Subpicosecond Midinfrared Spectroscopy of the P_fr Reaction of Phytochrome Agp1 from Agrobacterium tumefaciens

Abstract

Phytochromes are light-sensing pigments found in plants and bacteria. For the first time, the P_fr photoreaction of a phytochrome has been subject to ultrafast infrared vibrational spectroscopy. Three time constants of 0.3 ps, 1.3 ps, and 4.0 ps were derived from the kinetics of structurally specific marker bands of the biliverdin chromophore of Agp1-BV from Agrobacterium tumefaciens after excitation at 765 nm. VIS-pump-VIS-probe experiments yield time constants of 0.44 ps and 3.3 ps for the underlying electronic-state dynamics. A reaction scheme is proposed including two kinetic steps on the S₁ excited-state surface and the cooling of a vibrationally hot P_fr ground state. It is concluded that the upper limit of the E-Z isomerization of the C₁₅ = C₁₆ methine bridge is given by the intermediate time constant of 1.3 ps. The reaction scheme is reminiscent of that of the corresponding P_r reaction of Agp1-BV as published earlier.

INTRODUCTION

In plants and bacteria, a multitude of processes is controlled by phytochromes (1,2), a class of photoreceptor proteins with two photochemically interconvertible and thermally stable states P_r and P_fr that absorb in the red and far-red spectral region, respectively. This feature of phytochromes not only allows the investigation of two different reactions of the same chromophore within one binding environment, but also makes them prototypes for biomimetic bistable light-driven switches. Three-dimensional structures of the chromophore-binding environments of bacterial phytochromes, DrBphP (3,4) and RpBphP3 (5), have recently been resolved.

The primary processes of the two photoconversions involve a Z-E isomerization of the methine bridge between rings C and D of the bilin chromophore for the P_r reaction and an E-Z isomerization for the P_fr reaction (6,7). Each reaction pathway involves several intermediate states, with the first ones being lumi-R (P_r reaction) and lumi-F (P_fr reaction), which are formed quickly after photoexcitation. The following reaction steps occur on the microsecond to millisecond timescale and have been characterized by UV/VIS, FTIR, and resonance Raman (RR) spectroscopy using low-temperature trapping techniques (8–15). In all bacterial and plant phytochromes known so far, the relatively slow P_r photoreaction takes place within ∼5–100 ps as opposed to the relatively fast P_fr photoreaction which occurs on the timescale of a few picoseconds (16–18).

The primary photoreaction of the P_r form of different members of the phytochrome family has been subject to numerous investigations, including ultrafast VIS-VIS (17–22) and VIS-IR (23,24) pump-probe as well as fluorescence (25–27) spectroscopy. In contrast, the literature on the primary processes of the P_fr reaction is still very scarce. However, the high rate of its excited electronic-state decay has been quantified by ultrafast VIS-VIS spectroscopy (16–18). Further, fluorescence studies on the P_fr form of oat phytochrome do not show any detectable fluorescence (26), in line with a very efficient quenching of the excited electronic state.

In this work, we address the primary photoreaction of the P_fr form of the biliverdin-binding phytochrome Agp1 (Fig. 1) from Agrobacterium tumefaciens (28) by ultrafast mid-IR transient absorption spectroscopy. To our knowledge, this is the first ultrafast IR investigation of the P_fr reaction of a member of the phytochrome family. Details of the chromophore-protein interaction in Agp1-BV have already been obtained by site-directed mutagenesis and mass spectrometry, showing the covalent binding of the biliverdin (BV) chromophore via its ring A-vinyl side chain to the Cys²⁰ residue (29). In both stable states, P_r and P_fr, the chromophore has been found protonated via RR spectroscopy and flash photolysis experiments (30). Recent results using locked bilin chromophores (31,32) suggest that the chromophore configuration is ZZZssa in P_r and ZZEasa in P_fr (ZZZssa and ZZEasa denote the configuration (Z, E) and conformation (a, s) of the methine bridges in the order A-B, B-C, and C-D).

FIGURE 1.

