DBP7: Structure and Function of Synapses

DBP7: Structure and Function of Synapses

Back to table

A. Collaborating Investigator(s): Kristen M. Harris,¹ Terry Sejnowski,² Tom Bartol, ² J. Faeder³

B. Institutions: ²University of Texas a t Austin, ²Salk, ³Pitt

C. Funding Status: G1: R01MH104319 (Harris) 9/14-07/19; G2: R01MH095980 (Harris) 7/12-06/17

D. Driving relationship between TR&D2 and DBP7: Neuronal dendrites, axons and synapses are structurally distorted in individuals with mental retardation and other neurological disorders. We would like to interpret this distortion, but dendrites and spines differ widely in their appearance and composition in normal brains. Our overall goal is to characterize this structural variation towards understanding how neurons regulate, sustain, and alter synaptic connectivity as brain function develops and changes with learning, memory, and pathology. We have made significant progress using reconstruction from serial section transmission EM (3DEM) and in developing a new approach of transmission EM on the scanning electron microscope (tSEM) that greatly increases throughput of high quality images.^105,106 We discovered that the summed synaptic surface area is balanced along hippocampal dendrites, either having many small spines with small synapses or few large spines with large synapses.¹⁰⁷ Even following substantial structural synaptic plasticity during LTP induced by theta-burst stimulation (TBS-LTP), a rebalancing occurs to achieve equal summed synaptic surface area along control dendrites and LTP dendrites with enlarged synapses. These findings lead to the hypothesis that heterosynaptic competition for intrinsic resources regulates synapse number and size along adult neurons (Fig VII.6 A).

In our funded research we are determining mechanisms underlying the developmental onset of the late phase LTP (L-LTP) lasting more than 3 hours.¹⁰⁸ We have shown that when L-LTP is induced in developing animals at postnatal day (P)15, instead of enlarging existing synapses, more dendritic spines form. P15 is an age of rapid spinogenesis, hence there was substantial resources for adding new spines, that resulted in a greater summed synaptic input than under control conditions (Fig VII.6 B). We are also investigating the role of silent synaptic growth¹⁰⁹ as a basis for the augmentation of LTP that could form the basis for the enhanced efficacy of spaced over massed learning.¹¹⁰ Critical to these studies is the improvement of alignment and reconstruction tools which will be developed by TR&D2 team members, driven by our data.

Two example findings illustrate how axonal and dendritic structure and composition support synaptic plasticity in the hippocampus, and provide high motivation for the proposed collaborative efforts with TR&D2 to enhance our data collection and analyses. We discovered that only ~20% of the axons that pass next to dendrites actually form synaptic contacts.¹¹¹ We are investigating what intrinsic rules govern the number and size of synapses supported along axons. Recently we showed that spines arising from the same dendrite synapsing on the same presynaptic axon have the same synaptic surface areas, head volumes, vesicle numbers and neck diameters.¹¹² This finding suggests a powerful structure-function relationship. Axons in CA1 stratum radiatum were evaluated with 3DEM after TBS-LTP.¹¹³ The frequency of axonal boutons with a single postsynaptic partner was decreased by 33% at 2 hours, corresponding perfectly to the 33% loss of small dendritic spines (head diameters <0.45 µm). Presynaptic vesicles as well as transport packets between boutons were reduced for at least 2 hours after TBS-LTP. These findings show that specific presynaptic ultrastructural changes complement postsynaptic ultrastructural plasticity during LTP. Smooth ER (SER) forms a membranous network that extends throughout neurons. SER regulates intracellular calcium and the posttranslational modification and trafficking of membrane and proteins. Greater SER volume, folding, or branching reduces the movement of membrane cargo and local delivery of resources increases in the vicinity of complex SER.¹¹⁴ Recently we have found that TBS-LTP initiates SER remodeling in adult hippocampal dendrites. In spines, SER volume increased and more spines contained a spine apparatus (SA), which is composed of highly folded SER. Synaptic growth was greatest at these spines. In parallel, dendritic shaft SER was less branched after TBS-LTP. Dendritic segments with no SER-containing spines had fewer neighboring spines whereas those surrounding a spine apparatus had spine densities equal to controls (Fig VII.7). Thus, dendritic spines with an apparatus collaborate with neighboring, but compete with more distant, spines for critical resources.

