzed state probed at 225 nm is shown in the top trace of Figure 6C. This spectrum contains all Raman bands of eight Trp and eleven Tyr residues in BR. The 1615-cm⁻¹ band is attributed to the overlap of the W1 and Y8a bands. The bands at 1555, 1357, 1013, and 763 cm⁻¹ are assigned to the vibrational modes of Trp, W3, W7, W16, and W18, respectively. The other spectra in Figure 6C represent time-resolved UVRR difference spectra, obtained by subtracting the unphotolyzed BR spectrum from the spectrum measured at each delay time from −5 to 1000 ps. Upon photoexcitation, negative UVRR bands were clearly observed for Trp within the instrument response time. These negative bands indicate depletion of Raman intensity due to a change in the protein structure during the photoreaction. They decayed from 10 to 50 ps, indicating that the band intensity of Trp recovered due to structural changes subsequent to the instantaneous change. In the region from 100 to 1000 ps, the difference spectra did not change.

Figure 6D shows the temporal intensity changes of the W3 band observed for BR. Data analysis revealed that the intensity of the band instantaneously decreased within the instrumental response time and recovered with a time constant of approximately 30 ps. The retinal chromophore in BR isomerizes from the all-trans to 13-cis form within ∼0.5 ps, to produce the primary photoproduct.^94,95 The initial UVRR intensity depletion was attributed to the protein response to the chromophore isomerization. The subsequent intensity recovery of 30 ps is likely to reflect the protein response to isomerization of the chromophore.⁹⁸

The observed process can be attributed to the structural rearrangement of the protein moiety in the vicinity of the retinal chromophore. The primary protein response to the chromophore isomerization can be understood in terms of the spectral changes in the structural marker bands in the UVRR difference spectra. Negative bands appeared at the W3, W16, and W18 band positions in 1–100 ps (Figure 6C). These negative bands indicate that the intensities of the W3, W16, and W18 bands in the intermediate are smaller than those in the unphotolyzed state. The Trp bands, except for the W1 band, were enhanced via a Franck-Condon mechanism through resonance with the B_a and B_b electronic transitions.^103,104 The probe wavelength nearly matched the peak wavelength of the Raman excitation profile of Trp due to the B_a and B_b electronic transitions (224 nm).¹⁰⁵ The absorption band of the B_b transition was blue-shifted when the solvent polarizability of the Trp solution decreased.^106,107 This shift results in decrease of resonance enhancement. Therefore, the intensity loss of the W3, W16, and W18 modes is associated with a reduction in local polarizability around the Trp residue(s) in 30 ps.

The temporal changes in Trp band intensities among the five microbial rhodopsins described above are compared in Figure 6D–6H. Figure 6E, 6F, 6G, 6H, and 6I show temporal intensity changes of the W3 band observed for HR, SRI, SRII, dark-adapted ASR, and light-adapted ASR, respectively. The five rhodopsins are classified as light-driven ion pumps (BR and HR) and photosensors (SRI, SRII, and ASR). The intensities of the Trp band in these rhodopsins bleached upon photoisomerization of retinal within the instrument response time (∼3 ps) and recovered in tens of picoseconds. In the early picosecond time frame, the structural change in the protein moiety likely is localized around the retinal chromophore in the rhodopsins. Three conserved Trp residues are located in the retinal binding pocket. Two residues (Trp86 and Trp182 in BR) sandwich the polyene chain of retinal. Another Trp (Trp189 in BR) is positioned near the β-ionone ring. Due to selective Raman enhancement, the Raman bands of the Trp residues function as good probes to examine structural changes around the chromophore in microbial rhodopsins. The comparison shows that the rates of rearrangement of the protein moiety were insensitive to function and ion binding (for HR and SRI). Moreover, the protein response upon the cis-to-trans isomerization and the trans-to-cis isomerization was investigated for ASR because ASR exhibits photochromic photoreactions. Based on the temporal intensity changes of dark-adapted and light-adapted ASR, shown in Figure 6H and 6I, temporal intensity changes of ASR upon trans-to-cis isomerization and cis-to-trans isomerization were calculated (Figure 6I). Rearrangement rates were insensitive to the direction of ASR isomerization. These results suggest that the primary structural response of the protein moiety to the chromophore isomerization is very similar in microbial rhodopsins.

A similar primary structural response of a protein moiety to chromophore isomerization was observed for the five microbial rhodopsins. Two pathways were proposed for the propagation of structural changes from the chromophore to the protein moiety. One is a pathway through a hydrogen-bonding network including the protonated Schiff base. The orientation of the hydrogen bond of the protonated Schiff base changes upon photoisomerization and can quickly perturb the protein moiety. The other path involves the van der Waals contacts between retinal and the surrounding amino acid residues. In the time-resolved visible RR spectra, an intense Raman band due to the hydrogen-out-of-plane (HOOP) wagging mode of retinal was observed in the primary intermediate of BR.⁹⁴ The HOOP band gains its intensity when the polyene chain is distorted.¹⁰⁸ In fact, no intense HOOP band was observed for polyenes in solution because the polyene quickly adopts the stable planar form following photoisomerization in solution. A distorted polyene chain with a measurable lifetime in the protein would be due to a highly packed structure around the retinal. The propagation of structural changes through the van der Waals contacts is possible in such a highly packed environment around the chromophore.