Chemical structure of the biliverdin (BV) chromophore in ZZEasa configuration, as suggested for the P_fr state (31).

As has been demonstrated earlier (34,35), ultrafast mid-IR spectroscopy is an excellent method to monitor light-induced structural dynamics. Here, it allows us to follow the transient vibrational spectra of the molecular states along the primary P_fr photoisomerization in Agp1-BV with subpicosecond time resolution. Combined with the electronic-states dynamics as obtained by transient absorption measurements in the visible spectra, different reaction models for the primary P_fr reaction can be evaluated.

METHODS

Sample preparation

Agp1 from Agrobacterium tumefaciens was expressed and purified as already described elsewhere (28). A D₂O (pD = 7.8) buffer solution (20 mM Tris, 50 mM NaCl) was used for all experiments. The protein was concentrated by ultrafiltration (YM-50, Centricon, Houston, TX) to a viscous smear, homogeneously spread on a 1.5-inch-diameter CaF₂ window (2 mm thick) and sealed by a second, similar window. In the absorption maximum of the P_r form at λ_max = 700 nm, the optical density of the sample was between ∼0.5 OD and ∼1.0 OD, equivalent to a pathlength of ∼25–50 μm. This translates to an optical density of ∼2 in the amide I/II-region of the steady-state FTIR spectrum of the sample. Note that the photoinduced IR-difference signals (see discussion below) are three orders-of-magnitude smaller (∼1 mOD). To allow experiments in the region of high amide I background absorption, samples of lower optical density were used.

The sample was rotated and moved in the focus plane of the laser beam perpendicular to the direction of incidence during the experiment to provide fresh sample conditions for each laser pulse, i.e., to exchange the (microscopic) excited sample volume between two pump-and-probe events (1.6 ms) and to allow for sufficient recovery of the P_fr state. To avoid photoproduct buildup, firstly an excitation wavelength of λ_exc = 765 nm on the red edge of the P_fr absorption band was chosen. Secondly, the sample was irradiated by background light from a halogen lamp (K2500-LCD, red filter, Schott, Mainz, Germany), fitting the absorption spectrum of the P_r form (λ_max = 700 nm). All measurements were performed at room temperature. Sample integrity was confirmed by static FTIR and UV/VIS spectroscopy before and after the experiments. Except for general bleaching of the steady-state absorption in the visible up to ∼10% until the sample was discarded, no spectral changes were observed.

Pump-probe spectrometer

The short laser pulses for the pump-probe spectrometer were generated in nonlinear optical devices. A Ti:Sa-regenerative amplifier system (CPA 2001, Clark-MXR, Dexter, MI) was used as pump source for the whole experiment. The visible pump pulses were generated in a homebuilt noncollinear optical parametric amplifier (NOPA), yielding pulses tunable between 470 and 765 nm with pulse lengths routinely at 60 fs.

Infrared probe pulses were generated in a two-stage optical parametric amplifier with subsequent difference frequency mixing. The center wavelength of the probe pulses is tunable between 800 cm⁻¹ and 2500 cm⁻¹ (36). After passing through the sample, the probe pulses with a typical full width at half-maximum (FWHM) of 100 cm⁻¹ are dispersed in a polychromator and detected by a 32-element MCT array (Infrared Systems Development, Winter Park, FL), comprising a spectral window of ∼300 nm (≈90 cm⁻¹ at 1750 cm⁻¹ and ≈40 cm⁻¹ at 1180 cm⁻¹). The pump beam was chopped at 318 Hz (half the laser repetition rate), and pump-induced absorption differences were evaluated on a single-shot basis. Control measurements on a thin silicon wafer were performed before and after each experiment to determine the time zero and the FWHM of the instrument response function (typically 280 fs). The optical path through the front CaF₂ window and other parameters were identical to those of the phytochrome measurements. The spectral resolution, which varies slightly with the probe wavenumber, was typically 2.5 cm⁻¹.

