Macromolecular Architecture Of The Synapse

National Institute Of Neurological Disorders And Stroke

Investigators

Linked publications, trials & patents

Abstract

The postsynaptic density (PSD) at excitatory glutamatergic synapses is an extensive molecular machine and a key site of memory, information processing, and storage. To investigate the PSD, we rely on a method we developed to freeze-substitute dissociated rat hippocampal cultures and examine them in thin sections by EM tomography to show individual protein complexes within the PSD. Our EM tomography revealed that the core of the PSD is an array of membrane-associated, vertically oriented filaments containing the scaffold protein PSD-95 in an extended configuration and a polarized orientation. Furthermore, multiple horizontally-oriented protein filaments are linked to the vertical filaments. There is a differential distribution of two major types of glutamate receptors, namely NMDARs and AMPARs, in the PSD with NMDARs clustered at the central region of the PSD surrounded by AMPARs at the periphery, the latticework of filament scaffolding appears to link and support the distribution of these receptors. We have adapted two powerful techniques for identifying individual proteins in tomograms. In the last year, we made strides in developing new approaches with nanobodies. This allows us to gold particle label synaptic proteins of interest to identify them directly in tomograms. We have also developed an alternative to labeling proteins by immunogold. We have found much improved tag labeling procedure for EM tomography using APEX, a cloneable horseradish peroxidase, which catalyzes the oxidation of DAB into electron dense material in the presence of hydrogen peroxide. We developed an application method such that the reaction product is more localized to the APEX tag rather diffusing away and filling their cellular compartments. We labeled CaMKII, which is a kinase and the most abundant protein in the brain. Further, CaMKII is required for learning and memory. When activated by synaptic activity, CaMKII ubiquitously phosphorylates numerous synaptic proteins. CaMKII holoenzyme is assembled by 12 identical CaMKII subunits into a relatively large protein, which we reason would be easier to identify with APEX. Expressed in rat hippocampal neurons and imaged by dark-field STEM tomography, our CaMKII-APEX2 construct clearly labeled individual CaMKIIs. With this technique, we are learning about the distribution of CaMKII in spines and their response to synaptic activity down to the level of individual holoenzymes. This new approach is now being expanded to other relatively large PSD proteins such as Shank and Homer. Going forward, thanks to our access to NIH cryo-EM facility, we now have success using cryo-EM tomography to study unitary structures in PSD fragments created by sonication of isolated PSDs from rat brain; we are also using cryo-EM tomography to study frozen, hydrated PSDs in synapses from cultured hippocampal neurons. We are tremendously helped by our access to the NIH-HCP supercomputer cluster for data transfer and processing. We expect that the combination of our current EM techniques with cryo-EM will result in new discoveries. Our trans-synaptic project found contiguous connectivity of structural elements from the postsynaptic through to the presynaptic compartment. In fact, structures can be traced to connect postsynaptic scaffolding to presynaptic vesicles through the cleft. We hope to understand the composition and distribution of these trans-synaptic assemblies and that this will, in turn, provide an understanding of the basic construction of the synapse and limits of trans-synaptic alignment. In the last year, we have added three datasets to our collection of arduously segmented and annotated trans-synaptic synapses. These datasets include two inhibitory synapses, which is a departure from our typical focus on excitatory synapses in hippocampal cell cultures. Including inhibitory synapses has been illuminating when contrasting the structure and distribution of assembly components with excitatory assemblies. For instance, full intercellular assemblies outside the synapse tend to be one specific size, with assemblies within the synapse being up to six-fold larger. Within the synapse, excitatory trans-synaptic assemblies are three-fold larger than inhibitory trans-synaptic assemblies and more evenly distributed in the synapse. Inhibitory trans-synaptic assemblies tend to locate closer to the outside edge of the synapse, leaving the central area of the synapse sparse. We are excited to study the underlying morphological differences that underpin these findings. Raw data segmentation and annotation for this project have ended. In the coming months, the roughly five thousand accumulated structures segmented from the tomographic datasets will be analyzed. From here, we intend to find the distribution of assembly build-types in both excitatory and inhibitory synapses. We believe the differences between the synapse types will shed light on the function of these assembled structures. Last, we have accelerated segmentation and visualization of synaptic structures by automation. Tomographic projects in our lab require extensive manhours to process. To resolve this issue going forward, we are developing new automated segmentation techniques, with two distinct applications: analyzing synaptic fragments and segmenting individual synaptic macromolecular structures. Application of our automatic segmentation optimization method (ASOM) obtained detailed structures of fragments from sonicated PSD isolates imaged by cryo-EM. ASOM automatically removes noisy parts from each of our segmented PSD fragments. Furthermore, in order to segment tightly packed granular structures in intact PSDs, we improved ASOM by combining watershed segmentation widely used to separate connected structures automatically. Its application enabled automatic separation into individual modules from connected granular structures. More than 50% of the modules were within the range between 40 nm and 50 nm suggesting that the fragments are generated by mechanical separation of modular structures rather than mere degradation of the synapse due to sonication. ASOM is efficient for automatic segmentation of individual structures, but it does not allow us to automatically segment structures belonging to one class like filaments connected to the postsynaptic membrane like PSD-95 filaments, those connected to the presynaptic membrane, and those connected to both the membranes, etc. as a single step. Because the filaments vary in length and shape even though they come from the same class, they have been segmented by hand requiring tremendous time and effort. Here we improved ASOM further combining skeletonization, which has been an important process for pattern recognition. The improved ASOM allowed us to automatically segment filaments connected to only the postsynaptic membrane without any bifurcation as one step revealing that PSD-95 filaments are segmented by automation. Furthermore, the improved ASOM automatically segmented other distinct classes such as those connected to the presynaptic membrane, postsynaptic membranes, and/or vesicle membranes. We found that some of the automatically segmented filaments were linked to each other across the synaptic cleft forming complete tripartite assemblies including incomplete ones. The findings are consistent with those assemblies obtained by painstaking manual segmentation suggesting that this approach will expedite segmentation of the complete and incomplete assemblies and, thus, their detailed characterization.