Main

Epilepsy is a dynamic disease in which ictal events, the clinical manifestation of the pathology, paroxysmally interrupt an interictal state. In electroencephalographic (EEG) recordings, interictal events are synchronous neuronal discharges with durations in the tens of milliseconds, whereas ictal events, or seizures, are patterned discharges that last for several tens of seconds. Interictal discharges reflect pathologic changes in an epileptogenic brain1,2, but it is unclear whether they trigger or prevent ictal-like activities3. The biological basis for the transition from the interictal state to a seizure, or ictogenesis, is poorly understood, although seizure onset seems to be preceded by subtle clinical, metabolic and electrical changes4.

Different mechanisms and even different neuronal populations may be involved in the genesis of interictal and ictal-like events in the epileptic temporal lobe. In vitro data suggest that interictal events are generated in the CA3 region, whereas seizures may emerge from the entorhinal cortex5,6. Interictal population bursts depend on functionally excitatory signaling5,7. Depolarizing responses to GABA may be involved in the genesis of ictal events, as several convulsants trigger strong interneuron firing and the enhanced GABA release8 could switch responses toward depolarization9,10,11. Changes in excitatory synaptic mechanisms may also be involved in the transition to seizure-like activity12,13. Several studies have suggested that the initiation, properties and propagation of population activities generated by epileptic networks are modified before ictal-like activity is initiated14,15,16.

In much of this work, ictal-like activity has been induced by applying convulsants to healthy tissue slices. This approach may miss some aspects of the disease because an epileptic brain differs from healthy tissue. Synaptic connectivity is changed, glial cells are activated, and the expression of voltage-gated and transmitter-operated channels alters. Another change is an alteration in Cl− homeostasis in tissue from individuals with pharmacoresistant temporal lobe epilepsies. The neurotransmitter GABA has depolarizing effects1, resulting in part from a reduced expression of the K-Cl co-transporter KCC2, as well as an increased or persistent activity of its counterpart, NKCC1 (ref. 2), which contribute to interictal-like activity generated by the subiculum.

We explored the transition from the interictal state to ictal-like activity in the subiculum of humans with mesial temporal lobe epilepsy. Seizures were induced in slices of this tissue by combining alkalosis or low Mg2+ with an increase in cellular excitability. During the transition to ictal-like discharges, both in slices and in situ, we detected a specific form of population activity: pre-ictal population bursts. These larger amplitude synchronous events coexisted with interictal events. They were mediated by glutamatergic signaling and were preceded by pyramidal cell firing, whereas interneuron discharges consistently preceded interictal events, which depend on both GABAergic and glutamatergic signaling. Pre-ictal discharges involved more cells and spread faster than interictal events. Their emergence just before ictal-like events raises the possibility that they may be involved in ictal initiation. Indeed, the ability of electrically induced pre-ictal discharges to trigger seizures at their onset sites further supports this hypothesis.

Results

Induction of ictal-like events in the epileptic subiculum

The subiculum generated a spontaneous interictal-like activity in all of our experiments1,2. We transformed this activity into ictal-like discharges in 95 of 121 slices from 33 of 42 subjects by combining two pro-convulsant stimuli (Fig. 1). We combined a change of extracellular K+ from 4 to 6–8 mM (n = 31 epileptic subjects) or a reduction of external Ca2+ and Mg2+ from 2 to 0.5 mM (n = 11) with either a reduction of external Mg2+ to 0.25 mM (n = 20) or an external alkalinization imposed by increasing HCO3− from 25 to 65–85 mM with a concomitant reduction in NaCl from 124 to 85–65 mM (n = 22) (Supplementary Table 1). Treatments with single convulsants, including 4-AP (n = 4), bicuculline (n = 4), high K+ (n = 4) or zero Mg2+ (n = 4), were ineffective.

Figure 1: Ictal discharges generated in the human subiculum.
figure 1

Convulsants produced patterned ictal discharges restricted to the subiculum. (a) Multiple extracellular recordings of an ictal event in a slice containing the hippocampus, subiculum and entorhinal cortex. Electrode locations: 1, dentate gyrus; 2, CA2; 3, CA1; 4–5–6, subiculum; 7, presubiculum; 8, entorhinal cortex. (b) Ictal event structure. An ictal discharge recorded in a 65 mM HCO3− and 8 mM K+ solution. Top, pre-ictal discharges (larger events, filled circles) recurred before a fast low-voltage activity (gray line) at seizure onset followed by rythmic bursts (black line). Seizure onset is indicated by the dotted arrow. Middle, time frequency representation of the local field potential (LFP). Bottom, multi-unit activity (MUA) revealed recurring PIDs of short duration followed by the rapid unpatterned action potential discharges at seizure onset and then by oscillatory bursts.

Full size image

Ictal discharges were recorded from a region of ∼2 × 8 mm in the subiculum. They propagated toward, but not into, the sclerotic CA1 region and the presubiculum (Fig. 1a). Ictal-like activity was not recorded from the dentate gyrus, CA3 region or entorhinal cortex, although multi-unit activity in these areas was increased by the convulsants.

