Periodic Reporting for period 3 - MolDynForSyn (How dendritic mRNA and protein distributions shape synaptic plasticity)
Periodo di rendicontazione: 2023-09-01 al 2025-02-28
Understanding the computational principles of molecular dynamics inside neurons is essential because these give rise to synaptic plasticity, and any molecular deviation can impair circuit function. However, it is also very challenging because the molecular dynamics of different molecules, mRNAs, and proteins inside the dendrites of neurons need to be characterized and conceptually understood. Three aspects make this problem particularly challenging. First, one needs to build data analysis tools that can evaluate spatial distributions of mRNA puncta and protein concentrations across the dendrites and synapses to compare any two molecular distributions systematically. Second, we need to find common computational principles underlying the diversity of molecular distributions in dendrites to go beyond the characterization of individual molecular species. Third, we need to solve the conceptual challenge of how to connect the diverse molecular dynamics to the net synaptic plasticity outcomes unfolding across multiple synapses while integrating different spatial and temporal scales of investigation into a computational framework and linking these changes observed in healthy and disease phenotypes. This is why
understanding the molecular dynamics underlying synaptic plasticity is one of the major challenges in molecular and computational neuroscience.
New advances in imaging techniques allowed in the last decade to resolve the precise copy number of proteins per spine, allowed determining the precise ratio of mRNA copy numbers soma to dendrite for many mRNA species, and provided insights into critical dynamical parameters characterizing molecular motion in dendrites such as half-life, diffusion coefficients, active transport. Thus, the time is ripe to provide a system’s level of understanding across different classes of molecules, build computational models for molecular motion in neurons consistent with and motivated by experimental data, and translate these insights into biologically plausible synaptic plasticity models. To do so, the ERC project “MolDynForSyn” proposes to discover the computational rules governing the spatial and temporal distributions and synthesize different experimental observations into a comprehensive computational framework. The MolDynForSyn project aims to test the hypothesis that energy considerations are the computational reason explaining the observation that certain protein classes move their mRNAs out of the soma and into the dendrites while other classes “prefer” somatic translation. One of the aims of the projects is to understand the computational rules of how the mRNAs are distributed across the dendrite, how the active and diffusional transport, half-lives, and other physical properties of molecules play together such that the long-term stability of synaptic protein numbers can be achieved. One hypothesis the project pursues is that the less mobile proteins may prefer dendritic protein synthesis.
Understanding computational rules underlying molecular dynamics is exciting and challenging, but more is needed to understand the function of neurons. Since synaptic plasticity is a critical ingredient enabling neural computation, it is also essential to investigate how molecular dynamics shape synaptic plasticity outcomes. To this end, the MolDynForSyn project investigates how the molecular dynamics of translation, degradation, and trafficking can explain the experimentally reported synaptic plasticity outcomes at multiple stimulated synapses. It models the spatial footprint of synaptic plasticity in the dendrite emerging from multiple simultaneous plasticity events and considers their interactions emerging from protein synthesis, degradation, and protein spread. The project first derives the multi-spine heterosynaptic plasticity rules mathematically from molecular considerations and experimentally measured spine plasticity data. The project explores the circuit-level consequences of synaptic plasticity by studying the memory and signal processing in neural networks that include the non-linear synaptic plasticity properties of individual synapses. Considering the neural activity in neural circuits with interacting plasticity events that emerge from shared molecular turnover, the project aims to understand the computational benefits of molecular resource sharing and understand how the individual features of molecular dynamics translate into circuit-level outcomes.
The project addresses the role of dendritic space by studying how the speed and magnitude of the protein response may depend not only on the physical properties of the molecules themselves but also on the dendritic morphology (e.g. number of dendritic branch points, segment lengths, branching ratios) comparing the results obtained for linear dendrites with those obtained in branching dendritic trees. Physical properties of molecules can change from health to disease conditions. For example, in Huntington’s disease, the Htt protein turnover and synthesis is CAG-repeat dependent. Therefore, the project aims to investigate, using the examples of Huntington’s and Tuberous sclerosis diseases, how changes in protein half-lives and upregulation of protein translation can alter the protein composition at the synapses.
understanding the molecular dynamics underlying synaptic plasticity is one of the major challenges in molecular and computational neuroscience.
