A stereological study of synapse number in the epileptic human hippocampus

Hippocampal sclerosis is the most frequent pathology encountered in resected mesial temporal structures from patients with intractable temporal lobe epilepsy (TLE). Here, we have used stereological methods to compare the overall density of synapses and neurons between non-sclerotic and sclerotic hippocampal tissue obtained by surgical resection from patients with TLE. Specifically, we examined the possible changes in the subiculum and CA1, regions that seem to be critical for the development and/or maintenance of seizures in these patients. We found a remarkable decrease in synaptic and neuronal density in the sclerotic CA1, and while the subiculum from the sclerotic hippocampus did not display changes in synaptic density, the neuronal density was higher. Since the subiculum from the sclerotic hippocampus displays a significant increase in neuronal density, as well as a various other neurochemical changes, we propose that the apparently normal subiculum from the sclerotic hippocampus suffers profound alterations in neuronal circuits at both the molecular and synaptic level that are likely to be critical for the development or maintenance of seizure activity

Silencing microRNA-134 produces neuroprotective and prolonged seizure-suppressive effects

Temporal lobe epilepsy is a common, chronic neurological disorder characterized by recurrent spontaneous seizures. MicroRNAs (miRNAs) are small, noncoding RNAs that regulate post-transcriptional expression of protein-coding mRNAs, which may have key roles in the pathogenesis of neurological disorders. In experimental models of prolonged, injurious seizures (status epilepticus) and in human epilepsy, we found upregulation of miR-134, a brain-specific, activity-regulated miRNA that has been implicated in the control of dendritic spine morphology. Silencing of miR-134 expression in vivo using antagomirs reduced hippocampal CA3 pyramidal neuron dendrite spine density by 21% and rendered mice refractory to seizures and hippocampal injury caused by status epilepticus. Depletion of miR-134 after status epilepticus in mice reduced the later occurrence of spontaneous seizures by over 90% and mitigated the attendant pathological features of temporal lobe epilepsy. Thus, silencing miR-134 exerts prolonged seizure-suppressant and neuroprotective actions; determining whether these are anticonvulsant effects or are truly antiepileptogenic effects requires additional experimentation.

Functional and effective connectivity in MEG. Application to the study of epilepsy