The mean duration of ictal-like events was 20.7 ± 1.7 s (95% confidence interval, [19.9–21.5], n = 16). We found a similar duration for different convulsants: high HCO3− and high K+ (21.4 ± 9.4 s, median = 18.21, range [14.8–39.8], n = 6), high HCO3− and low Ca2+ and Mg2+ (19.8 ± 7.0 s, median = 20.1, range [12.2–27.7], n = 6), and low Mg2+ and high K+ (20.9 ± 6.7 s, median = 18.2, range [16.0–31.1], n = 4) (no significant difference, Kruskal-Wallis variance analysis, χ2(2) = 4.16, P = 0.12). Once initiated, ictal-like events recurred as long as the convulsant solution was perfused. The mean interval between them was 121.1 ± 5.2 s (95% confidence interval, [118.5–123.7], n = 15). Exclusive interictal-like activity was restored when the convulsant solution was withdrawn.

Ictal-like field potentials were similar to intracranial EEG recordings of mesial temporal lobe seizures in humans17,18 and animals19. In slices from 20 individuals, these seizures consisted of recurrent rhythmic bursts of frequency 3.7 ± 0.1 Hz (95% confidence interval, [3.6–3.7]). In slices from nine subjects, seizure onset involved an initial fast, low-voltage activity (duration = 4.1 ± 0.2 s, median = 4.4, range [1.6–7.3], n = 8; mean frequency of multiunit activity = 268.3 ± 2.6 Hz, median = 275.1, range [138.9–370.5], n = 8), which evolved into rhythmic bursts (Fig. 1b).

Ictal-like discharge pattern did not depend on the convulsant. Alkalinization was associated with rhythmic bursts in 65% of the experiments and with low-voltage fast activity in 35% of the experiments (n = 17). In low-Mg2+ experiments, rhythmic bursts were observed 75% and low-voltage fast activity 25% of the time (corrected χ2 = 0.03, P = 0.86). In tissue from three individuals, similar patterns of ictal-like activity were induced by either low Mg2+ or alkalinization, together with an increase in K+ (data not shown).

Pre-ictal discharges precede ictal-like events in vitro

Ictal-like events were preceded by recurring population events of larger amplitude than interictal discharges (IIDs; Fig. 1b). These pre-ictal discharges (PIDs) recall 'hypersynchronous' EEG events that precede intracranially recorded temporal lobe seizures17,18.

PIDs recurred 12.2 ± 1.5 s (95% confidence interval, [11.4–13.0], n = 15) before ictal-like events with a mean frequency of 2.3 ± 0.2 Hz (95% confidence interval, [2.2–2.4], n = 13). PIDs preceded seizure initiation in 7 of 7 low-voltage fast activity onset (nonparametric Spearman correlation, n = 15 seizures, ρ = −0.09, P = 0.76) and 12 of 15 rhythmic burst onset ictal patterns, recurring for a similar period at a similar frequency (n = 13 seizures, ρ = −0.60, P = 0.4). PIDs were easily distinguished from the onset of seizures with fast low-voltage activity. Changes in burst frequency and shape usually let them be distinguished from the start of ictal-like events consisting of rhythmic bursts, although their onset could be more progressive. These data indicate that ictal-like activities are generated by the subiculum and preceded by specific pre-ictal events.

In situ recording of PIDs

We searched for similar events in intracranial EEG records from the mesial temporal lobe20 of individuals with epilepsies associated with hippocampal sclerosis. Scalp recordings do not sample activity from these regions unless they propagate to the lateral neocortex. In all five of the individuals that we studied, seizures emerged from a single zone. With a 1-cm3 resolution for stereoEEG electrodes21, this zone should encompass the hippocampal formation (subiculum and head of the hippocampus), the amygdala and the entorhinal cortex. Recurring field events similar to PIDs preceded seizure onset in all nine seizures from the five subjects (Fig. 2a,b). PID-like events formed a separate population of events that were larger than IIDs (Fig. 2a,c). The mean amplitude of PIDs in vivo was 358 ± 18 μV (95% confidence interval, [355–361], n = 7,754 PIDs detected by a restrictive threshold algorithm from five in situ records), whereas that of IIDs was 140 ± 19 μV (95% confidence interval, [137–143], n = 3,7483 IIDs from five records) (Mann-Whitney, P < 0.001). The mean duration of PIDs was 203 ± 12 ms (95% confidence interval, [201–205], n = 7,754) and that of IIDs was 185 ± 6 ms (95% confidence interval, [181–189], n = 3,7483) (Mann-Whitney, P = 0.280; Fig. 2c,d).

Figure 2: IIDs, PIDs and ictal discharges in intracranial recordings.
figure 2

(a) Stereo EEG recording (electrode HipAg1) showing activity of the subiculum and head of the hippocampus at seizure onset. The seizure was preceded by recurring IIDs (blue circles) and PIDs (pink circles). It began (arrow) with fast low-voltage activity and continued with oscillatory rhythmic bursts. (b) Recordings from multiple electrode contacts on a referential montage. Traces from contacts with no epileptic activity are shown in green, those with isolated IIDs in blue, and those recording both IIDs and PIDs in pink. Left, IID sample (blue circle). Middle, PID sample (pink circle). Right, seizure onset (thick arrow). The thin arrow indicates an expanded trace from the HipAg1 contact in a. Seizure onset is highlighted in yellow. Electrodes are identified according to the recorded area (amyg, amygdala; Hip, hippocampus; TB, temporo-basal; T2, second temporal gyrus; OrFr, orbito-frontal), location (A, anterior; M, median; P, posterior; AP, antero-posterior) and hemisphere (g, left; d, right). The first contact is at the tip of the electrode and the last as it emerges from the cranium. (c) Amplitude distribution for all field potentials preceding a single ictal event by 30 min showing the distinct amplitudes of IIDs (25–125 μV) and PIDs (150–500 μV). (d) Three-dimensional reconstruction of electrode contacts from post-implantation magnetic resonance imaging showing sites where PIDs (pink), IIDs (blue) or no epileptic events (green) were recorded. An arrow highlighted in yellow indicates the site of seizure onset. The name of each electrode is shown at the site of emergence from the skull.

