The Per-Arnt-Sim (PAS¹) domain is a ubiquitous protein signaling module found in all kingdoms of life which allows organisms to sense and subsequently respond to environmental stimuli (1). This ~100 amino acid domain is often expressed in tandem with an enzymatic domain whose activity is regulated by the PAS domain in response to an external stimulus (2-5). The roles of PAS proteins are diverse; some members bind prosthetic groups though others do not. As a class they show low sequence homology in spite of striking tertiary structural similarities (6). Among these proteins are photoreceptors (7), redox sensors (8), and small molecule sensors (9). Heme-PAS proteins form an important subgroup of PAS domain proteins which contain a covalently bound heme prosthetic group and exhibit catalytic control of their enzymatic domains in response to dissolved diatomic gases (2, 3, 10-12). Several heme-PAS proteins have been characterized including phosphodiesterase A1 of Acetobacter xylinum (13), the Escherichia coli protein DOS (10), the putative carbon monoxide sensor NPAS2 (14) from mice and humans, and the FixL proteins of nitrogen-fixing Rhizobia (15). Of these, the FixL proteins from the nitrogen-fixing bacteria Bradyrhizobium japonicum (bjFixL) and Sinorhizobium meliloti (smFixL) have been the most widely studied (see ref 12 for review of heme-PAS proteins).

bjFixL is a dimeric oxygen sensor responsible for initiating regulation of transcription of genes encoding proteins involved in nitrogen fixation and oxidative stress (16). Each monomer contains a single N-terminal heme-PAS sensory domain and a C-terminal kinase domain. The autophosphorylation activity of the histidine kinase is inhibited by ligand binding at the heme (2, 3). The isolated heme-PAS domain of this protein, referred to as bjFixLH (Figure 1A), has been the subject of crystallographic studies which examine the mechanism of ligand recognition and signal transduction (17-20). In contrast to most globins, FixLs have a hydrophobic binding pocket and lack a distal histidine residue (Figure 1B). Upon binding of oxygen to deoxy bjFixLH, the heme alters its stereochemistry and the protein undergoes a conformational change both in the FG loop and in residues distal to the heme in the H and I -strands of the core sheet of the PAS domain. Rearrangement of the FG loop is accompanied by movement of Arg220 into the heme pocket where it forms an essential hydrogen bond with the bound oxygen (18, 19). Carbon monoxide binding to FixL does not result in this FG loop switch, but similar structural changes are evident in the H and I -strands. The FG loop in the CO-bound structure is only slightly shifted when compared to the O₂- and CN-bound structures (21); the side chain of Arg220 does not move into the pocket. These findings suggest that signaling upon ligand binding in bjFixLH is at least partially achieved through the H and I -strands distal to the heme. This region of the PAS -sheet has been implicated in signal transduction in several other PAS domains of diverse function (22-28).

Figure 1 (A) Ribbon cartoon of the structure of bjFixLH in crystals in space group R32 (PDB code 1XJ2). Secondary structure elements are identified in their order in the primary structure (17). (B) A ball and stick rendition of the heme pocket of bjFixLH. The CO-bound conformation is shown in yellow, the deoxy state in gray.

Carbon monoxide has been used for several decades as a molecular probe of protein dynamics associated with conformational change in heme proteins (29-32). The CO-iron bond in CO-bound heme is photolabile and can be ruptured by light with high quantum efficiency. This makes it possible to rapidly generate a state of the protein that is far from equilibrium, to initiate the relaxation of structure toward the unliganded deoxy state, and to monitor rerelaxation to the CO-bound state as CO ultimately rebinds. Using time-resolved Laue X-ray crystallography (see ref 33-36 for reviews) it is possible to visualize changes in the positions of all atoms within a protein during such processes, thus revealing the structural dynamics of the entire protein as it performs its function. Time-resolved crystallography has been extensively applied to both wild type and mutant forms of the oxygen storage and transport protein myoglobin (37-43), to a dimeric hemoglobin which retains cooperative allosteric binding in its crystal lattice from the clam Scapharca inaequivalvis (44), and to the bacterial blue-light photoreceptor PYP (23, 45, 46). In addition, initial time-resolved studies have been carried out on the photosynthetic reaction center (47).

