Spatiotemporal neuronal correlates of seizure generation in focal epilepsy

Summary

Purpose:  Focal seizures are thought to reflect simultaneous activation of a large population of neurons within a discrete region of pathologic brain. Resective surgery targeting this focus is an effective treatment in carefully selected patients, but not all. Although in vivo recordings of single-neuron (i.e., “unit”) activity in patients with epilepsy have a long history, no studies have examined long-term firing rates leading into seizures and the spatial relationship of unit activity with respect to the seizure-onset zone.

Methods:  Microelectrode arrays recorded action potentials from neurons in mesial temporal structures (often including contralateral mesial temporal structures) in seven patients with mesial temporal lobe epilepsy.

Key Findings:  Only 7.6% of microelectrode recordings showed increased firing rates before seizure onset and only 32.4% of microelectrodes showed any seizure-related activity changes. Surprisingly, firing rates on the majority of microelectrodes (67.6%) did not change throughout the seizure, including some microelectrodes located within the seizure-onset zone. Furthermore, changes in firing rate before and at seizure onset were observed on microelectrodes located outside the seizure-onset zone and even in contralateral mesial temporal lobe. These early changes varied from seizure to seizure, demonstrating the heterogeneity of ensemble activity underlying the generation of focal seizures. Increased neuronal synchrony was primarily observed only following seizure onset.

Significance:  These results suggest that cellular correlates of seizure initiation and sustained ictal discharge in mesial temporal lobe epilepsy involve a small subset of the neurons within and outside the seizure-onset zone. These results further suggest that the “epileptic ensemble or network” responsible for seizure generation are more complex and heterogeneous than previously thought and that future studies may find mechanistic insights and therapeutic treatments outside the clinical seizure-onset zone.

The guiding assumption of seizure genesis for the last 50 years has been that seizures initiate without warning when most or all neurons in an epileptic focus begin to fire synchronously (Penfield & Jasper, 1954; Kandel et al., 1991). Data collected over several decades have noted seizure-related activation of neuronal spiking both inside and outside the epileptic focus determined by the macroscopic local field-potential seizure discharge (Rayport & Waller, 1967; Verzeano et al., 1971; Wyler et al., 1982; Colder et al., 1996), yet the cellular assemblies comprising the seizure-onset zone (SOZ) and the cellular dynamics underlying seizure generation remain poorly understood. In part, this may due to the limited nature of the available data: It remains a technical challenge to obtain high fidelity microwire recordings over the time span of days required to capture spontaneous seizures (Brinkmann et al., 2009), and intraoperative recordings typically capture few spontaneous seizures and are confounded by the effects of anesthesia.

Over the past decade, the acquisition, storage, and analysis of long duration, wide bandwidth recordings from microelectrode arrays have become possible (Buzsaki, 2004; Bower et al., 2009; Brinkmann et al., 2009), and have led to the discovery of transient, pathologic, local field potential (LFP) oscillations in human epileptic brain, including high frequency oscillations (HFOs) (Bragin et al., 1999; Worrell et al., 2008; Schevon et al., 2009) and microseizures (Schevon et al., 2008; Stead et al., 2010), suggesting that wide bandwidth recordings may contain additional, clinically useful information (Engel et al., 2009).

In addition to LFPs, it is now possible to obtain long-term, continuous, recordings from populations of neurons and to explore the cellular correlates of spontaneous seizure generation over long timescales and across large regions of brain tissue (Buzsaki, 2004; Brinkmann et al., 2009). When recording at the spatial scale of clinical macroelectrodes (1–10 mm²) the characteristics of ictal LFP discharges are often highly reproducible from seizure to seizure. The spatially reproducible nature of focal seizures in many patients with medically resistant partial epilepsy is the basis for successful epilepsy surgery (Engel, 1996). However, unlike the stereotypical spatial patterns, the temporal dynamics of the interictal to ictal transition are more complicated. When recording from tissue volumes sampled by clinical macroelectrodes, ictal LFPs appear to begin suddenly and without warning. It is generally assumed that the neuronal correlate of seizure onset is a paroxysmal, widespread, synchronous discharge of a large population of neurons, but there is little direct evidence to support that seizures begin this way on a cellular level.

