Transition to seizures in the isolated immature mouse hippocampus: a switch from dominant phasic inhibition to dominant phasic excitation

Abstract

The neural dynamics and mechanisms responsible for the transition from the interictal to the ictal state (seizures) are unresolved questions in epilepsy. It has been suggested that a shift from inhibitory to excitatory GABAergic drive can promote seizure generation. In this study, we utilized an experimental model of temporal lobe epilepsy which produces recurrent seizure-like events in the isolated immature mouse hippocampus (P8–16), perfused with low magnesium ACSF, to investigate the cellular dynamics of seizure transition. Whole-cell and perforated patch recordings from CA1 pyramidal cells and from fast- and non-fast-spiking interneurons in the CA1 stratum oriens hippocampal region showed a change in intracellular signal integration during the transition period, starting with dominant phasic inhibitory synaptic input, followed by dominant phasic excitation prior to a seizure. Efflux of bicarbonate ions through the GABA_A receptor did not fully account for this excitation and GABAergic excitation via reversed IPSPs was also excluded as the prime mechanism generating the dominant excitation, since somatic and dendritic GABA_A responses to externally applied muscimol remained hyperpolarizing throughout the transition period. In addition, abolishing EPSPs in a single neuron by intracellularly injected QX222, revealed that inhibitory synaptic drive was maintained throughout the entire transition period. We suggest that rather than a major shift from inhibitory to excitatory GABAergic drive prior to seizure onset, there is a change in the interaction between afferent synaptic inhibition, and afferent and intrinsic excitatory processes in pyramidal neurons and interneurons, with maintained inhibition and increasing, entrained ‘overpowering’ excitation during the transition to seizure.

One of the central mysteries of epilepsy is the underlying neural mechanisms responsible for the transition into a seizure from the interictal state. Both clinical and laboratory evidence suggest that during temporal lobe epilepsy (TLE), the most prevalent form of the epilepsies (Engel, 1989), three functional brain states exist: interictal, preictal and ictal (the seizure episode) (Litt & Lehnertz, 2002). The preictal, or transitional state, is dynamically different from the ictal and interictal states (Iasemidis, 2003) and involves a progressive change in signal complexity (Babloyantz & Destexhe, 1986; Basar, 1998; Lehnertz et al. 1999; Chiu et al. 2005) preceding a seizure. However, the underlying dynamical cellular neurophysiological mechanisms leading to the spontaneous ictal state are still unclear.

Neuronal networks can display non-linear and complex behaviours that result in multiple stable states, with the capacity to undergo spontaneous transitions between these states. From a dynamical perspective, these transition periods can be abrupt, discontinuously ‘jumping’ from one state to another, as observed in most cases of absence seizures, or there can be a gradual evolution into the ictal state, as often observed with TLE (Lopes da Silva et al. 2003b). Using an in vitro model of TLE (Rafiq et al. 1993, 1995; Zhang et al. 1995; Derchansky et al. 2006) which generates recurrent seizure-like events, and in which the transition from interictal to the ictal state resembles clinically defined human epileptiform activity (Goetz & Pappert, 1999), we investigated the intracellular dynamics and the sequence of inhibitory and excitatory events during the preictal state.

At present, our knowledge about the balance between inhibition and excitation during and prior to the ictal state is unclear. Under epileptic conditions, studies show that multiple forms of inhibition are involved, depending on the location of inhibitory synapses on the neuron. For example, in animal models of epilepsy, dendritic but not somatic GABAergic inhibition is decreased (Houser & Esclapez, 1996; Cossart et al. 2001) and it has been hypothesized that this is the mechanism responsible for ictal generation. Computational modelling also shows similar time-varying interactions between interneurons and pyramidal cells and suggests that epileptiform activity can be explained by impaired GABAergic dendritic inhibition prior to ictal onset (Wendling et al. 2002).

Another hypothesis for ictal generation postulates that interneurons might be involved in synchronizing large neuronal populations. This synchronization is possible via their abundant connectivity to pyramidal neurons and other interneurons. In addition to synchronization, excitation might be achieved through the alteration of the intracellular chloride gradient after prolonged high-frequency activation of the GABA_A receptor, leading to an excitatory GABAergic synaptic response (Perreault & Avoli, 1988; Kaila et al. 1997; Taira et al. 1997). Bicarbonate efflux through GABA_A channels, which depolarizes the cell, has been proposed to be an additional synaptic mechanism mediating the excitatory GABAergic action (Lambert & Grover, 1995; Staley et al. 1995; Lamsa & Kaila, 1997; Uusisaari et al. 2002; Perez Velazquez, 2003).

Previous studies have shown that the CA3 region plays an important role in driving epileptogenesis and hyperexcitability (Walther et al. 1986; Tancredi et al. 1990), with the CA3 region leading the CA1 during the ictal state (Derchansky et al. 2006). The current study was focused on the CA1, ‘receiving’ end as we wanted to investigate the intracellular activity of the region being driven, which can potentially shed light on the intracellular processes by which large-scale neuronal networks become synchronized. To examine the neural dynamics during transition into seizures and to investigate the inhibitory/excitatory balance during the transition state, we used an in vitro model of epilepsy, recording recurrent seizures in pyramidal neurons and interneurons in the CA1 region of the intact, isolated mouse hippocampus and in hippocampal slices.

Methods

Hippocampal preparation

Male C57/BL mice (postnatal days (P) 8–16) were anaesthetized with halothane and decapitated in accordance with the Canadian Animal Care Guidelines. Briefly, the brain was extracted and placed in ice-cold (2–5°C), oxygenated (95% O₂, 5% CO₂) ACSF containing (mm): 123 NaCl, 2.5 KCl, 1.5 CaCl₂, 2 MgSO₄, 25 NaHCO₃, 1.2 NaH₂PO₄ and 15 glucose. The hippocampus was then dissected out by gently sliding the spatula ventral to the corpus callosum, making the insertion first at the septal hippocampal region. This dissection was performed while constantly perfusing the tissue with ice-cold, oxygenated ACSF. The tissue was then transferred to oxygenated ACSF at room temperature for a minimum of 1.5 h before being placed into the recording chamber (modified from Khalilov et al. 1997). For a more detailed method of the dissection, please see Derchansky et al. (2004).

Some experiments in this study utilized the hippocampal slice. Briefly, 500 μm thick slices from rats (P15–25) were obtained using a vibratome (Series 1000, Technical Products International, St Louis, MO, USA). Slices were incubated for at least 1 h before being transferred to an interface-type chamber for electrophysiological recording. For a more detailed method of the dissection, please see Jahromi et al. (2002).

Experimental environment

Epileptiform activity was induced in the intact hippocampus and hippocampal slice by perfusing the tissue with low-Mg²⁺ ACSF containing (mm): 123 NaCl, 5 KCl, 1.5 CaCl₂, 0.25 MgSO₄, 25 NaHCO₃, 1.2 NaH₂PO₄ and 15 glucose (10 glucose for slices). The intact hippocampus was placed in the recording chamber and maintained at 32°C, with humidified warmed oxygen flowing over the solution in the chamber. ACSF flowed over the tissue at a flow rate of 7 ml min⁻¹. Hippocampal slices were placed in an interface-type recording chamber and were also aerated with humidified oxygen.

