Protein structural dynamics can span many orders of magnitude in time. Photoactive yellow protein’s (PYP) reversible photocycle encompasses picosecond isomerization of the light-absorbing chromophore as well as large scale protein backbone motions occurring on a millisecond timescale. Femtosecond-to-millisecond time-resolved mid-infrared spectroscopy is employed here to uncover structural details of photocycle intermediates up to chromophore protonation and the first structural changes leading to the formation of the partially unfolded signaling state pB. The data show that a commonly thought stable transient photocycle intermediate is actually formed after a sequence of several smaller structural changes. We provide residue-specific spectroscopic evidence that protonation of the chromophore on a few hundreds of microseconds timescale is delayed with respect to deprotonation of the nearby E46 residue. That implies that the direct proton donor is not E46 but most likely a water molecule. Such details may assist the ongoing photocycle and protein folding simulation efforts on the complex and wide time-spanning photocycle of the model system PYP.
Photoactive yellow protein (PYP) is involved in the signal transduction pathway leading to negative phototaxis of the bacterium Halorhodospira halophila.1 The photoreceptor protein is an excellent model system to study functional protein dynamics and serves as the structural prototype of the PAS (PER-ARNT-SIM) domain superfamily.2 The reversible photocycle has been studied using many different time-resolved spectroscopies, such as UV/vis, infrared (IR), nuclear magnetic resonance (NMR), and x-ray diffraction and scattering methods, and has been found to start with isomerization of the buried p-coumaric acid (pCA) chromophore after absorption of a blue photon and to lead to a partial unfolding of the protein.3–11 These structural changes take place over many orders of magnitude in time.12 In this IR study, a large part of the photocycle is followed from the ultrafast femtosecond timescale up to 750 µs using mechanical delay scanning as well as two electronically synchronized laser systems. The measured dynamics involve multiple transient intermediate states and include structural processes, such as a picosecond trans–cis isomerization of the anionic pCA and its protonation on a few hundreds of microseconds timescale (see Fig. 1) that is accompanied by a spectral shift of PYP’s UV/vis absorption spectrum.10–14 Because the phenolic head of the ionized pCA in the ground state pG is embedded in an H-bond network (see Fig. 1), several potential proton donors are, in principle, available. In the literature, reports favor either the nearby E4615–18 or a water molecule,19,20 as modification of E46 does not prevent the protonation of pCA.21 In this work, the large amount of collected transient IR data are analyzed and reduced via a combination of global analysis (GA) and lifetime density analysis (LDA) methods,22–27 in order to extract spectral correlations in time of the monitored photocycle transitions. These correlations reveal details of the structural events eventually leading up to the protonated pCA. The obtained data are interpreted in light of the extensive body of available assignments and the implications for the widely accepted photocycle scheme are discussed.
A. Sample preparation
For expression of the wild-type protein, the system described in Ref. 32 was used. Purification was performed by an anion exchange column (Q Sepharose High Performance, GE Healthcare) with a salt gradient up to 300 mM NaCl for elution followed by a second purification step by size exclusion chromatography (Sephacryl S100 GE Healthcare). The purity of the protein was verified by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and the content of holo-PYP was about 100% judged by the ratio between the absorption at 446 and 280 nm. For determination of protein concentrations before each pump–probe experiment, the absorption at 446 nm was used with an extinction coefficient of 45500 M−1 cm−133 The protein was dissolved in the D2O buffer (50 mM NaPi, pD 8).
The integrity and quality of samples were also checked before and after each pump–probe experiment via FTIR (using the same sample cell as used for the laser experiments; 50 μm spacer) and UV/vis absorption spectroscopy (using a diluted protein sample in a 1 cm cuvette). No significant changes were observed. The functionality of the protein was also confirmed via the measurement of light-minus-dark difference FTIR absorption spectra (using a LDM-445-1600 diode laser, Lasertack).
B. Pump–probe experiments
The fs-to-ns-experiments are performed with an fs-laser system from Spectra Physics (Spitfire, producing 3 W of 100 fs pulses at 800 nm at 1 kHz). A home-built optical parametric amplifier (OPA) generates mid-IR probe pulses via difference frequency generation in AgGaS2. The 467 nm pump pulse is generated via tripling of the signal of a second OPA. A fused silica rod stretches the pump pulse duration to ∼600 fs to prevent sample damage from the high excitation density. The delay between pump and probe pulses is mechanically scanned from −20 to 866 ps (in 59 timepoints), and the experiments are from now on referred to as “fs” delays.