We present time-resolved X-ray diffraction studies of CO recombination in bjFixLH to reveal the nature of the structural changes and the time course by which they develop. By comparison with globins, little is known about the dynamics of CO recombination within the widely distributed heme-PAS domain, for which the bimolecular association rate constant for CO rebinding at ambient temperature is 100-fold lower than in wild-type sperm whale myoglobin (10). Time-resolved crystallographic datasets were collected using Laue diffraction at ambient temperature at 16 delay times from 1 s to 10 ms after CO photolysis in order to directly visualize the photolyzed bjFixLH structure, rebinding of CO ligand, and relaxation of the heme and the protein. Singular value decomposition (SVD) analysis of the resulting 16 time-resolved difference Fourier maps identifies the number and nature of significant structural components (46, 48, 49). The application of time-resolved crystallography to bjFixLH is particularly challenging for two reasons: bjFixLH crystals exhibit relatively modest diffracting power, and the structural changes associated with protein relaxation and CO binding turn out to be small in spatial extent (18, 21). However, substantial improvements in the time-resolved Laue method (50), data processing (51), noise reduction (52), and electron density visualization (39) allow us to successfully apply this technique to bjFixLH, whose crystals are of a moderate diffraction quality more typical of the majority of protein crystals than are those of myoglobin or PYP. In addition, we report transient absorbance flash-photolysis measurements of carbon monoxide rebinding in bjFixLH crystals and compare these with solution studies.

Materials and Methods

Protein Purification and Crystal Preparation. bjFixLH protein was expressed and purified as described (53). The crystals were grown, converted to the CO state, and mounted in capillaries under CO as described (21). Rhombohedral crystals of bjFixLH in space group R32 were used for time-resolved Laue diffraction due to their high symmetry and low mosaicity. The concentration of heme (and CO) in the crystals is 38 mM. Crystals were mounted 1-2 days prior to data collection, though no oxidation was observed spectroscopically over several weeks.

X-ray Data Collection. The rhombohedral bjFixLH crystals grow in a disk-like morphology and were mounted with the thin dimension facing the capillary wall in order to present the shortest dimension to the laser to achieve maximum photolysis. Crystals were photolyzed by a 7.5 ns (fwhm) laser pulse at 615 nm produced by a Nd:YAG pumped dye laser (Continuum) containing rhodamine 610/640 laser dye. The pulse was delivered to the crystal through either one or two optical fibers focused to a spot with ~1 mm diameter at the crystal. The total laser pulse energy was 1.5-2 mJ. A wavelength of 615 nm was used to penetrate the optically dense crystal on the red edge of the peak of the CO-bjFixLH absorption spectrum. The optical density at this wavelength of a typical crystal was 0.25. Temperature was maintained at 14 C at the crystal by a cooled gas stream to minimize the temperature rise arising from heat deposited by repeated laser flashes. Laue X-ray diffraction data were collected on a MarCCD 165 detector at the 14-ID-B station of BioCARS, at the Advanced Photon Source (APS), Argonne, IL. This beamline employs an APS Undulator A insertion device which was set at 25 mm gap. At this gap the first harmonic of the undulator lies at 1.085 Å wavelength. All experiments were conducted in pump-probe mode in which a laser pulse to initiate CO photolysis-the pump-was followed after an adjustable delay time by the X-ray pulse or pulse train-the probe. Delay times, defined by the time between the centers of the X-ray and laser pulses, ranged from 1 s to 10 ms. Delay times from 1 to 7 s used a 500 ns X-ray pulse train. For delay times of 10 s or longer, a 2 s X-ray pulse train was used to enhance the X-ray intensity and minimize the number of exposures. A typical Laue dataset consisted of 40 light (with laser flash) and 40 complementary dark Laue diffraction images spaced by 3 in angular setting of the crystal. Redundant datasets were collected at several delay times to allow averaging that maximized signal-to-noise and data completeness.

