Presence of synchrony‐generating hubs in the human epileptic neocortex

•Initiation of pathological synchronous events such as epileptic spikes and seizures is linked to the hyperexcitability of the neuronal network in both humans and animals. •In the present study, we show that epileptiform interictal‐like spikes and seizures emerged in human neocortical slices by blocking GABAA receptors, following the disappearance of the spontaneously occurring synchronous population activity. •Large variability of temporally and spatially simple and complex spikes was generated by tissue from epileptic patients, whereas only simple events appeared in samples from non‐epileptic patients. •Physiological population activity was associated with a moderate level of principal cell and interneuron firing, with a slight dominance of excitatory neuronal activity, whereas epileptiform events were mainly initiated by the synchronous and intense discharge of inhibitory cells. •These results help us to understand the role of excitatory and inhibitory neurons in synchrony‐generating mechanisms, in both epileptic and non‐epileptic conditions.


Introduction
Understanding the role of different neuron types in the generation of physiological and pathological synchronies is crucial for identifying what makes a brain region predisposed to generate hypersynchronous events, such as epileptic seizures and interictal spikes. Abundant data describe the properties and behaviour of cells and neuronal circuits during epileptic activity in animal models (de Curtis & Avanzini, 2001;McCormick & Contreras, 2001;Avoli et al. 2002;Trevelyan et al. 2015), although information about the human cells and disease is considerably less detailed (Avoli & Williamson, 1996;Avoli et al. 2005). Knowledge about the cellular and network mechanisms related to synchrony generation mainly derive from the hippocampus and the surrounding medial temporal areas, whereas the role of different neuron types in synchronization processes of other neocortical regions remains mainly uncovered. In the present study, we explored the firing pattern of human neocortical single cells and microcircuits during synchronizations, as well as their possible relationship with epilepsy. We induced epileptic seizures and interictal spikes with the GABA A receptor antagonist bicuculline (BIC) in human neocortical slices derived from epileptic and tumour patients and compared these pathological events with synchronous population activity spontaneously occurring in physiological solution. We aimed to obtain insight into the cellular mechanisms by analysing the discharge properties of clustered excitatory and inhibitory neurons during both epileptiform and physiological synchronous events.
The cellular features and the firing pattern of single neurons provide important information about how synchronous events are initiated. The bursting behaviour and the paroxysmal depolarization shift of excitatory principal neurons have been linked to the initiation of interictal discharge in animals (de Curtis & Avanzini, 2001). However, in humans, neither in vitro (Prince & Wong, 1981;Avoli & Olivier, 1989;Williamson et al. 2003), nor in vivo early work (Calvin et al. 1973;Ishijima et al. 1975;Wyler et al. 1982;Staba et al. 2002) could find direct evidence for the relationship between bursting behaviour and interictal spike generation. More recently, an in vivo study reported that cells with modulated firing during interictal spikes have higher burstiness index than non-modulated cells (Keller et al. 2010). In postoperative human neocortical tissue, pharmacologically induced interictal-like activity was used to reveal the cellular properties presumably contributing to the generation of pathological synchronies. The leading role of excitatory cells and bursting behaviour in the generation of hypersynchronous events was supported by the fact that bursting neurons and depolarization shift were observed in the Mg 2+ -free  and in the K + -channel blocker 4-aminopyridine (4-AP) (Mattia et al. 1995) models of epileptic activity. In the presence of the GABA A receptor antagonist BIC (Hwa et al. 1991) interictal-like activity was also reflected as bursts and a paroxysmal depolarization shift in human neocortical neurons at the intracellular level. By contrast to the Mg 2+ -free and the 4-AP models, where interictal-like and seizure activity spontaneously emerged, when applying BIC bath, electrical stimulus was needed to induce epileptiform synchrony.
Cellular synchronization mechanisms leading to the initiation of seizures in epileptic patients were investigated recently with the aid of intracortical microelectrodes. The dynamic balance of excitation and inhibition characterized the human neocortex in physiological conditions, which broke during epileptic seizures (Dehghani et al. 2016). Increased neuronal firing, up to several minutes before the onset of the seizure, was observed in both the neocortex (Truccolo et al. 2011) and the hippocampus (Lambrecq et al. 2017) of epileptic patients. Seizures with different onset patterns were shown to be generated by different mechanisms. The low voltage fast activity was the most often detected pattern in the neocortex (Perucca et al. 2014) and was shown to begin with the increased activity of inhibitory interneurons (INs) in the medial temporal lobe (Elahian et al. 2018) and other neocortical areas (Dehghani et al. 2016) of humans, as well as in animals (Weiss et al. 2019). The hypersynchronous seizure onset pattern was observed in patients with mesial temporal lobe epilepsy exclusively (Perucca et al. 2014) and was associated with enhanced excitatory processes in both human (Huberfeld et al. 2011) and experimental epilepsy (Kohling et al. 2016). Preceding the seizures of neocortical origin that spread to the medial temporal lobe, inhibitory but not excitatory neuronal discharge was decreased in the hippocampal formation . During the course of the seizures, heterogeneous firing patterns were detected in both the human hippocampus (Babb et al. 1987) and the neocortex, together with a high ratio of cells increasing their firing rate (Truccolo et al. 2011). Furthermore, spontaneous seizure-like events appeared in neocortical slices derived from epileptic patients, by applying kainate plus carbachol (Florez et al. 2015), as well as in the Mg 2+ -free model (Avoli et al. 1987;Mattia et al. 1995), with such seizures being reflected intracellularly as a large depolarizing shift and burst firing (Avoli et al. 1997).
Intracortical microelectrodes recording neuronal activity are implanted only in epileptic patients and, for obvious ethical reasons, such data cannot be obtained from healthy subjects. Our in vitro model has a considerable advantage compared to in vivo recordings: it has a double internal control. First, we can compare epileptic tissue with non-epileptic samples derived from tumour patients without clinical manifestations of preoperative epileptic activity. Second, we can compare synchronous activity spontaneously occurring in physiological solution [spontaneous population activity (SPA)] (Tóth et al. 2018) with epileptiform interictal-like discharges and seizures. SPA emerges in slices derived from patients both with and without epilepsy, shows several crucial differences compared to interictal spikes in vivo, and is considered to be associated with physiological processes (Tóth et al. 2018). However, limitations of the in vitro conditions are evident. The excised neocortical samples have partially cut internal (and do not possess external) connections and the bath solution is indisputably different from the cerebrospinal fluid of the living subjects. Although the resulting modified synaptic connections might affect the spontaneous activity of neurons, the double inter-nal control in our model system can help us to identify phenomena related to epileptic conditions. The present study investigated the firing properties of human neocortical neurons during physiological and epileptiform synchronies. Specifically, we aimed to describe the role of excitatory and inhibitory cells and circuits in synchrony generation by comparing the behaviour of neuronal networks and the discharge properties of single cells during SPA and disinhibitioninduced interictal spikes and seizures in a human in vitro model of epilepsy.

Ethical approval
The experiments included in the present study were conducted on resected human brain tissue. Written informed consent was obtained from all patients. Our studies conformed to the standards set by the latest revision of the Declaration of Helsinki, except for registration in a database. The procedures were approved by the Regional and Institutional Committee of the National Institute of Clinical Neuroscience, as well as by the Hungarian Ministry of Health and the Research Ethics of Scientific Council of Health (ethical licence number: ETT TUKEB 20680-4/2012/EKU). The study used a partly overlapping patient dataset with our previous work (Tóth et al. 2018).

Epileptic patients
Samples were resected from 30 epileptic patients (Table 1). We obtained epileptic neocortical tissue from frontal (n = 6 patients), temporal (n = 17 patients), parietal (n = 4 patients) and occipital (n = 3 patients) lobes. Most of the patients (n = 20) suffered from pharmacoresistant epilepsy (ResEpi, resistant epilepsy) for 28.1 ± 7.3 years on average. As in our previous study (Tóth et al. 2018), we categorized the remaining 10 patients as patients who were either seizure free with appropriate pharmacological treatment (TreatEpi, treatable epilepsy, n = 5 patients) or had one (provoked) seizure without the need for medication (NoMed, no need for medication, n = 5 patients) and were operated on to resect their tumour. Ten epileptic patients were diagnosed with cortical dysplasia, 10 patients with tumour of glial origin and three patients with carcinoma metastasis. The remaining seven patients had cavernoma (n = 3), gliosis (n = 2), haematoma (n = 1) and hippocampal sclerosis (n = 1) (Table 1). Histopathological changes (signs of dysgenesis or tumour infiltration) of the obtained tissue have been verified with Nissl staining; the neuronal marker, NeuN; the astroglial marker, glial fibrillary acidic protein; and a specific interneuron marker, parvalbumin-immunostain (Table 1) (Tóth et al. 2018). Epileptic patients comprised 17 females and 13 males, age range 18-68 years, mean ± SD = 35.1 ± 13.8 years.

