Spontaneous synchronous bursting of the CA3 hippocampus in vitro is a widely studied model of physiological and pathological network synchronization. The role of inhibitory conductances during network bursting is not understood in detail, despite the fact that several antiepileptic drugs target GABAA receptors. Here, we show that the first manifestation of a burst event is a cell type-specific flurry of GABAA receptor-mediated inhibitory input to pyramidal cells, but not to stratum oriens horizontal interneurons. Moreover, GABAA receptor-mediated synaptic input is proportionally smaller in these interneurons compared with pyramidal cells. Computational models and dynamic-clamp studies using experimentally derived conductance waveforms indicate that both these factors modulate spike timing during synchronized activity. In particular, the different kinetics and the larger strength of GABAergic input to pyramidal cells defer action potential initiation and contribute to the observed delay of firing, so that the interneuronal activity leads the burst cycle. In contrast, excitatory inputs to both neuronal populations during a burst are kinetically similar, as required to maintain synchronicity. We also show that the natural pattern of activation of inhibitory and excitatory conductances during a synchronized burst cycle is different within the same neuronal population. In particular, GABAA receptor-mediated currents activate earlier and outlast the excitatory components driving the bursts. Thus, cell type-specific balance and timing of GABAA receptor-mediated input are critical to set the appropriate spike timing in pyramidal cells and interneurons and coordinate additional neurotransmitter release modulating burst strength and network frequency.
Synchronized neuronal activity is an important feature of many regions of the brain under physiological and pathological conditions. In the hippocampus, synchronized bursts are correlated with specific behavioral states and occur during sharp waves (SWs), slow wave sleep, and interictal electrographic activity (interictal spikes) in epileptic animals or humans (Matsumoto and Ajmone-Marsan, 1964; Buzsaki et al., 1983; Suzuki and Smith, 1985, 1988). These rhythms are believed to depend on the local circuitry of the hippocampus and represent an intrinsic pattern of network activity (Traub and Miles, 1991). The expression of specific voltage-dependent conductances (Wong and Prince, 1981; Traub et al., 1991; Migliore et al., 1995) and the abundance of recurrent collateral excitatory connections between CA3 hippocampal pyramidal neurons (MacVicar and Dudek, 1982; Li et al., 1994) are critical features for the initiation of network bursting, which can be preserved in vitro (Traub and Miles, 1991). Indeed, burst firing in a single CA3 pyramidal cell generates a very powerful excitatory drive onto its postsynaptic targets (Miles and Wong, 1984) so that, under conditions of abolished fast GABAergic inhibition, an individual pyramidal neuron can recruit every other pyramidal cell of the CA3 network in a synchronized population burst (Miles and Wong, 1983). However, complete blockade of fast inhibition is not a necessary requirement for burst generation, and other in vitro models of synchronicity exist (Wilson and Bragdon, 1993). Partially restricted synchronized bursts (partial bursts) can be achieved in slices under a variety of different experimental conditions that do not completely abolish GABAergic input. For example, hippocampal slices exposed to elevated concentrations of external potassium engage in rhythmic activity (Rutecki et al., 1985; Traynelis and Dingledine, 1988; Traub and Miles, 1991; McBain et al., 1993), and bursts remain restricted to a portion of the network (Traub and Miles, 1991). During partial bursts, the non-bursting cells of the network display low-frequency oscillations driven by synchronous subthreshold synaptic potentials (Traub and Miles, 1991). Three main variables have been shown to play an important role in modulating this type of network burst in the hippocampus: (1) the specific intrinsic properties of the participating neurons (Chamberlin and Dingledine, 1989; Traub and Miles, 1991; Kamondi et al., 1998; Fricker and Miles, 2001); (2) the strength of glutamatergic input originating from the recurrent collateral positive feedback of the CA3 region (Staley et al., 1998; Bains et al., 1999); and (3) the strength of GABAergic input (Korn et al., 1987; Traub and Miles, 1991; Staley et al., 1998).
In this study, we have combined experimental work with modeling to understand the functions of specific synaptic input to CA3 pyramidal cells and interneurons during synchronous activity. Our results underscore the importance of cell type specificity of GABAA receptor-mediated input for the generation of properly timed and graded bursting in CA3 pyramidal cells and interneurons.
Materials and Methods
Slice preparation. Young rats (Sprague Dawley; age, postnatal days 13-19) were deeply anesthetized using isoflurane and killed by decapitation, in accordance with National Institutes of Health and institutional protocols. The brain was removed quickly and placed into ice-cold cutting solution of the following composition (in mm): 234 sucrose, 28 NaHCO3, 2.5 KCl, 1.25 NaH2PO4, 0.5 CaCl2, 7 MgCl2, 7 glucose, 1 ascorbic acid, and 3 pyruvic acid, saturated with 95% O2/5% CO2, at pH 7.4. The hemisected brain was then glued onto the stage of a vibrating microtome (Leica, Nussloch, Germany), and sections of 400 μm thickness were cut and stored in an incubation chamber for ∼1 hr at room temperature before use. The composition of the artificial CSF (ACSF) in the incubation chamber was (in mm) 130 NaCl, 24 NaHCO3, 3.5 KCl, 1.25 NaH2PO4, 1.5 CaCl2, 1 MgSO4, and 10 glucose, saturated with 95% O2/5% CO2, at pH 7.4. Slices were then transferred, as needed, to the recording chamber and observed under an upright Axioskop microscope (Zeiss, Oberkochen, Germany) equipped with a 60× water immersion differential interference contrast objective coupled with an infrared camera system (Dage-MTI, Michigan City, IN).
