Macromolecular Architecture Of The Synapse

点击翻译按钮获取中文摘要

Progress Summary: A key priority for this project is to develop methods of determining molecular identity within tomograms of the PSD. First, we developed a new technique using nanobody labeling for tomograms. A nanobody binds directly to the target protein with high specificity which allows us to use gold particle labels on synaptic proteins and identify them directly in tomograms. We have data on PSD-95, CaMKII, and Homer1b, and we are in the process of finishing this work and preparing an initial publication for this type of work. Second, we succeeded in fine-tuning a genetic labeling procedure for EM tomography using APEX2, a cloneable horseradish peroxidase, which catalyzes the oxidation of DAB into electron-dense material in the presence of hydrogen peroxide. CaMKII is a kinase required for LTP and is the most abundant protein in the brain. Using this APEX2 method in rat hippocampal neurons imaged by dark-field STEM tomography, individual APEX2 labeled CaMKII are readily identified in tomographic reconstructions of dendritic spines. As a result, we are beginning to understand the distribution of CaMKII in spines, on the membrane, and in the PSD, their self-association, and their response to synaptic activity. We are preparing this work for publication. We have made great strides in characterizing synaptic structures using high-pressure freezing and freeze substitution (HPF/FS). However, HPF/FS is time intensive with many points of failure and relies on stain to visualize the structure. With cryo-EM tomography (cryo-ET), we can trade a slight degradation in resolution for ease in preparation and pure visualization of structure. We are experimenting with three cryo-focused methods in collaboration with NIH cryo-EM facilities (NICE and MICEF). First, we acquired over 150 tomograms on the NCI 300 kV Krios cryo-ET microscope, specifically on frozen-hydrated isolated PSDs from rat brains or sonicated PSD fragments. Second, we are also using the cryo-focused ion beam (FIB) milling. With FIB milling we can shave neuronal processes and synapses to obtain 200-300 nm thick lamellar. Third, we are freezing synaptosomes isolated from the brain. Our goal with each cryo method is to develop the method to a point where we achieve results comparable to our HPF/FS techniques. Our transsynaptic assembly project investigates intracellular structures linked by cleft-spanning structures. In this project, transcleft structures and all connected transmembrane and intracellular structures are segmented and analyzed in tomograms of synapses from high-pressure frozen, freeze-substituted neuronal cultures. In renderings, cleft-spanning structures typically make continuous connections from one intracellular compartment to the other, forming what we call transsynaptic assemblies. This project has yielded several clear findings. First, nearly all transcleft objects have some intracellular component. Second, transsynaptic assemblies with large intracellular volumes and more than one intracellular component are very likely to be associated with synaptic vesicles. Third, transsynaptic assemblies share intracellular components and produce large domains of associated assemblies or just association domains. We believe association domains explain the underpinnings of the nanodomain phenomena and reveal a more complex picture of their composition and function, as our results show that less than half of assemblies associate with synaptic vesicles. Further, we classified and enumerated over three thousand structures. We designed an algorithm to display a structure at random, prompt the user for a description, and then parse descriptions for common morphological elements. We were able to use this information to find common structures associated within assemblies. This work was published this year. We hypothesize that there are functionally critical transsynaptic combinations of presynaptic, postsynaptic, and cleft molecules. Unfortunately, we do not have the technology to analyze the number of synapses necessary to confidently identify these with satisfying specificity. To get the number of structures necessary, we are pairing this project with the automated segmentation project. Automated segmentation is crucial for the future of previously discussed projects and electron tomography in all forms. With our automated segmentation project, our goal is to accelerate the segmentation and visualization of synaptic structures with automation. We developed an automatic segmentation optimization method (ASOM). With ASOM, we are processing many large tomograms. For one project, ASOM segmented detailed structures of fragments isolated from sonicated and control PSDs imaged by cryo-EM. However, many structures within PSDs segmented by ASOM are still interconnected in complicated ways. To examine those structures more in detail, we improved ASOM further by adding watershed segmentation, widely used to separate connected structures automatically. This enabled the automatic segmentation of hundreds of tightly packed granular structures in intact PSDs into individual modules. These results were recently published. Recently, we improved ASOM to automatically segment filaments connected to only the postsynaptic membrane in one step, revealing that PSD-95-like filaments can be segmented by automation. The improved ASOM automatically segments other distinct classes such as those connected to the presynaptic membrane, postsynaptic membranes, and vesicle membranes. Further, the automatic segmentation of transsynaptic components are consistent with those assemblies obtained by painstaking manual segmentation, demonstrating that this approach will contribute to expediting the segmentation of the assemblies. We need a platform for users to apply ASOM algorithms. Most segmentation tools do not have the most basic functions of ASOM. Therefore, we have been developing our own software package that streamlines the entire tomography data pipeline, from image alignment to visualization. In addition to ASOM, we will integrate object classification algorithms based on machine learning and AI. Our goal is to decrease the time needed to fully analyze a tomogram from several months to a few days. Recently, we successfully implemented an advanced reconstruction method called the Simultaneous Iterative Reconstruction Technique (SIRT), widely used for generating tomograms. Our method has improved accuracy and reduced noise for both conventional ET and cryo-ET. Our SIRT method produces EM tomograms more efficiently than IMOD while qualities of the tomograms were found to be equal to or better than those generated by IMOD. Also, we combined ASOM with skeletonization and found that ASOM automatically segmented distinct transsynaptic structures similar to those segmented by hand. More work is required to provide the same level of detail as hand segmentation. Currently, we are combining the advanced ASOM algorithm with new AI algorithms that are potentially applicable to various subcellular structures. This project will make our software package a more efficient and robust segmentation platform for more varied structures, thus expanding the scope of our investigations to include large amounts of datasets and various experimental conditions.