The mammalian hippocampus displays a peculiar pattern of fast (≈200 Hz) network oscillations superimposed on slower sharp waves. Such sharp wave–ripple complexes (SPW–R) have been implicated in memory consolidation. We have recently described a novel and unique method for studying SPW–R in naive slices of murine hippocampus. Here, we used this model to analyse network and cellular mechanisms of this type of network activity. SPW–R are usually generated within area CA3 but can also originate within the isolated CA1 region. Cellular synchronisation during SPW–R requires both excitatory and inhibitory synaptic transmission as well as electrical coupling, the latter being particularly important for the high-frequency component. Extracellular and intracellular recordings revealed a surprisingly strong inhibition of most CA1 pyramidal cells during SPW–R. A minority of active cells, however, increases action potential frequency and fires in strict synchrony with the field ripples. This strong separation between members and non-members of the network may serve to ensure a high signal-to-noise ratio in information processing during sharp wave–ripple complexes.
During ripples, a small fraction of neurons (≈10 % of simultaneously recorded hippocampal pyramidal cells; Ylinen et al. 1995) fires action potentials in tight synchrony with the oscillating local field potential (Buzsáki et al. 1992; Csicsvari et al. 1999). This highly selective and co-ordinated behaviour requires a specific and rapid interaction between participating cells in order to secure precise phase-coupling in the range of a millisecond. We have recently suggested that gap junctions are crucial for neuronal synchronisation during ≈200 Hz ripples. This notion was based on recordings of spontaneous network oscillations and electrical coupling potentials in rat hippocampal slices in vitro (Draguhn et al. 1998). Theoretical modelling suggested that gap junctions are located between the axons of hippocampal projection cells (Draguhn et al. 1998; Traub et al. 1999; Schmitz et al. 2001). Recent experimental and modelling work revealed that gap junctions are also critical for certain forms of gamma oscillations (Tamás et al. 2000; Traub et al. 2000; Hormuzdi et al. 2001; Deans et al. 2001) and that inhibitory synaptic potentials and electrical coupling can act together in the generation of fast rhythms (Traub & Bibbig, 2000; Tamás et al. 2000).
All experiments were performed on adult (4–12 weeks) C57-bl/6 mice. Animal procedures were approved by the Berlin state government (T 0386/98) and were in accordance with the guidelines of the National Institutes of Health. Mice were briefly anaesthetised with ether, decapitated and the brain removed. Brains were constantly kept under cooled (≈1–4 °C) artificial cerebrospinal fluid (ACSF, containing (mm): NaCl 129, KCl 3, MgSO4 1.8, CaCl2 1.6, glucose 10, NaH2PO4 1.25, NaHCO3 21, gassed with 95 % O2:5 % CO2, pH 7.4). After removal of the cerebellum, hemispheres were separated, glued to a vibratome chamber (Campden Instruments, Sileby, UK) and horizontal slices of 450 μm were cut. Recordings were performed at 35 ± 0.5 °C in a Haas-type interface chamber. Stable SPW–R activity could regularly be recorded from the slices after 1–2 h of equilibration. Extracellular electrodes had large tip diameters of 8–12 μm and were filled with ACSF before use.
Electrical stimulation of the Schaffer collateral was performed with bipolar platinum wires (diameter 50 μm) located in stratum radiatum. Intracellular recordings were performed with a bridge-balance amplifier (npi electronics, Tamm, Germany). Electrode (o.d. 1.2 mm) resistance was 40–90 MΩ. After impaling a cell, we injected negative current for several minutes until the membrane potential had stabilised and current injection could be gradually reduced to zero. Bridge balance was repeatedly adjusted during the experiment by optimising the voltage response to small negative current injections (100–200 pA). Intrinsic properties were assessed by negative and positive current injections of 1 s duration. Offset potentials were determined at the end of the experiment and were subtracted from the recorded values.
Octanol was applied directly to the ACSF; all other drugs were added from 1000-fold stock solutions in water or DMSO (6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) and CGP 55845A). Bicuculline and carbenoxolone were purchased from Sigma (Deisenhofen, Germany). (±)-2-Amino-5-phosphonopentanoic acid (±-APV) and all other drugs were from RBI/Tocris (Köln, Germany).