Electrophysiology

Extracellular recordings were obtained from the CA1 cell layer of the hippocampus in the middle septotemporal level using glass pipettes filled with 150 mm NaCl. For whole-cell electrophysiology, infrared imaging with differential interference contrast (IR-DIC) was used to visualize pyramidal cells and interneurons using an upright microscope (BX51, Olympus, Melville, NY, USA) under ×40 magnification (water-immersion objective) with an OLY-150IR camera-video monitor unit (Olympus). Whole-cell patch clamp recordings were performed using an Axopatch 200B amplifier (Axon Instruments, Union City, CA, USA) in the current clamp configuration, with 5–8 MΩ electrodes pulled from borosilicate capillary tubing (World Precision Instruments, Sarasota, FL, USA) with a Narashige pipette puller (PP-830, Japan).

Two whole-cell patch pipette solutions were used. First, the ‘high chloride’ (30 mm) solution contained (mm): 8 NaCl, 0.0001 CaCl₂, 10 NaHepes, 20 KCl, 110 potassium gluconate, 1 MgCl₂, 0.3 NaGTP, 2 NaATP. The second solution with ‘normal chloride’ (5 mm) varied in potassium gluconate and KCl composition and contained (mm): 140 potassium gluconate and 5 KCl. Neurobiotin (3%) was dissolved in all whole-cell patch pipette solutions for staining and subsequent morphological analysis of the recorded neurons. The perforated patch pipette solution contained (mm): 50 KCl, 2 Hepes, 0.1 EGTA and gramicidin (≤ 50 μg ml⁻¹, Sigma-Aldrich, Oakville, Canada). For whole-cell recordings, the resistance of the whole-cell seal was 2–4 GΩ before breaking through the membrane. After achieving a whole-cell recording, the resting membrane potential was recorded (RMP) and current pulses (± 100 pA, 900 ms, at 25 pA increments) were injected into the neuron to obtain the I–V relationship and the input resistance (R_input). Interneurons in the stratum oriens were characterized based on their morphology and electrophysiological response to hyperpolarizing and depolarizing current steps. Interneurons were defined as fast-spiking (FS) when their peak discharge frequency (peak frequency) in response to the peak depolarizing current step exceeded 50 Hz, and as non-fast-spiking (non-FS) when this peak frequency fell below 50 Hz. Categorization based on these criteria was previously used in characterizing hippocampal interneurons (Fujiwara-Tsukamoto et al. 2004). Pyramidal cells were identified by their characteristic spike-frequency adaptation (peak frequency = 15.0 ± 5 Hz), their morphological features as well as by visual inspection based on the location of the electrode in the pyramidal cell layer.

The reversal potential of the spontaneous preictal activity was determined by manually injecting slow negative and positive current to hyperpolarize and depolarize the cell over several seconds, and monitoring the resting membrane potential at which point the polarity of the activity changed. Electrical stimulation in the presence of APV and CNQX was performed in the CA3 region with a bipolar electrode (10–20 V, 50 μs). Pressure-injection of muscimol to the dendrites and soma of pyramidal cells was performed using a Picospritzer (General Valve, TX, USA).

Morphological reconstruction

After fixing the tissue in 4% paraformaldehyde, intact hippocampi were cut serially at 50 μm along the coronal plane using a vibratome. Sections were washed in PB and Tris-buffered saline (TBS; 0.05 m, pH 7.4), then blocked by normal goat serum (5%) containing Triton X-100 (0.5%) in TBS for 45 min. After three washes in TBS sections were treated with Alexa546-conjugated streptavidin solution (1 : 200, Molecular Probes, Burlington, Ontario, Canada) to visualize the Neurobiotin-filled cells. Sections were analysed under a fluorescent light microscope (Olympus AX70) and the filled cells were photographed (Spot digital camera).

To reconstruct the axonal and dendritic arborization, sections were rinsed in TBS, and incubated in avidin–biotinylated horseradish peroxidase complex (ABC, 1 : 200, Vector Laboratories, Burlington, Ontario, Canada) for 2 h at room temperature. The immunoperoxidase reaction was carried out using Ni²⁺-intensified 3,3′-diaminobenzidine 4-HCl (DAB, Sigma-Aldrich) as a chromogen and 0.05% H₂O₂. Sections were mounted, air-dried and cover-slipped, and the filled profiles were reconstructed using a camera lucida.

Pharmacology

All drugs were purchased from Sigma-Aldrich (Oakville), including 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX), bicuculline methiodide (BMI) and dl-2-amino-5-phosphonovaleric acid (APV). These agents were dissolved in double-distilled water at a specific concentration (CNQX, 10 mm; d- and dl-APV, 30 mm; BMI, 10 mm) and kept frozen in 100 μl aliquots. These were then dissolved into low-Mg²⁺ ACSF for the desired final concentration (CNQX, 10 μm; APV, 60 μm; BMI, 10 μm). Muscimol was dissolved in the extracellular pipette solution (500 μm) and lidocaine (lignocaine) N-methyl chloride (QX222) was dissolved in the intracellular pipette solution (10 mm). Ethoxyzolamide (ETZ) and acetazolamide were dissolved in dimethyl sulfoxide (DMSO; 0.01%) and aliquots of 100 μl were kept frozen and dissolved accordingly to obtain the final concentration (100 μm). DMSO (0.01%) had no effect on hippocampal field activity.

Calculating chloride equilibrium potentials (E_Cl)

For the three electrophysiological recording techniques using different intracellular chloride concentrations, the following Nernst equation for equilibrium potentials was used:

Where z =−1.0; T = 305 K, R = 8.314 J K⁻¹ mol⁻¹, F = 96 485 C mol⁻¹ and [X]_o and [X]_i are the extracellular and intracellular chloride concentrations, respectively. For perforated patch calculations, known intracellular chloride concentrations were used based on the age range of animals used in this study (Kandler & Friauf, 1995; Rohrbough & Spitzer, 1996; Ganguly et al. 2001; Staley & Smith, 2001; Ben-Ari, 2002).

Data analysis and event detection

Student's t tests were performed in SPSS (v10.0, SPSS Inc., Chicago, IL, USA). Statistical significance was set at P < 0.05. Power spectral analysis was performed in Clampfit 8.2 (Axon Instruments, Union City, CA, USA). Curve fitting was performed in Origin (v6.0, Microcal Software Inc., Northampton, MA, USA), and only curves with r > 0.95 were analysed. All results are shown as mean ± s.e.m.

Results

The preictal state is characterized by a dynamic switch from inhibition to excitation

Intracellular and extracellular recordings were obtained in the CA1 region of the intact isolated mouse hippocampus perfused with low-Mg²⁺ ACSF (Fig. 1Aa). Low-Mg²⁺ treatment resulted in recurrent seizures after 230 ± 50 s (P8–16, n = 160, Fig. 1Ab), with an average duration of 90 ± 4.0 s. Three characteristic states were observed in these recordings: interictal, preictal (transition state) and ictal. The intracellular and extracellular activity generating the ictal states in this model is dependant on excitatory synaptic transmission and is blocked by glutamatergic antagonists (Walther et al. 1986; Tancredi et al. 1990; Quilichini et al. 2002; Derchansky et al. 2004), and all preictal states evolved into ictal states.