The first analyses of single-unit activity during spontaneous seizures in humans with focal epilepsy found heterogeneous neuronal firing patterns at the onset of spontaneous seizures (Babb & Crandall, 1976; Babb et al., 1987). Although finding a heterogeneous range of activity patterns as predicted by the “epileptic neuron” hypothesis (Wyler et al., 1982), they did not find fixed, pacemaker, or “epileptic” neurons that initiated each recorded seizure in a patient. These studies (while ahead of their time) were hampered by the technical capabilities available at that time, and focused on cellular activity at seizure onset and propagation. They were not able to rigorously address the cellular and network ensemble dynamics on longer timescales, for example, minutes to hours, leading into spontaneous seizure. Remarkably they nonetheless found that relatively few neurons within the SOZ were actually involved in the seizure discharge, only 7% at seizure onset and no more than 40% of neurons recorded became activated even as the seizure evolved. Recently, Truccolo et al. (2011) investigated changes in cellular activity during the interictal-to-ictal transition by recording from a small (4 × 4 mm²), localized region of human epileptic neocortex using a penetrating microelectrode array (Donoghue, 2002). Similar to previous studies, they found a heterogeneous and sparse activation of neurons during seizures (Babb & Crandall, 1976; Babb et al., 1987) and found that neurons outside the SOZ were also activated during seizures (Wyler et al., 1982). Unlike previous studies, however, they found that recurrent seizures from individual patients were stereotypical at the level of cellular recordings, with characteristic “motifs” of neuronal activations leading up to and persisting through seizures. Determining whether such neuronal firing motifs occur in mesial temporal lobe seizures and reflect the initiation of seizures requires long-term (multihour) recordings over a broad anatomic reach.

Enlarging the spatiotemporal scale of recordings can also help address the consistent, but still surprising result shared by all studies of human neural recordings in epilepsy: that the majority of neurons do not alter their firing rates prior to, or even in response to, ongoing epileptiform LFP activity in the SOZ or surrounding tissue. The observation of sparse neuronal responses before and during interictal and ictal epileptiform discharges could reflect a distributed, “epileptic network,” but it could also reflect limited spatiotemporal sampling that simply misses seizure-related activation of a fixed population of “epileptic neurons.” Differentiating between these hypotheses could guide not only studies of the mechanisms of seizure generation, but also strategies for seizure prediction and therapeutic intervention.

Materials and Methods

Data acquisition

Patients undergoing chronic intracranial EEG (iEEG) monitoring for drug-resistant, mesial temporal lobe epilepsy (MTLE) were enrolled after written informed consent was obtained in this Mayo Clinic Internal Review Board–approved research protocol. Depth electrodes were implanted stereotactically into hippocampal structures as described previously (Worrell et al., 2008). Briefly, hybrid depth Pt-Ir electrodes (1.3 mm diameter, polyurethane; AD-Tech Medical Instrument Corporation, Racine, WI, U.S.A.) containing both clinical macroelectrodes (four or eight contacts, 2.3 mm diameter, 5 or 10 mm spacing, 200–500 Ω) and research microelectrodes (9 or 18 oriented radially on the shaft between macro contacts and a bundle of 9 extending from the tip, 40 μm diameter, 500–1,000 kΩ) were stereotactically placed into the medial temporal lobe using an occipital or lateral approach (Van Gompel et al., 2010). High-resolution (1.5 or 3 Tesla) T₁-weighted, volumetric magnetic resonance imaging (MRI) brain scans obtained before implantation surgery were combined with high-resolution, 64-slice computed tomography (CT) head scans obtained following implantation surgery, to localize macroelectrodes within the hippocampus. The location of microwire electrodes was inferred from their known relationship to clinical macroelectrodes. Local field potential (LFP) dynamics of focal seizure generation were obtained via continuous, wideband (near DC – 9 kHz, sampling at 32 kHz, 1 μV resolution) electrophysiologic recordings using hybrid depth electrode arrays composed of clinical macroelectrodes and research microelectrode arrays. Data were acquired on a Digital Lynx system (Neuralynx Inc., Bozeman, MT, U.S.A.), transmitted on a private Ethernet line and stored for offline analysis (Bower et al., 2009; Brinkmann et al., 2009).

