Activity patterns in the hippocampus

Movement on a linear track

We now consider the run along a linear track (cf. also Figure 2.19). During the traversal of the track, neurons which encode a position in this particular environment are activated in the same order as the the place fields are traversed, and thus define a spike pattern (sequence) that corresponds to that specific track. Interestingly, this sequence is replayed during sleep or resting phases following the exploration in a highly compressed manner (Wilson and McNaughton, 1994;

Nadasdy et al., 1999; Lee and Wilson, 2002; Ji and Wilson, 2007; Davidson et al., 2009, and cf.

also Figure 2.25).

Theta phase precession

Figure 2.26: Hippocampal spatio-temporal field. The firing rate of a hip-pocampal neuron as a function of position and phase. Modified from Mehta et al.

(2002) with permission.

The learning of spike patterns taking places during ex-ploration is most likely based on the phenomenon of phase precession, a form of temporal coding of position complementing the rate coding discussed above (cf. also Section 2.4.1 and Figure 2.19).

While exploration of space prominent rhythmic modula-tions of the local field potential (LFP) in the frequency range 4−10Hz, called theta oscillations, are observed, and the timing of single spikes of place cells reflects the position within the corresponding place field (O’Keefe and Recce, 1993; Skaggs et al., 1996; Mehta et al., 2002;

Maurer and McNaughton, 2007; Gupta et al., 2012):

When entering the place field spikes occur preferentially late in the theta-cycle and move to earlier and earlier times while the place field is traversed (cf. Figure 2.26).

Neurons with overlapping place fields are activated during one theta-cycle, and the phenomenon of phase precession yields a compressed representation of recently traversed, present and future locations in space: Neurons representing positions already passed, spike (on average) early in each theta cycle (i.e., at lower phases with respect to the theta rhythm), followed by neurons representing the present position and neurons corresponding to places that will be passed in the future. The average time difference between spikes of place cells with nearby place field centers is in the order of tens of milliseconds and therefore perfectly suited for inducing spike time dependent plasticity (cf. Section 2.1.4). Thus, the combination of the compressed representation of the sequence of locations by phase precession favors the emergence of feed-forward structures (Skaggs et al., 1996; Buzs´aki, 2006; Mehta et al., 1997; Bush et al., 2010). These structures are candidates for the later replay of the spike sequences during sleep (cf. also “Spatial exploration phase” in Chapter 7).

Sharp-Wave-Ripples

The replay of experienced spike patterns occurs in sleep (slow-wave sleep) or resting phases after exploration during short phases of strongly enhanced network activity, called “sharp waves”, superimposed by high-frequency oscillations, called “ripples” (Buzs´aki et al., 1992; Ylinen et al., 1995; Maier et al., 2003, 2011; Buzs´aki and Silva, 2012). The sharp-wave-ripple complex (SWR) has a duration of approximately 50−150ms and the ripple frequency is in the range of 100− 200Hz (cf. Figure 2.27). SWRs are reported all over the hippocampus and in parts of the entorhinal cortex, however, the superimposed oscillations are most prominent and have the highest frequency (up to 200Hz) in the CA1 region (Chrobak and Buzs´aki, 1996; Csicsvari et al., 1999a; Andersen et al., 2007; Sullivan et al., 2011).

Figure 2.27: Unfiltered extracellular local field potential (LFP) recording of SWRs in CA1 (top) and120−300Hz bandpassed-filtered ver-sion (below). Figure reproduced from Maier et al.

(2011) with permission.

A moderate fraction of approximately 10− 20% of pyramidal neurons (Ylinen et al., 1995;

Buzs´aki and Silva, 2012) contribute only a single spike or single burst (Buzs´aki et al., 1992) to the SWR event, nonetheless, the co-ordination of single neuronal discharges over the whole CA3-CA1-subicular complex-EC re-gion makes them the most synchronous event observed in the mammalian brain (Chrobak and Buzs´aki, 1996; Csicsvari et al., 1999a,b, 2000; Buzs´aki and Silva, 2012). Although, the sharp-wave is highly coordinated over all par-ticipating areas via the polysynaptic loop, the high-frequency oscillations (ripples), in par-ticular in CA1, are generated locally (Yli-nen et al., 1995; Csicsvari et al., 1999a; Maier et al., 2011; Sullivan et al., 2011).