E. Innovations: Our past collaborative work has involved truly heroic studies, requiring a brute-force manual approach to obtain the original 3DEM reconstructions.^112,115 The editing was done section by section with iterative 3D reconstructions to confirm the edits were in the correct locations. The improved alignment, segmentation, and model-editing tools of this proposal will support the specific aims of two funded grants in the Harris laboratory. They will make these in-depth analysis more routine and allow realistic MCell models to simulate biochemical signaling in the 3D subcellular structure of dendrites, axons, spines and synapses. Surface meshes used to represent cell membranes and subcellular structures must meet very strict geometric standards (e.g. water-tight, non-intersecting, manifold) like those we have created in our recent publications. We will add to these models the effects of synaptic plasticity.

F. Methods and Procedures: We will provide high resolution images and 3DEM reconstructions to test the alignment, segmentation, and editing routines proposed in this grant. Our tissue processing and 3DEM methods have been rigorously tested and published.^105,106 Example series have been posted to the OPENCONNECTOME.¹¹⁶ We will share data and images while accomplishing the Aims of our grants G1 and G2 (paraphrased here): Aim 1 (G1): To test the hypothesis that the abrupt onset of L-LTP at P12 is associated with first occurrence of dendritic spines using our rat model systems of in vivo perfusion-fixed hippocampus, TBS-LTP in hippocampal slices, and quantitative analyses from 3DEM. Aim 2 (G1): To test whether dendritic spines are induced by the first bout of TBS at P10, after which a second bout of TBS can produce L-LTP, but not at P8 when multiple bouts do not.

Aims 1-4 (G2): Recently, we have shown that initially saturated LTP can be subsequently augmented once a couple hours elapse between episodes of LTP induction.¹¹⁷ We have identified, nascent zones, which are dynamic regions with a postsynaptic density (PSD) that lack presynaptic vesicles (Fig VII.8).¹¹⁸ Immediately following induction, vesicles accumulate at prior nascent zones converting them to active zones. By 2 hours, nascent zones have returned. We will test the roles of protein synthesis, SER, and candidate molecules in building or stabilizing synapses during saturation and augmentation of LTP.

DBP6: Constructing a dynamic, spatial map of transcription and chromatin structure

DBP6: Constructing a dynamic, spatial map of transcription and chromatin structure

Back to table

A. Collaborating Investigator(s): Daniel Larson, ¹ Carl Kingsford,² Robert F. Murphy,² Ivet Bahar³

B. Institutions: ¹NIH, ²Carnegie Mellon U, ³Pitt

C. Funding Status of Project: NIH, NCI 1ZIABC011383-05 (Larson) (2011- )

D. Driving relationship between DBP6 and TR&D1 and 4: This DBP drives Aim 1 of TR&D4 and Aim 2 of TR&D1. The project stems from funded work in the Larson laboratory to study the dynamics and heterogeneity of genome structure as it relates to gene expression by systematically measuring the position, mobility, and transcription of genes in the human nucleus using multi-color imaging of nascent RNA in living cells. A key innovation of this DBP is the ability to image the history of transcriptional activity at two loci (Fig VII.5A-B). This two-gene assay is likely to become a crucial tool in imaging gene expression. These measurements will drive the development of new image-analysis (TR&D4) and structural modeling (TR&D1) tools. The Larson lab (Fig VII. 5A) will produce 3D movies where fluorescent spots reveal the location and intensity of transcription for pairs of genes. Analysis of these movies requires the development of new segmentation, registration, and modeling techniques to create models of the spatiotemporal relationships among labeled loci and between labeled loci and cellular landmarks (TR&D4). No existing packages can automatically process and model 3D movies of RNA transcripts and chromosomal domains in living cells. Images of 10,000 individual MCF7 cell lines, each with a pair of gene loci whose transcription is monitored with different fluorescent proteins, will be generated. This scale requires the development of fully automated image analysis techniques to create maps of the relationships between transcribing loci, which will drive TR&D4 Aim 1 and be incorporated into CellOrganizer.