Electrophysiological recordings. Conventional whole-cell current-clamp and voltage-clamp techniques were applied using a Multiclamp 700 amplifier (Axon Instruments, Union City, CA). Pipettes used for current-clamp experiments (resistance, ∼5 MΩ) were filled with the following solution (in mm): 125 K-methylsulfonate, 10 NaCl, 4 MgATP, 0.3 GTP, 16 KHCO3, and 0.5% biocytin, saturated with 95% O2/5% CO2 to a pH of 7.2 and 285-295 mOsm. Voltage-clamp recordings were performed on pyramidal cells and interneurons with electrodes of ∼2 MΩ resistance and filled with the same intracellular solution including 10 mm N-(2,6-dimethylphenylcarbamoylmethyl) triethylammonium bromide to minimize voltage-dependent conductances. Series resistance values were ∼10 MΩ and were not compensated. Dynamic-clamp recordings were performed using a Synaptic Module 1 (SM-1) conductance injection amplifier (Cambridge Conductance, Cambridge, UK) (Robinson and Kawai, 2000; Aradi et al., 2004) and a Multiclamp 700 amplifier (Axon Instruments) in bridge balance mode. The reversal potentials for the excitatory and inhibitory conductance commands were set to 0 and -63 mV, respectively. The electrical circuit of SM-1 computed the current to be injected into the soma of the cell. All recordings were performed at 33-35C° in a modified ACSF, in which the total concentration of KCl was raised to 8.5 mm. In addition, the ACSF used for dynamic-clamp experiments included 2-amino-5-phosphonovaleric acid (20 μm), 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide (10 μm), and SR-95531 (gabazine; 12.5 μm) (Tocris Cookson, Ellisville, MO). The temperature of the solution was monitored by a probe in the recording chamber and could be changed by a heating system applied to the perfusing solution before entering the bath (TC344A; Warner Instruments, Hamden, CT). The estimated reversal potential for GABAA receptor-activated current (EGABAA) was calculated to be -63 mV according to the following equation:
where [Cl-]i was 10 mm, [Cl-]o was 141.5 mm, was 16 mm, was 24 mm, and b was 0.2 (Bormann et al., 1987). However, taking into account that elevated potassium conditions induce a positive shift of fast IPSPs (Korn et al., 1987), we included in our estimated value of EGABAA a shift of ∼3 mV, similar to what was experimentally reported in another study using whole-cell recording conditions (McBain, 1995). High-chloride electrodes were filled with an intracellular solution in which K-methylsulfonate was substituted by equimolar KCl. Under these experimental conditions, the calculated EGABAA was ∼0 mV. Interneurons were selected in the stratum oriens and had bipolar horizontal appearance. Our results were obtained in the CA3c region, which has been reported to be the initiating focus of high-K synchrony in the rat adult hippocampus (Korn et al., 1987). Therefore, assuming a similar situation in the juvenile rat hippocampus, it should be kept in mind that events recorded close to the focus may not mirror those in the follower regions. Cells were usually maintained at a membrane potential of approximately -60 mV to reduce the occurrence of isolated action potentials in the interburst period.
Data analysis. Data were filtered at 3 kHz and digitized at 10-20 kHz using a Digidata 1322A analog-to-digital board. Analysis was performed using the pClamp (Axon Instruments), Origin (MicroCal, Northampton, MA), Excel (Microsoft, Seattle, WA), Spike 2 (CED, Cambridge, UK), Whole-Cell Program (courtesy of Dr. J. Dempster, University of Strathclyde, Glasgow, UK), and Prism (GraphPad, San Diego, CA) software packages. Analysis of the synchronous events was performed by threshold-triggered alignment of spontaneous events, using one channel as trigger so that the temporal relationships between the two channels were maintained intact. Because it is difficult to determine the exact timing of initiation of a burst, the interevent frequency was calculated from interevent intervals determined by the peaks of the bursts or subthreshold synchronous events. The analysis of the properties of synchronous events was performed on ∼1-3 min duration epochs under the appropriate experimental configuration (baseline control, drug application, or different holding potential). The duration of the epoch was selected based on the frequency of the events in the individual experiment. The decay time constant of the spontaneous EPSC was fitted by a monoexponential or biexponential function, as appropriate, and weighted time constants were used for comparison (Maccaferri et al., 2000). Statistical comparisons were performed using the appropriate t test, one- and two-way ANOVA with Bonferroni post hoc correction. Values are given as mean ± SE.
Determination of arealeft and arearight. To compare the complex kinetics of synaptic conductances between two cells recorded during a burst, we defined two indices (arealeft and arearight) that are related to the degree of similarity of the two recorded waveforms. “Arealeft” was defined as the difference between the areas of the normalized conductances [to -1 normalized unit (n.u.)] between the initial baseline and a reference point that was the peak of the recording in one of the cells. Therefore, arealeft quantifies the kinetics of nonoverlapping synaptic activation during the rising phase. “Arearight” was calculated similarly, but during the decay phase, i.e., from the peak back to the baseline (see left and right insets of Fig. 5, bottom).
