Designing Regularizers and Architectures for Recurrent Neural Networks

Cette thèse contribue a la recherche vers l'intelligence artificielle en utilisant des méthodes connexionnistes. Les réseaux de neurones récurrents sont un ensemble de modèles séquentiels de plus en plus populaires capable en principe d'apprendre des algorithmes arbitraires. Ces modèles effectuent un apprentissage en profondeur, un type d'apprentissage machine. Sa généralité et son succès empirique en font un sujet intéressant pour la recherche et un outil prometteur pour la création de l'intelligence artificielle plus générale. Le premier chapitre de cette thèse donne un bref aperçu des sujets de fonds: l'intelligence artificielle, l'apprentissage machine, l'apprentissage en profondeur et les réseaux de neurones récurrents. Les trois chapitres suivants couvrent ces sujets de manière de plus en plus spécifiques. Enfin, nous présentons quelques contributions apportées aux réseaux de neurones récurrents. Le chapitre \ref{arxiv1} présente nos travaux de régularisation des réseaux de neurones récurrents. La régularisation vise à améliorer la capacité de généralisation du modèle, et joue un role clé dans la performance de plusieurs applications des réseaux de neurones récurrents, en particulier en reconnaissance vocale. Notre approche donne l'état de l'art sur TIMIT, un benchmark standard pour cette tâche. Le chapitre \ref{cpgp} présente une seconde ligne de travail, toujours en cours, qui explore une nouvelle architecture pour les réseaux de neurones récurrents. Les réseaux de neurones récurrents maintiennent un état caché qui représente leurs observations antérieures. L'idée de ce travail est de coder certaines dynamiques abstraites dans l'état caché, donnant au réseau une manière naturelle d'encoder des tendances cohérentes de l'état de son environnement. Notre travail est fondé sur un modèle existant; nous décrivons ce travail et nos contributions avec notamment une expérience préliminaire.

Plasticité du cortex visuel: «homéodynamie» des connexions neuronales et modèle d’effets d’antidépresseurs