Non-epileptic patients
Nineteen patients diagnosed with brain tumour but without epilepsy (NoEpi, no epilepsy) were included in the study (Table 1). These patients, as stated in their anamnesis, did not show clinical manifestation of epileptic seizure before the date of their brain surgery. Neocortical tissue was resected from tumour patients from frontal (n = 7 patients), temporal (n = 4 patients), parietal (n = 6 patients) and occipital (n = 2 patients) lobes. Thirteen patients were diagnosed with tumours of glial origin: glioblastoma (n = 10) or anaplastic astrocytoma (n = 3). Four patients were operated to remove their carcinoma metastasis, one patient had cavernoma and one had neurocytoma (Table 1). The distance of the obtained neocortical tissue from the tumour (Table 1) had been assessed by the neurosurgeon, based on magnetic resonance images, intraoperative pictures and occasionally defined by a navigational system. Non-epileptic patients comprised eight females and 11 males, age range 31-79 years, mean ± SD = 58.5 ± 14.6 years.

Recordings
The extracellular local field potential gradient (LFPg) recording was obtained as described previously (Tóth 5644Á. Kandrács and others J Physiol 597.23 et al. 2018). Briefly, we used a 24 contact (distance between contacts: 150 µm) laminar microelectrode (Ulbert et al. 2001;Ulbert et al. 2004a;Ulbert et al. 2004b;Fabó et al. 2008;Wittner et al. 2009) and a custom-made voltage gradient amplifier of bandpass 0.01 to 10 kHz. Signals were digitized with a 32 channel, 16 bit resolution analogue-to-digital converter (National Instruments, Austin, TX, USA) at a 20 kHz sampling rate, recorded with a home written routine in LabView8.6 (National Instruments, Austin, TX, USA) (RRID: SCR 014325). The linear 24 channel microelectrode was placed perpendicular to the pial surface, and slices were mapped from one end to the other at every 300-400 µm.

Drugs
A-type γ-aminobutyric acid (GABA A ) receptor-mediated signalling was suppressed by bicuculline methiodide (BIC, 20 µM or 50 µM), obtained from Tocris Bioscience (Izinta Kft., Hungary). During most of the experiments, 100 mL of solution containing BIC was washed into the interface chamber, and 10 min long epochs were recorded continuously during the whole experiment (control -BIC application -washout). The washout period endured until the reappearance of the SPA. In the absence of a reappearing SPA, the washout persisted for 1 h.

Data analysis
Analysis of SPAs and BIC-induced events. The presence or absence of BIC-induced activity was noted in the case of every slice. In several cases, only one 10 min epoch was recorded but, in most cases, three to five epochs were recorded when applying the BIC bath, and the one from the middle/end was chosen for detailed analysis when recurrent BIC-induced events occurred. In the case of single events, the file containing the event was analysed. Data were analysed with the Neuroscan Edit4.5 (Compumedics Neuroscan, Charlotte, NC, USA) and bespoke routines for Matlab (MathWorks, Natick, MA, USA) (RRID: SCR 001622) and C++. The microelectrode covered all layers of the neocortex. Usually, channels 1-12 were in the supragranular, channels 13-15 in the granular and channels 16-23 were in the infragranular layers. Channel positions were determined according to the thickness of the neocortex of the given patient and corrected if necessary.
Detection of both spontaneous and BIC-induced population activity was performed on LFPg records after a double Hamming window spatial smoothing and a band-pass filtering between 3 and 30 Hz (zero phase shift, 12 dB per octave). Events larger than two times the SD of the basal activity were detected and included in the analysis. The largest amplitude LFPg peak of simple events was chosen as time zero for averaging and for peri-event time histogram (PETH) analysis. Two types of synchronous events have been detected during the application of BIC: interictal-like spikes (IISs) and seizure-like events. IISs were further divided into separate groups according to their spatial and temporal complexity. Temporally complex events consisted of multiple LFPg deflections spreading to the same cortical layers. Spatially complex events united waves developing from different locations. In the case of complex events (such as complex IISs and seizures), the first LFPg peak on the channel where its amplitude was the largest was chosen as time zero. The location and frequency of the events (see below) were determined in each case. Current source density (CSD; an estimate of population trans-membrane currents) and multiple unit activity (MUA) were calculated from the LFPg using standard techniques (Tóth et al. 2018). The duration of the synchronous events was assessed by two different methods because of the differences in the waveform characteristics of spontaneous and induced synchronous events. The length of SPA events was measured on the channel displaying the largest LFPg amplitude, at 50% of that amplitude. To obtain an estimation for the total duration of the SPAs, the half-maximum length was multiplied by two. This method could not be applied to BIC-induced events because the length of the LFPg deflections were unequal on the different channels; furthermore, the peak of the deflection was at considerably different time point on the different channels. Therefore, the length of IISs was calculated using the mean global field power (MGFP) from the channels where the event was present. The MGFP corresponds to the spatial SD of field potential amplitude values obtained with multiple channel recording (Lehmann & Skrandies, 1980;Skrandies, 1990). The average length was calculated at 5% of the maximal MGFP amplitude. In the case of temporally complex IISs, the length of the first component was calculated. The MGFP method could not be applied to SPAs because the LFPg amplitude of these events is low, and therefore the 5% height of the maximal values dropped below the noise level and provided imprecise measurement. The length of BIC-induced seizure-like activities was assessed by manual estimation using the butterfly plot of all channels. In the case of recurring seizures, the butterfly plot was generated from the averaged LFPg, to include all seizure events.
The recurrence frequency was determined in each recording with population events (SPA, IIS, seizure). Note that, in the case of single epileptiform events, we calculated the recurrence frequency from one, usually 10 min long epoch containing the event, although, in most of these cases, multiple epochs were recorded.
Time-frequency analysis. We analysed the ripple and fast ripple components of SPAs, IISs and seizures with the aid of routines written in Matlab. Original 20 kHz sampling rate records were low pass filtered at 700 Hz and then down-sampled to 2 kHz. Wavelet analysis was applied on epochs from −1000 ms to 1047.5 ms for SPAs and from −500 to 1547.5 ms for IISs with the LFPg peak of the SPA/IIS at time zero (detected as described above). In the case of seizures, the epoch(s) ranged from −500 ms to 15,883.5 ms. Time-frequency analysis was performed between 0 and 700 Hz on the electrode channels where the SPA/IIS/seizure was present, and baseline corrected to −300 to −100 ms. For each channel, the maximal power change (relative to the baseline) was determined within the range from 130 to 250 Hz (ripple frequency) and 300 to 700 Hz (fast ripple frequency) at time zero (Tóth et al. 2018). The frequencies where the power showed the maximum were also determined. Both the ripple and fast ripple power, as well as the frequencies, were averaged across the channels; thus, one ripple and one fast ripple power and frequency parameter were determined for each recording. This last step was necessary for the comparison of synchronous activities spreading to different numbers of channels.