Analysis to identify loci of coordinated and bursty translation from these maps will require the development of new computational and statistical techniques. No existing packages can identify such events from 4D traces of transcription, hence driving Subaim 2.2 of TR&D1. We will develop applications to take point traces derived from the image analysis to produce models of transcription activity conditioned on the position and transcription activity of nearby loci. From these models, significant transcription burst events will be detected. Construction of these models will require new methods for distance imputation, detection of high-confidence, reproducible distances, and clustering of measurements into heterogeneous structural classes. We have developed an open-source suite of analysis tools (Armatus-3C)⁹³ for genome structure measurements from chromosome capture data that will be significantly extended and adapted into a suite suitable for the dynamic point traces collected here (TR&D1). Finally, the dynamic imaging data produced by this DBP will provide a validation set that will drive Subaim 2.1 of TR&D1 for the application and extension of elastic network models (e.g. GNM) and ProDy API^94,95 (originally developed for protein dynamics) to evaluate chromatin motion from more widely available, but static, chromosome conformation capture measurements (e.g. Hi-C).

E. Resulting Innovations: This DBP will drive the following technical advancements and deliverables: (1) Pipeline for processing of 3D movies including a machine learning system for optimizing segmentation to a new cell/probe system. (2) Software package for fully automated construction of probabilistic point traces and probabilistic location maps from tens of thousands of 4D movies containing fluorescently labeled transcribing loci. (3) New analysis techniques, based on point process statistics and conditional models, to identify burst events consisting of co-localized co-transcribing genes, and a software package for the reporting of these events. (4) Transfer, extension and validation of GNM analysis tools to predict the motions of gene loci and their correlations (Fig III.9 in the TR&D1 section).

F. Methods and Procedures: Generation of live cell 3D movies for 10,000 gene pairs in estrogen-responsive MCF7 cells in order to better understand position, mobility, and transcription of genes. This approach is based on insertion of a DNA cassette that codes for RNA hairpins.⁹⁶ When transcribed, the RNA hairpins are specifically bound by a fluorescent phage coat protein, resulting in a fluorescent ‘spot’ which represents the active gene. The original MS2 system for RNA visualization has been in use for nearly 2 decades, but a second stem loop system based on the PP7 phage was recently developed and shown to be orthogonal to the MS2 system.^97-99 Using this concept, Larson has made successive advances that enabled: observation of the transcription of single genes in yeast and human cells,^100,101 simultaneous visualization of two segments of an individual RNA by labeling an intron and exon of a single transcript⁹⁸and direct insertion of stem loops into an endogenous locus in human cells using gene-trap technology.

Specific Aim 1: We will develop and apply fully automated image processing approaches (TR&D4) to create models of the spatiotemporal relationships between labeled loci and cellular landmarks using movies of 10,000 individual MCF7 cell lines, each with a pair transcribing gene loci, monitored with different fluorescent proteins. Although many of the analytical tools needed to analyze time-series images of this type have been implemented in some fashion in previous studies,^98,102 these implementations were low throughout or relied on user intervention. This project will drive the development in TR&D4 of high-throughput, fully automated and adaptable tools, thus providing tools that can systematically integrate the spatial, temporal and functional components of gene regulation. These tools will be developed in the context of live cell imaging, but will be useful to the community for other datatypes as well.