The performed microsecond (“ms”) delay experiments are done on two synchronized fs-laser systems (similar to the approach described elsewhere34) that electronically delay the pump pulse from −7.5 ns to 750 µs (in 84 timepoints). The same aforementioned Spectra Physics system (containing the master oscillator) now generates mid-IR probe pulses, whereas a second laser system (containing the slave oscillator; Coherent Mira and Legend Elite HE, producing 4.5 W of 100 fs pulses at 800 nm at 1 kHz) generates the pump pulses. The time resolution is determined by the electronic jitter and amounts to 20–30 ps as determined in GaAs. The pump pulse wavelength is set to 468 nm and generated via sum-frequency generation of the signal and fundamental beams in a β-barium borate (BBO) crystal. Two fused silica rods are used to stretch the pump pulses close to picosecond duration to prevent sample damage.
For both experiments, a mechanical chopper (HMS Elektronik) running at 500 Hz is placed in the pump beam path to block every second pump pulse (3 μJ). A custom-built Lissajous scanner moves the sample in a Lissajous pattern (having a size of about 1.3 × 1.3 cm2) so that each pump pulse excites a fresh sample spot. The set speed and pattern bring the sample back to the starting point in about half a minute. Difference spectra are collected in two different spectral windows with the probe OPA centered at λ0 = 6200 nm (νC = C region) or at λ0 = 5770 nm (νC = O region). The final fs and μs datasets are constructed by pixel-concatenation of the two partially overlapping spectral windows. The fs experiments overlap from 1652 to 1656 cm−1 (two pixels from either window), and a signal scaling factor of 0.7 (determined at 56 ps) is used for the low wavenumber region in order to match the intensities of the high wavenumber region. The μs experiments overlap from 1665 to 1670 cm−1 (two pixels from either window) and a signal scaling factor of 0.94 (determined at 1 ns) is used for the high wavenumber region to match the intensities of the low wavenumber region. The need for scaling is most likely caused by small differences in the spatial overlap of the beams. The relative polarization of the pump and probe beams is set to the magic angle. The signals are referenced on a 2 × 32 pixel MCT detector (Infrared Associates) and dispersed by a 150 l/mm grating (Triax 180, Horiba).
C. Data analysis
The time-resolved data presented in Figs. 2 and S1 are analyzed (i.e., reduced) with three methods, i.e., single pixel fitting, lifetime density analysis (LDA), and global analysis (GA). As all methods consistently lead to the same conclusions, the main focus of this paper will be on the LDA and its comparison to the single pixel fits. A comparison between the LDA (using Optimus27) and GA (using a sequential model in the Globe toolbox24) results is presented in Fig. S2. The LDA is performed on the time course of a single pixel via Tikhonov regularization, resulting in a lifetime density map (LDM; see Fig. 3) when all pixels are combined.27 These maps reveal spectral changes in time, analogous to decay-associated difference spectra (DADS) produced by GA. The data used for all methods are corrected for a constant background signal by subtraction of a timepoint before time zero (−20 ps and −7.5 ns for the fs and the μs data, respectively). Optimum regularization is determined by the elbow of the L-curve, given by the correlation between the smoothing and residual norms [see Fig. S2(g)].35,36 In order to be able to use the same L-curve criterion for all pixels, the input data are standardized (i.e., having a mean of zero and a standard deviation of 1) using Matlab’s z-score function.37 Without standardization, LDM features can shift in amplitude and/or time and may exhibit unrealistic oscillations in time on one or more pixels. These oscillations can be suppressed by using lasso regression, which imposes additional constraints.38,39 The spectral features discussed in this work are confirmed by lasso regression (shown in Fig. S3). Standardization can also lead to the fact that features may not line up on the wavenumber axis with spectral features visible in a Species-Associated Difference Spectrum (SADS), as the scaling procedure effectively decreases large signals and amplifies small signals. The LDM can, however, in principle, be rescaled to resemble the raw data (and SADS) better by reversing the standardization via multiplication by the raw data’s standard error. As an LDM then becomes dominated by the largest signals, only the LDA of the standardized data is discussed here. The rescaled data are, however, shown, nonetheless, in Fig. S4. The supplementary information contains further details on and comparisons of the different analysis procedures.