Data Reduction, Difference Fourier Maps, and Singular Value Decomposition Analysis. Laue diffraction data was processed using the Precognition Laue processing software package (54). All data were integrated to a resolution of 2.5 Å. Difference structure factor amplitudes for each time point, F, were calculated by subtracting the dark structure factor amplitudes from the light. Difference structure factors were then weighted according to their magnitudes and standard deviations as described in Ursby et al. (52) and rajer et al. (40). Phases for each reflection were calculated using the static room-temperature structure of the CO-bjFixLH complex (PDB code 1XJ2) (21). Difference Fourier maps were calculated for each dataset from a single crystal in order to minimize systematic error associated with crystal to crystal variation. A single F dataset at each delay time was produced by averaging redundant time points. The completeness and redundancy of all data is presented in Table 1, and statistics for the 16 difference datasets in Table 2. Restrained refinement of the CO-bound and photolyzed states of bjFixLH was carried out using datasets calculated from all data spanning delay times from 1 to 3 s, using the program REFMAC5 from the CCP4 package (55). Refinement statistics are presented in Table 3. Estimation of the extent of photolysis and extrapolation of electron density to 100% photolysis were carried out as described by Schotte et al. (39). Briefly, maps are calculated from extrapolated structure factors according to the formula F_phot = F_dark + (2/x)(F_light - F_dark) where F_dark and F_light are the experimentally determined CO-bound and partially photolyzed structure factor amplitudes and F_phot represents structure factors of a 100% photolyzed state. The fractional extent of photolysis, x, is calculated from electron density maps by estimating the point at which CO density is no longer observable and density of the heme iron is fully extended from the porphyrin in extrapolated maps. Figures of difference Fourier maps were created using the programs Ribbons (56) and pymol (57). SVD of time-resolved difference Fourier maps was carried out with the program SVD4TX (48). SVD decomposes time-dependent data into its orthonormal time-independent components in the form of left singular vectors, or lSVs, and identifies the time course associated with each component in a corresponding right singular vector, or rSV (48, 49, 58). Light and dark maps, mapped nonlinearly to brightness in Figure 4, were created using equivalent reflections by Friedrich Schotte using the program Lauemap (38). Cavity analysis of bjFixLH was performed with VOIDOO (59).

	Figure 2 Sixteen time-resolved difference electron density maps from 1 s to 10 ms after photolysis. Maps are contoured at ±3.5 and 4.5 where blue is positive and red is negative, and is the root-mean-square value of the difference density across the asymmetric unit.
	Figure 3 A superposition of time-dependent Folight and corresponding Fodark electron density maps. Shown are maps for the CO-bound ground state (magenta) and corresponding photolyzed state (green) at time delays from 1 s to 10 ms. Maps are mapped nonlinearly to color brightness. The overlap of magenta and green in this superposition presentation of maps blends to white. The magenta-to-green color gradient indicates the direction of motion of protein atoms. The orientation is shown in the upper right corner as a ribbon model of the molecule in gray, with van der Waals surface outlined in yellow.
	Figure 4 (A) CO recombination kinetics in bjFixLH crystals in space group R32, derived from transient absorbance change (black trace), a fit to a bimolecular expression (red), and the first right singular vector rSV1 derived from SVD analysis of time-resolved difference electron density (green). (B) The corresponding difference electron density lSV1 from SVD analysis, contoured at ±3.5 and ±4.5 where blue is positive and red is negative density. (C) The difference map created by averaging difference structure factors from all delay times from 1 to 3 s.

