Sevoflurane-induced high-frequency oscillations, effective connectivity and intraoperative classification of epileptic brain areas

Abstract

Objective:

To determine how sevoflurane anesthesia modulates intraoperative epilepsy biomarkers on electrocorticography, including high-frequency oscillation (HFO) effective connectivity (EC), and to investigate their relation to epileptogenicity and anatomical white matter.

Methods:

We studied eight pediatric drug-resistant focal epilepsy patients who achieved seizure control after invasive monitoring and resective surgery. We visualized spatial distributions of the electrocorticography biomarkers at an oxygen baseline, three time-points while sevoflurane was increasing, and at a plateau of 2 minimum alveolar concentration (MAC) sevoflurane. HFO EC was combined with diffusion-weighted imaging, in dynamic tractography.

Results:

Intraoperative HFO EC diffusely increased as a function of sevoflurane concentration, although most in epileptogenic sites (defined as those included in the resection); their ability to classify epileptogenicity was optimized at sevoflurane 2 MAC. HFO EC could be visualized on major white matter tracts, as a function of sevoflurane level.

Conclusions:

The results strengthened the hypothesis that sevoflurane-activated HFO biomarkers may help intraoperatively localize the epileptogenic zone.

Significance:

Our results help characterize how HFOs at non-epileptogenic and epileptogenic networks respond to sevoflurane. It may be warranted to establish a normative HFO atlas incorporating the modifying effects of sevoflurane and major white matter pathways, as critical reference in epilepsy presurgical evaluation.

1. Introduction

1.1. Motivation for the Present Study

Drug-resistant focal epilepsy effects millions of children around the world, and prompt intervention is needed to prevent cognitive delays (Laguitton et al., 2021; Morningstar et al., 2021; Sultana et al., 2021). This treatment often entails a two-stage surgery that includes multiple neurosurgeries, implantation of intracranial subdural and/or depth electrodes, and days of extra-operative intracranial EEG (iEEG) recording, to determine the boundaries of the presumed epileptogenic zone for subsequent resection (Asano et al., 2009; Hader et al., 2004; Jayakar et al., 2016; Mullin et al., 2016; Roth et al., 2021; Uribe-Cardenas et al., 2020). Compared to continued medical therapy, resective surgery leads to better seizure control in patients with drug-resistant focal epilepsy (Dwivedi et al., 2017; Widjaja et al., 2020; Wiebe et al., 2001). However, this two-stage procedure entails many significant risk factors such as infection and increased intracranial pressure associated with chronic iEEG electrodes (Belohlavkova et al., 2019; Hader et al., 2013; Mullin et al., 2016). To avoid these issues, it is instead preferable to conduct one-stage surgeries where resection immediately follows acute, intraoperative electrocorticography (ECoG) (Bansal et al., 2017; Krsek et al., 2013). However, the utility of intraoperative ECoG is currently limited because general anesthesia reduces the occurrence rate of interictal epileptiform discharges – such as spike-and-wave signals – that are suggested to estimate the epileptogenic zone (Asano et al., 2009; Bayram et al., 2021; Weiss et al., 2015). In addition, these signals are typically identified via visual inspection, which is time consuming and subject to rater bias (Asano et al., 2009; Hader et al., 2004; Jayakar et al., 2016; Kural et al., 2020; Roth et al., 2021; Uribe-Cardenas et al., 2020). To help facilitate the use of one-stage procedures and enhance intraoperative seizure focus localization, there is great need to develop objective ECoG epilepsy biomarkers that can be reversibly induced in a single surgical session. Achieving this goal is expected to reduce clinical diagnostic burden and optimize treatment cost-effectiveness, while maintaining postoperative seizure outcomes comparable with two-stage surgery.

1.2. Sevoflurane Anesthesia Modulates High-Frequency Oscillations

Interictal high-frequency oscillations (HFOs) at ≥80 Hz are a component of epileptiform spike-and-wave discharges that represent a potentially valuable neurophysiology biomarker for localizing the epileptogenic zone (Wang et al., 2013; van 't Klooster et al., 2015; Frauscher et al., 2017; Bernardo et al., 2018; Ohuchi et al., 2019; Kural et al., 2020; Schönberger et al., 2020). Their occurrence rates and phase-amplitude coupling (PAC) with delta waves on extraoperative iEEG recording are often increased in the seizure onset zone, and resection of such elevated-HFO areas is associated with good postoperative seizure control (Frauscher et al., 2017; Guth et al., 2021; Jacobs et al., 2008; Kuroda et al., 2021; Motoi et al., 2018; Nonoda et al., 2016; Wang et al., 2013; van 't Klooster et al., 2017). Sevoflurane – a Federal Drug Administration (FDA)-approved anesthetic used as standard care in pediatric operating rooms – is known to acutely activate epileptiform signals in patients with and without epilepsy, it is easily titratable, and its effects are rapidly reversible (Åkeson and Didriksson 2004; Bayram et al., 2021; Cornelissen et al., 2015, 2018; Edgington et al., 2022; Gibert et al., 2012; Iijima et al., 2000; Jääskeläinen et al., 2003; Komatsu et al., 1994; Kurita et al., 2005; Särkelä et al., 2007; Schultz et al., 2000; Stasiowski et al., 2019; Tanaka et al., 2017; Woodforth et al., 1997). Furthermore, recent intraoperative ECoG studies demonstrated that sevoflurane augmented HFO occurrence rates and coupling with delta waves preferentially in resected, epileptogenic sites (Orihara et al., 2020; Wada et al., 2022). However, these results do not describe how the spatiotemporal dynamics – effective connectivity (EC) – of HFO signal propagation respond to sevoflurane, nor the anatomical white matter pathways used for propagation. Gleaning this knowledge is expected to enhance HFO-based localization efforts, as clinical studies have shown that effective connectivity is elevated in the seizure onset zone (Guo et al., 2021; Park et al., 2017; Park et al., 2018; Parker et al., 2017; Rotondi et al., 2016; Shih et al., 2021; Tenney et al., 2018; Yin et al., 2020).

1.3. Study Aims

This observational study builds on our previous ECoG work that suggested sevoflurane diffusely increased HFO-delta phase-amplitude coupling, although most in the epileptogenic zone (Wada et al., 2022). We utilized modulation index (MI) to quantify the strength of coupling between delta phase and HFO amplitude, which is an outstanding summary measure of interictal spike-and-wave discharges (Kural et al., 2020; Kuroda et al., 2021). For this study, we treated MI as a supplemental reference to assess the diagnostic utility of HFO effective connectivity, and here we also quantified MI using HFOs defined at 80-300 Hz.

