Time-resolved serial crystallography captures high-resolution intermediates of photoactive yellow protein

See allHide authors and affiliations

Science  05 Dec 2014:
Vol. 346, Issue 6214, pp. 1242-1246
DOI: 10.1126/science.1259357


Serial femtosecond crystallography using ultrashort pulses from x-ray free electron lasers (XFELs) enables studies of the light-triggered dynamics of biomolecules. We used microcrystals of photoactive yellow protein (a bacterial blue light photoreceptor) as a model system and obtained high-resolution, time-resolved difference electron density maps of excellent quality with strong features; these allowed the determination of structures of reaction intermediates to a resolution of 1.6 angstroms. Our results open the way to the study of reversible and nonreversible biological reactions on time scales as short as femtoseconds under conditions that maximize the extent of reaction initiation throughout the crystal.

Watching a protein molecule in motion

X-ray crystallography has yielded beautiful high-resolution images that give insight into how proteins function. However, these represent static snapshots of what are often dynamic processes. For photosensitive molecules, time-resolved crystallography at a traditional synchrotron source provides a method to follow structural changes with a time resolution of about 100 ps. X-ray free electron lasers (XFELs) open the possibility of performing time-resolved experiments on time scales as short as femtoseconds. Tenboer et al. used XFELs to study the light-triggered dynamics of photoactive yellow protein. Electron density maps of high quality were obtained 10 ns and 1 µs after initiating the reaction. At 1 µs, two intermediates revealed previously unidentified structural changes.

Science, this issue p. 1242

X-ray structure analysis has successfully determined high-resolution structures of more than 100,000 proteins and nucleic acids. But these structures represent static pictures of the biomolecules, which during their reactions engage in rapid dynamic motion. Time-resolved macromolecular crystallography (TRX) (1) unifies structure determination with protein kinetics, as both can be determined from the same set of data (2, 3). TRX is traditionally performed using pump-probe experiments and the Laue method at a synchrotron source, in which light-sensitive molecules within a crystal at near-physiological temperature are illuminated by a laser pump pulse to initiate their reaction, followed by a polychromatic probe x-ray pulse. These experiments rely on the exceptional stability of synchrotron sources to measure small, time-dependent differences between diffraction patterns with and without the pump laser pulse. Synchrotron-based Laue diffraction experiments are currently restricted by the x-ray beam brilliance to strongly scattering, relatively large (typically 6 × 105 μm3) crystals, whose optical density makes high (>25%), uniform reaction initiation difficult. Further, the time resolution is limited to ~100 ps by the duration of the probe x-ray pulse. However, difference electron density (DED) maps from synchrotron-based TRX experiments have revealed that large structural changes occur in times shorter than 100 ps (47). Important structural changes associated with key chemical processes such as isomerization evidently occur in the range of femtoseconds to tens of picoseconds, inaccessible to synchrotron experiments. The advent of free electron lasers such as the Linac Coherent Light Source (LCLS) and the SPring-8 Angstrom Compact Free-Electron Laser (SACLA) has opened a new avenue for ultrafast time-resolved structural studies. These lasers emit femtosecond pulses of hard x-rays whose peak brilliance is higher than that available at the most advanced synchrotrons by a factor of 109.

The method of serial femtosecond crystallography (SFX) (8) has opened new opportunities for time-resolved structural studies (9, 10). In SFX, a stream of micro- or nanocrystals in their mother liquor at near-physiological temperature is delivered by a liquid jet injector (11) to the x-ray interaction region, where the diffraction pattern of a single tiny crystal is recorded by illuminating the jet with an individual x-ray pulse from the x-ray free electron laser (XFEL). Diffraction patterns are obtained rapidly (e.g., at 120 Hz) at the LCLS. Although enormous x-ray doses, up to 1000 times the room-temperature synchrotron “safe dose” (12), are deposited in the crystal by the femtosecond x-ray pulse, the processes that lead to destruction are sufficiently slow that the crystals diffract before they are destroyed (8, 13, 14). Structures are solved using thousands of diffraction patterns of individual crystals; these patterns extend to near-atomic resolution (15, 16). To conduct a time-resolved SFX (TR-SFX) experiment at the XFEL with femtosecond time resolution, a reaction must be initiated in a light-sensitive crystal by a femtosecond laser pump pulse, then probed after a time delay Δt by a femtosecond x-ray probe pulse (9, 17).

