Gontier, Camille Michel JeanClaude (2022). Statistical approaches for synaptic characterization. (Thesis). Universität Bern, Bern

Text
22gontier_cmjc.pdf  Thesis Available under License Creative Commons: Attribution (CCBY 4.0). Download (10MB)  Preview 
Abstract
Synapses are fascinatingly complex transmission units. One of the fundamental features of synaptic transmission is its stochasticity, as neurotransmitter release exhibits variability and possible failures. It is also quantised: postsynaptic responses to presynaptic stimulations are built up of several and similar quanta of current, each of them arising from the release of one presynaptic vesicle. Moreover, they are dynamic transmission units, as their activity depends on the history of previous spikes and stimulations, a phenomenon known as synaptic plasticity. Finally, synapses exhibit a very broad range of dynamics, features, and connection strengths, depending on neuromodulators concentration [5], the age of the subject [6], their localization in the CNS or in the PNS, or the type of neurons [7]. Addressing the complexity of synaptic transmission is a relevant problem for both biologists and theoretical neuroscientists. From a biological perspective, a finer understanding of transmission mechanisms would allow to study possibly synapserelated diseases, or to determine the locus of plasticity and homeostasis. From a theoretical perspective, different normative explanations for synaptic stochasticity have been proposed, including its possible role in uncertainty encoding, energyefficient computation, or generalization while learning. A precise description of synaptic transmission will be critical for the validation of these theories and for understanding the functional relevance of this probabilistic and dynamical release. A central issue, which is common to all these areas of research, is the problem of synaptic characterization. Synaptic characterization (also called synaptic interrogation [8]) refers to a set of methods for exploring synaptic functions, inferring the value of synaptic parameters, and assessing features such as plasticity and modes of release. This doctoral work sits at the crossroads of experimental and theoretical neuroscience: its main aim is to develop statistical tools and methods to improve synaptic characterization, and hence to bring quantitative solutions to biological questions. In this thesis, we focus on modelbased approaches to quantify synaptic transmission, for which different methods are reviewed in Chapter 3. By fitting a generative model of postsynaptic currents to experimental data, it is possible to infer the value of the synapse’s parameters. By performing model selection, we can compare different modelizations of a synapse and thus quantify its features. The main goal of this thesis is thus to develop theoretical and statistical tools to improve the efficiency of both model fitting and model selection. A first question that often arises when recording synaptic currents is how to precisely observe and measure a quantal transmission. As mentioned above, synaptic transmission has been observed to be quantised. Indeed, the opening of a single presynaptic vesicle (and the release of the neurotransmitters it contains) will create a stereotypical postsynaptic current q, which is called the quantal amplitude. As the number of activated presynaptic vesicles increases, the total postsynaptic current will increase in steplike increments of amplitude q. Hence, at chemical synapses, the postsynaptic responses to presynaptic stimulations are built up of k quanta of current, where k is a random variable corresponding to the number of open vesicles. Excitatory postsynaptic current (EPSC) thus follows a multimodal distribution, where each component has its mean located to a multiple kq with k 2 N and has a width corresponding to the recording noise σ. If σ is large with respect to q, these components will fuse into a unimodal distribution, impeding the possibility to identify quantal transmission and to compute q. How to characterize the regime of parameters in which quantal transmission can be identified? This question led us to define a practical identifiability criterion for statistical model, which is presented in Chapter 4. In doing so, we also derive a meanfield approach for fast likelihood computation (Appendix A) and discuss the possibility to use the Bayesian Information Criterion (a classically used model selection criterion) with correlated observations (Appendix B). A second question that is especially relevant for experimentalists is how to optimally stimulate the presynaptic cell in order to maximize the informativeness of the recordings. The parameters of a chemical synapse (namely, the number of presynaptic vesicles N, their release probability p, the quantal amplitude q, the shortterm depression time constant τD, etc.) cannot be measured directly, but can be estimated from the synapse’s postsynaptic responses to evoked stimuli. However, these estimates critically depend on the stimulation protocol being used. For instance, if interspike intervals are too large, no shortterm plasticity will appear in the recordings; conversely, a too high stimulation frequency will lead to a depletion of the presynaptic vesicles and to a poor informativeness of the postsynaptic currents. How to perform Optimal Experiment Design (OED) for synaptic characterization? We developed an Efficient SamplingBased Bayesian Active Learning (ESBBAL) framework, which is efficient enough to be used in realtime biological experiments (Chapter 5), and propose a link between our proposed definition of practical identifiability and Optimal Experiment Design for model selection (Chapter 6). Finally, a third biological question to which we ought to bring a theoretical answer is how to make sense of the observed organization of synaptic proteins. Microscopy observations have shown that presynaptic release sites and postsynaptic receptors are organized in ringlike patterns, which are disrupted upon genetic mutations. In Chapter 7, we propose a normative approach to this protein organization, and suggest that it might optimize a certain biological cost function (e.g. the mean current or SNR after vesicle release). The different theoretical tools and methods developed in this thesis are general enough to be applicable not only to synaptic characterization, but also to different experimental settings and systems studied in physiology. Overall, we expect to democratize and simplify the use of quantitative and normative approaches in biology, thus reducing the cost of experimentation in physiology, and paving the way to more systematic and automated experimental designs.
Item Type:  Thesis 

Dissertation Type:  Cumulative 
Date of Defense:  19 September 2022 
Subjects:  500 Science > 570 Life sciences; biology 600 Technology > 610 Medicine & health 
Institute / Center:  04 Faculty of Medicine > Preclinic Human Medicine > Institute of Physiology 
Depositing User:  Hammer Igor 
Date Deposited:  11 Oct 2022 12:36 
Last Modified:  11 Oct 2022 12:36 
URI:  https://boristheses.unibe.ch/id/eprint/3857 
Actions (login required)
View Item 