Determination of seizure onset

The seizure-onset time and seizure-onset zone (SOZ) were determined using standard, clinical practice of expert visual review of macroelectrode recordings (MS and GAW). Candidate seizure events on continuous iEEG recordings from both clinical macroelectrodes and research microelectrodes were identified with an automated detector using signal line-length thresholded for hypersensitive detection, combined with clinical notes and patient event reporting, and were ultimately confirmed by visual review of the iEEG recordings. Candidate events were subjected to post hoc, expert visual review to verify or reject detected events, with blinding to all channel and clinical information (Wilson et al., 2003; Gardner et al., 2007). The visual verification of a seizure discharge was based on the electrographic features of seizures (Schiller et al., 1998; Worrell et al., 2004; Bragin et al., 2007): (1) Paroxysmal change in the background iEEG activity with a characteristic spectral seizure-onset pattern, (2) temporal and spectral evolution of the seizure discharge, and (3) termination of seizure discharge. The clinical SOZ was defined by the clinical macroelectrodes that showed the earliest iEEG change at electrographic seizure onset.

After seizure onset was determined based on macroelectrode recordings, microelectrodes were grouped according to their position relative to the SOZ. Microelectrodes located between macroelectrodes in the SOZ were labeled “SOZ,” those on the same side of the brain, but outside the SOZ were labeled “ipsilateral,” and microelectrodes in the mesial temporal lobe contralateral to the SOZ were labeled “contralateral.”

Detection of action potentials

The quality of recordings was established first by visual review to identify “noisy” electrodes, which were dropped from further analysis. Data were extracted from each microelectrode and were analyzed offline to detect single neuron extracellular action potentials (i.e., to detect “spikes”). Continuous recordings (Fig. 1A) were filtered offline between 600 and 10,000 Hz using zero-phase-shift digital filtering. Action potentials were detected by applying a series of both absolute and relative thresholds in both voltage and time (Fig. 1B), as is commonly done in the analysis of extracellularly recorded unit activity (Harris et al., 2000). We have previously used this approach to detect action potentials during the interictal-to-ictal transition in a rodent model of epilepsy (Bower & Buckmaster, 2008). In the following description, negative voltage is “down,” so that the “peak” of an action potential is in the direction of negative voltage (i.e., graphically “down”) and the “valley” is in the positive direction (i.e., graphically “up”). Root mean square variation (“sigma”) was computed for 1 min windows and candidate action potentials identified as peak magnitudes greater than three sigma below the mean. From the most negative peak, candidate spikes increased from at least one sigma more positive than the peak value within 120 μs, increased at least 1.5 sigma from the peak within 150 μs, and reached a plateau (“valley”) within 500 μs. Finally, the peak-valley magnitude had to be >20 μV. The timestamps and waveforms of candidate spikes that satisfied these criteria were then stored to a MySQL database (http://www.mysql.com). Because these action potentials were not subsequently attributed to individual neurons, these action potentials were assumed to arise from multiple neurons (multiunit activity, or MUA, Fig. 1C).

Figure 1.  Long-duration, high-frequency recording of multiunit activity (MUA). (A) Fifty minutes of continuous data recorded from macroelectrodes and microelectrodes (depth electrode schematic shown in inset, described in Worrell et al., 2008) in hippocampus beginning 45 min prior to seizure onset (“0”). The hybrid depth electrodes are composed of four or eight clinical macroelectrodes (blue), nine microwires (black) exiting the tip of the depth, and 18 microwires arranged between the clinical macroelectrodes along the depth electrode shaft. (B) Action potentials were detected using a series of both relative and absolute thresholds in both voltage and time. Candidate action potentials had to satisfy each of the six criteria, or else they were rejected: a. peak >three standard deviations (“three sigma”), b. keep largest value, c. rise to peak >1 sigma, d. fall from peak >1.5 sigma, e. peak-valley time between 0.03 and 0.5 ms, f. |peak-valley| >20 μV. (C) Expanded view showing 5 s of filtered (600–10,000 Hz) microelectrode data (C1) 45 min prior to seizure onset with action potentials marked by “.” and an example shown at right (C2) seconds prior to seizure onset with a rejected action potential marked by “x”, each accepted action potential detection marked (“.”) and an example shown at right. Example action potentials are shown at right on a 5 msec timescale. (D) Firing rate for the time period shown in panel A for action potentials shown in panel B. The activity on this microelectrode was classified as “preictal increase.”