TR-SFX is challenging because of the very different properties of the x-ray pulses emitted by synchrotrons and by XFELs (10, 18). Time-resolved synchrotron studies take advantage of an x-ray beam with exceptional stability, where ideally a data set is collected on one large single crystal at essentially constant beam energy, bandwidth, photon flux, and volume of the crystal exposed to the x-rays. The resulting data consist of sets of consecutive light and dark images collected at the same orientation from the large single crystal. This consistency of data acquisition is important, as structure factor changes between the light and dark states are often very small. By contrast, several inherent pulse-to-pulse variations make TR-SFX at atomic resolution challenging: (i) The XFEL photon flux per pulse can vary by up to an order of magnitude; (ii) the peak energy and spectral content of the x-ray beam changes from pulse to pulse; and (iii) the crystal size is variable, and even if it were constant, the volume of the crystal interacting with the beam can change. These factors give rise to large fluctuations in the diffracted intensities. However, the resulting total error is inversely proportional to the square root of the number of diffraction patterns (18), and by collecting diffraction patterns from a large number of tiny crystals, high-quality x-ray data can be obtained (15, 16, 19).

Despite these challenges, TR-SFX offers several advantages over time-resolved Laue crystallography at a synchrotron: (i) Time resolution, largely set by the duration of the x-ray pulse in the femtosecond time range, is substantially higher; (ii) the diffraction-before-destruction principle overcomes the x-ray damage problem; (iii) each crystal diffracts only once and crystals are rapidly exchanged, which provides an easy way to address irreversible processes; (iv) the quasi-monochromatic FEL x-ray beam allows investigation of crystals with large unit cells; (v) diffraction patterns are less sensitive to crystal mosaicity than in the Laue method; and (vi) the small size of the crystals (often <10 μm) allows more uniform laser initiation of the reaction of the molecules in the crystal. These advantages were exploited in the first TR-SFX studies of the large protein complexes, the photosystem I–ferredoxin complex and photosystem II, as model systems (20, 21). Structural changes were recently discovered in a pump-probe TR-SFX study of the water-splitting complex in photosystem II at 5.5 Å resolution (22), but their interpretation remains provisional, in part because of the limited resolution.

We demonstrate that high-resolution, interpretable DED maps can be determined by applying TR-SFX to a well-studied model system: the bacterial blue light photoreceptor PYP (photoactive yellow protein). Upon absorption of a blue photon, PYP enters a photocycle in which numerous intermediates are occupied on time scales from femtoseconds to seconds (Fig. 1A) (2325). Structural changes on time scales longer than 100 ps have been investigated to high resolution by time-resolved crystallography using the Laue method at synchrotrons (4, 5, 26). Even barriers of activation have now been determined solely from temperature-dependent time-resolved x-ray data (3). The photocycle examined by time-resolved crystallography contains six intermediates denoted IT, ICT, pR1, pR2, pB1, and pB2. The strongest features appear in the DED maps when pR1 and pR2 are substantially populated, because the sulfur of Cys69 to which the chromophore is covalently attached is considerably displaced in both these intermediates. This occurs between ~200 ns and 100 μs (Fig. 1B). For our TR-SFX experiments, we selected a delay time of 1 μs in order to clearly reveal these features. A second data set was collected at a delay of 10 ns, where three distinct intermediates are substantially populated whose interpretation requires high-quality data. The DED maps were calculated from x-ray data analyzed by Monte Carlo integration over a large number of diffraction patterns, each of which is subject to all the stochastic fluctuations outlined above (18). Comparison with synchrotron Laue data established that the DED maps determined by the two methods are very similar. These findings open the way to high-resolution TR-SFX studies of light-driven processes and, by extension, to reversible and nonreversible reactions that may be initiated by other methods.

Fig. 1 Simplified PYP photocycle.

The cycle is shown from the perspective of a time-resolved crystallographer (3, 4, 26). The dark state pG is activated by absorption of a blue photon (450 nm) to pG* that begins the photocycle. The crystal structures of longer-lived intermediate states IT, ICT, pR1 (pRE46Q), pR2 (pRcw), pB1, and pB2 are known.