III. RESULTS AND DISCUSSION
Transient vis pump–IR probe data are collected using two different experimental approaches to cover the full photocycle dynamics spanning the femtosecond to millisecond time regime. Figure 2 shows selected transient spectra of the raw data. The fs delays (see the Pump-probe experiments section) probe the ultrafast femtosecond to nanosecond time window that is limited by the pulse length and by the length of the stage. The measured μs delays run from tens of picoseconds (limited by the electronic jitter) to hundreds of microseconds (limited by the laser repetition rate) using two electronically synchronized laser systems. The data from both types of experiment partially overlap in time and are consistent with each other. Figure 2 shows that spectral changes occur throughout the investigated spectral region and on all timescales. Figure S1 shows all available raw data as a surface plot. To extract useful information out of the data spanning nine orders in the magnitude of time, the fs and μs data were separately analyzed via both LDA and GA (see Figs. 3 and S2, respectively). These methods complement each other by representing different ways to look at the data. One important difference is that the GA imposes a model to fit the data, while LDA is model-free. Each resulting GA spectrum is also typically interpreted as a transient intermediate or species. In an LDM, this would correspond to time-synchronous features (highlighted in Fig. 3 by horizontal dashed yellow lines). Next to panels (A) and (B) in Fig. 3, the mainly populated photocycle species are also shown in their respective time ranges. A negative (blue) feature in Fig. 3 corresponds to an increase in spectral amplitude at a particular lifetime and a positive (red) feature to a disappearance. In this study, GA (using a sequential model as depicted in the photocycle; see Fig. 1) is used to determine the globally occurring time constants and LDA to extract details (both in time and in wavenumber) that may not have been picked up by the GA.
Spectral changes and lifetimes resulting from the GA of the fs timescale data [Fig. S2(d)] largely coincide with the appearance of LDM features in Fig. 3(b) (marked by the dashed horizontal lines). The same is valid for the μs data [Figs. S2(a) and 3(a)]. The quality of the fits as well as their residuals are shown in Figs. S5–S7. Figure 3(c) enlarges and zooms into the delays larger than a microsecond. Notably, peak lifetimes appearing around 200, 300, and 500 µs are observed in the few hundred microseconds region, where the GA only extracts a single lifetime of 386 µs [see Fig. S2(a)]. The GA lifetime thus represents the approximate lifetime of all smaller spectral changes that are revealed by the LDM in the same time range combined. Below, it will be shown that different peak times revealed by the LDM are indeed significant and they reveal important mechanistic details of PYP’s photocycle using the extensive body of available literature on model compounds as well as the protein.
A. Early photocycle transitions up to ns: From pG* to pR
After absorption of a visible photon, the pCA chromophore in pG is excited and the protein enters the photocycle (see Fig. 1). The data and its interpretation are presented in chronological fashion, starting by the fs experiment [Fig. 2(a)] that probes the photocycle almost up to the red-absorbing pR intermediate. The measured spectral window is insightful for proteins in general and PYP in particular, as it contains the νC = C/νC = O modes from the chromophore as well as those from protein amide I vibrations.
First, PYP’s electronically excited pG* state is formed, marked by the upshift of the CO of E46 [from 1725 to 1742 cm−1, see Fig. 2(a)]. This shift was assigned to the weakening of the hydrogen-bond between E46 and the phenolate ring of the chromophore.40,41 Next, the chromophore isomerizes from trans to cis and the H-bond between the chromophore’s carbonyl and the protein backbone is broken (it upshifts by about 20 to 1665 cm−1), forming the I0 intermediate in ∼2 ps.10,40–48 Failure to break this H-bond leads to reformation of pG via a ground state intermediate (GSI).31,43 The I0 intermediate then proceeds to pR in 1–3 ns due to relaxation processes of the chromophore and its H-bond network.30,31,41,43,49–51
In Fig. 3(b), most features of the 1, 2, and 34 ps lifetimes predominantly show GS recovery [see also Fig. 2(a)], evident by the red features. The slowest 34 ps lifetime appears long for an excited state lifetime; however, it has been observed in previous UV/vis studies as well.43,52 The three lifetimes thus correspond to three excited states that return (via a GSI31,43) to pG or enter the photocycle to form I0. One characteristic excited state feature is the decay of the upshifted E46 absorption at 1742 cm−1 that is assigned to its COOD group.40 In addition to pG recovery, a part of the excited state population successfully enters the photocycle and forms I0, evident by the persistent bleach of E46 at 1725 cm−1 [see Fig. 2(a)], accompanied by an increase of the bleach around 1610 cm−1, which is attributed to the disappearing C=C stretch mode of the trans chromophore, and thus indicates that the chromophore has isomerized. In addition, an upshift from 1635 to 1640 cm−1 to a broad feature around 1660 cm−1 is observed. This shift is assigned to pCA’s C=O vibration sensing the hydrogen-bond rupture between the chromophore and the backbone at C69.41
The fs LDM in Fig. 3(b) shows small features in the amide I region at 131 ps (centered around 1630 cm−1). Previously, the I0 → I0‡ transition was only found to be associated with a change in molar extinction coefficient in the visible spectral region (without changing its absorption maximum) on a similar timescale and suggested to be caused by protein relaxation in the vicinity of the chromophore.30 Other studies have not detected this transition.40,43 The slowest LDM features in the fs data correspond to ns formation of the next intermediate pR.