In an effort to implement continuous neurophysiology markers, we mainly sought to determine how well sevoflurane-modulated HFO effective connectivity could classify epileptogenic sites, during surgery. This study defined HFOs as high-frequency (80-300 Hz) augmentation determined by time-frequency domain analysis, as stated in the Methods section below; thus, HFOs in the present study were effectively equivalent to high-frequency activity at 80-300 Hz, a term proposed in Noorlag et al., (2019). We defined “epileptogenic sites” as those resected in patients who achieved International League Against Epilepsy (ILAE) class 1 seizure outcomes (Wada et al., 2022; Wieser et al., 2001). We tested the hypothesis that intraoperative HFO effective connectivity would be elevated as a function of sevoflurane anesthetic stage, and the degree of enhancement would be higher in epileptogenic sites. Our hypothesis was predicated on the observations that sevoflurane activates spike-and-wave discharges (discussed above), which are characterized by an HFO component, rapid spatial propagation, and resultant effective connectivity augmentation (Kural et al., 2020; Kugiumtzis et al., 2017). To compute effective connectivity, we utilized transfer entropy (TE): an algorithm that quantifies unidirectional information flow by calculating the ability of one site to alter the predictive capability about the future state of another (Ito et al., 2011). Furthermore, we developed a novel variation of dynamic tractography to visualize the sevoflurane-dependent dynamics of HFO signal propagation via direct white matter tracts, as a function of anesthetic stage. To do so, we applied diffusion-weighted imaging (DWI) tractography similar to that reported in our previous studies, in combination with HFO effective connectivity (Mitsuhashi et al., 2021; Silverstein et al., 2020; Sonoda et al., 2021). This technique may help clarify the anatomical highways used for epileptic HFO propagation and help justify future large cohort investigations into seizure outcome prediction using these biomarkers.

Within the spike-and-wave structure, HFOs are typically coupled to delta waves (Kural et al., 2020), local slowing is a hallmark feature of drug-resistant focal epilepsy, and sevoflurane is known to augment delta waves (Cornelissen et al., 2015, 2018; De Stefano et al., 2022). We hence tested the additional hypothesis that intraoperative delta-TE (defined for this study as 3-4 Hz; Nonoda et al., 2016; Iimura et al., 2018) would increase as a function of sevoflurane concentration and be highest in epileptogenic areas. Finally, the input for ECoG-based TE is a spectral amplitude (SA) time series, which represents the square-root of power for EEG frequency bands. We conducted a supplementary analysis to determine the effects of sevoflurane on delta and HFO-SA.

2. Methods

2.1. General Methods

This is an observational study of eight pediatric patients that underwent resective epilepsy surgery and subsequently achieved ILAE class 1 seizure outcomes (Wada et al., 2022). All data was gleaned from standard-of-care treatment for drug-resistant focal epilepsy, with absolutely no deviation from clinical protocols. All patients underwent the following procedures [i] intracranial electrode implantation, [ii] extraoperative iEEG recording to determine the resection margin, and [iii] intraoperative ECoG recording using the same implanted electrodes, in the presence of sevoflurane anesthesia, followed by [iv] focal cortical resection. In the present study, we retrospectively employed computational signal processing on [iii] intraoperative ECoG data (Figure 1). We quantified ECoG-defined HFO and delta-TE at an oxygen baseline, three time-points while sevoflurane was dynamically increasing, and another during a maintenance dose of sevoflurane at 2 minimum alveolar concentration (MAC). A linear mixed model determined if those measures increased as a function of sevoflurane anesthetic stage. A binary logistic mixed model deduced if sevoflurane-augmented HFO and/or delta-TE could significantly classify the epileptogenic sites. A supplementary analysis tested the same hypotheses using MI and SA.

Figure 1. Methodological summary.

A. Patients were implanted with either surface and/or depth intracranial electrodes to map epileptogenic and eloquent cortex. B. Raw intraoperative electrocorticography (ECoG) traces from a representative patient as sevoflurane was increased from an oxygen baseline (bottom) to 2 minimum alveolar concentration (MAC; top). C. ECoG data was then mathematically transformed from the time-voltage domain into the time-frequency domain; representative time-frequency transformed electrode under oxygen baseline (left) and sevoflurane (right). D. Electrodes were co-registered on each patient’s three-dimensional magnetic resonance image (MRI) reconstructed cortical surface for further analysis (patient 6 with right frontal distribution shown). Left and right images show lateral and medial views, respectively. For patient 6, A-B are depth electrodes and C-H are surface electrodes. Yellow dotted line represents the resection margin.

2.2. Patient Population

This study included eight young patients (ages 4-22 years old; 5 males) with a diagnosis of drug-resistant focal epilepsy. The eligibility criteria were identical to those in Wada et al., (2022), and patient characteristics can be found in Table 1. All patients underwent two-stage resective epilepsy surgery between October 2018 - March 2020 at the National Center for Neurology and Psychiatry, Tokyo, Japan. We obtained written consent from all patients or legal guardians for those under 18 years old and from those who were unable to provide their own written consent.

Table 1.

Patient profiles.

Patient Number	Age (years)	Sex	Anti-seizure medications	Number bipolar electrode pairs (epileptogenic; non- epileptogenic)	Resection area	Pathology
1	14	F	TPM, VPA	75 (35;40)	Left frontal	Gliosis
2	19	M	CBZ, LEV, LTG, PER	63 (29;34)	Left occipital	Ulegyria
3	11	F	LEV, LTG, VPA	120 (17;103)	Right frontal and parietal	Polymicrogyria
4	12	F	CBZ, CLB, TPM	114 (27;87)	Right temporal and parietal	Gliosis
5	5	M	ZNS, LEV, CLB	41 (17;24)	Left frontal and temporal	FCD type IIb
6	11	M	CBZ, VPA, LTG	62 (5;57)	Right frontal	FCD type IIa
7	4	M	CLB, LEV, LCM, PB	62 (34;28)	Right frontal	FCD type IIb
8	22	M	CBZ, LCM	84 (38;46)	Right temporal	Gliosis

CBZ = Carbamazepine; CLB = Clobazam; F = Female; FCD = Focal cortical dysplasia; LCM = Lacosamide; LEV = Levetiracetam; LTG = Lamotrigine; M= Male; PB = Phenobarbital; PER = Perampanel; TPM = Topiramate; VPA = Valproic acid; ZNS = Zonisamide.

