Our dream of watching a protein function in real time with 120 ps time resolution and near-atomic spatial resolution was first realized in 2003 using picosecond time-resolved Laue crystallography, an experimental methodology developed by the Anfinrud group at the European Synchrotron and Radiation and Facility (ESRF) in Grenoble, France. To advance this capability further, we initiated in 2005 a major effort to develop the infrastructure required to pursue picosecond time-resolved X-ray science at the Advanced Photon Source (APS) in Argonne, IL. This effort is summarized in a separate report. One of the critical components in this effort is TReX, an in-house software package designed to analyze time-resolved Laue data. This effort has proven to be much more demanding than envisioned at the outset, but represents a critical component in our research. We are currently overcoming remaining issues and are working on a major update to this software package: TReX-II. The first step in analyzing crystal diffraction data, whether acquired with monochromatic or polychromatic X-ray radiation, is the indexing of diffraction spots recorded on a two-dimensional detector. Once indexed, the Laue pattern can be predicted, after which its spot intensities need to be integrated, scaled, and merged with results from numerous crystal orientations. Finally, the merged results are Fourier transformed to generate time-resolved electron density maps. The integration methods employed by our group thus far are inspired by PROW, a software package developed by Dr. Dominique Bourgeois for PRofile integration of Overlapping and Weak spots. We have discovered that the precision and accuracy of the integration suffers from numerous systematic errors;these errors contribute much noise to the diffraction data, and adversely affect the quality of electron density maps constructed from those data. These problems became apparent when analyzing data acquired using a novel protocol in which 37 diffraction images at time points spanning 100 ps to 100 ms with 4 time points per decade were collected at a single orientation with a single crystal. To minimize the extent of radiation damage, which would destroy the protein crystal if that many images were acquired from a single spot on the crystal, we translated the crystal stepwise along its length and thereby distributed the radiation damage across the entire length of the crystal. This protocol requires long crystals, and was first tested using photoactive yellow protein crystals 0.6-mm in length. Examination of the spot intensities across the extensive time series unveiled numerous sources of systematic errors that compromise the accuracy of the integration methods currently employed by TReX. We are in the process of breaking away from PROW-based methods and are developing new integration and scaling methods that avoid the pitfalls that have plagued this and other Laue processing software packages. Significant progress has been made this past year in the development of TReX-II. We have implemented a ratio method for processing the diffraction data, have developed novel Singular Value Decomposition (SVD) methods for outlier rejection and noise filtration, and are refining global analysis methods capable of extracting the structures of intermediates along the reaction pathway. In the past, Laue data processing involved merging redundant observations from numerous orientations, a process that required image scaling and wavelength normalization. Because the scaling and normalization procedures are error prone, the quality of Laue data has traditionally been inferior to data acquired with monochromatic methods. To eliminate these sources of error, we have developed a ratio method in which spot intensities from time-resolved images are divided by corresponding spot intensities from reference images acquired under nearly identical conditions. Once ratios are computed, error-prone scaling and normalization are no longer required to merge redundant data.
We aim to produce movies of protein structure changes as they function over many decades of time. To that end, we acquire time-resolved diffraction images at times spaced logarithmically with 4 images per decade 1, 1.78, 3.16, 5.62. For example, in our most recent studies of photoactive yellow protein (PYP), the time series spanned from 100 ps to 316 ms, which covers more than 9 decades of time. Reference images (-1 ns time delay) are inserted between each group of four, and at the beginning and end of the series. The reference images track changes due to radiation damage, which lead to changes in the spot intensities. Radiation damage affects the unit cell parameters, which alter the wavelength for which the Bragg condition is met. Since the undulator spectrum is sharply peaked and quite asymmetric, the effect of radiation damage on the spot intensities can be complex. Nevertheless, we found that the intensity variation across the series of reference images can be analyzed by SVD, with a relatively small number of components being sufficient to reconstruct the reference intensities. Thus, we now treat each reflection as a vector in time, and can use the SVD results to identify and reject reflections that exhibit outliers along the series. This SVD filtering approach improves the accuracy of the ratios computed, and removes the effects of radiation damage from the data. Pump-induced structure changes cause some reflections to get brighter and some dimmer, but the average ratio is expected to be unity. However, that is true for only some time points: Laser photoactivation of a protein crystal deposits excess energy into the crystal, which heats the surface and causes the crystal to bend slightly. This distortion, which develops on the 100 ns time scale, increases the mosaicity of the crystal and reduces the integrated spot intensity. Therefore, the average ratio found when the crystal suffers from distortion is less than one. To compensate for this systematic error, we rescale the ratios so that their average is unity, and thereby compensate for this systematic error. SVD analysis is also used on the rescaled ratios to identify and reject outliers, and to develop a basis set of vectors, linear combinations of which can reproduce the time-dependent ratios of all reflections. Using a high-quality PYP data set consisting of 44 time points in a series, we are experimenting with both real and reciprocal space SVD filtering with an aim to develop analysis methods that maximize the quality of the time-resolved electron density maps and the quality of the movies. Finally, a global analysis method developed by our group for processing time-resolved spectroscopic data has been extended to time-resolved electron density data. This method allows us to recover by linear least-squares methods time-independent electron density maps for each state in a kinetic model. The signal-to-noise ratio (S/N) for these maps are enhanced relative to the individual snapshots in accordance with the number of time frames in which that structure is represented. Moreover, this approach allows us to refine rate coefficients for structural change. This methodology is being developed using a recently acquired PYP data set (March 2011), and the results obtained thus far are very promising. We are working to refine a kinetic model that is fully consistent with the time-resolved structure changes observed. One of the challenges remaining is to develop objective strategies for quantifying the absolute populations of intermediates in the kinetic model.
|Kaila, Ville R I; Schotte, Friedrich; Cho, Hyun Sun et al. (2014) Contradictions in X-ray structures of intermediates in the photocycle of photoactive yellow protein. Nat Chem 6:258-9|
|Schotte, Friedrich; Cho, Hyun Sun; Kaila, Ville R I et al. (2012) Watching a signaling protein function in real time via 100-ps time-resolved Laue crystallography. Proc Natl Acad Sci U S A 109:19256-61|