Interictal epileptiform discharges in focal epilepsy are preceded by increase in low-frequency oscillations

OBJECTIVE
Interictal epileptiform discharges (IEDs) constitute a diagnostic signature of epilepsy. These events reflect epileptogenic hypersynchronization. Previous studies indicated that IEDs arise from slow neuronal activation accompanied by metabolic and hemodynamic changes. These might induce cortical inhibition followed hypersynchronization at IED onset. As cortical inhibition is mediated by low-frequency oscillations, we aimed to analyze the role of low-frequency oscillations prior the IED using magnetencephalography (MEG).


METHODS
Low-frequency (1-8 Hz) oscillations pre-IED ([-1000 milliseconds (ms), IED onset]) were analyzed using MEG in 14 focal epilepsy patients (median age = 23 years, range = 7-46 age). Occurrence of local pre-IED oscillations was analyzed using Beamformer Dynamical Imaging of Coherent Sources (DICS) and event-related desynchronization/synchronization (ERD-ERS) maps constructed using cluster-based permutation tests. The development of pre-IED oscillations was characterized using Hilbert transformation.


RESULTS
All patients exhibited statistically significant increase in delta (1-4 Hz) and/or theta (4-8 Hz) oscillations pre-IED compared to baseline [-2000 ms, -1000 ms]. Furthermore, all patients exhibited low-frequency power increase up to IED onset.


CONCLUSIONS
We demonstrated consistently occurring, low-frequency oscillations prior to IED onset.


SIGNIFICANCE
As low-frequency activity mediates cortical inhibition, our study demonstrates that a focal inhibition precedes hypersynchronization at IED onset.


Introduction
Interictal epileptiform discharges (IEDs) serve as a diagnostic hallmark of epilepsy and arise from hypersynchronization of epileptogenic tissue in between seizures. A deeper understanding of the underlying neurophysiological dynamics might provide an important insight into the mechanisms underlying epilepsy.
Early studies demonstrated that IEDs arise from high-frequency bursts of action potentials superimposed on a slow depolarizing potential, the so-called paroxysmal depolarizing shift (Matsumoto and Marsan, 1964). The processes underlying IED development, from a microscopic to a macroscopic perspective, have been studied by several authors. These all demonstrate that a gradual neuronal assembly build-up occurs prior to IED onset. In a study of epileptogenic tissue action potentials together with local field potentials, Keller et al demonstrated a decrement in single unit firing up to 500 milliseconds (ms) before IED onset. These changes were accompanied by a decrease in electrocorticography (ECOG) broadband (0-50 Hz) spectral power, indicating that IED hypersynchronization is preceded by a period of micro-and mesoscopic neuronal inhibition (Keller et al., 2010). The notion that cortical inhibition precedes IED onset is supported by several studies. Optical imaging techniques reflecting cellular and membranous properties have demonstrated a simultaneous activation of inhibitory interneurons alongside a broadband ECOG spectrum power decrease up to 500 ms prior to IED onset . However, there are only a few previous studies on macroscopic (scalp electroencephalographic, EEG) time-frequency changes prior to IED onset in focal epilepsy, and these studies report conflicting results, such that both broadband decreased spectral power and increased broadband synchronization have been found Jabran et al., 2020).
Although these results primarily focused on changes occurring from À500 ms up to IED onset, EEG-fMRI studies have consistently reported that local irritative zone metabolic and hemodynamical changes begin much earlier. Jacobs et al found that focal epilepsy patients exhibit blood-oxygen-level-dependent (BOLD) changes indicating that cortical metabolic changes occur up to 14 seconds prior IED onset (Jacobs et al., 2009). In a study of over 100 EEG-fMRI cases, Hawco et al found a BOLD response peak at 1 or 3 seconds before IED onset (Hawco et al., 2007). Similar changes have also been reported in animal models, where metabolic changes occur 2 seconds prior to IED onset in a GABA-inactivation model of epilepsy (Osharina et al., 2017).
In spite of the published evidence that long-term inhibitory changes play a role in gradual neuronal build up terminating in IED onset, several aspects of this build up remain obscure, and warrants further examination. Although a few studies have been conducted on neurophysiological pre-IED changes, these have all analyzed broadband EEG changes. However, different frequency bands mediate different functionality. More specifically, lowfrequency oscillations are associated with cortical inhibition in both epilepsy and in cognitive functions (Conradsen et al., 2013;Herring et al., 2019;Ikeda et al., 2020;Jensen and Mazaheri, 2010;Trevelyan and Schevon, 2013). In this study we therefore aim to analyze the role of low-frequency oscillation, from the hypothesis that specifically low-frequency oscillations are systematically altered during the pre-IED period.
We analyzed this in resting-state magnetoencephalography (MEG) measurements of 14 focal epilepsy patients. With excellent temporal resolution, MEG is well suited for characterization of time-locked pre-IED changes. As previous studies have demonstrated that the pre-IED period is characterized both by broadband power changes as well as possible cortical inhibition, we hypothesized that IEDs might be immediately preceded by low-frequency oscillations. To test this hypothesis, we examine irritative zone pre-IED low-frequency oscillations using beamformer Dynamical Imaging of Coherent Sources (DICS) (Gross et al., 2001), and statistically test IED-related low-frequency oscillatory desynchronization-synchronization using cluster-based permutation . Furthermore, since IED onset signifies a hypersynchronization overcoming such a cortical inhibition, we also aimed to examine whether the pre-IED state was characterized by an increasing time-locked cortical inhibition. To test this, we analyzed the evolution of irritative zone low-frequency (delta, 1-4 Hz, and theta, 4-8 Hz) oscillations in the pre-IED state using Hilbert transformation. Previous neurophysiological studies on pre-IED changes have analyzed only the interval [-500 ms, T0], T0 at IED onset. Since there is evidence of metabolic/hemodynamic changes occurring before À500 ms, we analyzed [-1000 ms, T0] in order to also capture earlier changes, utilizing MEG excellent temporal resolution.