B. Multiple chromophore C=O conformations in pR
On a nanosecond timescale, the pR intermediate is formed (see Fig. 1), which is suggested to consist of multiple pCA conformations.53–56 Experimental evidence comes, for instance, from a pH-dependent study that reports two slightly different C=O environments of the deprotonated chromophore,53 manifested by two resonance Raman bands at 1655 and 1672 cm−1 in pR, which evolve from a single band at 1631 cm−1 in pG.57 The two states consist of pR1 (corresponding to the low wavenumber band at low pH), having only one intact H-bond to Y42, and pR2 (the high wavenumber band at high pH) featuring two intact H-bonds between the chromophore and the two nearby E46 and Y42 residues (see the chemical structures in Fig. 1). Ultimately, both forms lead to protonation of the chromophore and the formation of pB.
All fs-to-ns difference spectra in Fig. 2 consistently exhibit the main feature with a shoulder between 1655 and 1680 cm−1, which may be assigned to a simultaneous formation of the two mentioned pR conformations. Therefore, pR is represented by a single intermediate in Fig. 1.
On the few hundred microseconds timescale, only the high wavenumber feature around 1670 cm−1 is observed to increase [the broad blue feature in Fig. 3(c)]. This feature corresponds to a C=O of the chromophore with two hydrogen bonds at the phenol oxygen and is associated with the 386 µs lifetime in the GA [Fig. S2(e)]. At the same time, the bleach of the COOD group of E46 intensifies [the broad red feature around 1720 cm−1 in Fig. 3(a)], implying a correlated change/movement of the (C=O of the) chromophore and a disappearance of the hydrogen bond between the phenolate oxygen and E46. The low wavenumber pR1 features do, however, not change.
Based on the assignments of the pH-dependent Raman experiments, the asymmetric feature observed in the 1655–1680 cm−1 region of the IR data presented here thus provides evidence for structural heterogeneity to be present already on an ultrafast timescale and leads to the branched formation of two pR states. Because these states continue to evolve on a timescale that is beyond our experiments (see Sec. III D), they are represented by a single species in Fig. 1. Structural heterogeneity in PYP has previously been observed in cryogenic trapping experiments as well as at room temperature52,58,59 and can likely be attributed to alterations of the chromophore binding pocket.60 These alterations only have a small influence on the electronic transition of the chromophore, but it has been shown that it is even possible to preferentially populate a subpopulation by tuning the excitation wavelength (i.e., implying structural heterogeneity in pG).52 A direct comparison of, for instance, lifetimes obtained in this study (using 468 nm excitation) with lifetimes resulting from other studies (for instance, below the absorption maximum of 445 nm) may, therefore, require some degree of caution. Both pR states are found to evolve simultaneously, consistent with FTIR and scattering studies, where both are observed to lead to pB′.17,53–55
C. Photocycle transitions from pR to pB′
The photocycle transition from pR to pB′ occurs on a few hundreds μs timescale and is characterized by a distinct color change due to the protonation of the chromophore.13,14 The literature does, however, not agree on the proton donor as it is assigned either to the E46 residue that is H-bonded to the phenolic oxygen of pCA in pG or to a water molecule.15–20
The measured spectral region contains protonation state marker bands of both E46 and pCA (but not of water). If E46 were the proton donor, the feature assigned to protonated E46 would concomitantly disappear with the appearance of the protonated chromophore. If water (or another residue) was the donor, features do not have to be synchronous.