New advances in imaging techniques allowed in the last decade to resolve the precise copy number of proteins per spine, allowed determining the precise ratio of mRNA copy numbers soma to dendrite for many mRNA species, and provided insights into critical dynamical parameters characterizing molecular motion in dendrites such as half-life, diffusion coefficients, active transport. Thus, the time is ripe to provide a system’s level of understanding across different classes of molecules, build computational models for molecular motion in neurons consistent with and motivated by experimental data, and translate these insights into biologically plausible synaptic plasticity models. To do so, the ERC project “MolDynForSyn” proposes to discover the computational rules governing the spatial and temporal distributions and synthesize different experimental observations into a comprehensive computational framework. The MolDynForSyn project aims to test the hypothesis that energy considerations are the computational reason explaining the observation that certain protein classes move their mRNAs out of the soma and into the dendrites while other classes “prefer” somatic translation. One of the aims of the projects is to understand the computational rules of how the mRNAs are distributed across the dendrite, how the active and diffusional transport, half-lives, and other physical properties of molecules play together such that the long-term stability of synaptic protein numbers can be achieved. One hypothesis the project pursues is that the less mobile proteins may prefer dendritic protein synthesis.
Understanding computational rules underlying molecular dynamics is exciting and challenging, but more is needed to understand the function of neurons. Since synaptic plasticity is a critical ingredient enabling neural computation, it is also essential to investigate how molecular dynamics shape synaptic plasticity outcomes. To this end, the MolDynForSyn project investigates how the molecular dynamics of translation, degradation, and trafficking can explain the experimentally reported synaptic plasticity outcomes at multiple stimulated synapses. It models the spatial footprint of synaptic plasticity in the dendrite emerging from multiple simultaneous plasticity events and considers their interactions emerging from protein synthesis, degradation, and protein spread. The project first derives the multi-spine heterosynaptic plasticity rules mathematically from molecular considerations and experimentally measured spine plasticity data. The project explores the circuit-level consequences of synaptic plasticity by studying the memory and signal processing in neural networks that include the non-linear synaptic plasticity properties of individual synapses. Considering the neural activity in neural circuits with interacting plasticity events that emerge from shared molecular turnover, the project aims to understand the computational benefits of molecular resource sharing and understand how the individual features of molecular dynamics translate into circuit-level outcomes.
The project addresses the role of dendritic space by studying how the speed and magnitude of the protein response may depend not only on the physical properties of the molecules themselves but also on the dendritic morphology (e.g. number of dendritic branch points, segment lengths, branching ratios) comparing the results obtained for linear dendrites with those obtained in branching dendritic trees. Physical properties of molecules can change from health to disease conditions. For example, in Huntington’s disease, the Htt protein turnover and synthesis is CAG-repeat dependent. Therefore, the project aims to investigate, using the examples of Huntington’s and Tuberous sclerosis diseases, how changes in protein half-lives and upregulation of protein translation can alter the protein composition at the synapses.
The proposal is based on three main aims. Each part deals with one key challenge related to molecular turnover in neural dendrites. Next, I will describe the progress and achievements along each dimension.
In aim 1, we aimed to analyze the spatial distributions of proteins and mRNAs to build a computational framework explaining them using a set of dynamical equations. To this end, we have developed a new software package (SpyDen) to automate analysis of molecular images and determine the spatial distribution of proteins and mRNAs along dendritic stretches and inside spines to be able to model it. The software now allows tracing dendrites, determine their width and evaluate the molecular signal along a dendritic segment of interest. SpyDen can also determine the spine outline and detect both continuous and discrete molecular signals across different image resolutions, we have also benchmarked the algorithmically obtained results against manual annotation. We have also identified newly published large-scale super-resolution scans that provide single molecule resolution for tens of different neural proteins along dendrites that will help us to determine more precisely the spine-to-dendrite exchange in our models (Helm et al Nat Neurosci 2021, Unterauer et al 2024). We processed part of this data to determine the statistics of spine-dendrite ratios across proteins and mapped these to the mRNA soma-to-dendrite localization scores for the corresponding proteins. In addition, we analyzed the variability of spine sizes in pyramidal neurons over time pre and post stimulation with respect to the molecular variability in the synapses and dendrites. Building on this finding we studied the dendritic vs synaptic pools of plasticity related proteins and found that they are consistent with a lognormal distribution that has been previously observed for spine sizes. We have derived a system of differential equations that are valid for both endogenous distributions and plasticity related dynamics. We have derived the energy budget and optimal localization profiles for mRNAs and proteins as a function of their physical parameters such as half-times, diffusion coefficients and cross-validated five model predictions with experimental data. These results indicate that mRNAs are localized in the dendrites rather than remaining in the soma if the corresponding protein has a short half-life, a long amino acid chain or if the mRNA is long-lived.
