Time-resolved x-ray crystallography of heme proteins

Abstract

Heme proteins, with their natural photosensitivity, are excellent systems for the application of time-resolved crystallographic methods. Ligand dissociation can be readily initiated by a short laser pulse with global structural changes probed at the atomic level by X-rays in real time. Third generation synchrotrons provide 100ps X-ray pulses of sufficient intensity for monitoring very fast processes. Successful application of such time-resolved crystallographic experiments requires that the structural changes being monitored are compatible with the crystal lattice. These techniques have permitted observing allosteric transitions in real time for a cooperative dimeric hemoglobin.

1. Introduction

Time-resolved X-ray diffraction is a unique tool for investigation of real-time structural changes that molecules undergo while performing their function. While static structures providing atomic models, now available for many macromolecules, greatly aid our understanding of molecular function, the detailed mechanism by which macromolecules function still often cannot be discerned. Time-resolved crystallography can elucidate the mechanism by providing snapshots of molecules in action. Short and intense X-ray pulses at third generation synchrotron sources are used to probe, in real time, reactions that are initiated synchronously and rapidly in molecules within the crystal. Such time-resolved studies aim to identify reaction intermediates that are often very short-lived, determine their structures, and describe the complete reaction mechanism, including reaction rates that govern the concentrations of intermediates during the reaction (Schmidt et al., 2005a). As an alternative to time-resolved crystallography, various physical or chemical trapping methods are often applied to extend the lifetimes of reaction intermediates sufficiently so that they can be studied by the more conventional and less technically challenging static crystallography. Examples of such methods are lowering temperature, trapping of intermediates by freezing, pH or solvent modifications and chemical modifications of the macromolecule (including mutations), substrate or cofactor. Although trapping methods provide valuable insight into structures of intermediates, they at the same time perturb the reaction, requiring consequences of the perturbation to be evaluated. In addition to time-resolved crystallography, real-time information on structural changes for macromolecules containing a chromophore, such as heme proteins, is also provided by time-resolved spectroscopic studies, including absorption, resonance Raman and infrared spectroscopy. Although such techniques offer great sensitivity to small structural changes, most are limited to the chromophore environment. In addition, structural changes are not directly observed, but the measured spectral changes must be interpreted in terms of underlying structural change. Time-resolved crystallography therefore has a distinct advantage as it provides direct and global structural information, for the entire molecule, in atomic detail as a function of time. As such, it is ideally suited for following the propagation of structural changes throughout the protein from the active site (heme in heme proteins, for example), for exploring the events involved in an allosteric mechanism and for tracking the ligands on their migration pathways throughout a protein.

As in any other type of time-resolved measurement, successful identification of an intermediate requires that the fraction of molecules in the intermediate accumulates, at some time following the start of the reaction, to a level significant enough to be detected by a measurement with an appropriate time resolution. The magnitude of structural change between the known initial state and the intermediate also has to be large enough to be detectable, but small enough to be accommodated by the crystal lattice. As many macromolecules are active and perform their function in the crystalline state, the structural changes involved in these cases are clearly compatible with the crystal lattice. However, a key question is if the changes are large enough to be reliably detected. Another important issue for time-resolved studies is that at any given time during the reaction several intermediates are likely to be present in the crystal, unless intermediates are very well separated in time by vastly different lifetimes. In other words, there may not be a time window where only one intermediate is predominant. It is therefore very important to have a method of data analysis that will facilitate extracting the structures of individual intermediates from the measured mixture of states as a function of time.

During the last two decades, essential advancements have been made in the development of both high-intensity third generation synchrotron X-ray sources and necessary time-resolved diffraction instrumentation, as well as in methodology and software for data processing and analysis (Bourgeois et al., 1996; Rajagopal et al., 2004b; Ren et al., 1999; Schmidt et al., 2003). The technique is in its mature phase today, where detection of structural changes as small as 0.2–0.3Ǻ with time resolution of 100ps is possible, even when reaction is initiated in only 10–20% of molecules in the crystal (Anderson et al., 2004; Bourgeois et al., 2006; Bourgeois et al., 2003; Šrajer et al., 2001; Šrajer et al., 1996; Ren et al., 2001; Schmidt et al., 2005b; Schotte et al., 2003; Schotte et al., 2004). In addition, extraction of time-independent structural intermediates from measured time-dependent structure factor amplitudes has been successfully demonstrated (Ihee et al., 2005; Rajagopal et al., 2005; Rajagopal et al., 2004b; Schmidt et al., 2004; Schmidt et al., 2003).