Full size image

In situ, PIDs were restricted to the hippocampal formation, whereas IIDs spread to adjacent temporal cortex (Fig. 2b,d). PIDs were recorded from electrodes that detected fast, low-voltage discharges during seizure initiation and were never detected at distances greater than 10 mm from seizure initiation sites. PIDs preceded all nine of the seizures that we studied from five in situ recordings. We detected 11 ± 4 PIDs in the 30 s before seizure onset (median = 10.5, range: [2–25], n = 6 seizures). Thus PID occur in the human at seizure initiating sites, especially before seizure onset.

PIDs emerge during the transition to the ictal state

PIDs emerge at the transition from the interictal to the ictal state (Fig. 3a,b). In vitro, this transition occurred over 29.1 ± 12.2 min (median = 25.5, range [13–52], n = 8 slices) after convulsant application (Fig. 3a). The delay to the first ictal event was 24.7 ± 11.5 min for increased HCO3− and high K+ (median = 25, range [13–36], n = 3), 32.0 ± 8.5 min for increased HCO3− and low Ca2+ and Mg2+ (median = 32, range [26–38], n = 2), and 31.7 ± 17.6 min for low Mg2+ and high K+ (median = 22, range [21–52], n = 3) (no significant difference, Kruskal-Wallis analysis, χ2(2) = 1.11, P = 0.57).

Figure 3: PIDs emerge during the transition to ictal-like activity in vitro.
figure 3

(a) Extracellular recordings of the transition to seizure-like activity induced by increased external HCO3− (85 mM) and K+ (8 mM). E1 and E2 are recordings from two subicular electrodes. MUA frequency (upper trace) and the extracellular signal from E2 (lower trace) are shown. (b) Amplitude measurements for all field potentials recorded by electrode E2 during the transition show the emergence of larger PIDs, whereas the amplitude of inter-ictal events did not change. (c) Dual extracellular recordings showing IIDs (open circles, left) before convulsant application and coexpression of PIDs (filled circles, right) with IIDs during the transition. (d) Amplitude distribution for all field potentials during the 35-min transition period showed IIDs of amplitude 10–50 μV and PIDs of amplitude 125–175 μV. (e) Mean and s.d. of the amplitude (black), duration (red) and propagation speed (blue) of IIDs and PIDs at steady state. Amplitudes and propagation speed, but not durations, were significantly different (line with asterisk). (f) Propagation of IIDs and PIDs. Triple extracellular recording (E1, E2, E3; E1–E2 distance, 1 mm; E2–E3 distance, 0.7 mm; E1–E3 distance, 1.4 mm) during the transition to ictal-like events. Field potential (FP) amplitude is plotted from signals from each electrode and the propagation speed of fields during the transition (between E1 and E3) is shown above. Top traces show initial IIDs (left), emerging PIDs (middle) and fully developed PIDs (right). Gray lines link FP peaks.

Full size image

During the transition to ictal activity, PIDs emerged as field potentials of progressively increasing amplitude, concurrent with IIDs (Fig. 3a–d). As population activity increased, the amplitude of PIDs increased, whereas that of IIDs remained constant (Fig. 3b). The amplitude of stable PIDs (185 ± 106 μV, 95% confidence interval, [146–196], n = 66 events from 8 slices) was significantly higher than that of simultaneous IIDs (49 ± 24 μV, 95% confidence interval, [42–55], n = 66 events from 8 slices) (Wilcoxon test, Z(7) = 3.99, P < 0.01; Fig. 3e). The duration of PIDs (83 ± 48 ms, 95% confidence interval: [69–93], n = 8 slices) was similar to that of IIDs (77 ± 34 ms, 95% confidence interval, [63–82], n = 8) (Wilcoxon test, Z(7) = 0.34, P = 0.75; Fig. 3e).

PIDs and IIDs possessed different spatial and temporal attributes. The spatial extent of PIDs, along the longitudinal axis of the subiculum, was larger than that of IIDs (Fig. 3f). The propagation velocity of PIDs was 0.33 ± 0.19 m s−1 (95% confidence interval, [0.29–0.37], n = 92 events from 3 subjects), whereas that of IIDs was 0.10 ± 0.09 m s−1 (95% confidence interval, [0.08–0.12], n = 91 events from 3 subjects) (t test, t = 10.59, P < 0.001). Thus, PIDs emerging during the transition to ictal-like discharges involve more widely distributed groups of neurons than do IIDs, produce larger fields and spread faster.