3.2.2 Photoactive Yellow Protein

Photoactive yellow protein (PYP) is a recognized blue-light sensor that controls the phototaxis of the bacterium, Halorhodospira halophila.¹⁰⁹ The chromophore of PYP is p-coumaric acid (pCA), which is buried in a hydrophobic pocket, as shown in Figure 7A. After pCA in the ground pG state is photoexcited to the pG* state, trans-to-cis isomerization leads to a cyclic reaction involving a series of intermediate states.^110,111 In the photocycle, rearrangement of the hydrogen-bond network around pCA plays a crucial role in facilitating signal transduction. Figure 7B depicts structure of pCA and residues close to the phenolic oxygen. In the pG state, short hydrogen bonds (SHBs) form between pCA and Glu46, as well as between pCA and Tyr42.¹¹² High-resolution neutron crystallographic analysis revealed a difference between the two SHBs.¹¹³ The hydrogen atom between pCA and Glu46 locates between the phenolic oxygen of pCA and the carboxylate oxygen of Glu46 and does not form a covalent bond to either oxygen atom. In contrast, the hydrogen atom between pCA and Tyr42 forms a covalent bond to the phenolic oxygen of Tyr42. Thus, the SHB between pCA and Glu46 is categorized to a low-barrier hydrogen bond (LBHB), whereas the SHB between pCA and Tyr42 is not a LBHB. Using time-resolved UVRR spectroscopy, we investigated changes in the hydrogen-bond network around pCA.^25,114

Figure 7C shows the UVRR spectrum of dark-state PYP and transient UVRR difference spectrum of PYP measured at 3 ps. The difference spectrum of PYP measured at 3 ps shows that UVRR bands attributable to both the Tyr and Trp residues produce spectral changes caused by photoinduced structural changes. In this spectrum, negative Tyr Y8a, Y7a, and Y9a bands were observed at 1177, 1224, and 1624 cm⁻¹, respectively. The negative bands represent intensity loss relative to the intensity in the dark state. The intensity depletion of the Tyr bands arises from the blue shift of the excitation profile. An increase in hydrogen-bond strength gives rise to a blue shift in the excitation profiles of Tyr UVRR bands.¹⁰⁶ Therefore, the intensity loss of the Tyr bands suggests that the hydrogen bond formed by the Tyr residue is strengthened at 3 ps. The Y9a band exhibited a sigmoidal form attributable to an upshift in the Y9a band upon photoreaction. The frequency of the Y9a band indicates hydrogen-bond strength,^15,115 and thereby its upshift supports our interpretation that intensity loss in the Tyr bands results from the stronger hydrogen-bond formation. For Trp bands, a small decrease in intensity was observed for the W3 and W7 bands. This intensity loss was due to the blue shift in the Raman excitation profile of the Trp residue, which was attributed to a reduction in hydrophobicity in the region surrounding the residue.^106,107

The PYP contains five Tyr residues: Tyr42, Tyr76, Tyr94, Tyr98, and Tyr118. Tyr42 is the most probable candidate for causing the spectral change in the wild-type spectra, because its phenolic OH group is directly hydrogen-bonded to pCA. To identify the Tyr residue(s) responsible for the spectral change, we compared time-resolved difference spectra of the Y42F mutant to those of the wild-type.¹¹⁴ In the time-resolved difference spectra of Y42F, the negative bands of the Tyr vibration disappeared, indicating that the remaining four Tyr residues in Y42F mutant undergo little structural and/or environmental changes and that Tyr42 is solely responsible for the spectral changes upon photoexcitation of pCA. Thus, we succeeded in detecting the spectral change of a single Tyr residue in PYP and concluded that the hydrogen bond between pCA and Tyr42 is strengthened in the pG* state.

After discovering that Tyr42 is solely responsible for the changes of the Tyr bands upon photoexcitation of pCA, we next studied how the hydrogen bonds of pCA to Tyr42 and Glu46 affect each other. The UVRR intensities of the Y8a, Y8b, and Y9a bands for Tyr42 in wild-type PYP were stronger by 20 ± 5%, than those in the E46Q mutant, showing that the hydrogen bond between pCA and Tyr42 is strengthened by substituting Glu46 for Gln in the pG state.¹¹⁴ In contrast, in the pG* state, intensities of the bands for Tyr42 in wild-type PYP were similar to those in the E46Q mutant, indicating that the hydrogen-bond network around pCA of the wild-type is similar to that of E46Q mutant. This study demonstrated that the hydrogen bond between pCA and Tyr42 and between pCA and Glu46 were associated in the hydrogen-bond network.

Dynamics of the hydrogen-bond network are of interest in the allostery of PYP. To elucidate picosecond structural dynamics, the temporal intensity change of the Y8a band was examined (Figure 7D). Results showed that the intensity of the Y8a band decreased within the instrument response time, partially recovered with single exponential kinetics with a time constant of 8 ps, and maintained a constant negative value up to 1 ns. The photoisomerization instantaneously perturbed the protein structure around pCA, followed by structural rearrangement of the protein moiety and an increase in hydrogen-bond strength between pCA and Tyr42 in 8 ps.