2.3. Intracranial Electrode Placement

All patients received either surface platinum disk electrodes (10 mm center-to-center), stereo-depth electrodes (5 mm center-to-center), or a combination of the two. Intracranial electrodes were implanted on the affected hemisphere to determine the boundaries of the presumed epileptogenic zone via subsequent extraoperative iEEG recording; their placement was guided by noninvasive presurgical evaluation from a multidisciplinary team of physician specialists who synthesized neuroimaging, neurophysiology, and clinical data (Takayama et al., 2020; Wada et al., 2022). The spatial extent of electrode coverage was strictly based on clinical necessity, with no attempt to increase the number of electrodes for scientific curiosity.

2.4. Extraoperative iEEG Recording

After implantation of intracranial electrodes, patients were sent to the inpatient ward for invasive, extraoperative iEEG recording (Takayama et al., 2020; Wada et al., 2022). Intracranial EEG signals were recorded using a 1,000 Hz sampling rate and a bandpass filter from 0.016 to 300 Hz. We reconstructed the three-dimensional magnetic resonance image (MRI) of each patient’s brain and co-registered the electrodes to their corresponding anatomical locations, as previously reported (Figure 1; Nakai et al., 2017; Wada et al., 2022). Seizure events were captured in all eight children, and a board-certified epileptologist (M.I.) used this data to determine the location of seizure onset zones (Asano et al., 2009).

2.5. Intraoperative ECoG Immediately Before Initiating Cortical Resection

After completing the extraoperative iEEG procedures, patients were transferred to the operating room for resection of the presumed epileptogenic zone. All procedures were standard-of-care treatment, and ECoG recording under the various anesthetic stages and related analysis have been approved by the Institutional Ethical Committee at the National Center of Neurology and Psychiatry, Tokyo, Japan (approval number: A2021-050). As reported previously (Wada et al., 2022), intraoperative ECoG was performed using the same electrodes from the extra-operative iEEG sessions. General anesthesia was induced with propofol (median dose: 2.07 mg/kg), immediately followed by remifentanil for analgesia (median dose: 2.00 mcg/kg) and rocuronium for muscle relaxation. Patients were then intubated, and sevoflurane was dynamically increased from 0 to a 2 MAC maintenance dose. During the intraoperative recording epoch, patients were also given 100% oxygen. ECoG was performed prior to anesthetization, while sevoflurane was rising, and for an additional 4 minutes at 2 MAC (Wada et al., 2022). Note, all sevoflurane doses were within clinically accepted limits (Malan et al., 1995; Wada et al., 2022).

2.6. Definition of Anesthetic Stages

The intraoperative ECoG recording was divided into five stages for further analysis: [1] 4-minute oxygen baseline; [2-4] three 2-minute time periods while sevoflurane was dynamically increasing from 0 to 2 minimum alveolar concentration: ‘sev increasing first’, ‘sev increasing mid’, and ‘sev increasing last’; and [5] 4 minutes while sevoflurane was held at 2 MAC maintenance. We analyzed MI, SA, and TE at each anesthetic stage to determine which period optimized the classification of epileptogenic sites. Anesthetic conditions were determined via clinical necessity, with no attempt to increase concentration nor duration for research purposes.

2.7. Focal Cortical Resection

Finally, resection was carried out by a board-certified neurosurgeon aiming to remove the presumed epileptogenic zone (seizure onset zones and neighboring MRI-visible lesions), while preserving eloquent cortex to minimize development of cognitive or sensorimotor deficits. Importantly, the results discussed below were not available to influence treatment plans.

2.8. Definition and Identification of Epileptogenic Electrode Sites

As done previously (Wada et al., 2022), we retrospectively categorized resected electrode sites as ‘epileptogenic’ and those retained as ‘non-epileptogenic’. These distinctions were confirmed with intraoperative photographs, as our previous work showed high concordance between postoperative MRI-defined resection margins and peri-surgical photographs (Kuroda et al., 2021). Since all patients in this study achieved ILAE-defined class 1 outcomes, we can classify electrodes as such because seizure freedom implies that the entire epileptogenic zone was sufficiently removed.

2.9. Quantification of ECoG-Derived Modulation Index

Investigators (E.F. and N.K.) who were blind to the epileptogenic status of electrodes exported ECoG signals onto a bipolar montage. Next, preprocessing was performed to remove artifactual ECoG channels and recording epochs. This data was then input into the open-source winPACT toolbox of EEGLAB (https://sccn.ucsd.edu/wiki/WinPACT), to compute MI for each electrode at all anesthetic stages (Delorme and Makeig 2004; Miyakoshi et al., 2013; Kuroda et al., 2021; Wada et al., 2022). We calculated PAC, as done previously (Miyakoshi et al., 2013; Kuroda et al., 2021). This software tool Hilbert transforms the EEG time series and quantifies the strength of coupling between the instantaneous phase of delta waves (3-4 Hz) and HFO (80-300 Hz) amplitude. Higher MI values reflect stronger phase-amplitude coupling.

2.10. Quantification of ECoG-Derived Spectral Amplitude and Transfer Entropy

Bipolar montaging and preprocessing were carried out identical to the above. FieldTrip software (https://www.fieldtriptoolbox.org/) was then used to transform the ECoG data from the time-voltage to the time-frequency domain via the wavelet method (Oostenveld et al., 2011), for delta waves (3-4 Hz) and HFOs (80-300 Hz). Spectral amplitude - the square-root of power - was used here and z-scored based on the mean and standard deviation of all signals within the baseline anesthetic stage, for each individual patient. For SA statistical analysis, all time-points for each electrode were averaged into one composite SA value, for each anesthetic stage and frequency band. For TE, the z-scored SA time series data was input into a transfer entropy Matlab R2020a program (MathWorks, Natick, MA, USA) based on the algorithm detailed by Ito et al. (2011). We customized the script to be compatible with ECoG signals as follows: [1] the ECoG time series was divided into time bins with lengths at least 6 wave-cycles in duration, depending on the frequency band; [2] within each bin, if the z-scored SA value eclipsed a threshold of 2 for at least 3 wave-cycles then the bin was converted into a binary ‘1’; and [3] the binary time series data was fed into the Ito et al., (2011) algorithm, which determines if activity at a given electrode site can enhance the predictive capability of the future state of another site. Since TE inherently measures conditional mutual information (Schreiber 2000), represented as "bits", it was necessary to binarize the continuous ECoG data for compatibility with the equation. At any given electrode, all of its efferent (unidirectional TE emanating from a given site), afferent (unidirectional TE converging on a given site), or total TE connections (both efferent and afferent) were averaged to get a single value for statistical analysis, for each anesthetic stage and frequency band.

