Vezatin is essential for dendritic spine morphogenesis and functional synaptic maturation.

Vezatin is an integral membrane protein associated with cell-cell adhesion complex and actin cytoskeleton. It is expressed in the developing and mature mammalian brain, but its neuronal function is unknown. Here, we show that Vezatin localizes in spines in mature mouse hippocampal neurons and codistributes with PSD95, a major scaffolding protein of the excitatory postsynaptic density. Forebrain-specific conditional ablation of Vezatin induced anxiety-like behavior and impaired cued fear-conditioning memory response. Vezatin knock-down in cultured hippocampal neurons and Vezatin conditional knock-out in mice led to a significantly increased proportion of stubby spines and a reduced proportion of mature dendritic spines. PSD95 remained tethered to presynaptic terminals in Vezatin-deficient hippocampal neurons, suggesting that the reduced expression of Vezatin does not compromise the maintenance of synaptic connections. Accordingly, neither the amplitude nor the frequency of miniature EPSCs was affected in Vezatin-deficient hippocampal neurons. However, the AMPA/NMDA ratio of evoked EPSCs was reduced, suggesting impaired functional maturation of excitatory synapses. These results suggest a role of Vezatin in dendritic spine morphogenesis and functional synaptic maturation.

Input-specific learning rules at excitatory synapses onto hippocampal parvalbumin-expressing interneurons.

Hippocampal parvalbumin-expressing interneurons (PV INs) provide fast and reliable GABAergic signalling to principal cells and orchestrate hippocampal ensemble activities. Precise coordination of principal cell activity by PV INs relies in part on the efficacy of excitatory afferents that recruit them in the hippocampal network. Feed-forward (FF) inputs in particular from Schaffer collaterals influence spike timing precision in CA1 principal cells whereas local feedback (FB) inputs may contribute to pacemaker activities. Although PV INs have been shown to undergo activity-dependent long term plasticity, how both inputs are modulated during principal cell firing is unknown. Here we show that FF and FB synapses onto PV INs are endowed with distinct postsynaptic glutamate receptors which set opposing long-term plasticity rules. Inward-rectifying AMPA receptors (AMPARs) expressed at both FF and FB inputs mediate a form of anti-Hebbian long term potentiation (LTP), relying on coincident membrane hyperpolarization and synaptic activation. In contrast, FF inputs are largely devoid of NMDA receptors (NMDARs) which are more abundant at FB afferents and confer on them an additional form of LTP with Hebbian properties. Both forms of LTP are expressed with no apparent change in presynaptic function. The specific endowment of FF and FB inputs with distinct coincidence detectors allow them to be differentially tuned upon high frequency afferent activity. Thus, high frequency (>20 Hz) stimulation specifically potentiates FB, but not FF afferents. We propose that these differential, input-specific learning rules may allow PV INs to adapt to changes in hippocampal activity while preserving their precisely timed, clockwork operation.

Serotoninergic fine-tuning of the excitation-inhibition balance in rat visual cortical networks.

Fundamental brain functions depend on a balance between excitation (E) and inhibition (I) that is highly adjusted to a 20-80% set point in layer 5 pyramidal neurons (L5PNs) of rat visual cortex. Dysregulations of both the E-I balance and the serotonergic system in neocortical networks lead to serious neuronal diseases including depression, schizophrenia, and epilepsy. However, no link between the activation of neuronal 5-hydroxytryptamine receptors (5-HTRs) and the cortical E-I balance has yet been reported. Here we used a combination of patch-clamp recordings of composite stimulus-locked responses in L5PN following local electrical stimulations in either layer 2/3 or 6, simultaneous measurement of excitatory and inhibitory conductance dynamics, together with selective pharmacological targeting and single-cell reverse transcriptase-polymerase chain reaction. We show that cortical serotonin shifts the E-I balance in favor of more E and we reveal fine and differential modulations of the E-I balance between 5-HTR subtypes, in relation to whether layer 2/3 or 6 was stimulated and in concordance with the specific expression pattern of these subtypes in pyramidal cells and deep interneurons. This first evidence for the functional segregation of 5-HTR subtypes sheds new light on their coherent functioning in polysynaptic sensory circuits.

D-Serine and Glycine Differentially Control Neurotransmission during Visual Cortex Critical Period.

N-methyl-D-aspartate receptors (NMDARs) play a central role in synaptic plasticity. Their activation requires the binding of both glutamate and d-serine or glycine as co-agonist. The prevalence of either co-agonist on NMDA-receptor function differs between brain regions and remains undetermined in the visual cortex (VC) at the critical period of postnatal development. Here, we therefore investigated the regulatory role that d-serine and/or glycine may exert on NMDARs function and on synaptic plasticity in the rat VC layer 5 pyramidal neurons of young rats. Using selective enzymatic depletion of d-serine or glycine, we demonstrate that d-serine and not glycine is the endogenous co-agonist of synaptic NMDARs required for the induction and expression of Long Term Potentiation (LTP) at both excitatory and inhibitory synapses. Glycine on the other hand is not involved in synaptic efficacy per se but regulates excitatory and inhibitory neurotransmission by activating strychnine-sensitive glycine receptors, then producing a shunting inhibition that controls neuronal gain and results in a depression of synaptic inputs at the somatic level after dendritic integration. In conclusion, we describe for the first time that in the VC both D-serine and glycine differentially regulate somatic depolarization through the activation of distinct synaptic and extrasynaptic receptors.