Distinct properties of PIDs and IIDs

Intracellular records from subicular pyramidal cells (Fig. 4) suggest that PIDs and IIDs depend on distinct signaling mechanisms. In records from 14 cells in which IIDs were associated with hyperpolarizing synaptic events1,2, PIDs were always accompanied by a depolarization (Fig. 4a). The absence of a hyperpolarization suggests that GABAergic signaling, necessary for IIDs1, could be less important in generating PIDs13. Indeed, the GABAA receptor antagonists bicuculline (20 μM, n = 6) and picrotoxin (50 μM, n = 3) suppressed IIDs, but not PIDs (Fig. 4b,c and Supplementary Fig. 1b). The mean PID amplitude was 310 ± 167 μV before (95% confidence interval, [121–499], n = 26 events from 3 slices) and 296 ± 123 μV (95% confidence interval, [156–435], n = 26) after applying 20 μM bicuculline. The mean PID duration was 69 ± 37 ms before (95% confidence interval, [27–112], n = 26) and 86 ± 58 ms after (95% confidence interval, [20–152], n = 26) bicuculline application. Neither parameter changed significantly (Wilcoxon test; amplitude, Z(3) = 0.57, P = 0.63; duration, Z(3) = 0.48, P = 0.68). Once established, PIDs were not affected by blocking NMDA receptors (n = 6; Fig. 4b and Supplementary Fig. 1c), but were suppressed by the AMPA receptor antagonist 6-nitro-2,3-dioxo-1,4-dihydrobenzo[f]quinoxaline-7-sulfonamide (NBQX, 10–20 μM, n = 6; Fig. 4b and Supplementary Fig. 1d).

Figure 4: PIDs depend on glutamatergic signaling.
figure 4

(a) Intracellular recording from a subicular pyramidal cell (I) with a local extracellular recording (E). This cell received hyperpolarizing synaptic inputs during IIDs (open circle) both before and after convulsants were added. In contrast, it was depolarized during PIDs (filled circle) in the presence of convulsants. Intracellular action potentials were cut. (b) IIDs and PIDs coexisted during the transition to ictal discharges induced by 0.25 mM Mg2+ and 8 mM K+ (left trace). Bicuculline (bic) blocked IIDs, but not PIDs (middle left). The NMDA receptor blocker D,L-AP5 (100 μM) did not change PIDs (middle right), but the AMPA receptor antagonist NBQX (10 μM) suppressed them (right). (c) Distinct reversal potentials for synaptic events associated with IIDs and PIDs. The amplitude of postsynaptic potentials associated with IIDs (open circles) and PIDs (filled circles) is plotted against membrane potential during the transition. The data shown are from three different cells (black, blue and red). Inset, postsynaptic potentials associated with IIDs and PIDs at different membrane potentials. Intracellular (upper) and extracellular (lower). The mean reversal potential of IIDs was −59.1 ± 3.8 mV and the estimated reversal potential of PIDs was −15.1 ± 5.9 mV. (d) PIDs were induced by increasing excitability with high K+. Field potentials revealed that PIDs (filled circle) and IIDs (open circle) coexisted in 10 mM K+.

Full size image

These data indicate that subicular circuits generate two forms of synchrony during the transition to ictal-like activity: IIDs depend on glutamatergic and GABAergic signaling, whereas PIDs appear to be glutamatergic. To confirm this hypothesis, we compared the voltage dependence of compound synaptic potentials associated with both events (Fig. 4c). Synaptic events associated with IIDs reversed at −59.1 ± 3.8 mV, whereas we estimated the reversal potential of events associated with PIDs as −15.1 ± 5.9 mV (n = 3 cells). Furthermore, distinct glutamatergic circuits seemed to become effective when cellular excitability increases. In support of this hypothesis, PIDs emerged in addition to IIDs when external K+ was increased from 4 to 10 mM. Ictal events were not triggered in high K+, suggesting that PIDs do not suffice for seizure initiation (n = 10; Fig. 4d).

Moreover, the progressive emergence of PIDs and the long delay before the first ictal event suggest that some cellular or synaptic plasticity could be involved. In the same slices, the delay to a new plateau spike frequency in entorhinal cortex, which does not generate PIDs, after applying convulsant (n = 7, 0.25 mM Mg2+ and 8 mM K+ solution) was 33.3 ± 11.5 min (median = 34.1, range [16.2–48.0], n = 7), whereas the delay to full amplitude PIDs in the subiculum was 52.2 ± 18.8 min (median = 52.1, range [27.6–73.9.2], n = 7) (n = 7, t test, t = 4.48, P < 0.01; Supplementary Fig. 2). This difference suggests that processes distinct from an increase in excitability, possibly involving NMDA-dependant plastic changes22, could be involved in the transition to ictal activity. Established PIDs were not affected by the NMDA receptor antagonist D,L(−)-2-amino-5-phosphonovaleric acid (D,L-AP5, 100 μM, n = 6, Figs. 4b and 5 and Supplementary Fig. 1c). However, we found that neither PIDs nor ictal events were generated if D,L-AP5 was applied before the convulsant (low Mg2+ and high K+, n = 5; Fig. 5a). These data suggest that the activation of NMDA receptors during population bursts is needed for the emergence, but not maintenance, of PIDs. Furthermore, initiation of ictal events appears to depend on recurring established PIDs, but requires functional GABAergic and glutamatergic signaling, as they were suppressed by antagonists at GABAA, AMPA and NMDA glutamate receptors (n = 6, 4 and 4, respectively; Supplementary Fig. 1).

