During slow wave sleep, the neocortex engages in largely synchronized activity patterns, with alternation of periods of generalized elevated activity (“up-states”) and depressed activity (“down-states”) (e.g., Steriade and Buzsaki 1990; Cowan and Wilson 1994; Timofeev et al. 2001; Petersen et al. 2003). This process appears to be a network phenomenon that originates in the neocortex. Up-states are supported by excitatory intracortical connections (Steriade et al. 1993; Sanchez-Vives and McCormick 2000) balanced by an increased activity of inhibitory interneurons (Timofeev et al. 2001). The positive feedback induced by these recurrent connections makes the cortical network capable of sustaining coherent activity fluctuations at several time scales, from the slow (<1 Hz), regular oscillations observed in ketamine-anesthetized animals (Steriade et al. 1993; Destexhe et al. 1999), to the more irregular, slow (~1 Hz) fluctuations seen in the un-anesthetized animal (Timofeev et al. 2001; Petersen et al. 2003) and also in human surface electroencephalogram (Achermann and Borbely 1997), to the activity in the faster delta frequency range (2-4 Hz, e.g., Ball et al. 1977). The slower end of this frequency range is often referred to as “up/down-state fluctuations.” Such activities in different frequency domains can actually coexist in a coordinate fashion: for example, delta oscillations are especially prominent during up-states, with each up-state encompassing several delta cycles, and the amplitude of delta declining from the beginning to the end of the up-states.

The concomitant hippocampal EEG pattern, termed large irregular activity (LIA), is characterized by short (50-150 msec) episodes of coherent burst firing, accompanied by dendritic sharp wave (SPW) LFPs and high-frequency LFP oscillations in the CA1 pyramidal layer (100-200 Hz, ripple oscillations) (O'Keefe and Nadel 1978; Buzsaki et al. 1992; Csicsvari et al. 2000), amidst relatively suppressed activity. SPWs occur at random intervals with a mean frequency of ~1 Hz during LIA, and appear to be generated in CA3, the portion of the hippocampus richest in recurrent connections (Ishizuka et al. 1990).

SPWs influence the firing of neurons in the entorhinal cortex (the major cortical efferent of the hippocampus) (Chrobak and Buzsaki 1994, 1996), and are weakly correlated with the occurrence of thalamocortical spindle oscillations (Siapas and Wilson 1998), which are also related to the occurrence of up-states (Steriade 2000). Sirota et al. (2003) showed that cortical oscillations in somatosensory cortex and hippocampal activity are related on the short time scale: Ripple events are associated with cortical discharges occurring 50-100 msec earlier. Moreover, they showed that both spindle and delta oscillations affect hippocampal activity, with hippocampal synaptic inputs phase-locked to the cortical oscillations, possibly influencing the generation of SPWs.

Hippocampal memory trace reactivation is strongest during SPWs (Kudrimoti et al. 1999), indicating that trace reactivation might, indeed, reflect the information stored in the CA3 synaptic matrix (McNaughton 1983; Buzsaki 1989; Shen and McNaughton 1996). Hippocampal SPWs, therefore, are a possible mechanism for the transfer of reactivated memories to the neocortex, and for the organization of coordinated cortical retrieval and consolidation (Hoffman and McNaughton 2002). These considerations motivated the investigation of the relationship between global cortical activity at slow time scales and hippocampal sharp waves, which is the subject of the present work.

Some of these results have been presented in an abstract format (Battaglia et al. 2001).

To establish the relationship between SPWs and neocortical activity, multiple single unit activity was recorded from an array of 144 independently positionable electrodes chronically implanted on three rats. The electrode array spanned about two-thirds of the dorsal surface of the neocortex, including anterior cingulate, primary and secondary motor, somatosensory, dorsal parietal, retrosplenial, and visual areas (Fig. 1A). At the same time, the left CA1 pyramidal layer EEG (Fig. 2) was recorded from one electrode.

Figure 1 Recording location and yield of isolated units from a particularly successful recording session. Average yield per session was approximately 50 neurons. (A) Outline of the cortical structures spanned by the electrode array, as implanted on rats 1 and (more ...)

Figure 2 Example of simultaneous recording of hippocampal LFPs and 84 neocortical single units. This is a 10-sec excerpt from a rest period showing: LFP trace from the CA1 pyramidal layer, bandpass (100-300 Hz) filtered, with SPW/ripple complexes (top); the global (more ...)