Data processing and analysis
Original data were filtered at 3 kHz (except for quadruple extracellular recordings, which were filtered at 1 kHz), sampled at 5–10 kHz with a CED Micro1401 interface (CED, Cambridge, UK) and analysed off-line using the Signal Averager and Spike2 software (CED). Slow and fast components were extracted from the original field potential recordings by filtering at different corner frequencies (see Fig. 2). For detection of sharp waves in extracellular recordings, raw traces were low-pass filtered at 50 Hz and events were detected by setting a threshold at 4–5 times the standard deviation of event-free baseline data. In some experiments, sharp wave detection levels were adjusted arbitrarily and reliable detection was controlled by eye. For detection of ripple oscillations, raw data were filtered at 150–300 Hz (band-pass) and the threshold for detection was set at four times the standard deviation of event-free baseline noise. The detected events were then further sorted by the following criteria: ripples had to consist of at least three consecutive negative spikes below threshold with no less than 2 ms and no more than 11 ms interspike intervals (90 Hz < frequency of ripples < 500 Hz). With these routines, ripples and sharp waves were discovered with high fidelity when comparing the results with original data by eye-inspection. Lowering detection thresholds resulted in an increasing contamination with noise or questionable events. Unit discharges (extracellularly recorded single action potentials) were detected by high-pass filtering at frequencies above 500 Hz (see Fig. 2). Parameters for further analysis included the number of sharp waves or ripples per second, cross-correlation of data from parallel recordings in different areas and Fourier analysis of raw data. In order to analyse drug-induced changes or to compare properties of ripples in different hippocampal subfields, we constructed interevent-interval histograms by collecting the intervals between neighbouring spikes within a ripple episode over a sufficiently long stretch of data (10 min). The dominant frequency for a given recording was determined as the reciprocal of the value of the mean interspike interval. The temporal correlation between sharp waves, ripples and unit discharges was analysed by event-cross-correlation. Based on threshold detection algorithms, event markers were set for the positive peak of each sharp wave (from low-pass filtered signals), the negative peak of each ripple wave (from band-pass filtered signals) and for unit discharges (from high-pass filtered signals). We then constructed auto- and cross-correlograms from these events at different temporal resolutions (see Fig. 11 and Fig. 13).
Quantitative results are given as means ±s.d. Groups of data were compared using the non-parametric Wilcoxon test for paired and the Mann-Whitney U test for unpaired data. A value of P < 0.05 was regarded as significant.
Extracellular field potential recordings from mouse hippocampal slices revealed small spontaneous field potential fluctuations in most (> 90 %) specimens. After optimising the electrode position within the pyramidal cell layers of CA1 or CA3, the events were observed as positive or biphasic waves of ≈0.05–0.5 mV amplitude and ≈30–80 ms duration (Fig. 1).
Structure of spontaneous SPW–R
The power spectrum of the extracellular field potential (Fig. 1B) was dominated by a low-frequency component at ≈25 Hz, which is likely to reflect the repetitive spontaneous waves shown in Fig. 1A. Upon closer inspection we found a smaller, but distinct, second peak at frequencies between 200 and 250 Hz (Fig. 1B, inset) which corresponds well with the repetitive small negative voltage fluctuations on top of the slower potentials (Fig. 1A). In order to isolate the different frequency components of the signal we applied digital filters to the original data (Fig. 2), similar to Buzsáki et al. (1992). Band-pass (150–300 Hz) filtered traces isolated spindle-shaped fast oscillations reminiscent of ‘ripples’ as described in vivo by O'Keefe (1976), Suzuki & Smith (1987) and Buzsáki et al. (1992). High-pass filtering (> 500 Hz) isolated unit discharges which were frequently observed during SPW–R but were clearly less frequent than the superimposed ripples, i.e. they did not accompany each cycle of the field oscillation in a one-to-one manner. This finding indicates that the fast field potential oscillations are generated by multiple cells and thus reflect synchronised network activity. Finally, the low-pass filtered waveform resembled sharp waves in vivo as described by Buzsáki (1986). From here on, we therefore refer to the observed signals as in vitro sharp wave–ripple complexes (SPW–R).