Figure 5: NMDA receptor signaling is involved in seizure generation and in the emergence of PIDs, but not their maintenance.
figure 5

All extracellular recordings were made from the same site in the subiculum. (a) Left, spontaneous IIDs (open circles). D,L-AP5 did not affect IIDs (middle), but the convulsant solution (10 mM K+ / 0.25 mM Mg2+) did not induce PIDs or ictal-like events (right trace) when D,L-AP5 was present. (b) Left, after D,L-AP5 washout, the convulsant solution induced PIDs (filled circles) during the transition period (middle left) to ictal discharges (gray line, middle right). Application of D,L-AP5 after PIDs had emerged did not change them, but suppressed the initiation of ictal-like events (right).

Full size image

Synaptic networks underlying IIDs and PIDs

If different transmitter systems are involved, then different cell types might discharge as IIDs and PIDs are initiated. We examined this point using action potential shape23 to discriminate between pyramidal cells and interneurons in juxtacellular recordings24. Action potential duration in 6 of 45 recorded neurons was less than 0.7 ms and these cells were classified as interneurons (mean duration = 0.47 ± 0.17 ms, 95% confidence interval, [0.33–0.61], n = 6; Fig. 6). In 39 putative pyramidal cells, spike duration was greater than 0.7 ms (mean = 0.99 ± 0.20 ms, 95% confidence interval, [0.92–1.05], n = 39; Fig. 6a).

Figure 6: IIDs and PIDs are generated by distinct networks.
figure 6

(a,b) Firing of putative pyramidal cells (a) and interneurons (b) recorded juxtacellularly during IIDs (blue) and PIDs (red). Cell type was determined from action potential duration (upper traces) measured from the positive to the negative peak (pyramidal cells > 0.7 ms, interneurons < 0.7 ms). Spike timing for pyramidal cells and interneurons (shown as dots in the box), spike probability (histograms) and spike frequency (lower traces) are shown with respect to IID and PID field potentials (upper trace). The dotted line with an arrow tail indicates the onset of field potentials and the dotted line with an arrow head indicates its peak. Most pyramidal cells were inhibited during IIDs and fired during PIDs (left), whereas a subset of pyramidal cells (7 of 39) fired both during IIDs and before PIDs onset (right) (a). All interneurons fired before IID onset and after PID initiation (left), but some interneurons (2 of 6) fired before both IID and PID onset (b).

Full size image

We then sought to distinguish between cells that fired before PID- or IID-related field potentials, reasoning that they could be involved in their initiation, and other cells that fired during events or that did not fire at all (10 mM K+, ten subjects; Fig. 6). Most pyramidal cells were inhibited during IIDs (26 of 39), but almost all of them (37 of 39) fired during PID (χ2, P < 0.0001; Fig. 6a). A subset of pyramidal cells (7 of 39) increased firing in the 25 ms before the onset of PID field potentials (19 ± 26 Hz, 95% confidence interval, [9–28], n = 30 events from 3 cells; Fig. 6a). Six out of seven of these triggering pyramidal cells were located in the region where PIDs were initiated and only one was located in areas to which PIDs propagated. Firing during PIDs was detected for 18 of 24 cells of the focus, compared with 12 of 15 in propagation zones (Fisher's exact test, P = 0.22). None of the pyramidal cells that fired before PIDs discharged before IIDs (3 ± 5 Hz, 95% confidence interval, [1–5], n = 3 cells). In contrast, all recorded interneurons (n = 6) discharged during both IIDs1 and PIDs. All six cells fired for at least 25 ms before IID initiation (Fig. 6b), but only two interneurons fired in the 25 ms preceding PID initiation (Fig. 6b). Effective interneuron firing frequency in the 25 ms before IID onset was 77 ± 12 Hz (95% confidence interval, [72–81], n = 30 events from 3 cells) and 22 ± 13 Hz (95% confidence interval, [17–27], n = 3 cells) in the 25 ms preceding PIDs. These data suggest that interneurons may be involved in IID initiation (6 of 6 interneurons fired before IID versus 0 of 39 pyramidal cells, Fisher's exact test, P < 0.0001). The majority of pyramidal cells were silent during IIDs, but most of them fired during PIDs. Among them, a subgroup that fired systematically before PIDs may be involved in their initiation (0 of 39 pyramidal cells fired before IIDs, 7 of 39 fired before PIDs; Fisher's exact test, P = 0.01).

Dynamics and propagation of IIDs and PIDs

The timing of IIDs and PIDs suggests that the circuits generating them are somewhat separate. An IID was never initiated during a refractory period of ∼200 ms after a PID. Apart from that interaction, PID amplitude and frequency were independent of IID occurrence. Similarly, neither the amplitude nor the timing of IIDs were affected by a PID (n = 5 slices; Fig. 4a and Supplementary Fig. 3).

