Electron microscope tomographic reconstruction has proven to be an invaluable technique for studying the three structure (Perkins et al., 1997a and b). This project area includes four specific projects: 1) development of methods for processing large tomographic data at remote facilities; 2) implementing procedures that enable acquisition and processing of data from multiple tilt axes to improve the z axis resolution; 3) examining tomographic reconstructions of computer generated models; and working on the development of procedures for serial tomography. During the last year we have focused on the following: Double axis tilt reconstruction: The Z resolution in electron tomography can be significantly improved by employing a double axis tilt procedure. In this acquisition method, images are acquired by tilting on the y axis and then rotating the specimen 908 with a rotating specimen holder and performing a second tilt series acquisition. With a 160 tilt series, the loss of frequencies is reduced from 33% (single axis) to 16.1%, thereby increasing the Z resolution. In fact, this value is very close to that obtained using the more image intensive and difficult conical tilting procedure (13.4% missing region). We are implementing two different procedures for deriving reconstructions from double-tilt data that have recently been reported. We have the obtained programs from David Mastronarde at the Boulder NCRR funded Electron Microscope Resource and are experimenting with his procedure and integrating it into our tomographic processing stream. We are also developing software compatible with a our tomography programs to perform another double technique from the Albany resource. We plan to evaluate both these methods for double tilt tomography on models and data specimens generated at NCMIR. Methods for improved tomographic reconstruction and processing large tomographic datasets at remote facilities: Although, we have generally employed the R-weighted backprojection algorithm to reconstruct the 3D volume from the tilt-series projections, we have begun investigating the application of iterative techniques such as ART or SIRT following the R-weighted method to improve the reconstruction, using programs written at NCMIR. The application of the combination of the R-weighted and iterative methods to large reconstructions is computationally intensive. In order to expedite processing we have implemented the R-weighted, ART and SIRT reconstruction algorithms on the 400-node Intel Paragon, and more recently, on the 256-processor Cray T3E parallel supercomputer at SDSC. This parallel implementation was partially supported by the NCRR funded National Biological Computing Resource (NBCR) at SDSC. These single-axis tilt reconstruction algorithms are relatively straight forward to implement on massively parallel supercomputers. The reconstruction of each of the one voxel-thick planes orthogonal to the tilt axis of the volume is assigned to an individual node on the parallel computer. Using a small number of nodes (16-32) to minimize queuing delays, the parallel implementation is ten times faster than that of a high-performance, single-processor machine such as the Silicon Graphics Incorporated workstation with an R10000 processor. A script, written in PERL, enables the researcher to define processing parameters easily and to initiate the reconstruction remotely on the parallel computer, thereby bypassing the complexities of the interface to the parallel machines. During this last year the parallel reconstruction program has been used by investigators in collaborative projects examining 1) changes in the structure of dendritic spines following loss of synaptic input, alterations in the structure of cardiac muscle in an animal model of heart failure, and 2) an analysis of the complex three-dimensional structure of mitochondria, 3) synaptic transmission in cultured neurons and frog-hair cell receptors. In addition to facilitating research projects using tomography, the increased speed of computation afforded by the use of parallel supercomputers will be useful in comparing the relative merits of various reconstruction methods and in determining the optimal parameters for a given reconstruction algorithm, e.g. the number of iterations (see below). The design of the parallel tomography program is modular. therefore, it is relatively straight forward to incorporate more recent single-axis, tilt-reconstruction algorithms that may further improve the quality of tomographic reconstructions. We are currently implementing a method developed by Jose Carazo and colleagues using spherical basis functions (""""""""blobs"""""""") which is reported to decrease artifacts and improve the resolution of features within the reconstruction. Improved autotomography: We are now implementing a low-dose autotomography data acquisition system from Hans Tietz. We have implemented an interface to the IVEM microscope computer and computerized stage. Over the next year, we plan to adapt this control package to accommodate the larger image format of the 2k x 2k camera system (described above in the section on development of the slow scan camera and interface).
Showing the most recent 10 out of 384 publications