Figure 1. The preictal state is characterized by three substates in pyramidal cells.

Aa, configuration of dual recording electrodes (whole-cell (WC), normal chloride, field (extracellular)) placed at the CA1 region in the middle septotemporal level. Infra-red image of the CA1 pyramidal cell layer as observed from a top view of the hippocampus (bottom inset). Ab, field recording of recurrent seizure-like events with three states: interictal, preictal and ictal, with a distinction between the preictal and ictal states by a clearly denoted 20 Hz oscillation at ictal onset (♦, bottom inset). Ba, whole-cell recording from pyramidal cell (PC) with field activity during the preictal state exhibits hyperpolarizing (preictal_h), mixed hyper-/depolarizing (preictal_m), and depolarizing (preictal_d) activity prior to ictal onset. Bb, representative intracellular activity during the three preictal substates. The preictal_m state is composed of hyperpolarizing (h arrow) and depolarizing (d arrow) activity. C, prior to ictal onset (t = 0 s), hyperpolarizing activity (t =−20 s to −10 s) decreases in amplitude towards a reversal in polarity with a gradually increase in the depolarizing peak amplitudes (measured by the arrows), with an average rate of approximately 2.0 mV s⁻¹. The initial resting membrane potential was −40 mV. Coupled with the phasic changes during this state, there is also a slow depolarization in the membrane potential prior to ictal onset.

The boundary between the preictal and ictal states was clearly defined extracellularly. The preictal state had an average burst frequency of 1.0 ± 0.5 Hz, followed by the ictal state which started with higher frequency discharges (20 ± 5.0 Hz; Fig. 1Ab, ♦). In all recordings, the entire preictal state had an average duration of 25.0 ± 5.0 s, and as the preictal state approached the ictal state, the intracellular firing patterns became more complex, with the generation of higher-frequency burst firing prior to ictal onset. As observed in a previous study (Derchansky et al. 2004), interictal discharges developed over time and only emerged after the first 6–10 ictal states. Hence, for the recording period in our study, the interictal state was marked by a quiescent period as recorded extracellularly and was clearly distinguishable from the low-frequency preictal state.

Our main finding is that during the preictal state, pyramidal cells (n = 53, V_m=−65.0 ± 4.0 mV, R_input= 260.3 ± 20.0 MΩ) displayed three distinct preictal substates consisting of: (a) initial hyperpolarizing synaptic activity (hyperpolarizing: preictal_h, observed when the cell was more depolarized than E_Cl), followed by (b) a mix of hyper/depolarizing activity (mix: preictal_m). During the preictal_m state, synaptic events were initially hyperpolarizing, followed by a depolarization. These depolarizations gradually increased, leading to (c) a preictal_d state (depolarizing) in which all events were depolarizing, prior to ictal onset. No ictal states were observed without the development of the preictal_d state as ictal onset never emerged directly from the preictal_h or preictal_m states.

The preictal state started with hyperpolarizing preictal_h events with gradually decreasing amplitudes. Eventually a total reversal in the polarity of these events was observed during the preictal_d state followed by increased depolarizing amplitudes prior to ictal onset (Fig. 1C). The rate of these changes was measured at 2.0 ± 0.3 mV s⁻¹ (n = 30), and was independent of the neuronal subtype or the resting membrane potential of the neurons. In addition to the aforementioned phasic discharges, the preictal state was accompanied by a slow 4–10 mV depolarization in membrane potential prior to ictal onset.

Recordings from fast-spiking interneurons (Fig. 2Aa–c, n = 41, V_m=−62.0 ± 6.0 mV, R_input= 145.4 ± 10.0 MΩ) and non-fast-spiking interneurons (Fig. 2Ba–c, n = 66, V_m=−61.0 ± 5.0 mV, R_input= 245.0 ± 20.0 MΩ) revealed that in both interneuronal subtypes the preictal_h, preictal_m and preictal_d states were preserved prior to ictal onset, and were similar to those observed in pyramidal cells. However, distinct from pyramidal cells and due to the higher-frequency intrinsic firing properties of interneurons, the preictal_d state in these neurons was accompanied by burst discharges of up to 50 Hz, which was absent in pyramidal cells. To fully characterize the recorded neuronal subtypes, we reconstructed axonal and dendritic arborization of the filled neurons. This analysis revealed that fast-spiking interneurons were identified as basket cells (Fig. 2Aa, n = 20, peak frequency = 67.0 ± 5.0 Hz), whose dendrites were localized to strata oriens, radiatum and lacunosum-moleculare, while the axon arbor extended heavily into the stratum pyramidale. In contrast, trilaminar interneurons, another class of fast-spiking interneurons (reconstruction not shown), possessed dendrites that were located in the stratum oriens (n = 21, peak frequency = 63.0 ± 2.0 Hz). The axonal arbors of trilaminar interneurons were pervasive in strata oriens, pyramidal and radiatum. The non-fast-spiking interneurons in this study were O-LM interneurons (Fig. 2Ba, n = 66, peak frequency = 35.0 ± 4.0 Hz). These interneurons were distinguishable by their large axonal arbors located in the stratum lacunosum-moleculare and a dendritic tree constrained to the stratum oriens.

Figure 2. The preictal state in FS and non-FS interneurons also consists of three distinct subphases.

Aa, morphological reconstruction of a FS interneuron (basket cell). The basket cell interneuron has an extensive axonal arbor in the CA1 pyramidal layer and dendrites extending in the stratum oriens and radiatum, with an average peak frequency of 67.0 ± 5.0 Hz. o, stratum oriens; p, stratum pyramidale; r, stratum radiatum; l, stratum lacunosum-moleculare. Ab, whole-cell recording from a FS interneuron demonstrates similar intracellular dynamics during the preictal state recordings to those observed in pyramidal cells, with all preictal h, m and d states preserved (Ac) prior to ictal onset. Ba, morphological reconstruction of a non-FS interneuron (O-LM). The O-LM interneuron (dendrites, black; axons, grey) has a characteristic axonal arbor in the stratum lacunosum-moleculare and a dendritic tree in the stratum oriens, with an average peak frequency of 35.0 ± 4.0 Hz. Bb and c, similar to pyramidal cells and FS interneurons, whole-cell recording from a non-FS interneuron in the CA1 stratum oriens region demonstrates three distinct preictal substates prior to ictal onset.

A full electrophysiological profile of pyramidal cells, FS and non-FS interneurons is detailed in Table 1. Furthermore, morphological analysis revealed that a small number of recorded neurons belonged to the axo-axonic (n = 3, peak frequency = 57.0 ± 5.0 Hz), and bistratified (n = 4, peak frequency = 64.0 ± 5.0 Hz) interneuronal subgroups. The morphology of all interneurons in our study is similar to what has been previously described (Sik et al. 1995). It must be noted that all interneurons displayed a similar transition dynamic irrespective of their electrophysiological and morphological classification.