Nuestro cerebro contiene cerca de 1014 sinapsis neuronales. Esta enorme cantidad de conexiones proporciona un entorno ideal donde distintos grupos de neuronas se sincronizan transitoriamente para provocar la aparición de funciones cognitivas, como la percepción, el aprendizaje o el pensamiento. Comprender la organización de esta compleja red cerebral en base a datos neurofisiológicos, representa uno de los desafíos más importantes y emocionantes en el campo de la neurociencia. Se han propuesto recientemente varias medidas para evaluar cómo se comunican las diferentes partes del cerebro a diversas escalas (células individuales, columnas corticales, o áreas cerebrales). Podemos clasificarlos, según su simetría, en dos grupos: por una parte, la medidas simétricas, como la correlación, la coherencia o la sincronización de fase, que evalúan la conectividad funcional (FC); mientras que las medidas asimétricas, como la causalidad de Granger o transferencia de entropía, son capaces de detectar la dirección de la interacción, lo que denominamos conectividad efectiva (EC). En la neurociencia moderna ha aumentado el interés por el estudio de las redes funcionales cerebrales, en gran medida debido a la aparición de estos nuevos algoritmos que permiten analizar la interdependencia entre señales temporales, además de la emergente teoría de redes complejas y la introducción de técnicas novedosas, como la magnetoencefalografía (MEG), para registrar datos neurofisiológicos con gran resolución. Sin embargo, nos hallamos ante un campo novedoso que presenta aun varias cuestiones metodológicas sin resolver, algunas de las cuales trataran de abordarse en esta tesis. En primer lugar, el creciente número de aproximaciones para determinar la existencia de FC/EC entre dos o más señales temporales, junto con la complejidad matemática de las herramientas de análisis, hacen deseable organizarlas todas en un paquete software intuitivo y fácil de usar. Aquí presento HERMES (http://hermes.ctb.upm.es), una toolbox en MatlabR, diseñada precisamente con este fin. Creo que esta herramienta será de gran ayuda para todos aquellos investigadores que trabajen en el campo emergente del análisis de conectividad cerebral y supondrá un gran valor para la comunidad científica. La segunda cuestión practica que se aborda es el estudio de la sensibilidad a las fuentes cerebrales profundas a través de dos tipos de sensores MEG: gradiómetros planares y magnetómetros, esta aproximación además se combina con un enfoque metodológico, utilizando dos índices de sincronización de fase: phase locking value (PLV) y phase lag index (PLI), este ultimo menos sensible a efecto la conducción volumen. Por lo tanto, se compara su comportamiento al estudiar las redes cerebrales, obteniendo que magnetómetros y PLV presentan, respectivamente, redes más densamente conectadas que gradiómetros planares y PLI, por los valores artificiales que crea el problema de la conducción de volumen. Sin embargo, cuando se trata de caracterizar redes epilépticas, el PLV ofrece mejores resultados, debido a la gran dispersión de las redes obtenidas con PLI. El análisis de redes complejas ha proporcionado nuevos conceptos que mejoran caracterización de la interacción de sistemas dinámicos. Se considera que una red está compuesta por nodos, que simbolizan sistemas, cuyas interacciones se representan por enlaces, y su comportamiento y topología puede caracterizarse por un elevado número de medidas. Existe evidencia teórica y empírica de que muchas de ellas están fuertemente correlacionadas entre sí. Por lo tanto, se ha conseguido seleccionar un pequeño grupo que caracteriza eficazmente estas redes, y condensa la información redundante. Para el análisis de redes funcionales, la selección de un umbral adecuado para decidir si un determinado valor de conectividad de la matriz de FC es significativo y debe ser incluido para un análisis posterior, se convierte en un paso crucial. En esta tesis, se han obtenido resultados más precisos al utilizar un test de subrogadas, basado en los datos, para evaluar individualmente cada uno de los enlaces, que al establecer a priori un umbral fijo para la densidad de conexiones. Finalmente, todas estas cuestiones se han aplicado al estudio de la epilepsia, caso práctico en el que se analizan las redes funcionales MEG, en estado de reposo, de dos grupos de pacientes epilépticos (generalizada idiopática y focal frontal) en comparación con sujetos control sanos. La epilepsia es uno de los trastornos neurológicos más comunes, con más de 55 millones de afectados en el mundo. Esta enfermedad se caracteriza por la predisposición a generar ataques epilépticos de actividad neuronal anormal y excesiva o bien síncrona, y por tanto, es el escenario perfecto para este tipo de análisis al tiempo que presenta un gran interés tanto desde el punto de vista clínico como de investigación. Los resultados manifiestan alteraciones especificas en la conectividad y un cambio en la