For VIS-VIS pump-probe experiments, the same NOPA was used as pump source, and the probe pulses were generated by a second NOPA. The chopping scheme was the same as in VIS-IR experiments. After passing through the sample, the probe pulses (spectral width ∼20 nm) were dispersed by a monochromator (bandwidth 6 nm) and detected by a photodiode. Time zero and FWHM of the system response (100 fs) were determined by cross correlation in a BBO crystal or in a SiC-photodiode (two-photon absorption (37)).

The recorded data are the pump-induced absorbance changes ΔA(Δt, λ_pr); cuts along the time axis (Δt) are transients for fixed wavenumbers; and cuts along the probe wavenumber axis (λ_pr) are difference spectra at discrete delay times. The broadband IR probe pulses along with the detector array allow the detection of one spectral window at a given center wavelength (see above). Comparability and normalization of the IR data within the entire investigated spectral regions (Figs. 2 and 3) was achieved via sufficient overlap between adjacent spectral windows.

FIGURE 2.

Difference spectra of the carbonyl and ethylenic stretch regions of Agp1-BV P_fr after excitation at λ_exc = 765 nm at selected delay times.

FIGURE 3.

Difference spectra of the fingerprint region of Agp1-BV P_fr after excitation at λ_exc = 765 nm at selected delay times.

Negative absorbance changes display the disappearance of IR absorption and thus depopulation of the P_fr electronic ground-state vibrations (bleach bands). Positive absorbance changes indicate the absorption of newly populated vibrational states.

The obtained absorbance changes ΔA(Δt, λ_pr) between 0.4 ps and 50 ps were analyzed by a global multiexponential fit,

(1)

with A₀(λ_pr) being the pump-induced absorption changes for Δt → ∞ and A_i(λ_pr) being the decay associated spectra (DAS) of the corresponding time constants τ_i.

RESULTS

Transient IR absorption spectra of Agp1-BV after photoexcitation at λ_exc = 765 nm were recorded for the carbonyl and ethylenic stretch regions (1490–1760 cm⁻¹) and the fingerprint region (1177–1288 cm⁻¹). Difference spectra in the respective spectral regions at various delay times are depicted in Figs. 2 and 3. To correlate the transient vibrational signals (Fig. 4) with the electronic-state dynamics, ultrafast VIS-VIS pump-probe experiments were conducted on the same samples, with excitation at 765 nm and probing at 630, 655, 680, 705, 730, and 755 nm (see Fig. 5 for representative transients).

FIGURE 4.

Transient absorption changes of Agp1-BV P_fr after excitation at λ_exc = 765 nm at selected probe wavenumbers of 1554 cm⁻¹ and 1574 cm⁻¹. (Dashed line) Global fit with time constants of τ₁ = 0.3 ± 0.1 ps, τ₂ = 1.3 ± 0.1 ps, and τ₃ = 4.0 ± 0.1 ps. For the fit, only data after 400 fs delay time was used to exclude nonlinear artifacts before (perturbed free induction decay (57)) and at time zero (cross phase modulation (58)) as well as the system response.

FIGURE 5.

Transient absorption changes of Agp1-BV P_fr after excitation at λ_exc = 765 nm at selected probe wavelengths of 630 nm and 755 nm. (Dashed line) Global fit with time constants of t_V1 = 0.44 pm 0.05 ps, and t_V2 = 3.3 pm 0.1 ps. Data before 400 fs delay time was excluded from the fit.

Band assignment

The structural dynamics of the biliverdin chromophore are monitored by the temporal evolution of vibrational marker bands. The timescale of the detected absorbance changes and the fact that the most significant structural changes are to be expected from the chromophore strongly suggest an assignment of the detected signals to vibrations of the BV chromophore rather than the protein moiety. Taking into account results from experiments using locked chromophores (31,32), we base our band assignment on the commonly accepted assumption that the chromophore structure of the P_fr form is ZZEasa. In support of the literature-based assignments, density functional theory (B3LYP/6-31G**) frequency calculations (38) on the ZZEasa chromophore were performed. For the calculations, all pyrrole nitrogens were modeled as protonated (30), no counterion was used, and vibrational frequencies were scaled with a global factor of 0.9613 (39).