Table 1.

Electrophysiological characterization of pyramidal, and fast-spiking (FS) and non-FS interneurons

			FS
Cell type	Pyramidal	Non-FS O-LM	Trilaminar	Basket
n	53 (4)	66 (6)	21 (5)	20 (5)
RMP (mV)	−68.0 ± 4.0	−62.0 ± 3.0	−62.0 ± 6.0	−63.0 ± 4.0
Input resistance (MΩ)	260.0 ± 20.0	267.0 ± 20.0*	145.0 ± 10.0	110.0 ± 15.0
Membrane time constant (ms)	40.0 ± 3.0	32.0 ± 3.0*	18.0 ± 3.0	12.0 ± 4.0
Peak frequency (Hz)	15.0 ± 5.0	35.0 ± 4.0*	63.0 ± 2.0	67.0 ± 5.0
Adaptation (%)	40.0 ± 5.0	26.0 ± 3.0*	18.0 ± 4.0	20.0 ± 2.0
AP width (ms)	0.9 ± 0.1	0.8 ± 0.05*	0.5 ± 0.02	0.4 ± 0.01

The number of recorded neurons (n) and the number of morphologically reconstructed cells (in parentheses). RMP, resting membrane potential; input resistance calculated using linear regression of voltage changes with applied hyperpolarizing current steps (−100 pA, increments of 25 pA). Membrane time constant (τ) was measured by fitting the first 200 ms of the hyperpolarizing voltage step response with an exponential function. Peak frequency was measured as the discharge observed when a step current of 200 pA was applied to the neuron. Adaptation is the ratio of 1 − (number of APs in last 200 ms/number of APs in first 200 ms) during a step current of 200 pA, multiplied by 100. AP width is the duration of the action potential measured at half-amplitude. Values are in s.e.m.

^*Significant differences in values between FS and non-FS interneurons. All results are shown as mean ± s.e.m.

The preictal_h state phasic potentials are generated by recurrent IPSPs

Calculated E_Cl and observed E_rev in pyramidal cells

To ascertain the nature of the preictal activity, we measured the reversal potentials of these spontaneous events. The dynamics of the preictal_h state were sensitive to whole-cell intracellular chloride concentrations. By manually applying a slow ramp of depolarizing current injection in the current clamp mode, the membrane voltage was monitored for polarity reversal of the preictal_h activity (‘region of reversal’, ROR, Fig. 3A). With electrodes containing ‘normal’ chloride (5 mm), the calculated E_Cl was −65.0 mV, and the experimentally observed reversal potential (E_rev) of the preictal_h state in pyramidal cells was −58.0 ± 5.0 mV (n = 20). Electrodes containing high chloride (30 mm) had a calculated E_Cl of −30.0 mV and an observed E_rev of −37.0 ± 3.0 mV in pyramidal cells (n = 30, Table 2).

Figure 3. IPSPs generate the preictal_h state.

Aa, intracellular recordings of the preictal_h state in a pyramidal cell using whole-cell (WC), perforated patch, and whole-cell technique with high internal chloride (whole-cell high Cl_i) to measure the reversal potential of the spontaneous preictal_h activity by manually injecting current to slowly depolarize and hyperpolarize the cell. In a sequence of intracellular events, the region of reversal (ROR) was defined as the region between the last positivity (+) and the first negativity (−) when depolarizing the cell, or the region between the last negativity and first positivity when hyperpolarizing the cell. The peak amplitude (arrows) was measured at the respective membrane potential (V_m). Ab, the reversal potential is sensitive to the intracellular chloride concentration, as whole-cell high Cl_i had a more depolarized reversal potential than standard WC and perforated patch. B, average reversal potentials (E_rev) of the preictal_h state in pyramidal cells, FS and non-FS interneurons with the three intracellular recording methodologies. Perforated patch reveals the most hyperpolarized E_rev, while a more depolarized E_rev is observed with whole-cell high internal chloride recordings, as suggested by the Nernst E_Cl. C, sample intracellular responses to electrical stimulation at varying membrane potentials in the presence of glutamatergic blockers (60 μm APV, 10 μm CNQX) and somatic pressure-injection of muscimol in two interneurons (perforated patch recording). The ROR for the evoked responses is between −70 and −55 mV. Da, dual whole-cell (normal chloride) and field recordings from a non-fast-spiking interneuron during BMI (10 μm) application created more recurrent and fragmented ictal episodes. Db, interneuronal preictal activity in low-Mg²⁺ (•, expanded from Da) is completely abolished with the application of BMI (♦, expanded from Da).

Table 2.

The calculated chloride equilibrium potential (E_Cl) and the observed E_rev values during the preictal state in pyramidal cells and interneurons

A.	Normal chloride	High chloride	Perforated patch
Calculated ECl (mV)	−65.0	−30.0	−81.0 to −60.0
Evoked EGABAA-Muscimol (mV)			−72.0 ± 4.0 (normal ACSF)
			−64.0 ± 3.0 (low Mg2+ ACSF – soma, interictal)
			−55.0 ± 5.0 (low Mg2+ ACSF – dendrite, interictal)

B. Observed Erev of PSP (mV)
	Preictalh			Preictald
	Normal chloride	High chloride	Perforated patch	Normal chloride	High chloride	Perforated patch
Pyramidal cell	−58.0 ± 5.0†	−37.0 ± 3.0†	−70.0 ± 4.0†	−30.0 ± 6.0‡	−7.0 ± 3.0‡	−2.0 ± 10.0‡
FS interneuron	−44.0 ± 4.0*†	−33.0 ± 4.0†	−55.0 ± 6.0†	+33.0 ± 5.0‡	+29.0 ± 6.0‡	+31.0 ± 11.0‡
Non-FS interneuron	−45.0 ± 4.0*†	−32.0 ± 3.0†	−58.0 ± 6.0†	+29.0 ± 6.0‡	+21.0 ± 4.0‡	+40.0 ± 11.0‡

A. E_Cl was calculated independently of the neuron being recorded as this value depends only on the chloride concentration in the patch pipette (top row). The values of evoked PSP potentials from focally applied muscimol on the soma or dendrites of CA1 neurons in normal ACSF and during the interictal state in low Mg²⁺ ACSF were obtained. Note the more depolarized values in the dendrites.

B. Observed E_rev values during the preictal_h and preictal_d states in the three neuronal groups with three recording solutions (normal chloride, 5 mm; high chloride, 30 mm; perforated patch).

^*Significant differences in values between the calculated E_Cl and observed E_rev values during the preictal_h state.

^†Significant difference in values between the three recording pipette solutions across the same neuronal group during the preictal_h state.

^‡Significant difference between E_rev values during the preictal_h state and E_rev values in the preictal_d state. All results are shown as mean ± s.e.m.

In addition to the two whole-cell patch recording solutions, gramicidin perforated patch recordings were used to prevent contamination of the intracellular chloride concentration by the recording solution. This resulted in a calculated E_Cl between −81.0 mV and −60.0 mV (based on [Cl]_i of 5–10 mm;Ben-Ari, 2002), and an E_rev of −70.0 ± 4.0 mV in pyramidal cells (n = 15). Hence, whole-cell recordings with normal and high chloride concentrations in the patch electrodes, and also with perforated patch recordings, revealed that the reversal potential of the preictal_h state in pyramidal cells was in reasonable agreement with the calculated E_Cl.