topología de las redes en cerebros epilépticos, desplazando la importancia del ‘foco’ a la ‘red’, enfoque que va adquiriendo relevancia en las investigaciones recientes sobre epilepsia. ABSTRACT There are about 1014 neuronal synapses in the human brain. This huge number of connections provides the substrate for neuronal ensembles to become transiently synchronized, producing the emergence of cognitive functions such as perception, learning or thinking. Understanding the complex brain network organization on the basis of neuroimaging data represents one of the most important and exciting challenges for systems neuroscience. Several measures have been recently proposed to evaluate at various scales (single cells, cortical columns, or brain areas) how the different parts of the brain communicate. We can classify them, according to their symmetry, into two groups: symmetric measures, such as correlation, coherence or phase synchronization indexes, evaluate functional connectivity (FC); and on the other hand, the asymmetric ones, such as Granger causality or transfer entropy, are able to detect effective connectivity (EC) revealing the direction of the interaction. In modern neurosciences, the interest in functional brain networks has increased strongly with the onset of new algorithms to study interdependence between time series, the advent of modern complex network theory and the introduction of powerful techniques to record neurophysiological data, such as magnetoencephalography (MEG). However, when analyzing neurophysiological data with this approach several questions arise. In this thesis, I intend to tackle some of the practical open problems in the field. First of all, the increase in the number of time series analysis algorithms to study brain FC/EC, along with their mathematical complexity, creates the necessity of arranging them into a single, unified toolbox that allow neuroscientists, neurophysiologists and researchers from related fields to easily access and make use of them. I developed such a toolbox for this aim, it is named HERMES (http://hermes.ctb.upm.es), and encompasses several of the most common indexes for the assessment of FC and EC running for MatlabR environment. I believe that this toolbox will be very helpful to all the researchers working in the emerging field of brain connectivity analysis and will entail a great value for the scientific community. The second important practical issue tackled in this thesis is the evaluation of the sensitivity to deep brain sources of two different MEG sensors: planar gradiometers and magnetometers, in combination with the related methodological approach, using two phase synchronization indexes: phase locking value (PLV) y phase lag index (PLI), the latter one being less sensitive to volume conduction effect. Thus, I compared their performance when studying brain networks, obtaining that magnetometer sensors and PLV presented higher artificial values as compared with planar gradiometers and PLI respectively. However, when it came to characterize epileptic networks it was the PLV which gives better results, as PLI FC networks where very sparse. Complex network analysis has provided new concepts which improved characterization of interacting dynamical systems. With this background, networks could be considered composed of nodes, symbolizing systems, whose interactions with each other are represented by edges. A growing number of network measures is been applied in network analysis. However, there is theoretical and empirical evidence that many of these indexes are strongly correlated with each other. Therefore, in this thesis I reduced them to a small set, which could more efficiently characterize networks. Within this framework, selecting an appropriate threshold to decide whether a certain connectivity value of the FC matrix is significant and should be included in the network analysis becomes a crucial step, in this thesis, I used the surrogate data tests to make an individual data-driven evaluation of each of the edges significance and confirmed more accurate results than when just setting to a fixed value the density of connections. All these methodologies were applied to the study of epilepsy, analysing resting state MEG functional networks, in two groups of epileptic patients (generalized and focal epilepsy) that were compared to matching control subjects. Epilepsy is one of the most common neurological disorders, with more than 55 million people affected worldwide, characterized by its predisposition to generate epileptic seizures of abnormal excessive or synchronous neuronal activity, and thus, this scenario and analysis, present a great interest from both the clinical and the research perspective. Results revealed specific disruptions in connectivity and network topology and evidenced that networks’ topology is changed in epileptic brains, supporting the shift from ‘focus’ to ‘networks’ which is gaining importance in modern epilepsy research.

