Network functions characterization of random hippocampal neuronal networks using multielectrode array

In nervous systems, behaviors are generated through the activity of ensembles of neurons assembled in networks, where networks become the interface between the cellular and behavioral levels. Uncovering the mechanisms of action of these networks is thus essential in order to fully understand and characterize the nervous systems functions in normal and pathological conditions. A complete understanding of a given network requires information about its individual component cells, their functional properties, their wiring diagram and their dynamic interactions. Few known networks come close to providing complete information on all these aspects. Moreover, even if complete information were available it would still only provide insights into network functioning. This is because the structural and functional properties of networks are not static but plastic and can vary over time. Reductionist approaches have been successful at identifying and characterizing many components that could contribute to network functions. For example, the contribution of specific individual classes of cells to network functions can be inferred by selectively silencing or activating them by combining genetics and pharmacology (Wulff & Wisden, 2005). However, networks are not the linear sum of its individual parts but instead reflect the spatial and temporal interactions of non-linear properties. Individual cell components influence and are influenced by the network output(s). Thus, a system-level understanding is needed as advocated by cyberneticians (Bertalanffy, 1969) and systems biologists (Kitano, 2002), focusing on the system’s interactions and dynamics rather than on the characteristics of the single isolated parts. The subject of my PhD thesis is to study network functions of neuronal networks composed of dissociated rat hippocampal neurons grown on multielectrode arrays (MEA). On the one hand, in vitro cultured neuronal networks with random connections retain universal properties and principles present in artificial and natural neural systems. Indeed, such in vitro networks form extensive synaptic connectivity and display complex spontaneous activity and sensitivity to externally applied patterns of activity and are endowed with basic plasticity mechanisms. The organizing principles operating at the level of neuronal populations are intrinsic to neurons and are therefore manifested in in vitro cultured networks, thus providing a point of view of neuronal network functions and mechanisms, independent of the network’s topology. On the other hand, MEAs offer the possibility of probing the neuronal activity of a large number of neurons non-invasively and for extensive periods of time, which is a prerequisite to investigate network functions at the population level while keeping in focus the contribution of single neurons. In this thesis, network functions have been investigated from four different points of view. First, the generation of functional neuronal networks from embryonic mouse stem cells is presented along with a comparison of these networks with those of cultured hippocampal networks. This represents the first demonstration proving that embryonic stem (ES) cells-derived neurons are capable of generating functional neuronal networks. We show that the spontaneous and evoked activities are remarkably similar to those of cultured hippocampal networks. Moreover, ES-derived networks are endowed with basic information properties and are capable of discriminating stimuli of different intensity at a single trial level. Thus, ES-derived neurons provide a novel strategy to design/engineer functional networks with defined computational properties. Second, as recent work showed that spontaneous activity is a hallmark of cultured networks (Maeda et al., 1995; van Pelt et al., 2004), we provide a detailed analysis of its dynamics and compare it with that of intact leech ganglia. We found that the spontaneous activity of these two very different networks share several features, such as the presence of long-range correlations, and that their dynamics can be accounted for the self-organized criticality (SOC) framework (Bak et al., 1988; Jensen, 1998). The SOC theory originates from statistical mechanics and characterizes the system dynamics of interconnected nonlinear threshold units, in which each unit has one or more input/output(s). SOC dynamics has been found in a variety of neuronal preparations with specific functional architecture in vertebrates. Our results provide the first demonstration of SOC in random networks as well as in invertebrate networks, thus generalizing SOC dynamics, and indicating that such dynamics might be a generic property and ubiquitous phenomenon in a large variety of neuronal networks. Third, the time varying aspect of network functions was investigated. In particular, we studied how synaptic plasticity could be induced pharmacologically and the molecular mechanisms allowing its maintenance in cultured hippocampal networks. To this end, we used a top-down approach (recording of the electrical activity with MEA) in combination with a bottom-up one (identification of specific genes by using DNA microarrays) to characterize the time course of the electrical and cellular events up to 24h. Then we investigated how such plasticity affected network functions. In particular, we looked at changes of (i) the spontaneous firing pattern, (ii) the evoked-activity and (iii) the pattern of gene expression. It was found that a transient exposure (30 min.) to inhibitors of GABAA receptors, such as bicuculline and gabazine, was sufficient to induce network plasticity. On the one hand, inhibitors of GABAA receptors induced a synchronous bursting pattern of activity that persisted up to 24h following the drug washout. An early component of synchronization was blocked by inhibitors of the MAPK/ERK pathway, whereas a late component was blocked by inhibitors of transcription. Moreover, our results indicate that the regulation of hundreds – rather than tens or thousands – of genes takes place following the drug washout and suggest that a down-regulation of K+ and HCN channels likely underlie the early component of synchronization. On the other hand, the evoked activity was potentiated for several hours following the drug washout, demonstrating the possibility of pharmacologically inducing a form of plasticity in MEAs. This could represent an interesting alternative with regards to the classical electrical protocols for the induction of functional plasticity. Overall, the combination of MEA and DNA microarrays revealed to be a powerful approach to investigate the electrical and molecular events underlying neuronal plasticity in cultured networks. Finally, we investigated the computational abilities of cultured hippocampal networks by comparing several coding schemes and focused on the difference of two forms of population coding: distributed and pooled. By using classification criteria from pattern recognition theory, we found striking similarities with in vivo population coding strategies used in the somatosensory cortex of the rat (Petersen et al., 2003) and the monkey (Nicolelis et al., 1998). In order to be able to correctly identify a stimulus’ intensity and spatial location, distributed codes proved to be more efficient than the pooling of neuronal responses. Taken together, these results provide the first compelling evidence that simple computational tasks can rely on distributed population coding schemes in random cultured networks, as proposed some years ago (Potter, 2001). The results reported in the first two paragraphs of the Results section have been published in peer-reviewed journals. The third and fourth paragraphs represent articles in preparation.