Neocortical Seizure Termination by Focal Cooling: Temperature Dependence and Automated Seizure Detection


Address correspondence and reprint requests to Dr. S. Rothman at Department of Neurology, Room 12E/25, St. Louis Children's Hospital, One Children's Place, St. Louis, MO 63110, U.S.A. E-mail:


Summary:  Purpose: The therapy for focal neocortical epilepsy remains suboptimal. We have, therefore, worked to develop techniques to cool small regions of the neocortical surface for seizure mapping and, ultimately, for long-term suppression of focal seizures.

Methods: We induced focal neocortical seizures in halothane-anesthetized rats by the microinjection of 4-aminopyridine (4-AP) into the motor cortex. The dura over the injection site was cooled with a Peltier device, and the temperature at the interface between dura and Peltier was measured with a thermocouple. In some experiments, seizures were automatically detected by a computer program that activated the Peltier device.

Results: Monopolar EEG indicated that our seizures were focal and suppressed when cooling was applied directly over the injection site. The threshold temperature required to observe any reduction in seizure duration was 24°C. The temperature gradient across the cooled neocortex was sharp, with the temperature increasing to 31°C at 4 mm below the Peltier, which was cooled to 20°C. Automatic seizure detection reduced the total seizure duration from 43.4 ± 33.6 s to 5.6 ± 5.3 s.

Conclusions: Cooling terminates neocortical seizures when applied very close to the epileptogenic focus. The threshold for seizure termination (24°C) may be lower than the threshold for termination of normal cortical activity, suggesting that this technique will not dissociate the anticonvulsant effect of cooling from the disruption of normal behavior. However, when coupled with automatic seizure detection, focal cooling remains an attractive option for development as a treatment for focal epilepsy.

Many patients with focal neocortical epilepsy are inadequately controlled with presently available anticonvulsants (AEDs). Many of these patients are poor candidates for cortical resection, because they lack a readily identifiable epileptogenic focus. We have, therefore, become interested in the possibility of using small thermoelectric (Peltier) devices to improve the treatment of this group of patients. These devices could aid in presurgical mapping by pinpointing regions of neocortex in which cooling eliminated ictal activity. They might define the functional consequences of permanently ablating a region of neocortex by revealing temporary deficits in higher cortical function during reversible focal cooling (1,2). Additionally, it may be possible to develop a permanently implantable cooling device, coupled to seizure detection or anticipation software, which could be used to terminate or even prevent focal seizures (3–6).

We have already demonstrated that rapid cooling can quickly terminate in vitro and in vivo seizures (7,8). Our initial in vivo study showed that cooling the cortical surface to ∼20°C could abort 4-aminopyridine (4-AP)-induced seizures within a few seconds without causing any histologic damage. However, this study raised a variety of questions about neocortical cooling that required further examination. Accordingly, we have gone on to determine (a) the extent of seizure spread induced by 4-AP microinjection, (b) the proximity of the Peltier device to the seizure focus required for seizure termination, (c) the temperature gradient established by the Peltier device in the rat neocortex, (d) the degree of cooling necessary to abort seizures, and (e) the improvement in seizure control when a Peltier device is used in conjunction with an automatic seizure-detection algorithm.

The results of these studies provide more detailed information about the cooling parameters required for seizure control and support the hypothesis that small Peltier devices might improve the treatment of neocortical epilepsy.


Neocortical seizure model

We used a protocol approved by the Washington University Animal Studies Committee. We anesthetized adult male Sprague–Dawley rats, 350–400 g, with halothane and then placed them on a heating pad in a stereotaxic frame (David Kopf Instruments, Tujunga, CA, U.S.A.). We applied the halothane continuously through a nosepiece. We selected halothane because we and other investigators have successfully produced prolonged focal seizures in halothane-anesthetized rats (8,9). A craniotomy was performed while the rat was breathing 4% halothane, but the inspired halothane was reduced to <1% when we induced seizures. The skin was infiltrated with 2% lidocaine and a 5 × 10-mm cranial window was created over the anterior left hemisphere by using a dental drill. The window extended medially to the sagittal sinus and 5 mm anterior and posterior to the coronal suture. After creation of the window, the dura was gently opened to allow drug injection. During the drilling, the skull was continuously irrigated with artificial cerebrospinal fluid to prevent the underlying brain from overheating.