A single dose of kainic acid elevates the levels of enkephalins and activator protein-1 transcription factors in the hippocampus for up to 1 year

Neuronal plasticity plays a very important role in brain adaptations to environmental stimuli, disease, and aging processes. The kainic acid model of temporal lobe epilepsy was used to study the long-term anatomical and biochemical changes in the hippocampus after seizures. Using Northern blot analysis, immunocytochemistry, and Western blot analysis, we have found a long-term elevation of the proconvulsive opioid peptide, enkephalin, in the rat hippocampus. We have also demonstrated that an activator protein-1 transcription factor, the 35-kDa fos-related antigen, can be induced and elevated for at least 1 year after kainate treatment. This study demonstrated that a single systemic injection of kainate produces almost permanent increases in the enkephalin and an activator protein-1 transcription factor, the 35-kDa fos-related antigen, in the rat hippocampus, and it is likely that these two events are closely associated with the molecular mechanisms of induction of long-lasting enhanced seizure susceptibility in the kainate-induced seizure model. The long-term expression of the proenkephalin mRNA and its peptides in the kainate-treated rat hippocampus also suggests an important role in the recurrent seizures of temporal lobe epilepsy.

Developmental disorder associated with increased cellular nucleotidase activity

Four unrelated patients are described with a syndrome that included developmental delay, seizures, ataxia, recurrent infections, severe language deficit, and an unusual behavioral phenotype characterized by hyperactivity, short attention span, and poor social interaction. These manifestations appeared within the first few years of life. Each patient displayed abnormalities on EEG. No unusual metabolites were found in plasma or urine, and metabolic testing was normal except for persistent hypouricosuria. Investigation of purine and pyrimidine metabolism in cultured fibroblasts derived from these patients showed normal incorporation of purine bases into nucleotides but decreased incorporation of uridine. De novo synthesis of purines and cellular phosphoribosyl pyrophosphate content also were moderately decreased. The distribution of incorporated purines and pyrimidines did not reveal a pattern suggestive of a deficient enzyme activity. Assay of individual enzymes in fibroblast lysates showed no deficiencies. However, the activity of cytosolic 5′-nucleotidase was elevated 6- to 10-fold. Based on the possibility that the observed increased catabolic activity and decreased pyrimidine salvage might be causing a deficiency of pyrimidine nucleotides, the patients were treated with oral pyrimidine nucleoside or nucleotide compounds. All patients showed remarkable improvement in speech and behavior as well as decreased seizure activity and frequency of infections. A double-blind placebo trial was undertaken to ascertain the efficacy of this supplementation regimen. Upon replacement of the supplements with placebo, all patients showed rapid regression to their pretreatment states. These observations suggest that increased nucleotide catabolism is related to the symptoms of these patients, and that the effects of this increased catabolism are reversed by administration of uridine.

Postnatal neuronal proliferation in mice lacking Ink4d and Kip1 inhibitors of cyclin-dependent kinases

Development of the central nervous system requires proliferation of neuronal and glial cell precursors followed by their subsequent differentiation in a highly coordinated manner. The timing of neuronal cell cycle exit and differentiation is likely to be regulated in part by inhibitors of cyclin-dependent kinases. Overlapping and sustained patterns of expression of two cyclin-dependent kinases, p19Ink4d and p27Kip1, in postmitotic brain cells suggested that these proteins may be important in actively repressing neuronal proliferation. Animals derived from crosses of Ink4d- null with Kip1-null mice exhibited bradykinesia, proprioceptive abnormalities, and seizures, and died at about 18 days after birth. Metabolic labeling of live animals with bromodeoxyuridine at postnatal days 14 and 18, combined with immunolabeling of neuronal markers, showed that subpopulations of central nervous system neurons were proliferating in all parts of the brain, including normally dormant cells of the hippocampus, cortex, hypothalamus, pons, and brainstem. These cells also expressed phosphorylated histone H3, a marker for late G2 and M-phase progression, indicating that neurons were dividing after they had migrated to their final positions in the brain. Increased proliferation was balanced by cell death, resulting in no gross changes in the cytoarchitecture of the brains of these mice. Therefore, p19Ink4d and p27Kip1 cooperate to maintain differentiated neurons in a quiescent state that is potentially reversible.

Neuronal migration is retarded in mice lacking the tissue plasminogen activator gene

Neuronal migration is a critical phase of brain development, where defects can lead to severe ataxia, mental retardation, and seizures. In the developing cerebellum, granule neurons turn on the gene for tissue plasminogen activator (tPA) as they begin their migration into the cerebellar molecular layer. Granule neurons both secrete tPA, an extracellular serine protease that converts the proenzyme plasminogen into the active protease plasmin, and bind tPA to their cell surface. In the nervous system, tPA activity is correlated with neurite outgrowth, neuronal migration, learning, and excitotoxic death. Here we show that compared with their normal counterparts, mice lacking the tPA gene (tPA−/−) have greater than 2-fold more migrating granule neurons in the cerebellar molecular layer during the most active phase of granule cell migration. A real-time analysis of granule cell migration in cerebellar slices of tPA−/− mice shows that granule neurons are migrating 51% as fast as granule neurons in slices from wild-type mice. These findings establish a direct role for tPA in facilitating neuronal migration, and they raise the possibility that late arriving neurons may have altered synaptic interactions.