Patients
Fourteen patients with focal epilepsy were included. Clinical MEG recordings were performed as a part of epilepsy evaluation at the department of clinical neurophysiology at Karolinska University Hospital during 2017-2020, after informed consent. Both children (n = 6) and adults were included (median age = 23 years, range = 7-46 age, 7 females). For demographics, focus localization and diagnoses, see Table 1. The experiment was approved by the Swedish Ethical Review Authority (DNR: 2016(DNR: /1563 and was performed in agreement with the Declaration of Helsinki.

MEG data acquisition and procedure
An Elekta NeuroMag TRIUX 306-channel MEG system with 204 planar gradiometers and 102 magnetometers was used to record MEG data. Electrooculography (EOG) and electrocardiography (ECG) was recorded simultaneously. Recordings were carried out at Karolinska Institutet (NatMEG facility) in a two-layer magnetically shielded room (Vacuumschmelze GmbH, model Ak3B). Data was sampled at 5000 Hz with an online 0.1 Hz high-pass filter and a 1650 Hz low-pass filter. Head-position indicator coils (HPI) were used to sample head movement during the recording. These were attached to the patients' head and digitalized using Polhemus Fastrak position tracker. During the recording, patients were positioned in supine position, with eyes closed and were asked to stay awake. Alertness level was quantified using EOG activity and by inspection of occipital alpha (8-13 Hz) activity. The recording lasted one hour.

Data preprocessing
Data pre-processing was performed using MaxFilter (Taulu and Simola, 2006) signal-space separation with buffer length 10 seconds and cut-off correlation coefficient 0.98. Raw data was bandpass filtered at 1-40 Hz using a Butterworth filter.

IED source reconstruction
IEDs were identified visually by an experienced physician (KW, the main author). IEDs included spikes, spike and slow wave complexes, sharp waves and polyspikes. These were selected in accordance with IED criteria in (Kane et al., 2017) (See Fig. 1a for raw IED traces). Source imaging of averaged IEDs was, for all patients, performed using both MNE-Python inverse operator Minimum Norm Estimate (MNE) (Hämäläinen and Ilmoniemi, 1994;Gramfort et al., 2013) as well as equivalent current dipole (ECD) fit (Sarvas, 1987) using Curry 7 Neuroimaging Suite (Compumedics Neuroscan). ECD dipole fit and inspection of voltage maps of individual IEDs were performed before averaging to ensure uniform IED localization within single clusters. As single epileptic foci can generate more than one type of IEDs, IEDs with uniform voltage maps but varying morphology were included (Duez et al., 2019). Parameters for MNE were set to lose = 0.2 (weight of source variance of dipole parallel to cortical surface), depth = 0.8 (depth prior weight of the forward solution), regularization parameter = 1/snr 2 with snr = 3. Pooling was performed by taking the norm of free orientations. Two complementary source imaging techniques were chosen in order to achieve accurate IED source reconstruction.
MEG source imaging cannot be used to accurately delineate the irritative zone (the IED generating area (Jehi, 2018)). Co-registration of MEG and intracranial EEG has demonstrated that hypersynchronization of 3 cm 2 cortex elicits an IED that can be identified by visual inspection of MEG data (Oishi et al., 2002). Accordingly, the irritative zone was defined as a 3 cm 2 area centered around the coordinate exhibiting maximum source estimate power and ECD dipoles of averaged, clustered IEDs. This restrictive definition of the irritative zone was utilized to avoid inclusion of nonepileptogenic tissue. The patients' clinical MRI were utilized to create a full segmentation of the head and brain, which was performed using FreeSurfer Fischl et al., 1999). The MNE-C software watershed algorithm (Gramfort et al., 2013) determined skin, skull and brain surface boundaries based upon this segmentation. A source space and a single compartment volume conductor model based upon the surface boundaries were also created using MNE-C.