We produced recurrent, focal seizures by injecting 0.5 μl of a 4-AP solution (25 mM in artificial cerebrospinal fluid) using a commercial oocyte injection system (Drummond Scientific, Broomall, PA, U.S.A.) coupled to a glass micropipette (tip diameter, ∼100 μm) (10,11). The injection system was mounted on a micromanipulator that allowed us to administer the 4-AP 0.5 mm below the surface of the motor cortex, at a position 2 mm anterior to the bregma and 2.5 mm from the midline. The injection was carried out over a 5-min period, to minimize cortical trauma, and the pipette was left in place for 20 min to prevent the 4-AP from escaping through the injection site.


For most experiments, we placed two screw electrodes symmetrically over each hemisphere and differentially recorded the EEG between the two (8). The EEGs were recorded using standard amplifiers (Grass-Telefactor, West Warwick, RI, U.S.A.), and digitized (200 Hz) and stored using PC-based commercial hardware and software (Digidata and Axoscope; Axon Instruments, Union City, CA, U.S.A.). We typically began EEG monitoring before injection of the 4-AP and continued throughout the entire experiment. Seizure onset and termination were readily recognizable as abrupt changes in EEG frequency and amplitude, and we previously found excellent agreement between two independent observers (X-F.Y. and S.M.R.), who independently measured seizure durations (8). Except when specifically indicated, the control seizure durations used to compare control and cooled seizures were corrected for the latency to seizure detection for that specific cooled group, so that the amount of time required to detect seizures and activate cooling would not unfairly bias our results. In our calculations, we took average values for seizure duration for each rat, so that rats with a larger number of seizures would not unfairly bias our results. In several experiments, we used an array of five monopolar electrodes the better to define the extent of seizure spread from our focal injection.

Focal cooling

We focally cooled the neocortex with commercially available thermoelectric (Peltier) chips (Melcor, Trenton, NJ, U.S.A.). Two chips, each 3.5 × 3.5 × 2.4 mm were positioned together, glued to the end of a copper rod, and serially connected (8). The rod allowed the device to be mounted in a micromanipulator and also made an excellent heat sink. The Peltier chips were powered by an adjustable DC supply that limited current to <0.8 A. The manipulator was positioned so that the Peltier device just touched the dura over the 4-AP injection site. A 0.13-mm thermocouple (Omega Instruments, Stamford, CT, U.S.A.), attached to the surface of the Peltier device, was connected to a controller (Omega CN1001), allowing the temperature to be held at a preset level and simultaneously monitored while the EEG was recorded.

Automatic seizure detection

We developed our own seizure-detection program by using a commercially available analog-to-digital converter (ADC) and compatible software (NI 6024E and LabView 6; National Instruments, Austin, TX, U.S.A.), which runs on a 350-MHz Windows PC. The incoming EEG data are placed into an array (default size, 128), filtered by using a Blackman-Harris window, and transformed into the frequency domain by using a fast Fourier transform. The major spectral peaks in the 5- to 40-Hz frequency range (usually 5) are identified, and their values are totaled. This total signal level is compared with a user-defined “rising threshold.” We defined a “seizure” as a signal level in excess of this threshold for 20 consecutive cycles (100 ms at a sampling frequency of 200 Hz). When a seizure has been detected for a specified duration beyond 20 cycles (“on window”), a digital logic output is asserted and turns on the Peltier device. This “on window,” which was typically set to 5 s, prevented Peltier activation by a very brief seizure. When the signal level is below the “declining threshold” for six cycles (30 ms), the seizure has terminated, and the digital output is deasserted. However, the digital output remains asserted for a preset interval (“off window”) after seizure termination, ensuring that it outlasts the seizure duration. Even though the output remains asserted, the detection algorithm considers the seizure to be over and will begin looking for seizures again. Consecutive seizures that recur within the “off window” will be detected as single events.


We were first interested in determining whether our 4-AP model actually produced spatially restricted seizures, because it seemed possible that the 4-AP might diffuse far enough to cause an extended region of epileptogenicity. We switched from our standard differential EEG recording technique to recording the electrocorticogram with monopolar leads. Two chloride silver wires were place on the dura, 2 mm away from the 4-AP injection site on both sides, and three additional wires were place at 2-mm intervals farther to the right. Under these conditions, when seizures were detected, they were most intense in the two electrodes that bracketed the injection site, and were much less apparent in the three more distant electrodes (97 seizures in five rats; Fig. 1). This remained constant for an observation period of close to 2 h. This is, therefore, reasonable evidence that the seizures produced by 4-AP microinjections remain focal.

Figure 1.