In the carbonyl stretch region (see Fig. 2), the instantaneous bleach band at 1710 cm⁻¹ is assigned to the C₁₉ = O stretching vibration, based on the results in the literature (9,10,40) and our own calculations. The hydrogen bonding of the ring D carbonyl group to His²⁸⁰ (3,15) lowers its stretching frequency significantly, whereas the ring A carbonyl group is not hydrogen-bonded and shows up at higher wavenumbers in PhyA-PΦB (10). Furthermore, since the isomerization occurs at the methine-bridge linking rings C and D, the contribution of modes located on ring D to the difference spectra can be expected to be substantially larger.

The manifold of bleach bands with peaks at 1594 cm⁻¹, 1580 cm⁻¹, and 1574 cm⁻¹ can be assigned to C=C stretching vibrations of the π-system comprised by the pyrrole rings and the linking methine bridges. Previous FTIR experiments on isotope-labeled Cph1-PCB (40), as well as RR experiments on Agp1-BV (30) and PhyA-PΦB (13,14), and FTIR experiments on PhyA-PΦB (9) have found these bands and assigned them to C=C stretching modes. In addition, our own calculations show a multitude of modes with dominant C=C stretching character in this spectral region.

Two pyrrole breathing modes located mainly on rings B and C are favored for the assignment of the small bleach band at 1509 cm⁻¹. The frequencies of these two modes were calculated to 1491 cm⁻¹ and 1523 cm⁻¹. The involvement of their constituent atoms in the delocalized π-system of the chromophore backbone makes them sensitive to the isomerization.

The broad and short-lived positive signal between 1610 cm⁻¹ and 1700 cm⁻¹ appears almost instantaneously and decays with a time constant of 0.3 ps (see analysis below, Fig. 6). Thus, it is likely due to S₁ vibrational modes.

FIGURE 6.

Decay-associated spectra for the P_fr primary reaction of Agp1-BV in the carbonyl and ethylenic stretch regions as derived from a global fit.

In the fingerprint region, our own calculations show numerous C-H rocking and C-C stretching vibrations, so that the assignment of the difference bands in Fig. 3 at 1276 cm⁻¹, 1238 cm⁻¹ and 1184 cm⁻¹ to modes of that character is only qualitative.

Kinetic analysis

A global analysis of the complete time-resolved IR data was performed via Eq. 1. Three exponentials are necessary to fit the data, whereas a four-exponential approach does not increase the goodness of fit. This analysis yields time constants of τ₁ = 0.3 ± 0.1 ps, τ₂ = 1.3 ± 0.1 ps, and τ₃ = 4.0 ± 0.1 ps, and the corresponding decay-associated spectra (DAS) A₁, A₂ and A₃, as shown in Figs. 6 and 7 for the respective spectral regions. Singular value decomposition of the time-resolved difference spectra and triexponential fit of the most significant component (not shown) renders time constants of 0.3 ps, 1.0 ps, and 3.6 ps. This further corroborates the results of the global fit.

FIGURE 7.

Decay-associated spectra for the P_fr primary reaction of Agp1-BV in the fingerprint region as derived from a global fit.

The further discussion will make significant use of the shapes of the DAS, whose features can be compiled as follows (note that negative/positive amplitudes of the DAS represent absorbance strength, which increases/decreases with the respective time constants): A₁ shows signals that appear within the system response time of the experiment, thus positive signals are interpreted as signals from the first states detectable by the experiment. The short-lived and broad absorption bands centered at ∼1740 cm⁻¹, 1650 cm⁻¹, and 1265 cm⁻¹ are then likely to be caused by S₁ vibrational modes that are formed rapidly after photoexcitation. The negative contributions in A₁ at ∼1563 cm⁻¹ and 1230 cm⁻¹, together with its local minimum at ∼1697 cm⁻¹, do not show any significant overlap with the ground-state bleach signals at 1710 cm⁻¹, 1594 cm⁻¹, and 1238 cm⁻¹, but are systematically red-shifted. Furthermore, they show a systematic overlap with negative contributions of A₂ and positive contributions of A₃, suggesting that A₁ and A₂ represent processes that feed a state related to A₃. In the context of the described systematics, it is plausible to regard the local minimum of A₁ at 1697 cm⁻¹ as a negative contribution in A₁, in line with those at 1563 cm⁻¹ and 1230 cm⁻¹, since the broad positive S₁-absorption in A₁ can easily obscure negative bands.

