Seven patients suffering from intractable epilepsy participated in the study. Intracranial invasive monitoring was indicated and offered because of the presence of conflicting scalp EEG data (except in patient H3 who directly underwent resective neurosurgery). Patients underwent stereotactic depth electrode implantation which positions were based on clinical findings, previous scalp EEG and imaging studies.

Five patients with temporal lobe abnormalities and two patients with cortical dysplasia participated. Neuronal activity could be identified in four patients with hippocampal microstimulation and in one patient with cortical dysplasia located in the anterior cingulum (Table 1). In the case of patient H3, microstimulation was performed under general anesthesia (Target-controlled infusion, TCI, Base Primea, Fresenius-Vial, Brezins, F; 3.2 μg.ml⁻¹ propofol; Schnider et al., 1988, 1999; 0.3 ng.ml⁻¹ sufentanil; Gepts et al., 1995).

This work was conducted according to the ethical guidelines of the Declaration of Helsinki and was approved by the Ethical Committee of the University Hospitals of Geneva (CER 05-218, 14-076). Individually signed consent forms from patients were collected.

Microelectrodes, made each of eight microwires, were located at the tip of the macroelectrodes (WB09R-SP00×, Ad-Tech Instruments, Racine, WI, USA); macroelectrodes were all implanted stereotactically (each macroelectrode with eight macrocontacts, BF09R-SP05X). These platinum microelectrodes present an impedance of 1 MOhm, with a diameter of 40 μm.

Multi-Unit Activity was acquired using a 20 kHz sampling frequency recording system (Inomed Medizintechnik GmbH, Teningen, Germany; [150–3000Hz]). Noise threshold was fixed for comparison across all three conditions, i.e., before, during and after stimulation. Spike detection and clustering was realized with an algorithm introduced by Quiroga et al. (2004) and used in previous studies (Viskontas et al., 2009). The author RT visually inspected the clustering results.

The microstimulation was biphasic charge balanced pulses, cathodic first, 60 μs/phase, sent at a frequency of 1 Hz for a maximum of 2 min (NimEclipse System, Medtronic, Columbia Heights, MN, USA). The stimulation was monopolar, using one surface skin electrode as the return electrode (Neuroline ground, Ambu, Ballerup, Danemark). The voltage was adapted for a current target of 25 μA. Hence, the charge density did not exceed the accepted threshold of 150 μC/cm²/phase for platinum material, considering that microelectrodes were 40 μm in diameter (Merrill et al., 2005).

A preoperative MRI was performed with a Siemens Trio 3.0T scanner using a 32-channel brain coil. The technical protocol was T2 FSE coronal (TR/TE 7520/114 ms, in-plane resolution 0.5 × 0.4 mm, slice thickness 3 mm), 3D T1 mp2rage (T/RTE 5000/2.89, in-plane resolution 1 × 1 mm, slice thickness 1 mm), DTI (TR/TE 8000/84, in-plane resolution 2 × 2 mm, slice thickness 2 mm, 30 diffusion directions), and 3D FLAIR (TR/TE 5000/419, in-plane resolution 0.9 × 0.9 mm, slice thickness 0.9 mm).

Postoperative CT was performed with a Siemens Somatom Definition Flash (Siemens, Erlangen, Germany). Slice thickness was 1.25 mm.

Preoperative high-resolution 3D sequences obtained at 3T and postoperative CT series were fused using commercially available software (Integrated Registration, AW Volume Share 5, GE Healthcare). Maximum intensity projections (MIP), multi-planar reformatting (MPR) and volume rendering (VR) were performed for better localization of the electrodes. Fused images illustrated location of microelectrode arrays (not performed for patient H3, H6 and H7 Supplementary Figure S1, Table 2).

Analyses were performed using SigmaStat 3.11 (Systat Software Inc., Richmond, CA, USA). The differences in firing rates, described by the number of spikes per second, for every different condition before, during and after stimulation, were assessed by a Mann-Whitney Rank Sum Test, their distributions being not normal. Every different period, i.e., before, during and after, lasted at least for 1 min; specifically, at least 60 data points were collected per period. In addition group comparisons were performed with non parametric Friedman test comparing periods of exactly 60 s, as exact sample sizes are required for this statistic test. The first 60 s of the stimulation periods were compared to the latest 60 s of the pre-stimulation periods. The first 60 s of the post stimulation periods were compared to the latest 60 s of the stimulation periods.