Quantum chemical calculations revealed that hydrogen bonds of the phenolic oxygen of chromophore to Tyr42 and Glu46 had a strong effect on the low-lying excited states of the chromophore.¹¹⁶ The hydrogen bonds produced pronounced energetic shifts of the electronic state. Therefore, the observed change in the hydrogen bonds of the chromophore upon pG* formation likely affects a potential energy surface of the electronic excited state and promoted photoisomerization of the chromophore. The E46Q mutant had an isomerization rate similar to that of the wild-type, whereas the Y42F and E46A mutants had slower isomerization rates than wild-type. These results indicate that elimination of one of the two hydrogen bonds affects isomerization rate. The change in the hydrogen bonds of the chromophore upon the pG* formation was thought to promote efficient isomerization of the chromophore.

Clarification of the mechanism by which external stimuli effect site-specific structural changes is essential for understanding protein functions. To accomplish this, a measurement technique with high time resolution and site-selectivity is necessary. Time-resolved UVRR spectroscopy can probe the structural dynamics of specific sites in a protein structure by selectively enhancing the vibrational Raman bands assignable to aromatic amino acid side chains as well as to polypeptide bonds with picosecond time resolution. We demonstrated the ability of time-resolved UVRR spectroscopy to observe the primary protein response to chromophore photoreactions in photoreceptor proteins.

4. Energy Flow in Proteins

Macroscopically, heat dissipation is described by the law of heat conduction, Fourier’s law.¹¹⁷ Our research related to this was prompted by the simple question: what does heat dissipation look like when the spatial scale of the system is reduced to atomic size. On an atomic spatial scale, approximation of the continuum medium under Fourier’s law cannot be represented. Both the time and space scale of molecules are too small to establish the concept of heat. Strictly speaking, the temperature of a molecular system is defined only when the entire system is in thermodynamic equilibrium. However, the transiently “hot” molecules discussed in this section are not in thermodynamic equilibrium. In a first approximation, immediately after the nonradiative transition, the distribution can be described as microcanonical. All molecules have the same energy of a few tens of thousands cm⁻¹ within a small uncertainty. Each fundamental mode with a frequency less than 4000 cm⁻¹ may be considered as a small subsystem of vibrational quantum states in the molecular system. The energy of the vibrational state of the subsystem is much smaller than the total energy. In this case, a good approximation of the occupation probability of the subsystem is provided by N(ν_i) ∝ exp(−hν_i/k_BT),^118,119 where ν_i is vibrational frequency. Thus, the subsystem behaves as if it had a higher temperature in canonical ensemble, which is called the internal temperature of the excited molecules.

Energy is the origin of many fascinating dynamics in nature. Thus, the energy flow in a condensed phase is fundamentally important for chemical dynamics.^120–124 Providing a microscopic picture for energy dissipation on an atomic scale is very challenging. Temporally resolved mapping of energy flow on the molecular scale is a prerequisite for understanding the microscopic mechanism of energy flow. Our approach to studying energy flow was to excite a specific group and then monitor energy redistribution in the surroundings. The energy flow can be studied by observing both the energy decay of the initially excited group and the energy increase in the accepting groups. Using this approach, the energy flow pathways in a condensed phase can be elucidated. To observe the energy dissipation process on an atomic space scale, a heat source molecule (heater molecule) and a heat probe molecule (probe molecule), with a well-defined intermolecular distance and relative orientation are needed. However, in solution, controlling the distance and relative orientation between a pair of molecules is difficult. We solved this problem by using heme proteins as the observation system.

Heme proteins are ideal molecules for a system to study energy flow because the heme group exhibits ultrafast internal conversion (within 100 fs),¹²⁵ and thus excess vibrational energy can be deposited locally at the heme site immediately after photoexcitation. Excess energy as great as 25000 cm⁻¹ can be deposited into the heme by photoexcitation via the Soret transition. Thus, heme acts as a very efficient heat convertor. To observe subsequent energy relaxation processes, we adopted anti-Stokes UVRR spectroscopy.^126–128 Because of the resonance effect, visible and ultraviolet RR spectroscopy selectively probes Raman bands of prosthetic groups and aromatic amino acid residues, respectively, allowing site-selective detection of energy at the level of a single amino acid residue in a large protein molecule. In addition, time-resolved anti-Stokes Raman spectroscopy is selective for vibrationally excited populations and is suitable for studying vibrational energy. For example, internal temperature can be obtained from the intensities and cross sections of Stokes and anti-Stokes scattering.^34,129,130 In heme proteins, the distance and relative orientation between heme and amino acid residues are well characterized based on X-ray crystallographic data, and the distance between heme and the amino acid residues can be as long as 20 Å. The protein moiety of a heme protein is a “quasi-solvent,” with a well characterized three-dimensional structure, and can be modified by site-directed mutagenesis. Therefore, studies using this technique based on heme proteins can provide new insights for understanding the mechanism of vibrational energy transfer in condensed phases.