Quantitative parameters of spontaneous hippocampal SPW–R in vitro were derived from an analysis of parallel recordings from CA3 and CA1 in 12 slices from 11 animals (Fig. 3). Sharp waves occurred at similar mean frequencies of 2.7 ± 1.5 Hz in CA3 and 2.7 ± 1.1 Hz in CA1 (n= 12; Fig. 3A). Likewise, ripples were equally often detected in CA3 and CA1, respectively (Fig. 3B; CA3: 1.9 ± 1.5 Hz; CA1: 1.9 ± 1.0 Hz). These numbers are lower than the corresponding values for sharp waves, probably due to the different signal-to-noise ratio in the differently filtered traces (see Methods and Fig. 2). The occurrence of sharp waves was rhythmically modulated as visible from event-autocorrelation functions (Fig. 3C). Clear peaks were detected in 10 of 12 slices at intervals corresponding to 2.9 ± 0.3 Hz, in accordance with the mean frequency of sharp waves. The internal structure of the high-frequency oscillations was quantified by detecting the negative peaks within ripples in band-pass filtered traces. Analysis of the inter-peak intervals revealed a mean frequency of 193 ± 14 Hz in area CA3, which was below the frequency measured in CA1 (210 ± 16 Hz; P < 0.05; Fig. 3D; see also Fig. 3F for example histograms of interspike-intervals). The mean number of detected oscillation cycles per ripple was 4.1 ± 0.6 (CA3) and 4.0 ± 0.6 (CA1; see Fig. 3E; not significantly different, P > 0.9). Thus, SPW–R occur at similar frequencies along the hippocampal pyramidal cell layer, and have largely similar waveforms except for a slightly higher intraburst ripple-frequency in CA1.
Propagation and laminar profile of SPW–R
Parallel recordings from CA3 and CA1 revealed a close temporal coherence of the signals at both recording sites (see Fig. 4A). SPW–R in CA3 were consistently leading before CA1 (38 slices). In order to gain more insight into the spatial propagation of SPW–R we performed double, triple and quadruple extracellular recordings from different regions, including the dentate granule cell layer, the hilus and the subiculum (12 slices in varying combinations). From inspection of the raw traces (Fig. 4) and from cross-correlation functions (Fig. 5C) it became clear that sharp waves propagate from area CA3 towards CA1 and then to the subiculum, analogous to previous in vivo observations by Chrobak & Buzsáki (1996). We also observed coherent field potential fluctuations between CA3 and the dentate gyrus (DG, see Fig. 4A), but the temporal relationship was more variable. Within CA1, the velocity of propagation was estimated from paired or multiple recordings yielding values between ≈2 and 7 cm s−1 (four slices, see Fig. 5). In three paired recordings within CA1 we directly measured the time lag between the occurrence of SPW–R at both sites, and compared it to the time lag between the occurrence of evoked population spikes (stimulation of Schaffer collateral). Synaptically triggered population spikes had much shorter latencies between both sites than SPW–R, indicating that more complex, time-consuming mechanisms are involved in the propagation of the network activity (Fig. 5D; average difference in velocity of propagation ≈2.5-fold).
In order to reveal whether the signal is strictly generated within CA3, we performed cutting experiments while recording in parallel from CA3 and CA1. Disconnecting the Schaffer collateral resulted in a total (n= 2 slices) or strong (≈80 %, n= 4 slices) suppression of SPW–R in CA1, whereas the activity was much less (≈30 %) reduced in CA3 (Fig. 6B). The remaining SPW–R in CA1 were then generated independently from the activity in CA3, i.e. coherence between both regions was lost. Recordings from completely isolated CA1 minislices (n= 3) revealed that SPW–R can be generated, at a low frequency of ≈0.5 Hz, within CA1 (Fig. 6C). In intact slices, however, SPW–R regularly originate in CA3 and then propagate to the other fields.