In this review, we will discuss some general aspects of time-resolved crystallography applicable to all systems, but will focus on the technique as applied to heme proteins. The natural photosensitivity of the heme-CO bond, which was first described over a hundred years ago (Haldane and Smith, 1896), permits rapid triggering of ligand release in hemoglobins and myoglobins, rendering them particularly amenable to time-resolved crystallographic experiments. As another contribution in this issue will discuss methods as applied to myoglobin, we will restrict our coverage to hemoglobins here.

2. Experiment

Structural changes in macromolecules at room temperature span many orders of magnitude in time, from fs to sec and longer. To examine short-lived intermediates in real time, it is critical to initiate the reaction in all molecules very rapidly, in a time period that is significantly shorter than the lifetime of such intermediates. The fastest method of reaction initiation by far is the use of ultra-short laser pulses. This method is readily applicable to molecules that are inherently photosensitive and undergo a reversible reaction, like ligand photodissociation and rebinding in heme proteins. Alternatively, other molecules can be rendered light sensitive by chemically attaching photosensitive groups to substrates, cofactors or important protein residues. The goal of such “caging” is to make the molecule inert until light is absorbed by the photosensitive group, which in turn triggers the reaction. In this case the reaction will be irreversible. Also, reaction initiation in such systems is typically relatively slow (μs-ms) as it is governed by the release of the caged group and often involves diffusion of the released group into the active site.

Time-resolved X-ray diffraction experiments are of a pump-probe type: laser pulses are used as “pump” pulses, to trigger the reaction rapidly, while X-ray pulses are used to “probe” the reaction, at various time delays following the pump pulse. Time-resolution of the experiment is determined by the duration of the pump or probe pulses, whichever is longer. A critical requirement for the X-ray source in such experiments is a very high X-ray flux in short pulses, typically available at third generation synchrotron sources, such as Advanced Photon Source (Argonne National Laboratory, USA), European Synchrotron Radiation Facility (Grenoble, France) and SPring 8 (Japan). As synchrotron radiation consists of a continuous 100ps pulse train, for best time-resolution one needs to isolate a single 100ps pulse from such train and synchronize its arrival at the sample with the laser pump pulse. For slower reactions with less demanding requirement on time-resolution, an X-ray pulse train of a longer total duration can be used as a probe pulse to increase the probe pulse intensity and therefore provide improved signal-to-noise ratio for recorded diffraction images.

Mechanical crystal rotations used in standard oscillation crystallography are much too slow to probe fast, sub-second reactions. Instead, Laue diffraction of stationary crystals is used. Laue diffraction requires significantly wider energy bandwidth than the standard monochromatic diffraction. It has been shown that undulators are best X-ray sources for time-resolved Laue diffraction experiments (Bourgeois et al., 2000; Šrajer et al., 2000; Ren et al., 1999). The typical bandpass of undulators used today for time-resolved experiments is 3–5% at 12–15keV. Softer, lower energy X-rays increase radiation damage while harder, higher energy X-rays are diffracted and detected less efficiently. Undulators, as compared to other synchrotron sources with wider energy bandwidth such as bending magnets or wigglers, have numerous advantages. They have higher peak intensity and, due to the narrow bandwidth, result in lower polychromatic background as well as reduced spatial and harmonic overlap in typically crowded Laue diffraction patterns. Overall data quality obtained from undulator sources is superior as judged by all important indicators, such as R_merge, data completeness and map quality.

When a single 100ps X-ray pulse is used to record a diffraction image, a flux greater than 10¹⁰ photons/pulse is needed, focused to match the crystal size. If such high-flux and well-focused X-ray pulses are not available, the pump-probe sequence has to be repeated numerous times and diffraction data accumulated at the detector prior to the readout of a diffraction image. Such measurements with repeated pump-probe sequence are clearly possible only for fully reversible systems. But even for such systems, it helps to reduce the number of pump (laser) pulses as they are potentially damaging to the crystal. Single X-ray pulse data acquisition is essential for irreversible systems, where each image requires a single pump-probe sequence and a new crystal.