Data segmentation subdivision
Data was subdivided into pre-IED epochs and control epochs. Pre-IED epochs were defined as the epoch [-1000 ms, T0] with T0 at IED onset, the beginning of the first ascending phase of the IED (please see Fig. 1b) (Kane et al., 2017, Khoo et al., 2018. Control epochs were defined as the epoch [À2000 ms, À1000 ms]. Only IEDs with control epochs without IEDs were included. Any epoch with slow wave sleep activity was excluded.

Detection of irritative zone low-frequency oscillations pre-IED
Mathematically, oscillating time series are synchronized if their phases are locked, as quantified by coherence (Rosenblum et al., 1996;Bastos and Schoffelen, 2016). Most connectivity measurements aim at detecting synchronization between separate brain regions where coherence measures can be directly applied. Detection of an IED-locked local synchronization of an area, however, cannot be achieved by such methods. Instead, quantification of oscillations is most optimally done using Dynamical Imaging of Coherent Sources (DICS) beamforming (Gross et al., 2001;Hillebrand et al., 2012).
In short, DICS is computed through the following steps: 1) Complex spectral density C of signals x(t) and y(t) is determined using Welch's method of spectral density estimation and a Hanning window is applied to these signal segments.
where X(f), Y(f) are Fourier transforms of x(t), y(t) and Y* (f) denotes the complex conjugate of Y.
2) The matrix C where element C (i, j) is the complex spectral density of signals i and j. C thus include cross spectrum densities of all possible combinations of MEG signals. The diagonal element C (i, i) represent the power spectrum of signal i. Coherence between two signals is defined as Coh (i,j) (f) = abs (C (i,j) (f)) 2 /(C (i,j) (f)C (i,j) (f)). Both measures are averaged across the frequency bands analyzed 3) A three-dimensional grid that covers the cortical surface is created and C (i, i) and Coh (i,j) (f) is computed at each grid point. 4) Time courses of coherent brain regions are extracted, and Hilbert transform is used to calculate amplitude and phase coupling between these. Hereafter, significant phase synchronization of coherent time courses was determined.
The DICS reference point was computed using a sensor-based search. In short, a cortical reference is determined using a spatial filter applied after the strongest coherence between sensors is calculated. For further details, please see (Gross et al., 2001).
DICS within delta (1-4 Hz) and theta (4-8) frequency bands was computed for the entire cortical mantle for both pre-IED and control epochs using MNE-Python. The delta and theta bands were analyzed separately. Hereafter, source estimates of the irritative zone was extracted for both conditions using built-in MNE-Python functions. Welch student's t-test was computed in order to determine if the source estimates of these conditions differed significantly. If, for any patient, statistically significant changes were found within both the delta and the theta band, this analysis was also performed for the frequency band 1-8 Hz.
In order to investigate whether these changes were specific to the irritative zone, the contralateral hemisphere was also analyzed. An of size 3 cm2 was extracted from the homologous brain region of the irritative zone on the contralateral side. In addition, ten equisized (3 cm2) extrafocal sites were analyzed in each patient. Considering the irritative zone as the center of a circle with radius 5 centimeter, the sites were placed on the perimeter with 5 centimeters spacing. The distance between these areas were chosen to avoid field spread from adjacent sites to influence the analyses (Schoffelen and Gross, 2009). DICS within the frequency band that exhibited significant changes at the irritative zone was computed for pre-IED and control epochs for each site. Hereafter, the pre-IED epochs for all extrafocal sites were compared to control epochs for all extrafocal sites using Welch student's test. To further test whether any pre-IED low-frequency changes were unique to the irritative zone, extrafocal control epochs were compared to irritative zone control epochs. For each patient, mean of the ten extrafocal control epochs was computed and compared to the irritative zone control epoch using Welch student's test.