In aim 2, we found that synaptic potentiation immediately after induction (after 2mins) was in general dependent on the number of simultaneously stimulated spines and their distance. Our analysis revealed that the potentiation initially increased with the number of stimulated spines, as they used common dendritic resources, but the potentiation seemed to decrease as the number of stimulated spines grew further. . We observed that the increase in potentiation was consistent with benefits of common protein resources, e.g. resources from a translational burst, but as the numbers of spine competitors grew further the protein resources allocated to each spine decreased and the plasticity ability was reduced. Our simulations show that size of the dendritic protein pool relative to the resources available in each spine is a major determinant of synaptic sizes and spiking response. Overall, we have built two plasticity models operating at different levels of abstraction. One model taking into account the initial (lognormal) size of a synapse, detailed molecular turn-over and molecular state dynamics for plasticity outcomes and one more coarse-grained model describing the multi-spine plasticity outcome with a small number of dynamical equations (Chater and Eggl et al Comm Biol 2023).
Aim3 linked the changes in protein dynamics and dendritic morphology to functional differences and disease phenotypes. In WP1 we addressed how the protein response is modulated by dendritic morphology such as segment length, branching ratios etc. We found that if a molecule is actively transported, it can accumulate at the dendritic tips and exhibit a bimodal distribution, with one peak located close to the soma and a second peak at the end of the dendrite. In addition, we noticed that dendritic segment length itself is a critical modulator of spatial distributions of proteins and mRNAs and has the ability to alter the shape of the protein distribution. Combining our results obtained in Aim1 with the goals in Aim3 we noticed that energy optimal distributions of mRNAs can be different depending on the length of the considered dendrites, such that somatic mRNA localization may be energetically preferred in small dendrites while the same molecule can prefer dendritic mRNA localization in longer dendrites. We verified this prediction by calculating the predicted total mRNA fraction preferring soma to dendrites and compared it to the reported dendritic lengths. In WP2 are building a model considering a state-dependent translation rate, where translation increased as a function of repeat length. We determined model parameters that can reproduce the mRNA amounts in mutant relative to wild type. Here, we found that in order to be consistent with reported protein amounts in mutants and wild type the amount of mutant mRNA needed to decrease. By reexamining the data published in Krauss et al 2013, we were able to verify this model prediction. To address the role of active transport and translational changes in another monogenetic disease, we calculated the change in synaptic and dendritic protein numbers associated with increased active transport, up-regulated translation and activity dependent translation in dendrites. To this end, we built a three-state-model of active transport that translates disease-induced modifications in run durations, motor binding and unbinding rates into changes of diffusion and directional drift and used it to predict protein distributions as a function of dendritic distance from soma.
In aim 1, we aimed to analyze the spatial distributions of proteins and mRNAs to build a computational framework explaining them using a set of dynamical equations. To this end, we have developed a new software package (SpyDen) to automate analysis of molecular images and determine the spatial distribution of proteins and mRNAs along dendritic stretches and inside spines to be able to model it. The software now allows tracing dendrites, determine their width and evaluate the molecular signal along a dendritic segment of interest. SpyDen can also determine the spine outline and detect both continuous and discrete molecular signals across different image resolutions, we have also benchmarked the algorithmically obtained results against manual annotation. We have also identified newly published large-scale super-resolution scans that provide single molecule resolution for tens of different neural proteins along dendrites that will help us to determine more precisely the spine-to-dendrite exchange in our models (Helm et al Nat Neurosci 2021, Unterauer et al 2024). We processed part of this data to determine the statistics of spine-dendrite ratios across proteins and mapped these to the mRNA soma-to-dendrite localization scores for the corresponding proteins. In addition, we analyzed the variability of spine sizes in pyramidal neurons over time pre and post stimulation with respect to the molecular variability in the synapses and dendrites. Building on this finding we studied the dendritic vs synaptic pools of plasticity related proteins and found that they are consistent with a lognormal distribution that has been previously observed for spine sizes. We have derived a system of differential equations that are valid for both endogenous distributions and plasticity related dynamics. We have derived the energy budget and optimal localization profiles for mRNAs and proteins as a function of their physical parameters such as half-times, diffusion coefficients and cross-validated five model predictions with experimental data. These results indicate that mRNAs are localized in the dendrites rather than remaining in the soma if the corresponding protein has a short half-life, a long amino acid chain or if the mRNA is long-lived.
In aim 2, we found that synaptic potentiation immediately after induction (after 2mins) was in general dependent on the number of simultaneously stimulated spines and their distance. Our analysis revealed that the potentiation initially increased with the number of stimulated spines, as they used common dendritic resources, but the potentiation seemed to decrease as the number of stimulated spines grew further. . We observed that the increase in potentiation was consistent with benefits of common protein resources, e.g. resources from a translational burst, but as the numbers of spine competitors grew further the protein resources allocated to each spine decreased and the plasticity ability was reduced. Our simulations show that size of the dendritic protein pool relative to the resources available in each spine is a major determinant of synaptic sizes and spiking response. Overall, we have built two plasticity models operating at different levels of abstraction. One model taking into account the initial (lognormal) size of a synapse, detailed molecular turn-over and molecular state dynamics for plasticity outcomes and one more coarse-grained model describing the multi-spine plasticity outcome with a small number of dynamical equations (Chater and Eggl et al Comm Biol 2023).