Modeling CA3 pyramidal cells. The multicompartmental model of bursting CA3 pyramidal cells developed by Traub et al. (1991) was reconstructed (Aradi et al., 2004) with a few modifications. We performed simulations using NEURON software (version 5.4) (Hines and Carnevale, 1997), running on Linux OS. Briefly, the CA3 pyramidal cell model consisted of 19 compartments and six voltage-activated channels: Na+, Ca2+, delayed rectifier K+, Ca2+-dependent afterhyperpolarizing K+, Ca2+- and voltage-dependent K+, and A-type K+ channels. [For the details of the compartmentalization of the model cell and for the kinetics of the channels, see Traub and Miles (1991).] In addition, intracellular calcium dynamics from Migliore et al. (1995) were downloaded from the ModelDB database of SenseLab (http://senselab.med.yale.edu/senselab/modeldb) and incorporated to give a more realistic description of the CA3 pyramidal cell. Some additional adjustments of the parameters of the spontaneously bursting Traub model of the CA3 pyramidal cell were necessitated to achieve stable membrane potential at -60 mV in the absence of synaptic inputs and obtain robust bursting behavior in response to the complex combinations of excitatory and inhibitory drive present in the elevated potassium in vitro model of synchronized network oscillations. These adjustments included the reversal potential for K+ set to -74.5 mV, and the reversal potential for the leak current set to -55.1 mV; for every compartment, the maximum conductance of the delayed rectifier K+ channel was multiplied by 1.45, the maximum conductance of the Ca2+ channel was multiplied by 4.01, the maximum conductance of the Ca2+- and voltage-dependent K+ channel was multiplied by 0.72, and the maximum conductance of the Ca2+-dependent K channel was multiplied by 0.6.
The excitatory and inhibitory synaptic inputs to the CA3 pyramidal cell were simulated using the actual average experimental traces of EPSCs and IPSCs measured in voltage-clamp configuration at -60 and 0 mV, respectively, during the network oscillations. Both the excitatory and inhibitory synaptic conductances were scaled to different maximum conductance values detailed in Results. The IPSC conductance was given to the soma, whereas the EPSC conductance was given to the proximal apical dendrites. The reversal potential for the EPSCs was set to 0 mV, whereas for the IPSCs it was -60 mV.
Visualization of recorded cells and reconstruction. Methods were similar to the ones described by Maccaferri and Dingledine (2002). Briefly, slices were fixed for 1-10 d in a 4% paraformaldehyde PBS solution at 4°C. Endogenous peroxidase activity was removed by incubating the slices in 10% methanol and 1% H2O2 PBS solution. Biocytin staining was processed using an avidin-HRP reaction (Vectastain ABC Elite kit; Vector Laboratories, Burlingame, CA), and axon visualization was improved using a PBS solution containing NiNH4SO4 (1%) and CoCl2 (1%). Slices were not resectioned but directly mounted on the slide using an aqueous mounting medium (Vectashield; Vector Laboratories). Alternatively, they were first dehydrated and then mounted on the slide using a toluene solution (Permount; Fisher Scientific, Fair Lawn, NJ). Slices were observed at 40× magnification and photographed using the AxioVision software package (Zeiss).
Synchronized bursting and effect of GABAA receptor blockade
Hippocampal slices were exposed to elevated potassium (8.5 mm) ACSF. Whole-cell recordings from CA3 pyramidal neurons often displayed spontaneous bursts (Fig. 1A). Hyperpolarization of the cell by current injection prevented firing but did not abolish the occurrence of spontaneous events, suggesting that, under our experimental conditions, rhythmic activity was driven by the network and not by intrinsic conductances. This hypothesis was directly tested by recording simultaneously from pairs of neighboring pyramidal neurons (Fig. 1B,C). Double recordings confirmed the occurrence of synchronized events, which could be a combination of bursts in both cells, burst in one cell and subthreshold events in the other (Fig. 1B), or subthreshold events in both cells.
What is the role of GABAA receptor-mediated input during this pattern of activity? We addressed this question by recording from spontaneously active pairs (n = 4) or single CA3 (n = 3) pyramidal neurons before and after the addition of the GABAA receptor antagonist SR-95531 (gabazine; 12.5 μm) (Heaulme et al., 1986). Application of gabazine to the slice triggered a transient period of irregular firing, which was followed by a new steady-state level of network frequency (Fig. 2A). During this newly established steady state, the frequency of spontaneous synchronous events decreased from 0.62 ± 0.07 Hz in control to 0.21 ± 0.02 Hz (p < 0.05; n = 11). When normalized to control conditions, frequency decreased to 0.34 ± 0.03 (n = 11) in the presence of the drug, as shown by the cumulative probability plot of the instantaneous network frequency (Fig. 2B, left). In contrast to this decrease in frequency, the strength of the events, quantified by their area, increased from 1.9 ± 0.4 V⋆msec in control to 5.7 ± 0.6 V⋆msec in gabazine (n = 11; p < 0.05). Normalization to control conditions shows that gabazine increased the area of the events by 3.8 ± 0.6 times (Fig. 2B, right). These results indicate that blockade of fast inhibition promotes multiple effects leading to a change of burst strength and frequency. Slices displaying ictal activity (Dzhala and Staley, 2003) or in which gabazine application triggered cyclic electrographic seizures (Khazipov et al., 2004) without recovering to an interictal-like bursting rhythm were discarded from this analysis.