Analysis of the initiation site and the spread of
BIC-induced events. We determined the initiation site, as well as the spreading direction and speed of the BIC-induced events between the neocortical layers. Note that, in our analysis, spreading direction refers to interlaminar spread (i.e. propagation across cortical layers). In most cases, the amplitude of the LFPg transient evoked by BIC was large on certain channels but close to baseline (and therefore not detectable) on others. Additionally, an increase in cell firing was associated with the events. To consider both the deflection on the field potential signal and the cellular activity, we applied a method to detect the start of each event on each channel. After applying a 0.5 Hz high-pass filter, we calculated the power of the LFPg signal. The start of the event was defined via a two-step thresholding process. For each recording, we defined two different thresholds. Threshold 1 determined which peaks should be considered as part of the event, whereas threshold 2 determined where the first peak deviates from the baseline. In the case where threshold 1 was crossed multiple times during one event, the first crossing point was used. We chose threshold 1 manually for each file. We set the threshold to a level ensuring that the first main peak of the BIC-induced events was detected and matched to the previous detection (see above). The variation in MUA and LFPg amplitudes (during the baseline, as well as during the BIC-induced event) did not allow a general threshold rule. Note that threshold 1 was not used to detect the BIC-induced events within the recording because it was made separately. It was only used within the algorithm to define the beginning of each recording. Starting from threshold 1 time point, the signal was followed to earlier and earlier time points, until it fell below threshold 2 (which was set as low as possible to find the deflection point from the baseline). This time point was used as the start of the event. This process was repeated for every channel and every event. No start was detected for a specific peak if it did not surpass threshold 1. In the case of temporally and temporally + spatially complex events, we analysed only the first appearance of the event on a given channel.
The spreading speed of the events across the channels was calculated by dividing the delay between the starting points of the events by the distance between the two electrode contacts in question. This speed value was determined for every channel pair, up to five contacts apart. For each event, the speed was calculated by taking the median of these pairwise speed values.
Cell clustering. Single cells were clustered from records high-pass filtered at 80 Hz or at 500 Hz with the aid of a bespoke program for Matlab (Wave Solution). Only neurons with a clear refractory period of at least 1.5 ms were included. Cell clustering during physiological conditions and during the BIC bath was performed separately. Although we made consecutive (usually 10 min long) recordings, the identification of the clusters separated by 30-40 min was very uncertain or even impossible. The cell activity observed in the physiological solution was considerably changed during the application of BIC. The spontaneously active cells became silent during the application of the drug, whereas other neurons started to discharge (more details are provided in the Results). Changes in the extracellularly recorded action potential (AP) waveform were observed in vivo, during population activity: the amplitude of the APs decreased by ß12% when the firing frequency of the cell was high (Stratton et al. 2012). We considered this phenomenon and clustered our cells accordingly. Furthermore, similar to a previous study , we had difficulties in clustering single units during epileptic seizures because of the distorted APs. In these recordings, only cells with relatively high amplitude and recognizable AP were clustered, whereas small (and noisy) clusters were excluded. During the high frequency oscillatory phases of the seizures, cluster detection might be imprecise; therefore, conclusions about cell firing during seizure were made carefully.
AP waveform analysis and separation of principal cells (PCs) and interneurons (IN) was performed with the aid of a bespoke routine in Matlab. Two independent criteria were used to separate PCs and INs, unbiased by the tissue of origin. The duration of PC APs is significantly higher than that of INs (Wilson & McNaughton, 1993;Csicsvári et al. 1999b). We measured the AP width at one-half of the largest LFPg amplitude (half-width). The cell was considered to be PC if this value was larger than 0.4 ms, and IN if it was less than 0.2 ms. Cells with AP width between 0.2 and 0.4 ms were defined as unclassified cells (UC). Discharge dynamics was also considered when separating cell types (Barthó et al. 2004;Peyrache et al. 2012). A high peak at 3-10 ms followed by a fast, exponential decay on the autocorrelogram, was characteristic of 'intrinsically bursting' PCs. If the peak was lacking but there was a sustained firing, or the peak was >10 ms, the cell was considered to be a 'regular firing' PC. The remaining cells with half-widths of >0.4 ms were categorized as PCs with 'unclear firing' . A slow rise together with a slow decay identified INs. In the hippocampus and neocortex of the living animal, the firing frequency of the cells gives additional information on their identity: PCs have a significantly lower firing rate than INs (Buzsáki & Eidelberg, 1983;Csicsvári et al. 1999a). In vitro procedures considerably alter the living conditions of the cells. Therefore, we did not consider this criterion for the identification of the cell type.
Extracellularly recorded very short duration and triphasic APs are indicative of axonally running APs, and might be categorized as INs (Robbins et al. 2013). We examined this question in our neuron database and found six of 772 cells (0.8%) with very short half-width (<0.1 ms). These cells indeed showed a triphasic waveform, even after transforming the local field potential gradient recording into a 'referential-like' recording by referencing all channels to channel 1 (the one at the pial surface).
Analysis of single cell firing was performed with a home written routine in Matlab. The average firing frequency, interevent interval (event = AP), and a measure for burstiness (percentage of APs within bursts) were calculated for each cell. Bursts were defined as a set of (at least) three APs within 20 ms. Bursts containing more than three APs could be longer than 20 ms, although each group of three consecutive APs had to be within a 20 ms period. Moreover, the first AP of the burst had to be preceded and the last AP had to be followed by a 20 ms silent period (Staba et al. 2002).
Cell firing during SPA and BIC-induced events. We examined the discharge of neurons relative to the LFPg peak of the SPA, IIS and seizure events with two approaches. We made these analyses in each recording containing SPAs, IIS(s) and seizure(s) and single cell firing in slices from the ResEpi (n = 15 SPA, n = 14 IIS, n = 6 seizures) and the NoEpi (n = 14 SPA, n = 13 IIS, n = 3 seizures) groups. First, we aimed to describe the proportion of single cells that participate in the generation of SPA and IIS events. Accordingly, we generated PETHs for every cell/SPA, cell/IIS or cell/seizure comparison, with a bin size of 5 ms, from −150 to +50 ms around the LFPg peak of the SPAs and from −400 to +200 ms around the LFPg peak of the IISs and in the case of seizures. From these PETHS, we calculated the average firing frequency of the neurons during the events as well as during the baseline period. In the case of SPAs, the time window of the events was from −50 to 50 ms, whereas the baseline ranged from −150 to −50 ms before the LFPg peak of the SPAs. In the case of BIC-induced population activities, the time window of the events was defined between −100 to 200 ms, whereas the baseline ranged from −400 to −100 ms before the LFPg peak of the IISs or the seizures. The normalized firing change was calculated for each cell in the form of A/(A + B), where A is the average firing frequency during the time window of the events and B is the firing frequency during the time window of the baseline. Thus, all values fell between zero and one. A neuron was considered to have an increased firing if its firing change value exceeded 0.6 (which equals to an increase to 150% of its baseline firing rate).
With the second approach, we analysed how excitatory and inhibitory cell types participate in the initiation of SPA, IIS and seizure events. Therefore, we computed combined PETHs for all PC APs, for all INs APs and for all UC APs of the given recording, from −100 to +200 ms around the LFPg peak of the SPAs and −200 to +1000 ms around the LFPg peak of the IISs and seizures. First, we determined, in each recording, which cell type (PC, IN or UC) starts firing during the SPA/IIS/seizure. Furthermore, we estimated the contribution of the discharge of the different cell types in the initiation of the synchronous events. We calculated the area under the curve of the PC/IN/UC firing of the PETH relative to the total firing during the interval of −50 to +50 ms around the LFPg peak of SPAs, and −100 to +200 ms around the LFPg peak of IISs and seizures.

Statistical analysis
Statistical significance was determined using either Statistica, version 13 (Tibco Software Inc. Palo Alto, CA, USA) (RRID: SCR 014213) or Matlab. In the case of normal distributions (verified with the Kolmogorov-Smirnov and Lilliefors test), a t test was performed to compare two groups or a one-way ANOVA (with Tukey's honest different significance post hoc test) was performed to compare multiple groups, respectively. If the normality test failed, we used Mann-Whitney U test or Kruskal-Wallis ANOVA for comparing two or multiple groups, respectively. In the main text, we report the mean ± SD for clarity, whereas, in the tables, we present both the median (first and third quartiles), as well as the mean ± SD. To test for unequal proportions in contingency tables, we used http://vassarstats.net. In the case of 2 × 3 contingency tables, Fisher's exact probability test was applied if the total size of the data set was not greater than 300, and a chi-squared test was applied if it exceeded 300.

Patient groups
Patients were distributed into four groups by experienced neurologists (Table 1) (Tóth et al. 2018): (i) patients with pharmacoresistant epilepsy (resistant epilepsy, ResEpi); (ii) patients with generalized or focal tonic-clonic seizures who were seizure free with appropriate medication (treatable epilepsy, TreatEpi); (iii) patients with one generalized tonic-clonic seizure or with occasional (provoked) seizures, and with no need for medication (no medication, NoMed) (these patients were operated to resect their tumour); and (iv) patients without preoperative seizures (no epilepsy, NoEpi). Glial tumours are known to be highly epileptogenic. Therefore, the anamnesis of tumour patients was carefully checked, and they were asked about their possible epileptic episodes by expert neurologists. Patients with the smallest doubt of having preoperative paroxysmal event(s) were categorized as NoMed (or TreatEpi) patients. Furthermore, preoperative clinical EEG recordings on patients with glial tumours could not confirm the presence of any epileptic activity (Tóth et al. 2018). Patients belonging to the first three groups (ResEpi + TreatEpi + NoMed) were considered to be epileptic, whereas patients in the NoEpi group are referred to as non-epileptic. To examine changes related to epilepsy, we compared results deriving from the ResEpi group with those from the NoEpi group.

Emergence of spontaneous activity in physiological bath
SPA was generated in the human epileptic ( Fig. 1 and Table 2) and non-epileptic neocortex in vitro, in physiological bath solution (Köhling et al. 1998;Kerekes et al. 2014;Pallud et al. 2014). Note that SPA designates the totality of recurring synchronous events in a given recording. The number of events per recording varied from 21 to 1229. Single population events will be referred to as 'SPA event' . The emergence rate of SPA was similar in slices derived from epileptic tissue affected vs. not affected by the cortical dysgenesis and it did not differ in samples derived from patients having tumours of glial origin vs. having carcinoma metastasis (Tóth et al. 2018). Occasionally, notably larger and more complex events also spontaneously emerged in tissue from epileptic patients, which were considered to be interictal-like discharges (not shown, n = 2/43 slices) (Tóth et al. 2018). We examined 26 slices with SPA and 17 slices without SPA derived from 30 epileptic patients, and 20 slices with SPA and six slices without SPA in tissue from patients in the NoEpi group (Table 2). The GABA A receptor antagonist BIC abolished the SPA in all cases. In 19 of 25 cases, SPA reappeared when BIC was washed out, whereas four slices from ResEpi and two slices from the NoEpi group did not generate SPA after the BIC washout. SPA appeared after the BIC washout in one slice from the NoMed group, which originally did not generate SPA. Interictal-like discharges spontaneously emerging in physiological bath solution also disappeared during the application of BIC but were not further analysed in the present study.