A critical part of time-resolved experiments is an X-ray shutter train that isolates single X-ray pulses or longer pulse trains from the continuous stream of synchrotron pulses (Bourgeois et al., 1996). Typically, a fast rotating chopper is used in series with a slower, single opening shutter that isolates a single opening of the chopper. For example, at the BioCARS facility, located at Sector 14 at the APS, an ultra-fast chopper (Forschungszentrum Jülich, Germany) is capable of isolating X-ray pulses that are 153.4ns apart, during the standard, 24-bunch mode of the APS storage ring (Fig. 1). Ability to utilize this standard rather than a special operating mode of the storage ring is highly beneficial as significantly more beamtime becomes available for conducting these technically challenging experiments.

Figure 1.

Skematic diagram of the timing for time-resolved experiments at BioCARS beamline 14-ID, APS. An ultra-fast chopper with an opening of only 300ns, along with a slower millisecond shutter, permits isolation of a single 100ps X-ray pulse during standard, 24-bunch, mode of the APS storage ring. Time-resolved experiments are carried out using a laser pulse that is synchronized with that X-ray pulse.

Crystals for time-resolved crystallography are typically mounted in thin-walled glass capillaries as data collection is conducted at or near room-temperature. Protein crystals contain a very large number of molecules, on the order of 10¹³–10¹⁴, and therefore typically have high optical density (OD) in the regions of the absorption spectrum where the chromophore absorbs significantly. In order to photo-initiate the reaction by laser pulses throughout the crystal, smaller (thinner) crystals have to be used and the laser wavelength tuned to a spectrum region where OD≪1. For ns time-resolved experiments, where high laser pulse energies are typically available, one can afford 1–2mJ/pulse, defocused to a size larger than the crystals and at the wavelength where the crystal OD is 0.1–0.2. Both low OD and large laser beamsize enhance the uniform photo-initiation throughout the crystal. For the same reason, it is also advantageous to illuminate the crystal by laser light from the opposite sides (Knapp et al., 2006). With shorter, sub-ns laser pulses, typical pulse energies are 10–100μJ/pulse. With the number of available photons just barely matching the number of molecules in smaller crystals in this case, all photons need to be absorbed. The laser beam has to be focused to match the size of the crystal. Typically a small X-ray beam is used to probe only the laser-illuminated surface layer of the crystal (Schotte et al., 2004). This way only the volume of the crystal that is penetrated by the laser light is probed by X-rays.

When crystals are exposed to laser pulses, both increased crystal mosaicity and crystal motion are often observed, as evident by elongation of the diffraction spots. The crystal motion is believed to be heating-related. To reduce the motion, a sufficient wait-time between pump-probe cycles as well as the minimal number of such cycles for each diffraction image should be considered. Also, immobilizing the crystal inside the capillary is highly beneficial, as successfully accomplished for Hb crystals by using polyvinyl or epoxy (Knapp et al., 2004).

A complete time-resolved data set spans four-dimensions: three traditional reciprocal space dimensions and time. Similar to a standard monochromatic data set, a polychromatic Laue data set at each time delay contains images collected at many different crystal orientations to completely sample the reciprocal space. The angular step in crystal orientation for Laue data depends on the bandpass of the X-ray source and is typically 2 – 3° for undulator sources. The number of time points required to characterize a reaction that is being investigated and how the time points are sampled ultimately depend on the number of intermediates and their lifetimes. A good starting point is to collect 5 points per decade in time, equally spaced in logarithmic time. The time interval between consecutive pump-probe cycles depends both on lifetimes of intermediate states and time needed for dissipation of heat deposited by the laser pulses. It is clearly necessary to allow sufficient time for the recovery of the initial state. Typical repetition rate used for heme proteins is 1–3Hz.