Ylinen et al. (1995) have shown that ripples are largely confined to the pyramidal cell layers and are much less prominent in dendritic regions. We recorded laminar profiles with 2–4 electrodes positioned in the dendritic and somatic layers of CA1 and revealed a similar behaviour in mouse hippocampal slices: ripple amplitude was maximal in the pyramidal layer, and showed a sharp decrease towards the stratum radiatum and stratum oriens (Fig. 7B; n= 8). The underlying sharp waves were distributed more broadly between different layers and appeared maximal in the stratum radiatum and in stratum pyramidale. Sharp waves and ripples both reversed phase between the principal cell layer and the dendritic region, similar to previous findings from in vivo recordings (Buzsáki et al. 1983; Ylinen et al. 1995). These data are quite compatible with the idea that sharp waves reflect compound excitatory postsynaptic field potentials in the stratum radiatum of CA1 and that ripples are generated in the pyramidal cell layer (Buzsáki et al. 1992). The cellular mechanisms underlying this complex network activity can, however, not unambiguously be derived from such laminar field potential recordings, and we therefore proceeded to the use of pharmacological tools.
Synaptic mechanisms in SPW–R
The shape of spontaneous sharp waves in CA1 is reminiscent of field excitatory postsynaptic potentials (EPSPs). Moreover, field EPSPs generated by weak electrical stimulation of the Schaffer collateral pathway were frequently superimposed by small negative deflections at ≈200 Hz, which are reminiscent of ripples (see Fig. 8). We thus wondered whether glutamatergic synaptic transmission is involved in the generation or propagation of SPW–R, and applied the non-NMDA glutamate receptor antagonist CNQX (20–30 μm). As expected, stimulus-induced field EPSPs were eliminated by the drug. At the same time, CNQX abolished the spontaneous SPW–R activity in both areas, CA1 and CA3 (n= 5, reversible in three experiments after washout for at least 20 min; see Fig. 8). SPW–R can be generated locally within CA3 and CA1, as shown in our cutting experiments. The block of SPW–R in both subfields can therefore not be solely attributed to a block of excitatory synaptic transmission via the Schaffer collateral. It is not clear from this experiment, though, whether CNQX selectively blocked the generation of sharp waves (which consequently could not elicit ripples) or whether both patterns of network activity were independently suppressed. In contrast to CNQX, the NMDA-receptor antagonist ±-APV (30 μm) neither altered the frequency of occurrence nor the inner structure of SPW–R, indicating that NMDA-receptors are not crucially involved in the generation of this pattern (n= 4 slices).
It has been suggested that network synchronisation during ripples depends on fast inhibitory postsynaptic potentials (IPSPs), which generate consecutive phases of high and low discharge probability in their target cells (Ylinen et al. 1995; Buzsáki, 1997). We therefore blocked IPSPs by administrating the GABAA receptor antagonists bicuculline (20 μm; n= 3) or gabazine (3 μm; n= 26). In all slices analysed, these drugs caused a rapid transition of SPW–R into larger epileptiform bursts (Fig. 9). Similar to SPW–R, the pathological discharges carried a high-frequency oscillation at around 200 Hz on top of an underlying excitatory wave (Fig. 9B). Thus, regular SPW–R do not form in the absence of GABAA receptor-mediated synaptic inhibition but the generation of fast (≈200 Hz) rhythms is not impeded.
Previously, we have reported that gap junctions play a critical role in the synchronisation of ≈200 Hz network oscillations (Draguhn et al. 1998; Traub et al. 1999). We therefore tested the effects of two different uncoupling agents (carbenoxolone and octanol) on SPW–R. At 200 μm, carbenoxolone strongly reduced the number of SPW–R in CA1 to a remaining frequency of 10 ± 8 % of control (n= 4; Fig. 10A and C). A similar and parallel reduction was seen in CA3 using a second recording electrode (n= 3). The effect of carbenoxolone developed slowly over several minutes, and was not reversible within washout periods of > 1 h. Likewise, octanol (1 mm) suppressed the frequency of SPW–R in CA1 to 4 ± 6 % of control (n= 5; Fig. 10B and D). Neither carbenoxolone nor octanol had any systematic effect on the ‘inner’ frequency of the remaining ripple oscillations. The effect of octanol reversed after washout of the substance. The remaining sharp waves in the presence of octanol had larger amplitudes and, to our surprise, were mostly not superimposed by high-frequency oscillations (Fig. 10B2). We therefore analysed whether ripples were more potently suppressed by uncoupling agents than sharp waves. Indeed, the percentage of sharp waves with associated ripple oscillations was decreased by both drugs. Carbenoxolone reduced the fraction of SPWs with detectable superimposed ripples from 32 to 18 % (n= 4). Likewise, octanol reduced this fraction from 75 to 34 % (n= 5; Fig. 10C and D). Thus, gap junction blockers exert differential effects on sharp waves and ripples, which points towards a selective and differential role of electrical synapses in the generation of both phenomena.
