Evidence of an inhibitory restraint of seizure activity in humans

Abstract

The location and trajectory of seizure activity is of great importance, yet our ability to map such activity remains primitive. Recently, the development of multi-electrode arrays for use in humans has provided new levels of temporal and spatial resolution for recording seizures. Here, we show that there is a sharp delineation between areas showing intense, hypersynchronous firing indicative of recruitment to the seizure, and adjacent territories where there is only low-level, unstructured firing. Thus, there is a core territory of recruited neurons and a surrounding 'ictal penumbra'. The defining feature of the 'ictal penumbra' is the contrast between the large amplitude EEG signals and the low-level firing there. Our human recordings bear striking similarities with animal studies of an inhibitory restraint, indicating that they can be readily understood in terms of this mechanism. These findings have important implications for how we localize seizure activity and map its spread.

Figure 1. Failure of propagation of full ictal events in mouse brain slices.

(a) Five successive ictal events (shown left to right) recorded in two layer 5 pyramidal cells (PCs) 600 μm apart. Only in the last three events, does PC 1 make the transition from inhibitory to excitatory barrages, which is indicative of being incorporated into the ictal event. A very important point arising from a previous study is that the upward deflections are not purely inhibitory, merely predominantly so. The level of inhibition simply occludes the very large concurrent excitatory drive at this time, which can be seen instead if the cell is clamped close to the GABAergic reversal potential. (b–f) Another example of δ-frequency, interictal-like activity, with concurrent low magnification Ca²⁺ network imaging. The more extended field of view allows us to visualize the failure of propagation, as the view incorporates territories that are recruited to the ictal event and other regions that resist recruitment. (b) Schematic showing the field of view in the brain slice. (c) Field of view; the two electrodes are visible, located in layer 5. (d) Eight minute recording showing two full ictal events, the first of which is only manifest as δ-frequency interictal activity in electrode 2. (e,f) Detailed views of the two ictal events, showing the time period of two Ca²⁺ network imaging movies (Supplementary Movies 1 and 2).

Figure 2. Ca²⁺ network imaging of failure of ictal propagation in mouse cortical slices.

(a) Mean neuronal Ca²⁺ fluorescence (centre-surround subtraction10) from 65 cells during three different epochs (epochs defined from concurrent voltage clamp recording of a layer 5 pyramidal cell within the field of view; all imaged cells within 200 μm of the recorded cell). Epoch 1 was a period of ictal activity elsewhere in the slice, which failed to invade the local territory. Epoch was baseline activity, and epoch 3 shows a full ictal event that did recruit the field of view. The grey shaded area shows the upper and lower extent of 5 s.e.m. signal for the 65 cells. Note: how even during the 'Failure of propagation', the mean signal comfortably exceeds baseline (mean±5× s.e.m.). (b) Brain slice loaded with OGB1. Two movies of this slice are in the Supplementary Information, showing a full ictal event and another event that failed to propagate into this territory. Representative traces of neuronal activity for these two events are shown in c and d, together with a voltage clamp recording of a local pyramidal cell (V_hold=−30 mV). In the V_clamp recording, note the upward deflections (predominant inhibition) for the failure, and the downward deflections for the successful propagation, as shown also in the example in panel a. (e) Coherence matrix for the Ca²⁺ signal of 65 imaged neurons, showing a general lack of coherence when the ictal event fails to invade, in contrast with the highly coherent signal when the ictal event does invade (number of coherence measures=2080; non-participation coherence (mean±s.d.)=0.494±0.172; post-recruitment coherence=0.914±0.075; t_s=102.0; P<<0.001, Student's t-test). (f) Line scan of a propagating ictal wavefront, derived from ×10 magnification Ca²⁺ network imaging. Spatial profiles of the wavefront (dotted line) were derived during periods of stability. (g) The spatial profile of activity measured in eight brain slices.

Figure 3. Simultaneous MEA and EEG seizure recordings share common features in the EEG frequency range.

(a,b) Patient C7 initiation, as recorded from one MEA channel (a) and the nearest active EEG channel (b, electrode 22, Supplementary Fig. 1). Spectrograms computed using the Morlet wavelet transform (1–50 Hz) together with the filtered time series signal are shown. This seizure followed the most common neocortical onset pattern, reflected in both the MEA and EEG channels, with an initial epileptiform discharge followed by rhythmic beta activity that gradually slows to the theta range and gains in amplitude. The seizures recorded from patients C2, C3 and C4 followed this same general pattern. (c,d) Simultaneous MEA and adjacent EEG recordings from patient C5. This seizure, following an electrographic pattern that is less common but still well known in neocortical EEG recordings, was characterized by an initial monomorphic delta rhythm that gradually increased in frequency to the theta range, then slowed to a 3–4-Hz spike and wave rhythm before offset.

Figure 4. Seizure propagation in vivo in a human.