2.11. Diffusion-Weighted Imaging and Dynamic Tractography

DSI Studio software (https://dsi-studio.labsolver.org/) was used to quantify DWI white matter tractography, as done previously by our group (Mitsuhashi et al., 2021). In short, open-source DWI data from 1,065 patients in the Human Connectome Project (HCP 1065; Yeh et al., 2018) was input into DSI Studio to create a standardized whole-brain tractography template in Montreal Neurological Institute (MNI) space. This template was made using whole-brain seeding with the following parameters: tracking threshold of 0.7, angular threshold of 70 degrees, and step-size of 0.3 mm. Board-certified neurosurgeons (N.K. and K.S.) mapped all electrodes to each patient’s reconstructed 3-dimensional MRI brain surface. FreeSurfer software (https://surfer.nmr.mgh.harvard.edu/) then spatially normalized the patient electrode coordinates into MNI-space. Each normalized electrode site was finally converted to its corresponding region on the Lausanne brain atlas (Hagmann et al., 2008) and overlayed on the HCP 1065 whole-brain tractography template. In DSI Studio, the electrode sites were used as regions-of-interest (ROIs) to compute a generalized fractional anisotropy (GFA) connectivity matrix for each patient. The white matter connectome between all electrode sites from an individual patient could then be visualized via DSI Studio by calculating streamlines connecting all site-pairs displaying positive GFA values in the connectivity matrix. Finally, dynamic tractography combined the patients’ DWI connectomes with TE to visualize major white matter pathways carrying HFO propagations, as a function of anesthetic stage. In short, white matter streamlines connecting each GFA positive site-pair were colored based on the HFO-TE values describing effective connectivity for the same pair. This visualization was repeated for each level of sevoflurane.

2.12. Statistical Analysis

Relationship between anesthetic stages and TE-based neural propagations:

We employed a linear mixed model to address the hypothesis that intraoperative HFO and/or delta-TE will increase as a function of sevoflurane stage. The fixed effect predictors were [1] age, [2] sex, [3] hemisphere, [4] number of anti-seizure medications, and [5] sevoflurane stage, patient and intercept were considered random effects predictors, and the dependent variable was either HFO or delta-TE. A fixed effect predictor greater than zero suggests that TE increases as a function of sevoflurane stage. We considered a two-sided p-value < 0.025 as significant.

Utility of TE-based neural propagations in classifying epileptogenicity:

We utilized a binary logistic mixed model to answer the hypothesis that intraoperative HFO and/or delta-TE modulated by sevoflurane can classify electrode epileptogenic status. The fixed effect predictors were [1] age, [2] sex, [3] hemisphere, [4] number of anti-seizure medications, and [5] TE, patient and intercept were considered random effects predictors, and the dependent variable was the epileptogenic status (yes/no) of electrode sites. We repeated this for each of the five anesthetic stages, along with both frequency bands, and considered a two-sided p-value < 0.005 as significant.

Analyses on modulation index (MI) and HFO/delta spectral amplitude:

A supplemental analysis using identical linear and binary logistic mixed models was carried out to test the same hypotheses for MI and SA.

2.13. Judging the Effect of a Potential Bias Between Patients

Investigators suggest that the epileptogenic zone is often large in patients with frontal lobe epilepsy (Salanova et al., 1994; Smith et al., 1997), so one might hypothesize that the relative size of the epileptogenic sites could be a potential bias in the present study. Therefore, we conducted an ancillary analysis using Spearman's rank correlation to assess whether the observed discrimination performance of each diagnostic biomarker was associated with the relative size of the epileptogenic zone. The dependent variable was the effect size of the difference in biomarker values between the epileptogenic and non-epileptogenic sites. In each patient, Cohen's d was employed to estimate effect sizes (Chacón et al., 2021; Cohen 1988; Kotiuchyi et al., 2020; Whelan et al., 2018). It was defined as the absolute value of: [(the mean biomarker value in the epileptogenic sites) – (the mean biomarker value in the non-epileptogenic sites)] / (the standard deviation for all sites). The independent variable was defined as: (the number of epileptogenic sites) / (the total number of electrode sites), for each patient. We employed a False Discovery Rate (FDR; Benjamini and Hochberg 1995) correction because of the repeated testing for five different dependent variables (i.e., modulation index_{3-4 Hz and 80-300 Hz}, delta-SA_{3-4 Hz}, HFO-SA_{80-300 Hz}, delta-TE_{3-4 Hz}, and HFO-TE_{80-300 Hz}) and five anesthethic conditions.

3. Results

3.1. Sevoflurane Increases Modulation Index

We first sough to establish MI as a reference marker for classifying epileptogenicity, in line with our previous study (Wada et al., 2022); here we defined HFO at 80-300 Hz, and MI quantified the strength of coupling between HFO_{80-300 Hz} amplitude and the phase of delta waves_{3-4 Hz}. Linear mixed model analysis showed that MI significantly increased as a function of increasing sevoflurane concentration, across all 621 pooled electrode sites, regardless of epileptogenicity (fixed effect estimate = 0.079; 95% confidence interval [CI] = 0.070 – 0.087; t-value = 18.09; p-value < 0.001; degrees of freedom [df] = 3,033; Supplementary Figure 1A). Binary logistic mixed model analysis further demonstrated that at the ‘sev increasing last’ anesthetic stage, MI significantly classified the epileptogenic status of electrodes (df = 615; t-value = 4.049; p-value < 001; odds ratio [OR] = 3.261; 95% CI = 1.838 – 5.786; Supplementary Figure 1B). A detailed breakdown of the statistical analysis for each anesthetic stage can be found in Supplementary Table 1.

3.2. Delta Spectral Amplitude

We then determined the effects of sevoflurane on delta spectral amplitude. Linear mixed model results showed that delta-SA at all sites (n = 621) was significantly elevated by increasing the concentration of sevoflurane (fixed effect estimate = 0.329; 95% CI = 0.299 – 0.359; t-value = 21.656; p-value < 0.001; df = 3,096; Supplementary Figure 2). Binary logistic mixed model analysis delved into the ability of delta rhythms to classify epileptogenicity, and relatively smaller delta-SA augmentation significantly classified the epileptogenic status of electrodes during all conditions except the baseline (Supplementary Figure 2; Supplementary Table 2).

3.3. Delta Transfer Entropy Rated Effective Connectivity

For effective connectivity, delta-TE was not significantly modulated as a function of increasing sevoflurane concentration (fixed effect estimate = 6.465E-5; 95% CI = −2.889E-5 - 0.000; t-value = 1.355; p-value = 0.175; df = 3,096; Figure 2A). Binary logistic mixed model analysis showed that relatively smaller delta-TE augmentation significantly identified epileptogenic sites at the ‘sev increasing last’ segment (df = 615; t-value = −3.451; p-value = < 0.001; OR = 2.100E-46; 95% CI = 2.131E-72 – 2.068E-20; Table 2; Figure 2B).