Benzodiazepine ligands rapidly influence GABAA receptor diffusion and clustering at hippocampal inhibitory synapses.

Benzodiazepines (BZDs) are widely used in the treatment of a variety of neurological and psychiatric conditions including anxiety, insomnia and epilepsy. BZDs are thought to act predominantly by affecting the gating of GABAA receptor channels, resulting in enhanced GABA-mediated currents in neurons. However, mutations mimicking the effect of BZDs on GABAAR channel gating have been shown to also impact the membrane dynamics and synaptic anchoring of the receptors. Here, using single molecule tracking combined with electrophysiological recordings, we show that BZD ligands rapidly influence the dynamic behavior of GABAARs in hippocampal neurons. Application of the inverse BZD agonist DMCM rapidly increased the diffusion and reduced the clustering of GABAARs at synapses, resulting in reduced postsynaptic currents. Conversely, the BZD full agonist diazepam had little effect at rest but reduced lateral diffusion and increased synaptic stabilization and clustering of GABAARs upon sustained neuronal activity, resulting in enhanced potency of inhibitory synapses. These effects occurred in the absence of detectable changes in gephyrin clusters, suggesting they did not reflect a rapid dispersion of the synaptic scaffold. Thus, alterations of the diffusion and synaptic anchoring of GABAARs represent a novel, unsuspected mechanism through which BZDs rapidly modulate GABA signaling in central neurons.

Learning a generative model of images by factoring appearance and shape.

Computer vision has grown tremendously in the past two decades. Despite all efforts, existing attempts at matching parts of the human visual system's extraordinary ability to understand visual scenes lack either scope or power. By combining the advantages of general low-level generative models and powerful layer-based and hierarchical models, this work aims at being a first step toward richer, more flexible models of images. After comparing various types of restricted Boltzmann machines (RBMs) able to model continuous-valued data, we introduce our basic model, the masked RBM, which explicitly models occlusion boundaries in image patches by factoring the appearance of any patch region from its shape. We then propose a generative model of larger images using a field of such RBMs. Finally, we discuss how masked RBMs could be stacked to form a deep model able to generate more complicated structures and suitable for various tasks such as segmentation or object recognition.

Visual deprivation suppresses L5 pyramidal neuron excitability by preventing the induction of intrinsic plasticity.

In visual cortex monocular deprivation (MD) during a critical period (CP) reduces the ability of the deprived eye to activate cortex, but the underlying cellular plasticity mechanisms are incompletely understood. Here we show that MD reduces the intrinsic excitability of layer 5 (L5) pyramidal neurons and enhances long-term potentiation of intrinsic excitability (LTP-IE). Further, MD and LTP-IE induce reciprocal changes in K(v)2.1 current, and LTP-IE reverses the effects of MD on intrinsic excitability. Taken together these data suggest that MD reduces intrinsic excitability by preventing sensory-drive induced LTP-IE. The effects of MD on excitability were correlated with the classical visual system CP, and (like the functional effects of MD) could be rapidly reversed when vision was restored. These data establish LTP-IE as a candidate mechanism mediating loss of visual responsiveness within L5, and suggest that intrinsic plasticity plays an important role in experience-dependent refinement of visual cortical circuits.

Representational power of restricted boltzmann machines and deep belief networks.

Deep belief networks (DBN) are generative neural network models with many layers of hidden explanatory factors, recently introduced by Hinton, Osindero, and Teh (2006) along with a greedy layer-wise unsupervised learning algorithm. The building block of a DBN is a probabilistic model called a restricted Boltzmann machine (RBM), used to represent one layer of the model. Restricted Boltzmann machines are interesting because inference is easy in them and because they have been successfully used as building blocks for training deeper models. We first prove that adding hidden units yields strictly improved modeling power, while a second theorem shows that RBMs are universal approximators of discrete distributions. We then study the question of whether DBNs with more layers are strictly more powerful in terms of representational power. This suggests a new and less greedy criterion for training RBMs within DBNs.

[Acquiring new information in a neuronal network: from Hebb's concept to homeostatic plasticity]. / Acquisition de nouvelles informations dans un réseau neuronal: du concept hebbien à la plasticité homéostatique.