Pre-IED ERD-ERS (event-related desynchronizationsynchronization) maps
Oscillatory changes time-locked to an event representing neuronal desynchronization or synchronization can be readily visualized using event-related desynchronization-synchronization (ERD-ERS) maps (Pfurtscheller and Lopes da Silva, 1999). Here, ERD-ERS maps were constructed to reveal if any increase in lowfrequency oscillations related to IED onset. A cluster-based permutation test designed for statistical testing of EEG/MEG-data (Maris and Oostenveld, 2007) was applied to determine if these were statistically significantly increased during pre-IED epochs compared to the control epoch defined as [-2000 ms, À1000 ms]. To handle any boundary effects, ERD-ERS maps were originally computed for the extended time interval [-2500 ms, 500 ms]. The first and last 500 ms was hereafter discarded to remove any boundary effects.
This was done as follows using MNE-Python functions: 1) All epochs were bandpass filtered within delta and theta frequency bands. 2) The time-frequency content of the epochs was determined using DPSS tapers as this technique is well suited for analysis of smaller sample sets and reduce frequency leakage (van Vugt et al., 2007). The number of DPSS tapers was set to 3 and frequency smoothing to 8 Hz in accordance with MNE-Python default settings (mne.time_frequency.tfr_multitaper ()).

3) Baseline correction was performed by subtraction of baseline values and division of the mean of these values. 4) A non-parametric cluster level paired t-test was applied to
determine if there were any statistically significant pre-IED low-frequency oscillation compared to control epochs. In short, cluster-based permutation tests (one-sided) is done accordingly (for details, see (Maris and Oostenveld, 2007): 4a) Trials from the two different conditions to be compared are collected in one set. 4b) A random partition were trials from this set are randomly drawn is created. Hereafter, test statistics is calculated on these random partitions. Threshold for cluster inclusion was set using the F-threshold that corresponds to p-value 0.05 given the number of trails. 4c) Step 4b is calculated for a random large number of partitions. A histogram of test statistics is constructed. The p-value is thus defined as the proportion of random partitions with larger test statistics than the observed one. A p-value < 0.05 concludes that the experimental conditions (here: pre-IED epochs versus control epochs) were significantly different.
Since ERD-ERS analyses are defined for sensor data Lopes da Silva, 1999, Graimann et al., 2002), ERD-ERS were computed for the subset of sensors covering the irritative zone and showing a visually identifiable IED.

Evolution of irritative zone pre-IED oscillations
We analyzed the development of pre-IED irritative zone lowfrequency oscillations, and whether the pattern of development of pre-IED oscillation differed significantly from baseline. To this end, a longer baseline ([-21 s (seconds), À1 s]) was used. This baseline was subdivided into 20 consecutive one-second control epochs. We hypothesized that low-frequency oscillation behavior should be random in these control epochs. Given the assumption that these epochs exhibited random behavior that could be described by a probability distribution, we computed if the pre-IED low-frequency oscillation behavior differed significantly from this probability. For all patients, the pre-IED epochs as well as the 20 control epochs were analyzed according to the following: Raw sensor data was bandpass filtered for delta and theta frequency bands. Minimum norm source estimates of the irritative zone were computed as described in section 2.4 IED source reconstruction. The analytic signal of the irritative zone source estimate was determined using Hilbert transformation. The envelope of this signal was extracted, and a linear fit of the envelope peaks was performed. The sign of the correlation coefficient of the linear fit was determined. A positive correlation coefficient equals an increasing linear fit, and thus an increase of oscillation amplitude throughout the analyzed epoch. Conversely, a negative correlation coefficient indicates a decrease of oscillation amplitude. Hereby, a 1x14matrix of all patients pre-IED correlation coefficient signs was constructed alongside a 20x14 matrix of all patients control correlation coefficient signs. Thus, row i of the matrices represent time [-(i + 1) s, -i s] and contains corresponding correlation coefficient signs of each patient. These binary arrays with two possible outcomes (positive or negative) can thus be modeled by a binomial distribution describing the probability of the number of such outcomes. A binomial distribution X $ Bin(n, p) is determined by parameters n and p, n the number of independent trails and p the probability of outcome 'positive'. Thus, to test any statistically significant difference between pre-IED and control conditions, a binomial distribution X was fitted to the control matrix. Parameter n was set to 14 (patients). Parameter p was calculated as (mean of number of 'positive')/n. A binomial test, a two-sided test of the nullhypothesis, was used to compute the probability (p-value) of pre-IED matrix given the fitted binomial distribution.