4.1 Energy Release from Heme

First, energy release from the heme group in protein was examined.⁷⁷ The anti-Stokes time-resolved RR difference spectra of heme between −5 and 50 ps are shown in Figure 8A. Anti-Stokes intensities were greatest at a 1-ps delay, when the ν₃, ν₄, ν₅, ν₆, and ν₇ bands were observed at 1469, 1353, 1118, 783, and 669 cm⁻¹, respectively. These bands lost intensity as delay time increased and reached equilibrium intensities at a 10-ps delay. Figure 8B shows temporal changes in the anti-Stokes ν₄ and ν₇ band intensities. Anti-Stokes intensity developed within the instrument response time. Deconvolution of instrument response from the decay of anti-Stokes intensity using a Gaussian fit to the cross-correlation signal gave decay constants of 1.1 ± 0.6 ps for the ν₄ band and 1.9 ± 0.6 ps for the ν₇ band. Because no intensity change occurred in the Stokes ν₄ and ν₇ bands in the 3 to 50-ps time range (Figure 2D), the observed intensity decay of the anti-Stokes ν₄ and ν₇ bands can be ascribed to vibrational energy relaxation. Thus, our data demonstrated that heme acts as a heater molecule, and the report on energy relaxation activated many theoretical research groups and inspired theoretical research on energy relaxation in proteins.^131–139

4.2 Energy Dissipation in Proteins

Next, we observed excess energy diffusion in the protein moiety by selectively observing the anti-Stokes resonance Raman spectra of Trp residues in the protein. In a “proof-of-principle” report, we accomplished direct observation of vibrational energy flow in cytochrome c and demonstrated that time-resolved anti-Stokes UVRR Raman spectroscopy is a powerful tool for monitoring vibrational energy flow in proteins.¹²⁶ Cytochrome c was chosen because it has a single tryptophan residue (Trp59 in bovine cytochrome c) near the heme group, as shown in Figure 9A. The ferric form of cytochrome c is photo-inert, and so effective deposition of excess energy is possible through photoexcitation of the heme unit. Time-resolved anti-Stokes UVRR spectra of Trp59 revealed the kinetics of energy flow from the heme group and energy release of the residue in cytochrome c.

Figure 9B shows the time-resolved anti-Stokes UVRR difference spectra obtained with pump and probe pulses with wavelengths of 405 and 230 nm, respectively. The top trace is a probe-without-photolysis spectrum, which represents the anti-Stokes UVRR spectrum for ferric cytochrome c, and contains the UVRR bands for Trp at 759 (W18), 876 (W17), and 1015 cm⁻¹ (W16), and a band for Tyr at 1181 cm⁻¹ (Y9a). The band at 1599 cm⁻¹ was not assigned to any modes of aromatic amino acid residues. The band at 983 cm⁻¹ indicated by the asterisk was due to the sulfate ion added as an internal standard for determining Raman intensity. Pump-induced difference spectra were obtained for cytochrome c (Figure 9B). The 405-nm pump pulse excited the heme into an electronic excited state. The spectrum at 5 ps shows pump-induced positive difference features observed for the Trp and Tyr anti-Stokes Raman bands, which disappeared within 30 ps in the anti-Stokes UVRR difference spectra. The temporal behavior of the integrated intensity of the W18 and W16 bands in the time-resolved difference spectra relative to those in the probe-without-photolysis spectrum are shown in Figure 9B and 9C, respectively. Intensity changes in the W16 and W18 bands were fitted by a convolution of the instrument response with an exponential rise and an exponential decay, I₁[exp(−t/τ_decay) − exp(−t/τ_rise)]. Time constants of rise and decay for the anti-Stokes W16 band were 5.6₇ ± 3.0 and 5.6₈ ± 3.0 ps, respectively. For the anti-Stokes W18 band, the time constants of 5.5 ± 2.5 and 5.6 ± 2.5 ps were obtained for the rise and decay, respectively. Upon the photoexcitation, Stokes W16 and W18 bands also changed their intensities, presumably due to a hydrogen bond between heme and Trp59, indicating that the Raman cross sections were temporally changed. By taking into account the changes of Raman cross sections, rates for rise and decay of the vibrationally excited populations for Trp59 to be 1–3 and ∼8 ps, respectively, were obtained. The data demonstrated that our technique is powerful for studying vibrational energy exchange in proteins. In addition, our technique demonstrated that the kinetics of energy flow in the protein moiety are independent of the amount of excess energy.¹²⁶

Energy dissipation from the protein to water (as the solvent) has been studied in femtosecond time-resolved infrared¹⁴⁰ and transient phase grating studies^141,142 that monitored the heating of water caused by photoexcitation of protein. These studies showed that the excess energy was transferred to the water interface through the protein matrix in less than 20 ps. The time constant of energy release from Trp59 was smaller than that of water heating. This result is consistent with the supposition that the energy released from Trp59 is transferred to the solvent water through the remaining protein moiety. Direct energy transfer from Trp59 to the solvent water was very unlikely because Trp59 was buried inside the protein and was not exposed to the solvent.