Observed E_rev in interneurons

A similar chloride-dependent sensitivity of the preictal_h state was observed in interneurons. No significant difference in E_rev values between FS and non-FS interneurons was observed (Table 2B, left columns). The observed E_rev in FS interneurons during the preictal_h state was −44.0 ± 4.0 mV (normal internal chloride), −33.0 ± 4.0 mV (high internal chloride) and −55.0 ± 6.0 mV (perforated patch; Fig. 3B). Non-FS interneurons had E_rev values of −45.0 ± 4.0 mV (normal internal chloride), −32.0 ± 3.0 mV (high internal chloride) and −58.0 ± 6.0 mV (perforated patch). This suggests that a chloride conductance via synaptically mediated IPSPs generates the activity observed during the preictal_h state.

Evoked IPSPs

We also measured the evoked E_IPSP by electrically evoking IPSPs in pyramidal cells and interneurons (Fig. 3C, n = 6, perforated patch) in the presence of the glutamatergic blockers APV (60 μm) and CNQX (10 μm), and by somatic pressure-injection of the GABA_A agonist muscimol in pyramidal cells (n = 5) and interneurons (n = 6; Fig. 3C). The evoked E_IPSP was −72.0 ± 4.0 mV in normal ACSF, but depolarized to −64.0 ± 3.0 mV after application of low Mg²⁺ ACSF. These measurements are consistent with the findings of other investigators that demonstrate that the resting membrane potential and the E_IPSP remain within 1–4 mV of each other between P5 and P31 (Banke & McBain, 2006).

GABA_A receptor activation is necessary for the generation of the preictal state

Since GABAergic IPSPs are involved in the generation of the preictal_h state, we examined the effects of GABA_A blockade by the GABA_A antagonist bicuculline methiodide (BMI, 10 μm, n = 45). Whole-cell recordings from pyramidal cells and interneurons showed that the interictal and preictal states were completely abolished with the addition of BMI to the low Mg²⁺ ACSF, and the ictal state developed discontinuously from quiescent neuronal activity without any interictal or preictal potentials (Fig. 3D). Furthermore, the previously observed slow membrane potential depolarization observed during the preictal state was completely abolished with BMI treatment. This is different from the previously observed gradual and continuous transition from the preictal state to the ictal episode in low Mg²⁺ ACSF without BMI. The ictal activity in this treatment was fragmented and was 30.0 ± 5.0% shorter in duration than ictal activity observed in low Mg²⁺ ACSF without BMI, while the frequency of occurrence of ictal episodes tripled.

The preictal_d phasic potentials show marked depolarized E_rev values

The reversal potentials for the preictal_d events were significantly more depolarized than those observed during the preictal_h state in all neurons. The preictal_d events showed chloride sensitivity only in pyramidal cells in that there was a significant E_rev difference between normal chloride (−30.0 ± 6.0 mV) and high chloride (−7.0 ± 3.0 mV, Table 2B; right columns) intracellular solutions. Conversely, interneurons demonstrated chloride-independent reversal potentials with no significant differences between the two intracellular whole-cell solutions and the perforated patch recordings, across both interneuronal subgroups. Furthermore, the observed E_rev values in interneurons during the preictal_d state were significantly more depolarized than the preictal_d reversal potential observed in pyramidal cells (by approximately 60 mV). Table 2 describes these differences in more detail, both across neuronal subtypes and the experimental intracellular solutions used.

Carbonic anhydrase inhibitors do not affect the preictal state

The depolarized reversal potential of the preictal_d state suggests that EPSPs are involved in the generation of this state. However, to exclude the possibility that this state is mediated by reversed IPSPs generated by an efflux of bicarbonate through GABA_A receptors, the hippocampus was perfused with the membrane-permeable carbonic anhydrase inhibitor, ethoxyzolamide (ETZ, 100 μm, n = 30). ETZ was used as a previous study demonstrated that this agent did attenuate some of the depolarizing GABAergic-mediated activity (Perez Velazquez, 2003). After 15–20 min of ETZ application, there was no change in recurrent ictal activity, frequency or duration. This observation was maintained through the entire duration of the experiment. All the preictal states observed in low Mg²⁺ ACSF with ETZ were conserved, including the three preictal substates h, m and d (Fig. 4A). The reversal potentials observed in interneurons and pyramidal cells during all preictal substates did not significantly change with ETZ application. Another carbonic anhydrase inhibitor, acetazolamide (50 μm, n = 10, all cell types), had similar effects (data not shown).

Figure 4. EPSPs and not the bicarbonate-mediated reversed-IPSPs generate the preictal_d state.

Aa, whole-cell recording from a pyramidal cell reveals that application of the carbonic anhydrase inhibitor ETZ (100 μm) does not attenuate spontaneous activity and maintains the interictal, preictal and ictal states. Ab, the three preictal substates are maintained, and the cell reveals an ROR similar to that of a pyramidal cell in low-Mg²⁺ ACSF. Dashed line indicates intracellular activity at the same membrane potential, with reversed polarities, just before and after the administration of the slow hyperpolarizing and depolarizing current steps. B, somatic pressure-injection of the GABA_A agonist muscimol (•, 500 μm, 10 ms, 10 p.s.i.) to pyramidal cells. Hyperpolarizing GABAergic responses are maintained throughout the preictal state as pharmacologically evoked by somatic activation of the GABA_A receptors. These responses are also evoked in FS interneurons (C) and non-FS interneurons (D) by somatic application of muscimol (•) throughout the preictal substates. All whole-cell recordings were performed using normal chloride intracellular solution.

Both somatic and dendritic GABAergic hyperpolarizing responses can be evoked during the entire preictal state

Next, we examined whether the recurrent depolarizing events observed during the preictal_d state were mediated by the excitatory action of GABA_A vis-à-vis reversed IPSPs (rIPSPs) by evoking a GABAergic postsynaptic response during this state. If reversed IPSPs were the driving forces during the preictal_d state, we would expect a depolarizing GABAergic response during this state. To test this hypothesis, we pressure-injected muscimol (500 μm, 10 ms, 10 p.s.i.) onto the soma of pyramidal cells (n = 6) and interneurons (n = 6 per subtype) throughout the entire preictal state. During all the preictal states, evoked GABAergic hyperpolarizing responses were still maintained with muscimol application in both pyramidal cells and interneurons (Fig. 4B–D). This further suggests that compound EPSPs play a large role in the preictal_d state.