On the other hand, the negative contributions of A₃ coincide well with the most prominent ground-state bleach signals at 1710 cm⁻¹, 1594 cm⁻¹, and 1238 cm⁻¹, in addition to further minor bleach signals at 1511 cm⁻¹, 1277 cm⁻¹, 1260 cm⁻¹, and 1186 cm⁻¹. Overall, A₃ shows a shape that is typical for vibrational cooling, with negative contributions overlapping with ground-state absorption bands, and positive contributions that are systematically red-shifted and asymmetrically broadened to the red side of the spectrum. Such characteristic patterns have been observed with transient IR spectroscopy on azobenzene (41) and protonated Schiff base retinal in solution (42) and have been attributed to vibrational cooling in the electronic ground state. Similar processes have been reported for many molecular systems in the condensed phase (43). The time constant τ₃ in our experiment complies with the time range found for electronic ground-state vibrational cooling in protein systems (24,34,44).

The A₀ spectra of the global analysis of the infrared data represent the residual signals for virtually infinite delay time, and thus should show the difference spectra of lumi-F–P_fr. Although the low isomerization quantum yield (see below) hampers the identification of lumi-F in the A₀ spectrum, qualitative agreement is found between the A₀-carbonyl region and low-temperature FTIR difference spectra of PhyA-PΦB (9,10).

From the transient absorption experiments in the visible with probe wavelengths of 630, 655, 680, 705, 730, and 755 nm, two global time constants of τ_V1 = 0.44 ± 0.05 ps and τ_V2 = 3.3 ± 0.1 ps were derived. Singular-value decomposition analysis as well yields two time constants of 0.6 ps and 3.3 ps, in good agreement with the global fit results. The transients given in Fig. 5 show the recovery of the electronic ground-state bleach (λ_max ≈ 750 nm), which is probed at 755 nm, and the decay of the S₁ excited-state absorption, which is probed at 630 nm and which accompanies the ground-state bleach decay. It should be kept in mind that the P_fr state of Agp1-BV exhibits significant absorbance between 600 and 800 nm (28). The traces at both wavelengths show contributions from both detectable time constants τ_V1 and τ_V2, whereas the amplitude of τ_V2 is substantially larger for the recovery of the ground-state bleach compared to the excited-state absorption decay. The contribution of the τ_V2 component in the transient signal at 630 nm is considered due to ground-state processes (see below).

The relative amount of recovery of the initial ground-state bleach signals in the IR allows the determination of the isomerization quantum yield independently of a specific reaction scheme (24). For this purpose, the IR transients at 1710 cm⁻¹, 1594 cm⁻¹, and 1238 cm⁻¹ were quantitatively evaluated. Their average bleach recovery suggests a quantum yield for lumi-F formation of ∼8%, which deviates from the reported value of 0.4% for the P_r formation (28). Note that the only scenario leading to an overestimation of the lumi-F quantum yield is an overlap of the (initial) bleach bands by short-lived absorption bands directly after photoexcitation. This can be excluded for all but the 1710 cm⁻¹ bleach bands, and is thus very implausible to cause an error of one order of magnitude.

DISCUSSION

The analysis of the ultrafast data allows the construction of several reaction schemes for the primary processes, three of which (Schemes A–C, see Fig. 8) are taken into consideration, with Scheme C being discussed in more detail. In the following, τ_i and τ_Vi are used as given above.

FIGURE 8.