D. Deprotonation of E46 on an early microsecond timescale
The LDA of the μs data shows pronounced spectral changes in the microsecond time window. Remarkably, the GA reveals three μs components [4, 61, and 386 µs; see Fig. S2(a)], while the LDA indicates that the GA components actually consist of more close-lying lifetimes [see the yellow dashed lines in Fig. 3(c)]. Importantly, the LDM peak distribution lifetimes can be used to construct a detailed schematic picture of several occurring protonation events. As mentioned above, the model-free nature of the LDA does, however, not provide clues on their interconnectivities.
The shortest μs component is apparently not associated with (one or more of) the pR states, as no feature is observed that can be assigned to a C=O mode of pCA (it shifts from 1631 cm−1 in pG to 1655 and 1672 cm−1 in pR57) but only shows two time-correlated features around 1740 cm−1 (red) and 1592 cm−1 (blue) in Fig. 3(c) (earliest yellow dashed line). The high wavenumber feature belongs to the upshifted C=O mode of E46 (see above) and resulted from weakening of the hydrogen-bond between E46 and the phenol group of pCA. Striking is, however, that the disappearance of the feature at 1740 cm−1 does not lead to recovery of the C=O mode around 1725 cm−1. Instead, we observe a correlated appearance of the 1592 cm−1 feature, which most likely arises from the deprotonated COO− moiety of E46. Such a deprotonation event leads to features below 1600 cm−1 due to bond order reduction and produces two asymmetric and symmetric COO− stretch modes around 1400 cm−1 and around 1560–1600 cm−1, respectively.61–63 The exact wavenumber of the high wavenumber mode has been shown to depend on the O–C–O angle, the number of H-bonding partners, as well as on the H-bonding strength.64 For acetate in H2O, for instance, shifts by about 80 cm−1 for bidentate H-bonds are observed for the high wavenumber band. The 1592 cm−1 feature is thus consistent with a deprotonated E46 residue. Protonation of the chromophore on the early μs timescale can also be excluded, as the expected changes of other chromophore modes are not detected.16 Fig. 4 shows a schematic overview of the observed microsecond components, with the 4 µs process depicted in gray.
The next 14 µs component shows similar features as those of the 4 µs component, but the increase in the feature near 1626 cm−1 makes the assignment less straightforward. Inserting an isotope at the C=O of E46 should shift the mentioned features and could thus lead to an unambiguous assignment. The next 60–70 µs lifetime exhibits red and blue features around 1650–1675 and 1638 cm−1, respectively [Fig. 3(c)]. It is, therefore, assigned to pG recovery from (one or more) pR (states). The 1638 cm−1 wavenumber is most likely shifted with respect to pG (1631 cm−1) due to spectral overlap with the positive red feature around 1626 cm−1.
The μs data thus point to the deprotonation of (a fraction of) E46 already on an early microsecond timescale, originating from the upshifted E46 population. However, what is the destination of the proton? As mentioned above, the proton does not arrive at the chromophore. As there are no other early microsecond features in the measured spectral region, its destination is unclear. It could be below the available signal-to-noise ratio, or it could go to a water molecule near the chromophore binding pocket. Eventually, the chromophore does, however, become protonated, but this happens on a few hundreds of microseconds timescale (see below).
E. Movement of E46 on a 175 µs timescale
Before the chromophore is protonated, the rich dynamics of the C=O mode of E46 already reveal changes in and around the chromophore binding pocket. Upon the formation of pR on a nanosecond timescale, the C=O mode was previously seen to downshift by about 9 from 1728 cm−1, indicating a strengthening of the H-bond to the chromophore.16,17 In addition to the downshifted population, the SADS of pR also has an upshifted feature.41,65 This upshift to 1740 cm−1 is best seen in the blue 2 ns SADS in Fig. S2(e) but also visible in the LDM of Fig. 3(a). The presence of the two positive features may indicate a broadening of the pG absorption around 1728 cm−1 and could also represent two populations. Upon closer inspection, the two features essentially follow the dynamics of the (two) mentioned pR states (i.e., from their formation up to the nanosecond timescale). The next (red) 60–70 µs features in Fig. 3(c) show partial recovery of pCA C=O features around 1650–1680 cm−1 and might reflect recovery to pG from pR. Above, we have already seen that the upshifted E46 population (in pR) deprotonates on a 4 µs timescale. In the GA, the remaining downshifted population disappears with a 61 µs lifetime, concomitant with a rising feature around 1750 cm−1 [the vertically elongated blue feature in Fig. 3(c)]. This upshift has previously been interpreted as the movement of E46 into a more hydrophobic environment,16,17 after originally having a strengthened H-bond to the chromophore (the downshifted fraction). The LDA shows a similar picture, although the peak time of the 1750 cm−1 absorption at 175 µs has most likely its red (disappearing) counterpart superimposed on the broad red feature around 1720 cm−1 around 300 µs. This broad red feature at 300 µs indicates a disappearance of the protonated E46 population altogether (it deprotonates; see below for more details), as it does not have a time-synchronous corresponding blue feature above 1700 cm−1. In other words, multiple transient microsecond states of E46 are thus formed during the photocycle. Figure 4 provides a pictorial view on these micro-transitions during pR to pB′. One pR fraction deprotonates on a 4 µs timescale, while a second fraction remains protonated but moves into a more hydrophobic environment with a lifetime of 175 µs [and subsequentially disappears with a 2 ms lifetime according to Fig. S2(a)]. A third fraction deprotonates with a lifetime of ∼300 µs. As the LDA is inherently a model-free method, potential connectivities between the fractions can only be inferred from a synchronous appearance/disappearance of features (e.g., it may or may not be that only the upshifted E46 population deprotonates). The relative time of appearance of each process can, however, be confidently extracted.