Disruption of the m1 receptor gene ablates muscarinic receptor-dependent M current regulation and seizure activity in mice

Muscarinic acetylcholine receptors are members of the G protein-coupled receptor superfamily expressed in neurons, cardiomyocytes, smooth muscle, and a variety of epithelia. Five subtypes of muscarinic acetylcholine receptors have been discovered by molecular cloning, but their pharmacological similarities and frequent colocalization make it difficult to assign functional roles for individual subtypes in specific neuronal responses. We have used gene targeting by homologous recombination in embryonic stem cells to produce mice lacking the m1 receptor. These mice show no obvious behavioral or histological defects, and the m2, m3, and m4 receptors continue to be expressed in brain with no evidence of compensatory induction. However, the robust suppression of the M-current potassium channel activity evoked by muscarinic agonists in sympathetic ganglion neurons is completely lost in m1 mutant mice. In addition, both homozygous and heterozygous mutant mice are highly resistant to the seizures produced by systemic administration of the muscarinic agonist pilocarpine. Thus, the m1 receptor subtype mediates M current modulation in sympathetic neurons and induction of seizure activity in the pilocarpine model of epilepsy.

HIV type-1 infection of the cotton rat (Sigmodon fulviventer and S. hispidus)

Cotton rats (Sigmodon hispidus and S. fulviventer) are susceptible to many viruses that infect humans (e.g., poliovirus, respiratory syncytial virus, influenza virus, adenovirus, and parainfluenza virus) and have been influential in developing therapeutic clinical intervention strategies for many viral infections of man. This study set out to determine whether cotton rats are susceptible to infection with HIV type 1 (HIV-1). Results indicate that HIV-1 does infect the cotton rat and S. fulviventer is more susceptible than S. hispidus. The virus was passaged from animal to animal for a total of three serial passages; but HIV replicated poorly in vivo, was only detectable as proviral DNA, and never exceeded one provirus per 1.8 × 105 cotton rat peripheral blood mononuclear cells. Infection induced a distinct and characteristic anti-HIV antibody response that, in some animals, included neutralizing antibodies, recognized all of the major HIV-1 antigens and the antibodies lasted out to 52 wk post-infection. Neonate S. fulviventer were not more susceptible to infection than adults. In vitro culture studies produced indirect evidence of viral replication by detection of viral gag gene RNA in reverse transcriptase–PCR assays on viral culture supernatants. Collectively, these results indicate that HIV-1 can replicate in a nontransgenic rodent and that this system may have potential as an animal model for HIV-1 infection if viral replication rates can be improved in vivo.

Abnormal neurotransmission in mice lacking synaptic vesicle protein 2A (SV2A)

Synaptic vesicle protein 2 (SV2) is a membrane glycoprotein common to all synaptic and endocrine vesicles. Unlike many proteins involved in synaptic exocytosis, SV2 has no homolog in yeast, indicating that it performs a function unique to secretion in higher eukaryotes. Although the structure and protein interactions of SV2 suggest multiple possible functions, its role in synaptic events remains unknown. To explore the function of SV2 in an in vivo context, we generated mice that do not express the primary SV2 isoform, SV2A, by using targeted gene disruption. Animals homozygous for the SV2A gene disruption appear normal at birth. However, they fail to grow, experience severe seizures, and die within 3 weeks, suggesting multiple neural and endocrine deficits. Electrophysiological studies of spontaneous inhibitory neurotransmission in the CA3 region of the hippocampus revealed that loss of SV2A leads to a reduction in action potential-dependent γ-aminobutyric acid (GABA)ergic neurotransmission. In contrast, action potential-independent neurotransmission was normal. Analyses of synapse ultrastructure suggest that altered neurotransmission is not caused by changes in synapse density or morphology. These findings demonstrate that SV2A is an essential protein and implicate it in the control of exocytosis.