Results
For demographics, IED characteristics, source localization and MRI findings, see Table 1

Detection of irritative zone low-frequency oscillations pre-IED
Examining alterations in low frequency irritative zone pre-IED oscillations, our results show that all patients exhibit a statistically significant increase in power of oscillations within the delta and/or theta band in the irritative zone during the pre-IED state compared to control epoch. Four and six patients exhibited an increase within the delta and theta band, respectively, and four patients exhibited an increase within both bands separately, as well as within the 1-8 Hz frequency band (collectively referred to as ''low-frequency oscillations"). All but two patients exhibited an increase in power of low-frequency oscillations during the entire pre-IED time course [-1000 ms, T0], compared to the control time course [-2000 ms, À1000 ms]. These two patients exhibited an increase in power of low-frequency oscillations during the time course [-980 ms, T0] and [-900 ms, T0], respectively. See Table 2a for mean pre-IED DICS oscillatory power, mean baseline DICS oscillatory power and pvalue for the difference between pre-IED ([-1000 ms, T0]) and baseline ([-2000 ms, À1000 ms]) oscillatory power.
Comparing pre-IED and control low-frequency oscillations at the ipsilateral extrafocal sites demonstrated that 12 patients exhibited no significant difference in DICS power levels between these conditions. In one patient, the pre-IED epoch exhibited a statistically significant increase in DICS levels compared to the control epoch. One patient exhibited a statistically significant decrease in   Table 2b).
Comparing pre-IED and control epochs on the contralateral side (brain regions homologous to the irritative zone) demonstrated no significant difference in 13 patients. The remaining patient exhibited a significant decrease in low-frequency DICS power levels compared to the control epoch (please see Table 2c).
Comparing control condition low-frequency DICS power levels between the irritative zone and the extrafocal sites demonstrate that three patients did not exhibit any significant difference between these sites during the control condition. Of the remaining 11 patients, six patients exhibited a significantly increased lowfrequency DICS power levels at the irritative zone compared to the extrafocal sites, during control epoch (see P-value B in Table 2b). Fig. 3a. Equivalent current dipoles of IEDs (interictal epileptiform discharges) for patients 1-9. Source localization of averaged IEDs for patients 1-9 performed using Curry 7 Neurosuite. For all patients, the equivalent current dipoles coincided with minimum norm (MNE) source estimates. The irritative zone was defined as a 3 cm 2 region centered around the coordinate of maximum MNE source estimate. Please note that the right/left convention is opposite of radiological right/left convention.

Pre-IED (event-related desynchronization-synchronization) ERD-ERS map constructed using cluster-based permutation test
Statistically testing the presence of irritative zone pre-IED lowfrequency oscillations compared control epochs, all patients exhibited a statistically significant (p-value < 0.05) increase in lowfrequency synchronization between individual IEDs during the pre-IED interval [-1000 ms, T0] compared to the control epoch [-2000 ms, À1000 ms]. ERD-ERS maps demonstrating statistically significant pre-IED synchronizations are found ( Fig. 4; statistically significant low-frequency synchronization in red). To further illustrate the increase low-frequency power oscillations pre-IED, the time courses [-2000, T0] where averaged across frequencies 1-8 Hz and plotted in Fig. 5. Note that the pre-IED power increases to reach a maximum at IED onset at T0.

Evolution of irritative zone pre-IED oscillations
All patients exhibited a positive correlation coefficient direction pre-IED. None of the 20 controls exhibited similar unanimous patterns with only positive directions for all patients.