Reaction schemes for the P_fr primary reaction of Agp1-BV (see text). Solid bars denote excited electronic states, open bars denote electronic ground states.

Scheme A includes a structural relaxation from the Franck-Condon region to a S₁-state P_fr* within τ₁, where a branching occurs. The productive pathway of the branching leads directly to the first metastable photoproduct lumi-F, whereas the nonproductive part decays back to a vibrationally hot electronic ground-state O_fr, which repopulates the (cold) P_fr vibrational ground state. The decay of P_fr* is supposed to occur with a rate constant of 1/τ₂ and the O_fr→P_fr transition with a rate of 1/τ₃. This reaction scheme is compatible with the analysis of the infrared data in that the positive contributions in A₃, which we attribute to a population of O_fr, always coincide spectrally with negative contributions from A₁ and A₂. Since the rate constants attributed to A₁ and A₂ are relatively similar, these two processes cannot be viewed as completely decoupled, hence the formation of O_fr shows kinetic contributions from the decay of S^FC and P_fr*.

The deficiencies of Scheme A are twofold: Firstly, for the photoreaction to occur on an ultrafast timescale, one would assume the gradient of the S₁ surface in the Franck-Condon region to be relatively large. This implies that the initial structural relaxation (not necessarily along the C₁₅ = C₁₆ torsional coordinate) occurs significantly faster than the system response time of the experiment and also faster than τ₁. Secondly, the decay of the excited-state absorption as detected in the VIS-VIS experiment shows τ_V1 as dominant time constant, which is very close to τ₁. Thus, we can assume τ₁ to represent the dominant S₁ decay process, which contradicts its interpretation as movement out of the Franck-Condon region.

Scheme B is considered as a sequential reaction on the S₁ excited-state potential energy surface involving two S₁ substates, P_fr** and P_fr*. From the Franck-Condon region, the reaction proceeds via P_fr** and P_fr*, with a branching to lumi-F and to the electronic ground-state P_fr (in its vibrational ground state) that occurs with the decay of P_fr*. Here, the time constants are assigned as follows: Relaxation from the Franck-Condon region with τ₁, reaction from P_fr** to P_fr* with τ₂, decay of P_fr* and formation of lumi-F and P_fr with τ₃. An additional reaction path between P_fr** and P_fr cannot be excluded.

Scheme B not only suffers from the same deficiencies as described above for Scheme A. Additionally, it cannot be assumed in general that the decay of the excited electronic-state P_fr* leads directly to a (vibrationally) relaxed electronic ground state. For this unrelaxed ground state not to be traceable in the time-resolved infrared data, either its lifetime has to be small compared to τ₃, or its vibrational spectrum has to coincide with that of P_fr* or of P_fr. However, the lifetime of the vibrationally excited electronic ground state can be estimated to ∼3–4 ps by comparison to the P_r reaction (24). Further, since the additional state is assumed to be structurally or vibrationally unrelaxed, its spectrum cannot coincide with that of P_fr, and a coincidence with the P_fr* vibrational spectrum seems far-fetched.

Scheme C accommodates our observations much better and avoids the difficulties of Schemes A and B. Here, the step from the Franck-Condon region to P_fr** is assumed to be much faster than the system response time of the experiment and thus undetectable. With the decay of P_fr**, a first branching occurs with reactions to P_fr* and to the vibrationally unrelaxed electronic ground-state O_fr. The time constant τ₁ is assigned to the decay of P_fr**. P_fr*, with a lifetime of τ₂, decays with a branching between the formation of the first metastable photoproduct lumi-F and the formation of a structurally unrelaxed form of the educt-state P_fr, subsumed with O_fr. O_fr reacts back to the vibrational ground-state P_fr, via vibrational cooling and structural relaxation within τ₃. Since O_fr is populated from P_fr** and P_fr*, it is suggested to comprise vibrationally hot as well as structurally unrelaxed P_fr ground states that cannot be spectrally separated and share closely similar kinetics in their relaxation to P_fr.