The most straightforward way of collecting time-resolved data is to collect complete laser-on and laser-off data sets for a given time delay between the laser and X-ray pulses, before moving to the next time delay. Since the signal of interest is the difference in diffraction before and after the photo-activation, it is best to collect both data sets on the same crystal and to interleave the laser-on and laser-off images for each crystal orientation. As crystal damage by the X-ray and laser pulses limits the maximum number of images that can be collected from one crystal, typically not all desirable time delays can be collected on the same crystal. As a result, crystal-to-crystal variations as well as the laser intensity fluctuations and drifts introduce potential systematic errors in time domain due to variations in reaction initiation in this scheme of data collection. These effects can have a detrimental influence on determination of time constants when the entire series of time delays is examined. To minimize such errors across the time domain, more recent experiments employ time delay as the fast variable for time-resolved data collection (Schmidt et al., 2005a). In this procedure, the time delay is scanned for a given crystal orientation, starting with the laser-off image and collecting a series of laser-on images that span the desired time domain. The time scan is then repeated for a new crystal orientation. Due to the radiation damage issues, several crystals are normally required for a complete space-time data set. On each crystal, a data set that is partial in reciprocal space but which covers all time delays is collected. Since the goal is to determine accurately the differences between the structure factor amplitudes of the initial dark state, F_dark, and the state at a particular time delay following the laser pulse, F_light(t), such differences ΔF(t)=F_light(t)−F_dark should be determined first, before they are merged across all crystals to obtain a time series of complete difference data sets. Overall, this data collection strategy constitutes the best method for accurately measuring the differences in structure factor amplitudes as a function of time, ΔF(t), as they change due to the formation and decay of intermediate states.

3. Data Processing and Analysis

Processing of time-resolved data can be roughly divided into three steps: 1) Laue data processing to derive time-dependent structure factor amplitudes from recorded diffraction images; 2) calculation of time-dependent difference electron density maps and analysis of such maps; 3) determination of structures of intermediate states.

Software for Laue data processing has played a critical role in the success of time-resolved crystallography. Several problems that are specific to the Laue diffraction method had to be addressed and resolved: spatial overlap of diffraction spots in typically crowded Laue diffraction patterns, wavelength normalization and resolving the harmonic overlaps. The so-called wavelength normalization is needed due to the fact that the intensity of the polychromatic radiation, the scattering power of the crystal and detector sensitivity are all wavelength dependent. The same reflection or its symmetry mate may be stimulated by a different wavelength depending on the crystal orientation. In order to merge data, the intensities therefore need to be brought to a common scale. The result of wavelength normalization is a λ-curve that combines all wavelength-dependent effects in diffraction intensities as recorded at the detector. Harmonic or energy overlap results from reflections that lie on a radial line (starting at the origin) in the reciprocal space. Such reflections are stimulated by different energies but scatter in exactly the same direction and therefore overlap exactly at the detector. These characteristic features of Laue diffraction data, traditionally considered “problems”, have been addressed very successfully by the Laue processing software developed in mid 1990-ies. Several excellent Laue processing packages exist today: Precognition (Renz Research Inc), LaueView (Ren and Moffat, 1995), PrOW (Bourgeois, 1999), Daresbury Laboratory Laue Processing Suite (Arzt et al., 1999; Campbell, 1995). Currently, the main ongoing software efforts center on automation of data processing with minimum user input and intervention to facilitate fast, on-line Laue data processing. The quality of Laue data today and electron density maps derived from the Laue data is comparable to standard monochromatic data.

In the next step of time-resolved data analysis, time-dependent difference electron density maps Δρ(t) are calculated in the following way. The structure factor amplitudes (SF) of the initial, dark state |F^D(hkl)| and a corresponding set of time-dependent structure factor amplitudes |F(hkl,t)| are used to calculate time-dependent difference amplitudes, ΔF(hkl,t) = |F(hkl,t)| − |F^D(hkl)|, for each time point t. Phases ${\phi }_{\mathit{hkl}}^D$, used together with difference amplitudes ΔF(hkl,t) for calculating difference maps Δρ(t), are obtained from the known dark state structural model. In order to reduce the experimental and data processing errors, several weighting schemes for difference amplitudes have been proposed when calculating difference maps (Šrajer et al., 2001; Ren et al., 2001; Schmidt et al., 2003; Ursby and Bourgeois, 1997). In such maps the features that appear above 3σ level are considered significant, where σ value is the root mean square (RMS) deviation of the difference density from the mean value in the asymmetric unit. Negative electron density features in such maps represent the loss of electrons and therefore indicate areas where atoms have moved from their positions in the dark state following photolysis, while positive features represent the gain of electrons and indicate areas where atoms have moved to. When SF amplitudes are given on the absolute scale, difference electron density in selected regions can be integrated to provide the information about the total number of electrons displaced from or into a particular volume in space as a function of time. Program Promsk (Schmidt et al., 2005b), for example, can be used to integrate difference density within a specified mask, generated by supplying atomic coordinates and a radius of integration around the coordinates for the structural region of interest. It is sometime beneficial to integrate |Δρ(t)| rather than Δρ(t) values in order to obtain the sum of both positive and negative signals in a particular region. In this case a similar integration of |Δρ(t)| is needed for a difference map free of signal to estimate the noise contribution and subtract it from the integrated |Δρ(t)| values for the signal. Time courses of Δρ(t) or |Δρ(t)| provide important information on formation and decay of structural intermediates, as will be illustrated in the next section, on the example of time-resolved studies of dimeric Scapharca hemoglobin.

