Velasco et al.¹⁴⁶ identified the centromedian nucleus of thalamus as a promising target for electrical stimulation. To date, this target represents the most stimulated site in the brain for epilepsy, but no large, controlled trial has yet been completed of centromedian stimulation to document efficacy. The largest group of patients with centromedian stimulation was published by Velasco et al.¹⁴⁵ A total of 49 patients with a variety of intractable seizures were subjected to bilateral centromedian stimulation at intensities of 2.5 to 5 V, with stimulation frequencies ranging from 60 to 130 Hz. In the study population, centromedian stimulation was effective against generalized tonic–clonic seizures, tonic seizures, and atypical absence seizures, but not against complex partial seizures. A small cross-over trial of centromedian stimulation in seven patients with mixed seizure types²⁹ showed a trend toward benefit, but not of statistical significance.

The subthalamic nucleus is a primary target for electrical stimulation to ameliorate tremor and Parkinson symptoms.⁴ Laboratory studies suggest that subthalamic stimulation also can benefit seizures in animal models of epilepsy produced by kindling,⁷⁰ genetic absence,¹⁴⁷ fluorothyl,⁷⁰ and kainic acid.⁷⁹ Nine patients treated with subthalamic stimulation for epilepsy have so far been reported.^5,13,21,79 Six of the nine were said to show more than a 75% improvement in seizure frequency, although each of these reports was uncontrolled.

The anterior thalamic nucleus provides efferents to superior medial frontal and cingulate cortex. In turn, the cingulum white matter projects to the entorhinal cortex, and thereby to the hippocampus. Stimulation of the anterior thalamic nucleus therefore is well placed to influence certain frontal and temporal seizures. Initial work on thalamic stimulation for epilepsy was performed by Cooper, with published results on several dozen patients.^19,143 These studies reported beneficial outcome in several patients, but details and long-term follow-up results were not provided in the publications. Sussman et al.¹²⁸ reported in abstract form the benefit of anterior thalamic stimulation at 100 Hz and approximately 5 V in three of five patients tested. Subsequently, pilot studies of anterior nucleus thalamic stimulation were performed on 14 patients from four different study sites.^40,48,58

Figure 1 shows the changes in seizure frequency for the patient group before and after stimulation. The mean seizure frequency at 12 months of stimulation was 55% ± 39% of baseline, and 57% of patients exhibited at least a 50% improvement in seizure frequency (responders). Among those believed to have frontal or temporal seizure foci, the responder rate was 67%.⁴⁰ Five patients had seizures capable of producing falls; four of these had a marked reduction in such seizures.⁵⁸ None of the 14 patients in the pilot trial experienced a serious complication. However, placement of deep brain stimulation electrodes is known to comprise a measurable risk for hemorrhage⁹ and infection.¹³² One patient implanted at a fifth institution using a similar thalamic stimulation pilot protocol had a hemorrhage leading to contralateral hemiparesis. Hodaie et al. in Toronto⁴⁸ have suggested that implantation itself might account for much of the perceived benefit, either by a microlesion effect or placebo effect.

Based on anecdotal and uncontrolled evidence of possible efficacy for anterior thalamic stimulation, a pivotal, multicenter, randomized, and blinded trial has been launched, called SANTE, for Stimulation of the Anterior Nucleus of Thalamus for Epilepsy (see www.epilepsycontrol.com). Eligible patients are those with medically uncontrolled partial or secondarily generalized seizures, not believed amenable to surgical resection. Seizures must be severe enough to interfere seriously with quality of life and must occur at a frequency of at least six per month. The SANTE trial utilizes bilateral implantation of deep brain-stimulating electrodes in the principal portion of the anterior nuclei, connected to a dual-channel Intercept (Medtronic) stimulation module. Stimulation parameters are 145 pulses per second with 90-μsec (approximately 0.1 msec) duration pulses, charge-balanced negative and positive on for 1 minute and off for 5 minutes. Voltage is set to 5 V in the treated group and 0 V in the placebo group. Since the patients, caregivers, or treating medical team cannot perceive the stimulation, this is a true double-blind study. After a 3-month baseline phase, patients receive either 5 V or 0 V stimulation for 3 months, then open-label therapy via a limited number of
parameter sets for 10 additional months. At time of this writing, the study had passed the midstudy “futility analysis” by the unblinded data-monitoring safety board, and the investigators have been given the go-ahead to complete the trial. Final outcome of the trial is not yet known.