It should be pointed out that the construction of the reaction schemes is based on the three time constants and the related DAS that were derived from the highly structured IR data. The two time constants of the VIS transients are nevertheless consistent with the dynamics described in Scheme C. Apparently the fit cannot distinguish between τ₁ and τ₂, but yields a weighted average τ_V1. This can be rationalized by the fact that the electronic absorption spectra are generally broad and unstructured, in contrast to vibrational spectra. Thus, the differences between the P_fr** and the P_fr* VIS spectra along with their similar lifetimes do not allow their separation into two kinetic components. The longer lifetime τ_V2 = 3.3 ps is in fair accordance with τ₃ = 4.0 ps obtained in the IR. Following Scheme C, τ_V2 has then to be attributed to processes on the electronic ground-state surface: 1), the recovery of the vibrationally fully relaxed P_fr electronic ground state, monitored at 755 nm; and 2), the decay of the structurally unrelaxed fraction of O_fr, which is suggested to contribute significantly to the transient absorbance at 630 nm.

Similar time constants have been observed in previous VIS-VIS experiments on the P_fr form of PhyA-PΦB (17) (0.68 ps and 4.0 ps) and the P_fr form of Cph1-PCB (18) (0.54 ps and 3.2 ps). These results suggested a reaction with two consecutive steps, but no specific reaction schemes were presented. For PhyA-PΦB, it remained unclear whether the longer time constant describes an electronic ground- or excited-state process. Similarly, the Cph1-PCB data did not allow an unequivocal assignment of the kinetics to electronic ground- or excited-state dynamics.

So far in this discussion, photoinduced intramolecular processes as transitions between different electronic-state potential energy surfaces, and conformational changes as well as vibrational relaxation of the chromophore, have been addressed. In addition, intermolecular energy flow from the chromophore to the surrounding protein and finally to the buffer solution will lead to an increased temperature of the microscopic sample volume. Timescales for energy conduction and heat diffusion, respectively, through protein matrices, have been found to be on the timescale of our experiment, e.g., 7–20 ps for hemoglobin (45). One temperature effect could be a change in the absorption bands of the surrounding water, leading to a spectrally rather unspecific baseline shift. This was not observed in our experiments. Spectrally specific contributions could originate from heating of the protein backbone, which for bacteriorhodopsin have been shown to yield IR difference bands in the amide II region at ∼1550 cm⁻¹ and (much smaller) in the amide I region (46). The difference signals for long times (A₀ spectrum) do not give indications for such bands. Further spectrally specific changes at long delay times (e.g., 50 ps) could be due to the vibrational spectrum of the chromophore, having released its excess energy to the environment. However, the band shift induced by such a small temperature change (a few Kelvin) is negligible, considering the relatively low chromophore concentration (41,46). We therefore conclude that neither the determination of the quantum yield nor the observability of the lumi-F state are affected by a temperature effect.

The identification of A₃ (and thus also of τ₃) as a relaxation process of a vibrationally unrelaxed P_fr electronic ground state leads to the conclusion that the first metastable photoproduct, which we assume to be lumi-F, can only be populated from the longest living electronic excited-state P_fr*, which decays within τ₂ = 1.3 ps. Considering the low quantum yield of the lumi-F formation, the fact that the cooling process is readily observed strongly suggests it to be part of the nonproductive pathway. Thus, an involvement of τ₃ in the isomerization reaction is very unlikely. Although the low quantum yield does not allow the detection of lumi-F product bands, this suggests the E-Z isomerization to occur along the excited-state reaction pathway, with an upper limit for the isomerization time of τ₂ = 1.3 ps. A stepwise isomerization via a ground-state reaction is implausible, since time-resolved experiments on the P_fr state of other phytochromes (17,18) with higher isomerization quantum yields were also unable to detect spectral changes beyond the order of magnitude of τ₃.