Singular Value Decomposition (SVD) provides a substantially more complete method to address the questions of the number of global intermediates, time-independent structures of such intermediates, rates of formation and decay of intermediates and the overall kinetic mechanism for the process studied by time-resolved crystallography (Ihee et al., 2005; Rajagopal et al., 2005; Rajagopal et al., 2004b; Schmidt et al., 2004; Schmidt et al., 2003). Both SVD and post-SVD analyses are needed and they applied to time-dependent difference electron density maps to obtain information about intermediates and the kinetic mechanism. A comprehensive description of these methods can be found in Schmidt et al., 2005b. In short, SVD is a method of global analysis that is applied to a data matrix composed of N time-dependent difference electron density maps, Δρ(t). The SVD procedure decomposes such a matrix into N time-independent Δρ maps or left singular vectors (lSV) and corresponding N time courses or right singular vectors (rSV). Each vector pair has a singular value associated with it, which weights the contribution of the lSVs to the experimental difference maps. The actual difference signal is typically contained in only few significant vectors, associated with singular values of large amplitude. The remaining singular vectors contain only noise. The SVD analysis therefore provides an effective noise filter as the input series of maps can be approximated by S/N improved maps Δρ′(t), reconstituted from significant singular vectors only. The number of significant singular values and vectors is also related to the number of intermediates in the reaction. As the rSV time courses are linear combinations of the true time-dependent concentrations of intermediates, they provide information on the number of relaxation processes and the associated relaxation rates that are associated with the time-dependant concentrations. From a global fit of all significant rSVs by a sum of exponential relaxations, with rates common to all rSVs, the number of relaxation processes and associated rates in the reaction are determined. The number of relaxation rates is the lower bound on the number of intermediates.

With the number of relaxations and associated rates determined from the rSVs, post- SVD analysis provides a method for determination of time-independent difference maps, Δρ_i, corresponding to intermediates (Rajagopal et al., 2005; Rajagopal et al., 2004b; Schmidt et al., 2005b; Schmidt et al., 2003). At this stage of the data analysis, assumption of a candidate reaction mechanism is required for the number of intermediates determined by the SVD analysis. For the assumed mechanism, the time-independent difference maps of intermediates are synthesized as a linear combination of significant lSV maps (Schmidt et al., 2005b). The mixture of intermediates in the experimental time-dependent difference maps has therefore been separated into the difference maps corresponding to pure intermediates. In an iterative procedure, plausible reaction mechanisms are examined and the best mechanism determined by comparing the measured time-dependent difference maps, Δρ(t), and difference maps, Δρ_calc(t), calculated for each candidate mechanism by using corresponding time-independent difference maps as well as time-dependent concentrations of pure intermediates (Schmidt et al., 2005b).

Finally, the structures of reaction intermediates are refined. For this purpose conventional electron density maps are calculated for each intermediate from the so-called extrapolated SFs (structure factors). Extrapolated SFs are obtained by vector summation of calculated SFs for the dark state and the difference SFs for an intermediate, where the difference factors result from the Fourier transform of the difference map of the intermediate. The amplitude of the difference SF is therefore amplified to correspond to 100% photo-initiation. Such conventional electron density map therefore represents just the intermediate, with no contribution from the dark state. The structures of intermediates are then modeled and refined using these conventional maps.

4. A case study: Scapharca dimeric hemoglobin

The dimeric hemoglobin (HbI) from the blood clam Scapharca inaequivalvis is very well suited for the investigation of allosteric structural transitions by time-resolved crystallography. The allosteric reaction from high affinity R-state to the low affinity T-state can be triggered by a laser flash, which causes the release of bound CO as in other heme proteins. Importantly, allosteric transition in HbI involves more localized structural changes than some other allosteric proteins; as a result, cooperative oxygen binding has been observed in the crystalline state (Mozzarelli et al., 1996) and full ligand-linked transitions, including the small (~3°) subunit rotation, can occur within crystals grown in the CO-liganded form (Knapp and Royer, 2003). Despite the rather limited structural changes, these transitions have large functional ramifications, with the R-state estimated to bind oxygen about 300-fold more tightly than the T-state (Royer et al., 1996).