It is possible that a differential chloride gradient collapse occurs in the dendrites and not in the soma. To examine whether rIPSPs could underlie the preictal_d state, pressure-injections of muscimol were applied to the distal dendrites of pyramidal cells in the stratum radiatum. Since the intact hippocampus is a thick structure, visually confirming that the muscimol-injecting electrode was in the stratum radiatum was technically challenging. Therefore, we utilized the hippocampal slice for this experiment. Slices perfused with low-Mg²⁺ ACSF demonstrated recurrent ictal activity with all three preictal substates present (Khosravani et al. 2003), similar to those observed in the intact hippocampus (P8–21, n = 30, Fig. 5A). As noted in the somatic layer, pressure-injection of muscimol to the dendritic region showed that evoked GABAergic hyperpolarizing responses were maintained throughout the preictal_d state (n = 6, Fig. 5B), with a reversal potential of −55.0 ± 5.0 mV. These data further suggest that the preictal_d state is generated by compound EPSPs rather than an excitatory dendritic or somatic GABAergic drive.

Figure 5. Hippocampal slices generate a preictal dynamic similar to that of the intact hippocampus, and hyperpolarizing GABAergic responses can be pharmacologically evoked by dendritic GABA_A activation of the receptor throughout the preictal state.

Aa, IR image from the CA1 stratum pyramidale (SP) cell layer in a hippocampal slice (400 μm), with a patched CA1 pyramidal cell (PC). Membrane voltage responses of the pyramidal cell to positive and negative current injections prior to seizure induction showing spike frequency adaptation that is common to these cell types (bottom inset). Ab, whole-cell recording (normal chloride) with field activity during the preictal state exhibits the three preictal substates observed in the intact hippocampus prior to ictal onset (♦, bottom inset) with the representative intracellular activity as observed during the three substates. Ba, IR image of the stratum pyramidale (SP) and stratum radiatum (SR) of the hippocampal slice with an intracellular recording electrode and the pressure-injector (PI) positioned in the dendritic region. Bb, dendritic pressure-injection of the GABA_A agonist muscimol (•, 500 μm, 10 ms, 10 p.s.i.) onto the pyramidal cell evokes hyperpolarizing GABAergic responses during the preictal_m and preictal_d substates, as shown by two recordings. C, sample of intracellular responses during the interictal state, at varying membrane potentials, to dendritic muscimol pressure-injection with a reversal potential of −55.0 mV.

The inhibitory drive is maintained during the preictal state

Since pressure-injection of muscimol resulted in stable hyperpolarizing GABAergic responses throughout the preictal state, we further investigated whether the change from IPSPs during the preictal_h state to EPSPs in the preictal_d state was generated by a lack of inhibitory input, or whether the inhibitory drive was still present, but was masked by a stronger excitatory drive. Intracellular application of QX222, a quaternary analogue of lidocaine used to attenuate electrically and pharmacologically evoked glutamatergic EPSPs in the CA1 region (Puil & Carlen, 1984), showed a stable inhibitory drive, which persisted throughout the entire preictal and ictal states in all neuronal subtypes (10 mm, n = 6, Fig. 6). Since QX222 blockade of EPSPs is more readily achieved when neurons are depolarized, neurons were kept at a resting membrane potential of −30.0 ± 10.0 mV with current injection. Furthermore, since EPSPs in this in vitro model are not generated exclusively by an inward Na⁺ conductance and include Ca²⁺ entry through NMDA receptors (DeLorenzo et al. 1998), EPSPs were not fully attenuated, but rather weakened, which was sufficient to uncover the persistent inhibitory drive. Although weakened, the remaining presence of the EPSPs resulted in a reduction in IPSP amplitude (45 ± 5.0%) and duration during the preictal_d state and during the ictal state.

Figure 6. Spontaneous, recurrent IPSPs are maintained throughout the entire preictal state.

A, independent whole-cell (normal chloride) recordings from two interneurons in different hippocampi. The normal action potentials (upper trace) are attenuated with QX222 (middle trace). Electrically evoked responses were recorded at −50, −60 and −70 mV and demonstrate the attenuation of EPSPs at these resting membrane potentials (boxed region, stimulus artifact omitted for clarity, represented by •). B, IPSPs are unmasked with intracellular application of QX222, which attenuates EPSPs during the preictal and ictal states, recorded in an interneuron (IN) with correlated field activity. C, IPSPs are maintained throughout the entire preictal state (♦, expanded from B). D, intracellular activity from the early preictal (•) and late preictal (^) reveals that application of QX222 does not completely attenuate all excitatory depolarization (d arrow), but weakens the effect of compound EPSPs that mask the underlying IPSPs.

Preictal dynamics in the CA3 region differ from those observed in the CA1 region

We have thus far explored the behaviour of the postsynaptic CA1 neurons during the preictal state and next, we investigated the preictal dynamics of the CA3 region, which plays an important role in epileptogenesis and hyperexcitability (Walther et al. 1986; Tancredi et al. 1990). In a previous study, we used multisite CA3 and CA1 extracellular recordings to demonstrate that the CA3 region leads the local CA1 region during the ictal state (Derchansky et al. 2006) and we now explored the intracellular correlate of the CA3 region and its relationship to local CA1 field activity during the preictal state.

Whole-cell recordings from pyramidal neurons in the CA3 region, with a field recording in the CA1 region (n = 7, Fig. 7A) revealed that the overall IPSP-to-EPSP transition pattern prior to ictal onset was preserved in this region and was similar to the pattern observed in the CA1 region (Fig. 7B). However, there were certain distinctive features in CA3 neurons during the preictal state which were not observed in the CA1 neurons and a clear distinction between the preictal h, m and d states was not as apparent in these neurons. During the preictal_h state, pyramidal cells in the CA3 region fired action potentials prior to the CA1 field activity with an average latency of 14.0 ± 1.1 ms (Fig. 7B–D). These dynamics differed to those observed in the CA1 neurons, as CA1 pyramidal cells only displayed IPSPs with no intrinsic bursting discharges during the preictal_h state. Overall, CA3 neurons were more hyperexcitable than those observed in the CA1 region as these recordings revealed a mixture of both intrinsic bursts of action potentials and recurrent IPSP/EPSPs.

Figure 7. Preictal intracellular dynamics of CA3 pyramidal cells.

A, configuration of dual recording electrodes (PC, whole-cell, field (extracellular)) placed at the CA3 and CA1 region in the middle septotemporal level. B, whole-cell pyramidal cell recording with field activity during the preictal and ictal states. C, the preictal state (♦, expanded from B) exhibits a mixture of synaptic activity and action potentials, which later develop into EPSPs prior to ictal onset, similar to the preictal_d state observed in CA1 neurons. D, intracellular CA3 activity leads the CA1 field (•, expanded from C), while IPSPs follow field activity. Vertical lines indicate onset of intracellular CA3 activity.

In 43% of CA3 pyramidal cells, activity during the preictal state consisted of an EPSP, followed by an IPSP, which differed from the intracellular activity observed in CA1 pyramidal cells, which displayed an initial IPSP followed by an EPSP during the preictal_m state. Interestingly, during the preictal_h state, the IPSPs in the CA3 neurons lagged the CA1 field activity, even though the intrinsic CA3 action potentials discharged prior to the CA1 region (Fig. 7D). During the preictal_d state, intracellular activity was consistent with that observed in CA1 neurons during this state and was characterized by recurrent EPSPs. Similar to the dynamics observed in the CA1 region, ictal onset was always preceded by the preictal_d state, as we never observed the initiation of the ictal state directly from the preictal_h state.