These observations are largely consistent with a previous step-scan FTIR study of Brudler et al., where the spectral integrals of the bleach and upshifted E46 features are compared (in H2O; our data are measured in D2O).17 By Brudler et al., it was concluded that the pB formation follows a branched model with only a quarter of the photo-excited molecules populating the upshifted band in 113 µs (in H2O) and moving E46 into a more hydrophobic pocket. This population subsequently deprotonates on a millisecond timescale (which is beyond our accessible timescale but partially resolved by the GA, nonetheless). The remaining major population was found to deprotonate directly in 113 µs (in H2O, we find 300 µs in D2O), followed by overall structural changes on a millisecond timescale. This would correspond to pR2. The other populations were not observed in previous studies probably because their fractions are relatively small. As expected, deuteration thus has a larger impact on the deprotonation than on the movement of E46. We also performed the experiment in H2O and found only one 105 µs component in the GA (not shown). The corresponding LDM also reveals that the upshift occurs prior to an increase in the bleach.
F. Chromophore protonation
The next important question arises: where does the proton of E46 go to? It is established that the chromophore becomes protonated on a few hundreds of microseconds timescale.9,15–17,54,66 The LDA shows that the proton has apparently not arrived yet at the chromophore on the 100 µs timescale, as no induced absorption bands assigned to a protonated chromophore are found below 1600 cm–1 [only disappearing red features are seen in Fig. 3(c); see below for assignments of protonation markers]. As we observe a fraction of the E46 population deprotonating on a 4 µs timescale, the proton is apparently not directly transferred to the chromophore. Transfer via a nearby water molecule and/or via another amino acid to the chromophore is conceivable. Alternatively, E46’s proton could go into the solvent, which is in our view the most likely scenario because, otherwise, the proton needs to be “trapped” near the chromophore for hundreds of microseconds. In that case, one or more amino acids in or near the chromophore binding pocket would likely have sensed its presence and given rise to a signal in the measured spectral window.
We found that the second population of E46 deprotonates on a 300 µs timescale, which is in a similar time range where the chromophore is becoming protonated9,15–17,54,66 and which is also the timescale on which only one of the pR states evolves [i.e., pR2, see above, implying a correlated movement of the (C=O of the) chromophore]. Because pR2 is previously assigned to a chromophore configuration with two H-bonds donated to its phenolic oxygen (with Y42 and E46, see above), it would be consistent with E46 being the chromophore’s proton donor.
The appearance of the protonated pCA is reported to be accompanied by several characteristic spectral changes, i.e., the coupled νC-C/νC = C stretching mode at 1555 cm−1 in pG becomes a doublet at 1559/1586 cm−1 in pB′ and 1562/1586 cm−1 in pB (in D2O).15,67 Importantly, the sign of the protonated pCA feature at 1566 cm−1 [blue in Fig. 3(c)] and the sign of the deprotonated E46 residue at 1720 cm−1 (red) around 300 µs are opposite. At first sight, these changes imply that the protonation of the chromophore occurs on the same timescale as that of the deprotonation of E46. A closer look at the single pixel time courses, however, shows that they are not synchronous [see Fig. 5(a)]. The single pixel time course at 1566 cm−1 (black markers) is clearly slower than the evolution at 1728 cm−1 (blue markers). A two-exponential fit at 1566 cm−1 requires a small long-lived component and a dominant shorter 476 µs component (black line). The inclusion of a fixed 10 ms lifetime leads to a small improvement of the fit at long delays with respect to a single exponential fit [see Fig. S8(a)]. Nonetheless, no other microsecond lifetimes are required for a satisfactory fit, indicating the absence of contributions of other spectrally overlapping features. Such overlap hampers, for instance, an unambiguous assignment of the deprotonated COO− group of E46 on the hundreds of microseconds timescale (it appears at 1592 cm−1 around 4 µs, see above), as other chromophore and protein bands may appear in the same spectral region as well and could, therefore, potentially lead to multi-exponential time courses.
