Information Processing and Transfer at the Hippocampus–Entorhinal– Neocortical Interface

As information is propelled along the multisynaptic feedforward loops of the entorhinal–hippocampal system, each stage adds unique features to the incoming information (Figure 7.1). Such local operations require time, and are generally re ected by macroscopic oscillations. In each oscillatory cycle, recruitment of principal neurons is temporally protracted and terminated by the buildup of inhibition. In addition to providing a temporal framework in which information can be packaged, oscillatory coupling across networks can facilitate the exchange of information and determine the direction of activity ow. Potential mechanisms in the entorhinal–hippocampal system supporting these hypotheses are described. Oscillations Provide the Structure for Information Processing in the Hippocampus Two major network patterns dominate the hippocampal system: theta oscillations (4–10 Hz) and sharp waves with their associated ripples (140–200 Hz). Theta and sharp-wave patterns also de ne states of the hippocampus: the theta state is associated with exploratory (“preparatory”) movement and REM sleep, whereas intermittent sharp waves mark immobility, consummatory behaviors, and slow-wave sleep. These two competing states bias the direction of information ow to a great extent, with neocortical–hippocampal transfer taking Chapter from "Dynamic Coordination in the Brain: From Neurons to Mind," edited by C. von der Malsburg, W. A. Phillips, and W. Singer. Strüngmann Forum Report, vol. 5. ISBN 978-0-262-01471-7. Cambridge, MA: The MIT Press. 102 G. Buzsáki and K. Diba place mainly during theta oscillations and hippocampal–neocortical transfer during sharp waves (Isomura et al. 2006). The extracellularly recorded theta oscillation is the result of coherent membrane potential oscillations across neurons in all hippocampal subregions (Buzsáki 2002). Theta currents derive from multiple sources, including synaptic currents, intrinsic currents of neurons, dendritic Ca2+ spikes, and other voltage-dependent membrane oscillations. Theta frequency modulation of perisomatic interneurons provides an outward current in somatic layers and phase-biases the power of ongoing gamma frequency oscillations (30–100 Hz), the result of which is a theta-nested gamma burst. Excitatory afferents form active sinks (inward current) at con ned dendritic domains within cytoarchitecturally organized layers in every region. Each layer-speci c excitatory input is complemented by one or more families of interneurons with similar axonal projections (Freund and Buzsáki 1996; Klausberger and Somogyi 2008), forming layer-speci c “theta” dipoles. The resulting rich consortium of theta generators in the hippocampal and parahippocampal regions is coordinated by the medial septum and a network of long-range interneurons. Furthermore, the power, coherence, and phase of theta oscillators can uctuate signi cantly in a layer-speci c manner as a function of overt behavior and/or the memory “load” to support task performance (Montgomery et al. 2009). Figure 7.1 Multiple loops of the hippocampal–entorhinal circuits. The long loop connecting the layer 2 entorhinal cortex (EC), granule cells (gc), CA3, CA1, and subiculum (S) back to the layer 5 EC is supplemented by multiple shortcuts and superimposed loops. The shortest loop between the EC and hippocampus is the path from the layer 3 EC to CA1 and back to the layer 5 EC. Excitatory traf c in the multiple loops is controlled by a large family of interneurons, whose connections are not loop-like (Freund and Buzsáki 1996). mc: mossy cells of the hilus; A: amygdala; RE: nucleus reuniens of thalamus; pFC: prefrontal, anterior cingulate cortex. Chapter from "Dynamic Coordination in the Brain: From Neurons to Mind," edited by C. von der Malsburg, W. A. Phillips, and W. Singer. Strüngmann Forum Report, vol. 5. ISBN 978-0-262-01471-7. Cambridge, MA: The MIT Press. Oscillation-supported Information Processing and Transfer 103 Theta-nested gamma oscillations are generated primarily by the interaction between interneurons and/or between principal cells and interneurons. In both scenarios, the frequency of oscillations is largely determined by the time course of GABAA receptor-mediated inhibition. Neurons that discharge within the time period of the gamma cycle (8–30 ms) de ne a cell assembly (Harris et al. 2003). Given that the membrane time constant of pyramidal neurons in vivo is also within this temporal range, recruiting neurons into this assembly time window is the most effective mechanism for discharging the downstream postsynaptic neuron(s) on which the assembly members converge. Although gamma oscillations can emerge in each hippocampal region, they can be coordinated across regions by either excitatory connections or long-range interneurons. The CA3–CA1 regions appear to form a large coherent gamma oscillator, due to the interaction between the recurrently excited CA3 pyramidal cells and their interneuron targets in both CA3 and CA1 regions. This “CA3 gamma generator” is normally under the suppressive control of the entorhinal–dentate gamma generator, and its power is enhanced severalfold when the entorhinal– dentate input is attenuated (Bragin et al. 1995). Entorhinal circuits generate their own gamma oscillations by largely similar rules, and these (generally faster) rhythms can be transferred and detected in the hippocampus. When the subcortical modulatory inputs decrease in tone, theta oscillations are replaced by large amplitude eld potentials called sharp waves (SPW). SPWs are initiated by the self-organized population bursts of the CA3 pyramidal cells (Buzsáki et al. 1983). The CA3-induced depolarization of CA1 pyramidal cell dendrites results in a prominent extracellular negative wave, from which the SPW derives its name, in the stratum radiatum. The CA1 SPWs are associated with fasteld oscillations (140–200 Hz), or “ripples” con ned to the CA1 pyramidal cell layer (O’Keefe and Nadel 1978; Buzsáki et al. 1992). At least two factors contribute to the eld ripples. First, the synchronous discharge of pyramidal neurons generates repetitive “mini populations spikes” that are responsible for the spike-like appearance of the troughs of ripples in the pyramidal cell