During recording sessions, rats 1 and 2 slept or rested for two periods (Rest 1 and Rest 2) of 25-40 min. Between these periods, they performed a behavioral task. Rat 3 was only recorded during one rest period of 25-40 min in each session and did not perform a behavioral task. Only the rest period data were analyzed for this report. Up to 96 cells were recorded simultaneously (Fig. 1B; total 2812 cells, in 61 sessions; 2298 from rat 1, 288 from rat 2, 226 from rat 3). Some of the cells may have been recorded from more than once across sessions, given that not all electrodes were moved every day (firing rate: mean: 1.94 ± 2.89 Hz; min: 0.005 Hz, max: 31.06 Hz).

Neocortical neuronal activity exhibited coherent fluctuations spanning all of the recorded areas. Spectral analysis of the total spike activity revealed discrete bouts of oscillations in the spindle range and in the delta/slow range during most of the recorded rest sessions (see, e.g., Fig. 3A). These episodes of oscillations accounted for 35% of the total recording time.

Figure 3 Global fluctuations of cortical activity and hippocampal sharp waves. (A) Power spectral density analysis of 700 sec of cortical activity. The power spectrum of combined instantaneous firing rate for the 84 recorded cells was estimated using a sliding (more ...)

During troughs of slow/delta oscillations, generalized neuronal silence was often observed for a duration of ~200-400 msec (Figs. 2 and 3B,C). The same degree of synchrony was observed across the sampled recorded areas during spindle oscillations (data not shown). Fluctuations on a longer time scale were also present, and were reflected in the autocorrelogram of the global population activity, which decayed with a time constant of 2.4 ± 0.5 sec (Fig. 3D). “Up-states” were defined arbitrarily as those 1-sec periods in which the number of “active” cells (one or more spikes) exceeded the 95% confidence threshold, computed from the randomly and independently shuffled interval series of all cells (see Materials and Methods). In all, 15.2% ± 4.3% of the intervals in Rest 1 and 13.0% ± 3.0% in Rest 2 exceeded this threshold, with activity in at least 58.3% ± 5.0% and 57.5% ± 5.1% of neurons, respectively. The recording session with the smallest effect yielded 7.6% of above-threshold intervals (p < 0.0001).

The sharp-wave-triggered Peri-event time histograms (PETH) of the global cortical activity showed that SPW events were accompanied by a brief burst of neocortical firing (Fig. 4). PETHs computed during periods of identified delta/slow oscillations were different from those computed during periods in which no delta/slow oscillations were present. During delta/slow oscillations (Fig. 4A,B), sharp wave events were preceded by a dip in average cortical firing lasting for ~500 msec. The average cortical firing was at the baseline level immediately before the dip. After the sharp wave event, the average cortical firing was increased, and slowly decayed to baseline with a time constant of the order of 3 sec. In epochs that did not include delta/slow oscillations, the sharp wave-triggered PETH of cortical firing exhibited different features (Fig. 4C,D): Although there was still a transient increase at the time of the sharp wave event (albeit not as strong as during delta/slow oscillations), no “fast” dip was observed preceding sharp waves. Instead, an exponential decline in average activity was observed preceding the sharp wave, with a time constant of the order of seconds. Baseline firing resumed shortly after sharp waves. Conversely, the generation of SPW events was affected by the neocortical state (Fig. 4E): the frequency of sharp wave occurrence was higher during “down-states” (1.12 ± 0.22 Hz; p < 10^-6) than during “up-states” (0.77 ± 0.20 Hz). The “down/up” transition-centered PETH of SPW events also revealed an increase in SPW frequency at the transition, which started 1-2 sec ahead of the transition.

Figure 4 SPW interaction with neocortical neural activity. (A) Peri-event time histogram (PETH) of cortical population activity during the periods of identified global oscillations in the delta/slow range, centered on hippocampal sharp wave events. A transient (more ...)

