Oscillatory activity is a hallmark of neuronal network function in various brain regions, including the olfac-tory bulb, thalamus, hippocampus and neocortex1. The frequency of network oscillations covers more than three orders of magnitude, from slow oscillations in the delta (0.5–3 Hz) and theta (3–8 Hz) ranges to fast oscillations in the gamma (30–90 Hz) and ultrafast (90–200 Hz) ranges1. Within this spectrum, gamma oscillations have received particular attention, because their relationship to higher brain functions is most evident2,3. Gamma oscillations have been proposed to represent reference signals for temporal encoding4,5, sensory binding of fea-tures into a coherent percept2, and storage and recall of information6,7. Conversely, disruption of gamma oscil-lations could underlie some psychiatric disorders, such as schizophrenia8,9.To understand how oscillations contribute to higher brain functions, it is essential to first consider the basic underlying mechanisms. A key requirement for the generation of network oscillations is regular and syn-chronized neuronal activity. If a neuron fires action potentials in a regular manner, rhythmic activation of output synapses generates a periodic fluctuation in the intracellular membrane potential of all postsynaptic target cells10. If several neurons fire action potentials both regularly and synchronously, this fluctuating out-put signal is amplified, defining temporal windows of increased and reduced excitability in a larger population of target cells. At the same time, the rhythmic synaptic activation pattern results in a fluctuating field potential signal, which can easily be measured using extracellular recording electrodes11. The divergence of synaptic con-nections leads to a high level of spatial coherence of network oscillations. Such highly coherent oscillations might be ideal reference signals for temporal encoding and sensory binding in large neuronal ensembles2,5.Although gamma oscillations occur in all cortical areas1,2, they have been particularly well studied in the hippocampus11–13. There are several reasons for this. First, the power of extracellularly recorded gamma oscillations is higher in the hippocampus than in other brain regions, owing to the simple laminated archi-tecture of the hippocampal circuit14. Second, gamma oscillations in the hippocampus are evoked under specific behavioural conditions, such as exploration, when they typically coexist with theta oscillations12. This allows researchers to analyse the relationship between network oscillations and behaviour. Finally, as the hippocampus is essential for spatial navigation and episodic memory, the relevance of network oscillations for coding, storing and recalling information can be investigated7,15,16. Although hippocampal gamma oscil-lations have been studied for decades, the underlying neuronal and synaptic mechanisms have only recently come to light.In this article, we aim to summarize the synaptic mechanisms of gamma oscillations in the cortex, with a primary focus on the hippocampus. We review the dependence of gamma oscillations on synaptic inhibi-tion and the role of fast-spiking, parvalbumin-expressing γ-aminobutyric acid (GABA)-containing interneu-rons. Next, we explain how mutual inhibition leads to *Physiologisches Institut der Universität Freiburg, Abteilung 1, Hermann Herder Strasse 7, D-79104 Freiburg, Germany. ‡Institut für Anatomie und Zellbiologie der Universität Freiburg, Abteilung Neuroanatomie, Albertstrasse 17, D-79104 Freiburg, Germany. Correspondence to P. J. e-mail: [email protected]:10.1038/nrn2044Divergence The number of postsynaptic target neurons innervated by a particular neuron. By contrast, convergence is the number of presynaptic neurons innervating a given neuron.Spatial coherence The correlation between signals at two different locations for all times (whereas temporal coherence is the correlation between signals at two different times for the same location). The term was originally defined in physics, but is also widely used in neuroscience.Synaptic mechanisms of synchronized gamma oscillations in inhibitory interneuron networksMarlene Bartos*, Imre Vida‡ and Peter Jonas*Abstract | Gamma frequency oscillations are thought to provide a temporal structure for information processing in the brain. They contribute to cognitive functions, such as memory formation and sensory processing, and are disturbed in some psychiatric disorders. Fast-spiking, parvalbumin-expressing, soma-inhibiting interneurons have a key role in the generation of these oscillations. Experimental analysis in the hippocampus and the neocortex reveals that synapses among these interneurons are highly specialized. Computational analysis further suggests that synaptic specialization turns interneuron networks into robust gamma frequency oscillators.REVIEWSNATURE REVIEWS | NEUROSCIENCE VOLUME 8 | JANUARY 2007 | 45© 2007 Nature Publishing GroupControlacbControl Control+100 nMKainate+50 μMGYKI 53655+3 μMBicucullineExcitationdependenceCarbacholCA3Inhibitiondependence+20 μMCarbachol+20 μM Carbachol+10 μM Bicuculline+20 μM NBQXBicuculline10 mV300 msPower (μV2)1000020 60 10075502540 80 120Frequency (Hz)100 μV100 ms100 μV100 ms200 μV100 msPNIN ININElevated K+CA1, CA3KainateCA3mGluRCA1Parvalbumin A calcium-binding protein that contains EF-hand (helix–loop–helix) motifs. In the hippocampus, parvalbumin is selectively expressed in fast-spiking basket cells and axo-axonic cells. Although the function of parvalbumin is not fully understood, its expression represents a reliable marker for interneuron identification.Network models Computational models of neuronal networks, in which individual neurons (integrate-and-fire or conductance-based elements) are coupled by inhibitory synapses, excitatory synapses or gap junctions.Gap junctions Morphologically specialized electrical and biochemical connections between two cells, which are formed by transcellular channels. A gap junction channel is composed of two hemichannels (connexons), each of which consists of six subunits (connexins). Gap junctions are blocked by octanol and carbenoxolone; however, these blockers are not absolutely specific.Acute hippocampal slices 200–400-μm-thick sections of the hippocampus, typically cut with a tissue slicer in the transverse plane. In comparison to the in vivo brain, the acute slice offers easy access in electrophysiological