Discussion

In this study, we describe the cellular dynamics of ictal/seizure transition in an in vitro model of temporal lobe epilepsy which generates recurrent seizures. This transition is characterized by a dominant inhibitory drive followed by dominant excitation. To the best of our knowledge, this is the first detailed neurophysiological investigation characterizing this state. The preictal state was classified into three substates (h, m, d), in which IPSPs were generated by dominant GABAergic inputs, switching later to a dominant excitatory drive prior to ictal onset. Blocking GABA_A receptors completely abolished the entire preictal state and resulted in discontinuous sudden transitions into temporally fragmented ictal episodes. Intermittent pressure-injections of a GABA_A agonist showed that the excitatory actions of GABA_A, as observed by other investigators, did not significantly contribute to the polarity switch of the spontaneous postsynaptic potentials during the preictal state in our model. Since the excitatory action of GABA is a result of prolonged activation of the GABA_A receptor brought about by high frequency activity (Perreault & Avoli, 1988; Kaila et al. 1997; Taira et al. 1997), we propose that this phenomenon was absent during the preictal state due to this state's lower-frequency activity, which is insufficient to create a depolarizing GABA response. Rather, it is at the ictal onset where higher-frequency activity is observed. Hence, the observed transition dynamics from preictal h → m → d must be explained by alternative mechanisms.

Changes in the balance of entrained spontaneous afferent inhibition and excitation as a possible mechanism for the dynamics of the preictal state

The spatial and temporal interactions between IPSPs and EPSPs play an important role in determining neuronal activity, since the integration of IPSPs and EPSPs can result in either a summative depolarizing or hyperpolarizing effect, or the shunting of excitatory currents through open inhibitory receptors. This dynamic has been observed when blockade of GABAergic inputs uncovers an increased excitatory drive in epileptic hippocampi (Behr et al. 2000; Otsu et al. 2000; Dalby & Mody, 2001).

Another example of this integration phenomenon was observed in the visual cortex where excitatory currents were dampened by hyperpolarizing IPSPs (Ferster & Jagadeesh, 1992). A recent study in the barrel cortex showed that interneurons mediate feedforward inhibition by powerfully shunting thalamocortical excitation (Sun et al. 2006). A computational simulation of neocortical pyramidal cells concluded that approximately 200 inhibitory synapses are required to prevent strong synaptic excitation and that shunting may occur during periods of maximal excitatory activity (Bush & Sejnowski, 1994).

In spinal motor networks, a dynamic balanced combination of simultaneous inhibition and excitation has been demonstrated to drive spiking activity (Berg et al. 2007) and to underlie the ‘Up states’ in the prefrontal cerebral cortex in vivo (Haider et al. 2006). We postulate that during the transitions to seizure observed in this model, there is a functional change in signal integration. Even though it is still experimentally difficult to separate the inhibitory and excitatory influences during complex epileptic events, we have shown that the excitatory drive, which is potentially less active or is masked either by a summative hyperpolarization or shunted by open inhibitory receptors during the preictal_h state, emerges during the preictal_d state, when EPSPs overcome inhibitory input. Another possible mechanism responsible for the dynamics observed during the preictal state is the use-dependent depression of IPSPs resulting from postsynaptic mechanisms such as a depolarizing shift in the E_IPSP, decreased IPSP conductance and decreased input resistance (McCarren & Alger, 1985). Although these factors might contribute to the transition to seizure, it is not the primary mechanism responsible for ictal onset in our model, since somatic and dendritic applications of muscimol show robust postsynaptic inhibitory GABAergic responses throughout the preictal state, and the QX222 data indicate a maintained inhibitory drive. Hence, although IPSP depression might be present, it cannot fully account for our observations.

Instead, increased entrained concurrent phasic afferent synaptic excitation (possibly with concomitant increased intrinsic excitation, such as dendritic calcium spikes) can explain the dynamics observed during the preictal state, primarily because it is known that synaptic input from the CA3 region plays an important role in driving epileptogenesis and hyperexcitability (Walther et al. 1986; Tancredi et al. 1990), and our previous work indicates that the CA3 region leads the CA1 during the ictal state (Derchansky et al. 2006). This study was designed to investigate the ‘receiving’ end and not the driver (CA3), which was the rationale for recording in the CA1 region, as we wanted to investigate the intracellular activity of the region being driven and investigate the process by which large-scale neuronal networks become (hyper)synchronized.

The presence of a slow depolarization of the membrane potential during the preictal state, and the attenuation of this depolarization with the application of BMI suggests that although phasic GABAergic responses remain inhibitory, tonic GABAergic inhibition could play an excitatory role during the transition into the ictal state. Recently, it has been demonstrated that tonic depolarizing extra-synaptic GABA_A conductance in the neonatal hippocampus promotes neuronal excitability (Marchionni et al. 2007) and we suggest that coupled with the observed increase in phasic excitatory inputs during the preictal_d state, this conductance plays an important and additive role in ictal onset. Future experiments can utilize subunit-specific antagonists for extra-synaptic GABAergic receptors to investigate their effect on the slow preictal depolarization.

Although this first hypothesis relies on the differential breakdown of the chloride gradient across the neuron (e.g. synaptic versus extra-synaptic locations) to generate the slow depolarization, it is also possible that the slow depolarization observed during the preictal period is non-GABAergically mediated and an alternative hypothesis suggests that because the addition of BMI causes the system to switch directly to the ictal state and eliminates its preictal counterpart, the slow depolarization associated with the preictal state is simply eliminated with the preictal period. This suggests that in the absence of inhibitory input to the neuron, the system is sensitive to depolarizing inputs and does not require a slow depolarization to reach its ictal state.

Potential mechanisms responsible for observed differences in preictal_d reversal potentials

Differential changes in synaptic drive in pyramidal cells and interneurons could explain the more hyperpolarized preictal_d reversal potential in CA1 pyramidal cells (−30.0 ± 6.0 mV, normal chloride) than in interneurons (31.0 ± 2.0 mV, chloride-independent). This suggests that pyramidal cells are innervated either by stronger afferent inhibition or by weaker excitation during the preictal_d state. This also suggests that different interneuronal subgroups might be innervating pyramidal cells and CA1 interneurons. This interneuronal subgroup could be more active during the preictal_d state, resulting in more simultaneous GABAergic activity with EPSPs, giving rise to a more hyperpolarized reversal potential in pyramidal cells.

The growing evidence for differential interneuronal populations innervating pyramidal cells and other interneurons could explain the different functional electrophysiological changes observed during the preictal period. Studies investigating somatodendritic targets of interneurons reveal that local axo-axonic interneurons synapse onto an average of 686 pyramidal cells (Buhl et al. 1994), while a single basket cell interneuron innervates approximately 1500–2000 pyramidal cells and only 60 other PV+ interneurons (Sik et al. 1995). Other inhibitory neurons innervating pyramidal cell dendrites include back-projecting, bistratified, O-LM, Schaffer collateral associated, CCK+, hippocamposeptal, CR+ and trilaminar interneurons. Another interneuronal family which specializes in innervating only other interneurons includes the IS-1, IS-2 and IS-3 neurons (Acsady et al. 1996a,b; Gulyas et al. 1996; Hajos et al. 1996).