BIC-induced epileptiform activity
The application of a BIC bath resulted in the generation of spontaneous epileptiform activity in 27 of 43 and in 15 of 26 slices in tissue derived from epileptic and non-epileptic patients, respectively (not different, chi-squared, p = 0.87) ( Table 2), which disappeared when BIC was washed out. This contrasts with previous results (Schwartzkroin & Haglund, 1986;Hwa et al. 1991), where electrical stimulation was also needed to achieve the emergence of interictal-like spikes in human epileptic neocortical slices during BIC application. Bicuculline-induced activity appeared either as IIS or as seizure-like activity (Fig. 1). Note that the term 'interictal-like discharge' is used for spontaneously occurring epileptiform activity in vitro (see above), whereas IIS is used to designate BIC-induced interictal-like activity. Note also that, when referring to IIS, we mean the totality of interictal-like spike events in a given recording. In the ResEpi group, IISs invaded either the entire width of the neocortex, or were restricted to the supragranular + granular (supra-gran), or to the granular + infragranular (gran-infra) layers (Fig. 2). In all other patient groups, IISs always spread to the entire neocortex ( Fig. 2D and Table 3). Based on temporal and spatial complexity, we further classified BIC-induced events as simple or complex events. In the ResEpi group, slices generated different types of IISs (temporally and spatially simple-simple, complex-simple, complex-complex, respectively), whereas, in the NoEpi group, only simple-simple IISs were detected ( Fig. 2 and Table 3). Temporally complex IISs consisted of more than one LFPg deflections, with a total duration of up to 5 s. Spatially complex IIS events (n = 2 activities) were variable in their spatial pattern. In both cases, the first spike of the spatially complex events usually invaded the entire width of the cortex, whereas, during the following spikes of the IISs, the supragranular and infragranular layers were differently involved from event to event (Fig. 2E).
Seizures were 15-28 s long, temporally complex epileptiform events, consisting of recurring interictal-like

. Different population activities in the human neocortex in vitro
A-C, SPA (A) was generated in human neocortical slices derived from patients with or without epilepsy. During the application of the GABA A receptor antagonist BIC, spontaneous epileptiform activity emerged, such as IIS (B) or seizures (C). SPAs were usually restricted to a subset of neocortical layers (see the LFPg traces, CSD and MUA maps), whereas IISs and seizures mainly invaded the entire width of the neocortex. IISs and seizures were larger LFPg and CSD amplitude events, with highly increased neuronal firing (higher MUA amplitude) than SPAs. An increased HFO activity in the ripple and fast ripple bands (arrows) was characteristic of SPAs, whereas, during IIS We distinguished four categories based on the presence/absence of SPAs in physiological bath solution and the subsequent presence/absence of BIC-induced activity. All Epi = ResEpi + TreatEpi + NoMed. Entire = involving the entire width of the neocortex, supra-gran = involving supragranular + granular layers, gran-infra = involving granular + infragranular layers.
spikes, always invading the entire width of the neocortex (Fig. 3). The occurrence of seizures was somewhat higher in the ResEpi group (n = 7/31; 22.6% of all BIC-induced activities) than in the NoEpi group (n = 3/27; 11.1%, significantly not different, chi-squared, p = 0.25). In six slices (five from ResEpi, one from NoEpi), we detected spatially complex seizures (i.e. parts of the seizure pattern were confined only to the supragranular + granular or to the granular + infragranular layers). The other four seizures (n = 2 in ResEpi slices and n = 2 in NoEpi slices) were spatially simple events. Three slices consecutively generated two different activities (one slice in ResEpi: first temporally complex, then simple IISs; one slice in TreatEpi: first IIS, then seizure; one slice in NoEpi: first seizure, then IIS). Furthermore, one slice from ResEpi simultaneously generated two independent and spatially distinct IISs (one supra-gran and one gran-infra). In this case, the activity emerging in the supragranular layers was similar to SPAs generated in physiological solution. Considering these as distinct activities, we analysed 16 IISs and seven seizure activities in slices derived from ResEpi, four IISs and one seizure from TreatEpi, one IIS and one seizure from NoMed and 13 IISs and three seizures from NoEpi.
In summary, we distinguished four types of synchronous activities in the human neocortex in vitro. SPAs (i) were spontaneously generated in physiological solution in all patient groups, whereas spontaneous interictal-like discharges (ii) appeared only in epileptic patients (not investigated in the present study; for details, see Tóth et al. 2018). In BIC bath, these two spontaneous activities disappeared, and two other types and seizures, all frequency bands showed an increased power. For this IIS example, we detected a long-lasting fast ripple power increase (arrow). The CSD, MUA and HFO heat maps were computed from SPA and IIS averages, whereas that of epileptic seizure was performed from a single event. Note the different colour and time scales. D, recurrence frequency of SPAs was significantly higher, whereas the LFPg and MUA amplitudes were significantly lower than that of epileptiform events. Furthermore, MUA amplitudes of NoEpi IIS and seizure were significantly higher than that of ResEpi slices. * P < 0.05 [Colour figure can be viewed at wileyonlinelibrary.com]

Figure 2. Different types of BIC-induced IISs in the human neocortex
IISs were grouped based on their temporal and spatial complexity. In most cases, IISs spread to the entire width of the neocortex (A), although spatially more restricted IISs (B and C) were also detected in ResEpi samples. Temporally complex events consisted of more than one LFPg deflection, such as complex-simple (A and B) and complex-complex IISs (A). Three of the IIS events shown on (A) are magnified in the second row. Note the difference in the length of the two simple-simple events (same time scale). In the case of spatially complex events, the neocortical layers were separately activated: complex-complex events on (A), with a high variability from event to event (E). D, distribution of the different types of IISs in relation to the patient groups. In NoEpi tissue, only simple-simple IISs were detected, whereas, in ResEpi samples, several different types of IISs occurred. Supra-gran, supragranular + granular; Gran-infra, granular + infragranular; simp, simple; comp, complex of synchronies emerged in tissue from both epileptic and non-epileptic patients: IISs (iii) and seizures (iv). IISs always had a spatially and temporally simple morphology in NoEpi slices, whereas they showed several combinations of spatially and temporally simple and complex manifestations in the tissue of ResEpi patients.

Characterization of SPAs and BIC-induced activities
To shed light on characteristics related to epileptic mechanisms, in this analysis, we compared slices derived from patients in the ResEpi group with those from the NoEpi group. In 12 of 16 (75%) and six of 13 (46%) cases in the ResEpi and NoEpi groups, respectively, the slices generated recurrent IISs, whereas, in the remaining four and seven cases, respectively, one single IIS event was detected during the entire recording interval (see Methods). The recurrence frequency of SPAs was 0.81 ± 0.54 Hz and 1.08 ± 0.64 Hz in slices derived from ResEpi and NoEpi groups, respectively ( Fig. 1D and Table 4); see also Tóth et al. (2018). The recurrence frequency of all IISs was significantly lower than that  of SPAs (ANOVA, P < 0.0001): in ResEpi slices, it was 0.09 ± 0.15 Hz (5.23 ± 9.03 min −1 ), whereas it was 0.02 ± 0.02 Hz (0.88 ± 1.09 min −1 ) in NoEpi slices (ResEpi and NoEpi are significantly not different, t test, p = 0.16). Recurrent seizures emerged only in tissue from epileptic patients (6/7 seizure activity in the ResEpi group; 1/1 in TreatEpi group), whereas all three seizures generated in slices from NoEpi group were single seizures. The frequency of seizures was very low: 0.007 ± 0.007 Hz (0.42 ± 0.42 min −1 ) in ResEpi slices and 0.002 ± 0.001 Hz (0.14 ± 0.05 min −1 ) in NoEpi slices (significantly different from SPA, Kruskal-Wallis ANOVA, P < 0.0001, although not different from IIS). For easier reading, we report the mean ± SD in the text, whereas both the median (first and third quartiles) and the mean ± SD are provided in the tables.
The duration of the different synchronous activity types (SPA, IIS, seizure) were calculated using different methods, thus, they cannot be compared. The average length of the SPAs was 68.8 ± 36.6 ms in ResEpi slices and 41.6 ± 18.9 ms in NoEpi slices (ResEpi and NoEpi are significantly different, t test, P < 0.01). The average duration of the IISs was 0.36 ± 0.23 s in the ResEpi group and 0.44 ± 0.16 s in the NoEpi group. The average lengths of seizures were 21.75 ± 4.44 s and 24.19 ± 3.48 s in ResEpi and NoEpi groups, respectively (significantly not different between ResEpi and NoEpi, t test, p = 0.42 for IIS length, p = 0.43 for seizure length). Because the length of the different activity types was calculated with different methods, we do not compare them. However, note the considerable differences (Fig. 1).
CSD analysis showed that sink-source pairs were restricted to the layers where the LFPg transients were observed (Fig. 1). The most frequent CSD pattern associated with SPA was a sink-source pair, or a source-sink-source triplet in the supragranular layer, because most of the SPAs were detected in the supragranular layer (see also Köhling et al. 1999). In the case of IISs/seizures, sinks and sources were found in all layers of the neocortex because these activities usually invaded the entire width of the neocortex. As in an earlier study (Köhling et al. 1999), the number and exact location of CSD sinks and sources was variable during both SPAs and epileptiform activities. Sinks and sources usually appeared simultaneously during SPAs, whereas during epileptiform activity the initial sink or source was followed by other sinks and sources. Sink or source or sink-source pair were observed as initial CSD deflection during IISs and seizures. The initial CSD transient could appear in any layer of the neocortex, and we could not find any link between the CSD pattern and the simple or complex nature of the IISs.
The power of the LFPg signal was increased in both the ripple (130-250 Hz) and the fast ripple (300-700 Hz) bands during SPAs, as in our previous study (Tóth et al. 2018). During BIC-induced IISs, the power of the signal was elevated in all examined frequency ranges, including ripple and fast ripple bands ( Fig. 1 and Table 5.). In the case of seizures, we could not detect peaks in the ripple and fast ripple bands because the LFPg power was homogeneously increased in all frequencies. SPAs showed an increased power of high-frequency oscillations (HFOs, both ripple and fast ripple bands) in 17 of 19 cases in the ResEpi group and in three of six cases in the TreatEpi group, whereas ripple and fast ripple power was enhanced in 17 of 22 and 14 of 22 cases, respectively, in recordings from NoEpi cases. Ripple and fast ripple power increase during SPAs in the ResEpi group was 3.67 ± 1.82 dB and 3.27 ± 1.57 dB, respectively, and was similar in the NoEpi group: 3.09 ± 1.59 dB and 3.07 ± 1.60 dB in ripple and fast ripple bands, respectively. HFO power increased during all IIS activities in recordings from tissue derived from epileptic patients (ResEpi n = 15 activities, TreatEpi n = 4 and NoMed n = 1), whereas it was increased in 11 of 13 cases in recordings from the NoEpi group. The power increase of HFOs linked to IISs was significantly higher than during SPA (12.53 ± 5.00 dB and 11.75 ± 4.67 dB for ripples and fast ripples, respectively, in the ResEpi group, and, in NoEpi: 16.05 ± 3.32 dB and 15.77 ± 3.30 dB for ripples and fast ripples, respectively) (both ripple and fast ripple power were significantly different between SPAs and IISs, Mann-Whitney U test, P < 0.0001, and NoEpi fast ripple power was significantly different from ResEpi and TreatEpi, one-way ANOVA and Tukey's honest significant difference test, P < 0.05). The peak frequency of both ripples and fast ripples linked to SPAs or to IISs was lower in the ResEpi than in the NoEpi group, although the differences were not significant (Table 5). In summary, remarkable differences were observed between the properties of SPAs, IISs and seizures. The recurrence frequency was lower, whereas the LFPg and MUA amplitudes and the high frequency power were higher in the case of IISs and seizures compared to SPAs. When comparing patient groups, only several features of the synchronies were different. A tendency to generate a higher percentage of recurring epileptiform activity was observed in slices from ResEpi compared to NoEpi patients. SPA length and IIS fast ripple power were higher, whereas IIS and seizure MUA were lower in ResEpi tissue than in NoEpi tissue.