In order to terminate clinical seizures by responsive stimulation, the seizure must be detected and treated before it spreads to a degree that it cannot be controlled with focal stimulation. This suggests that this therapy will be most effective if stimulation is delivered as early and precisely in the epileptic discharge as possible, usually before the onset of clinical symptoms. Responsive stimulation thus requires sufficiently accurate seizure prediction algorithms (SPAs) for EEGs recorded from intracranial electrodes.

All recent SPAs utilize a “sliding window analysis” in which a window of recorded EEG activity is mathematically analyzed using either linear or nonlinear algorithms. Linear algorithms calculate particular features directly from the EEG, including autocorrelation, spectral band analysis, curve length, accumulated energy, and high-frequency epileptiform oscillation^78,92 nonlinear algorithms. The EEG sliding window is reconstructed in three-dimensional phase space and analyzed using techniques such as the short-term maximum Lyapunov exponent, dynamic similarity, or correlation dimension.^78,92 All of these techniques have been purported to have their advantages in seizure prediction. However, as discussed at the recent Second Annual Seizure Prediction Workshop in April 2006,⁷⁷ there is still much work to be done in terms of understanding and applying the basics of seizure dynamics. Future techniques will likely apply multichannel, multialgorithm integrated analysis, utilizing a continuous probability curve rather than binary thresholding. In addition, the relevance of high-frequency epileptiform oscillations prior to seizure onset will need to be determined and incorporated into future SPAs. It is likely that future seizure prediction and detection in an individual patient will utilize more than one SPA, and that the optimal SPA application profile will differ from patient to patient.

In humans, cortical stimulation has long been known to abort spontaneous or induced epileptiform activity during brain mapping in surgical patients. Penfield and Jasper first applied focal electrical stimulation over 50 years ago to terminate spontaneous seizures detected by electrocorticography (ECoG) at the time of resective surgery. More recently, stimulation has been used to shorten or terminate epileptiform afterdischarge activity in patients implanted with intracranial subdural electrode arrays. During the brain-mapping procedures in these implanted patients, brief electrical stimulations were applied to the cortex, with the stimulus amplitude progressively increased until a clinical alteration or subclinical electrographic afterdischarge resulted. In 17 patients with subdural electrodes who underwent functional mapping, control afterdischarge duration was compared with afterdischarges that received brief bursts of stimulation.⁷² A biphasic 50-Hz pulse was presented for 0.3 to 2 seconds at the same stimulus intensity used to generate the afterdischarge. The duration of the afterdischarges was significantly shorter in those patients receiving stimulation (Table 1).

Brief pulse durations (0.5 to 1 second) were more effective in terminating afterdischarges than were stimulations of longer duration. This and other studies suggest that the exact stimulation parameters and timing may be very important in determining whether stimulation will disrupt or further entrain electrographic discharges.

Limited animal model and human experience pertains to responsive stimulation. In the genetic absence epilepsy rat of Strasbourg (GAERS) rat, which has a genetic disposition to spontaneous generalized absence seizures, responsive stimulation delivered to the subthalamic nucleus upon detection of
spike-and-wave epileptiform activity was effective in suppressing seizures.¹⁴⁷ Interestingly, continuous stimulation had no effect on seizure frequency in this model. Responsive stimulation has also suppressed spontaneous seizures in cats¹⁰⁰ and in rat hippocampal slices.⁹⁴

While still in its infancy, the field of responsive stimulation already has examples of effective clinical application in human epilepsy patients. Osorio et al.⁹⁷ at the University of Kansas applied high-frequency electrical stimulation that was delivered either directly to the epileptogenic zone (local closed loop, n = four patients) or to the bilateral anterior thalamic nuclei (remote closed loop, n = four patients) in response to every other automated seizure detection. The eight patients had a baseline video-EEG monitoring with implanted subdural and depth electrodes that localized epileptic foci and quantified seizure frequency using a linear simple genetic algorithm (SGA). Patients determined to have multifocal onsets were implanted into the bilateral thalami for remote stimulation, while local therapy was delivered to patients with a precisely localized epileptogenic focus. A total of 1,491 stimulations were delivered, 0.2% of which triggered afterdischarges. The mean reduction in seizure rate in the local closed-loop group was 55%; in the three responders the mean decrease was 86%, with two patients rendered seizure free during the local stimulation therapy. In the remote thalamic closed-loop stimulation, the mean seizure reduction rate was 41%, with a 74% reduction in the two responders.

