Developing 3D novel edge detection and particle picking tools for electron tomography

Over the past 3 billion years, photosynthetic organisms including higher plants, cyanobacteria and microalgae have evolved intricate light capturing interfaces capable of harnessing the huge energy resource of the sun (>2000x global energy demand) to produce the biomass, food and fuels which supports life on earth. As the global population expands, and with it food and fuel demand, it is becoming increasingly important to understand the structure and function of the complex and dynamic machinery of photosynthetic organisms. This is because these intricate photosynthetic systems provide invaluable blueprints for the design of next-generation solar-driven microalgae-based and bio-inspired artificial systems. Microalgae-based systems can be located on non-arable land to produce food, fuel and high value products, in many cases using salt water. Microalgae systems have already achieved demonstration scale and the ability to produce crude oil at a price of ~$230 barrel. Through further optimization, renewable oil prices of $100 barrel could be achievable, opening the path to commercial deployment. The first step of all biofuel production is light capture and optimizing its efficiency. However, this requires a detailed structural knowledge of photosynthetic interfaces spanning the cellular to atomic resolution range. The advances of diverse, multi-scale imaging techniques, including high-resolution single particle analysis (SPA), crystallography and electron tomography now provides a path to reveal the structure and dynamics of the photosynthetic machinery. Electron tomographic data provide unprecedented opportunities to resolve cells to molecular resolution while modern SPA, electron and X-ray crystallography and NMR can resolve the atomic resolution structure of proteins. This project is focused on bridging the gap between these techniques by developing advanced edge detection algorithms for automated tomogram segmentation to the molecular level, to facilitate molecular docking of atomic resolution protein structures and so, the development of atomic resolution atlases of the photosynthetic machinery. Chapter 1 reviews the technology drivers for the development of solar fuel systems, the process of photosynthesis, advances in imaging and image processing technologies. The noise contamination of these images and more specifically the different types of noise reduction algorithms used to reduce its effects are reviewed. Edge detection and segmentation algorithms including manual and semi-automated segmentation, thresholding, gradient-based edge detectors, canny edge detetcor, the snake algorithm, the watershed transform, bilateral edge filter, Laplacian of Gaussian (LoG) and arbitrary Z-crossings are next reviewed to evaluate the state of the art in terms of tomogram segmentation and challenges that must be overvome to achieve molecular segmentation. Chapter 2 describes the design and development of the automated bilateral edge (3D BLE) filter for detection of macromolecular complexes. 3D BLE was found to detect objects with edge widths as low as 2 pixels. The 3D BLE filter was shown to perform well and speed up automated segmentation significantly. Chapter 3 describes the design, development and performance testing of the Rapid, automated 3D Z-crossing algorithm (RAZA). The advantage of RAZA over the 3D BLE filter is that it yields mathematically discrete contours that delineate each segmented object. In case of RAZA, each object has defined parameter values in terms of height, width, length, surface area and volume, which can be used as structural fingerprints. The 3D BLE produced non-connected edges, where as RAZA was able to deal with this problem. RAZA successfully detected the Golgi apparatus, mitochondria and mature insulin granule comparative to objects in a benchmarking test set of manually-segmented pancreatic beta cells and also resolved the contours of macromolecular objects such as ribosomes and the extrinsic domains of membrane proteins. Chapter 4 further refines the concept of using structural fingerprints to automate the selection of specific classes of objects in tomograms. This provides a bridge between electron tomography, SPA, electron and X-ray crystallography and NMR on the path to the development of atomic resolution cellular atlases. Specifically, the novel RAZA particle selection (RAZAPS) method was developed. RAZAPS defines each segmented object according to a set of measurable, geometric parameters and based on these, all similar particles of interest were searched, identified and quantified in an automated manner. As examples, GroEL truth set as well as mitochondria and ribosome-like macromolecules in cellular tomograms were used to perform automated quantification of organelles and complexes of interest. It was also successfully shown that using a few ribosomes like templates, thousands of molecules having similar structural characteristics could be extracted and contoured for tomographic sub-volume averaging. In chapter 5 this work and future work on the path to detecting membrane proteins in membranes is summarized to direct the next phase of work of resolving photosystems in thylakoid membranes. The use of advanced direct electron detection systems and denoising filters in combination with RAZAPS now make this a tantalizing prospect which could significantly assist with the targeted engineering of the photosynthetic apparatus of microalgae and assist with the design of bio-inspired solar fuel systems of the future.