Binning, perievent time histograms, and synchrony

Seizures selected for analysis were preceded by at least 2 h without seizures to avoid postictal changes that might confound the results. Action potentials were binned in 1 min, 30 s, and 10 s bins centered on seizure onset to generate peri-event time histograms (PETHs) extending from 45 min prior to 15 min following seizure onset. Bin size did not affect results (see Table 2), so 1-min bins were chosen for all subsequent analyses. To control for differences in firing rate, firing rates were normalized to the period 45–30 min prior to seizure onset, which was called the “baseline” period (Fig. 1D). Significant changes in firing rate were detected by repeated-measures analysis of variance (ANOVA) with Dunn-Sidak correction (Sokal & Rohlf, 1995), comparing bins preceding seizure onset to firing rates observed during the baseline period.

Synchrony between action potentials on different electrodes was computed using cross-correlations (de la Rocha et al., 2007) for 10-ms bins over 1 s and averaged for 1 min, 30 s, and 10 s bins centered on seizure onset. Cross-correlations involving MUA are known to give larger values than correlations involving independent sources (Aertsen et al., 1989), so only relative comparisons were considered to be reliable. As with firing rates, significance was computed using repeated-measures ANOVA with Dunn-Sidak correction.

Results

Twelve spontaneous seizures were recorded from seven patients (N = 2, 2, 1, 2, 1, 2, 2) undergoing intracranial monitoring for epilepsy surgery for medically refractory seizures arising from the mesial temporal lobe whose data contained multiple, low-noise microelectrode recordings. Only data from seizures separated by at least 2 h of seizure-free activity were analyzed (Table 1). Of these, five patients had electrodes implanted bilaterally to determine the hemisphere in which seizures initiated. Four of the seven patients had resective surgery based on these recordings. The remaining three patients had seizures originating in both hippocampi and were not candidates for temporal lobectomy. In all seven patients, the combination of strip and depth electrodes clearly demarcated a hippocampal origin for each seizure. Recordings were obtained from a total of 207 microelectrodes (N = 33, 33, 27, 33, 23, 62, 63) located on depth electrodes implanted in the mesial temporal lobe.

