Estudio del impacto de la dinámica neuronal en el procesamiento de información en la capa granular del cerebelo

Résumé

The cerebellum is a critical brain area for sensorimotor and also non-motor functions such as cognitive and emotional processes. Cerebellar lesions contribute to pathological syndromes such as autism or schizophrenia. However, the primitives under which the cerebellum, and the whole brain, operate at a functional and dysfunctional level are still unclear. To address the complexity of the "diseased" brain system, it is necessary to extract the relevant underlying molecular mechanisms. The availability of large volumes of biomedical data usually makes it difficult to extract this relevant information and interpret it comprehensively. In this thesis, we have made a preliminary experimentation to analyze genetic correlations between diseases with different clinical symptomatologies and/or clinical prognosis (and still based on similar molecular mechanisms). For this purpose, we have developed a methodology for the identification and functional annotation of the most relevant genes in disease. This methodology integrates current systems biology methods, such as protein-protein interaction (PPI) networks, together with multidimensional data sets from different biological levels. The objectives of this first part of the thesis are: the identification of potential diagnostic biomarkers (corresponding to key nodes in the biological and molecular processes of the interactome); the deductive analysis of multidimensional data as an alternative to other search systems; and the extraction of connections between disorders (comorbidities) that are a priori unrelated and that usually escape these traditional systems. Although the methodology is of general purpose, we have applied it to a set of diseases called channelopathies, where ion channels are altered and which generate a wide phenotypic variability. We conclude that our methodology is flexible, fast and easy to apply. Furthermore, it is able to find more correlations between relevant genes than other two traditional methods. Understanding the cerebellar operation in information processing requires decoding the intrinsic functional dynamics of healthy neurons. The tools provided by computational neuroscience allow developing large-scale computational models for the study of these information processing primitives. The most abundant and smallest neurons not only in the cerebellar input layer, but also in the whole brain, are the cerebellar granule cells (GrCs). These neurons play a crucial role in the creation of somatosensory information representations. Their firing characteristics are related to synchronization, rhythmicity and learning in the cerebellum. One of these features is the frequency of enhanced bursting (i.e., spiking resonance). This complex firing pattern has been proposed to facilitate input signal transmission in the thetafrequency band (4-12Hz). However, the functional role of this feature in the operation of the granular layer (the input layer of the cerebellar cortex) is still unclear. Moreover, inherent complex dynamics such as resonance are usually ignored in most efficient computational models. The main goal of this thesis is the creation of different mathematical models of cerebellar GrCs that meet two requirements: to be efficient enough to allow the simulation of large-scale neuron networks, and to be biologically plausible enough to enable the evaluation of the functional impact of these nonlinear dynamics on the information transmission. Indeed, a high degree of biological realism in efficient models allows research at levels where in vivo or in vitro experimental biology is limited. Methodologically, in this thesis we have chosen the "adaptive exponential integrateand- fire" (AdEx) type of model as the simplified neuron model (it has only two differential equations and few parameters) that meets both realism and low computational cost. This model fits quite well the firing characteristics of real cells, but some of its parameters cannot be directly fitted with measurable experimental values. Therefore, an optimization method is necessary to best fit the parameters to the biological data. We have focused on addressing this challenging optimization problem. First, we have developed a parametric optimization methodology based on genetic algorithms (GA) applied to the case of GrC. We have presented the obtained AdEx neuron models and demonstrated their suitability to reproduce not only the main firing properties of real GrCs (including resonance), but also emergent features not defined in the GA (within the cost function to be optimized). Second, we evaluated four alternative algorithms, which are the most widely used and successful in other fields such as engineering. Finally, in the last part of this thesis we have presented an advanced optimization methodology based on multimodal algorithms. The advantage of this approach is that, after a single optimization process, instead of obtaining an only one candidate numerically outperforming the other candidates, as in the previous cases (a single solution), we obtain a sparse population of different neuron models. That is, a heterogeneous population of neurons of the same type with intrinsic variations in their properties. From this set of promising neuron models, the researcher can choose and filter based on the desired biological plausibility (and neuronal parameter configuration). Thus, we also studied how the target properties of the neuron could be obtained with diverse internal parameter configurations. We explored the parameter space and its impact on the subset of neuronal properties that we aim to reproduce.