The isomerization quantum yield of the P_fr primary reaction, as derived here via the primary recovery of the P_fr educt-state, is very similar to that of the P_r reaction (24). In contrast to the P_r reaction, where the yield of the primary reaction (lumi-R formation) was found to be equal to that of the P_fr formation (28), the corresponding values of the P_fr reaction differ in more than one order of magnitude, i.e., 8% vs. 0.4%. This suggests the possibility of a short-cut reaction originating in an intermediate state along the P_fr→P_r reaction path and leading back to the P_fr form. In consequence, this dark reaction would have to include a thermally driven Z-E isomerization. Double-bond isomerizations of free chromophores often require the energy of an absorbed photon (47,48). Thus, such a short-cut reaction was unexpected. However, dark conversion of, e.g., bacteriorhodopsin (49), PYP (50), or phytochromes (28) are examples for thermally driven isomerization reactions that take place within the protein environment. The dark conversion of Agrobacterium phytochrome Agp2 and some other bacterial phytochromes proceeds from P_r to P_fr (51,52). These examples show that also a Z→E dark isomerization around the C₁₅ = C₁₆ double bond of the bilin chromophore is not impossible. This study suggests that such a dark isomerization is an integral part of the Agp1 photocycle. It seems that Agp1 is an exception in this respect, because the overall quantum yield of the P_fr-to-P_r conversion of other phytochromes is >10% and thus in the range of the P_fr to lumi-F quantum yield estimated here (28). The rather complex P_fr-to-P_r reaction of Agp1 is thus most likely the consequence of an evolutionary process which finally resulted in the rather low P_fr-to-P_r quantum yield of this phytochrome.

Comparison of reaction Scheme C of the P_fr primary reaction (Fig. 8) with the scheme of the P_r primary reaction (24) yields close similarities. The schemes for both reactions are basically identical, with the only difference in the values of the time constants. The decay of the first structurally relaxed excited electronic-state, P_fr** in the P_fr scheme and A in the P_r scheme, with 0.3 and 0.7 ps, respectively, and the relaxation on the electronic ground state, i.e., the decay of O_fr and O_r, with 4.0 and 3.3 ps, respectively, show very similar time constants. In contrast, the lifetime of the longest-lived S₁ species differs significantly from 1.3 ps in the P_fr reaction to 33.3 ps in the P_r reaction. The very short S₁ lifetime in the P_fr reaction is consistent with the vanishingly small fluorescence quantum yield of the P_fr reaction as compared to the P_r reaction (26). Note that the similarity of the reaction schemes for P_r and P_fr implies in turn specific similarities of the two sets of DAS—A_i (P_r) (24) and A_i (P_fr). In fact, qualitative match is found concerning their sign and relative spectral position. Observed differences, e.g., in terms of spectral width and absolute spectral position are not unexpected, since both reactions originate from different chromophore configurations and thus exhibit different ground- and excited-state vibrational spectra.

The observation of a slow P_r reaction and a fast P_fr reaction in Agp1-BV is in line with earlier results on the primary photoreactions of different phytochromes (16–18). Similarly distinct timescales for the forward- and backward-direction of cis-trans isomerizations have been found in the photochemistry of stilbene and azobenzene. In azobenzene, the isomerization processes are completed within 10 ps for the trans-cis and within 1 ps for the cis-trans direction (53). In stilbene, the cis-trans isomerization takes <1 ps and the trans-cis isomerization >10 ps, dependent on the solvent (54). The asymmetry of the reaction kinetics with respect to the initial configuration reflects the differently shaped regions of the excited-state potential energy surfaces that are accessed upon photoexcitation of the respective cis- or trans-state. Certainly, fundamental aspects of cis-trans photoisomerization already discussed for stilbene and azobenzene in solution can be applied to bilin chromophores in a protein environment. However, the example of retinal proteins (55,56) demonstrates how drastically the specific properties of the strongly anisotropic environment realized by the protein moiety can alter the photoinduced isomerization kinetics of a protein-bound chromophore.

In conclusion, femtosecond IR spectroscopy has brought forward a new and more specified reaction scheme of the P_fr photoisomerization of Agp1-BV. Whether or not a unique reaction scheme applies to the P_fr reaction of many phytochromes, if not of phytochromes in general, is up to future studies.