Aim3 linked the changes in protein dynamics and dendritic morphology to functional differences and disease phenotypes. In WP1 we addressed how the protein response is modulated by dendritic morphology such as segment length, branching ratios etc. We found that if a molecule is actively transported, it can accumulate at the dendritic tips and exhibit a bimodal distribution, with one peak located close to the soma and a second peak at the end of the dendrite. In addition, we noticed that dendritic segment length itself is a critical modulator of spatial distributions of proteins and mRNAs and has the ability to alter the shape of the protein distribution. Combining our results obtained in Aim1 with the goals in Aim3 we noticed that energy optimal distributions of mRNAs can be different depending on the length of the considered dendrites, such that somatic mRNA localization may be energetically preferred in small dendrites while the same molecule can prefer dendritic mRNA localization in longer dendrites. We verified this prediction by calculating the predicted total mRNA fraction preferring soma to dendrites and compared it to the reported dendritic lengths. In WP2 are building a model considering a state-dependent translation rate, where translation increased as a function of repeat length. We determined model parameters that can reproduce the mRNA amounts in mutant relative to wild type. Here, we found that in order to be consistent with reported protein amounts in mutants and wild type the amount of mutant mRNA needed to decrease. By reexamining the data published in Krauss et al 2013, we were able to verify this model prediction. To address the role of active transport and translational changes in another monogenetic disease, we calculated the change in synaptic and dendritic protein numbers associated with increased active transport, up-regulated translation and activity dependent translation in dendrites. To this end, we built a three-state-model of active transport that translates disease-induced modifications in run durations, motor binding and unbinding rates into changes of diffusion and directional drift and used it to predict protein distributions as a function of dendritic distance from soma.
As we described above, we have accomplished several goals of the project. Toward the end of the project, we expect to reveal the computational rules underlying the molecular dynamics inside dendrites and show how they give rise to synaptic plasticity and network activity outcomes. The project will impact how molecular imaging data is analyzed and how molecular dynamics in neurons are simulated. It will also help view synaptic plasticity less as a single synapse phenomenon and more like a spatial footprint affecting the whole dendrite.
Similar to digital twins of complex physical objects in other disciplines (e.g. climate, astronomy, architecture), computational models of molecular trafficking can help build and test intuition. With tens of thousands of different molecules, each localizes in slightly different ways and presents varying copy numbers. Finding common computational principles and studying how they affect neural computation is a challenge. My group is also developing several novel data analysis approaches and simulations for molecular dynamics and is identifying common principles across seemingly unrelated experimental observations.
First, we integrated different image analysis targets into a single software framework, considering single molecule puncta data as well as continuous signals across spines and dendrites to be able to compute their statistical distributions. Second, we established a computational framework based on minimal energy principles to explain similarities and differences across mRNA and protein localization in dendrites. Third, we developed multi-scale models that allow simulating synaptic plasticity events across different sites and help anticipate the expected plasticity footprint.
We are translating our computational insights into surprising consequences for the interpretation of brain-wide pulse-chase molecular labeling experiments and super-resolution spatial proteomics screens. To this end, we have started collaborations with the laboratories of Prof. Rizzoli at the University of Goettingen and Prof. Svoboda at the Allen Institute and have successfully transferred our computational insights into specific predictions at the interface of molecular dynamics, morphology, and neural activity.
Similar to digital twins of complex physical objects in other disciplines (e.g. climate, astronomy, architecture), computational models of molecular trafficking can help build and test intuition. With tens of thousands of different molecules, each localizes in slightly different ways and presents varying copy numbers. Finding common computational principles and studying how they affect neural computation is a challenge. My group is also developing several novel data analysis approaches and simulations for molecular dynamics and is identifying common principles across seemingly unrelated experimental observations.
First, we integrated different image analysis targets into a single software framework, considering single molecule puncta data as well as continuous signals across spines and dendrites to be able to compute their statistical distributions. Second, we established a computational framework based on minimal energy principles to explain similarities and differences across mRNA and protein localization in dendrites. Third, we developed multi-scale models that allow simulating synaptic plasticity events across different sites and help anticipate the expected plasticity footprint.
We are translating our computational insights into surprising consequences for the interpretation of brain-wide pulse-chase molecular labeling experiments and super-resolution spatial proteomics screens. To this end, we have started collaborations with the laboratories of Prof. Rizzoli at the University of Goettingen and Prof. Svoboda at the Allen Institute and have successfully transferred our computational insights into specific predictions at the interface of molecular dynamics, morphology, and neural activity.