Cellular behaviour during SPW–R
Network synchrony is achieved by entraining the membrane potentials of multiple neurons into a common rhythm, which defines alternating phases of high vs. low discharge probability. It is therefore important to analyse the firing pattern of individual neurons during SPW–R.
For this purpose, we performed extracellular recordings from the CA1 and CA3 pyramidal cell layers, dissected the different frequency components (sharp waves, ripples and unit discharges) and encoded them as events, using threshold detection algorithms (Fig. 11A; see Methods). From these data it became apparent that unit discharges were more frequent during SPW–R than during non-SPW–R episodes. Again, units and ripples did not match in a one-to-one fashion, excluding the possibility that the rapidly oscillating field potentials reflect the activity of a single cell (Fig. 11A, bottom). Indeed, units of similar size were usually found only once or twice within a single SPW–R, as is typical for pyramidal cells (Csicsvari et al. 1999). Cross-correlation diagrams from units versus sharp waves revealed a prominent increase of unit activity during SPW–R (Fig. 11B, left). The peak number of events in the cross-correlation histograms was 22 ± 12-fold above baseline, indicating a strong increase in action potential frequency during SPW–R (n= 7 experiments from CA1 and n= 7 from CA3; results similar for both regions). Inspection of the traces indicated that units were mostly located within the negative-going phase of the field ripple. This phase-coupling between single cells and the underlying network activity was confirmed by cross-correlations between unit discharges and field ripples (Fig. 11B, right). Indeed, histograms from all 14 recordings were clearly modulated at ≈5 ms intervals indicating that action potentials occur at fixed phases of the ripple oscillation.
Finally, we performed intracellular recordings from putative CA1 pyramidal cells. These cells (n= 20) showed a regular spike frequency accommodation, an input resistance of 24–88 MΩ and resting membrane potentials (RMP) of −63 ± 7.9 mV. When recording the local field potential with a closely positioned extracellular electrode, all recorded cells exhibited potential deflections accompanying the field SPW–R. More than half of the potentials were depolarising, whereas the others were either hyperpolarising or biphasic when recorded at RMP (Fig. 12A and B1). Only 50 % of the recorded cells had a ‘typical’ or predominating waveform pattern (seven cells with depolarising potentials, two with hyperpolarising and one cell with biphasic responses; Fig. 12B2). The other half of the cells showed a mixture of different waveforms, mostly biphasic potentials together with either depolarising or hyperpolarising potentials (Fig. 12B2, ‘mixed’). We rarely found both hyperpolarising and depolarising potentials together in the same cell. Although depolarising potentials were the most frequently observed SPW–R-associated behaviour and were even the dominant waveform pattern in 7/20 cells, most cells (19/20) did not reach the threshold for action potential generation. Indeed, the depolarising potentials contained a major inhibitory component. This became apparent when we depolarised cells close to firing threshold so that action potentials were elicited at high frequency: in this situation action potential firing ceased during SPW–R in 15/20 cells (Fig. 13B1). Such effective inhibition during SPW–R was also present in cells which, at RMP, showed depolarising potentials.