Les informations sensorielles sont traitées dans le cortex par des réseaux de neurones co-activés qui forment des assemblées neuronales fonctionnelles. Le traitement visuel dans le cortex est régit par différents aspects des caractéristiques neuronales tels que l’aspect anatomique, électrophysiologique et moléculaire. Au sein du cortex visuel primaire, les neurones sont sélectifs à divers attributs des stimuli tels que l’orientation, la direction, le mouvement et la fréquence spatiale. Chacun de ces attributs conduit à une activité de décharge maximale pour une population neuronale spécifique. Les neurones du cortex visuel ont cependant la capacité de changer leur sélectivité en réponse à une exposition prolongée d’un stimulus approprié appelée apprentissage visuel ou adaptation visuelle à un stimulus non préférentiel. De ce fait, l’objectif principal de cette thèse est d’investiguer les mécanismes neuronaux qui régissent le traitement visuel durant une plasticité induite par adaptation chez des animaux adultes. Ces mécanismes sont traités sous différents aspects : la connectivité neuronale, la sélectivité neuronale, les propriétés électrophysiologiques des neurones et les effets des drogues (sérotonine et fluoxétine). Le modèle testé se base sur les colonnes d’orientation du cortex visuel primaire. La présente thèse est subdivisée en quatre principaux chapitres. Le premier chapitre (A) traite de la réorganisation du cortex visuel primaire suite à une plasticité induite par adaptation visuelle. Le second chapitre (B) examine la connectivité neuronale fonctionnelle en se basant sur des corrélations croisées entre paires neuronales ainsi que sur des corrélations d’activités de populations neuronales. Le troisième chapitre (C) met en liaison les aspects cités précédemment (les effets de l’adaptation visuelle et la connectivité fonctionnelle) aux propriétés électrophysiologiques des neurones (deux classes de neurones sont traitées : les neurones à décharge régulière et les neurones à décharge rapide ou burst). Enfin, le dernier chapitre (D) a pour objectif l’étude de l’effet du couplage de l’adaptation visuelle à l’administration de certaines drogues, notamment la sérotonine et la fluoxétine (inhibiteur sélectif de recapture de la sérotonine). Méthodes En utilisant des enregistrements extracellulaires d’activités neuronales dans le cortex visuel primaire (V1) combinés à un processus d’imagerie cérébrale optique intrinsèque, nous enregistrons l’activité de décharge de populations neuronales et nous examinons l’activité de neurones individuels extraite des signaux multi-unitaires. L’analyse de l’activité cérébrale se base sur différents algorithmes : la distinction des propriétés électrophysiologiques des neurones se fait par calcul de l’intervalle de temps entre la vallée et le pic maximal du potentiel d’action (largeur du potentiel d’action), la sélectivité des neurones est basée sur leur taux de décharge à différents stimuli, et la connectivité fonctionnelle utilise des calculs de corrélations croisées. L’utilisation des drogues se fait par administration locale sur la surface du cortex (après une craniotomie et une durotomie). Résultats et conclusions Dans le premier chapitre, nous démontrons la capacité des neurones à modifier leur sélectivité après une période d’adaptation visuelle à un stimulus particulier, ces changements aboutissent à une réorganisation des cartes corticales suivant un patron spécifique. Nous attribuons ce résultat à la flexibilité de groupes fonctionnels de neurones qui étaient longtemps considérés comme des unités anatomiques rigides. En effet, nous observons une restructuration extensive des domaines d’orientation dans le but de remodeler les colonnes d’orientation où chaque stimulus est représenté de façon égale. Ceci est d’autant plus confirmé dans le second chapitre où dans ce cas, les cartes de connectivité fonctionnelle sont investiguées. En accord avec les résultats énumérés précédemment, les cartes de connectivité montrent également une restructuration massive mais de façon intéressante, les neurones utilisent une stratégie de sommation afin de stabiliser leurs poids de connectivité totaux. Ces dynamiques de connectivité sont examinées dans le troisième chapitre en relation avec les propriétés électrophysiologiques des neurones. En effet, deux modes de décharge neuronale permettent la distinction entre deux classes neuronales. Leurs dynamiques de corrélations distinctes suggèrent que ces deux classes jouent des rôles clés différents dans l’encodage et l’intégration des stimuli visuels au sein d’une population neuronale. Enfin, dans le dernier chapitre, l’adaptation visuelle est combinée avec l’administration de certaines substances, notamment la sérotonine (neurotransmetteur) et la fluoxétine (inhibiteur sélectif de recapture de la sérotonine). Ces deux substances produisent un effet similaire en facilitant l’acquisition des stimuli imposés par adaptation. Lorsqu’un stimulus non optimal est présenté en présence de l’une des deux substances, nous observons une augmentation du taux de décharge des neurones en présentant ce stimulus. Nous présentons un modèle neuronal basé sur cette recherche afin d’expliquer les fluctuations du taux de décharge neuronale en présence ou en absence des drogues. Cette thèse présente de nouvelles perspectives quant à la compréhension de l’adaptation des neurones du cortex visuel primaire adulte dans le but de changer leur sélectivité dans un environnement d’apprentissage. Nous montrons qu’il y a un parfait équilibre entre leurs habiletés plastiques et leur dynamique d’homéostasie.

Learning representations for Information Retrieval

La recherche d'informations s'intéresse, entre autres, à répondre à des questions comme: est-ce qu'un document est pertinent à une requête ? Est-ce que deux requêtes ou deux documents sont similaires ? Comment la similarité entre deux requêtes ou documents peut être utilisée pour améliorer l'estimation de la pertinence ? Pour donner réponse à ces questions, il est nécessaire d'associer chaque document et requête à des représentations interprétables par ordinateur. Une fois ces représentations estimées, la similarité peut correspondre, par exemple, à une distance ou une divergence qui opère dans l'espace de représentation. On admet généralement que la qualité d'une représentation a un impact direct sur l'erreur d'estimation par rapport à la vraie pertinence, jugée par un humain. Estimer de bonnes représentations des documents et des requêtes a longtemps été un problème central de la recherche d'informations. Le but de cette thèse est de proposer des nouvelles méthodes pour estimer les représentations des documents et des requêtes, la relation de pertinence entre eux et ainsi modestement avancer l'état de l'art du domaine. Nous présentons quatre articles publiés dans des conférences internationales et un article publié dans un forum d'évaluation. Les deux premiers articles concernent des méthodes qui créent l'espace de représentation selon une connaissance à priori sur les caractéristiques qui sont importantes pour la tâche à accomplir. Ceux-ci nous amènent à présenter un nouveau modèle de recherche d'informations qui diffère des modèles existants sur le plan théorique et de l'efficacité expérimentale. Les deux derniers articles marquent un changement fondamental dans l'approche de construction des représentations. Ils bénéficient notamment de l'intérêt de recherche dont les techniques d'apprentissage profond par réseaux de neurones, ou deep learning, ont fait récemment l'objet. Ces modèles d'apprentissage élicitent automatiquement les caractéristiques importantes pour la tâche demandée à partir d'une quantité importante de données. Nous nous intéressons à la modélisation des relations sémantiques entre documents et requêtes ainsi qu'entre deux ou plusieurs requêtes. Ces derniers articles marquent les premières applications de l'apprentissage de représentations par réseaux de neurones à la recherche d'informations. Les modèles proposés ont aussi produit une performance améliorée sur des collections de test standard. Nos travaux nous mènent à la conclusion générale suivante: la performance en recherche d'informations pourrait drastiquement être améliorée en se basant sur les approches d'apprentissage de représentations.