Conventional crystallographic analyses complemented by site-directed mutagenesis and ligand-binding experiments have revealed three key structural transitions that contribute to the functional differences between the R and T states of HbI: movement of Phe 97 (corresponding to position F4 of myoglobin) from the subunit interface to the proximal pocket upon ligand loss (Knapp et al., 2005; Pardanani et al., 1997), redistribution of interface water molecules (Pardanani et al., 1998; Royer et al., 1996) and movement of the heme groups towards the interface upon ligand loss (Knapp et al., 2001). We carried out time-resolved crystallographic experiments of this molecule to elucidate the time-dependent interplay among these functionally important structural transitions (Knapp et al., 2006).

Key to the success of these experiments was the ability of HbI crystals to tolerate laser induced ligand release and reversible ligand rebinding for many (thousands) of cycles. Crystals were mounted in thin walled glass capillaries and immobilized under a poly(vinyl) film (Knapp et al., 2004), in order to minimize crystal movements that had been observed upon laser exposure in preliminary experiments. Due to rapid binding of oxygen compared with CO and oxidation of the heme iron in the presence of oxygen, it is important to remove oxygen from the capillaries; this was accomplished by purging the capillaries with CO before adding crystal stabilizing solution that had been saturated with sodium dithionite.

Photodissociation of the bound CO molecule was induced by 7ns laser pulses (FWHM) from a Nd:YAG pumped dye laser (Continuum). Crystals were simultaneously illuminated from two opposite sides, at 615nm. Three different time-series, each using time-delay as the fast variable (see Experiment section), were employed to elucidate the structural transitions that occur between 5ns and 80μs following photolysis. For the first time-series, data were collected with time-delays between laser and x-ray pulses of 5ns, 25ns, 75ns, 200ns, 700ns and 3μs. For each of four different crystals, 21 orientations, with a 9° increment between angular settings of the crystal, were used to collect both dark images (no laser illumination) and the six time delays after the laser illumination. The two other time series covered time delays between 2μs and 80μs. Our results produced less noisy time-courses for structural changes, demonstrating that this procedure, in which data are collected at all time delays from the same set of crystals, minimizes systematic errors between different time delays.

Difference Fourier maps for the M37V mutant of HbI used for these studies revealed the time-dependent structural changes that follow photolytic release of ligand. Successful photolysis is evident in the earliest time-delay (5ns) by strong negative density (−14σ and −17σ for subunits A and B, respectively) at the ligand position (Fig. 2). Integration of this density indicates roughly 40% photolysis, which decays to about 10% by 1μs. The results from early time delays (5–200ns) suggest that structural changes are largely, but not entirely, limited to the ligand binding site, whereas difference maps at later time points in the μs time range show propagation of the structural changes into the subunit interface (Fig. 2A). Therefore, the ligand-linked transition can be usefully divided into two phases: an early intermediate phase during the nanosecond time period and an allosteric phase.

Figure 2.

Difference Fourier maps between Scapharca HbI* (photoproduct) and HbI-CO at time-delays of 5ns and 60μs. (A) Ribbon diagram of HbI-CO dimer (gray) with side chains for His F8 (cyan), Phe F4 (yellow) and two key interface water molecules (small cyan spheres) are shown along with the difference Fourier map contoured at +3.5σ (blue) and −3.5σ (red). Density at 5ns suggests that early structural changes are largely concentrated at the ligand binding site, with major interface allosteric structural transitions occuring by 60μs. An exception is the density for the pair of R-state water molecules shown in cyan (arrows), which show clear negative density by 5ns. (B) An α-carbon trace (gray) for the CD region and E and F helices along with the heme group (salmon) and key side chains from subunit A. (Map contoured at +3.5σ (blue) and −3.5σ (red).) Photolysis is clearly evident by the strong negative density observed at the CO binding site for both subunits A (−14σ) and B (−17σ) at 5ns. (C). Difference density for the region around Phe F4, with map contoured at +2.5σ (blue) and −2.5σ (red). The R to T transition of Phe F4 has not started by 5ns, but is complete by 60μs. From Knapp et al. (2006) Proc Natl Acad Sci U S A 103, 7649–54. Copyright (2006) National Academy of Sciences, USA.