Specific Aim 2: We will identify bursting and co-localized transcriptional events from the location maps derived from Aim 1 via the development of a framework for identifying confident, significant spatial relationships between loci. We will impute missing distances and construct a conditional random field (CRF) that models expression as a function of neighboring measured genes and their relative distances. Bursty events will be those sets of genes with high joint probability of being expressed when their spatial distances are small. Point process statistics will also be used and extended to identify spatial clustering.

Specific Aim 3: We will relate predicted motion derived from MCF7 Hi-C measurements using extensions of GNM modeling techniques (TR&D1) to the observed motion via live cell imaging. Mobility measures such as mean square displacement (MSD) and distance fluctuations between pairs of genes, step-size distributions, and angular displacements^103,104 will be computed from the segmented transcription sites. The two modalities of measurements (Hi-C vs. imaging) will be used to confirm conclusions.

DBP4: Spatiotemporal modeling of T cell signaling

DBP4: Spatiotemporal modeling of T cell signaling

Back to table

A. Collaborating Investigator(s): Christoph Wülfing,¹ Peter Cullen,¹ Paul Verkade,¹ Robert F. Murphy,² Deva Ramanan,² James Faeder³

B. Institutions: ¹University of Bristol, ²Carnegie Mellon University, ³Pitt

C. Funding Status: European Res Council PCIG11-GA-2012-321554 (Wülfing) 8/12 – 7/16; Wellcome Trust 102387/Z/13/Z (Wülfing) 8/14 – 9/17; Wellcome Trust 201254/Z/16/Z (Wülfing) 5/16 – 4/19; Medical Res Council GW4 BioMed DTP (Wuelfing) 9/16 – 8/19; Wellcome Trust Senior Investigator Award 104568/Z/14/Z (Cullen).

D. Driving relationship between TR&D and DBP. This project drives Aims 2 and 3 of TR&D4 and Aim 1 of TR&D3. During the current grant period, this DBP has resulted in the addition of important new capabilities to CellOrganizer. The first is a new class of generative models for proteins that are not contained in organelles, and in which the ability to compare protein models constructed from different images/movies is achieved through registration to a common event (in this case, formation of the immunological synapse). The second is the ability to perform causal modeling of spatiotemporal relationships between sensors. This will be further extended in the proposed renewal, and new capabilities will be developed, as described in detail below.

E. Innovations: DBP4 has led to new insight into actin dynamics at the T cell synapse in the tuning of T cell activation by costimulation; will provide a new paradigm for learning models of signaling pathways.

F. Methods and Procedures: The goal of DBP4 in the original MMBioS proposal was to build models enabling the understanding of the way in which various molecules in the T cell signaling system influence each other in time and space. Such mutual influence is critical in the tuning of T cell activation. While antigen recognition is mediated by the T cell receptors (TCR), the outcome of T cell activation is greatly influenced by how the TCR signal is amplified or suppressed with critical roles in autoimmune disease and the immune response to cancer. The specific aims were: (1) To construct models of the relationship between the spatiotemporal distributions of potential signaling molecules; (2) To construct biochemical cell simulations that capture the sequence of events involved in T cell signaling. The first aim consisted of two subaims, both of which have been accomplished. The first built on initial work aimed at creating registered maps of the spatiotemporal distribution of signaling intermediates just before and after formation of a synapse between a T cell and an antigen presenting cell. Many complications were encountered with aligning and morphing many different proteins to a common template so that they could be compared. These were all overcome, the pipeline has been created as a new functionality in CellOrganizer, and complete maps have been obtained for 9 proteins (actin and 8 actin regulators) under two conditions, a strong T cell stimulus where the TCR signal is amplified through costimulation as compared to a stimulus where the dominant costimulatory receptor, CD28, is blocked. Resolving actin regulation by costimulation, WAVE2 and cofilin showed the most significant changes (Fig VII.4) when comparing maps, a result that would not have been possible without the quantitative comparisons made possible by the computational pipeline. Cell biological reconstitution experiments confirmed their roles. This work has led to a major paper recently published in Science Signaling.⁸³ The 2nd subaim was to learn the relationships between the spatiotemporal patterns of different proteins. We adopted a more systematic approach to construct a generative model that captures the relationships between different sensors in different regions of the cell. A manuscript describing this work has recently been submitted. These methods are described in TR&D4 along with the proposed further developments. Work on the second aim was delayed while the maps were refined, but will be completed by the end of the 5th year.