CO Flash-Photolysis Kinetics in bjFixLH Crystals. CO flash photolysis optical measurements on crystals were carried out using a home-built microspectrophotometer (60). The pump laser pulse was provided as in the X-ray diffraction experiments. The change in absorption at 543 nm due to CO recombination after photolysis was measured using weak CW light from a He-Ne laser, where the change in light intensity transmitted by the crystals was recorded by a photomultiplier tube (Hamamatsu R955) and an oscilloscope (500 MHz Tektronix 744A). One thousand time-courses were collected and averaged to improve the signal-to-noise ratio.

Results and Discussion

Evolution of Structural Change and Carbon Monoxide Rebinding within the Heme Pocket. The time-resolved crystallographic data series follows the evolution of structural change over 4 decades in time as CO rebinds after photolysis (Figure 2). All maps from the early delay times (1 to 3 s) are very similar. The maps show the slow decay of a long-lived photolyzed state, characterized by a large negative feature on the CO, a highly displaced, domed porphyrin and heme iron, and the collapse of the distal hydrophobic residues, Leu236, Ile238, and Ile215 into the heme pocket after departure of the CO. Also evident are distinct difference features on the backbone atoms of the H and I -strands at Leu236 and Val253 on the surface of the protein. Such features were previously identified in the refined structures of the CO and deoxy states in static CO - deoxy difference maps (21). Also notable are difference features on the side chain of the proximal histidine and the F-helix, on propionate 6 of the heme and on residues Pro212, His213, and Ile216 in the FG loop. This solvent-exposed region has exceptionally high temperature factors which are further elevated upon ligand binding (21). As a consequence, difference features in the FG loop and heme propionates are low in magnitude (from 2.8 to 3, where is the rms value of the difference electron density across the asymmetric unit). The photolyzed state persists relatively unchanged in structure from 1 s to 100 s. From 100 to 300 s, the first signs of ligand rebinding are evident through a decrease in magnitude of the features associated with the CO, the heme iron, and the backbone atoms of the distal -sheet residues. From 300 s to 1 ms, substantial CO rebinding takes place, evidenced by rapid decay of most distal side chain features, followed by loss of the features originating from the movement of the side chain of Leu236. Only a trace of the large negative CO feature remains at 5 ms, by which time the CO has nearly fully rebound. The 10 ms difference map is featureless and displays only noise: CO rebinding and structural relaxation to the reactant (the CO-bound state) is complete.

Structural Changes in the bjFixL PAS Domain. In order to reveal the smaller global conformational changes that occur in the protein following photolysis, electron density maps were constructed from the time-resolved diffraction data using the visualization technique of Schotte et al. (39). This approach allows a different presentation of data to better visualize subtle structural changes by mapping electron density nonlinearly to color brightness. Structural changes are presented as a color gradient across atoms that move, while nonlinear mapping of density highlights weak features. These maps, a superposition of the F_o^light and F_o^dark Fourier maps instead of F difference maps, reveal structural relaxation throughout bjFixLH upon photolysis (Figure 3).