We next designed experiments with the aim of evaluating how activation of GABAA receptors expressed on the surface of the two recorded cells impacted network-driven bursting. We recorded synchronized activity in pairs of cells using different intracellular solutions and then applied gabazine to the slice. In one cell of the pair, we used a standard intracellular recording solution with a chloride concentration of 10 mm (low [Cl-]i). Under this condition, EGABAA was approximately -60 mV. The other neuron of the pair was recorded with a high-chloride concentration intracellular solution (135 mm; high [Cl-]I; estimated EGABAA, ∼0 mV). We reasoned that the different driving force for GABAA receptor-mediated currents in the two recorded neurons would affect synchronized activity that is modulated directly by receptors expressed in the two recorded cells. In contrast, network functions controlled by GABAA receptors expressed in the rest of the network would remain similar. In addition, application of gabazine should have opposite effects on functions mediated primarily by the receptors located on the two recorded cells.
Before gabazine application, the frequency of network-driven events was identical under the two recording configurations (0.85 ± 0.12 Hz in low [Cl-]i vs 0.85 ± 0.12 Hz in high [Cl-]i; n = 4; p > 0.05). However, we found that events were larger in the cells recorded with the high-chloride solution (Fig. 3A). The area of the synchronous events under low-versus high-chloride recording conditions was 2.6 ± 0.9 versus 12.1 ± 2.5 V⋆msec, respectively (n = 4 double recordings; p < 0.05). Gabazine application decreased the frequency of synchronous events in both cells to a similar degree. Network frequency decreased from 0.85 ± 0.12 Hz to 0.32 ± 0.03 Hz (n = 4; p < 0.05) in cells recorded with low-chloride solutions and from 0.85 ± 0.12 Hz to 0.36 ± 0.05 Hz (n = 4; p < 0.05) in cells recorded with high-chloride solution (Fig. 3B, left). Normalization to control reveals that frequency decreased to 0.38 ± 0.03 in low-chloride versus 0.43 ± 0.04 in high-chloride recording conditions (n = 4; p > 0.05). Gabazine-induced modulation of the strength of network-driven events, however, was clearly dissociated by the two different recording conditions (Fig. 3B, right). Gabazine increased the area of the synchronous events recorded with low-chloride electrodes from 2.6 ± 0.8 V⋆msec to 6.8 ± 1.9 V⋆msec (n = 4; p < 0.05; corresponding to an increase of 2.82 ± 0.35 after normalization), whereas a decrease of burst strength was observed in the cell recorded with a high-chloride pipette. Under these latter conditions, the area of the synchronized events decreased from 12.1 ± 2.5 V⋆msec to 5.5 ± 1.8 V⋆msec (n = 4; p < 0.05; corresponding to a decrease of 0.46 ± 0.10 after normalization). In conclusion, these first sets of experiments (Figs. 2, 3) indicate that GABA release occurs during network oscillations and can powerfully activate GABAA receptors in the slice, which play an important role in regulating burst strength and frequency.
Interneuron activity during network bursts
Synaptic release from presynaptic terminals of interneurons is the main source of GABA acting on postsynaptic receptors (Buhl et al., 1994; Freund and Buzsaki, 1996) It has been shown that during elevated potassium-induced network activity, stratum oriens horizontal interneurons are driven by glutamatergic periodic inward currents (McBain, 1995). However, the temporal relationship between these interneurons and pyramidal cell firing has not been explored in detail. Because GABAA receptor-mediated input modulates bursts in pyramidal cells, we assumed that interneurons must also be active. To verify this hypothesis, we recorded simultaneously from horizontal stratum oriens interneurons and pyramidal cells (Fig. 4A). Horizontal stratum oriens interneurons are easily recognizable in a living slice and can target several different postsynaptic domains of pyramidal cells (Maccaferri et al., 2000). Interneurons displayed strong burst generation in 10 of 12 recordings and subthreshold synaptic oscillations in the remaining two. In all cases, events were strongly synchronized to pyramidal cell activity. In four cases, suprathreshold activity was concomitantly observed in both cells. Initiation of interneuron firing always preceded and outlasted action potential generation in the pyramidal cells (Fig. 4B), similarly to what has been reported in human epileptic tissue and normal primate slices (Schwartzkroin and Haglund, 1986) As shown in Figure 4C, the frequency of network events was virtually identical in pyramidal cells and interneurons (0.81 ± 0.07 vs 0.82 ± 0.07 Hz, respectively; n = 12; p > 0.05), but the strength of the synchronized events was markedly different (1.4 ± 0.2 V⋆msec in pyramidal cells vs 3.7 ± 0.2 V⋆msec in interneurons; n = 12; p < 0.05). The mean ratio of the interevent frequency in interneurons over pyramidal cells was 1.0 ± 0.0 (n = 12), and the mean ratio of the area was 3.7 ± 0.9 (n = 12). The tight synchronization between interneurons and pyramidal cells and the fact that we consistently recorded from spatially close neurons (Fig. 4D) suggest that the recorded cells were mostly driven by the same set of afferent fibers. These results also indicate that stratum oriens interneurons are activated more powerfully than pyramidal cells during synchronized bursting and that spike timing within a bursting cycle is different between principal cells and interneurons. The observed spike timing in interneurons would predict that synaptic inhibitory activity should precede and even outlast excitatory inputs to the postsynaptic targets of the recorded cells.