Initiation and propagation of BIC-induced activities across cortical layers
Most of the IISs (Fig. 2) and all seizures (Fig. 3) invaded the entire width of the neocortex. In the case of spatially complex events, supragranular and infragranular layers were separately activated (Figs 2E and 3A-B). Supra-and infragranular layers presumably have different roles in interictal spike generation: in vivo spontaneous interictal spikes propagated from other brain areas usually showed an initial activation in the granular or the supragranular areas, whereas de novo generated spikes were initiated in the infragranular layers (Ulbert et al. 2004a). Although propagation from distant sites is excluded in slice preparations, the question arises as to whether certain layers have a greater probability of inducing synchronous events, whereas others follow, or whether BIC-induced events appear synchronously throughout the entire width of the neocortex. We aimed to explore this question and therefore determined the initiation site of the IISs and found several different interlaminar spreading patterns. The initiation site could be in any of the neocortical layers ( Fig. 4A): in supragranular layers (n = 12 IISs from ResEpi, n = 9 from NoEpi), in granular layers (n = 4 from ResEpi, n = 2 from NoEpi) or in infragranular layers (n = 6 from ResEpi, n = 2 from TreatEpi). In two recordings (one from ResEpi, the other from NoEpi), we detected two different types of IISs: one was initiated in the supragranular layers and spread to the infragranular layers, and one was generated in the granular layers and spread to the two other layers (Fig. 4E). In all other cases, the initiation layer and spreading directions were stable within one recording. Spatially complex IISs were generated in the granular (n = 1) or in infragranular layers (n = 1), however, this was not constant during the recording (Fig. 2E).
We performed a more detailed analysis on how IISs spread across the layers of the neocortex in 12 recordings with recurrent IISs (9 from ResEpi, 1 from TreatEpi, 2 from NoEpi) and in three with recurrent seizures (all from ResEpi). In this analysis, we included recordings with at least six (but up to 166) IIS events, to determine a precise propagation speed and exclude any large bias coming from the eventuality of single epileptiform events. Because seizures occurred with a considerably lower frequency, we analysed recordings with three to five seizures. Note that our analysis reveals the features of the propagation of epileptic events in the depth (across the layers) of the neocortex and not the horizontal spread over larger areas. We determined the initiation time point of every event on each channel, and also examined its spreading direction and speed (Fig. 4B-E). In most cases (10/12 recordings), we observed a jitter among the channels over time (i.e. the starting point of the IIS was on different channels in the course of the consecutive events). Furthermore, the activation sequence of the channels was similarly variable as the starting point, although the spreading direction remained constant.
The propagation speed of IISs was also examined within the neocortical column. We determined a propagation speed for each event and examined how it changes with time. Usually, it was constant throughout the recording (Fig. 4B); however, we found intracortical propagation with increasing (n = 3) (Fig. 4C) and decreasing (n = 1) (Fig. 4D) speed as well. The increase/decrease in speed was the result of either a faster/slower activation of neighbouring channels (n = 1 increased, n = 1 decreased) or to a decrease in the jitter observed at channel activation (n = 2 increased) (Fig. 4B). The average propagation speed of IISs varyied between 19.7 and 98.7 mmc, with a mean of 51.8 ± 23.7 mm s −1 in ResEpi slices and 74.3 ± 39.0 mm s −1 in NoEpi slices. We observed IISs with low (<30 mm s −1 ) and high (>75 mm s −1 ) propagation speed in both ResEpi and NoEpi slices. In both cases with the two different IIS activation patterns (see above), one IIS had considerably higher propagation speed than the other. When the propagation speed increased over time, it could reach up to a threefold increase (from ß40 to ß120 mm s −1 ). Seizures in ResEpi tissue propagated with a speed of 41.2 ± 12.6 mm s −1 . The layer of initiation and the propagation speed were not related: high and low speed could be associated with IISs initiated in any layers. We could not find a correlation between the propagation speed of the IISs and the aetiology of the patients. High, medium and low speed IISs were observed in ResEpi slices derived from patients with dysplasia, tumour or hippocampal sclerosis. High speed IISs were detected in NoEpi slices derived from patients with carcinoma metastasis or with glial tumour.
Our observation is that BIC-induced population events can be generated in any cortical layer. Although the starting channel can change over time, the layer of initiation remains stable. The direction and the speed of the intracortical propagation were usually constant, with some exceptions when propagation speed increased or decreased throughout the recording.

Firing characteristics of single cells
In this analysis, we included all files with population activity (SPA/IIS/seizure) and single cell activity from the ResEpi and NoEpi groups. Recordings lacking either synchronous event or single cells were excluded, and only ResEpi and NoEpi recordings were compared to focus on epilepsy-related characteristics. We clustered 384 and 349 single cells in recordings from ResEpi and NoEpi groups, respectively. In ResEpi, 193 single units (from

Figure 4. Initiation and propagation of BIC-induced IISs within the neocortex
A, IISs could be initiated in every layers of the human neocortex: supragranular (left), granular (middle) and infragranular (right) layers. B-E, the initiation region and the spreading direction of the recurring IISs were determined. We detected the starting point of every IIS event on every channel (red lines) and calculated the propagation speed of the events within the neocortical column (right, bottom). Left: original recording of an event, also showing the absolute values with the starting points (red lines). Bottom: consecutive IIS events are visualized with the detected raster plots, two of which (marked with blue star and triangle) are magnified in the middle. The colour intensity of the grey raster lines is in correlation with the LFPg amplitude of the event on that channel (colour scale in E). The heat map on the right (top) illustrates the spreading speed of the IIS. Each column is an IIS event; each row corresponds to a channel. Red colours show the initiating channels, blue colours show the late activating channels. The intensity of the colours is related to the LFPg amplitude of the event on the given channel (colour scale is in E). The propagation speed (right, bottom) was stable in most cases (B) but an increasing (C) or decreasing (D) speed was also observed. Two types of IIS with different propagation patterns (E, blue star and blue triangle) were simultaneously occurring in two recordings (one from ResEpi, one from NoEpi). The change in the propagation speed was often related to the modification in the jitter of channel activation (compare the event marked with the blue star with the event marked with the blue triangle in C). [Colour figure can be viewed at wileyonlinelibrary.com] 16 slices derived from 13 patients, with a mean ± SD of 10.9 ± 6.7 neurons per slice and 13.4 ± 7.9 neurons per patient) were clustered in control conditions and 191 (from 18 slices derived from 15 patients, with 11.1 ± 5.4 neurons per slice and 14.0 ± 11.2 neurons per patient) during application of BIC, whereas, in NoEpi, 182 cells (from 14 slices derived from 11 patients with 13.3 ± 12.7 neurons per slice and 16.9 ± 20.8 neurons per patient) in control and 167 (from 14 slices derived from 10 patients, with 11.6 ± 7.6 neurons per slice and 16.3 ± 14.7 neurons per patient) in a solution containing BIC (Table 6).
In most cases, we clustered cells from one slice per patient but, in samples derived from five patients (2 ResEpi and 3 NoEpi patients), we could analyse two or three slices per patients. To increase the number of cell/IIS comparison, we included two further recordings from the TreatEpi group with high numbers of neurons (n = 39 single cells in 2 slices) and IIS events. When comparing cellular discharge in physiological conditions and in BIC bath, we noted several patterns of firing change.
In physiological solution, spontaneously discharging cells were observed mainly in the supragranular and the infragranular layers. During BIC application, the majority of these cells became silent, whereas other, previously undetected, spontaneously discharging neurons appeared on other recording channels (Fig. 5). We noted that cells located in the granular layer often showed a very characteristic bursting-like firing pattern ( Fig. 5A and  C), which was very rarely seen in control conditions. In several cases in BIC bath, neurons discharged only during the BIC-induced events and were silent between the events (n = 1/12 with seizures in ResEpi, and n = 7/33 with IISs, 6 from ResEpi, 1 from NoEpi) (Fig. 5B), even if spontaneous neuronal activity was observed in physiological solution before the application of the BIC. We separated excitatory PCs and INs based on their AP shape and firing patterns (Fig. 5D-E). For the analysis of the cellular properties in relation to epilepsy, cells in the TreatEpi group were excluded, and only ResEpi and NoEpi were compared. The ratio of detectable (i.e. spontaneously firing) PCs was higher than that of INs during physiological conditions in tissue from both epileptic and non-epileptic patients  Table 6). The ratio of intrinsically bursting PCs was low in both patient groups: in ResEpi 1.6% in control conditions and 2.1% in BIC, in NoEpi 2.7% in physiological conditiond and 1.8% in BIC bath.
The firing frequency of neurons was somewhat higher in ResEpi tissue than in NoEpi tissue (Table 6). Furthermore, in physiological bath, INs showed a slightly higher average firing frequency than PCs, in both ResEpi slices and NoEpi slices. Differences in the firing frequency were not significantly different. In BIC bath, the overall firing frequency of the cells did not change, although the vast majority of the cells fired with a very high frequency during the BIC-induced events. This latter phenomenon was reflected in the significantly increased burstiness together with the decrease of the interevent interval of all neuron types during the application of BIC, compared to control conditions (Table 6).
In summary, other neurons were spontaneously active during physiological conditions than in BIC bath. Higher numbers of PCs than INs were spontaneously firing in physiological bath, whereas this ratio was reversed during the application of BIC. The average firing frequency remained unchanged in BIC bath, with a modified firing pattern: the neurons intensely discharged during the epileptiform events and tended to be more silent between the events.