The steric interaction between CO and the distal side chains flanking the ligand pocket and the change in stereochemistry of the heme are the two main factors that initiate protein relaxation after photolysis. Relaxation of protein structural elements can be primarily associated with either factor or with their combination. The movement of the distal -sheet is driven by steric interactions with the ligand. Movement of the distal side chains of -sheet residues into the pocket necessitates movement of backbone atoms and produces a longer-range conformational change. As noted in difference maps described above (Figure 2), the side chains of Leu236 and Leu238 move into the hole vacated by the newly photolyzed CO. The main chain atoms of these residues also move, as do the other residues in the distal -sheet to which their main chains are hydrogen-bonded. Changes in the distal -sheet of the molecule extend over the face of the sheet and are most prominent in residues of the I -strand at residue Arg254, more than 15 Å distant from the location of the bound CO. A lateral movement of the heme is also evident, a consequence of the relaxation of steric hindrance between the CO-liganded heme and the side chain of Leu236. This movement shifts the heme away from the FG loop toward the interior of the PAS domain and is also evident in the coupled movement of His200 and the side chain of Phe176. The steric interaction between the heme and Phe176 displaces the C helix to which Phe176 belongs. Ultimately, this helix and both the CD and GH loops are shifted outward from the core of the domain. On the proximal side of the heme, structural changes in His200 and the F helix are driven by heme doming and the movement of the heme iron out of the porphyrin plane upon spin state change. Heme doming is indicated by positive and negative difference electron density features flanking the heme plane. The main chain atoms of the F helix are displaced away from the heme, while the side chains of Tyr203 and Met192 move toward the heme to fill space created by upward movement at the porphyrin periphery. Finally, the FG loop moves in response to both factors: the doming and lateral movement of the heme propionate groups along with the steric interaction between Ile215 with ligand. The central portion of the FG loop, from residues His213 to Ile215, moves toward the heme pocket after photolysis, while residues Ile216 to Ile218 move away from the heme pocket, driven upward by the movement of the propionates as the heme changes from a planar to a domed conformation. Conformational differences are evident in the water molecules coordinated to the FG loop at the side chains of residues His213 and Arg206. Thus, conformational relaxation upon CO photolysis involves structural changes at considerable distance from the site of ligand binding in several regions of the protein.

Optical Studies of CO Recombination in bjFixLH Crystals. We examined CO recombination spectroscopically in crystals of bjFixLH to characterize CO recombination in crystals, to relate this to known solution behavior, and to provide a framework for interpreting our crystallographic experiments. CO can indeed be efficiently photolyzed in crystals of bjFixLH at moderate laser energy per pulse (~1.25-2.5 mJ/mm²). This was of concern because studies of myoglobin suggest that its crystals are considerably more difficult to photolyze than solution and there is a risk of causing irreversible damage to the crystals (61). Transient absorbance spectra show the rebinding of CO after photolysis of the R32 crystal form (Figure 4A). Also shown is a fit to an expression describing bimolecular CO rebinding where the ligand and the protein are equal in concentration, C/(1 + kt), under the conditions which hold in the crystal where the concentrations of photolyzed bjFixLH and free CO are equal. A purely bimolecular scheme is appropriate since, despite significant unimolecular, geminate recombination for O₂ and NO in bjFixLH, no geminate recombination of CO has been observed (62, 63). In the expression noted above, the rate k represents k_CO[CO]_phot, where k_CO is the second-order rate coefficient and [CO]_phot is the concentration of CO molecules, initially bound to heme, that escape into the solvent following photolysis. The value for k derived from the fit is 7.35 × 10² s^-1. A value for [CO]_phot of 19 mM is derived from the extrapolated electron density of the photolyzed state (discussed in detail below), which indicates that ~50% of all CO molecules in the crystal yield a photolyzed CO to the solvent. Hence the value for k_CO is 3.9 × 10⁴ M^-1 s^-1, in a very good agreement with the second-order rate coefficient of 5 × 10⁴ M^-1 s^-1 measured in solution (53). This suggests that the effects of crystallization mother liquor and the intermolecular forces in the crystal lattice on CO recombination are small. An even better mathematical fit to our spectroscopic data is achieved by a stretched exponential expression of the form C exp[(-kt)] (64). Such an expression introduces an additional parameter . We do not feel that this more complex fit is justified in the light of the modest signal-to-noise of our data and the absence of a compelling physical model for stretched exponential behavior.

SVD Analysis of Time-Resolved Electron Density. In order to relate our time-dependent optical and crystallographic results, we performed SVD on the bjFixLH difference electron density maps (Figure 2). Application of SVD to our time-dependent difference electron density reveals one and only one significant singular value that accounts for all difference signals in the data. Thus only two states are present, namely, the CO-bound reactant and the product, photolyzed state (Figure 4B). The time-resolved data in the case of bjFixLH do not reveal any new, short-lived intermediate structures, in marked contrast with PYP (23, 46). The time course of the decay of the lSV1 component of the data, namely, rSV1, agrees to within experimental error with the time course of CO rebinding measured optically (Figure 4A, green line). Evidently the optical and structural data are monitoring closely related recombination processes.

