Adaptive electric field control of epileptic seizures

Abstract

We describe a novel method of adaptively controlling epileptic seizure-like events in hippocampal brain slices using electric fields. Extracellular neuronal activity is continuously recorded during field application through differential extracellular recording techniques, and the applied electric field strength is continuously updated using a computer-controlled proportional feedback algorithm. This approach appears capable of sustained amelioration of seizure events in this preparation when used with negative feedback. Seizures can be induced or enhanced by using fields of opposite polarity through positive feedback. In negative feedback mode, such findings may offer a novel technology for seizure control. In positive feedback mode, adaptively applied electric fields may offer a more physiological means of neural modulation for prosthetic purposes than previously possible.

Fig. 1.

Schematic of the hippocampal slice and electrodes in the perfusion chamber (modified Haas style) as viewed from the side (middle) and above (top), with the wiring for the field feedback amplifier indicated and a block diagram of the controller. The brain slices rest on a nylon mesh just below the top surface of the perfusate of ACSF. The atmosphere above the perfusate is warmed to the bath temperature of 35°C and saturated with 95% O₂/5% CO₂. An electric field is imposed on the slice by a set of Ag–AgCl electrodes embedded in the floor of the chamber. The current applied between these parallel-plate field electrodes is feedback-controlled so that the potential difference between the sensing electrodes is proportional to a program voltage. An additional pair of electrodes is used as recording ground. Also shown (bottom) is the alignment of the electric field with the somatic–dendritic axis of a neuron that results in suppression. With this alignment, the electric field polarizes the neuron, leaving the transmembrane potential of the soma more negative with respect to the extracellular space. If the spike initiation zone of the neuron is near the soma, as is typical for the pyramidal cells studied here, the cell will become less excitable. Reversing the field polarity will depolarize the soma and bring the neuron closer to threshold.

Fig. 2.

Power spectral density (PSD) for recorded activity and applied field stimulus when the stimulus was a low-frequency random signal (A) and a typical feedback control signal (B). For display purposes, the stimulus PSD was vertically scaled such that its amplitude matched that of the recorded activity PSD at low frequencies. In both cases, the stimulus PSD falls off quickly (∼f⁻²) for frequencies, f, above ∼4 Hz, in contrast to the neuronal activity PSDs, which have significant spectral power up to 350 Hz. Also shown are the PSDs of the recorded neuronal activity after removal of an estimate of the stimulus artifact. These signals are indistinguishable from the original recording for frequencies above ∼2 Hz. B, The raw signal lies slightly below the processed signal for low frequencies. These results indicate that the applied field is not simply masking the neuronal activity in the recording process during control. The stimulus artifact accounts for <5% of the RMS recorded signal amplitude.

Fig. 3.

Adaptive control of seizure activity using applied electric fields. In A–C, the main trace is the raw extracellular potential recording. Insets are tracings of activity, filtered to illustrate the high-frequency activity, shown at expanded scales. In each case, a dashed line is used to demarcate when control is turned on.A, B, Examples of seizure suppression from separate experiments using electric fields applied as a negative feedback parameter. Electrographic seizures are observed as an increase in high-frequency activity atop large low-frequency deflections (Traynelis and Dingledine, 1988). B, Seizures occur interspersed among frequent short network bursts (Rutecki et al., 1985). C, Example of seizure induction achieved using positive feedback.

Fig. 4.

Event detection results for a single 90 min recording with different electric field stimuli applied. Thebottom trace indicates feedback gain (G;left axis) or amplitude (A; right axis) of the applied stimulus. Greek letters andcolors indicate the type of stimulus: baseline (black, no letter); full-wave feedback control (red, α); half-wave-rectified feedback control (green, β); constant amplitude suppressive field (light blue, γ); low-frequency noise (dark blue, δ); suppressive half-wave-rectified low-frequency noise (orange, ε); positive feedback control (magenta, μ). Two types of event detection were used to identify synchronous neuronal activity from the recorded field potentials. RMS events were detected from variations in the RMS power in the frequency band 100–350 Hz. DC events were detected by threshold detection after low-pass filtering of the recordings at 10 Hz. The character of both types of events, as quantified by their average and maximal amplitudes as well as their duration, was visibly changed from baseline when control was applied. No events of either type were observed during the final and longest (16 min) application (α3) of full-wave control.

Fig. 5.

Traces and spectrograms of activity with and without control for the same experiment as Figure 4. A, Activity (bottom trace) and applied field (top trace) from the final application of full-wave control (α₃) from Figure 4 and the baseline preceding it.B, C, A 15-sec-long trace andspectrogram of a seizure-like event (B) and of activity during control (C) from A. The top traces in B and C are the activity, high-pass-filtered at 100 Hz. The spectrograms(B–D) are calculated in overlapping vertical frequency bins 50 Hz tall from 25–350 Hz, and in overlapping horizontal time windows 0.05 sec wide. D, Spectrogram for a longer period illustrating the contrast between baseline and controlled activity.

Fig. 6.

Examples of activity during nonfeedback electric field stimulus for the same recording as Figure 4. For each set, thetop trace is of the recorded activity, whereas thebottom trace shows the applied field. A, Application of a constant-amplitude (DC) suppressive field (Fig. 4, γ). B, Application of a full-wave low-frequency noise field (Fig. 4, δ). C, Application of a half-wave-rectified low-frequency noise field (Fig. 4, ε). In each case, large neuronal events are observed, although the full-wave noise field did have the effect of breaking up the seizure-like events into shorter durations. The horizontal axis is time in units of seconds.

Fig. 7.

Comparison of PSD of recorded activity during control (lines with symbols) as compared with baseline (lines without symbols). The control corresponds to the final control application in Figure 4, and the baseline corresponds to the final baseline application. PSDs were calculated in overlapping 1.64 sec (2¹⁴point) windows. The power averaged over the windows is shown inA, whereas the window-to-window variance of power is shown in B. For both measures, the controlled activity falls well below that of the baseline activity.

Fig. 8.

Statistics of the RMS power of recorded activity in the frequency band 100-350 Hz, calculated in 1.64 sec windows, for baseline (■), full-wave control (○), and half-wave-rectified control (▵). Statistics correspond to all applications, independent of gain for the recording of Figure 4. A, The normalized histogram. B, Cumulative probability. It is clear that the baseline activity has many windows with much higher power than either type of control. These windows correspond to the first phase of the seizures. Inset, The normalized histogram of power calculated with logarithmically spaced bins, abscissa, power; ordinate, frequency) for baseline (■) and full-wave control (○). From this plot, it is observed that deviations to both high and low power are eliminated during full-wave control. The windows with extremely low power correspond to the latter phase of the seizures and the recovery times after them. C, The power variance versus average power is plotted for these three conditions. The two types of control are statistically well distinguished from that of the baseline activity.

Fig. 9.

Examples of network activity when control is released. In A–C, theinset is the activity for the full control period, indicated in gray, plus the baseline periods before and after. A, The trace corresponds to the same experiment as Figure 3A, with half-wave-rectified control. The network oscillates between excitation similar to seizure onset and suppression by the controller. When control is released, this activity proceeds immediately into a full seizure-like event.B, C, Traces from another experiment in which half-wave-rectified control (B) was compared with nonrectified control (C). For half-wave rectification, seizures were observed very soon (0–3 sec) after control was released, as compared with 12–18 sec for nonrectified control. The time base for eachinset is the same and is indicated in A. The inset vertical scale is half that of the main traces.