Table 1.   Detailed patient information

Patient no.	Intracranial electrodes	Interictal EEG location	Seizures (anatomic location of IOZ)	MRI	Surgery, pathology, and outcome
Posterior approach depth electrode: 8 macrocontacts and 18 microwires. Lateral approach depth electrode: 4 macrocontacts and 20 microwires. RTD, right temporal depth; LTD, left temporal depth; AD, Anterior depth; MD, middle depth; PD, posterior depth. ATL, anterior temporal lobectomy; AMD, amygdala; HC, hippocampus; EC, entorhinal cortex; MTS, mesial temporal sclerosis; IOZ, ictal onset zone; nIOZ, nonictal onset zone; μSz, microseizure; μPED, microperiodic epileptiform discharge; ILAE, International League Against Epilepsy outcome score (20).
1 Age onset: 12 years Age: 58 years RF: febrile seizures	Right and left temporal Depths: RTD and LTD 8-contact hybrid	Bitemporal: LTD 1, 2, 3, 4 RTD 1, 2	L. temporal seizures LTD 4, 5 (HC) R. temporal seizures RTD 1, 2 (AMD and HC)	Bilateral HC Atrophy L>R	Resective surgery: no Path: N/A Outcome: N/A
2 Age onset: 13 years Age: 37 years RF: none	Left temporal Strips: 3 neocortical 8- contact strips (superior, middle and inferior temporal gyrus) Depths: AD and PD hybrids	Left temporal: AD 1, 2, 3, 4 PD 1, 2, 3	#1 left temporal: AD 1, 2 PD 1, 2	Normal	Resective surgery: L. ATL Path: gliosis Outcome: ILAE-1 F/u = 18 months
3 Age onset: 3 years Age: 27 years RF: bacterial meningitis 3 years	Right and left temporal Depths: RTD and LTD 8-contact hybrid	N/a	R. temporal seizures RTD 1, 2 (AMD) L. temporal seizures LTD 1, 2 (HC)	Bilateral HC Atrophy (L>R)	Resective surgery: no Path: N/A Outcome: N/A
4 Age onset: 38 years Age: 47 years RF: none	Right and left temporal Depths: RTD and LTD 8-contact hybrid	Bitemporal: LTD 1, 2 RTD 1, 2	Left temporal seizures LTD 1, 2 (AMD and AHC) Right temporal seizures RTD 1, 2, 3, 4 (AMD and AHC)	Normal	Resective surgery: no Path: N/A Outcome: N/A
5 Age onset: 15 years Age: 22 years RF: febrile convulsion 9 months	Right and left temporal Depths: RTD and LTD 8-contact hybrid	Bitemporal: LTD 1, 2, 3, 4 RTD 1, 2	L. temporal seizures LTD 1, 2 (AHC)	Bilateral HC Atrophy and T2 signal L>R	Resective surgery: L. ATL Path: MTS Outcome: ILAE-1 F/u = 18 months
6 Age onset: 22 years Age: 58 years RF: closed head injury with convulsion at age 13 y	Left temporal Strips: three 1 × 8 strips (superior(LSS), middle (LMS) and inferior (LIS) temporal gyri) Depths: AD and PD 4-contact hybrids	L. temporal: LMS 2 AD 1, 2, 3, 4 PD 1, 2, 3	L. temporal: AD 1, 2 PD 1, 2 LMS 2, 3	Normal	Resective surgery: L. ATL Path: subpial gliosis Outcome: ILAE-1 F/u = 24 months
7 Age onset: 12 years Age: 21 years RF: none	Left temporal Grid: 24 contact (4 × 6) Strip: 1 × 8 anterior temporal (ATS) Depths: AD, MD and PD 4-contact hybrids	L. temporal: ATS 1, 2, 3, 4 AD 1, 2 MD 1, 2	L. temporal seizures AD 1, 2 (AMD) MD 1, 2 (AHC)	Normal	Resective surgery: L. temporal corticectomy. Path: subpial gliosis Outcome: ILAE-1 F/u = 28 months

Forty-three microelectrodes were located in the SOZ, 85 microelectrodes were located in the same mesial temporal lobe, but not within the SOZ, and 79 microelectrodes were located contralateral to the SOZ. Wide bandwidth recordings from microelectrodes interposed between clinical macroelectrodes (Worrell et al., 2008; Stead et al., 2010) allowed the detection of action potentials from multiple, individual neurons. Spike waveforms and firing rate characteristics were computed and used to identify microelectrodes that contained MUA. Recording sessions of five patients contained two seizures and the microelectrode recordings during each seizure (here defined as a “trace”) were examined independently, producing a total of 436 ictal “traces” available for analysis. Visual review (GAW) identified 38 traces that contained too much noise and were excluded from further analysis. One hundred thirty-six traces had an average detection rate of <0.01 Hz and were dropped from future analysis, because neuron activity was too low, either because these electrodes were located in white matter tracts or because of local tissue damage. Traces with sufficiently high firing rates to allow analysis (N = 262) remained stable, both in terms of firing rate and waveform shape (Table 2). Eighty-two traces came from microelectrodes located in the SOZ, 71 traces came from the ipsilateral mesial temporal lobe, but outside the SOZ, and 109 traces came from microelectrodes located in the mesial temporal lobe contralateral to the SOZ.

Table 2.   Percentage of electrodes providing seizure-related information.