Figure 2. Delta effective connectivity responds to sevoflurane anesthesia.

A. Delta transfer entropy values of pooled individual electrode sites (n = 621), as a function of sevoflurane concentration. The black trend line represents the equation predicted from the linear mixed model analysis. B. Each patient’s (n = 8) average delta transfer entropy for epileptogenic (yellow; n = 202 total sites) and non-epileptogenic (green; n = 419 total sites) sites, at each anesthetic stage. Yellow and green dots represent the average delta transfer entropy value of all epileptogenic or non-epileptogenic electrode sites, respectively, for individual patients. Lines connect a given patient’s non-epileptogenic and epileptogenic values. Asterisks denote binary logistic mixed model significance (p < 0.005) for classifying epileptogenic sites. For each patient and anesthetic stage, effect size was estimated using Cohen’s d defined as the absolute value of: [(the mean biomarker value in the epileptogenic sites) – (the mean biomarker value in the non-epileptogenic sites)] / (standard deviation for all sites). The average Cohen’s d across all eight patients is shown under the corresponding anesthetic stage. O2 = oxygen baseline; SI1 = sevoflurane increasing first third; SI2 = sevoflurane increasing mid third; SI3 = sevoflurane increasing last third; Sev2 = sevoflurane 2 minimum alveolar concentration (MAC).

Table 2.

Delta transfer entropy binary logistic mixed model classification of epileptogenic sites.

Anesthetic Stage	t-value	p-value	Odds Ratio	95% CI (lower)	95% CI (upper)
Oxygen Baseline (df = 615)	1.855	0.064	5.357E71	6.150E-5	4.666E147
Sev Increasing First (df = 615)	−0.045	0.964	0.398	1.294E-18	1.224E17
Sev Increasing Mid (df = 615)	−2.245	0.025	1.747E-29	1.205E-54	2.53E-4
Sev Increasing Last (df = 615)	−3.451	< 0.001	2.100E-46	2.131E-72	2.068E-20
Sevoflurane 2 MAC (df = 615)	−1.483	0.139	6.274E-40	7.567E-92	5.203E12

Significant classifiers (p < 0.005) are bolded. CI = confidence interval; df = degrees of freedom; MAC = minimum alveolar concentration.

3.4. HFO Spectral Amplitude

We next wanted to determine the effects of sevoflurane on HFOs. Beginning with spectral amplitude, linear mixed model analysis of all pooled electrode sites (n = 621), regardless of epileptogenicity, demonstrated that HFO-SA slightly, albeit significantly, decreased as a function of progressive sevoflurane stage (fixed effect estimate = −0.088; 95% CI = −0.102 - −0.073; t-value = −11.95; p-value < 0.001; df = 3,096; Supplementary Figure 3). It should be noted that a large subset of electrodes showed an increase of HFO-SA as sevoflurane became more concentrated (Supplementary Figure 3). Binary logistic mixed model analysis demonstrated that HFO-SA could not classify the epileptogenic status of electrodes (Supplementary Table 3).

3.5. HFO Transfer Entropy Rated Effective Connectivity

We subsequently set out to investigate the effects of sevoflurane anesthesia on HFO effective connectivity rated by TE. Linear mixed model results showed that HFO-TE at all sites (n = 621), regardless of epileptogenicity, significantly increased as a function of enhanced sevoflurane stage (fixed effect estimate = 3.710E-4; 95% CI = 2.87E-4 - 4.56E-4; t-value = 8.604; p-value < 0.001; df = 3,096; Figure 3A). Binary logistic mixed model analysis demonstrated that elevated HFO-TE significantly classified the epileptogenic status of electrodes, at all anesthetic stages, although most successfully at sevoflurane 2 MAC (df = 615; t-value = 7.177; p-value < 0.001; OR = 1.582E115; 95% CI = 4.738E83 – 5.280E146; Table 3; Figure 3B). Figure 3C visualizes the anatomical distribution of patient 6’s high-value TE connections in relation to the resection margin, for both HFOs and delta waves at each anesthetic condition. The HFO-TE signals are first non-specifically distributed, but at sevoflurane 2 MAC, they are confined to the epileptogenic area (Figure 3C).

Figure 3. Sevoflurane activates high-frequency oscillation (HFO) effective connectivity most in the epileptogenic zone.

A. HFO transfer entropy (TE) for all pooled individual electrodes (n = 621) as a function of sevoflurane concentration. The black trend line represents the equation predicted from the linear mixed model analysis. B. Each patient’s (n = 8) average HFO transfer entropy for epileptogenic (yellow; n = 202 total sites) and non-epileptogenic (green; n = 419 total sites) sites, at each anesthetic stage. Yellow and green dots represent the average HFO transfer entropy value of all epileptogenic or non-epileptogenic electrode sites, respectively, for individual patients. Lines connect a given patient’s non-epileptogenic and epileptogenic values. Asterisks denote binary logistic mixed model significance (p < 0.005) for classifying epileptogenic sites. For each patient and anesthetic stage, effect size was estimated using Cohen’s d defined as the absolute value of: [(the mean biomarker value in the epileptogenic sites) – (the mean biomarker value in the non-epileptogenic sites)] / (standard deviation for all sites). The average Cohen’s d across all eight patients is shown under the corresponding anesthetic stage. C. Right hemisphere of patient 6 with electrodes mapped to reconstructed three-dimensional cortical surface. HFO (red arrows) and delta (blue arrows) effective transfer entropy connections above 6 standard deviations between electrode sites, at each anesthetic condition. Yellow dotted lines outline the resection margin (i.e., the epileptogenic zone defined in the present study). O2 = oxygen baseline; SI1 = sevoflurane increasing first third; SI2 = sevoflurane increasing mid third; SI3 = sevoflurane increasing last third; Sev2 = sevoflurane 2 minimum alveolar concentration (MAC).

Table 3.

HFO transfer entropy binary logistic mixed model classification of epileptogenic sites.

Anesthetic Stage	t-value	p-value	Odds ratio	95% CI (lower)	95% CI (upper)
Oxygen Baseline (df =615)	3.615	< 0.001	6.787E184	2.602E84	1.770E285
Sev Increasing First (df = 615)	3.738	< 0.001	6.056E42	2.014E20	1.821E65
Sev Increasing Mid (df = 615)	3.350	< 0.001	2.097E51	1.713E21	2.565E81
Sev Increasing Last (df = 615)	3.299	0.001	7.270E95	6.157E38	8.585E152
Sevoflurane 2 MAC (df = 615)	7.177	< 0.001	1.582E115	4.738E83	5.280E146

Significant classifiers (p < 0.005) are bolded. CI = confidence interval; df = degrees of freedom; HFO = high-frequency oscillation; MAC = minimum alveolar concentration.