In four cells, we analysed the voltage dependence of the cellular correlates of SPW–R. The potentials reversed between −61 and −73 mV (Fig. 13A1–3), consistent with a major contribution of GABAA receptor-mediated inhibitory postsynaptic potentials. Administration of the GABAB receptor antagonist CGP 55845 A (2 μm) did not alter the voltage dependence, excluding a major role of GABAB receptors in SPW–R (Fig. 13A3; n= 3 cells with similar results). Interestingly, at higher temporal resolution we found small phasic inflections in the intracellular potentials that were synchronous with the field ripples (Fig. 13A1) and that may correspond to inhibitory postsynaptic potentials. This was also reflected in phasically modulated cross-correlograms of cellular potentials and ripples, as shown in Fig. 13A4. In four of the 20 cells, depolarisation by current injection elicited action potential firing that persisted during SPW–R. These action potentials fell into a sharply defined phase of concomitant SPW–R (Fig. 13B2).
Taken together, these cellular data reveal a pronounced inhibition of most CA1 pyramidal cells during SPW–R and strong phase-coupling of action potentials between SPW–R and the active neurons.
We examined the properties and cellular mechanisms of hippocampal sharp wave–ripple complexes in vitro. Our results show that this waveform pattern depends on excitatory and inhibitory synaptic transmission as well as on electrical coupling. The high-frequency component (≈200 Hz ripples) is more sensitive to gap junction blockers than the underlying sharp waves, indicating that ripple oscillations per se are dependent on functional gap junctions. At the cellular level SPW–R mediate a surprisingly strong inhibition of most CA1 pyramidal neurons, and the minority of active cells fires in strict synchrony with the ripple cycles. This mechanism is suited to enhance the signal-to-noise ratio between members and non-members of the local SPW–R network.
In this study we made use of our recent observation of spontaneously occurring SPW–R in naive mouse hippocampal slices (Maier et al. 2002). These local field potentials closely resemble the corresponding SPW–R pattern in the rodent hippocampus in vivo (O'Keefe, 1976; O'Keefe & Nadel, 1978; Buzsáki et al. 1983, 1992; Buzsáki, 1986; Suzuki & Smith, 1988; Ylinen et al. 1995; Chrobak & Buzsáki, 1996). We have previously reported spontaneous ≈200 Hz network oscillations in hippocampal slices from rats (Draguhn et al. 1998). This activity, however, did not display underlying sharp waves. The reasons for this discrepancy are currently unknown but it is feasible that mouse slices contain more intact circuitry (Insausti, 1993) or that cellular excitability in this preparation is higher. Under certain experimental conditions, sharp waves have been observed in slices from different species, including rats (Schneiderman, 1986; Schwartzkroin & Haglund, 1986; Köhling et al. 1998; Papatheodoropoulos & Kostopoulos, 2002a,b; Wu et al. 2002; Kubota et al. 2003) but have not yet been described to be associated with ripple-oscillations. In the present study, we observed entire SPW–R similar to those recorded in living rodents. Such waveforms occur spontaneously and reliably after 1–2 h of rest following tissue preparation, indicating that network functions have to be restored before this complex pattern of activity can occur. While the precise mechanisms of ‘recovery’ are unknown, our observation is similar to the preconditions for certain forms of long-term potentiation in vitro (see for example Frey et al. 1995). Another apparent difference between SPW–R in mouse hippocampal slices and ≈200 Hz field potential oscillations in slices from rats is the sensitivity of SPW–R towards blockers of chemical synaptic transmission, whereas the oscillations in rat tissue occur even in the absence of chemical synaptic transmission (Draguhn et al. 1998). It is feasible that glutamatergic and GABAergic synapses are required to bring about sharp waves which then, in turn, trigger ripples. The fast oscillations themselves may still be independent from chemical synaptic transmission. This hypothesis is supported by their persistence in the presence of GABAA receptor antagonists, as well by their high sensitivity towards gap junction blockers (see below).