The asynchronous time evolution shown in Fig. 5(a) is also apparent in the LDM of Fig. 3(c). On the same hundreds of microseconds timescale, there are actually three (induced) blue features below 1610 cm−1 around 440 µs, all of them appearing with a lifetime that is delayed with respect to the (disappearing) red feature at 1720 cm−1 around 330 µs. This delay is also evident by different peak distribution lifetimes in Fig. 5(b), where slices through the LDM at 1566 and 1728 cm−1 are compared. The corresponding residuals of the LDA are shown in Fig. S8(b), showing some degree of misfit for both pixels at the longest delays. Although the standardization of the data allows a single shared regularization factor using the elbow-criterion, modest improvements in the fit are possible by lowering the regularization factor from 0.6 [as used in Figs. 3(a) and 5(b)] to 0.15 [see Figs. S8(c)–S8(e)]. However, the LDA peak lifetimes still do not coincide in Fig. S8(e). The small lifetime peaks in both curves preceding the main features below 100 µs in Fig. 5(b) are considered artifacts from the LDA, as they depend on the choice of regularization factor [compare to Fig. S8(e)] and similar lifetimes are not found using exponential fitting [see Fig. 5(a)].
Although the lifetimes resulting from the exponential fits and the LDA are not identical, both consistently show that the dynamics of the two pixels are not synchronous and thus imply that E46 cannot be the proton donor of pCA on the hundreds of microseconds timescale. It is noted that the GA detects a single 386 µs time constant, being about the average of the pCA and E46 lifetimes found in the analyses presented in Fig. 5. Although the GA lifetimes in Fig. S2(a) are close or even beyond the available maximum timescale of 750 µs, the chromophore protonation lifetime is consistent with the 100–400 µs time constants that are reported in D2O as well as H2O in literature.9,15–17,54,66 It is also noteworthy that the three about 175, 300, and 450 µs peak times that are seen in the LDM are all within or near the reported range of lifetimes. Disregarding differences in sample preparation and measurement conditions, the spread may be also explained by the used experimental methods, as each technique has different observables. For instance, resonance Raman spectroscopy only probes the chromophore and not the surrounding amino acid residues, while IR spectroscopy senses both. The lifetimes measured with different methods may thus not correspond to transitions between real stable transient species (i.e., where two species are separated by a single energy barrier) but possibly to transitions between micro-intermediates. While the system is evolving through all these intermediates, certain methods are only sensitive to a part of the changes.
G. The chromophore’s proton donor is not E46 but likely water
Therefore, if E46 is not the proton donor, there must be a different one. Additional evidence for the presence of an alternative proton source comes from mutation studies. In the E46Q mutant, for instance, protonation of the chromophore occurs on a fivefold faster timescale than in the wild type,68 even though E46Q cannot donate a proton.16 The photocycles (at neutral pH) of both the mutant and the wild-type proteins are, however, similarly fast,68 from which it can be concluded that E46 does not necessarily needs to be the proton donor.42 This may thus imply the existence of multiple or branched pathways leading to the protonation of the chromophore. Additional pH-dependent solvent deuteration studies report either the involvement of a water molecule in the proton transfer process or that of an equilibrium model involving the chromophore and E46.69 When closely inspecting the data from a previous step-scan FTIR study,16 where the protonation of the chromophore was reported to directly correlate with the time of ionization of E46, an earlier appearance of the E46 feature may actually also be visible (it appears to be systematically shifted to earlier times, although it was interpreted to have the same time constant). Data from that study and our work are thus consistent. Finally, a previous extensive review of static and transient data20 also concludes that the solvent is the most likely proton donor for the chromophore, but direct (spectral) evidence was lacking so far.