Leading role of thalamic over cortical neurons during postinhibitory rebound excitation

The postinhibitory rebound excitation is an intrinsic property of thalamic and cortical neurons that is implicated in a variety of normal and abnormal operations of neuronal networks, such as slow or fast brain rhythms during different states of vigilance as well as seizures. We used dual simultaneous intracellular recordings of thalamocortical neurons from the ventrolateral nucleus and neurons from the motor cortex, together with thalamic and cortical field potentials, to investigate the temporal relations between thalamic and cortical events during the rebound excitation that follows prolonged periods of stimulus-induced inhibition. Invariably, the rebound spike-bursts in thalamocortical cells occurred before the rebound depolarization in cortical neurons and preceded the peak of the depth-negative, rebound field potential in cortical areas. Also, the inhibitory-rebound sequences were more pronounced and prolonged in cortical neurons when elicited by thalamic stimuli, compared with cortical stimuli. The role of thalamocortical loops in the rebound excitation of cortical neurons was shown further by the absence of rebound activity in isolated cortical slabs. However, whereas thalamocortical neurons remained hyperpolarized after rebound excitation, because of the prolonged spike-bursts in inhibitory thalamic reticular neurons, the rebound depolarization in cortical neurons was prolonged, suggesting the role of intracortical excitatory circuits in this sustained activity. The role of intrathalamic events in triggering rebound cortical activity should be taken into consideration when analyzing information processes at the cortical level; at each step, corticothalamic volleys can set into action thalamic inhibitory neurons, leading to rebound spike-bursts that are transferred back to the cortex, thus modifying cortical activities.

Cultured melanocytes from dilute mutant mice exhibit dendritic morphology and altered melanosome distribution

Mutant alleles at the dilute unconventional myosin heavy chain locus cause diluted coat color, opisthotonic seizures, and death. The dilute coat color phenotype is caused by irregular clumping of pigment in the hair, but amounts of melanin are unchanged from wild-type controls. The melanocyte phenotype has been described as adendritic, since hair bulb and Harderian gland melanocytes appear to be rounded in tissue sections. These observations do not exclude the possibility that the processes lack pigment, since the melanocyte shape was judged by the distribution of melanin. We have tested this hypothesis by culturing primary melanocytes from dilute mutant and wild-type mice. The mutant melanocytes do not lack processes; instead, they exhibit a concentrated perinuclear distribution of melanosomes, while wild-type melanocytes have a very uniform cytoplasmic distribution of melanosomes. Electron micrographs show no detectable differences in melanosome morphology or maturation between dilute and wild-type melanocytes. Immunofluorescence experiments indicate that the dilute protein is concentrated in regions of the cytoplasm that contain melanosomes. These experiments show that the dilute myosin is necessary for the localization of melanosomes, either by active transport or tethering.

Molecular determinants of coordinated proton and zinc inhibition of N-methyl-d-aspartate NR1/NR2A receptors

Modulation of the N-methyl-d-aspartate (NMDA)-selective glutamate receptors by extracellular protons and Zn2+ may play important roles during ischemia in the brain and during seizures. Recombinant NR1/NR2A receptors exhibit a much higher apparent affinity for voltage-independent Zn2+ inhibition than receptors with other subunit combinations. Here, we show that the mechanism of this apparent high-affinity, voltage-independent Zn2+ inhibition for NR2A-containing receptors results from the enhancement of proton inhibition. We also show that the N-terminal leucine/isoleucine/valine binding protein (LIVBP)-like domain of the NR2A subunit contains critical determinants of the apparent high-affinity, voltage-independent Zn2+ inhibition. Mutations H42A, H44G, or H128A greatly increase the Zn2+ IC50 (by up to ≈700-fold) with no effect on the potencies of glutamate and glycine or on voltage-dependent block by Mg2+. Furthermore, the amino acid residue substitution H128A, which mediates the largest effect on the apparent high-affinity Zn2+ inhibition among all histidine substitutions we tested, is also critical to the pH-dependency of Zn2+ inhibition. Our data revealed a unique interaction between two important extracellular modulators of NMDA receptors.