We demonstrated that time-resolved anti-Stokes UVRR spectroscopy is a powerful tool for monitoring vibrational energy flow in a protein. This technique was further improved by combining it with site-directed mutagenesis.¹²⁷ The position of a Trp residue in a protein can be changed by amino acid substitution. Thus, by comparing data from mutants with different distances from the Trp residue to the heme, mapping energy flow in a protein is possible by moving the position of the probe residue. Thus, heme proteins provide significant advantages for studying energy flow in the condensed phase, as opposed to the solution phase because the distance and relative orientation of the heater (heme) and probe groups (Trp residue) can be fixed in proteins but not in solution (where the two molecules would diffuse freely). Wild-type sperm whale Mb has two Trp residues, Trp7 and Trp14. First, mutants devoid of Trp residues were prepared by replacing Trp7 and Trp14 with Tyr and Phe, respectively.¹⁴³ Then, we prepared three mutants with a single Trp at a specific position. Figure 10A shows the Mb mutants prepared. One mutant had a Trp residue at position 68, which is in the vicinity of the heme. The center-to-center distance between the heme and Trp68 is 6.8 Å (PDB ID, 2OH9). Another mutant has a Trp residue at position 28, with a distance to the heme of 12.4 Å (PDB ID, 2OH8). In the third mutant, Trp7 was replaced with Tyr while Trp14 remained unaltered; the distance from the heme to Trp14 is 15.0 Å. The Trp residues in the three mutants were located in a similar direction from the heme, but their distances from the heme differed significantly. A comparison of the data using these mutants allowed examination of distance dependence of energy flow from the heme to the Trp residues. Crystallographic data of these mutants¹⁴⁴ showed that the three-dimensional structures of the mutants are very close to that of wild-type Mb.

Figure 10B and 10C depict the temporal evolution of the anti-Stokes intensity of the W18 and W16 bands in the time-resolved difference spectra relative to that in the probe-without-photolysis spectrum. The anti-Stokes intensities of the mutants reflected the amount of energy delivered to the Trp residues at different positions. Data in Figure 10B show that anti-Stokes intensities decreased as heme–Trp distance increased. This can be explained by the classical thermal diffusion model showing that excess energy becomes spatially less dense as the energy diffuses in the protein. Changes in the anti-Stokes intensity bands were fitted by a convolution of instrument response with an exponential rise and exponential decay for the Trp68 and Trp29 mutants. Time constants of the rise of the W18 band were 3.0 ± 0.4 and 4.0 ± 0.6 ps for the Trp68 and Trp28 mutant, respectively. For the W16 band, time constants of 3.0 ± 0.7 and 4.9 ± 1.1 ps for the rise were obtained for the Trp68 and Trp28 mutant, respectively. Thus, the intensity rise for Trp28 was consistently slower than that for Trp68, indicating that excess energy takes longer to arrive at position 28 (Trp28, 12.4 Å) than at position 68 (Trp68, 6.8 Å). Thus, the observed distance dependence is qualitatively consistent with classical thermal diffusion. This study is the first example of the direct observation of energy dissipation in a protein moiety with the spatial resolution of an amino acid residue.

Next, these observational data were compared with quantitative data calculated based on the model by Li and Champion. They presented a classical two-boundary thermal transport model to simulate the thermal dynamics of transient cooling in chromophoric biomolecules.¹⁴⁵ We calculated the temperature distribution in each protein based on the two-boundary heat transport model. Figure 10E shows calculated temperature as a function of time and position of the Trp residue. The temperature dependence of the anti-Stokes Raman intensity is expressed by the Boltzmann factor. Based on the calculated temperature, the Boltzmann factor for the W18 and W16 modes was determined to investigate the temporal evolution of the anti-Stokes intensities. The broken lines in Figure 10B and 10C represent temporal profiles of the Boltzmann factor based on the model. The model reproduced the temporal behavior of the anti-Stokes intensities of the W16 and W18 bands for both the Trp68 and Trp28 mutants, but did not reproduce temporal behavior of the Trp14 mutant. Thus, the classical heat transport model is unable to reproduce the entire data set, suggesting that a more realistic model at the molecular level is necessary for describing energy flow in proteins. These experimental data validate Leitner’s statement that energy flow is intrinsically anisotropic due to the geometry of proteins.¹⁴⁶

Energy transfer to amino acid residues contacting the heme group was also investigated. We prepared three Mb mutants, F43W, V68W, and L89W. The Trp residue at position 43, 68, or 89 in each mutant is located close to the heme, represented by space-filling purple spheres in Figure 10A. The distance from the heme to the Trp43, Trp68, or Trp89 residue in each mutant was 6–7 Å. The anti-Stokes W18 band intensities of the Trp43, Trp68, and Trp89 residues in the time-resolved difference spectra of each mutant were plotted against delay time (Figure 10E), to compare their temporal behaviors. The vertical axis represents pump-induced change in the W18 band intensity relative to that in the probe-without-photolysis spectrum. For all mutants, band intensity increased up to 6–8 ps, and then decayed to almost zero within 50 ps. Temporal changes at the three Trp positions were similar, implying that vibrational energy was transferred from the heme to the three positions (43, 68, and 89) with similar rates.

The distances between heme and Trp43, Trp68, and Trp89, along the main chain of the protein are very different. If excess energy is transferred to Trp residues through the main chain, build-up times of the vibrationally excited population would be completely different among the three positions, but that was not observed. Nonbonded contacts of Trp43, Trp68, and Trp89 with the heme were evident from the X-ray crystallographic data (Figure 10A).^144,147 The similar rates of energy transfer from heme to the Trp residues suggest that the energy is transferred not through the heme–His93 covalent linkage and the protein main chain, but through atomic contacts between the heme group and the residues.