To estimate the proportion of individual cells exhibiting statistically significant modulation around SPWs, the firing rates before, during, and after each SPW event were compared with an estimate of the SPW-independent firing rates (Fig. 5; Table 1), for both delta/slow epochs and non-delta/slow epochs. For all intervals, there was an elevated number of significantly (two-tailed t-test, p < 0.05) up- or down-modulated (or both) cells. The proportion of up- and down-modulated cells was in most cases different from chance; the significance of such differences was tested by a binomial sign test, whose results are reported in Table 1. The delta/slow epochs showed similar numbers of up- and down-modulated cells before the sharp waves (consistent lack of deviation from baseline observed in the PETH prior to sharp waves), and larger numbers of up-modulated cells at the sharp wave times and in the subsequent interval (consistent with the long-lasting increase from baseline registered in the PETH). At the time of the sharp waves, there were significantly fewer down-modulated cells than chance. In non-delta/slow epochs, there was an elevated number of down-modulated cells in the pre-SPW intervals, a prevalence of up-modulated cells at sharp wave times, and similar numbers of up- and down-modulated cells in the interval following the sharp waves. The pattern was similar in each of the experimental animals (Table 1). To assess whether these effects were localized in only some of the recorded cortical areas, the 12 × 12 electrode array was divided into nine square regions (3 × 3). The number of SPW-modulated cells was above chance in all regions studied both in delta/slow epochs and non-delta/slow epochs, although there were regional differences (p < 0.02, χ² test). The effects seemed to be strongest in the cortical regions spanning the midline, but further study would be required to rule out variables other than region in the generation of this effect. The z-scores for sharp wave-modulation of individual cells in the two rest periods were only moderately correlated (r = 0.48 in the interval corresponding to SPWs, and lower in the other intervals), indicating that different ensembles were activated during different SPW events.

Figure 5 Single unit activity modulation by SPWs. (A) Percentage of cells with firing rates that were significantly (p < 0.05) up-modulated (white bars) or down-modulated (black bars) in intervals centered on, or surrounding, the SPWs, as they were detected (more ...)

Table 1. Detail of the numbers of up- and down-modulated cells during delta/slow and non-delta/slow epochs

The probability of SPW occurrence also fluctuated weakly on time scales comparable to those of neocortical slow oscillations, as shown by the autocorrelogram of SPW times, which decays with a time constant of 5.51 ± 0.06 sec (Fig. 3E).

The three main results of this study are the observation of deeply synchronized fluctuations in neuronal firing in the delta/slow frequency range across almost the entire neocortex, an observation that extends previously published results (Destexhe et al. 1999), of the existence of longer time-scale fluctuations in nonanesthetized animals, and, most importantly, of the relationship between these cortical phenomena and hippocampal sharp waves. A coupling between cortical delta rhythm and sharp waves was previously demonstrated by Sirota et al. (2003) in somatosensory cortex. It is shown here that longer time-scale coherent fluctuations in cortical activity, namely, the transitions between what have been referred to as “down-states” and “up-states” were correlated with an increased probability of occurrence of sharp waves in the hippocampus (Fig. 4E), and sharp waves were more likely to occur during down-states than during up-states. Because the coupling between cortical activity and hippocampal activity is of a probabilistic nature, there was no one-to-one correspondence between cortical transitions and sharp waves, which explains why the modulation observed in the PETHs is relatively small. The small size of the effect may have prevented Pelletier et al. (2004) from observing a relation between neocortical firing and sharp waves in the entorhinal cortex, which are tightly related to the hippocampal ones (Chrobak and Buzsaki 1996), because of their smaller sample size. The effect was nevertheless observed in a large fraction of the cells, thus it could not be attributed to a modulation of a small subset of cells (Fig. 5). These two facts taken together suggest that different, relatively small, groups of cells are recruited by each sharp wave episode. The small size of this group may reflect the sparse coding of information in neocortical modules, and may be partly caused by the high inhibitory pressure in the entorhinal cortex (Pelletier et al. 2004). The correlation between hippocampal and neocortical dynamics was also evident when the change in sharp wave frequency associated with cortical activity states was considered: sharp wave events were 20%-30% more likely in correspondence with down-to-up transitions compared with baseline (Fig. 4E). Sharp waves were also more likely during down-states than during up-states.

The intracellular physiology literature has introduced the concept of “up” and “down” states as membrane potential fluctuations between two well-defined levels. In fact, in some regions or experimental conditions, the membrane potential fluctuations are bimodal in nature (e.g., as observed in the striatum by Wilson and Groves 1981). Bimodal fluctuations of membrane potential have been observed in intracellular recordings in anesthetized and naturally sleeping animals (Timofeev et al. 2001; Petersen et al. 2003) in the delta range, with frequencies corresponding to those of the population firing rate oscillations shown here. The picture emerging from population firing rates is, however, somewhat different: The population firing rate during delta/slow oscillations did not show bimodality, as the oscillations are characterized by an abrupt upward transition and a slower decline (see, e.g., Fig. 2). This might be owing to a gradual loss in coherence across neocortical cells (perhaps because of different durations of up-states in different neocortical networks) or to firing rate adaptation, which may slow down cell firing as the cortex remains in an up-state. Nevertheless, the data are consistent with the conclusion that sharp waves were correlated with discrete transitions from relatively low to relatively high states of activity. The precise shape of the sharp-wave-triggered PETH depends on the current dynamical state of the neocortex: During delta/slow oscillations, the PETH stays at baseline until few hundreds of milliseconds from the sharp wave, whereas, following the sharp waves, it decays back to baseline in a few seconds, starting from a more elevated value. In periods without delta/slow oscillations, the PETH is basically flat after the sharp wave, but exhibits a slowly increasing depression leading to up the sharp wave (2-3 sec).