Synaptic excitation and inhibition during SPW–R
The AMPA/kainate receptor antagonist CNQX (but not the NMDA receptor antagonist ±-APV) completely abolished spontaneous activity, indicating that glutamatergic transmission is involved in the generation of ≈200 Hz network oscillations or of sharp waves. Small stimulus-induced field EPSPs in CA1 show some similarity to SPW–R: they are superimposed by ripple-like voltage deflections (see Fig. 8) and both EPSPs and SPW–R travel along the trisynaptic hippocampal output loop (Chrobak & Buzsáki, 1996; Fig. 4 and Fig. 5). However, SPW–R are not identical to field EPSPs as indicated by their slower propagation velocity (which is comparable to certain forms of epileptiform activity; cf. Wong & Prince, 1990). In contrast to orthodromically propagating EPSPs, SPW–R also propagate ‘backward’ into the dentate gyrus (our observations and those of Buzsáki, 1986), similar to several other physiological and pathological patterns of network activity (Müller & Misgeld, 1991; Bragin et al. 1995b; Penttonen et al. 1997). It should also be noted that ripples do not strictly depend on excitation by sharp waves but can occur without underlying slower field potentials both in vivo (Ylinen et al. 1995) and in vitro (Draguhn et al. 1998).
How are sharp waves or SPW–R generated? We found that they usually emerge within CA3, an area containing multiple mutual excitatory connections between pyramidal cells. These recurrent connections can trigger synchronised activation (Miles & Wong, 1986) and might generate compound field EPSPs underlying sharp waves. Similar connections exist, at lower density, between CA1 pyramidal cells (Deuchars & Thomson, 1996) which might explain the generation of SPW–R at lower frequency in CA1 minislices. Alternatively, sharp waves may result from synchronised GABA release from electrically coupled interneurons (Traub, 1995; Avoli, 1996; Traub et al. 2001) for which there is good experimental evidence (Fukuda & Kosaka, 2000; Venance et al. 2000). Consistent with this hypothesis, sharp waves are blocked by GABAA receptor antagonists (see also Papatheodoropoulos & Kostopoulos, 2002a,b) as well as by gap junction blockers (see below) and they occur less frequently in tissue from connexin 36 (Cx36)-deficient mice (Maier et al. 2002). Recent in vivo recordings from another strain of Cx36-deficient mice did not, however, reveal any difference in SPW–R frequency (Buhl et al. 2003). The reasons for these different findings have not yet been revealed.
Fast (≈200 Hz) network oscillations can occur in the absence of phasic synaptic inhibition. This became evident from our experiments with blockers of GABAA receptors, which converted SPW–R into larger interictal-like discharges (Dingledine & Gjerstad, 1980; Gutnick et al. 1982) that were clearly superimposed by ≈200 Hz network oscillations. Thus, in contrast to certain forms of gamma- (Whittington et al. 1995; Wang & Buzsáki, 1996; Whittington et al. 2000) and theta- (Cobb et al. 1995) oscillations, IPSPs are not crucial for neuronal synchronisation at very high frequencies (see also Jones & Barth (2002) for fast and ‘very fast’ (> 400 Hz) oscillations in the somatosensory cortex in vivo). The absence of regular SPW–R in the presence of GABAA receptor blockers may either be attributed to a selective block of the underlying sharp wave or may be a consequence of the reduced excitability following each epileptiform discharge. Under normal conditions, SPW–R do generate strong synaptic inhibition (Fig. 13; Ylinen et al. 1995; Grenier et al. 2001), which may act to sharpen synchrony (Traub & Bibbig, 2000) or to suppress action potential generation in pyramidal cells outside the SPW–R network. The latter function may ensure a high signal-to-noise ratio, i.e. exclusive activation of a well-defined memory-encoding cell ensemble during SPW–R (Wilson & McNaughton, 1994; Kudrimoti et al. 1999; Nádasdy et al. 1999).
SPW–R and electrical coupling
Spontaneous SPW–R were sensitive to gap junction blockers, confirming our previous proposal that gap junctions co-ordinate neuronal activity during ripples (Draguhn et al. 1998; Traub et al. 1999) and consistent with their suppression by the uncoupling anaesthetic halothane in vivo (Ylinen et al. 1995; Grenier et al. 2001). At present, absolutely specific gap junction blockers are not available and therefore such results should be interpreted with caution. However, the substances used in our present experiments have only moderate (octanol) or virtually no (carbenoxolone) side-effects on cellular excitability in CA1 (Draguhn et al. 1998; Schmitz et al. 2001). Interestingly, both carbenoxolone and octanol suppressed high-frequency ripples more readily than the underlying sharp waves, similar to the effect of halothane observed by Ylinen et al. (1995). Thus, uncoupling agents so far provide the only pharmacological tool that distinguishes between ripples and the underlying sharp waves, indicating that the mechanisms of synchronisation differ between these phenomena. The subtype and precise localisation of gap junctions synchronising SPW–R are currently unknown. We have recently reported that the waveform of SPW–R is unchanged in mice devoid of Cx36 but SPW–R occur at lower frequency (Maier et al. 2002; but see Buhl et al. 2003, for contrasting in vivo data). This gap junction subunit is primarily expressed in hippocampal interneurons (Venance et al. 2000; Meier et al. 2002) and may contribute to the generation of SPW–R rather than to their precise synchronisation. Our further hypothesis that gap junctions are located between axons of pyramidal cells (Draguhn et al. 1998; Traub et al. 1999; Schmitz et al. 2001) is still awaiting direct morphological proof and has not been addressed in this study.