The excitation induced by hippocampal low-frequency microstimulation, head or subiculum, supports the epileptogenic effect of low-frequency macrostimulation at this anatomical target (Boëx et al., 2007). Specifically, low-frequency macrostimulation (5 Hz) of the hippocampus resulted in an increase of interictal epileptiform discharges that can potentially generate seizures. In agreement with this literature, the present study suggests that the epileptogenic effect of low-frequency hippocampal macrostimulation observed in patients with epilepsy could be related to excitation at the neuronal level. A possible explanation could be a glutamaergic release induced by stimulation in agreement with principal hippocampal cell types (Cavus et al., 2016). Because of the low proportion of hippocampal GABAergic cells, these cells need to be specifically targeted to produce inhibition of the hippocampus as performed with optogenetic studies of the hippocampus (Krook-Magnuson et al., 2013).

If low-frequency hippocampal macrostimulation (5 Hz) or microstimulation (1 Hz) are both consistent with an epileptogenic effect in patients, this seems in contradiction with a decrease in neuronal activity reported with other hippocampal low-frequency stimulations performed in patients with epilepsy shown to be efficient in reducing interictal discharges (Alarcón et al., 2012; during 0.1 Hz, 1 ms, 6–8 mA, monophasic square pulse). Studies performed in kindled rats, supported inhibition or depression effects of low frequency direct hippocampal stimulation (Mohammad-Zadeh et al., 2007; Zhang et al., 2009; both monophasic square pulses). Nevertheless, within these studies, the applied stimuli differ markedly from the stimuli used in the present group of patients. Stimuli applied in all chronic DBS studies performed in human are all charge-balanced pulses, such as pseudo-monophasic (e.g., Medtronic stimulators, Medtronic Inc., Minneapolis, MN, USA; Boston Scientific stimulators, Boston Scientific, Marlborough, MA, USA; St. Jude Medical, St. Paul, MN, USA) or biphasic pulses (e.g., the present study) to avoid lesion in the neural tissue (Lilly et al., 1955; Mortimer et al., 1970). Indeed, no charge-balanced stimuli, such as monophasic-square pulses can create damage to the tissue because of the resulting products of non-reversible electrochemical reactions as known and applied in human safe electrical protocols (Merrill et al., 2005). As a consequence, lesions of the surrounding cells can contribute to produce inhibition of neuronal activity (Piallat et al., 2009) and contributed to overestimating the inhibitory effects reported in animal studies using low-frequency stimulation (Brummer and Turner, 1977).

While low frequency stimulation of the head of the hippocampus or of the subiculum does not appear to be anti-epileptogenic (Boëx et al., 2007), ventral hippocampal commissure (Kile et al., 2010; Rashid et al., 2011) or fornix (Koubeissi et al., 2013) stimulation could be anti-epileptogenic. Local neuronal excitation can have different effects on the networks to which these neurons are part of. They are certainly different between hippocampal per se or hippocampal commissural stimulations.

Cortical dysplasia is also a common cause of medically intractable epilepsy in both children and adults (Goldberg and Coulter, 2013). In patient D1, microstimulation of the cortical dysplasia led to a sustained increase of the firing rates. Hence, microstimulation can also modulate neocortical neuronal activity. Changes in cortical firing rate have been suggested as a means of seizure prevention (Truccolo et al., 2011), as supported by long-term low-frequency macrostimulation that has been shown to be anti-epileptogenic in patients with cortical epileptic focus (Matsumoto et al., 2005; Elisevich et al., 2006; Yamamoto et al., 2006; Hsu et al., 2011). The mechanism underlying the increase in firing rate with low-frequency stimulation can be found in an increased cortical excitability resulting from a lack of inhibitory interneurons. Indeed, experimental models of cortical dysplasia suggest a decrease of inhibitory interneurons as one of the main causes of epileptogenesis with an imbalance between excitatory and inhibitory activity (models obtained by irradiating rats in utero, Calcagnotto et al., 2005) and an alteration in firing rates and patterns of these interneurons (Zhou and Roper, 2011).

Studies of cortical microstimulation in monkeys have been described. High-frequency microstimulation can produce what has been recently described as “neural hijiacking” where the electrical stimulation eliminates and replaces the ongoing natural activity without summation, again demonstrating the possibility of cortical neural excitation with microstimulation (Griffin et al., 2011; Cheney et al., 2013).

In conclusion, microstimulation effects on microelectrode recordings were here studied for the first time in patients with epilepsy. Low-frequency microstimulation of the hippocampus or of cortical dysplasia was demonstrated as sufficient to increase spike frequencies, exciting the neuronal activity of cerebral areas involved in refractory epilepsy. In agreement with studies applying realistic stimuli, local neuronal excitation appears as the principal local mechanism of action of safe DBS. The effects of this local neuronal excitation are then dependent principally of the cerebral networks these neurons are involved and on the frequency of the stimulation, which were not studied here.