Binomial distribution fitting
Denominating positive correlation coefficients as 1, and negative correlation coefficients as 0 and adding these across the 14 patients gives results for pre-IED and controls as presented in Table  3a and 3b

Discussion
In this study, we examined the presence and buildup of changes in low-frequency oscillations during pre-IED epochs in 14 focal epilepsy patients. We hypothesized that such oscillations might reflect a cortical inhibition occurring prior to IED onset. Our results show that the pre-IED state was systematically characterized by statistically significant increases in delta and theta band synchronization. This synchronization was shown to increase during the 1000 ms preceding IED onset (see Figs. 4 & 5). To investigate whether the occurrence of pre-IED low-frequency oscillations were specific to the irritative zone, we analyzed low-frequency oscillation characteristics at extrafocal sites: in the contralateral hemisphere in the homologous region of the irritative zone, and in ten sites in the ipsilateral hemisphere -outside of the irritative zone. In the contralateral hemisphere, 13 patients exhibited no difference between pre-IED and control conditions while one patient exhibited decreased low-frequency powers compared to the control condition. In the ipsilateral hemisphere, 13 out of 14 patients did not exhibit a pre-IED-specific increase in lowfrequency oscillations compared to the control epochs at these extrafocal sites (12 patients did not have significant difference between conditions, and one had decrease in low-frequency powers compared to control condition). Thus, only the irritative zone exhibited a unanimous upregulation of low-frequency oscillations prior to IED onset. We also compare control epoch low-frequency behavior of the irritative zone and the extrafocal sites. Three patients did not exhibit any significant differences between these regions, while 55% of patients exhibited increased low-frequency power at the irritative zone (see Table 2a, 2b) compared to the extrafocal sites during control conditions. Such differences could be due to both epilepsy-associated interictal regional slowing, as well as physiologically occurring neurophysiological variations in amplitude and frequency, especially in younger subjects (Kane et al., 2017;Tao et al., 2011). Thus, Fig. 3b. Equivalent current dipoles of IEDs (interictal epileptiform discharges) for patients 10-14. Source localization of averaged IEDs for patients 10-14 performed using Curry 7 Neurosuite. For all patients, the equivalent current dipoles coincided with minimum norm (MNE) source estimates. The irritative zone was defined as a 3 cm 2 region centered around the coordinate of maximum MNE source estimate. Please note that the right/left convention is opposite of radiological right/left convention. Fig. 4. Pre-IED (interictal epileptiform discharge) ERD-ERS (event-related desynchronization/synchronization) maps. ERD-ERS maps of a subset of sensors covering the irritative zone comparing pre-IED [-1000 ms, T0] (T0 at IED onset) to control epochs [-2000 ms, À1000 ms] in frequencies 1-8 Hz (percentage increase). Red/blue: Clusters exhibiting statistically significant (p-value < 0.05) synchronization/desynchronization (normalized power). Non-significant clusters are masked. individual regional differences in frequency content of the control epochs is to be expected within a focal epilepsy population including both children and adults.
Only a few other EEG studies, and no MEG studies before ours, have analyzed pre-IED power changes. In contrast to the results reported here, these EEG studies demonstrated a decrease in pre-IED power. It should however be noted that these studies analyzed broadband 1-50 Hz changes without a subdivision into smaller frequency bands Jabran et al., 2020;Keller et al., 2010). Speculatively, it is possible that this broadband power decrease obscured the delta and theta band power increase reported here. Although these studies were conducted using different acquisition modalities, the choice of EEG or MEG should not influence study results.
Several studies on pre-IED dynamics have suggested that the pre-IED state is characterized by cortical inhibition. Since low frequencies are associated with cortical inhibition, we suggest that the results reported here support this notion. Within normal brain physiology, low frequencies gate information-routing through mediation of cortical inhibition (Jensen and Mazaheri, 2010). This can also be induced by external direct current stimulations, where low frequency pulses can be used to interrupt visual gamma activity (Herring et al., 2019). Similar observations have been made also related to epilepsy, where therapeutic intervention has shown that low frequency TMS (transcranial magnet stimulation) can be used to inhibit seizure generation (Kile et al., 2010). Furthermore, cellular recordings have demonstrated that neurons exhibit a period up to ten seconds of inhibition prior to the paroxysmal depolarizing shift, and similar changes can be seen surrounding an ictal core (so called surround inhibition; (Schevon et al., 2012;Trevelyan and Schevon, 2013). Other studies have revealed that this surround inhibition is mediated by a low frequency LFP, and