(a) Multiunit activity in four electrodes, spaced at 800 μm intervals in patient C7, showing the spatial progression of the ictal wavefront and illustrating the pre-recruitment, recruitment and post-recruitment seizure phases. Black arrowheads denote the progression of the ictal wavefront, and delineate the pre-recruitment from the post-recruitment period. The recruitment index refers to the temporal ordering according to the time of the sharp increase in firing ('recruitment'). Inset: layout schematic showing the locations of the four selected channels, with the convention that the wire exits on the left. (b) Raster plot showing the colour-coded multiunit activity for all electrodes in C7, which showed unit activity ('hot' colours active; 'cold' colours inactive), ordered according to the time of recruitment. (c) The delay in electrode recruitment plotted versus the distance from the first recruited (index) electrode. The slope of the line of best fit indicates a propagation speed of 0.83 mm s⁻¹. (d) Low-frequency (2–50 Hz) signal from the same electrodes shown in panel a. Note the absence of any visually notable features in this bandwidth that correspond with the clear recruitment process in the multiunit activity (filled arrowheads), although note also that the transition into the clonic phase (open arrowheads) corresponds to an increase in amplitude that can be seen as a delayed progression. (e) Overlays of the 2–50 Hz signal recorded at all microelectrodes for short periods before and after recruitment. Note: the smear of signals, due to propagation across the array, and also the relatively greater degree of pre-recruitment variability compared with post-recruitment. Propagation of these events was consistent in different subjects, and over 2 orders of magnitude (>100×) faster than ictal wavefront propagation. This 100-fold difference in propagation speed mirrors exactly the situation in the 0 Mg²⁺ model where forward and backward projecting synaptic barrages that propagate with identical speeds, roughly 100-fold faster than the ictal wavefront. (f) Estimates of the spatial extent of the wavefront. For the time of peak ictal activity in each individual electrode, we plot the distribution of distances to the nearest pre-recruitment electrode (see Methods for further detail).

Figure 5. Hypersynchronous activity and electrode coherence following recruitment.

(a) Action potential—field potential phase plots showing a marked transition in the phase timing of action potentials, from the pre-recruitment phase when action potentials occur with equal probability at all phases of the field potential, to the post-recruitment period when firing phase is very skewed, indicating that firing is strongly influenced by the rhythm of the seizure. (b) The low frequency (2–50 Hz) signature of ictal invasion is a dramatic increase in phase coherence between electrodes (WPLI, weighted phase lag index, paired t-test, T=83.87, P0.001, n=165).

Figure 6. Electrophysiological indicators of non-participating cortical territories in human seizures.

(a) Multiunit activity and (b) 2–50 Hz signal during a seizure in patient C4. Although some electrodes clearly show firing activity, this was at a low level, with no obvious spatial pattern or wave of recruitment. (c) Action potential—field potential phase plots showing that the firing phase of the units remains in the 'pre-recruitment' pattern throughout the event, at no time showing any significant phase angle preference (note different radial scales in early and late seizures). (d) The inter-electrode coherence measure shows a minimal increase from early to late seizure periods, in marked contrast to the striking increase in coherence associated with ictal invasion of the MEA territory (see Fig. 5b).

Figure 7. Stereotyped firing patterns are another hallmark of territories that are recruited to seizures.

(a) Recording from a single microelectrode implanted in a cortical area fully recruited into seizures (patient C5), showing LFP in the frequency range of clinical EEG in panel b, and multiunit activity in panel c. For clarity, only the initial portions of both seizures are shown. Note: the highly conserved field potential (panel b) and multiunit activity (c). Panel d shows units from a different channel, while e shows the cumulative firing trajectories for every channel with a firing rate exceeding 12 spikes s⁻¹. Trajectories were highly correlated across the three recorded seizures (Spearman coefficients 0.71±0.14 for 84 channels). The black line is the trajectory for the electrode raster in d, and the colours for other electrodes are preserved in the two plots for ease of comparison. (f) The same plots for a territory not incorporated into the seizure (patient C4). (g) We show recordings from the same electrode during two seizures. Note that the LFP pattern is again highly conserved, but the unit activity is not, as evident in both the single electrode examples (h,i) and also the high, seizure to seizure variability of trajectories (Spearman coefficients 0.38±0.26 for 30 channels across seven seizures) (j). Again we only show trajectories for electrodes showing greater than 12 spikes s⁻¹ during all seizures, and so j excludes examples like that shown in h, which are even more extreme.

Figure 8. Sharp drop in activity variance as wavefront incorporates cortical territories.

(a) The Fano factor, a measure of firing heterogeneity, has been used previously to characterize neuronal spiking patterns during seizures. In territories that are fully recruited to a seizure, the Fano factor increases sharply during the period when the ictal wavefront passes across the MEA, but once it has incorporated the entire array, the metric drops sharply to below baseline, even though firing rate remains very high (red arrows). (b) In territories that are penumbral to the seizure throughout its duration, the Fano factor remains persistently raised above baseline.