Neuronal firing related to synchronous activity
We aimed to examine the mechanisms of synchrony initiation in the human neocortex; therefore, we investigated the activity of single neurons during SPAs, IISs and seizures with several approaches.
First, we dtermined whether a combined discharge of single cells forms a build-up period before SPAs and BIC-induced events, as in the rodent hippocampus before interictal-like events (Menendez de la Prida et al. 2002;Cohen et al. 2006;Wittner & Miles, 2007). Neuronal firing preceding the SPA events was only occasionally seen. Build-up discharge was observed in seven of 21 SPAs in ResEpi tissue and in six of 22 SPAs in NoEpi tissue. The existence of the neuronal discharge before the events was rare within one recording; we noted cellular firing before in 1.6 ± 4.9% of the events in ResEpi tissue and only in 0.1 ± 0.1% of the events in NoEpi tissue.
In recordings with IISs, we observed neuronal firing before the events more often: in seven of 17 IISs in the ResEpi and in seven of 14 IISs in the NoEpi group (Fig. 4A). We detected cell firing before the events in 21.9 ± 40.6% of the events in the ResEpi, and in 30.3 ± 41.2% in the NoEpi group. Increased neuronal firing was observed up to several minutes before seizures in epileptic patients in both the neocortex (Truccolo et al. 2011) and the hippocampus (Lambrecq et al. 2017). By contrast, during BIC-induced seizures in vitro, cells were always silent before the seizures (n = 7 in ResEpi and n = 3 in NoEpi), they started to fire abruptly with the first large LFPg component, and they usually showed an intense firing at the late, decreasing component (Fig. 3C).

Single cell firing during SPA
We examined how single cells behaved during SPAs emerging in physiological solution. To obtain insight into the synchronization mechanisms, we analysed how Figure 5. The effect of BIC on APs A-C, single cell firing during the application of physiological solution (left) and BIC bath (middle and right). Black rectangles indicate SPAs on the left (note the difference in amplitude compared to IISs on the right; the same scales apply in all cases). Spontaneously occurring neuronal discharges were observed mainly in the supragranular and infragranular layer in physiological solution. By contrast, in the presence of BIC, the majority of these cells became silent (e.g. cells on the dark and light blue, as well as the orange and red channels in A; all cells in B; and cells on the light blue and orange channels in C). In addition, other neurons started to spontaneously discharge (see green, orange and purple channels in A, dark blue and green channels in C). Several (mainly granular) cells showed a characteristic bursting-like behaviour (green channel in A, dark blue channel in C). In several slices, most of the cells stayed silent when applying BIC, and showed excessive discharge only during the IISs (B). D-E, interneurons, bursting principal cells and regularly spiking principal cells were identified by the shape of their APs and the autocorrelogram of their firing. These characteristics remained unchanged in the presence of BIC (E).
[Colour figure can be viewed at wileyonlinelibrary.com] neurons change their firing during SPAs by computing the firing change during the events compared to baseline periods (Fig. 6A). The normalized firing change of all cells was 0.55 ± 0.30 in ResEpi tissue (n = 136 cells in 15 files with SPA) and 0.64 ± 0.31 in NoEpi tissue (n = 124 cells in 14 files with SPA) (Fig. 6B). Single cells were considered to have an increased firing if their normalized firing increase was equal or more than 0.6 (an increase to at least 150%). Slightly more cells increased their firing during SPA in NoEpi slices (56%) than in ResEpi slices (39%) ( Table 7). We did not find significant differences between the cell types, similar percentages of PCs, INs and UCs showed increased discharge during both ResEpi and NoEpi SPA, with comparable normalized firing change. Because the activity of burst firing PCs was related to epilepsy and to epileptiform discharges (Connors, 1984), we separately investigated the activity of this special neuron group. Out of the three bursting PCs detected in ResEpi tissue, the firing of two was increased during SPA, whereas one remained unchanged. In the NoEpi tissue 2/5 bursting PCs increased their firing during SPA (Fig. 6A and Table 7).
Next, we examined how excitatory or inhibitory cells participate in the generation of SPAs. Therefore, we analysed the combined firing of all PCs, all INs and all UCs within each recording with SPA in both ResEpi tissue and NoEpi tissue. In line with the above finding, such that around one-half of the cells showed increased firing during SPA, in several cases, no clear population firing enhancement was visible during SPA (n = 6/15 SPA in ResEpi tissue and 5/14 SPA in NoEpi tissue) (Fig. 6C  and D). Furthermore, we could not determine a constant sequence of cell type activation because both excitatory or NS, not significant.
inhibitory neuronal population could start firing, whereas the other followed ( Fig. 6C and D) 7E and Table 8).
In summary, around one-half of the cells increased their firing rate during SPA, in both ResEpi tissue and NoEpi tissue. No significant differences were found between patient groups or cell types. Either PCs or INs could start discharging at the beginning of the SPAs, whereas the other cell types followed. PCs contribute to the overall firing slightly more than INs.

Single cell firing during BIC-induced activity
The results focussing on the role of excitatory and inhibitory cell types in the initiation of synchronies related to epilepsy provided controversial data. When GABA A receptors were blocked by BIC in slice preparations, bursting pyramidal cells were shown to induce interictal-like events in both the hippocampus (Cohen et al. 2006;Wittner & Miles, 2007) and the neocortex (Connors, 1984) of rodents. On the other hand, interneurons discharged before pyramidal cells during spontaneously occurring interictal-like activity in the human subiculum in vitro (Pallud et al. 2014). Furthermore, a heterogeneous cell population showed a complex interplay during interictal spikes in the human neocortex of epileptic patients (Keller et al. 2010). The beginning of seizures was also associated with glutamatergic (Huberfeld et al. 2011) or GABAergic (Elahian et al. 2018) processes, depending on the seizure morphology. To clarify the role of excitatory and inhibitory neuronal populations in the generation of epileptiform activity, we determined the normalized firing increase of neurons, as well as the contribution of PCs and INS during IISs and seizures and compared these data with those related to SPAs.
The normalized firing change during IISs was significantly higher than during SPA (Kruskal-Wallis ANOVA, P < 0.001), and even higher during seizures (significantly different, P < 0.001), in both ResEpi slices and NoEpi slices (ResEpi IIS: 0.86 ± 0.31, seizure: 0.97 ± 0.12; NoEpi IIS: 0.95 ± 0.17, seizures: 1.0 ± 0.0) ( Table 7). During ResEpi IIS and seizures, 86% and 99% of the cells increased their firing rate, respectively (SPA-IIS-seizure significantly different, chi-squared, P < 0.001), whereas, during NoEpi IIS and seizures, 95% and 100% of the cells increased their firing rate, respectively (SPA-IIS-seizure significantly different, Fisher's exact test, P < 0.001). Similar to SPAs, no differences were seen between cell types: similar proportions of PCs, INs and UCs showed increased firing with similar values of normalized firing increase (Table 7). Concerning intrinsically bursting PCs, three of four and three of three, respectively, were found to increase their firing rate in ResEpi and NoEpi IIS. No bursting PCs were clustered in the recordings with seizures.
Next, we performed PETHs by computing the discharge of all PC, all IN and all UC cells separately, in each recording with IISs and seizures, as well as examined the timing of intrinsically bursting PCs, separately. To increase the number of IIS examined, beside the 14 recordings in the ResEpi group and the 13 in NoEpi group, we included two more recordings with high numbers of IIS events from the TreatEpi group. By contrast to our expectations, which assumed (bursting) PCs initiate IISs (Connors, 1984;Wittner & Miles, 2007), we found that, in most cases, INs fired earlier than PCs in our human neocortical disinhibition model of epilepsy. IN firing preceded the discharge of other cell types (PC and UC) during IISs in seven of 14 recordings in the ResEpi group and in five of 13 recordings in the NoEpi group (Fig. 7). PCs began the discharge pattern in three and two cases in ResEpi and NoEpi slices, respectively. In all the remaining recordings (n = 4 in ResEpi, n = 2/2 in TreatEpi and n = 6 in NoEpi), UC cells started to fire during the IISs, whereas other cells followed (ResEpi and NoEpi significantly not different, Fisher's exact test, P > 0.6). Four intrinsically bursting PCs were discharging during IISs in ResEpi, three in TreatEpi and three in NoEpi slices. All ResEpi, all TreatEpi and one NoEpi bursting PCs fired later than the INs detected in the same recording ( Fig. 7B and C) and the remaining two bursting PCs in the NoEpi tissue discharged later than all UCs (n = 7) of the same recording (Fig. 7D).
Similar to IIS, INs tended to discharge before PCs during the initial phase of seizures. In the ResEpi group, INs started to fire together with UCs in three of six cases, whereas PCs followed. PCs and UCs initiated seizures in two of six and one of six cases, respectively. In the NoEpi group, in all cases (n = 3) INs started to discharge (once together with UCs), whereas PC followed.
We analysed how a neuronal population of a given slice generates SPA and epileptiform activity. Because we observed SPA in the majority of the slices before inducing IIS/seizure, we determined whether the same neuron type initiated SPA and IIS/seizure in the same slice. In samples from ResEpi patients, we observed similar initiator neuron type in two of 14 slices, and in two of 13 in NoEpi slices.