Because the SVD analysis indicates that only two structural states exist in our time-resolved difference maps, the early time points of the time-resolved data prior to substantial CO recombination represent this state most clearly. In order to assess the effect of the SVD analysis, the lSV1 difference map was compared with the averaged map derived from the Fs obtained from the 1 to 3 s delay times (Figures 4B and C). As expected, the two maps are very similar and show the same principal structural features. The map derived from SVD (Figure 4B), however, has substantially better signal-to-noise ratio since the total noise is distributed among the remaining 15 left singular vectors. For example, the weak features associated with movement of the heme propionate 6, the FG loop, and the proximal side of the heme are greatly enhanced in the lSV1 map, as are long-range structural changes associated with the movement of the H and I-sheets of the protein at residue Arg254. Thus the difference map derived from SVD best represents the structure of the photolyzed state of bjFixLH 1 s after photolysis of CO.

Structural Relaxation of bjFixLH to the Deoxy State. In order to confirm whether complete structural relaxation of bjFixLH to the deoxy state occurs upon photolysis of the CO, as indicated by our SVD analysis, we compared the time-independent difference maps derived from SVD with a difference map calculated from experimental structure factor amplitudes obtained from static CO-bound and deoxy bjFixLH crystals (21). We also compared the coordinates of the photolyzed state obtained by refinement of the time-resolved Laue data with coordinates of the authentic deoxy state obtained by refinement of static crystallographic data (Figure 5, panels C, D, and F). Equivalent global conformational differences are evident in both the difference maps and the refined coordinates. In the electron density difference maps (Figure 5, panels A and B), structural change is evident from the heme to the distal -strands. Difference electron density is present at lower levels in the experimental time-resolved maps than in calculated static electron density difference map due to the noise in the former. The noise present in the experimental maps, coupled with elevated temperature factors for the photolyzed state, gives rise to some apparent differences between the maps. These differences are associated with regions of the structure with exceptionally high B-factors. Temperature factor differences of greater than 20% have been noted between the liganded and unliganded forms of bjFixLH, specifically in the FG loop (21). A prominent difference between the maps which can be attributed to this effect can be seen at the side chain of Ile215 in the heme pocket. Common features are evident on the main chain atoms of Arg246 and in the FG loop of the molecule, however. These are the conformational changes most distant from the heme noted earlier (21), which suggests that propagation of structural changes in photolyzed bjFixLH is complete. Differences between refined coordinates of the CO-bound and photolyzed species show very similar changes in the distal -strands and FG loop. The static CO and deoxy structures on the one hand, and time-resolved dark and photolyzed structures on the other, show the same conformational differences (Figure 5C and 5D). Furthermore, there are no major differences between either the static deoxy structure and the refined photoproduct state, or the static CO-bound and its time-resolved counterpart (Figure 5E and 5F). In all cases differences are confined to surface side chains of residues with high temperature factors. Thus, in both difference electron density maps and refined coordinates of the photoproduct, key conformational changes between the deoxy and CO structures are represented. This confirms that bjFixLH completely relaxes to a conformation identical to the static deoxy state within 1 s after photolysis.

Figure 5 (A) Fc(deoxy) - Fc(CO) experimental electron density difference maps calculated from static structure factor amplitudes reported in ref 21. (B) Electron density difference map F(photoproduct) - F(CO) derived from SVD of time-resolved electron density difference maps (lSV1). (C) CO-bjFixLH (from static measurements, PDB code 1XJ2) colored by difference in atomic position relative to static deoxy-bjFixLH (PDB code 1XJ3). Increasing red color corresponds increasing distance between refined coordinates (see key at bottom of figure, distance in angstroms). (D) CO-bjFixLH refined from "dark" Laue data (without laser flash) colored by difference in atomic position relative to refined photolyzed-bjFixLH coordinates. (E) CO-bjFixLH coordinates from "dark" Laue data colored by difference in atomic position relative to static CO-bjFixLH, PDB code 1XJ2. (F) Photolyzed-bjFixLH colored by difference in atomic position relative to deoxy-bjFixLH.