Synaptic dynamics of GABAA-mediated versus excitatory input in pyramidal neurons
We performed a set of experiments with the aim of dissecting out the dynamics of fast inhibitory and excitatory conductances during the same bursting cycle in pyramidal cells (Fig. 5). By effectively reducing inhibitory or excitatory currents in one cell by voltage clamping at the relevant reversal potential, we could compare the kinetics of the isolated GABAergic or glutamatergic input and their temporal relationship. When two neighboring cells were held at -60 mV, excitatory burst currents were overlapping (Fig. 5, top left inset) as determined by the values of arealeft and arearight (see Materials and Methods for definitions of arealeft and arearight). The calculated values from eight double recordings were 2.0 ± 3.6 n.u.⋆msec (arealeft) and 10.9 ± 4.0 n.u.⋆msec (arearight). Next, we evaluated the dynamics of GABAA receptor-mediated conductances in one cell of the pair held at 0 mV (which is the assumed reversal potential for excitatory conductances), while recording the other cell at -60 mV (Fig. 5, top middle inset). Under this configuration, the values for arealeft and arearight were -38.9 ± 14.3 n.u.⋆msec and -39.0 ± 4.1 n.u.⋆msec, respectively (n = 8; p < 0.05), demonstrating that when close to the reversal potential of excitatory transmission, the inhibitory component temporally precedes and outlasts the excitatory waveform in the other neuron. We then switched configuration again, so that this time -60 mV was applied to the cell previously held at 0 mV and vice versa (Fig. 5, top right inset). This manipulation reversed the values for arealeft and arearight to 22.7 ± 7.8 n.u.⋆msec and 31.2 ± 10.2 n.u.⋆msec (n = 8; p < 0.05). We concluded that the time course of fast inhibitory and excitatory conductances activated during a burst in pyramidal cells is indeed different. In particular, inhibitory conductances have a longer duration and precede and outlast the excitatory waveforms. To verify that currents recorded at 0 mV were mediated by GABAA receptors, in seven double recordings we applied gabazine and found that the spontaneous outward currents recorded at 0 mV were completely abolished (Fig. 6). We further corroborated these results in another series of experiments by recording from a pair of cells both held at -60 mV. However, one cell of each pair was patched with a low-[Cl-]i pipette (10 mm), whereas the second cell was recorded with a high-[Cl-]i electrode (135 mm). We reasoned that the cell recorded with the low-[Cl-]i pipette would reveal the time course of the excitatory conductances only, whereas the second neuron recorded with the high-[Cl-]i electrode would reveal a combination of the dynamics of both excitatory and inhibitory conductances, weighted by their strength. As shown in Figure 7, the kinetics of the currents were clearly different in the two cells, suggesting earlier activation and later termination of GABAA-mediated conductances. Indeed, the kinetic difference between the time courses of the synaptic currents in the two recorded neurons was significantly reduced when gabazine was applied. We quantified this difference by calculating the arealeft and arearight indices for the cells recorded with the high- and low-[Cl-]i electrode. Arealeft switched from -18.2 ± 5.7 n.u.⋆msec in control solution to 1.1 ± 1.3 n.u.⋆msec in the presence of gabazine (n = 7; p < 0.05), whereas arearight changed from -30.1 ± 8.4 n.u.⋆msec to -13.1 ± 7.5 n.u.⋆msec, after application of the drug (n = 7; p < 0.05), suggesting that the earlier activating and later decaying waveform was mediated by GABAA receptors.
Direct effect of fast inhibitory conductances on pyramidal cells during a bursting cycle
The specific dynamics of fast inhibitory and excitatory conductances during a burst suggest that the latency to the first spike during a synchronous cycle may be controlled by GABAA receptor-mediated conductances. To address this question, we compared the latency to the first spike in the absence and in the presence of gabazine (Fig. 8). We quantified the effect of gabazine on the latency to the first spike by measuring the time difference (Fig. 8, Δ latency) of the point on the trace located halfway from baseline to spike threshold. Spike threshold was defined as the membrane potential level when dV/dt exceeds 10 V/sec (Fricker and Miles, 2001). Gabazine had a prominent effect on the latency to the first spike in pyramidal cells, as expected (Δ latency was 13.4 ± 4.8 msec; n = 8) (Fig. 8A). In contrast, no effect of gabazine could be found on Δ latency in five individually recorded interneurons (-0.3 ± 0.6 msec; n = 5; p < 0.05). As a positive control to show that gabazine had an effect on the slices in which interneurons were recorded, we measured the changes in frequency and area of interneuronal bursts. As expected, and similarly to what is shown in Figures 2 and 3, the frequency of bursts in interneurons decreased from a control value of 0.83 ± 0.13 Hz to 0.29 ± 0.06 Hz (p < 0.05; n = 5) (Fig. 8C) in the presence of gabazine. In addition, gabazine increased the area of the bursts from 5.1 ± 0.8 V⋆msec to 7.5 ± 1.2 V⋆msec (p < 0.05; n = 5) (Fig. 8C). These data suggest that early inhibition during a burst plays a role in regulating spike timing in pyramidal cells but not in interneurons. It is, however, important to note that gabazine application in the slice is likely to trigger a variety of indirect effects (see also Discussion). For example, the augmented area of the events recorded in interneurons could depend on the increased bursting in pyramidal neurons and not necessarily be the result of direct blockade of fast inhibition on the interneurons themselves.