Count	Patient
	1	2	3	4	5	6	7	All
For each patient, “total” describes the number of electrodes times the number of recorded seizures. Visual inspection identified “good” quality recordings. Following detection of action potentials, electrodes with sufficient firing “rate” were grouped into seizure-related classes: “pre_inc”/”pre_dec” recordings with significantly increased/decreased firing prior to seizure onset and “ict_inc” recordings with significantly increased firing following seizure onset. Seizure-related changes in activity were computed using three different bin sizes: 1 min, 30 s, and 10 s and the results for each bin size are arranged in the table as 1-min/30 s/10 s.
Total	66	66	27	66	23	62	126	436
Good	63	66	26	50	19	58	116	398
Rate	46	58	13	40	9	40	56	262
pre_inc	3/3/2	1/1/1	3/4/4	2/2/2	3/1/0	2/1/1	6/7/7	20/19/17   (7.6%)
pre_dec	0/0/0	2/1/0	1/0/0	14/15/12	1/1/0	0/0/0	7/8/7	25/25/19  (9.5%)
ict_inc	5/0/0	28/22/11	0/0/1	1/5/0	5/6/5	1/4/2	0/2/3	40/39/22 (15.3%)

Only 32.4% of all traces showed seizure-related changes in action potential detection rate prior to or following the seizure onset, demonstrating that the majority of neurons were unaffected by ongoing seizures recorded from surrounding tissue. Four classes of seizure-related changes in neuron firing rate were observed (Fig. 2): (1) those that showed increased firing rates prior to seizure onset (“preictal increase,” N = 20, 7.6%), (2) those with decreased detection rates prior to seizure onset (“preictal decrease,” N = 25, 9.5%), (3) those with increased detection rates immediately following seizure onset (“ictal,” N = 40, 15.3%), and (4) those with a constant firing rate prior to and following the seizure (“unchanged,” N = 177, 67.6%). One hundred sixty traces came from microelectrodes located on the shaft of the depth electrode and 102 traces came from microelectrodes extending from the tip of the bundle, but seizure-related changes in activity were observed as frequently on microelectrodes located on the shaft (N = 56 of 160, 35.0%) of the depth electrode as on microelectrodes extending from the tip (N = 29 of 102, 28.4%) of the depth electrode (chi-square, 2 × 2 contingency, p = 0.27). Considering the locations of microelectrodes and the relative percentage of electrodes in each classification group (Fig. 3), more “preictal increase” and “ictal increase” microelectrodes were observed in the SOZ and more “preictal decrease” microelectrodes were observed in the ipsilateral mesial temporal lobe than would be expected by chance (p < 0.001, chi-square). Surprisingly, seizure-related activity of all three types was observed not just within and outside the SOZ, but also in the hippocampus contralateral to where seizures began. Mapping of changes in MUA during the interictal-ictal transition to the electrode locations during a single seizure in one patient (Fig. 4) shows that seizure-related changes in activity appear to be distributed along the length of the depth electrode with some microelectrodes in the SOZ not responding to the seizure, whereas the largest change relative to interictal activity was observed in one of the more distal microelectrodes.

Figure 2.  Seizure-related activity observed for all recorded data. Each row shows a perievent (seizure) time histogram (PETH) with 1-min bin width for one microelectrode around one seizure (i.e., a “trace”) beginning 45 min prior to seizure onset. Grayscale is normalized to firing during 15 min beginning 45 min prior to onset ranging from black (no firing) to white (twice the baseline firing rate or greater). Traces from the SOZ are at top, ipsilateral MTL outside SOZ middle, and contralateral MTL bottom. Rows are grouped by seizure-related firing rate changes: hollow arrows for preictal changes, solid arrows for postictal changes, and “NC” for “no change.”

Figure 3.  All types of seizure-related firing rate changes were observed in all recorded brain regions. For each activity-related group, grey bars show the expected relative ratios by each anatomic grouping (i.e., grey bars in each group sum to 100%). Colored bars placed on top show the actual percentage of contralateral (blue), ipsilateral outside SOZ (green), and SOZ (red) relative ratios. Although SOZ microelectrodes predominate in the preictal-increase and ictal-increase groups and ipsilateral microelectrodes dominate in the preictal-decrease group, examples of each seizure-related activity change are observed in all anatomic groups.