More recently, patients undergoing subdural and depth electrode monitoring for seizure localization and functional mapping were enrolled in a trial testing an externalized version of an implantable responsive neurostimulator (eRNS, NeuroPace, Inc., Mountain View, CA).⁶⁴ This device used linear SPAs that could be tuned to patient-specific epileptiform activity to deliver electrical stimulation through up to eight contacts of a combination of subdural and/or depth electrodes implanted at the epileptogenic zone. Of 50 enrolled patients, 40 received responsive stimulation. The experience with four of these patients was subsequently reported. In this trial, electrographic seizures were altered and suppressed in these patients during trials of neurostimulation, with no major side effects. In one patient, stimulation appeared also to improve the baseline EEG.

Implantation of an internalized version of this responsive neurostimulator (RNS) system was subsequently investigated in a multicenter clinical trial assessing feasibility of the device’s clinical implantation and implementation. For this trial, enrollment included subjects 18 to 65 years with intractable partial-onset seizures and localized epileptogenic-onset region or regions. Subjects with at least 12 simple partial (SP) sensory or motor, complex partial (CP), or generalized tonic–clonic (GTC) seizures over an 84-day baseline period qualified for implant. The responsive stimulator was connected to up to four contact leads (subdural and/or depth), which were targeted to the seizure focus. This trial’s results have been reported in limited fashion. A single center’s experience with eight implants resulted in 45% reduction in seizure frequency in seven of eight patients over more than 9 months’ follow-up.³⁰ For the multicenter trial, efficacy was assessed during the most recent 84 days for which a subject could have received therapy. Defining response as at least a 50% reduction in seizures, the responder rate for 56 subjects was 36% for CP, 50% for GTC, and 36% for totally disabling (TD) seizures. The median percentage reduction was CP 28%, GTC 50%, and TD 30%; seizure reduction was significant for CP (p <0.005), GTC (p <0.02), and TD (p <0.001). In 65 implanted subjects, including 17 device replacements, there were no serious unanticipated device-related adverse events. Responsive neurostimulation was well tolerated by the patients.³⁷

In follow-up to the feasibility trial, a clinical efficacy trial began enrollment in the fall of 2006. In contrast to available implantable deep brain stimulation systems, in which the pulse generator is implanted below the clavicle, the RNS system has two four-contact subdural strip and/or depth electrodes that are attached to a seizure detection and stimulation delivery device that is implanted directly into the skull (Fig. 2). In our center’s experience, the most commonly encountered current applications for the RNS trial are dominant mesial temporal seizure onsets in patients with preserved hippocampal function, bitemporal epilepsy with mesial temporal onsets, and seizure onsets arising from a functionally “eloquent” area such as essential language cortex.

Responsive stimulation therapy for epilepsy must be efficacious, well tolerated, and safe in order to offer an alternative to resective surgery or other stimulation modalities in patients resistant to antiepileptic medication. Therapy should be targeted at the epileptogenic region or at seizure propagation pathways and leave other regions of the brain unaffected. It need not interfere with normal brain function, have an acceptable false-positive detection rate, and be stable as a therapy over time. It must also not result in chronic brain damage or the generation of new epileptic foci.

In addition to responsive stimulation, a device with accurate SPA technology will likely be paired in the future with novel treatment strategies such as convection-enhanced local drug delivery¹²⁶ or focal cooling¹¹³ to treat epilepsy in closed-loop responsive fashion. Future therapy may be delivered to the epileptic focus, to the epileptogenic region, to propagation pathways, or to deep brain structures. Whether a local responsive treatment can both suppress seizures and inhibit epileptogenesis remains to be determined.

As discussed above, SPA technology will certainly improve and need to be optimized on a case-by-case basis. The ability of responsive stimulation to then terminate detected seizures may depend on a variety of stimulation parameters including exact timing of delivery, current intensity, stimulus duration, the stimulus waveform, pulse frequency, and the delivery of the
stimulation in relation to the morphology of the epileptiform activity. These are all areas requiring further basic and human investigation.

The exact localization of the epileptogenic zone is critical for the future efficacy of responsive therapies. Currently, many mesial temporal and nearly all neocortical responsive stimulation patients require invasive monitoring to determine their seizure onset location or locations. Advances in techniques such as magnetic resonance imaging (MRI)–based localization of interictal and ictal activity have great promise in determining where to apply responsive treatment.

TMS has the potential for excellent spatial and temporal resolution. A 0.1-msec stimulation induces a current lasting the same duration and confined to 1 cm² of cortex with a maximum depth of about 2 cm.^57,112 When the motor cortex is stimulated, the excitation affects corticocortical neurons predominantly with subsequent activation of corticospinal neurons. The produced movement is a motor-evoked potential (MEP) with measurable conduction time and amplitude. Beyond the stimulation of motor cortex, all cortex that is adjacent to the cranium may be stimulated. Depending on the stimulation parameters, TMS may produce transient dysfunction for purposes of functional mapping or may temporarily alter the intrinsic excitability of the local gray matter. It is this capacity to alter the intrinsic excitability of cerebral tissue that is the foundation of TMS as a potential treatment for epilepsy.