Our data clearly reveal that the chromophore’s proton cannot directly originate from the nearby E46 residue. It is also important to mention here that the observed relative delay between the ionization and protonation markers is near impossible to extract via the widely applied GA method, as each spectrum predominantly embodies the largest spectral changes occurring during a relatively narrow time window. A SADS essentially represents the sum of many contributions that do not necessarily appear or disappear at exactly the same time. Such a view would be consistent with a photocycle following not a smooth but a more rugged potential energy landscape, involving multiple small steps, as envisioned to occur during protein folding.70,71
H. Photocycle transitions beyond pB′
The μs data report on the photocycle up to 750 µs and probe the pR → pB′ → pB transitions. The pB intermediate is PYP’s signaling state, which can also be photo-accumulated in an FTIR spectrometer (see Fig. S9). This spectrum resembles the long-lived SADS of the μs data, which in turn is consistent with the ms pB-pG literature spectrum.16 Although the spectra shown in Fig. S9 are not identical below 1610 cm−1, both the SADS and FTIR data are entirely positive in the 1565–1610 cm−1 region and both show the fully developed and almost equal-sized bleaches around 1690 and 1725 cm−1. In addition, the amide I modes in both spectra are observed to be especially prominent,16 which is only the case for pB-pG.16,67,72 The features near the end of the timescale in the LDM of Fig. 3(b) seem to correspond to the sum of the two slowest SADS [i.e., the orange 2 ms and red long-lived spectra in Fig. S2(a)]. In other words, the slowest LDM features represent a mixture of the pB′/pB states. In contrast, the GA is able to separate the two states spectrally. The 2 ms SADS shows that the upshifted E46 band at 1752 cm−1 disappears (it even turns negative) and the protonated chromophore marker band near 1562 cm−1 rises further. The GA thus indicates that the moved E46 population disappears at the same time the chromophore protonates. Only experiments that include the final part of the photocycle will reveal if the two processes are exactly synchronous or not, but these results are consistent with the dynamics of the pR1 state.
The presented time-resolved IR difference spectra of PYP reveal a detailed picture of its light-induced structural changes. The many orders in the magnitude of the probed timescales and the large number of structure-sensitive bands provided by IR spectroscopy require data reduction methods that assist their interpretation. The presented analysis allows us to identify spectral correlations in time that shed new light on the chromophore’s protonation event on a few hundreds of microseconds timescale and involve only one of the pR states (i.e., pR2). On that timescale, it is surprisingly found that the proton donor of the chromophore cannot be the nearby E46 residue that donates an H-bond to the phenolic group of the chromophore in pG. It is more likely that the proton comes from a water molecule or potentially from another residue, as the ionization of E46 precedes the chromophore’s protonation by hundreds of microseconds. A consistent picture now emerges from the applied data analysis methods, which explains previously apparently contradicting interpretations of data obtained using different observables (Raman vs IR, for instance).
The additional details the LDA method provides not only show that the photocycle of PYP is incredibly complex but also invites us to rethink the typical view on transitions between “stable” structural intermediates. As discussed here for PYP, such transitions may often consist of multiple smaller structural changes that all occur in a relative short time window. Different experimental techniques, which use different observables, may be sensitive to one or more of these micro-transitions, explaining why the time constant assigned to the overall transition often depends on the experimental technique or the particular signal the analysis is focusing on (which would not be the case for a transition over a single high barrier between intermediates). Introduction of additional observables via IR labels73,74 that probe different parts of the structure has great potential, in particular, in combination with LDA, to reveal and analyze such heterogeneities, leading to a more complete understanding of the nature of transitions and intermediates in protein photocycles and kinetics in general.
See the supplementary material for raw time-resolved and steady-state data and more details on the data analysis methods and their associated figures.
We greatly acknowledge the help of Manuel Pescher of our institute and Erwin Köhler and Stefan Schreiber of the Electronics Workshop of the Institute of Nuclear Physics in building the laser synchronization electronics. We also thank Sabrina Oesteritz and Hans-Werner Müller for their help during protein expression and purification, the group of Reinhard Dörner for lending us a high-resolution oscilloscope, and Chavdar Slavov for his help with the LDA. J.B. thanks the Alexander von Humboldt Foundation for the Sofia Kovalevskaja Award. The authors thank the Deutsche Forschungsgemeinschaft for funding (Grant Nos. INST 161/722-1 FUGG and 466145756).
Conflict of Interest
The authors have no conflicts to disclose.
L.J.G.W.v.W. and L.B. contributed equally to this work.
The data that support the findings of this study are available from the corresponding author upon reasonable request.