Figure 4.  Seizure-related activity on microelectrodes is highly variable with regard to seizure-onset zone (SOZ). (LEFT) Schematic of a depth electrode for one seizure containing macroelectrodes (bars), microwires extending from the tip of the electrode (lines), and microwires placed between macroelectrodes on the shaft (dots). Light gray rectangle shows the clinically defined SOZ as determined by low-frequency clinical criteria. Arrows from macroelectrodes and circled microelectrodes point to EEG shown on the right. (RIGHT) EEG and firing rates beginning 45 min prior to seizure onset (“0”). EEG for microelectrodes has been high-pass filtered to show unit activity, but seizures were clearly observed on microelectrodes near the SOZ commensurate with seizure activity observed on the macroelectrodes. Vertical scale is the firing rate normalized to firing rate during 15 min beginning 45 min prior to seizure onset. Note that the scale on the top row is much larger than the other rows. Symbols at right show seizure-related activity change category: hollow arrows show preonset changes, solid arrows show postonset changes, and “NC” shows “no change.”

If the pathologic network underlying seizure generation was a fixed neuronal assembly, then seizure-related MUA changes for the constituent neurons would be similar across seizures within a patient. Two seizures were recorded within the same, continuous recording of five patients, allowing an analysis of the interseizure variability of MUA. There were a total of 118 traces where seizure-related classification (e.g., “preictal increase,” see Materials and Methods) was computed for two different seizures. Other than showing that the “unchanged” classification was the most likely category for any given seizure, seizure-related activity was not consistent for the second seizure, regardless of categorization for the first seizure. Of 43 microelectrodes that displayed seizure-related changes in the first of two seizures, six showed a seizure-related change in the second (14.0%), but three of these microelectrodes were classified into different categories. One “preictal increase” and two “ictal increase” traces were consistent across seizures. Four microelectrodes that were classified as “unchanged” during the first seizure were classified as “preictal increase” in the subsequent seizure (i.e., “unchanged-preictal”), 15 were classified as preictal decrease, and 12 microelectrodes were classified as ictal increase. The observation of both “unchanged” and seizure-related changes in activity from the same microelectrode during different seizures rejects the possibility that all “unchanged” microelectrodes resided in brain regions unaffected by seizures. A comparison of ensemble activity between two seizures in the same patient (Fig. 5) showed that MUA patterns bear little resemblance to one another despite similar LFP recordings from a nearby macroelectrode.

Figure 5.  Seizure-to-seizure variability, despite similar macroscopic recordings. Top and bottom panels show 45 s of data around seizure onset from sequential seizures during the same recording session, separated by 5 h. EEG recording on the top line of each panel shows data from the same macroelectrode during each seizure. Colored plots below show firing rates for 30 microelectrodes, normalized to the firing rate of each electrode for 15 min beginning 45 min prior to seizure onset. Flat lines show microelectrodes that recorded firing during the baseline period, but showed little or no activity during the seizure.

Cross-correlations between groups of microelectrodes were computed for the 45-min preceding seizure onset to 5 min following onset (Fig. 6). Compared to the baseline period (45–30 min prior to seizure onset), correlation coefficients differed for action potentials from microelectrodes contralateral to (0.0026 ± 0.0022), ipsilateral to (0.0044 ± 0.0023), and inside the SOZ (0.0109 ± 0.0087) during the 15-min baseline period beginning 45 min prior to seizure onset (ANOVA, p < 0.001). Changes in synchrony were observed prior to and continuously into seizure onset only for contra-SOZ pairs, only for the last 30 s prior to seizure onset, and not for any other groups. This suggests that a global change in action potential synchrony is not required to initiate seizures. Cross-correlations between ipsilateral and SOZ traces were increased relative to baseline following seizure onset and often persisted for several minutes, suggesting an increase in synchrony across a broad population of neurons.

Figure 6.  No change in cross-correlation observed prior to seizure onset. Traces were grouped according to the location of the microelectrode relative to the macroelectrode-determined seizure onset zone: contralateral to (Contra), ipsilateral to (Ipsi), and within (SOZ) the seizure-onset zone. (A) Average time-shift normalized cross-correlation across all patients. Significant correlations observed prior to seizure onset did not persist continuously up to seizure onset. (B) Example from one patient showing no preictal or postictal changes in cross-correlation.