The preferential generation of low-threshold calcium spikes (LTS) and dendritic sodium spikes in interneurons could be another potential mechanism that could explain these neurons' more depolarized preictal_d reversal potentials. Neocortical interneurons demonstrate burst firing due to the activation of dendritic T-type calcium channels, which serve to amplify small synaptic inputs and trigger LTS at the soma (Goldberg et al. 2004). Calcium imaging in the olfactory bulb reveals that LTS are associated with calcium transients in the proximal and distal dendrites and that together with high-frequency dendritic sodium spikes, these conductances act in regulating intracellular calcium concentrations (Pinato & Midtgaard, 2005). Hence, we suggest that aside from the balance in afferent inhibition and excitation during the preictal_d state, the underlying conductances in interneurons and pyramidal cells during this state could be different, as reflected by the differences in the reversal potentials.

The ‘clocking’ functionality of the preictal state

Abolishing the preictal state with a GABAergic antagonist resulted in more recurrent and fragmented ictal episodes with no interictal activity, and an attenuation of the slow membrane depolarization prior to ictal onset. This indicates that the transitory state is not necessary for ictal onset and generation, but acts as a clock to keep the ictal periods similar in duration rather than being temporally scattered and irregular. When the dominant inhibitory drive is abolished, leaving only excitatory input in a hyperexcitable environment, there is a higher degree of neuronal activity, suggesting that the function of the interneuronal drive, as observed during the preictal_h state, is to dampen and control such activity.

Using extracellular field recordings, we have demonstrated that in this recurrent seizure model, the CA3 region consistently leads the CA1 region (Derchansky et al. 2006). Here we demonstrate that during the early preictal state in the CA3, even though the CA3 neuron fires before the CA1 field activity, the IPSP in this neuron actually lags the CA1 field activity. Other experimental models of limbic seizures showed that spontaneous interictal activities originating from the CA3 region act to suppress ictal episodes in the entorhinal cortex (Bragdon et al. 1992; Barbarosie & Avoli, 1997; Bragin et al. 1997). It has been demonstrated that the CA3-driven glutamatergic interictal activity modulates GABA-mediated interictal activity, which controls the rate of ictogenesis (Barbarosie et al. 2002). Without such a control mechanism, as revealed by GABAergic blockade and physical isolation of the CA3 region, ictal episodes become exacerbated. Other experimental evidence for this notion has been reported in the barrel cortex, where axo-axonic interneurons keep network excitability from going out of control (Zhu et al. 2004). A loss of functionality of these neurons prevents dampening of excessive excitation, thereby increasing the propensity towards epileptogenesis (DeFelipe, 1999).

The interictal and ictal states

In tissue obtained from patients suffering from TLE, ‘pacemaker cells’ in the subiculum discharge synchronously with interictal bursts (Cohen et al. 2002). These bursts are GABAergically mediated and the activity of these pacemaker cells, which are composed of interneurons and a subgroup of pyramidal cells, has a reversal potential of approximately −55 mV in the presence of glutamatergic antagonists. Since most patients in this study suffered from hippocampal sclerosis, there was a loss of excitatory input to the subiculum from the CA1 region, resulting in continuous bursts of interictal activity with no dynamic transition to seizure. This mechanism could underlie the continuous interictal bursts and supports our notion that for a transition from interictal to the ictal state to occur, dominant excitatory input via EPSPs is needed. Also, the preictal_h state in our model was generated by IPSPs, but if the resting membrane potential of the neurons was more hyperpolarized than the E_Cl, the observed activity during this state will be depolarizing, via reversed IPSPs, similar to the observed activity generated by the subiculum pacemaker cells. Furthermore, since the preictal_h state follows the interictal state, it is reasonable to consider that both states are generated by similar GABAergic drives.

The ictal state has been extensively investigated in hippocampal slices and recordings from interneurons and pyramidal cells in this state show that neurons receive direct glutamatergic inputs, possibly from recurrent CA1 collaterals, as well as synchronous excitatory GABAergic inputs generated by a depolarized GABA_A reversal potential (Kaila et al. 1997; Perez Velazquez, 2003; Fujiwara-Tsukamoto et al. 2006). In the tetanic stimulation model of epilepsy, stratum oriens and pyramidale interneurons form a positive feedback circuit with local pyramidal cells in order to generate post-tetanic discharges (Fujiwara-Tsukamoto et al. 2004) and a study investigating the complex cell-type-specific neuronal interactions in the formation of seizure patterns has revealed a dynamic pattern of excitation and inhibition during spontaneous seizures (Ziburkus et al. 2006). Using simultaneous intracellular recordings from pyramidal cells and interneurons, the study revealed an increase in interneuronal activity during the preictal period, followed by a period of spike trains observed in pyramidal cells during the ictal state.

Another study investigating the ictal state using low Mg²⁺ ACSF administered low-frequency single or paired-pulse electrical stimulation at the Schaffer collaterals to elicit ictal discharges (Kohling et al. 2000). Due to the hyperexcitable nature of the tissue in the low Mg²⁺ medium, these minimal stimulations induced high-frequency γ oscillations similar to those observed in our model in the ictal state following the preictal_d state. These high-frequency oscillations are thought to be responsible for the collapse of the chloride gradient and the excitatory action of GABA during the ictal states. Hence, in our model, the spontaneous low frequency preictal_d state replaces the low-frequency evoked electrical stimulation in previous models and serves to initiate the high-frequency ictal onset.

Preictal dynamics and clinical relevance

Understanding the dynamics and neurophysiology of the preictal epileptic state is important for the study of synchrony generation in pathophysiological systems as well as for a more pragmatic application of this knowledge to seizure prediction and control. Recently, there has been a resurgence in studies focusing on the characterization of the preictal state using non-linear mathematical tools and applying these measures to neuronal dynamical systems (Lehnertz, 1999; Litt & Lehnertz, 2002; Iasemidis, 2003; Lopes da Silva et al. 2003a; Sabesan et al. 2003; Chiu et al. 2005; Mormann et al. 2006). It is during the preictal transition state that administering a therapeutic intervention, such as electrical stimulation, may be effective in preventing seizure onset (Khosravani et al. 2003; Morrell, 2006). Our data show that complete abolishment of the GABAergic preictal state in BMI does not lead to cessation of epileptic activity in this model. Furthermore, utilizing carbonic anhydrase inhibitors does not attenuate the preictal state in our model. Our study did find that the reversal potential for IPSPs did become more depolarized, but not enough to fully explain the transition to seizures.

In our model, as long as GABAergic inhibition was intact, we always observed the sequence of preictal h → m → d before ictal onset. We propose that with a global excitatory stimulus such as low Mg²⁺ perfusion, interneurons, which have a lower firing threshold than pyramidal cells (i.e. they are naturally more depolarized than pyramidal cells) will discharge prior to the more hyperpolarized pyramidal cells. This explains the IPSP–EPSP sequence we observed during the preictal state. We further suggest that the switch from the preictal_h to the preictal_d state precipitates seizure onset. We interpret this process as an extreme manifestation of underlying normal network dynamics.