Next, we took advantage of computational modeling techniques, in which manipulations do not suffer from these indirect effects. We used the time course of the voltage-clamp recordings shown in Figure 5 to simulate GABAA receptor-mediated (gIPSC) and fast excitatory (gEPSC) conductances. We fed gIPSC and gEPSC to a model of a CA3 pyramidal cell and evaluated the role of the ratio between gIPSC and gEPSC on the threshold and timing for burst initiation. As shown in Figure 9A, increasing the relative contribution of gIPSC over gEPSC converted full bursts to subthreshold synaptic potentials and regulated spike timing within a burst. Full-blown bursts produced by the simulations were similar, but shorter than experimental records (Fig. 9A). This is probably because of the fact that gEPSC was estimated from voltage-clamp recordings at -60 mV, in which the contribution of NMDA receptors (NMDARs) to synaptic conductance is minimal, if not completely absent (Hestrin et al., 1990). This result is also consistent with the notion that NMDAR-mediated synaptic transmission is essential for seizure generation in the CA1 area but is not required for interictal-like bursts in the CA3 hippocampal subfield (Traynelis and Dingledine, 1988) The gIPSC/gEPSC ratio was crucial in setting the spike timing within the burst (Fig. 9B): this effect was observed for all the different values of conductances tested and therefore is likely to reflect a general network mechanism (Traub and Miles, 1991).
Synaptic events in pyramidal neurons and stratum oriens horizontal interneurons during network activity
Network dynamics are determined by the integrated pattern of activity of all its cellular components. The results shown in Figure 8 suggest that the burst initiation may be differently regulated in pyramidal cells versus interneurons. Therefore, we decided to approach this question directly by recording and comparing the dynamics of synaptic inputs on pyramidal cells versus stratum oriens horizontal interneurons with respect to both of their excitatory and inhibitory components. First, we studied spontaneous EPSCs (sEPSCs) occurring during interburst intervals (Fig. 10A, double arrow). sEPSCs recorded on pyramidal cells and interneurons held at -60 mV showed clear differences, similar to what has been reported for evoked EPSCs in quiescent slices (Hestrin et al., 1990; Sah et al., 1990). From a total of 12 double recordings, the 10-90% rise times of these events were 2.2 ± 0.2 msec in pyramidal cells versus 0.6 ± 0.1 msec in interneurons (p < 0.05), the decay time constant (τ) was 8.9 ± 0.5 versus 3.6 ± 0.7 msec (p < 0.05), and the charge transfer was 0.50 ± 0.03 versus 0.17 ± 0.01 pC (p < 0.05), respectively. In contrast, excitatory inputs during burst activity (bEPSC) were not significantly different between the two cell types. Despite a larger variability of the events recorded in the interneuron population, no statistical differences could be found in the mean charge transfer (52 ± 6.7 pC in pyramidal cells vs 82 ± 28 pC in interneurons; p > 0.05) (Fig. 10B, left) or time course (see below) (Fig. 11). Taken together, these results indicate that, despite differences in sEPSC unitary components, the longer duration of bEPSCs filters out their kinetic differences. In addition, the specific integrative properties of pyramidal cells and interneurons might contribute to produce a similar excitatory input, which explains the tight synchronicity of their activation, as shown in Figure 4. Next, we analyzed burst events under voltage-clamp configuration at Vh = 0 mV (bIPSCs) (Fig. 10B, right) in pyramidal cells and interneurons. bIPSCs were larger in pyramidal cells when compared with interneurons (the charge transfer measured was 167 ± 11 vs 31.4 ± 5.8 pC; respectively; n = 10; p < 0.05). Given the importance of the gIPSC/gEPSC ratio we have shown before (Fig. 9), it is important to note that the ratio of the average bIPSC charge transfer over the average charge transfer of bEPSCs was larger in pyramidal cells than interneurons (∼3 in pyramidal cells compared with ∼0.4 in interneurons). In addition, a barrage of synaptic activity increasing in a ramp manner often preceded the main event in pyramidal cells, whereas such activity could not be observed in interneurons (Figs. 10B, right; 11).
The differences of the dynamics of the synaptic inputs during a burst were dependent on the holding membrane potential, which would isolate excitatory and inhibitory components (Fig. 11). When both cells were held at -60 mV, the time course of the recorded conductances was very similar. Under this configuration, the difference between the areas delimited by the normalized waveforms in the two cell types was as follows: arealeft, -2.7 ± 3.1 n.u.⋆msec; arearight, 0.2 ± 8.7 n.u.⋆msec; n = 10. Changing the holding potential in both cells to 0 mV caused a significant increase of the area resulting from the subtraction of the normalized waveforms. Under this experimental condition, arealeft and arearight were 48 ± 5 and -49 ± 16 n.u.⋆msec, respectively (n = 10; p < 0.05). These results indicate that inhibitory input during a bursting cycle is temporally cell type specific and differs between pyramidal cells and interneurons.