Evident in the difference Fourier maps is the formation of a tertiary intermediate within 5ns as each R-state subunit responds to the presence of an unliganded heme group. This intermediate is characterized by a buckled heme group, with the iron displaced from the heme plane by about 0.4Å as is apparent from strong positive peaks in the electron density maps and confirmed by difference refinement (Terwilliger and Berendzen, 1995). Rapid structural changes are also observed in the F helix, particularly around the heme-linked proximal histidine (Fig. 2B). Density features, and difference refinement, indicate that the F-helix moves in the direction of the dimeric interface, which results in a disruption of water molecules unique to the R-state. Most clearly affected are two R-state water molecules (Fig. 2A) that are located near the F4 and F7 side chains and also hydrogen bond to heme propionates. It appears that their disordering is needed to facilitate the R to T movement of the heme groups towards the interface and thus may lay the foundation for the allosteric transition.

Key structural transitions characteristic of the T-state commence after 1μs. These include movement of the heme groups towards the interface, highlighted by the density peaks for the iron atom, movement of the Phe F4 side chain from the interface into the proximal pocket and accumulation of water molecules at T-state specific locations in the subunit interface. Integration of the difference Fourier density for the T-state position of Phe F4 at its maximum accumulation (10–30μs) suggests ~8% population of the T-state Phe F4. This matches the deliganded heme population in the same time interval. Phe F4 has therefore switched from the R to T state in all deliganded subunits, suggesting nearly complete R to T transition in those subunits without ligand 10–30μs following the laser flash. Comparison of the time-courses of the structural changes associated with the R to T transitions indicates that all three central movements (heme, Phe F4, water molecules) follow very similar time-courses (Fig. 3). Thus, these movements are tightly coupled, suggesting a rather concerted R to T transition.

Figure 3.

Time-dependent integrated difference electron density values (Δρ(t)) following photolysis of Scapharca HbI for several structural regions, with values averaged over subunits A and B. (A) Integrated Δρ(t) values for the loss of bound CO (red circles) and the photodissociated CO at the distal pocket docking site (CO*, blue circles). Both features exhibit similar biphasic behavior that suggest an exponential geminate ligand rebinding phase and a bimolecular rebinding phase. (B) Integrated absolute values of difference electron densities, |Δρ(t)|, for helices E-H and the CD loop region. The F helix and the CD loop region show nanosecond changes corresponding to the formation of an early intermediate. Both the E and F helices as well as the CD loop exhibit an increase in signal in the microsecond time range that is similar to the signals associated with the rearrangements of Phe F4 (C) and allosteric water molecules (D). Simultaneous fits of data in B-D by a common exponential function in the microsecond time region are shown as solid lines. The very similar time-courses observed suggest a concerted allosteric transition. From Knapp et al. (2006) Proc Natl Acad Sci U S A 103, 7649–54. Copyright (2006) National Academy of Sciences, USA.

The time-resolved crystallographic experiments on Scapharca dimeric hemoglobin reveal an early intermediate whose structural properties suggest that it facilitates the transition between R and T states. The structural transitions to form the T-state molecule are observed to occur in the μs time range, with the functionally key structural changes following very similar time-courses indicative of a highly coupled transition (Knapp et al., 2006). These experiments demonstrate feasibility of following functionally relevant structural transitions in heme proteins by time-resolved crystallography.

5. Conclusions

With significant technical and software developments over the last decade, time-resolved macromolecular crystallography technique has reached a mature phase, with the demonstrated ability to detect relatively small structural changes even at levels of reaction initiation of only 15–40%. The important development of essential methods for global analysis of time-resolved data, such as SVD and more recently developed cluster analysis (Rajagopal et al., 2004a), are also well under way. The technique provides an important tool for insight into functionally important structural relaxation processes, ligand migration and allosteric action on the atomic level, with 100ps time resolution at the most intense present-day X-ray sources such as synchrotrons. One of the main challenges for time-resolved crystallography is the application of the technique to a wider range of molecules of biological interest. Most such macromolecules are not inherently photosensitive and system-specific efforts are needed to determine a suitable reaction initiation method, including the methods such as caging that make triggering of the reaction by light possible. Other challenges for the technique involve studies of irreversible processes, further improvements in time resolution to sub-100ps with new intense X-ray source such as XFELs and combining experimental results from time-resolved crystallography with computational and theoretical approaches to provide complete description of reaction pathways, including the transition states.