We collected TR-SFX data for the dark and excited states in an interleaved mode by using two 20-Hz nanosecond lasers to initiate the reaction. Every third x-ray pulse probed a laser-excited crystal, resulting in a light-dark-dark scheme (fig. S1). The time-resolved data sets were obtained for pump-probe delay times of 10 ns and 1 μs, at a laser pulse energy of 15 μJ focused to a 150-μm beam diameter at the sample (~800 μJ/mm2). The microcrystal hit rate varied between 1% and 18%, and 60% of the resulting diffraction patterns were successfully indexed (table S1). Data quality, as judged by the R-split values, progressively improved with the number of indexed patterns (table S2 and fig. S4). R-split values for the dark data approached the overall value of 6.5% for 65,000 indexed patterns, remained below 10% to ~2 Å resolution, and was still acceptable (22.4%) at the highest resolution of 1.6 Å. Because the data collection scheme generated twice as many dark patterns as light patterns, the light data showed a somewhat larger value of R-split (9.2% for 32,000 indexed patterns). DED maps were calculated from weighted difference structure factor amplitudes (2) with phases derived from a structural model of PYP refined against the x-ray FEL dark data (see supplementary materials). The initial model was obtained from Protein Data Bank (27) entry 2PHY.

The results are shown in Figs. 2 and 3. Monte Carlo integration robustly determines intensities for the strongly diffracting PYP crystals. Synchrotron-quality DED maps with positive and negative peaks around 10 standard deviations σ (3) are obtained with only 4000 indexed light patterns (see supplementary materials). Because the unit cell of PYP crystals is relatively small and the crystals are a few micrometers in size, intensity fringes between Bragg reflections caused by the shape transform of the crystals (28) are absent. In addition, the mosaicity of PYP crystals is extremely small, so that many reflections may be collected almost as full reflections. As a result, sufficiently accurate intensities are obtained from fewer diffraction patterns. Strong DED features can already be observed at very large R-split values (table S2) corresponding to higher experimental noise in the structure amplitudes. If the noise in the amplitudes is too high, DED features deteriorate (29, 30). To extract faint DED features associated with small structural changes of PYP, diffraction patterns should preferably be collected with a laser-on, laser-off sequence to accumulate an equal number of light and dark patterns, aiming for redundancies on the order of 1500 in the highest-resolution shell.

Fig. 2 Stereo view of the light-dark 1.6 Å difference electron density map at 1-μs time delay, superimposed on the dark PYP structure (cyan).

Contour levels: red/blue –3σ/+3σ. Chromophore and some important chromophore pocket residues are shown in yellow and marked in (A). (A) Red arrow: Plume of structural displacements extends to Met18, close to the N-terminal helix, which may be strongly displaced at longer times (26, 34). (B) View rotated by ~90°. A large part of the molecule does not display sizable DED features and remains structurally unaltered.

Fig. 3 Comparison of electron density and DED maps in the chromophore pocket obtained by TR-SFX and the Laue method.

The dark state is shown in yellow in all maps. (A and D) Electron density maps for the PYP dark state obtained with TR-SFX and Laue, respectively (contour level 1.1σ, 1.6 Å resolution). The PCA chromophore and nearby residues are marked in (A). Arrow: Double bond in the chromophore about which isomerization occurs. (B) TR-SFX DED map at 10 ns. Light green structure: ICT intermediate. Features marked by dotted arrows belong to additional intermediates not shown. (C) TR-SFX DED map at 1 μs. Pink and red structures: structures of pR1 and pR2 intermediates, respectively. (E) Laue 32-ns DED map correlates best to the TR-SFX 10-ns map. (F) Laue 1-μs DED map. Contour levels of the DED maps: red/white –3σ/–4σ, blue/cyan +3σ/+5σ, except for (C) where cyan is +7σ. See fig. S8 for stereo versions of (B) and (C).