The dramatic success of the first aim has led us to propose significant evolution in the project over the following five years. We propose to add an investigation of the attenuation of T cell signaling by inhibitory receptors⁸⁴ as enabled by collaborations recently established at the University of Bristol. This is of great scientific importance as inhibitory receptors both limit autoimmune disease and suppress the immune response to cancer.^85-89 We also propose to complement our cell-wide subcellular signaling distributions with imaging of vesicles and proteomic data. The majority of the cellular pool of inhibitory receptors resides in vesicles, incapable of engaging their ligands.^90,91 Vesicular trafficking to the plasma membrane thus is of great importance in inhibitory receptor function. In addition, as mechanisms of inhibitory receptor trafficking and signaling are largely unresolved, proteomic approaches to complement the imaging data will provide the necessary genome-wide coverage for unbiased functional discovery. Dr. Wülfing has recruited two biological collaborators from the University of Bristol, Dr. Peter Cullen, a leader in vesicular trafficking and proteomics, and Dr. Paul Verkade, a leader in electron microscopy (EM). The goal is to construct accurate spatiotemporal simulations of T cell signaling that consider both costimulatory and inhibitory receptor signaling. The specific aims are:

Aim 1. To construct a predictive model of the biochemical interactions among molecules involved in T cell signaling and to test and refine that model using experimental manipulations. This aim will drive Aim 2 of TR&D4 and Aim 1 of TR&D3. We will begin by constructing a tentative reaction network of the molecules for which we have detailed maps, use initial estimates for rate constants, carry out reaction-diffusion simulations, compare the resulting molecule distributions to our maps, adjust the rate estimates, and iterate. Information from other sources will be added to provide initial values or constraints. Dr. Wülfing’s group will provide movies of additional signaling molecules as prompted by the models. To allow incorporation of vesicle-based inhibitory receptors, Dr. Wülfing’s group will provide TIRF data on inhibitory receptor insertion into the plasma membrane and imaging data on the distribution of groups of vesicles. Dr. Verkade will complement these data by immuno-EM to quantify inhibitory receptor localization. Dr. Cullen’s group will collect proteomics data to characterize vesicles containing specific immunoregulatory receptors through the identification of their luminal contents. This will allow separation of vesicles into classes with distinct compositions. The information that specific groups of molecules are trafficked in distinct vesicles will be used to further constrain the proposed reaction network. Taking advantage of Dr. Cullen’s expertise in the regulation of vesicular trafficking, specific vesicle populations will then be perturbed and the information used to further refine the network. Dr. Wülfing’s group has already generated first constructs to manipulate the localization of key signaling intermediates and inhibitory receptors, and will collect movies for signaling molecules after expression of these constructs. It is important to note that there is previous work on modeling and simulating T cell signaling for an experimental system in which antigen is presented on a flat surface to T cells,⁹² a form of T cell activation that imposes substantially different biophysical constraints on T cell signaling than the more physiological interactions with APC that we analyze. We will compare both models.

Aim 2. To explore the relationship of synapse geometry to signaling pathways. 2 This aim will drive aim 3 of TR&D4. Dr. Verkade’s group will collect large sets of electron microscope images of synapse geometry at various times during signaling under various conditions. Models of this geometry will be created and used to improve the simulations above. They will also collect immune-EM images for specific molecules so that the distributions between fluorescence and EM can be registered. Dr. Wülfing’s group will provide the samples for the EM analysis for various conditions and constructs.