Aspergillus clavatus tremorgenic neurotoxicosis in cattle fed sprouted grains

Beef and dairy cattle from four different herds in southern and central Queensland fed hydroponically-produced sprouted barley or wheat grain heavily infested with Aspergillus clavatus developed posterior ataxia with knuckling of fetlocks, muscular tremors and recumbency, but maintained appetite. A few animals variously had reduced milk production, hyperaesthesia, drooling of saliva, hypermetria of hind limbs or muscle spasms. Degeneration of large neurones was seen in the brain stem and spinal cord grey matter. The syndrome was consistent with A clavatus tremorgenic mycotoxicosis of ruminants. The cases are the earliest known to be associated with this fungus in Australia. They highlight a potential hazard of hydroponic fodder production systems, which appear to favour A clavatus growth on sprouted grain, exacerbated in some cases by equipment malfunctions that increase operating temperatures.

Visual thalamocortical projections in the flying fox: Parallel pathways to striate and extrastriate areas

We studied thalamic projections to the visual cortex in flying foxes, animals that share neural features believed to resemble those present in the brains of early primates. Neurones labeled by injections of fluorescent tracers in striate and extrastriate cortices were charted relative to the architectural boundaries of thalamic nuclei. Three main findings are reported: First, there are parallel lateral geniculate nucleus (LGN) projections to striate and extrastriate cortices. Second, the pulvinar complex is expansive, and contains multiple subdivisions. Third, across the visual thalamus, the location of cells labeled after visual cortex injections changes systematically, with caudal visual areas receiving their strongest projections from the most lateral thalamic nuclei, and rostral areas receiving strong projections from medial nuclei. We identified three architectural layers in the LGN, and three subdivisions of the pulvinar complex. The outer LGN layer contained the largest cells, and had strong projections to the areas V1, V2 and V3. Neurones in the intermediate LGN layer were intermediate in size, and projected to V1 and, less densely, to V2. The layer nearest to the origin of the optic radiation contained the smallest cells, and projected not only to V1, V2 and V3, but also, weakly, to the occipitotemporal area (OT, which is similar to primate middle temporal area) and the occipitoparietal area (OP, a third tier area located near the dorsal midline). V1, V2 and V3 received strong projections from the lateral and intermediate subdivisions of the pulvinar complex, while OP and OT received their main thalamic input from the intermediate and medial subdivisions of the pulvinar complex. These results suggest parallels with the carnivore visual system, and indicate that the restriction of the projections of the large- and intermediatesized LGN layers to V1, observed in present-day primates, evolved from a more generalized mammalian condition. (C) 2004 IBRO. Published by Elsevier Ltd. All rights reserved.