Figure 2 shows an overview of the 1-μs light-minus-dark DED map superimposed on the refined PYP dark structure. The largest DED peak is located on the sulfur of the chromophore (Fig. 2A) and has a highly significant negative value of –22σ (where σ is the root mean squared value of the DED across the asymmetric unit). The largest positive DED peak (18σ) is close to the same sulfur. These features reveal substantial displacement of the sulfur 1 μs after laser excitation, associated with a change in chromophore configuration from trans to cis. Chromophore isomerization triggers further protein conformational changes extending to the periphery of the protein (red arrow). The positive (blue) and negative (red) DED peaks are contiguous and can be interpreted in terms of atomic models (Fig. 3, B and C). When the resolution is reduced below 3 Å, the difference signal disappears (fig. S7) and clear interpretation of structural changes in PYP becomes impossible. We note, however, that the minimum resolution needed to observe structural changes is likely system-dependent, and an extension to other systems must await further work.

In Fig. 3, A and D, the electron densities of the dark PYP structure derived from SFX are compared with those from the Laue method. The maps are of comparable quality and can be interpreted by the same reference structure (yellow). Shown in Fig. 3, B and C, are DED maps obtained at time delays of 10 ns and 1 μs from TR-SFX data. The DED maps are initially interpretable with intermediate structures derived from our earlier Laue experiments at BioCARS beamline 14-ID at the Advanced Photon Source (3, 4, 26). Negative DED features in both panels are accounted for by the yellow reference (dark) structure. Each DED map arises from a mixture of intermediates. For the 10-ns map, we show here only the ICT intermediate (4) in green as a guide to the eye. Figure 3, D to F, shows corresponding maps collected by Laue diffraction at 0°C. The Laue 32-ns DED map correlates well with the 10-ns TR-SFX map. Many features are present in both the TR-SFX and Laue maps, and it is evident that they show the same mixture of intermediates. Particularly intriguing is the displacement of Glu46 caused by the initial isomerization of the chromophore, shown in both maps. Note that the mixture of intermediates, whose concentrations change rapidly around 10 ns, prevents the refinement of individual intermediate structures until a complete time series of DED maps (35, 26) becomes available. In the 1-μs map (Fig. 3C), two intermediates, pR1 and pR2, are present with substantial occupancy (orange and pink structures, respectively). Because the concentration of these intermediates does not change over several decades in time (fig. S5A), structural characterization of this mixture is possible. The refinement of this mixture with an appropriate pair of conformations reveals hitherto unidentified structural changes (compare fig. S8A with fig. S9). In the TR-SFX maps, the DED features are much more pronounced, with stronger positive and negative features that are also much better connected spatially, and thus are more readily interpretable than the Laue maps.

The fraction of molecules that entered the photocycle can be determined by fitting calculated DED maps of the two intermediates populated at the 1-μs time delay to the observed DED map (see supplementary materials). About 22% of the molecules populate the pR2 intermediate and 18% the pR1. The extent of reaction initiation by the nanosecond laser pulse is therefore 40%, versus the 10 to 15% typically achieved in synchrotron experiments (3). The higher extent of reaction initiation with the smaller crystals used in TR-SFX demonstrates one of its key advantages for time-resolved studies and provides an explanation for the higher-quality DED maps.

In a time-resolved Laue experiment on a large single crystal of PYP, 3 to 10 complete pump-probe sequences, with waiting times of a few seconds between each sequence to allow completion of the photocycle, are necessary to accumulate a sufficiently exposed diffraction pattern before detector readout. When the nanosecond laser is used for reaction initiation, the laser beam is usually smaller than the crystal size and does not penetrate fully and uniformly through the large, optically dense crystals. This localized application of the repeated, intense laser pulses induces transient strain in the crystals, which results in radially streaked Laue spots. Strain imposes an upper limit on the useful laser energy, which for PYP crystals is around 4 to 5 mJ/mm2 at 485 nm. This sets a practical limit on the extent of reaction initiation. Further, the repeated laser pulses have a damaging effect on the crystals, which imposes a limit on the total number of laser pulses that a crystal can tolerate (31).