Individual TMS pulses provide a means to measure cortical excitability through several parameters, including intensity thresholds, post-MEP (central) silent periods, and paired-pulse techniques that are similar to those used in basic neurophysiology.¹³¹ These techniques assess different aspects of cortical excitability. Threshold studies apparently represent pyramidal cell membrane excitability as the results are affected by sodium and calcium channel–blocking medications. Central silent period studies likely represent intracortical inhibition and paired-pulse studies are influenced by networks beyond the local cortex. The results of both are dependent on glutamatergic, dopaminergic, and GABAergic (γ-aminobutyric acid) activity.

Single and paired-pulse stimulations do not produce a sustained alteration to the cortical excitability, but such alterations may occur when the stimulation is delivered in trains of pulses. This technique is called repetitive TMS (rTMS). Low-frequency rTMS (LF-rTMS) is defined as pulse trains at or below 1 Hz and high-frequency rTMS (HF-rTMS) as pulse trains above 1 Hz and usually around 20 Hz. The 1-Hz division has practical implications because LF-rTMS is associated with decreases in cortical excitability and HF-rTMS is associated with increases. The occurrence of seizures as an adverse effect of TMS most often occurs during HF-rTMS. Therefore, LF-rTMS is the technique that has been most explored as a treatment for epilepsy. This approach partly depends on abnormal excitability of epileptic cortex, as documented in clinical and animal studies. LF-rTMS delays the development of seizures in pentylenetetrazol-exposed and kindled rats.¹

In humans, the reports of rTMS efficacy for seizures have varied considerably with respect to the types of patients included and outcome. Direct electrical stimulation through subdural grid electrodes can decrease both interictal epileptiform discharges and electrographic seizures in patients.^60,116 In a case report of a patient with a focal cortical dysplasia, 100 pulses of LF-rTMS were delivered to the area over the dysplasia twice a week for 4 weeks with a resulting 70% reduction in seizure frequency and a 77% reduction in interictal epileptiform discharges.⁸⁸ A report of two patients with epilepsia partialis continua describes each receiving 15 minutes of intermittent HF-rTMS and one becoming seizure free for 2 weeks. Both patients demonstrated a decrease in seizure-related hyperperfusion on single photon emission computed tomography (SPECT).³⁹

A series of five patients with intractable focal epilepsy received LF-rTMS over 3 months with modest efficacy.¹¹ Collective reduction in seizure frequency was 23%, but only one individual within the group demonstrated a significant reduction, which was 43%. The others showed no significant change in mean daily seizure number. In a series of nine patients with various types of focal epilepsy, 1,000 pulses of LF-rTMS were delivered to the vertex daily for 5 days. This produced a seizure frequency reduction >50% for three patients, a 20% to 50% reduction for three patients, a 20% reduction for one patient, and no improvement for the remaining two patients.¹³³ Those who benefited continued at the reduced frequency for 6 to 8 weeks. A series of seven patients with extratemporal lobe epilepsy received stimulation with 810 pulses of LF-rTMS daily for 5 days.⁵⁹ The average frequency of complex partial seizures decreased by 36%, but the decrease in total seizures did not reach statistical significance.

Theodore et al. delivered 30 minutes of LF-rTMS each day for a week to the scalp over the identified epileptic focus in a series of 24 patients.¹³⁵ Using a controlled, blinded study design, they observed a 36% reduction in seizure frequency for those patients with neocortical epilepsy and a 12% reduction for those with mesial temporal lobe epilepsy, but the difference was not statistically significant. A similarly designed study with baseline, intervention, and follow-up periods included four patients and observed a seizure frequency reduction >50% in three patients.¹¹⁵ This study differed from the prior by focusing the stimulation with frameless stereotaxy at a well-defined epileptogenic region, which may explain its more uniform results.

Overall, rTMS has demonstrated promise as a treatment for some forms of epilepsy. The rapid decrement of magnetic field strength with distance suggests that rTMS might be most useful with superficial epileptogenic regions, such as with cortical dysplasias. Further development of rTMS as a treatment will require controlled studies that are designed to determine efficacy and identify patient characteristics that are most associated
with successful treatment. Technical development of rTMS should establish optimal stimulation parameters for affecting seizure frequency by addressing the questions of stimulation intensity, frequency, duration, and targeting. If future clinical investigations identify rTMS methods that produce consistently superior seizure control results, epilepsy care would have the additional benefit of gaining a safe and highly tolerable treatment.