Amplitudes of band intensity changes (Figure 10E) were distinctly different among the three Trp positions: amplitude was largest in the order Trp68 > Trp43 > Trp89. These observed differences suggest that energy flow at the three positions is also different. Two possibilities exist for the origin of this difference: the rate of energy flow into or out of each Trp residue is different. A careful examination of the X-ray crystallographic structures of the mutants revealed that amplitude increased as Trp residue exposure to the solvent water decreased, which was the dominant influence.¹²⁸ X-ray crystallographic structures of the mutants showed that Trp68 and Trp43 were buried inside the protein, while Trp89 was near the protein surface, facing the solvent.^144,147 Amplitudes of the anti-Stokes band intensity changes (Figure 10E) were inversely correlated with values for solvent accessible surface area (SASA): the larger the SASA value, the smaller the amplitude. This correlation was explained by assuming that the solvent water efficiently accepts the energy of the Trp residues, an assumption consistent with our results for the role of propionate groups in vibrational energy relaxation.^148,149 Computational studies indicated that the propionate groups of heme function as an efficient channel for energy transfer from heme to solvent water.^133,138 Inspired by these studies, the vibrational energy relaxation of heme lacking propionate group(s) were investigated, with results demonstrating that the rate of heme energy relaxation is decelerated upon removal of the propionate groups.^148,149 Therefore, solvent water is an efficient acceptor of vibrational energy. Accordingly, the intensity difference observed among the three positions showed that energy flow out of the Trp residues dominates. A detailed analysis identified a good correlation between SASA and energy relaxation rate from the Trp residue, shown in Figure 10F.

This study established the importance of atomic contacts in vibrational energy flow in proteins. The dominant channel for energy flow from the heme group to the protein moiety is not through the covalent linkage of heme–His93 and the protein main chain, but instead through atomic contacts between the heme and residues. We propose that atomic contact between an amino acid residue and solvent water is an important channel for energy dissipation in proteins.

The technique of energy mapping based on the anti-Stokes UVRR intensity of Trp residues fully utilizes two unique characteristics of Raman spectroscopy, site-specific observation of the resonance Raman effect and selective observation of the vibrationally excited population through anti-Stokes scattering. We further developed this technique by combining it with site-directed mutagenesis, which allows the long-anticipated ability to map energy flow in a protein with a spatial resolution of a single amino acid residue. Systematic application of our general methodology to proteins with different structural motifs may significantly improve understanding of energy flow in proteins.

5. Deepened Insights into Protein Dynamics, Energetics, and Architecture

5.1 High Packing Density of Protein Structures: Functional Compactness

In allosteric regulation, a change at one site in the protein structure induces conformational changes, which alter structure and function at a distant site, much like dominos toppling in long lines. For example, ligand association/dissociation at an allosteric site can alter enzymatic activity or binding affinity in a distal region. This site-to-site communication is of great interest to molecular science to understand the nature of protein dynamics and regulatory mechanisms.

A precise balance between flexibility and stability are required for allosteric communication. To regulate protein function, proteins must be both stable enough to retain their native three-dimensional structures and flexible enough to switch conformational states through external perturbations such as molecule binding or photoreactions in prosthetic groups. Understanding of allostery is advanced by determining how protein motions on different timescales relate to each other and contribute to this balance.

Packing density of the interior of a native protein is high.¹⁵⁰ For a given protein, the polypeptide chain is folded compactly into a characteristic three-dimensional structure in aqueous solution. Although the structures are complicated, the packing density of the protein is nearly maximized, indicated by isothermal compressibilities of proteins that are one order of magnitude smaller than those of organic liquids.¹⁵¹ The maximized packing density prompted the hypothesis that not only covalent and hydrogen bonds but also nonbonded contacts play important roles in structural propagation of allosteric transitions. Since structural changes propagate by atomic displacements, the propagations of atomic displacements would stop upon encountering internal void spaces in the protein structure. Therefore, the absence of internal void spaces is crucial for successive structural changes to propagate to spatially distinct sites, i.e., proteins are molecules of functional compactness.

We observed that size of the frequency shift of ν(Fe–His) in Mb mutants was similar even if the covalent bond between the imidazole ring and the polypeptide chain was absent. Such a frequency shift was not observed for the heme group when dissolved in micelles (described in Section 3.1.1). These results indicate that, even without the covalent bond, structural changes in the protein matrix can be induced. Since modifications of the heme vinyl group, or a mutation at the valine residue contacting heme, significantly reduced the cooperativity of Hb,^152–156 indicating contact between the residue and vinyl group of the heme was considered essential for transmitting the information about oxygen binding to another subunit. In microbial rhodopsins, we observed changes in Trp residue spectra in 10 ps, which were observed commonly for the five microbial rhodopsins studied, which suggested that structural changes around the Trp residues are induced by atomic contacts between the retinal chromophore and the surrounding amino acid residues.

Allostery is associated with mechano-stress in proteins. Recently, Mitchell et al. studied crystal structures and NMR spectra of proteins in various regulatory and ligand binding states.¹⁵⁷ They calculated and analyzed distributions of strain throughout several proteins. Strains reveal allosteric and active sites, and suggest that quasi-two-dimensional strained surfaces mediate mechanical couplings between them. From a mechanical point of view, allosteric proteins are ‘mechanical transducers’ that transmit regulatory signals between distant sites. Thus, control of enzymatic activity by mechanical stress is interesting in the context of allosteric mechanisms. For example, mechano-sensitive enzymes, such as titin-like kinases, have been investigated.^158,159 The enzyme-DNA chimeras, in which a DNA molecular spring attached to the enzyme exerts a force in a known direction, were used for measurements of the kinetics of catalyzed reactions under a non-destructive mechanical stress on an enzyme.^160–164 An asymmetric effect of mechanical stress on the forward and reverse reactions catalyzed by an enzyme was reported recently.¹⁶⁵ The enzymatic activity was controlled through a photochromic structural change in an azobenzene derivative, which can adopt a cis- or trans-configuration when illuminated by UV or blue light, respectively.¹⁶⁶ Proteins that act as mechanical transducers have very high atomic density in the three-dimensional structure.