Models of Sharp-Wave-Ripples

Experiments show that the pyramidal neurons fire action potentials synchronized to the local field potential oscillations (ripples) during the SWRs (Buzs´aki et al., 1992; Csicsvari et al., 1999b; Maier et al., 2011). Likewise the spiking probability of various types of interneurons change during the SWR events (Klausberger et al., 2003, 2004, 2005; Klausberger, 2009). In particular, basket cells (targeting the soma of pyramidal neurons — and therefore having a powerful direct inhibitory control) and bi-stratified cells (targeting the dendrites in stratum radiatum and stratum oriens) show increased spiking activity in SWR episodes (Klausberger et al., 2003, 2004).

This observations lead to the idea that transiently increased excitatory input (e.g., from CA3 to CA1 via the Schaffer collaterals) during sharp-waves excites oscillations in the interneuron network of basket cells (Ylinen et al., 1995; Buzs´aki and Chrobak, 1995; Brunel and Wang, 2003; Geisler et al., 2005). The induced oscillations in the sharp-wave ripple frequency range then entrain the phasic spiking of the pyramidal cells (Ylinen et al., 1995; Buzs´aki and Chrobak, 1995; Brunel and Wang, 2003; Geisler et al., 2005). Alternatively, it was proposed that the high-frequency oscillations are based on axo-axonal gap junctions (Traub et al., 1999; Traub and Bibbig, 2000; Maex and Schutter, 2007). Here, it is assumed that the axons of the pyramidal neurons form a sparse (electrically coupled) network, where spikes may propagate and multiply in the presence of transiently increased depolarizing input. This yields rhythmic generation of bursts of axonal spiking, which excites the pyramidal cell somata to spike after antidromic and orthodromic propagation.

The third model (considered in Chapter 7) is based on fast dendritic spikes (Memmesheimer, 2010): Spontaneous fluctuations (or appropriate stimulations) cause (weakly) synchronized firing of a subset of pyramidal neurons. The impact of this synchronized pulse on postsynaptic neurons is amplified by synchrony sensitive dendritic spikes (cf. also Section 2.2.2), and thus generates an even more synchronized pulse which in turn may cause synchronous spiking in the subset of postsynaptic neurons, etc. By this mechanism larger and larger groups of neurons spike synchronously and establish neuronal oscillations resembling an SWR events. The oscillation frequency (time difference between subsequent groups) is determined by the conductance delays and the initiation time of dendritic sodium spikes. For sparsely and locally coupled networks (like CA1) and plausible parameters for dendritic spike generation (cf. Ariav et al., 2003; M¨uller et al., 2012) the oscillation frequency is in the range of approximately 200Hz, for more globally coupled networks (and thus larger average conductance delays; like CA3) it decreases (Memmesheimer, 2010), consistent with experimental findings (Chrobak and Buzs´aki, 1996; Ylinen et al., 1995;

Csicsvari et al., 1999a; Sullivan et al., 2011). This may also explain, why the ripple frequency in in vitroslice is generally higher than those detectedin vivo (Maier et al., 2003; Nimmrich et al., 2005; Both et al., 2008), since long-range connections are decreased during the preparation.

We remark that we discuss the plausibility of this model comprehensively in the light of recent experimental findings in Chapter 7.

Replay

Replay of experienced spike sequences during sleep has been demonstrated in a wealth of exper-imental studies (Wilson and McNaughton, 1994; Nadasdy et al., 1999; Lee and Wilson, 2002;

Ji and Wilson, 2007; Davidson et al., 2009, and others) and is thought be initiated in CA3 (Chrobak and Buzs´aki, 1996; Csicsvari et al., 1999a, 2000; O’Neill et al., 2008). In this com-paratively densely coupled region an initial coincident activation (spontaneously or induced) of a subgroub of cells (“initiator cells” Buzs´aki, 1989) is assumed to trigger the activation of the whole previously stored sequence (cf. also Section 2.4.2 about signal transmission in feed-forward networks). Then the activity is projected to region CA1 via the Shaffer Collaterals, where the high-frequency oscillations are superimposed by a local network mechanism (see subsections above).

A replay in CA1 itself based on recurrent connections is assumed to be unlikely because of the sparse recurrent connectivity. However, recentin vivo experiments with mutants having a pro-jection from CA3 to CA1 (Shaffer collaterals) that can be temporally deactivated (Nakashiba et al., 2008, 2009), have shown that replay during SWRs in CA1 still persists if CA1 is deaffer-ented from CA3 (Nakashiba et al., 2009).