Dynamic analysis of field potentials generated at distinct sites during the initiation of ictal-like activity demonstrated further differences between PIDs and IIDs. We measured the time delay and amplitude and used an index of cross correlation to compare PIDs and IIDs recorded by pairs of electrodes separated by 1–2 mm in slices from six different epileptic subjects (Fig. 7). Both the amplitude and cross-correlation index were larger for PIDs than for IIDs. Furthermore, latencies between PIDs recorded at different sites were shorter and less variable than those for IIDs, which were often initiated from several foci. At the transition to ictal-like discharges, the latencies of field events recorded from separated electrodes were further reduced, indicating a faster propagation (Fig. 7c,d).

Figure 7: Dynamics of population activity during the transition to ictal events.
figure 7

(a) Dual extracellular recordings made during the interictal period (white line), the pre-ictal period (black line) and the onset of an ictal-like event (green line). Interictal events (open circle) and pre-ictal events (filled circle) during the transition are shown below on an expanded timescale. Time frequency analysis of the extracellular signal is shown below. (b) Cross-correlation index versus time lag for IIDs (empty circles) and PIDs (filled circles). (c) Three-dimensional plot of the cross-correlation index, time lag between sites, and the amplitude of the field potential for IIDs, PIDs and initial ictal discharges. (d) Histograms of the amplitude, time lag and cross-correlation index for each type of event. Error bars indicate s.d. Lines with an asterisk indicate a significant difference (P < 0.05).

Full size image

Finally, we asked whether PIDs, which recurred before seizure onset, could effectively initiate seizures. One test of this point would be to examine the effects of blocking PID on seizure occurrence, but blocking PID with AMPA receptor antagonist also blocked ictal events (Supplementary Fig. 1). We therefore initiated sequences of PIDs by focal stimulation and asked whether they reliably induced seizures shortly after their generation.

We found that single low-intensity stimuli (3–6 V, 0.05 ms) delivered near the PID initiation site initiated a similar event (n = 7 slices; Fig. 8). Attempting to mimic PID timing, we then tested the effects of 1–2-s sequences of 2–5-Hz stimuli. PID sequences induced in this manner triggered seizures in 38 of 45 trials at the PID onset site (n = 7 slices; Fig. 8b). Similar low-intensity sequences were never effective at sites of propagation (0 of 25 trials, Fisher's exact test, P < 5 × 10−13; Fig. 8c). High-intensity stimulation of remote regions (15–50 V, 0.05 ms) could initiate back-propagated PID-like events at PID onset sites. Sequences of these events also triggered ictal-like events (n = 13/20 trials in 3 slices; Fig. 8d). The duration of stimulus-induced ictal-like events was similar to that of spontaneous events (Friedman's χ2(2), P = 0.26). The pre-ictal period, defined as the time during which PIDs occurred, was shortened by stimulation-induced PIDs. For spontaneous ictal-like events, it was 10.61 ± 11.35 s (95% confidence interval, [10.34–10.88], n = 39 events, 4 slices). For events induced by stimulation at the PID initiation site, the delay was 1.97 ± 0.53 s (95% confidence interval, [1.96–1.98], n = 45 events, 4 slices) and it was longer for events induced by stimulation at follower sites, 10.39 ± 8.44 s (95% confidence interval, [10.19–10.59], n = 25 events, 4 slices) (Friedman's χ2(1), P < 0.01). The duration of the pre-ictal period, normalized to that of spontaneous ictal events, was shorter for stimulation at PID foci than at distant sites (0.28 ± 0.13 versus 1.09 ± 0.5, Friedman's χ2(1), P < 0.01). These data indicate that the site of PID initiation is a privileged zone for seizure initiation by stimuli that mimic clusters of PIDs.

Figure 8: Repeated PIDs trigger seizure-like events.
figure 8

(a) Dual extracellular recordings of an ictal event preceded by PIDs. The lower black trace was recorded at the site of onset of PIDs and ictal events. The upper gray record was made from a follower region. Right, expanded traces from the period indicated with the asterisk. (b) Electrical stimulation (2 Hz, 2 s, black sign) near the site of PID initiation elicited PIDs (expanded at right), which induced a seizure-like event. (c) Identical electrical stimulation (gray sign) in a region of PID propagation did not trigger seizure-like events. (d) High-intensity bipolar electrical stimulation (large gray sign) in the region of PID propagation generated PIDs at their initiation site. A seizure-like event followed. The double-headed arrow indicates the delay between stimulation and back-propagated PID onset. (e) Probability of triggering an ictal-like event by moderate intensity stimuli at the PID focus (black), at a follower site (white) or by strong stimuli at the follower site generating back-propagated PIDs (gray). *P < 10−6, **P < 10−12.

Full size image

Discussion

Our data suggest that the subiculum generates ictal-like activity in tissues from individuals with temporal lobe epilepsies. Seizures were preceded by a specific form of population activity: pre-ictal discharges emerged during the transition-to-seizure period (a period in the tens of minutes) and further triggered seizures by recurring in the seconds preceding their onset. Similar events are present at the seizure onset zone in intracranial recordings made from people with temporal lobe epilepsy. During ictogenesis, PIDs occur contemporaneously with IIDs. In situ, they are restricted to sites of seizure initiation, whereas IIDs are more widespread. PIDs are associated with larger field potentials than IIDs, suggesting that they involve more neurons, and they propagate more quickly. They depend on glutamatergic rather than the mixed depolarizing GABA/glutamatergic mechanisms underlying IIDs and are preceded by pyramidal cell firing rather than by interneuron firing. This population activity may be a crucial feature of ictogenesis and helps to define both the spatial and temporal aspects of seizure initiation.