One possible scenario that would produce this pattern is depicted in Figure 6. In this oversimplified scheme, transitions between up-states and down-states are described by a Markov process, so that the state durations follow an exponential distribution, and at least a fraction of the sharp waves coincide with the down-to-up transitions, the others being distributed randomly in time. If the up-state has a much shorter mean duration than the down-state, the sharp wave-triggered PETH of cortical activity has a shape similar to that observed during delta oscillations. If down-states are shorter than up-states, the PETH resembles that observed in no-delta/slow periods. Thus, whereas the abrupt increase in firing at the time of the sharp wave event is observed in both cases (in the model, because of the assumption of coincidence of sharp waves and down/up transitions), the difference in the shape of the PETH in these two dynamical regimes appears to reflect a different relative duration of up- and down-states, as might be expected in deeper or more superficial sleep stages, possibly because of cholinergic modulation of leakage K⁺ conductances (Compte et al. 2003).

Figure 6 Qualitative picture of the interaction between up-state and down-state transitions in cortical firing and hippocampal sharp waves and how the apparently gradual trends in the average PETH functions shown in Figure 4 can be accounted for by averaging over (more ...)

At present, there is no clear evidence for a causal relationship between these phenomena, even though a modulation in the neocortical EEG has been shown to precede sharp wave events by 50-100 msec (Sirota et al. 2003). Subtle changes in neocortical input, related to the slow fluctuations, may affect the probability of triggering an SPW in the hippocampus. More specifically, a decrease in cortical input might lead to hyperpolarization and low spike rates of CA3 cells, and to a de-inactivation of Ca⁺⁺ channels (Buzsaki et al. 1992), leading to higher excitability of the hippocampal neurons. A sharp wave event may then be triggered by some fluctuation in the hippocampal input (e.g., by a neocortical transition to an up-state). On the other hand, it has been observed (Lewis and O'Donnell 2000) that stimulation of hippocampus or the ventral tegmental area can trigger up-states. Thus, the appropriate framework to describe the hippocampal/neocortical interactions might well be that of a closed loop, with causal relationships actually going both ways. During down-states, the cortex may have a role in mediating the increase in excitability in the hippocampus, which may lead to sharp waves, which could explain the increased sharp wave frequency during down-states. The burst discharge associated to a sharp wave in the CA3 network may converge on a stored memory state, making that memorized information available to the neocortex (Shen and McNaughton 1996). Both the neocortex, especially at the end of a down-state (Compte et al. 2003), and the hippocampus are in an extremely excitable state, prone to explode in coherent, self-sustained activations. Therefore, they can readily follow the input coming from the other structure: The hippocampal drive during a sharp wave could increase the probability of a transition from a down-state to an up-state. Conversely, as mentioned above, the neocortical down/up transition may provide the triggering input causing the sharp wave. Common inputs from other brain structures may influence the generation of these related events in the hippocampus and neocortex.

The set of network relationships highlighted in the present study might be relevant to the memory consolidation process: Sharp waves, reflecting the stored patterns of hippocampal activity (Kudrimoti et al. 1999), could elicit coordinated retrieval of corresponding activity configurations stored in the neocortex that are indexed by the hippocampal cue (for review, see McNaughton et al. 2003). The neocortical input may, in turn, influence the memory content retrieved by the hippocampus at that time.

We thank K. Bohne, K. Poneta, L. Gronenberg, J. Lammers, S. Cowen, M. Tatsuno, and K. Hoffman for help with running the experiments; K. Stengel (Neuralynx, Inc.) for technical assistance; and A. Fuglevand, K. Gothard, and K. Hoffman for suggestions on the manuscript. This work is supported by NIH, MH46823, and MH01565. F.P.B. was partially funded by HFSP LT150/99B. G.R.S. was partially funded by NIH NIGMS 5 T32 GM08400.