To explore the significance of cell type-specific fast inhibition, we ran simulations and injected conductances derived from actual voltage-clamp recordings to a model pyramidal cell at different time windows (Fig. 12). We tested the effect of cell type specificity of fast GABAergic conductances on the timing of the first spike in a burst (Fig. 12A). As shown in Figure 12, B and C, the prediction of these simulations was that both the strength and kinetics contribute to delay action potential initiation within a burst in pyramidal neurons. Dynamic-clamp experiments performed in pyramidal cells fully confirmed these predictions: feeding an inhibitory conductance derived from pyramidal cells had a different effect on the latency to the first spike when compared with waveforms derived from interneurons (Fig. 12B, bottom) (data from nine pyramidal cells under dynamic-clamp conditions; p < 0.05).
In this study, we have answered three fundamental questions related to network bursting under conditions of preserved fast GABAergic inhibition: (1) What is the role of GABAA receptors expressed on CA3 pyramidal cells during network bursting? (2) What is the natural pattern of activation of fast inhibitory and excitatory conductances during a burst cycle? (3) Does cell type-specific inhibition play any role in the regulation of network bursting?
Direct role of GABAA receptors on CA3 pyramidal cells during network bursting
We have taken advantage of the whole-cell configuration to control the level of chloride in the recorded cells and manipulate the reversal potential of GABAA receptor-mediated currents. Our data demonstrate that GABAA receptor activity in the recorded neurons affects the strength of the burst but not the synchronicity with the rest of the network. In agreement with this result, computational modeling shows that the ratio of inhibitory over excitatory conductances acts as a gate to allow bursting versus subthreshold oscillatory activity. Therefore, a convergence of results indicates that GABAA receptors expressed on pyramidal cells can powerfully modulate network bursting. Voltage-clamp analysis of synchronicity in adult animals, however, has suggested a much smaller role for fast GABAergic conductances (Staley et al., 1998). This discrepancy most likely reflects the different developmental stage of the network. GABAA receptors are phosphorylated by different kinases (Poisbeau et al., 1999) and modulated by hormones (Lambert et al., 2003), environmental metabolic products (Pasternack et al., 1996), and drugs (Havers and Luddens, 1998). All these factors could play important roles in gating and/or grading bursting in individual neurons without affecting their participation to the general network burst frequency. This mechanism is likely to play an important role in the spatial selection of bursting neuronal assemblies in vivo. For example, not all CA3 pyramidal neurons burst during SWs (Buzsaki, 1986; Csicsvari et al., 2000): such a selection mechanism could have important consequence for the induction of long-term synaptic plasticity (Buzsaki, 1989). It is important to underscore that our experimental approach coupled with modeling is very different from the more traditional experiments based on unspecific blockade of GABAA receptors in the circuit. Indeed, unspecific blockade of GABAA receptors in a slice changes both burst strength and interburst frequency (Fig. 2) (Korn et al., 1987; Staley et al., 1998; Khazipov et al., 2004). The change in network frequency makes the differences between bursts in the presence and absence of GABAA receptor-mediated inhibition extremely difficult to interpret. For example, a change in the frequency of spontaneous events will initiate bursting at a different phase of the burst afterhyperpolarization (Alger and Nicoll, 1980; Chamberlin and Dingledine, 1989) and lead to the recruitment of different cell assemblies (Traub and Miles, 1991). Furthermore, the reduced frequency of spontaneous bursts is likely to interfere with the short-term plasticity (Salin et al., 1996) and glutamate release probability (Staley et al., 1998) of excitatory recurrent collateral connections. In addition, the use of unspecific GABAA receptor antagonists may affect intracellular chloride homeostasis and thus enhance GABAB receptor-mediated inhibition (Lopantsev and Schwartzkroin, 1999, 2001). Similarly, blockade of fast inhibition on interneurons (Hajos and Mody, 1997) may lead to increased firing and consequently augmented GABAB receptor-mediated inhibition (Traub and Miles, 1991; Miles and Wong, 1984; Scanziani, 2000). Finally, as we show here, GABAA receptor blockade results in transient and irregular network hyperexcitability before a new steady state is reached (Fig. 2). This transient period of irregular hyperexcitability may trigger long-term changes of inhibitory or excitatory synaptic transmission (Bains et al., 1999; Chevaleyre and Castillo, 2003). In conclusion, application of GABAA receptor antagonists on the entire slice triggers a cascade of indirect events that makes it impossible to assess the direct role of fast inhibition on pyramidal cell bursting. Here, we also show that the effect of GABAA antagonists on network frequency of synchronous events in the recorded cells requires blockade of receptors in the entire slice. The most parsimonious explanation is that increased interneuron firing results in enhanced presynaptic and/or postsynaptic GABAB-mediated inhibition (Scanziani, 2000). Indeed, the slow kinetics of postsynaptic GABAB receptor-operated conductances (Otis et al., 1993) would be predicted to increase the interburst duration.