We found in vitro that different convulsant treatments induced seizure-like activities similar to intracranial recordings of ictal events in slices from the same individuals. Ictal-like activities sometimes included an early period of fast low-voltage oscillations or consisted of rhythmic population bursts. Both patterns were preceded by PIDs. From in situ data, it has been suggested that these distinct patterns depend on different contributions from hippocampus and cortex25, but our recordings indicated that an isolated subiculum supports both patterns. Although intracranial macro-electrodes may not possess sufficient spatial resolution to distinguish entorhinal cortex and subicular activity21, microelectrode recordings suggest that at least interictal activities are generated in the subiculum26,27. Our findings indicate that the subiculum alone generates ictal-like activity. In previous slice work on human tissue, ictal activities were also induced in the dentate gyrus by high K+ (refs. 28,29) and in neocortical dysplasias by 4-AP30.

In vitro, PIDs emerge during a long transition period before ictal-like events are initiated and recur in the seconds preceding seizure onset. Intracranial EEG recordings have also revealed large-amplitude field potentials just before a seizure31,32. We found that PIDs were restricted to electrodes sampling the seizure onset zone. Similar field potentials of larger amplitude precede ictal events in several in vitro models of epilepsy6,14,33,34,35. Moreover, intracranial recordings found that PID-like events preceded seizures in multiple focal human epilepsies originating in temporal17 or extra temporal36 hippocampal17,18 or neocortical37 regions, and associated with sclerosis17,38, malformation39, dysplasia or were found in macroscopically normal tissue36.

PIDs and IIDs differ pharmacologically. IIDs depend on both glutamatergic and depolarizing GABAergic signaling1 and our data suggest that interneuron firing reliably preceded IID initiation (Fig. 6). In contrast, PIDs apparently depend on glutamatergic synapses. PIDs were blocked by AMPA receptor antagonists, reversed at depolarized potentials and were preceded by discharges in a subset of pyramidal cells near their initiation site40. Presumably, recurrent excitatory synapses between subicular pyramidal cells are responsible41. The existence of distinct groups of initiating cells may explain both the contemporaneous generation of PIDs and IIDs and the weak temporal interactions between them (Supplementary Fig. 3). A glutamatergic basis does not exclude an interneuronal participation in PIDs. However, both the voltage dependence of synaptic events (Fig. 4) and the timing of interneuron firing (Fig. 6) suggest that GABAergic systems are secondarily activated compared with IIDs. A reduced inhibitory restraint could account for the differences in propagation velocity42 and size of the neuronal population involved in PIDs and IIDs (Fig. 3). Excitatory field events similar to PIDs have been detected just before seizures in animal models6,8,15,43,44. PIDs do not involve barrages of hyperpolarizing synaptic events, which may act as an inhibitory restraint to protect against the spread of seizure-like activities45. The PIDs described here differ in that they were initiated at the site of seizure onset in epileptic subiculum rather than before events spreading through healthy cortex in the presence of convulsants. PIDs seem to depend on firing in principal cells with a late interneuron discharge, in contrast with the early interneuron firing that generates a strong feedforward inhibition to restrain seizure spread.

Our results indicate that PIDs precede ictal events in vitro and that their spatial distribution in situ is aligned with seizure initiating regions. Our data suggest that a NMDA-dependant, cellular or synaptic plasticity may underly PID emergence during the transition to seizure. Furthermore, stimulated PID-like events can initiate ictal-like discharges at their onset site. We hypothesize that seizures are triggered by repeated PIDs, involving larger groups of neurons than IIDs, which initiate changes in glutamatergic or GABAergic signaling. One mechanism could be a depolarizing shift in the reversal potential for Cl−-mediated GABAergic events10. We found that GABAergic signaling is needed for ictal-like events, even if PIDs are resistant. Alternatively, or in conjunction, repeated PIDs may enhance synaptic excitation12,13 sufficiently to recruit neurons to an ictal event.

We have described and dissected a previously unknown epileptiform activity in human epileptic tissue in vivo and in vitro. In the subiculum, emerging PIDs coexist with IIDs during the transition to seizure and, once established, have a role in seizure initiation. These epileptiform transient events are generated by distinct networks. IIDs are the expression of a basal epileptic defect resulting from epileptogenic processes. PIDs, in contrast, are dynamically generated during the transition period and contribute to ictogenesis at the seizure focus.

Methods

Epileptic tissue.

Temporal lobe tissue blocks containing the hippocampus, subiculum and part of the entorhinal cortex were obtained after operations on 45 people with pharmacoresistant mesial temporal lobe epilepsies associated with hippocampal sclerosis (age 18–52 years, seizures for 3–35 years). Surgery consisted of resecting the amygdala, hippocampus, the anterior parahippocampal gyrus, and sometimes the temporal pole, anterior basal and lateral neocortex. All of the individuals gave their written, informed consent and the Comité Consultatif National d'Ethique approved our protocol.