Neocortical seizures induced by 4-aminopyridine (4-AP) injection are localized. A: Diagram of skull showing location of cranial window and site of monopolar electrocorticogram leads. The five silver-wire EEG leads (E–A, from left to right) were referenced to a screw electrode in the occiput. The 4-AP was injected between leads E and D. B: An ictal electrocorticogram shows that the seizure activity was best seen in the two leads bracketing the injection site (E, D), barely detected 4 mm away (C), and inapparent 6–8 mm away (B, A).

Although our previous work had shown quite convincingly that focal cooling could rapidly terminate focal seizures when the Peltier was placed directly over the injection site, we were concerned that the effect of cooling might not be so restricted. We decided to see whether cooling on the side of the injection site had any effect on seizure duration. When we moved the cranial window to the right, so that the edge of the Peltier device was 5 mm away from the injection site, we saw no effect on seizure duration. Control seizures lasted 76.7 ± 12.9 s, and cooled seizures were 72.8 ± 19.5 s (n = 69 and 30 seizures, respectively; p > 0.1). It appears that in our seizure model, the cooling must be applied directly over the seizure focus to be effective (Fig. 2).

Figure 2.

Surface cooling induces steep temperature gradients in the cortex. A: Diagram of device constructed to measure temperature directly below Peltier device. B: Temperature beneath Peltier is lowest in the cortex immediately adjacent to the device, but extends to a depth of 4 mm from a baseline of ∼30°C at the dural surface. Each point represents the average of three measurements taken from five separate animals.

The inability of the Peltier device to reduce seizure duration when it was placed >5 mm from the 4-AP injection site suggested that the neocortical temperature gradient was quite steep. However, we wanted to map the gradient more accurately with a thermocouple. We slightly modified our cooling device and holder, putting a 1-mm space between the two Peltier wafers and drilling a hole in the copper bar to which they had been glued (Fig. 2A). This allowed us to advance a 33-g needle containing a thermocouple accurately, directly below the Peltier in small steps controlled by a micromanipulator. When we cooled the cortical surface to 20°C, the temperature at 1 mm was 24°C, increasing to 31°C at 4 mm (Fig. 2B). The fairly sharp temperature gradient indicates that a cooling device must be placed very close to the focus, as expected from the experiments described earlier.

Another important issue, not yet specifically addressed, is the actual magnitude of temperature reduction required for seizure cessation. At the onset of seizures, we reduced the cortical surface temperature to preset levels between 20°C and 26°C to determine the cortical cooling required to reduce seizure duration. We found that 20°C and 22°C produced rapid and insignificantly different changes in seizure duration (Fig. 3). At 24°C, seizure reduction was inconsistently affected, and at 26°C, we could see no influence of cooling on seizures. We have not yet gone below 20°C, because of concern about the possibility of producing cortical damage.

Figure 3.

Temperature dependence of seizure cessation. A: Significantly different from control and 26°C (p < 0.001), but not each other (p > 0.05). B: Significantly different from control and 26°C (p < 0.01). C: Not significantly different from control, p > .0.05).

Finally, we wished to study the potential utility of automated seizure detection in improving the efficacy of focal cooling. We first established the reliability of our seizure-detection algorithm. We compared the duration of seizures manually detected with the duration obtained from the digital output of the computer (Fig. 4). We included both cooled and control seizures and found excellent agreement between the two measurements (r = 0.97). These measurements took into consideration the “on” and “off” windows of the detection program and excluded recurrent seizures that occurred at intervals of <15 s, the typical “off window” of our program. Over an interval in which we observed 58 seizures by manual inspection, the computer detected 60 seizures, giving a false-positive rate of only 3.5%; there were no false negatives. When we extended our observations, we found that automatic cooling reduced seizure duration from 43.4 ± 33.6 s to 5.6 ± 5.3 s (n = 34 and 24 seizures, respectively; p < 0.001; Fig. 5).

Figure 4.

Durations of seizures measured manually or by computer are highly correlated.

Figure 5.

Efficacy of automatic seizure detection. Seizure (A) quickly decreases after the onset of cooling (B) triggered by seizure-detection algorithm (C). The interval between the arrow in C and the square wave indicates the “on window,” the time between seizure detection and onset of the digital output activating the Peltier device.


The results of the present experiments support our earlier observations that focal cooling can rapidly terminate neocortical seizures (8). Although the prior experiments yielded very robust results, we were concerned about several issues we had not previously addressed. We had assumed that the 4-AP microinjection would elicit very focal seizures, but had no confirmatory experimental evidence. It was possible that the drug would rapidly diffuse and affect an entire hemisphere. The results with electrocorticography demonstrate that the seizures were very focal and failed to spread more than a few millimeters from the site of injection. This suggests that our 4-AP model may turn out to have general utility in epilepsy research. We are aware that seizure foci may not be so discrete in human neocortical epilepsy, but it may still be feasible to use cooling on a larger region of cortex.