layer. Second, the rhythmic positive “wave” components re ect synchronously occurring oscillating inhibitory postsynaptic potentials (IPSPs) in the pyramidal cells because the CA3–CA1 pyramidal cells strongly drive perisomatic interneurons during the SPW. In the time window of SPWs, 50,000–100,000 neurons discharge synchronously in the CA3–CA1–subicular complex–entorhinal axis. The population burst is characterized by a threeto vefold gain of network excitability in the CA1 region, preparing the circuit for synaptic plasticity (Csicsvari et al. 1999a). SPWs have been hypothesized to play a critical role in transferring transient memories from the hippocampus to the neocortex for permanent storage (Buzsáki 1989), and this hypothesis is supported by numerous experiments demonstrating that the neuronal content of the SPW ripple is largely determined by recent waking experiences (Wilson and McNaughton 1994; Foster and Wilson 2006; Csicsvari et al. 2007). Chapter from "Dynamic Coordination in the Brain: From Neurons to Mind," edited by C. von der Malsburg, W. A. Phillips, and W. Singer. Strüngmann Forum Report, vol. 5. ISBN 978-0-262-01471-7. Cambridge, MA: The MIT Press. 104 G. Buzsáki and K. Diba Reciprocal Information Transfer by Oscillations Oscillations and neuronal synchrony create effective mechanisms for the storage, readout, and transfer of information between different structures. Oscillations impose a spatiotemporal structure on neural ensemble activity within and across different brain areas, and allow for the packaging of information in quanta of different durations. Furthermore, oscillations support the bidirectional ow of information across different structures through the changing of the temporal offset in the oscillation-related ring (Amzica and Steriade 1998; Chrobak and Buzsáki 1998b; Sirota et al. 2003; Buzsáki 2005; Siapas et al. 2005). Most importantly, exchanging information across structures by oscillations involves mechanisms different from what is usually meant by the term “information transfer.” In the usual sense, transfer of information involves two structures, or systems, which can be designated as the “source” (sender) and “target” (receiver). Typically, the information transfer process is assumed to be unidirectional and passive: the source sends the information to an ever-ready recipient. In systems coupled by oscillations, however, the method appears to be different; we refer to this process as “ reciprocal information transfer” (Sirota and Buzsáki 2005; Sirota et al. 2008). The reciprocal process implies that a target structure takes the initiative by temporally biasing the activity in the sender (information source) structure (Sirota et al. 2003; Fries 2005; Sirota and Buzsáki 2005; Isomura et al. 2006; Womelsdorf et al. 2007). Biasing is achieved by the strong output (“duty cycle”) of the receiver so that the information, contained in ner timescale gamma-structured spike trains, reaches the recipient structure in its most sensitive state (the “perturbation” cycle), ideal for reception. Below, we illustrate this principle using the state-dependent communication between the hippocampus and neocortex. In the waking state, transfer of neocortical information to the hippocampus can be initiated by the hippocampus via theta-phase biasing of neocortical network dynamics, as re ected in the local eld potential (LFP) by transient gamma oscillations in widespread and relatively isolated neocortical areas (Sirota et al. 2008). As a result, locally generated gamma oscillations from multiple neocortical locations are time biased so that the information contained in their gamma bursts arrive back at the hippocampus at a phase of the theta cycle optimal for maximal perturbation of hippocampal networks and plasticity (Huerta and Lisman 1996; Holscher et al. 1997). In the CA1 region, this corresponds to the positive (least active, i.e., recipient) phase of the theta cycle (Csicsvari et al. 1999b). This is also the phase at which a hippocampal neuron discharges when the rat enters its place eld (O’Keefe and Recce 1993). In short, through theta-phase biasing, the hippocampus can affect multiple neocortical sites so that it can effectively receive information from the resulting neocortical assemblies, by way of the entorhinal cortex (EC), at the optimal time frame. Chapter from "Dynamic Coordination in the Brain: From Neurons to Mind," edited by C. von der Malsburg, W. A. Phillips, and W. Singer. Strüngmann Forum Report, vol. 5. ISBN 978-0-262-01471-7. Cambridge, MA: The MIT Press. Oscillation-supported Information Processing and Transfer 105 The direction of information transfer during slow-wave sleep at the hippocampal–neocortex axis is largely opposite to that in the waking theta state. As discussed earlier, in the absence of theta oscillations, the CA3–CA1 region gives rise to SPWs. The synchronous discharge of CA1 neurons and downstream subicular and EC neurons provides a most effective output to the neocortex (Chrobak and Buzsáki 1994). During sleep the information source (sender) is the hippocampus but, again, the transfer of information is initiated by the receiver (now the neocortex). This latter process has been hypothesized to be critical in consolidating the learned information acquired in the waking state (Buzsáki 1989). A caveat in this two-stage model of information transfer and memory consolidation is the absence of a mechanism to coordinate and guide the hippocampal output temporally to the ongoing activity in the neocortical circuits that gave rise to the experience-dependent hippocampal input. In other words, a mechanism must exist to allow the hippocampal output during sharp wave ripples (SWRs) to address the relevant neocortical circuits. Below, we outline one such potential mechanism. While the hippocampus is generating SPWs during slow-wave sleep, large areas of the neocortex oscillate coherently at a slow frequency (0.5–1.5 Hz) (Steriade et al. 1993a, b; Destexhe et al. 1999). During these slow oscillations, large areas of the neocortex and paleocortex (Isomura et al. 2006) toggle coherently between active (UP) and silent (DOWN) states, although isolated cortical modules can also shift between states, independent of surrounding areas (Figure 7.2). The DOWN–UP transitions can trigger K complexes and