Discussion

The MUA during the interictal-to-ictal transition associated with focal seizure generation showed a heterogeneous response, with MUA unchanged, increased, and decreased both inside, outside, and contralateral to the SOZ. Firing rates on the majority of microelectrodes that recorded MUA, however, did not change either prior to or during the seizure. This finding was independent of anatomic location of the microelectrodes, or their relationship to the SOZ. Even for the microelectrodes with preictal MUA changes, the pattern was not repeated in subsequent seizures despite the similarity of the ictal LFP (Figs 4 and 5). Ensemble activity from two macroscopically similar seizures (Fig. 5) suggests that the neuronal assemblies underlying their generation do not necessarily overlap and that neurons that might be considered part of the epileptic network during one seizure may not fire at all during a subsequent seizure. Finally, the absence of increased global synchrony (Fig. 6) suggests that any change in synchrony prior to seizure onset occurs within a restricted population of neurons.

This heterogeneous pattern of neuronal firing in seizure generation agrees with previous observations during spontaneous seizures from both humans (Babb & Crandall, 1976) and rats with epilepsy (Bower & Buckmaster, 2008). The firing rates of granule cells in epileptic rat dentate gyrus displayed heterogeneous seizure-related firing rates, and although a group of units showed increased firing at seizure onset, the largest group showed no change at all before or during the seizure (Bower & Buckmaster, 2008). Unlike earlier human studies of neuronal activity at and following seizure onset (Verzeano et al., 1971; Babb & Crandall, 1976), the increased spatial sampling and longer recording durations in this study provide a systematic test of whether specific, “epileptic” neurons become active prior to seizures based on neuronal firing rates during long (>1 h) interictal periods and compared to the interictal-ictal transition. The strict action potential detection criteria used in this study could produce instances of false-negative action potentials (i.e., “missed spikes”), but it is unlikely to have affected the results, because missed action potentials would bias toward decreased firing rates, particularly around the time of seizures when the extracellular concentrations of ions may change, thus distorting waveforms. Our data contain several clear examples of increased firing rates in all patients. Therefore, if the current approach of spike detection did bias detection rates, our results should give a more conservative estimate of single neuron firing increases prior to seizures.

The observation that the majority of recorded neurons are not involved in seizures and that “preictal increase” microelectrodes are present inside and outside the SOZ (Fig. 3) suggests that a more spatially distributed and sparse network of neural tissue is involved in ictogenesis. The observation that more “ictal” neurons were observed in the SOZ than would be predicted by chance, however, is consistent with the SOZ as the site of initial seizure propagation. The observation of increased “preictal decrease” responses ipsilateral to, but outside of, the SOZ could reflect an inhibitory surround response at seizure onset, but it should be noted that this result is heavily weighted by observations from a single patient (patient 4, Table 2). The presence of all types of responses in all three categories of tissue (SOZ, ipsi, and contra) argues against the assumption, however, that ictogenesis occurs solely in the SOZ.

Seizures are defined by the spatiotemporal characteristics of pathologic local field potentials, and the emphasis on macroelectrode recordings and success of epilepsy surgery has fostered these definitions. Epilepsy, however, is a disease of cells—neurons and glia—not field potentials, and the concept that focal seizures originate from a well-defined population of neurons within a fixed epileptic network is not necessarily what should be expected from neuronal microcircuitry. The results do not support a model of focal epileptic brain and ictogenesis characterized by a pathologic collection of neurons localized to the SOZ that become highly synchronized and serve as fixed, seizure-generating “seeds” that recruit the majority of surrounding neurons. The findings in this study are consistent with seizures arising from an “epileptic network” (Bragin et al., 2000) with multiple neuronal assemblies within that network capable of seizure initiation from multiple locations that evolve into the stereotypic, macroscopic ictal LFP.

Acknowledgments

We appreciate the technical support provided by Cindy Nelson and Karla Crockett. This research was supported by the National Institutes of Health R01-NS063039(GW), Mayo Clinic Discovery Translation Grant, Minnesota Partnership for Biotechnology and Medical Genomics.

Disclosures

The authors have no conflicts of interest. We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.