Figure 7. Single-cell behaviour during IIS
Different types of initiation patterns were observed regarding the firing increase during IISs. PETHs were calculated for all PCs (red), INs (blue) and UCs (grey) related to IIS events (right). By contrast to our expectations, PC firing increase at the start (A) was seen only in a minority of cases (21% in ResEpi, 15% in NoEpi). Our observation was that INs preceded the discharge of other cell types (B) in most cases (50% in ResEpi,38% in NoEpi). In some cases, In the remaining slices, different cell types initiated SPAs than IISs/seizures.
To determine the contribution of excitatory and inhibitory discharge in epileptiform synchronies, we calculated the percentage of APs given by PCs and INs during the initial firing of IISs and seizures. As in the ratio of clustered PCs and INs, we found a switch in excitatory and inhibitory contribution in the firing compared to SPAs; only 16.5 ± 29.0% and 12.2 ± 15.4% of the APs came from PCs in ResEpi and NoEpi IIS,whereas 43.4 ± 34.3% and 46.4 ± 29.1% of the APs derived from INs, respectively (Table 8). Burst firing PCs contributed to the total firing in 1.7 ± 3.7% in ResEpi and in 1.1 ± 3.6% in NoEpi IIS. In the case of seizures, the values for PCs and INs were not different from IISs (ResEpi seizures: 18.8 ± 30.9% from PCs, 32.2 ± 17.7% from INs, NoEpi seizures: 3.6 ± 4.3% from PCs, 59.9 ± 26.8% from Ins).
Next, we analysed the correlation between the propagation speed of IISs (see above) and the cellular contribution. Unclassified cells discharged at the initial phase of the two NoEpi and one TreatEpi slice examined for propagation speed. In the ResEpi slices, we could not find any relation between highly contributing cell type and propagation speed. High speed IISs were initiated by either INs or all cell types together, medium speed IISs were initiated by PCs, or IN + UCs or PC + UCs. Low speed IISs were initiated by PCs, INs or by IN + UCs.
We found that almost all neurons, independently from their type, increased their firing rate during IISs and seizures. In most cases, INs started to discharge at the onset of the epileptic synchronies, whereas other neurons followed, but PCs and UCs were also found to fire at the initial phase of the hypersynchronous events. During both IISs and seizures INs provided higher ratio of APs to the overall firing than PCs, which is in contrast to SPAs where PCs contributed more than INs.

Synchrony-generating hubs in the human neocortex
When applying the GABA A receptor antagonist BIC, tissue slices in the ResEpi group tended to generate seizures and IISs more often, and with somewhat higher recurrence frequency, than slices derived from the NoEpi group. Furthermore, we detected several combinations of temporally and spatially simple and complex IISs in ResEpi tissue, whereas only simple-simple IISs were recorded in NoEpi samples, spreading to all layers of the neocortex. In slices from the ResEpi group, spatially more restricted IISs were also observed (supragranular or infragranular IISs) and supragranular and infragranular neuron populations were separately activated during temporally and spatially complex IISs (Fig. 2E). This suggests that epileptic neural circuits (possibly as a consequence of the epileptic reorganization) have a greater probability of initiating hypersynchronous events than non-epileptic networks. In the human epileptic neocortex, smaller neuronal populations of the cortical network were also able to generate IISs, compared to NoEpi tissue. These delimited neural circuits, similar to the hippocampal 'hub' cells or cell populations (Feldt Muldoon et al. 2013;Bui et al. 2015), or the 'pathologically interconnected neuron clusters' (Bragin et al. 2000) of epileptic rodents, might act as initiators for the generation of hypersynchronous events.

Effects of BIC on neuronal firing
Considerably higher numbers of INs were detected in a perfusion solution containing BIC, compared to physiological conditions, where the majority of the spontaneously firing cells were excitatory principal cells. This might be explained by the difference in the synaptic input of the different cell types. In the rodent hippocampus, the proportion of inhibitory synapses arriving at CA1 pyramidal cells was 5.3% (Megias et al. 2001). All interneuron types examined so far received higher ratio of inhibitory inputs than pyramidal cells: 6.4% of the boutons forming synapses with parvalbumin-containing interneurons were GABAergic (which cell type received a considerably high number of glutamatergic boutons), 29.4% of the synapses terminating on calbindin-positive interneurons were inhibitory, 20.7% were GABAergic on calretinin-positive (Gulyás et al. 1999) and 35.85% on cholecystokinin-containing interneurons (Mátyás et al. 2004). Because GABA A receptors are blocked by BIC (Simmonds, 1980), the transmission of fast GABAergic hyperpolarizing potentials from inhibitory cells to their postsynaptic targets is largely reduced and, consequently, cells are released from their usual inhibition. Because the UC cells (C) or UC cells together with INs (D) started to fire and the other cells followed (29% in ResEpi, 46% in NoEpi). The firing of intrinsically bursting PCs (purple in B, C and D) always followed the activation of other cells detected in the same recording. Below the single sweeps, raster plots show the firing of the cells clustered in the given recording. Note the variability in single cell firing during consecutive events. In the case of spatially complex IISs (A and D), cell discharge was associated with the activation pattern of the neocortical layers. E, contribution of PC (red) and IN (blue) discharge to the total cell firing during SPAs, IISs and seizures in ResEpi (left) and NoEpi (right) samples. PCs had a higher impact on the overall firing during SPAs than INs, whereas IN firing was considerably more pronounced than PC firing during both IISs and seizures. [Colour figure can be viewed at wileyonlinelibrary.com]

Contribution of PC firing (%)
proportion of the GABAergic input terminating on INs is higher, than on PCs, this might result in the more pronounced increase in excitability of these cells, and thus in the appearance of high numbers of spontaneously active INs. Although the above data are derived from the rodent hippocampus, comparable synaptic input proportions might characterize the human neocortical neurons as well.
The possibility that BIC distorts the shape of APs cannot be excluded and might also account for the high proportions of INs detected in the disinhibited human neocortex. BIC was shown to reduce the afterhyperpolarization phase of the APs but not the duration of the depolarization and repolarization phases (Seutin et al. 1997). Because we used the length of the AP at the half of the maximal amplitude to distinguish between excitatory and inhibitory cell types (and not the peak-to-through duration, which would be distorted by bicuculline), this possibility is improbable.