As mentioned before, TE for each electrode consists of both efferent (influencing other sites) and afferent (being influenced by other sites) connections, which were combined for the above analysis. We also looked at individual efferent and afferent TE for delta waves and HFOs, which significantly classified epileptogenic sites during the same anesthetic stages as the combined TE described above (Supplementary Figures 4-5; Supplementary Tables 4-7). The only exception was that HFO-afferent-TE was not a significant identifier at ‘sev increasing mid’ and ‘sev increasing last’. In addition, efferent TE better classified epileptogenicity than afferent TE, but the combined TE was the most successful (Supplementary Tables 4-7; Tables 2-3).

3.6. Judging the Effect of a Potential Bias Between Patients

To assess the relationship between the proportion of epileptogenic sites and biomarker classification ability, Spearman’s rank correlation was carried out for all five biomarkers (MI, delta-SA, HFO-SA, delta-TE, and HFO-TE) and all five anesthetic conditions. After FDR p-value correction, the ancillary analysis failed to show a significant association between the relative size of the epileptogenic zone and the observed discrimination performance of a given diagnostic biomarker (all FDR-corrected p-values > 0.05).

3.7. ECoG Effective Connectivity – Anatomical White Matter Visualization

Finally, dynamic tractography combined DWI and ECoG to visualize the degree of HFO effective connectivity between site-pairs being transmitted through corresponding white matter connections, at each anesthetic stage (Figure 4). Patient 6’s electrodes with a right-frontal disribution are shown as Lausanne brain atlas regions overlayed on the standardized whole-brain tractography template (Figure 4 Top). Figure 4 (bottom) shows HFO-TE effective connectivity augmentation along the white matter connectome for patient 6 especially elevated during the ‘sev increasing first’, ‘sev increasing mid’, and ‘sevoflurane 2 MAC’ anesthetic stages.

Figure 4. White matter – high-frequency oscillation (HFO) correlation.

(Top) Diffusion weighted imaging tractography. Electrode sites with a right frontal distribution from patient 6 converted to Lausanne brain atlas regions-of-interest (ROIs) and overlayed on a standardized whole-brain tractography template derived from 1,065 patients in the Human Connectome Project. (Bottom) The white matter tractography network from patient 6’s electrode ROIs, with HFO transfer entropy (TE) values superimposed on corresponding tracts as color, for each sevoflurane stage. Light and dark blue tracts denote HFO transfer entropy values above and below the oxygen baseline TE maximum, respectively. The top row is a right-lateral-sagittal view, the middle row is a superior-axial view, and the bottom row is an anterior-coronal view. P = posterior; A = anterior; R = right; L = left.

4. Discussion

4.1. Significance and Innovation

To our knowledge this is the first study to investigate the effects of sevoflurane anesthesia on ECoG-based HFO and delta effective connectivity, for the purpose of intraoperatively localizing epileptogenic brain regions. It is also the first study to visualize the concordance between sevoflurane-activated HFO effective connectivity and underlying anatomical white matter connectivity. The main significance of this work is three-fold: [1] it provides positive preliminary data that helps warrant a larger prospective study to assess the utility of sevoflurane-based intraoperative localization of the epileptogenic zone, which could significantly reduce diagnostic burden and treatment cost by lessening the need for extraoperative iEEG; [2] it demonstrates the feasibility of using dynamic tractography to visually track sevoflurane-activated HFO signal propagation along white matter pathways; and [3] it starts to illuminate how sevoflurane differentially impacts HFO and delta effective connectivity in epileptogenic versus non-epileptogenic neural tissue.

Although the literature is relatively scarce, the present study results are in-line with previous work describing sevoflurane-induced HFO augmentation. Multiple observational studies of drug-resistant focal epilepsy patients demonstrated that sevoflurane increased the occurrence rate of spikes and HFOs in epileptogenic tissue (Orihara et al., 2020, 2022). We demonstrated a similar phenomenon: that sevoflurane enhanced HFO effective connectivity most in epileptogenic sites, and this effect was optimized at sevoflurane 2 MAC. These preliminary results continue to strengthen the hypothesis that sevoflurane-activated HFOs may be reliable intraoperative epilepsy biomarkers. While the exact role of spontaneous HFOs is currently unknown, they have been noted in both pathologic and physiologic states (Frauscher et al., 2017, 2018; Guth et al., 2021; Jacobs et al., 2008; Kerber et al., 2014; Kuroda et al., 2021; Motoi et al., 2018; Nonoda et al., 2016; Wang et al., 2013; van 't Klooster et al., 2017). Studies of epilepsy patients implanted with intracranial EEG have shown physiologic HFOs present across a berth of brain regions, which vary in occurrence rate depending on location (Frauscher et al., 2018; Kerber et al., 2014; Wang et al., 2013). Furthermore, human and rodent recordings have demonstrated the presence of physiological HFOs in medial temporal structures, and it is thought that these signals are involved with memory consolidation (Axmacher et al., 2008; Buzsáki 2015; Buzsáki et al., 1992).

The overlap between physiologic and pathologic HFOs presents a key challenge in utilizing these signatures as a proxy for epileptogenic tissue. Even in our current study, non-epileptogenic sites displayed increased HFO effective connectivity, as a function of sevoflurane concentration. Certain prospective studies, including a randomized single-blind trial, have failed to demonstrate that HFOs are superior to existing markers, such as spikes, in guiding surgical resection of epileptogenic tissue (Jacobs et al., 2018; Zweiphenning et al., 2022). This underscores the need to find ways of differentiating between endogenous and epileptiform HFOs, especially when considering surgical resection. One possible metric could be the degree of sevoflurane-induced HFO effective connectivity augmentation because the present results suggest that sites with a greater increase of HFO-TE can significantly classify epileptogenic electrodes. In addition, various intracranial EEG studies showed that HFOs associated with spikes are specific seizure onset zone markers (Roehri et al., 2018; Wang et al., 2013), and Kerber et al., (2014) similarly demonstrated that the type of EEG background was a key determinant of epileptiform HFOs. Another distinguishing factor may be frequency range; investigations in human patients suggest that fast-ripples (250-500 Hz) may be more indicative of epileptiform activity compared to ripples at 80-250 Hz (Nevalainen et al., 2020; Schönberger et al., 2020). This dichotomy could be due to different cellular mechanisms. It is thought that ripples are the result of coordinated inhibition sculpting principal neuron firing, and fast-ripples may emerge from a breakdown of GABAergic influence leading to asynchronous bursting of multiple neural populations (Buzsáki 2015; Buzsáki et al., 1992; Jefferys et al., 2012; Jiruska et al., 2017). Granted, these relationships remain murky and future studies are warranted to find more accurate ways of pinpointing epileptiform HFOs; the results are expected to enhance localization of epileptogenic brain regions and ultimately improve seizure and cognitive outcomes following surgery for drug-resistant focal epilepsy.