Cellular behaviour during SPW–R
Unit and intracellular recordings revealed three principal results: (1) action potential frequency in the pyramidal cell layer is massively (≈20-fold) enhanced during SPW–R; (2) nevertheless, most individual pyramidal cells (15/20) undergo strong inhibition during SPW–R; and (3) firing of neurons in the pyramidal cell layer is strictly phase-coupled to the negative phase of individual ripple cycles.
Our data are consistent with previous extra- (Buzsáki et al. 1992; Csicsvari et al. 1998) and intracellular (Ylinen et al. 1995; Grenier et al. 2001) recordings in vivo but show a surprisingly strong inhibition of non-participating pyramidal cells. Even those cells which responded with depolarising potentials to SPW–R at resting membrane potential could be massively inhibited at the same time. This became apparent when we induced action potential firing by depolarising such cells close to threshold: in this situation, still, most of them were silenced by field SPW–R (Fig. 13B1). Thus, the cellular correlates of SPW–R may reflect overlapping inhibitory and excitatory inputs which, around threshold, result in net inhibition for most cells. This pronounced inhibition may be caused by high-frequency discharges of inhibitory interneurons which can follow ripples almost in a 1:1 fashion (Ylinen et al. 1995; Csicsvari et al. 1998), consistent with the phasic inflections in pyramidal cell potentials (Fig. 13A1).
Units did not follow every ripple wave, showing that the field oscillation represents genuine multicellular network activity. Although the nature of the discharging cells was not determined, their behaviour is reminiscent of pyramidal cells in vivo (Csicsvari et al. 1999). Discharge frequency increased sharply during the events (Fig. 11B; compare with Csicsvari et al. 1998) and action potentials were precisely phase-coupled to the negative phases of ripples (Fig. 11C; compare Buzsáki et al. 1992; Ylinen et al. 1995; Csicsvari et al. 1998). Similarly, the 5/20 putative pyramidal cells which did discharge at resting membrane potential (n= 1) or after active depolarisation (n= 4) showed a constant phase relation between action potentials and underlying SPW–R (Fig. 13B2). It should be noted that Ylinen et al. (1995), as well as Grenier et al. (2001), have been able to change this phase relation in in vivo recordings from rats by filling cells with chloride, suggesting that the phase relation is determined by GABAA receptor-mediated potentials.
In summary, our data reveal that SPW–R in mouse hippocampal slices are characterised by an intricate interplay of excitatory, inhibitory and electrical transmission. They are accompanied by pronounced synaptic inhibition of most pyramidal cells but enhance discharge probability of the few participating cells which are sharply phase-coupled to the network. The selection of active versus silent cells may depend on previous experience, and could therefore provide the key to the function of SPW–R in information processing. The future challenge will thus be to define the mechanisms by which cells are determined to participate in SPW–R or to stay silent.
This work was supported by a grant from the Deutsche Forschungsgemeinschaft (DFG Dr 326/1–3). We thank Dr Roger D. Traub (New York) and Dr Uwe Heinemann (Berlin) for helpful discussion, Dr Herbert Siegmund and Dr Hans-Jürgen Gabriel for their help in data analysis and Jan Börgermann and Petra Rook for experimental support. A preliminary account of these data has been presented at the Proceedings of The Physiological Society, Leeds, UK, 2002.