Natural pattern of activation of inhibitory conductances during a burst cycle
Network modeling studies have suggested that a burst cycle is always initiated by a single or a few “ancestral” CA3 pyramidal neurons bursting spontaneously (Traub and Miles, 1991). Under this assumption, action potentials in these ancestral cells would result in glutamate release and suprathreshold activation of a fraction of their postsynaptic targets (i.e., pyramidal cells and GABA-releasing interneurons). Spike timing in these first targets would be primarily determined by their specific EPSP-spike coupling properties and, in turn, would then determine the activation of GABAergic and glutamatergic conductances in the next round of postsynaptic neurons, which are disynaptically connected to the initiating cells. EPSP-spike coupling properties are different between pyramidal cells and specific classes of interneurons and depend on their specific intrinsic and synaptic properties (Fricker and Miles, 2001; Maccaferri and Dingledine, 2002). If interneuron firing preceded and outlasted firing in pyramidal cells (Fricker and Miles, 2001; Maccaferri and Dingledine, 2002), then activation of GABAA receptor-mediated conductances would be expected to have a broader duration than excitatory input. However, the progressive recruitment of additional cells of the CA3 circuit would make the interplay between synaptic input and intrinsic properties much more complex and even difficult to predict. Here, we provide the first direct experimental quantification of the natural pattern of activity of fast inhibitory and excitatory conductances during the same burst cycle. The finding that the mean pyramidal cell bIPSC precedes the mean pyramidal cell bEPSC is apparently paradoxical but consistent with the “ancestral pyramidal neuron(s)” hypothesis of Traub and Miles (1991). Our results also clearly show that horizontal stratum oriens cells are powerfully active during network bursting. This broad class of interneurons includes cells with different axonal patterns that target several distinct postsynaptic domains on target pyramidal cells (Maccaferri et al., 2000). Therefore, it is not surprising that fast inhibitory conductances can have a significant impact on burst regulation. For example, besides gating suprathreshold activity, our simulations suggest that the ratio between inhibitory and excitatory burst conductances contributes significantly to setting the appropriate spike timing during a burst cycle (Fig. 9).
Cell type-specific inhibition and regulation of network bursting
A high degree of target cell specificity has been described for a few classes of GABAergic interneurons. Specific classes of axoaxonic and perisomatic targeting cells (Somogyi et al., 1983; Acsady et al., 2000), for example, establish synaptic contacts preferentially on excitatory neurons when compared with interneurons. Nevertheless, interneurons specifically targeting other interneurons have also been described (Freund and Buzsaki, 1996). These classes of interneurons would be predicted to generate a cell type-specific biased GABAergic input. Here, we show that, in addition to intrinsic properties, cell type-specific inhibitory input further contributes to the maintenance of the appropriate spike timing between pyramidal cells and stratum oriens horizontal interneurons. Our experiments demonstrate that a pyramidal cell-specific input is functional during synchronous bursting and that the ratio between the overall GABAA receptor-mediated and fast excitatory inputs is larger in pyramidal cells compared with stratum oriens horizontal interneurons. Recently, Kogo et al. (2004) have shown cell type-specific depression of GABAergic input in stratum oriens horizontal interneurons, mediated by activation of group III metabotropic glutamate receptors (mGluRs). Therefore, it is also possible that group III mGluRs might depress fast inhibition in stratum oriens horizontal interneurons during synchronous activity. What interneuron subtype(s) originate(s) the pyramidal cell-specific input? At present, we can only speculate that our observation of a selective flurry of GABAergic input at the beginning of a burst cycle reminds of the observation that axoaxonic cells fire immediately before the initiation of a SW cycle in the CA1 region in vivo (Klausberger et al., 2003). If this was the case, the domain-specific location of this input would be perfectly suited to prevent or delay firing. Our simulations and dynamic-clamp experiments suggest that this type of input could be important to maintain the appropriate spike timing in pyramidal cells, which is an essential variable in the induction of some forms of long-term synaptic plasticity (Bi and Poo, 1998). The occurrence of early spiking during the burst could trigger back-propagating action potentials inducing associative synaptic depression that could lead to the shutdown of the network-driven activity (Bi and Poo, 1998; Bains et al., 1999). Finally, it is interesting to note that the composition of GABAA receptors expressed on the axon initial segment of pyramidal cells is enriched in α2 subunits (Nusser et al., 1996), suggesting the possibility of specific pharmacological modulation. In contrast, the cell type-specific differences in kinetics of excitatory synaptic transmission during the interburst intervals are strongly attenuated during burst events, which are of much longer duration (Fig. 10). This network-state specificity is likely to be important for the maintenance of synchronicity between pyramidal cells and interneurons.
In conclusion, we have unveiled new mechanisms of burst control in the CA3 hippocampus, highlighting a precise role for cell type-specific GABAergic input. This new knowledge could prompt future development of different pharmacological strategies in the treatment of brain conditions associated with burst hyperexcitability and synchronicity, such as epilepsy.
This work was supported by the National Institute of Mental Health (Grant 5R01MH067561 to G.M.) and the National Institute of Neurological Disorders and Stroke (R. Dingledine) during its preliminary phase. We thank Dr. C. McBain (National Institute of Child Health and Human Development, National Institutes of Health) and Drs. M. Bevan, A. Contractor, D. McCrimmon, and M. Martina for constructive criticism of this manuscript. In addition, we thank Dr. R. Dingledine (Emory University, Atlanta, GA) for help and support in pilot experiments leading to this project. We also thank Dr. C. J. Heckman for help in the cross-correlation analysis and Dr. J. Dempster (University of Strathclyde, Glasgow, UK) for providing us with the Whole-Cell Program analysis package.
Correspondence should be addressed to Dr. Gianmaria Maccaferri, Department of Physiology, Northwestern University Medical School, 303 East Chicago Avenue, Tarry Building, Room 5-707 M211, Chicago, IL 60611. E-mail:.
Copyright © 2004 Society for Neuroscience 0270-6474/04/249681-12$15.00/0