such low frequency activity can also give rise to ictal discharges . To this background, we suggest that the low frequency oscillations reported here represent such pre-IED cortical inhibition. We also demonstrated that the power of low-frequency oscillations increased during the pre-IED epoch (see Tables 3a, 3b), Table 2a Irritative zone mean pre-IED (interictal epileptiform discharge) and mean control DICS (Dynamical Imaging of Coherent Sources) power (unit: Ampere meter) and p-value for Welch student's t-test comparing power levels of pre-IED [-1000 ms, T0] and control [-2000  Delta and theta 4.18e-24 3.18e-24 *** 14 Theta 5.05e-24 3.77e-24 *** P-value: * = <0.05 ** = <0.005 *** = <0.001 + = Significant changes found separately in delta and theta bands ++ = Mean power calculated for frequency band 1-8 Hz while similar changes were unseen during control epochs. These findings indicate the pre-IED cortical inhibition might grow until IED onset. Although IEDs and seizure activity constitute two different cortical processes, it can be noted that low frequency activity also plays a role in seizure dynamics. Interestingly, it has been suggested that slow activity might be linked to an ictal inhibitory penumbra (Eissa et al., 2016). It is possible that low-frequency oscillations might reflect cortical inhibition, both in seizure and IED development.
It is noteworthy that while other studies on macroscopic time frequency pre-IED changes found effects only at À400 ms, our results demonstrated changes already from time À1000 ms. Analyzing the evolution of irritative zone pre-IED oscillations demonstrates that the low-frequency power increases from À1000 ms up to IED onset. However, one sub-study also indicated that the [-2000 ms, À1000 ms] epoch might be characterized by a decrease within delta and theta bands. As we have not explored this further, these findings should be considered with caution and addressed in a subsequent study (see Table 3a, 3b).
Several hemodynamical studies as well as cellular recordings (Hawco et al., 2007;Jacobs et al., 2009;Osharina et al., 2017) indicate changes as early as 10 seconds prior to IED onset. Thus, it is possible that neurophysiological pre-IED changes occurring at [-1000 ms, 0] might arise from this long-term irritative zone modulation.
We also demonstrated that low-frequency oscillations were unique to the irritative zone in 12 out of 14 patients. In one of the remaining patients, the examined extrafocal regions also exhibited increased power in low-frequency oscillations during the pre-IED epoch compared to control. It is possible that this patient did not only exhibit focal cortical inhibition, but global pre-IED inhibition. Finally, one patient exhibited increased power in extrafocal low-frequency oscillatory activity during control than during pre-IED epochs, indicating that pre-IED low-frequency oscillations were unique to the irritative zone. However,  Mean number of control positive correlation coefficients = 6.6 gives p = 0.47. Thus, the fitted binomial distribution is X $ Bin (14, 0.47). The binomial test of the pre-IED correlation coefficients deviating from the expected distribution gives a p-value < 0.01. interpretation of results from individual patients should be done with caution. The included patients exhibited heterogeneous underlying etiologies. Consequently, included IEDs were of varying appearance as this depends on epilepsy diagnosis (Smith, 2005). Epilepsy constitutes a syndrome with etiologies ranging from monogenetic mutations to cortical malformations (Staley, 2015). It is thus possible that individual diagnoses as well as patient age could be associated with different underlying pathophysiology. However, several studies on pre-IED changes have included heterogeneous populations and none of these found that pre-IED dynamics depended on underlying etiology or patient age. (Hawco et al., 2007;Jabran et al., 2020;Jacobs et al., 2009). Thus, it is likely that pre-IED changes can be compared across different patient characteristics.
Our study has limitations. Cluster-based permutation tests have become an attractive and popular statistical method for analyzing EEG/MEG data, with both high Type II and low Type 1 error rates. However, although permutation cluster tests can be used to test whether a statistically significant difference exists, they should not be used for high-precision characterization of time points or frequency bands where power changes occurs. It is also well known that usage of different time frequency methods might give rise to different results (Wacker and Witte, 2013). Thus, in this study, complementary techniques giving additional characterization of pre-IED irritative zone time-frequency changes were also employed. Furthermore, some included patients exhibited a low number of IEDs. Although this constitutes a study limitation, including patients with a varying number of IEDs reflects the general focal epilepsy population.
In summary, our results indicate that the pre-IED state is characterized by a systematic increasing cortical inhibition that terminates at IED onset. However, these findings need to be corroborated by future studies. As epilepsy is a network disorder, pre-IED changes should also be analyzed within the epileptic network.