Synaptic plasticity is the cellular mechanism underlying the phenomena of learning and memory. Much of the research on synaptic plasticity is based on the postulate of Hebb (1949) who proposed that, when a neuron repeatedly takes part in the activation of another neuron, the efficacy of the connections between these neurons is increased. Plasticity has been extensively studied, and often demonstrated through the processes of LTP (Long Term Potentiation) and LTD (Long Term Depression), which represent an increase and a decrease of the efficacy of long-term synaptic transmission. This review summarizes current knowledge concerning the cellular mechanisms of LTP and LTD, whether at the level of excitatory synapses, which have been the most studied, or at the level of inhibitory synapses. However, if we consider neuronal networks rather than the individual synapses, the consequences of synaptic plasticity need to be considered on a large scale to determine if the activity of networks are changed or not. Homeostatic plasticity takes into account the mechanisms which control the efficacy of synaptic transmission for all the synaptic inputs of a neuron. Consequently, this new concept deals with the coordinated activity of excitatory and inhibitory networks afferent to a neuron which maintain a controlled level of excitability during the acquisition of new information related to the potentiation or to the depression of synaptic efficacy. We propose that the protocols of stimulation used to induce plasticity at the synaptic level set up a "homeostatic potentiation" or a "homeostatic depression" of excitation and inhibition at the level of the neuronal networks. The coordination between excitatory and inhibitory circuits allows the neuronal networks to preserve a level of stable activity, thus avoiding episodes of hyper- or hypo-activity during the learning and memory phases.

Involvement of NR2A- or NR2B-containing N-methyl-D-aspartate receptors in the potentiation of cortical layer 5 pyramidal neurone inputs depends on the developmental stage.

In the cortex, N-methyl-D-aspartate receptors (NMDARs) play a critical role in the control of synaptic plasticity processes. We have previously shown in rat visual cortex that the application of a high-frequency stimulation (HFS) protocol used to induce long-term potentiation in layer 2/3 leads to a parallel potentiation of excitatory and inhibitory inputs received by cortical layer 5 pyramidal neurones without changing the excitation/inhibition balance of the pyramidal neurone, indicating a homeostatic control of this parameter. We show here that the blockade of NMDARs of the neuronal network prevents the potentiation of excitatory and inhibitory inputs, and this result leaves open to question the role of the NMDAR isoform involved in the induction of long-term potentiation, which is actually being strongly debated. In postnatal day (P)18-23 rat cortical slices, the blockade of synaptic NR2B-containing NMDARs prevents the induction of the potentiation induced by the HFS protocol, whereas the blockade of NR2A-containing NMDARs reduced the potentiation itself. In P29-P32 cortical slices, the specific activation of NR2A-containing receptors fully ensures the potentiation of excitatory and inhibitory inputs. These results constitute the first report of a functional shift in subunit composition of NMDARs during the critical period (P12-P36), which explains the relative contribution of both NR2B- and NR2A-containing NMDARs in synaptic plasticity processes. These effects of the HFS protocol are mediated by the activation of synaptic NMDARs but our results also indicate that the homeostatic control of the excitation/inhibition balance is independent of NMDAR activation and is due to specialized recurrent interactions between excitatory and inhibitory networks.

Homeostatic control of the excitation-inhibition balance in cortical layer 5 pyramidal neurons.

Homeostatic regulation in the brain is thought to be achieved through a control of the synaptic strength by close interactions between excitation and inhibition in cortical circuits. We recorded in a layer 5 pyramidal neuron of rat cortex the composite response to an electrical stimulation of various layers (2-3, 4 or 6). Decomposition of the global conductance change in its excitatory and inhibitory components permits a direct measurement of excitation-inhibition (E-I) balance. Whatever the stimulated layer was, afferent inputs led to a conductance change consisting of 20% excitation and 80% inhibition. Changing synaptic strengths in cortical networks by using a high-frequency of stimulation (HFS) protocol or a low-frequency of stimulation (LFS) protocol (classically used to induce long-term potentiation or long-term depression at the synaptic level) were checked in order to disrupt this balance. Application of HFS protocols in layers 2-3, 4 or 6, or of LFS protocols in layer 4 induced, respectively, long-term paralleled increases or long-term paralleled decreases in E and I which did not change the E-I balance. LFS protocols in layers 2-3 or 6 decreased E but not I and disrupted the balance. It is proposed that regulatory mechanisms might be mainly sustained by recurrent connectivity between excitatory and inhibitory neuronal circuits and by modulation of shunting GABA(A) inhibition in the layer 5 pyramidal neuron.

Learning eigenfunctions links spectral embedding and kernel PCA.

In this letter, we show a direct relation between spectral embedding methods and kernel principal components analysis and how both are special cases of a more general learning problem: learning the principal eigenfunctions of an operator defined from a kernel and the unknown data-generating density. Whereas spectral embedding methods provided only coordinates for the training points, the analysis justifies a simple extension to out-of-sample examples (the Nyström formula) for multidimensional scaling (MDS), spectral clustering, Laplacian eigenmaps, locally linear embedding (LLE), and Isomap. The analysis provides, for all such spectral embedding methods, the definition of a loss function, whose empirical average is minimized by the traditional algorithms. The asymptotic expected value of that loss defines a generalization performance and clarifies what these algorithms are trying to learn. Experiments with LLE, Isomap, spectral clustering, and MDS show that this out-of-sample embedding formula generalizes well, with a level of error comparable to the effect of small perturbations of the training set on the embedding.