Our studies on energy flow in proteins revealed that atomic contacts are important for energy transfer. The anti-Stokes Raman intensities of the Trp residue in the three mutants increased with similar rates after depositing excess vibrational energy at the heme, despite the differences in distance between the heme and each substituted Trp residue along the main chain of the protein. This indicates that vibrational energy is not transferred through the main chain of the protein, but through atomic contacts between the heme and Trp residue. The amplitude of the band intensity change correlated with the extent of exposure of the Trp residue to solvent water, indicating that atomic contacts between an amino acid residue and solvent water play an important role in the vibrational energy flow in a protein.

Atoms are packed much more densely in typical globular proteins, and these protein structures can be described in terms similar to those used for dense liquids. Our observations demonstrated that the highly packed nature of protein architecture plays an important role in protein dynamics and energy flow. Interestingly, the feature of dense atomic packing is important for protein folding into a unique structure. Gō proposed the consistency principle of protein folding, which stated that various energy terms contributing to the stabilization of the native state of globular proteins are consistent with each other in the first approximation in the native state.^167,168 This means that each energy term is individually minimized at the lowest total energy value, which ensures the stability of the native structure. A similar hypothesis, the principle of minimal frustration suggested by Bryngelson and Wolynes, states that energetics frustration is minimized.¹⁶⁹ Accordingly, consistency in interactions within the protein structure results in close atomic packing with minimal frustration. Figure 11 shows relationship between folding, structure, and dynamics discussed in terms of the nature of the high packing density.

In contrast to a coarse-grained view, protein structures exhibit inconsistencies and frustration when a finer-grained view is taken on both spatial and energy scales, as evident by the multiple-minima feature of the energy surface. Kitao et al. showed that, if a proper set of collective coordinates is selected, the energy surface along most of the collective axes (e.g., 95%) is essentially harmonic with a single-minimum feature, while the energy surface along a small number of large amplitude principal axes has a hierarchical multiple-minima feature,^170,171 indicating that the latter modes of proteins are frustrated. Therefore, external perturbations can induce anisotropic motions and trigger the protein to switch from one state to another along the intrinsic collective directions. Such modes can be a switch for allosteric transitions.

5.2 Descriptions of Protein Motions

An ultrafast structural change at the prosthetic group perturbs the protein structure in a step-function manner. Therefore, the photodissociation of gas ligands in heme proteins and photoisomerization and photoexcitation in microbial rhodopsins and PYP provide good opportunities to study the protein response to reaction at the prosthetic group. In our studies, the protein response was observed as the frequency shift of the ν(Fe–His) mode with the ∼100-ps time constant in Mb and Hb, and spectral changes of aromatic amino acids with the ∼10-ps time constant in microbial rhodopsins and PYP. Although the response function was not expected to be simple, two limiting cases can be considered with respect to the length scale and dynamics of the protein response function for reaction at the prosthetic group. On a long timescale for the motions, protein motion between local minima in its structure is averaged over a certain time interval, and, thus, atomic displacements are diffusive in nature. Therefore, conformational changes along the reaction coordinate appear as a thermally activated Brownian process. These processes can be described by Kramers’ theory based on the viscosity dependent rate of structural changes in Mb (Section 3.1.1). A large number of nearly degenerate conformational substates leads to a distribution of barrier heights and pathways so that structural relaxation is highly nonexponential, as observed in the geminate recombination of CO^172,173 and other ligands in heme proteins.^174,175 Because proteins have a highly associative nature, this protein response function on a long timescale often is modeled as analogous to glass relaxation dynamics.¹⁷⁶

In the other limiting case, in the short-time dynamics of the protein response to the reaction forces, the dominant contribution to atomic displacements in proteins is nondiffusive motion. In the extreme, the nascent gradients on a potential energy surface that develop during ligand photodissociation and chromophore photoisomerization leads to collective displacement of a large number of atoms. In this event, the structural changes are best described as a displacement of atoms in a superposition of the low-frequency collective modes of the protein, rationalized by the fluctuation–dissipation theorem.⁵⁹ This description is validated by the accurate modeling of the short-time behavior of liquids such as inertial motions provided by the modal approach (e.g., instantaneous-normal-mode approach).^177,178 Atoms in proteins are covalently bonded. The use of a modal description for the short-term dynamics of protein motion is more rigorous than it is for liquids. For example, normal mode analysis by Seno and Gō revealed that 57% of the tertiary structural changes can be accounted for by the displacement of six spatially extended modes with frequencies ranging from 5 to 12 cm⁻¹.¹⁷⁹ Numerous examples have shown that functionally important transition pathways of protein structures often follow the trajectories of one or a few low-frequency normal modes.^180,181 Data related to ultrafast protein responses contribute to verification of the linear response approach. An example was demonstrated by observation⁵⁰ and calculation⁶⁰ on primary protein responses upon CO dissociation (Section 3.1.1).