Furthermore, the current study underscores the importance of considering dynamic HFO propagation – effective connectivity - for delineating epileptogenic networks. Alternatively, HFO spectral amplitudes failed to significantly classify epileptogenicity. We believe this is due to the fact that HFO-SA exhibited a bimodal response to sevoflurane, and the greater degree of repression overshadowed the subset of sites displaying HFO enhancement. It is thus plausible that the electrodes showing sevoflurane-induced HFO-SA elevation may represent epileptogenic areas, because our TE results suggest that information transfer involving highly augmented HFOs significantly classified epileptogenicity.

Comparing the fidelity of iEEG and ECoG, a recent study by Weiss et al., (2021) of 16 patients with drug-resistant epilepsy suggested that the spatial distribution of HFO occurrence rates during non-rapid eye movement (REM) sleep more accurately identified epileptogenic tissue than sevoflurane-activated HFOs. However, the former approach still entails risk factors associated with chronic iEEG implants and days of extraoperative recording. Thus, one could imagine that when there is clear concordance between EEG, neuroimaging, and clinical data, it may be advantageous to consider using one-stage sevoflurane-guided localization to avoid the pitfalls of extraoperative iEEG. Even in more complicated cases with little concordance between various diagnostic modalities and/or localization near functionally important brain regions, the traditional two-stage procedure may still benefit from sevoflurane-based localization to help intraoperatively optimize iEEG placement and/or resection margins. Taken together, our results suggest that sevoflurane-activated HFO effective connectivity may be a good candidate biomarker for intraoperative classification of epileptic brain regions, and this phenomenon was optimized at sevoflurane 2 MAC. However, larger cohort studies and a normative atlas of HFO-TE at various sevoflurane levels are needed to validate the present findings before widespread implementation.

Considering slow waves, investigations in both human patients and rodent models likewise demonstrated that sevoflurane anesthesia was associated with increased delta power (Cornelissen et al., 2015, 2018; Guidera et al., 2017). Although current dogma posits that delta slowing is a feature of epilepsy (De Stefano et al., 2022), our current results suggest that, in the context of sevoflurane, one must exercise caution because we found that epileptogenic sites may have relatively less delta-SA and TE augmentation. Larger cohort studies are needed to confirm this notion, but it is possible that sevoflurane-induced delta amplitude and TE augmentation may not be a good guide for iEEG electrode placement. Previous work suggests that delta waves and the down state of slow waves are associated with neural inhibition (Harmony 2013; Tartaglia and Brunel 2017). Since HFOs represent active neural tissue, it could make sense that sites with relatively less delta-inhibition (i.e., the epileptogenic sites in this instance) will exhibit more HFOs. In addition, both previous and the current study suggest that sevoflurane-augmented phase-amplitude coupling between HFOs and delta waves can intraoperatively identify epileptogenic sites (Wada et al., 2022). Chamadia et al., (2019) showed that sevoflurane even induces delta phase-amplitude coupling in healthy human subjects. Likewise, extraoperative iEEG recordings during slow wave sleep showed enhanced HFO-delta phase-amplitude coupling in epileptogenic areas (Kuroda et al., 2021). Our present results further support the role of intraoperative MI as a possible biomarker to compliment HFO-based localization of the epileptogenic zone. While these results are promising, a larger cohort study is needed to verify the initial findings.

Given that TE is designed to evaluate the degree to which one site can influence the future state of another (i.e. unidirectional information transfer), our current results may suggest that epileptogenic sites exhibit a stronger ability to project and respond to HFO electrogenic influence. One can also interpret effective connectivity as a summary measure of the spatiotemporal dynamics of HFO signals; in that case, our results also imply that sevoflurane-driven HFO signals rapidly propagate throughout epileptic networks, and we were able to successfully visualize white matter pathways carrying such activity via dynamic tractography. Supporting this notion, a human resting state iEEG study by Arnulfo et al., (2020) showed that inter-regional HFO synchronization was strongest for contacts closest to the cortical white matter lamina, while the opposite was true for low-frequency synchrony. Other work has similarly demonstrated that ictal HFO propagation latency in epileptic spasms positively correlates with the length of white matter tracts and fractional anisotropy values (O’Hara et al., 2022). Taken together, the ability to visualize white matter pathways supporting sevoflurane-induced HFO signal propagation could be a useful tool for aiding surgical disconnection of epileptic brain networks.

4.2. Possible Mechanisms

Due to the small sample size of this current cohort, we cannot make definitive conclusions, but our working hypothesis as to the mechanism underlying sevoflurane-HFO epileptogenic localization is outlined in Figure 5. In short, this preliminary data suggests that increasing sevoflurane first causes diffuse HFO activation, followed by enhanced HFO-delta phase-amplitude coupling and eventual refinement of HFO propagation specifically in epileptic networks. Although the exact mechanism underlying sevoflurane-induced activation of HFOs and delta waves remains unknown, it is generally accepted that sevoflurane potentiates inhibitory GABAergic neurotransmission (Alkire et al., 2008; Mapelli et al., 2021; Ogawa et al., 2011; Xu et al., 2020). Considering such gamma-aminobutyric acid (GABA) activity, HFOs may result from synchronous inhibitory inputs onto principal cells that cause rapid firing of action potentials (Jiruska et al., 2017; Karlócai et al., 2014). Epileptogenic neural tissue is reported to express increased levels of the NKCC1 ion channel, which is responsible for accumulating intracellular chloride (Cl⁻; Huberfeld et al., 2015; Kahle and Staley 2008; Liu et al., 2020); furthermore, bumetanide – a NKCC1 blocker – has been shown to reduce NKCC1 protein expression, along with the frequency of epileptic seizures in human patients (Gharaylou et al., 2019; Khale and Staley 2008). In a microenvironment of hyper-elevated GABA, NKCC1 channels extrude excess intracellular Cl⁻ ions leading to paradoxical depolarization and interictal epileptiform events (Huberfeld et al., 2015; Kahle and Staley 2008; Liu et al., 2020). In the context of sevoflurane, it is possible that the elevated level of Cl⁻ ions from enhanced GABAergic transmission may cause epileptogenic tissue with perturbed NKCC1 expression to undergo such paradoxical depolarization, leading to HFO generation.