As is the case in entorhinal slices, 4-AP produces ictal-like discharges and not just inter-ictal spiking (10,12). The seizures last for ∼1 min, spontaneously terminate, and continually recur for >2 h. They therefore provide a consistent model for seizures that may be relevant to focal human epilepsy. It is potentially relevant that 4-AP blocks a neuronal potassium current, because one group of genetic human epilepsies, albeit generalized seizures, also is associated with potassium channel mutations (13–15).

Our early work also had omitted any controls on the effective cooling radius of our Peltier devices. Whereas we had assumed that it was essential for the cooling device to be in contact with the seizure focus, we had no evidence supporting this hypothesis. We have now shown that placing the Peltier chip a few millimeters away from the site of convulsant injection nullifies the effect of cooling. We must verify that this result generalizes to other forms of focal epilepsy, but we are optimistic that focal cooling may be a useful technique for localization of epileptic foci in humans.

The steep temperature gradient induced by cooling likely explains the requirement for the Peltier chip to be in close proximity to the epileptic focus. We, like others, have found that the effects of cooling are very discrete (2,16). This is almost certainly due to high local cortical blood flow nullifying the effect of any cooling device not in direct contact with the cortex or dura. Based on our detecting cooling to a depth of 4 mm, we would anticipate at least some efficacy if the technique were used in human neocortex, which varies in thickness from 1.5 to 4.5 mm (17). It may even be possible to increase the efficacy of focal cooling for seizure control by going below 20°C. Thus far, no brain injury has been detected at this level of cooling, although very severe cooling will disrupt the blood–brain barrier and injure underlying cortex (8).

We were very curious about the threshold temperature required to terminate seizures in our model. Our previous experiments had reduced temperature to ∼20°C, but we had not attempted to hold a specific temperature. We added feedback to our system, so we could maintain a set temperature in the present experiments. We found that there was no difference in seizure duration when temperature was reduced to either 20°C or 22°C. However, cooling to 24°C had a more variable effect on the duration of seizures, with some seizures virtually unaffected; at >24°C, cooling had no effect. Our results on seizures appear almost identical to the results described by Lomber et al. (16), who had to cool cat cortex to <24°C to affect visual cortex activity. Based on our current model, we are pessimistic that there is a level of temperature reduction that will allow maintenance of normal cortical function and still abort seizures. However, the 4-AP model may be especially severe, so that conditions responsible for focal epilepsy in humans may be sensitive to more limited cooling.

The reduction in seizure duration with automatic seizure detection was gratifying. With our relatively straightforward algorithm, seizures could be detected in 1–2 s and therefore aborted within 5–6 s. We recognize that our algorithm is relatively simple, but it was extremely reliable for detecting experimental seizures in rats. We could have improved the response speed even more, had we reduced the length of the “on window.” However, we were reluctant to do this, because the program would have detected such brief seizures that we would have had difficulty visually confirming the presence of paroxysmal activity. More sophisticated seizure-detection algorithms would likely work even faster, with an even lower probability of false positives (18,19). Combining automatic seizure detection with focal cooling during cortical mapping in the monitoring phase of epilepsy surgery might allow identification of epileptogenic foci. This data could permit clinicians to make more informed decisions about surgical resections of an epileptogenic region. It would be very laborious to attempt to obtain this information from direct visual EEG monitoring and manual activation of cooling.

The ultimate goal of cooling (or any other focal therapy for epilepsy), however, is to prevent seizures before they even begin. This now seems like a feasible goal. At least three different neurophysiology research groups have developed EEG analysis algorithms that anticipate seizures before the development of clear ictal discharges (3,6,20). Seizures can be predicted at least several minutes before EEG or behavioral seizure activity, suggesting that the availability of a safe, rapid intervention strategy would allow seizure prophylaxis. We believe that focal cooling may be the most attractive seizure prophylactic technique now in development. The many practical problems related to focal cooling must be surmounted, including improved techniques for heat dissipation and current generation. The rate of progress in the microelectronics industry is so rapid that there is reason to be very optimistic that these obstacles can be overcome.

Acknowledgment: This work was supported by a grant from Citizens United for Research in Epilepsy, Inc. (CURE), the Stein Fund for Pediatric Neurology Research, and NS14834 from the NIH.