Ecology and evolution of primate colour vision

More than one hundred years ago, Grant Allen suggested that colour vision in primates, birds and insects evolved as an adaptation for foraging on colourful advertisements of plants-fruits and flowers. Recent studies have shown that well developed colour vision appeared long before fruits and flowers evolved. Thus, colour vision is generally beneficial for many animals, not only for those eating colourful food. Primates are the only placental mammals that have trichromatic colour vision. This may indicate either that trichromacy is particularly useful for primates or that primates are unique among placental mammals in their ability to utilise the signals of three spectrally distinct types of cones or both. Because fruits are an important component of the primate diet, primate trichromacy could have evolved as a specific adaptation for foraging on fruits. Alternatively, primate trichromacy could have evolved as an adaptation for many visual tasks. Comparative studies of mammalian eyes indicate that primates are the only placental mammals that have in their retina a pre-existing neural machinery capable of utilising the signals of an additional spectral type of cone. Thus, the failure of non-primate placental mammals to evolve trichromacy can be explained by constraints imposed on the wiring of retinal neurones.

Specialization in pyramidal cell structure in the sensory-motor cortex of the vervet monkey (Cercopethicus pygerythrus)

Recent studies have revealed systematic differences in the pyramidal cell structure between functionally related cortical areas of primates. Trends for a parallel in pyramidal cell structure and functional complexity have been reported in visual, somatosensory, motor, cingulate and prefrontal cortex in the macaque monkey cortex. These specializations in structure have been interpreted as being fundamental in determining cellular and systems function, endowing circuits in these different cortical areas with different computational power. In the present study we extend our initial finding of systematic specialization of pyramidal cell structure in sensory-motor cortex in the macaque monkey [Cereb Cortex 12 (2002) 1071] to the vervet monkey. More specifically, we investigated pyramidal cell structure in somatosensory and motor areas 1/2, 5, 7, 4 and 6. Neurones in fixed, flat-mounted, cortical slices were injected intracellularly with Lucifer Yellow and processed for a light-stable 3,3'-diaminobenzidine reaction product. The size of, number of branches in, and spine density of the basal dendritic arbors varied systematically such that there was a trend for increasing complexity in arbor structure with progression through 1/2, 5 and 7. In addition, cells in area 6 were larger, more branched, and more spinous than those in area 4. (c) 2005 IBRO. Published by Elsevier Ltd. All rights reserved.

Nervous and muscle system development in Phascolion strombus (Sipuncula)

Recent interpretations of developmental gene expression patterns propose that the last common metazoan ancestor was segmented, although most animal phyla show no obvious signs of segmentation. Developmental studies of non-model system trochozoan taxa may shed light on this hypothesis by assessing possible cryptic segmentation patterns. In this paper, we present the first immunocytochemical data on the ontogeny of the nervous system and the musculature in the sipunculan Phascolion strombus. Myogenesis of the first anlagen of the body wall ring muscles occurs synchronously and not subsequently from anterior to posterior as in segmented spiralian taxa (i.e. annelids). The number of ring muscles remains constant during the initial stages of body axis elongation. In the anterior-posteriorly elongated larva, newly formed ring muscles originate along the entire body axis between existing myocytes, indicating that repeated muscle bands do not form from a posterior growth zone. During neurogenesis, the Phascolion larva expresses a non-metameric, paired, ventral nerve cord that fuses in the mid-body region in the late-stage elongated larva. Contrary to other trochozoans, Phascolion lacks any larval serotonergic structures. However, two to three FMRFamide-positive cells are found in the apical organ. In addition, late larvae show commissure-like neurones interconnecting the two ventral nerve cords, while early juveniles exhibit a third, medially placed FMRFamidergic ventral nerve. Although we did not find any indications for cryptic segmentation, certain neuro-developmental traits in Phascolion resemble the conditions found in polychaetes (including echiurans) and myzostomids and support a close relationship of Sipuncula and Annelida.

Methods of studying the planar distribution of objects in histological sections of brain tissue

This article reviews the statistical methods that have been used to study the planar distribution, and especially clustering, of objects in histological sections of brain tissue. The objective of these studies is usually quantitative description, comparison between patients or correlation between histological features. Objects of interest such as neurones, glial cells, blood vessels or pathological features such as protein deposits appear as sectional profiles in a two-dimensional section. These objects may not be randomly distributed within the section but exhibit a spatial pattern, a departure from randomness either towards regularity or clustering. The methods described include simple tests of whether the planar distribution of a histological feature departs significantly from randomness using randomized points, lines or sample fields and more complex methods that employ grids or transects of contiguous fields, and which can detect the intensity of aggregation and the sizes, distribution and spacing of clusters. The usefulness of these methods in understanding the pathogenesis of neurodegenerative diseases such as Alzheimer's disease and Creutzfeldt-Jakob disease is discussed. © 2006 The Royal Microscopical Society.