Figure 5. A working hypothesis: sevoflurane activates high-frequency oscillations (HFOs) preferentially in epileptogenic brain regions.

During oxygen baseline there are low-level background HFOs and delta waves resonating in the brain. Increasing the concentration of sevoflurane anesthesia activates widespread spike-and-wave epileptiform activity, along with the HFO and delta wave components. During the ‘sev increasing last third’ epoch, the sevoflurane-induced HFOs and delta waves become hyper-synchronized in epileptogenic tissue by an unknown process that may involve resistance to delta effective connectivity augmentation. This phase-amplitude coupling could create windows of disinhibition at sevoflurane 2 minimum alveolar concentration (MAC), when HFO effective connectivity is most amplified in epileptogenic neural tissue, and these signals may partially utilize white matter tracts to propagate throughout the brain.

Alternatively, in vitro studies demonstrate that while sevoflurane does generally potentiate GABAergic activity, high enough concentrations can impede inhibitory signaling and actually increase excitability of projection neurons (Eckle et al., 2013; Hapfelmeier et al., 2001; Mapelli et al., 2021; Suzuki and Smith 1988). A recent study in mice likewise demonstrated that early-life exposure to sevoflurane enhances excitatory output of hippocampal CA1 neurons by reducing GABAergic transmission (Lin et al., 2021). This may help explain why, in the current study, HFO effective connectivity was most predictive of epileptogenesis at 2 MAC. It is possible that the higher concentration of sevoflurane paradoxically blunted GABAergic signaling, which then disinhibited HFO propagation. For some reason, epileptic tissue seems to be more susceptible to this phenomenon, which was reflected by the relatively smaller delta-TE and larger HFO-TE that characterized epileptogenic sites near sevoflurane 2 MAC (Figures 2B, 3B). In fact, it is well documented that application of the GABA antagonist bicuculline leads to the generation of epileptiform discharges, and penicillin is known to induce spikes through a similar mechanism (Suzuki and Smith 1988; Schwartzkroin and Prince 1978). However, future studies are needed to elucidate the exact mechanism driving sevoflurane-induced activation of HFOs in epileptogenic brain regions.

4.3. Methodological Constraints and Future Directions

Although we tried to control all possible variables, a few issues with our experimental design should be addressed. First, one could argue that the epileptogenic zone (i.e., in this case the resected electrodes in patients who achieved ILAE class 1 seizure freedom) is not a practical intraoperative epilepsy biomarker because it is only confirmed after surgery. One possible way to avoid this issue in a prospective study is to rather consider the seizure onset zone, which is determined by board certified clinicians prior to surgery. However, the epileptogenic zone is theoretically defined as “the minimal amount of cortex that must be resected to produce seizure freedom” (Lüders et al., 2006). In many cases, that is suggested to include the seizure onset zone plus other areas such as lesions and spiking regions; meaning removal of only the seizure onset zone may not be sufficient for postoperative seizure control. Hence, we opted to sufficiently localize the epileptogenic zone, to better match the surgical goal. In addition, the epileptogenic zone is typically large in frontal lobe epilepsy (Salanova et al., 1994; Smith et al., 1997), so one could hypothesize that this may bias the ability of a given biomarker to classify epileptogenicity. However, based on the ancillary Spearman’s rank correlation, we failed to find objective evidence that the observed discrimination performance of a diagnostic biomarker depended on the relative size of the epileptogenic zone. Even still, the majority of patients in this study were diagnosed with frontal lobe epilepsy, so the present results may disproportionately reflect activity inherent to this region. Future studies with more patients, greater diagnostic variability, and a denser distribution of electrode coverage on other brain lobes are necessary to address this potential concern.

Another possible criticism is that we did not use the patients’ own DWI for tractography. Instead, we opted to use data from HCP 1065 (Yeh et al., 2018), which is a standardized white matter connectome derived from the average of 1,065 patient brains. Although this method may sacrifice some correlation power, we felt it would better generalize the results and ensure that the tractography data was well verified. One other limitation with our DWI – TE correlation is that it exclusively visualizes monosynaptic propagation via white matter pathways. Our group and others have used single-pulse electrical stimulation in patients implanted with intracranial electrodes to infer the existence of both monosynaptic and multisynaptic circuits, depending on propagation time (Silverstein et al., 2020; Veit et al., 2021), so it can be assumed that a portion of the TE-defined connections are multisynaptic. Our dynamic tractography also does not visualize intracortical signal spread through grey matter (Kharas et al., 2022). Future models would benefit by considering the multitude of different propagation pathways. Finally, one could argue that this study may not entail enough participants. It is certainly true that a larger cohort is needed to substantiate these findings, but even with eight patients, we were still able to determine statistical significance using a mixed model approach.

In the future, we plan to definitively determine whether these intraoperative ECoG-based biomarkers can predict epileptogenicity and postoperative seizure outcome. Power analysis suggests that 66 patients are needed to detect a medium effect size of Cohen’s d (i.e., 0.35) with an alpha of 0.05 and power of 0.8. Despite these few caveats, the current study laid important groundwork toward elucidating the effects of sevoflurane on HFO and delta effective connectivity, helped visualize HFOs propagating along white matter pathways, and further demonstrated the clinical validity of using sevoflurane-activated HFOs for intraoperative localization of the epileptogenic zone. We believe these results justify future, large cohort studies to solidify our findings, and they prompt the need for establishing normative atlases of intraoperative HFO-TE at different concentrations of sevoflurane, to better understand its clinical relevance.

Highlights.

• Sevoflurane diffusely augmented intraoperative high-frequency oscillation (HFO) effective connectivity, yet most in the epileptogenic zone.

• Intraoperative HFO effective connectivity best classified the epileptogenic sites at sevoflurane 2 MAC.

• HFO effective connectivity can be visualized on major white matter pathways, as a function of sevoflurane concentration.

Abbreviations

ASM

Anti-seizure medication

CI

Confidence interval

Cl⁻

Chloride

df

Degrees of freedom

DWI

Diffusion-weighted imaging

EC

Effective connectivity

ECoG

Electrocorticography

GABA

gamma-aminobutyric acid

GFA

Generalized fractional anisotropy

iEEG

Intracranial electroencephalography

ILAE

International League Against Epilepsy

HCP

Human Connectome Project

HFO

High-frequency oscillation

MAC

Minimum alveolar concentration

MI

Modulation index

MNI

Montreal Neurological Institute

MRI

Magnetic resonance image

OR

Odds ratio

PAC

Phase-amplitude coupling

REM

Rapid eye movement

ROI

Region of interest

SA

Spectral amplitude

TE

Transfer entropy

Data Availability Statement

All data from this study are available from the corresponding author upon reasonable request.