A Comparison with Myoglobin. Time-resolved crystallographic studies of the oxygen storage protein myoglobin have shown that relaxation to the deoxy state takes place over a large time range, with structural events taking place from picoseconds to milliseconds (37-43). This transition occurs on a rugged energy landscape populated by interconverting conformational substates (see ref 65 for review). This complexity gives rise to stretched exponential kinetics. bjFixLH has a significantly different protein fold and ligand binding properties than myoglobin, making its structural dynamics of interest. Our results show that structural relaxation from the CO-bound to the deoxy form of bjFixLH is rapid and is complete within 1 s, despite the fact that overall CO recombination in bjFixLH is approximately 100-fold slower than sperm whale myoglobin (10). By comparison with myoglobin the energy landscape of CO-recombination in wild type bjFixLH appears relatively flat as assessed by ultrafast UV/visible spectroscopic techniques (62, 63), but becomes more complex upon addition of the distal coordinating residue found in most globins (66). However, recent femtosecond infrared measurements suggest that CO recombination in bjFixLH is more complex on faster time scales (van Wilderen, Key, et al., in preparation). Time-resolved crystallography at delay times shorter than 1 s is necessary to visualize if multiple structural processes occur following photolysis, or to confirm the apparent simplicity of this gas-modulated conformational switch.

In myoglobin, CO release from the protein matrix is characterized by a network of docking sites within the protein that are transiently occupied by outgoing CO. Sites have been found in myoglobin using static (67-71) and time-resolved techniques (37, 38, 40, 41). For example, the half-life of the CO docked at the proximal (Xe1) docking site is ~10 s (see Figure 11 of ref 29). A transiently docked CO molecule in bjFixLH with a similar half-life and peak occupancy would be detectable in our experiment. We therefore examined difference maps from 1 to 10 s after photolysis to identify comparable CO docking sites in bjFixLH, in which CO would appear as a positive electron density difference peak. However, we could not positively identify any, despite the presence of a small hydrophobic cavity adjacent to the heme, bordered by the side chains of Phe176, Met234, and Val253 (not shown). Unlike myoglobin for which several cavities exist, this is the only cavity we detected in bjFixLH, and it might be too small (2.1 ų) to transiently accommodate an outgoing CO molecule. Further experiments using xenon binding (67) or cryotrapping (68, 69) may be necessary to determine if other cavities exist in bjFixLH in crystals.

Implications for Signal Transduction. Prior studies of bjFixLH have suggested an ensemble ligand detection mechanism in which the structural signaling mechanism is not restricted to a single region of the protein and displays semiredundant aspects (21). Structural changes noted in the FG loop and in the -sheet distal to the heme are coupled to the movement of the main chain atoms of Leu236 in the I strand, which allows the FG loop, the distal coordinating residue for O₂ (Arg220), and the -sheet residues to act as a group. We observe significant difference density near Leu236 after exit of photolyzed ligand from the protein, and the associated shift of the residues of the H and I -strands. Conformational changes are apparent at propionate 6 and in the FG loop residues Ile215 to Ile218. Thus the conformational elements of this ensemble sensing mechanism are present in our time-resolved difference maps: doming of the heme and subsequent propionate 6 movement and ligand shape sensed by steric clash of Leu236 with the bound ligand. Though additional support for an ensemble mechanism of this type has recently been reported (65), the means by which these structural changes inhibit kinase activity remains unclear.

Acknowledgment

J.K. thanks Sudarshan Rajagopal and Marius Schmidt for assistance with SVD analysis, and Friedrich Schotte and Philip Anfinrud for preparation of brightness-mapped electron density maps.