Synchrony-generating mechanisms in the human neocortex
In the present study, we investigated the excitatory and inhibitory processes during presumably physiological synchrony such as SPA (Tóth et al. 2018), as well as during interictal-like spikes and seizures. Considerable differences were observed between SPA and BIC-induced epileptiform activities. SPAs had a clearly lower level of synchrony than pathological events: only one-half of the cells increased their firing, and cells had a normalized firing increase of ß0.50-0.60. As expected, IISs were more synchronized events, with ß90% of cells showing enhanced firing and a normalized firing increase of ß0.85-0.95. Seizures displayed the highest degree of synchrony, involving practically all cells, which had a normalized firing increase of 0.97-1.00. The role of the different excitatory and inhibitory neuron types in the generation of hypersynchronous events has been long investigated but has remained a matter of debate. Both excitatory bursting pyramidal cells (Connors, 1984) and INs (Pallud et al. 2014) were found to initiate interictal-like discharges in neocortical slices, whereas neurons responded with diverse firing patterns to interictal spikes in the neocortex of epileptic patients (Keller et al. 2010). Similarly, seizure onset was linked to both excitatory (Huberfeld et al. 2011) and inhibitory (Elahian et al. 2018) processes, depending on the morphology of the seizure. We found that the role of excitatory and inhibitory neurons was different in physiological vs. pathological synchronies. PCs had a higher contribution to the overall firing than INs during SPA (PC: ß32% vs. IN: 15-24%), which reversed during epileptiform activities (both IIS and seizure): INs contributed with higher percentages to the overall firing than PCs (IIS PC: ß15% vs. IN: ß45%; seizure PC: ß3-18% vs. IN: ß32-60%).
It should be noted that the relatively low occurrence rate of the seizure-like activities and the heterogeneous origin of the tissue samples do not allow us to draw far-reaching concusions about the generation of seizures. A methodological difficulty further complicates the investigation of seizure initiation mechanisms. High electrical activity of the brain, such as an epileptic seizure, was shown to modify the properties of extracellular APs used for cell clustering, and therefore deteriorated the identification of single cells . Merricks et al. (2015) recorded using a multielectrode array with large (ß400 µm) interelectrode distance, which corresponds to multiple but single electrode recording. A higher spatial electrode density, such as a tetrode configuration, largely improves the identification of single cell APs (Gray et al. 1995). We used a multiple channel electrode with a considerably smaller contact distance (150 µm), allowing a better neuron separation than the above study, although it should be noted that our clustered cells discharging during seizures are possibly less precise than neurons detected during other conditions. However, the impact of seizure activity on cell clustering is probably similar in the case of PCs and INs, and thus our results concerning the contribution of the different cell types in the initiation of seizures support our assumption.
Based on our findings, we propose that SPAs emerge on the basis of a complex and balanced interaction of excitatory and inhibitory networks. This refers to in vivo physiological conditions in human patients, where a dynamic balance of excitation and inhibition characterized the neocortex (Dehghani et al. 2016). The higher impact of PC firing suggests that excitatory networks might have a leading role (which is not necessarily required for the initiation itself) in this type of synchrony.
Epileptiform IISs and seizures appeared spontaneously, when the application of BIC artificially impaired the balance of excitation and inhibition. These hypersynchronous events involved the vast majority of neurons, and the elevated contribution of IN discharge indicates that they were mainly initiated by inhibitory cells.

Non-synaptic mechanisms to synchronize cellular activity in the disinhibited neocortex
We found an intense inhibitory cell discharge at the initial phase of BIC-induced epileptiform events (both IISs and seizures). This raises the question of how the neocortical circuit can be synchronized if the downstream step of the interneuronal firing (i.e. the transmission through postsynaptic GABA A receptors) is blocked. Our results showing that inhibitory cell firing can lead these synchronization processes implies that non-synaptic cellular interactions are also involved in the generation of disinhibition-induced hypersynchronous events in the human neocortex. Non-synaptic interactions are transmitted through gap junctions, ephaptic effects, changes in extracellular ions (Jefferys, 1995) and neuropeptides (van den Pol, 2012). Indeed, BIC-induced epileptic activity is regulated by a neural network connected through gap junctions in the guinea pig brain preparation (de Curtis et al. 1998). Gap junctions specifically connect parvalbumin-positive interneurons in the rat neocortex (Galarreta & Hestrin, 1999), with such a cell type being considered to 'control rhythm' in physiological conditions (Freund & Katona, 2007). Neurons are not in close apposition in the human neocortex, and so an ephaptic effect is possibly less important here than in the hippocampus, where it can clearly participate in synchronization processes (Dudek et al. 1986). An increase in extracellular K + concentration has been shown to enhance the excitability, as well as the spontaneous discharge rate, of hippocampal neurons (Cohen & Miles, 2000), and thus a high level of extracellular K + is commonly used to induce epileptiform activity in vitro. However, the increase in K + concentration during disinhibition-induced epileptic activity is rather the consequence of the paroxysmal discharges because it starts to elevate after the onset of the epileptic bursts (Heinemann et al. 1977). Neuropeptides present in inhibitory cells might also modify the activity of other neurons in the presence of BIC. However, neuropeptides, such as neuropeptide Y, somatostatin, cholecystokinin, etc., mainly act by reducing neurotransmitter release from the presynaptic terminals (van den Pol, 2012). Especially, neuropeptide Y was found to decrease both glutamate and GABA release in the rat neocortex, which explained its anti-epileptic effect (Bacci et al. 2002). In summary, an electrical connection through gap junctions is the most probable non-synaptic mechanism to synchronize interneuron populations in the human neocortex when GABA A receptors are blocked, whereas the other mechanisms appear to be improbable. This theory remains to be tested in the future.

Human disease vs. models
BIC-induced epileptic activities were hypersynchronous events, with a synchronous and intense firing of INs during the initial phase. This was an unexpected finding in the case of IISs because, in the same BIC model in rodents, synchrony was shown to develop as a result of the loss of inhibitory control, and neurons were recruited into spikes via recurrent excitatory pathways (Miles & Wong, 1987). Excitatory bursting PCs were demonstrated to initiate BIC-induced IISs in both neocortical (Connors, 1984) and hippocampal (Cohen et al. 2006;Wittner & Miles, 2007) slices of rodents. This clearly differs in the disinhibited human neocortex because all intrinsically bursting PCs were found to discharge later than other cells of the same recording. Data regarding the generation mechanisms of interictal spikes in epileptic patients are very sparse. Around one-half of the extracellularly detected cells participated in the spikes in both the neocortex (Ishijima et al. 1975;Keller et al. 2010) and the hippocampus (Alvarado-Rojas et al. 2013), with a heterogeneous firing pattern (Keller et al. 2010). PCs and INs were not distinguished in these studies; thus, information about excitatory and inhibitory signalling in the generation of interictal spikes in epileptic patients is lacking.
Seizures occurring in human neocortical slices in our BIC model consisted of continuously recurring interictal-like spikes, lasting for 15-28 s. This morphology resembles in vivo seizures composed of periodic spikes (Perucca et al. 2014) or spike and wave complexes (Truccolo et al. 2014). Seizures with hypersynchronous spike onset in animal models (Salami et al. 2015;Kohling et al. 2016), as well as preictal discharges in human slices (Huberfeld et al. 2011), were related to excitatory processes. By contrast, the dominance of inhibitory cell firing characterized the onset of BIC-induced seizures (such as that of IISs) in our experiments, for which the mechanism was seen at low voltage fast activity seizures in epileptic patients (Dehghani et al. 2016;Elahian et al. 2018) and in rodents (Gnatkovsky et al. 2008).
The above differences and inconsistencies found in our experiments, as well as previously, might be explained by species differences and/or by the different conditions and methods used, as well as the different cortical regions investigated. Bursting pyramidal cells in the guinea pig (Connors, 1984) vs. interneuronal firing in human neocortical slices (present study) initiating BIC-induced interictal spikes points to species differences. Pharmacological manipulations and in vitro circumstances clearly differ from in vivo conditions (BIC-induced vs. spontaneously occurring epileptic activity in patients). The largely reduced GABAergic neurotransmission in vitro results in synchronization mechanisms different from those of the intact brain, where the dynamic balance of excitation and inhibition breaks and transforms into epileptic seizure (Dehghani et al. 2016). Furthermore, differences in the examined cortical region might also account for the dissimilarities observed in the results. The generation mechanism of hypersynchronous spikes (Huberfeld et al. 2011) and seizures (Salami et al. 2015) was related to excitatory processes in the three-layered hippocampal formation, whereas epileptiform activity with similar morphology (i.e. manifested as single or multiple spikes) was associated with an enhanced interneuronal firing in the six-layered neocortex (present study). Our results suggest that the disinhibition model of epileptic activity shows considerable differences compared to the human disease. Furthermore, as shown in the present study, the same disinhibition model operates with different mechanisms in human tissue compared to rodents.

Conclusions
Our data indicate that physiological synchronous population events emerged on the basis of a complex interaction between excitatory and inhibitory networks in the human neocortex. The disinhibition of the neural circuit, through the impairment of the excitation-inhibition balance, resulted in the spontaneous generation of hyper-synchronous epileptiform activities, such as IISs and seizures. IISs were uniform in NoEpi samples: they always invaded the entire neocortex and were simple events. In tissues derived from epileptic patients, IISs were heterogeneous, either spreading to the entire width of the neocortex or restricted to a subset of layers, and several variations of temporally and spatially simple and complex events were also detected. This suggests that smaller neuron populations are able to generate epileptiform activity in the neocortex of epileptic patients than in peritumoural tissue, which might act as synchrony initiator circuits.
In the generation of physiological events, excitatory neuron firing might have a leading role, whereas epileptiform activity was mainly initiated by inhibitory cell discharge. This latter finding suggests that a non-synaptic mechanisms might also play an important role in synchrony generation, most presumably comprising the electrical connection between certain interneuron populations through their gap junctions. Our results regarding hypersynchronous events initiated by inhibitory cells differ from the results obtained in animal models or in epileptic patients, implying that different neural mechanisms might be activated in the disinhibition model than in the human disease. Therefore, conclusions should be drawn carefully when using pharmacologically induced, in vitro models of epilepsy, by considering the limits of extrapolation from model to disease and from animals to humans.