Laminar distribution of the pathological changes in the cerebral cortex in variant Creutzfeldt-Jakob disease (vCJD)

To determine the pattern of cortical degeneration in cases of variant Creutzfeldt-Jakob disease (vCJD), the laminar distribution of the vacuolation ("spongiform change"), surviving neurones, glial cell nuclei, and prion protein (PrP) deposits was studied in the frontal, parietal and temporal lobes. The vacuolation exhibited two common patterns of distribution: either the vacuoles were present throughout the cortex or a bimodal distribution was present with peaks of density in the upper and lower cortical laminae. The distribution of the surviving neurones was highly variable in different regions; the commonest pattern being a uniform distribution with cortical depth. Glial cell nuclei were distributed largely in the lower cortical laminae. The non-florid PrP deposits exhibited either a bimodal distribution or exhibited a peak of density in the upper cortex while the florid deposits were either uniformly distributed down the cortex or were present in the upper cortical laminae. In a significant proportion of areas, the density of the vacuoles was positively correlated with either the surviving neurones or with the glial cell nuclei. These results suggest similarities and differences in the laminar distributions of the pathogenic changes in vCJD compared with cases of sporadic CJD (sCJD). The laminar distribution of vacuoles was more extensive in vCJD than in sCJD whereas the distribution of the glial cell nuclei was similar in the two disorders. In addition, PrP deposits in sCJD were localised mainly in the lower cortical laminae while in vCJD, PrP deposits were either present in all laminae or restricted to the upper cortical laminae. These patterns of laminar distribution suggest that the process of cortical degeneration may be distinctly different in vCJD compared with sCJD.

Central neural mechanisms mediating human visceral hypersensitivity

Although visceral hypersensitivity is thought to be important in generating symptoms in functional gastrointestinal disorders, the neural mechanisms involved are poorly understood. We recently showed that central sensitization (hyperexcitability of spinal cord sensory neurones) may play an important role. In this study, we demonstrate that after a 30-min infusion of 0.15 M HCl acid into the healthy human distal esophagus, we see a reduction in the pain threshold to electrical stimulation of the non-acid-exposed proximal esophagus (9.6 ± 2.4 mA) and a concurrent reduction in the latency of the N1 and P2 components of the esophageal evoked potentials (EEP) from this region (10.4 ± 2.3 and 15.8 ± 5.3 ms, respectively). This reduced EEP latency indicates a central increase in afferent pathway velocity and therefore suggests that hyperexcitability within the central visceral pain pathway contributes to the hypersensitivity within the proximal, non-acid-exposed esophagus (secondary hyperalgesia/allodynia). These findings provide the first electrophysiological evidence that central sensitization contributes to human visceral hypersensitivity.

Functional interconnectivity between the globus pallidus and the subthalamic nucleus in the mouse brain slice

In accordance with its central role in basal ganglia circuitry, changes in the rate of action potential firing and pattern of activity in the globus pallidus (GP)-subthalamic nucleus (STN) network are apparent in movement disorders. In this study we have developed a mouse brain slice preparation that maintains the functional connectivity between the GP and STN in order to assess its role in shaping and modulating bursting activity promoted by pharmacological manipulations. Fibre-tract tracing studies indicated that a parasagittal slice cut 20 deg to the midline best preserved connectivity between the GP and the STN. IPSCs and EPSCs elicited by electrical stimulation confirmed connectivity from GP to STN in 44/59 slices and from STN to GP in 22/33 slices, respectively. In control slices, 74/76 (97%) of STN cells fired tonically at a rate of 10.3 ± 1.3 Hz. This rate and pattern of single spiking activity was unaffected by bath application of the GABAA antagonist picrotoxin (50 μM, n = 9) or the glutamate receptor antagonist (6-cyano-7-nitroquinoxaline-2, 3-dione (CNQX) 10 μM, n = 8). Bursting activity in STN neurones could be induced pharmacologically by application of NMDA alone (20 μM, 3/18 cells, 17%) but was more robust if NMDA was applied in conjunction with apamin (20-100 nM, 34/77 cells, 44%). Once again, neither picrotoxin (50 μM, n = 5) nor CNQX (10 μM, n = 5) had any effect on the frequency or pattern of the STN neurone activity while paired STN and GP recordings of tonic and bursting activity show no evidence of coherent activity. Thus, in a mouse brain slice preparation where functional GP-STN connectivity is preserved, no regenerative synaptically mediated activity indicative of a dynamic network is evident, either in the resting state or when neuronal bursting in both the GP and STN is generated by application of NMDA/apamin. This difference from the brain in Parkinson's disease may be attributed either to insufficient preservation of cortico-striato-pallidal or cortico-subthalamic circuitry, and/or an essential requirement for adaptive changes resulting from dopamine depletion for the expression of network activity within this tissue complex. © The Physiological Society 2005.