Stereo EEG recordings were performed with depth electrodes to assess seizure onset areas in five subjects before epilepsy surgery (Ad-Tech Medical Instruments). Signals from electrodes, of diameter 1 mm, with 4–8 contacts, 1 mm wide, 2.3 mm long and 10 mm apart, were recorded with an EEG video system (Nicolet 6000, Viasys-Nicolet), sampling 63 iEEG channels at 400 Hz (bandwidth 0.05–150 Hz). Electrode positions were derived from postoperative magnetic resonance imaging artifacts. Recordings were analyzed with Spikoscope and EEG View/LTM software (Nicolet). Seizures and the 3–5 min preceding them were analyzed.

Tissue preparation.

Post-surgical tissue was transported in a cold, oxygenated solution containing 248 mM D-sucrose, 26 mM NaHCO3, 1 mM KCl, 1 mM CaCl2, 10 mM MgCl2 and 10 mM D-glucose, equilibrated with 5% CO2 in 95% O2. Hippocampal-subicular-entorhinal cortex slices or isolated subicular slices (400 μm thick) were cut with a vibratome (HM650V, Microm). They were maintained at 37 °C, equilibrated with 5% CO2 in 95% O2, in an interface chamber perfused with a solution containing 124 mM NaCl, 26 mM NaHCO3, 4 mM KCl, 2 mM MgCl2, 2 mM CaCl2 and 10 mM D-glucose. NBQX and D,L-AP5 were used to block glutamatergic signaling and bicuculline or picrotoxin were used to block GABAA receptors. Ictal-like activity was induced either by increasing NaHCO3 to 65–85 mM with a concomitant reduction in NaCl from 124 to 85–65 mM or by reducing Mg2+ to 0.25 mM. These changes were coupled with an increase in external K+ to 6–8 mM or, for the high HCO3− and low Cl− case, a decrease of Mg2+ and Ca2+ to 0.5 mM. A stable expression of both IIDs and PIDs was induced using a 10 mM K+ solution.

Recordings.

Up to four tungsten electrodes etched to a tip diameter of ∼5 μm46 were used for extracellular recording. Signals were amplified 1,000-fold and filtered to pass frequencies of 0.1 Hz to 10 kHz (AM systems, 1700). Intracellular records were made with glass microelectrodes containing 2 M potassium acetate and beveled to a resistance of 50–100 MΩ. Signals were amplified with an Axoclamp 2B amplifier in current-clamp mode2. Juxtacellular records24 were made with glass micro-electrodes of resistance 3–5 MΩ filled with the perfusion solution. Intracellular and extracellular signals were digitized at 10 kHz with a 12–bit, 16-channel A-D converter (Digidata 1200A, Axon Instruments), and monitored and saved to a PC with Axoscope (Axon Instruments).

Data analysis.

Recordings were analyzed with Clampfit (Axon Instruments), Spikoscope (http://spikoscope.sourceforge.net/) software and routines written in LabView (National Instruments), Spikoscope and Matlab (MathWorks). Extracellular recordings were filtered post hoc to display action potentials or field events46. Field potential events, including interictal and pre-ictal events, were detected with an 'up-only' algorithm and a user-defined threshold.

The dynamics of changes in extracellular signals at the transition to ictal-like discharges were computed from normalized cross-correlations between extracellular records using the fast Fourier transform (1,024 points Hanning window, mean detrending). Synchronization dynamics was estimated with a sliding cross-correlogram (moving window, 350 ms; overlap, 345 ms) or a fixed window cross-correlation (window, 350 ms). Latencies between pairs of records were calculated from the peak offset of the cross-correlation function.

Extracellular recordings were low-pass filtered (500 Hz, zero-phase shift) to detect changes in field potentials during the transition to ictal-like activity. The amplitude and timing of all detected field events was measured and independent time series were constructed for IIDs and PIDs after a two-threshold separation based on amplitude. The timing of PIDs measured at different periods during the transition was used as a reference. Typically, 5–15 IIDs occurring before and after PIDs were plotted against amplitude and inter-event interval.

Juxtacellularly recorded action potentials were correlated with a local field potential. Cells firing juxtracellular spikes with less than 0.7 ms between their positive and negative peaks47,48,49 were presumed to be interneurons. Spikes were detected using a threshold detection algorithm (>5:1 signal to noise ratio). Temporal relations between single spikes and IIDs (IIDs or PIDs) were examined by aligning all individual field discharges on their maxima and constructing discharge-triggered spike histograms. The instantaneous spike frequency was measured by convolving the timing of each detected event with a Gaussian function with a 20 ms s.d. A significant change in spike frequency was defined as an absolute fluctuation over 3 s.d. of the baseline firing and was defined relative to the time of field maxima. Time-frequency analyses were performed using a wavelet transform50.

Statistics.

JMP 7.2 software (SAS Institute) was used and results were expressed as mean ± s.d. with significance of P < 0.05. Exact P values are given unless P < 0.0001. Distributions were described by 95% confidence intervals or median and range values when n < 10. Fisher's exact test was used to compare small samples. Nonparametric Kruskal-Wallis ANOVA was used to compare non-homogeneously distributed data. Then, post hoc nonparametric Mann-Whitney rank-sum test was performed to determine differences between groups. For related samples, the nonparametric Friedmann test was used. ANOVA on ranks and the nonparametric Wilcoxon signed rank test were used for repeated measures.