Neither limit applies to TR-SFX. Because the crystals are so small, the laser pulse can easily penetrate through them. For example, the absorption length at 450 nm, where the crystal is most optically dense, is ~3 μm (3). Most crystals used in this study are smaller and can be near-uniformly illuminated with little local strain. The quasi-monochromatic collimated x-ray beam results in a sharp intersection of the Ewald sphere with each reciprocal lattice point. Even if the crystal mosaicity were to increase transiently, it would only increase the extent of partiality of each spot. Indeed, radially streaked diffraction spots do not appear in our data. In TR-SFX, each microcrystal is illuminated only once by the laser pulse, which allows the laser pulse energy to be increased to otherwise unacceptable levels. For example, the 0.8 mJ/mm2 laser energy density at 450 nm used here would produce irreversible bleaching if applied in a few repetitive pulses to macroscopic crystals (32). Once excited, more than 80% of the PYP molecules return to their dark state on the picosecond time scale without entering the photocycle (24, 33). At the laser pulse power we used, each molecule in the microcrystals is matched by an average of about 24 photons. Thus, each PYP molecule has the potential to be repeatedly activated within the nanosecond pulse. Because the excited-state lifetime is ~500 fs (24), the effective photolysis yield with our nanosecond laser may in principle reach levels several times the primary quantum yield (fig. S6). The intermediates earlier in the photocycle with lifetimes up to 10 ns have red-shifted absorption maxima around 500 nm (25). Because we excite at 450 nm, the probability that one of these intermediates absorbs a second laser photon is negligible. Despite the very high nanosecond laser flux, no sign of photo-induced damage is evident in our TR-SFX diffraction patterns. As a result, excellent DED maps are obtained of the quality needed for structural interpretation.

The way is now open to study reactions with ultrafast time resolution (24). For example, using femtosecond laser pump pulses to initiate the reaction in PYP may allow time resolution in the subpicosecond regime, as recently demonstrated by wide-angle x-ray scattering experiments on a bacterial photosynthetic reaction center at LCLS (17). Reaching this time resolution would take structural biology into uncharted territory (24). PYP displays rich femtosecond chemistry, but little is known experimentally about the corresponding atomic structures and the way the elementary chemical process of isomerization proceeds. Moreover, on the femtosecond time scale, coherent phenomena may become evident that bear on how the PYP chromophore undergoes its primary photoabsorption. On the ultrafast time scale, the penetration depth under intense femtosecond optical excitation may be dominated by nonlinear cross sections while the photolysis yield becomes fundamentally limited by the primary quantum yield of the PYP chromophore. Intense optical pulses with very short duration (e.g., 10 fs) will create only small population levels, which might be substantially increased by stretching and shaping the laser pulse (24). To capture this small population, every x-ray pulse must probe a crystal—such as a microcrystal—that has been excited as completely as possible. These experiments may elucidate the elementary mechanism of light absorption in chromophores, with implications for photosynthesis, light sensing, and the process of vision.

Our results establish that high-resolution TR-SFX is readily possible. Reaction initiation by light to explore fast and ultrafast isomerization might be extended in the future to chemically triggered reactions, which would open the door to the application of high-resolution TR-SFX at x-ray laser sources to a wide range of biologically and pharmaceutically important proteins.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S10

Tables S1 to S4

References (3555)

  • * Present address: Linac Coherent Light Source, SLAC National Accelerator Laboratory, Sand Hill Road, Menlo Park, CA 94025, USA.

  • Present address: Physics Department, University of Wisconsin, Milwaukee, WI 53211, USA.

References and Notes

  1. Acknowledgments: Supported by NSF career grant 0952643 (M.S.), NIH grant R01GM095583 (P.F.), NIH grant R24GM111072 (V.S., R.H., and K.M.), and NSF Science and Technology Centers grant NSF-1231306 (“Biology with X-ray Lasers”). The work of M.F. and his team was performed in part under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under contract DE-AC52-07NA27344 and supported by LLNL Lab-Directed Research and Development Project 012-ERD-031. We thank T. White for making the newest version of CrystFEL available to us, R. G. Sierra and H. DeMirci for help setting up crystal preparation in their labs, S. Lisova for making injector nozzles, and D. Deponte for help with the injector setup. M.S. thanks R. Hovey and S. Tripathi for help on early attempts to produce microcrystals. The TR-SFX measurements were carried out at the Linac Coherent Light Source (LCLS) at SLAC National Accelerator Laboratory. LCLS is an Office of Science User Facility operated for the U.S. Department of Energy Office of Science by Stanford University. J.S. is an inventor on a patent applied for by Arizona State University that covers the gas dynamic virtual nozzle. Coordinates and (difference) structure factors are deposited in the Protein Data Bank under accession numbers 4WL9 and 4WLA.
View Abstract

Navigate This Article