Epilepsy in mice deficient in the 65-kDa isoform of glutamic acid decarboxylase

γ-Aminobutyric acid (GABA), the major inhibitory neurotransmitter in the mammalian brain, is synthesized by two glutamate decarboxylase isoforms, GAD65 and GAD67. The separate role of the two isoforms is unknown, but differences in saturation with cofactor and subcellular localization suggest that GAD65 may provide reserve pools of GABA for regulation of inhibitory neurotransmission. We have disrupted the gene encoding GAD65 and backcrossed the mutation into the C57BL/6 strain of mice. In contrast to GAD67−/− animals, which are born with developmental abnormalities and die shortly after birth, GAD65−/− mice appear normal at birth. Basal GABA levels and holo-GAD activity are normal, but the pyridoxal 5′ phosphate-inducible apo-enzyme reservoir is significantly decreased. GAD65−/− mice develop spontaneous seizures that result in increased mortality. Seizures can be precipitated by fear or mild stress. Seizure susceptibility is dramatically increased in GAD65−/− mice backcrossed into a second genetic background, the nonobese diabetic (NOD/LtJ) strain of mice enabling electroencephalogram analysis of the seizures. The generally higher basal brain GABA levels in this backcross are significantly decreased by the GAD65−/− mutation, suggesting that the relative contribution of GABA synthesized by GAD65 to total brain GABA levels is genetically determined. Seizure-associated c-fos-like immunoreactivity reveals the involvement of limbic regions of the brain. These data suggest that GABA synthesized by GAD65 is important in the dynamic regulation of neural network excitability, implicate at least one modifier locus in the NOD/LtJ strain, and present GAD65−/− animals as a model of epilepsy involving GABA-ergic pathways.

Glutamic acid decarboxylase and glutamate receptor changes during tolerance and dependence to benzodiazepines

Protracted administration of diazepam elicits tolerance, whereas discontinuation of treatment results in signs of dependence. Tolerance to the anticonvulsant action of diazepam is present in an early phase (6, 24, and 36 h) but disappears in a late phase (72–96 h) of withdrawal. In contrast, signs of dependence such as decrease in open-arm entries on an elevated plus-maze and increased susceptibility to pentylenetetrazol-induced seizures were apparent 96 h (but not 12, 24, or 48 h) after diazepam withdrawal. During the first 72 h of withdrawal, tolerance is associated with changes in the expression of GABAA (γ-aminobutyric acid type A) receptor subunits (decrease in γ2 and α1; increase in α5) and with an increase of mRNA expression of the most abundant form of glutamic acid decarboxylase (GAD), GAD67. In contrast, dl-α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid (AMPA) receptor GluR1 subunit mRNA and cognate protein, which are normal during the early phase of diazepam withdrawal, increase by approximately 30% in cortex and hippocampus in association with the appearance of signs of dependence 96 h after diazepam withdrawal. Immunohistochemical studies of GluR1 subunit expression with gold-immunolabeling technique reveal that the increase of GluR1 subunit protein is localized to layer V pyramidal neurons and their apical dendrites in the cortex, and to pyramidal neurons and in their dendritic fields in hippocampus. The results suggest an involvement of GABA-mediated processes in the development and maintenance of tolerance to diazepam, whereas excitatory amino acid-related processes (presumably via AMPA receptors) may be involved in the expression of signs of dependence after withdrawal.