Simultaneous estimation of global background synaptic inhibition and excitation from membrane potential fluctuations in layer III neurons of the rat entorhinal cortex in vitro

It is becoming clear that the detection and integration of synaptic input and its conversion into an output signal in cortical neurons are strongly influenced by background synaptic activity or "noise." The majority of this noise results from the spontaneous release of synaptic transmitters, interacting with ligand-gated ion channels in the postsynaptic neuron [Berretta N, Jones RSG (1996); A comparison of spontaneous synaptic EPSCs in layer V and layer II neurones in the rat entorhinal cortex in vitro. J Neurophysiol 76:1089-1110; Jones RSG, Woodhall GL (2005) Background synaptic activity in rat entorhinal cortical neurons: differential control of transmitter release by presynaptic receptors. J Physiol 562:107-120; LoTurco JJ, Mody I, Kriegstein AR (1990) Differential activation of glutamate receptors by spontaneously released transmitter in slices of neocortex. Neurosci Lett 114:265-271; Otis TS, Staley KJ, Mody I (1991) Perpetual inhibitory activity in mammalian brain slices generated by spontaneous GABA release. Brain Res 545:142-150; Ropert N, Miles R, Korn H (1990) Characteristics of miniature inhibitory postsynaptic currents in CA1 pyramidal neurones of rat hippocampus. J Physiol 428:707-722; Salin PA, Prince DA (1996) Spontaneous GABAA receptor-mediated inhibitory currents in adult rat somatosensory cortex. J Neurophysiol 75:1573-1588; Staley KJ (1999) Quantal GABA release: noise or not? Nat Neurosci 2:494-495; Woodhall GL, Bailey SJ, Thompson SE, Evans DIP, Stacey AE, Jones RSG (2005) Fundamental differences in spontaneous synaptic inhibition between deep and superficial layers of the rat entorhinal cortex. Hippocampus 15:232-245]. The function of synaptic noise has been the subject of debate for some years, but there is increasing evidence that it modifies or controls neuronal excitability and, thus, the integrative properties of cortical neurons. In the present study we have investigated a novel approach [Rudolph M, Piwkowska Z, Badoual M, Bal T, Destexhe A (2004) A method to estimate synaptic conductances from membrane potential fluctuations. J Neurophysiol 91:2884-2896] to simultaneously quantify synaptic inhibitory and excitatory synaptic noise, together with postsynaptic excitability, in rat entorhinal cortical neurons in vitro. The results suggest that this is a viable and useful approach to the study of the function of synaptic noise in cortical networks. © 2007 IBRO.

Effect of dopamine on synchronous neuronal oscillations in the globus pallidus-subthalamic nucleus network