Finally, the validity of the modal description of protein responses is related to the time scale of vibrational energy relaxation. The criteria needed to discriminate collective from diffusive protein motions can be obtained by timescale vibrational energy relaxation in proteins. For example, if the vibrational modes of the protein are strongly anharmonically coupled, vibrational redistribution within the protein can occur very rapidly. At the same time, the number of low-frequency normal modes that account for the atomic displacements in the protein response increases, causing the modal description to become less appropriate. This was argued for stochastic processes in a condensed phase.^182,183

6. Summary and Future Outlook

I enjoy studying protein dynamics from the view point of molecular science. But one might ask whether studies on proteins can provide any general understanding about molecules and molecular systems. Good reasons exist to believe that the answer is yes.

Structural dynamics is a central issue of molecular science. Strong connections exist between protein dynamics and condensed-phase dynamics involving chemical reactions. For example, observation of primary protein responses (Section 3) is analogous to dynamic Stokes shift measurements for solvation dynamics.^184,185 Section 4 describes our studies on energy flow in proteins. These studies are related to investigations on vibrational energy relaxation in liquids, which were thoroughly examined for solutions of diatomic and polyatomic molecules.^120,122 In addition, proteins, as static structures, often are investigated at atomic resolution using X-ray crystallography, NMR spectroscopy, and cryo-electron microscopy. The structures can be modified artificially using site-directed mutagenesis. These are distinct advantages of physicochemical studies on proteins compared to such studies on liquids. Therefore, questions about proteins, in addition to their importance in protein studies, can provide important information and answer questions and can stimulate innovative studies in molecular science. For this reason, the most detailed understanding of solution-phase reaction dynamics may come from studies of biological systems.

Creation of functional molecules and gaining an understanding of their functions also are important subjects in molecular science. Proteins perform functions that artificial molecules cannot do. In oxygen binding by Hb, for example, the heme group and the protein matrix profoundly affect each other in an elegant manner; the protein’s structural changes form a strongly coupled system in which tertiary relaxation cascades into the quaternary structural transition. Biomechanics of the tertiary structural changes in ion pumps are important for a detailed understanding of directional ion transport. The protein undergoes conformational changes so that transient ion binding sites change ion affinities in a concerted fashion. Knowledge about the mechanism of allostery can help design competent proteins that exhibit switch-like behavior. Much work remains to be done to fully understand and utilize the principle of allostery. Once the principle is understood, the ability to interlock multiple functional units in molecular systems may be possible. Advances in understanding allostery also are being applied to the engineering of sensitive modular switches, in which an input domain is connected to an output domain. These switches may be useful in sensing molecules that are silent to conventional detection techniques and in executing functions dependent on molecular concentration.

I am optimistic that studies on proteins will contribute more extensively to our general understanding of molecules and molecular systems, and lead to the development of innovative and useful molecules. Advances in both theoretical and experimental methods of studies over a broad range of timescales and length scales of the protein response are required to reach this ultimate goal.

Acknowledgment

I express my gratitude to colleagues whose names appear in the references. I am especially grateful to Dr. Misao Mizuno, Dr. Haruto Ishikawa, Dr. Akira Sato, and Dr. Masato Kondoh. My sincere thanks go to Prof. Teizo Kitagawa, a professor emeritus at the Institute for Molecular Science, for leading me to studies of protein allostery and vibrational spectroscopy, and to the late Prof. Robin M. Hochstrasser, from the University of Pennsylvania, for showing me the beauty and fun of ultrafast chemistry of condensed phases. This work was supported by Grants-in-Aid for Scientific Research (Nos. JP15350013, JP17350009, JP20350007, JP23350007, and JP26288008) from the Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research on the Priority Areas “Molecular Science for Supra Functional Systems” (No. JP19056013), and on Innovative Areas “Soft Molecular Systems” (No. JP25104006) from the Ministry of Education, Culture, Sports, Science and Technology and support for Precursory Research for Embryonic Science and Technology from the Japan Science and Technology Agency. Finally, my heartfelt thanks go to my wife for her constant support and encouragement.

References

Yasuhisa Mizutani

Yasuhisa Mizutani was born in Kuwana, Mie in 1964. He earned his B. Eng. and M. Eng. degrees in 1987 and 1989, respectively, from Kyoto University, under the supervision of Prof. Koichiro Nakanishi. He received his Ph.D. degree from the Graduate University for Advanced Studies in 1992 under the supervision of Prof. Teizo Kitagawa. After that, he completed a postdoctoral fellowship at the Institute for Molecular Science and another at the University of Pennsylvania, where he worked with Prof. Robin M. Hochstrasser. In 1994, he joined the faculty at the Institute for Molecular Science as a research associate in the group of Prof. Teizo Kitagawa. He was appointed as an associate professor of Kobe University in 2001 and a full professor of Osaka University in 2006. His research interest has been focused on the application of time-resolved resonance Raman spectroscopy to a wide range of functionally important protein dynamics and energetics. He received the 11th Inoue Research Award for Young Scientists of Inoue Foundation for Science in 1995, the 17th Morino Science Award from the Morino Foundation for Molecular Science in 2001 and The Chemical Society of Japan Award for Creative Work from the Chemical Society of Japan in 2013.