Changes in the pattern of activity of neurones within the basal ganglia are relevant in the pathophysiology and symptoms of Parkinson’s disease. The globus pallidus (GP) – subthalamic nucleus (STN) network has been proposed to form a pacemaker driving regenerative synchronous bursting activity. In order to test whether this activity can be sustained in vitro a 20o parasagittal slice of mouse midbrain was developed which preserved functional connectivity between the STN and GP. Mouse STN and GP cells were characterised electrophysiologically by the presence or absence of a voltage sag in response to hyperpolarising current steps indicative of Ih and the presence of rebound depolarisations. The presence of evoked and spontaneous post-synaptic GABA and glutamatergic currents indicated functional connectivity between the STN and GP. In control slices, STN cells fired action potentials at a regular rate, activity which was unaffected by bath application of the GABAA receptor antagonist picrotoxin (50 μM) or the glutamate receptor antagonist CNQX (10 μM). Paired extracellular recordings of STN cells showed uncorrelated firing. Oscillatory burst activity was induced pharmacologically using the glutamate receptor agonist, NMDA (20 μM), in combination with the potassium channel blocker apamin (50 -100 nM). The burst activity was unaffected by bath application of picrotoxin or CNQX while paired STN recordings showed uncorrelated activity indicating that the activity is not produced by the neuronal network. Thus, no regenerative activity is evident in this mouse brain preparation, either in control slices or when bursting is pharmacologically induced, suggesting the requirement of other afferent inputs that are not present in the slice. Using single-unit extracellular recording, dopamine (30 μM) produced an excitation of STN cells. This excitation was independent of synaptic transmission and was mimicked by both the Dl-like receptor agonist SKF38393 (10 μM) and the D2-like receptor agonist quinpirole (10 μM). However, the excitation was partially reduced by the D1-like antagonist SCH23390 (2 μM) but not by the D2-like antagonists sulpiride (10 μM) and eticlopride (10 μM). Using whole-recordings, dopamine was shown to induce membrane depolarisation. This depolarisation was caused either by a D1-like receptor mediated increase in a conductance which reversed at -34 mV, consistent with a non-specific cation conductance, or a D2-like receptor mediated decrease in conductance which reversed around -100 mV, consistent with a potassium conductance. Bath application of dopamine altered the pattern of the burst-firing produced by NMDA an apamin towards a more regular pattern. This effect was associated with a decrease in amplitude and ll1crease in frequency of TTX-resistant plateau potentials which underlie the burst activity.

On the mechanisms of cerebellar synaptic plasticity

Changes in the strength of signalling between neurones are thought to provide a cellular substrate for learning and memory. In the cerebellar cortex, raising the frequency and the strength of parallel fibre (PF) stimulation leads to a long-term depression (LTD) of the strength of signalling at the synapse between PFs and Purkinje cells (PCs), which spreads to distant synapses to the same cell via a nitric oxide (NO) dependent mechanism. At the same synapse, but under conditions of reduced post-synaptic calcium activity, raised frequency stimulation (RFS) of PFs triggers a long-term potentiation of synaptic transmission. The aims of the work described in this thesis were to investigate the conditions necessary for LTD and LTP at this synapse following RFS and to identify the origins and second messenger cascades involved in the induction and spread of LTP and LTD. In thin, parasagittal cerebellar slices whole cell patch clamp recordings were made from PCs and the effects of RFS of one of two, independent PF inputs to the same PC were examined under a range of experimental conditions. Under conditions designed to reduce post-synaptic calcium activity, RFS to a single PF input led to LTP and a decreases in paired pulse facilitation (PPF) in both pathways. This heterosynaptic potentiation was prevented by inhibition of protein kinase A (PKA) or by inhibition of NO synthase with either 7-nitroindazole (7-NI) or NG Nitro-L-argenine methyl ester. Inhibition of guanylate cyclase (GC) or protein kinase G (PKG) had no effect. A similar potentiation was observed upon application of the adenylyl cyclase (AC) activator forskolin or the NO donor spermine NONOate. Both of these treatments also resulted in an increase in the frequency of mEPSCs, which provides further evidence for a presynaptic origin of LTP. Forskolin induced potentiation and the increase in mEPSC frequency were blocked by 7-NI. The styryl dye FM1-43, a fluorescent reporter of endo- and exocytosis, was also used to further examine the possible pre-synaptic origins of LTP. RFS or forskolin application enhanced FM1-43 de-staining and NOS inhibitors blocked this effect. Application of NONOate also enhanced FM1-43 de-staining. When post-synaptic calcium activity was less strictly buffered, RFS to a single PF input led to a transient potentiation that was succeeded by LTD in both pathways. This LTD, which resembled previously described forms, was prevented by inhibition of the NO/cGMP/PKG cascade. Modification of the AC/cAMP/PKA cascade had no effect. In summary, the direction of synaptic plasticity at the PF-PC synapse in response to RFS depends largely on the level of post-synaptic calcium activity. LTP and LTD were non-input specific and both forms of plasticity were dependent on NOS activity. Induction of LTP was mediated by a presynaptic mechanism and depended on NO and cAMP production. LTD on the other hand was a post-synaptic process and required activity of the NO/cGMP/PKG signalling cascade.