Tracking multi-site seizure propagation using ictal high gamma activity

Abstract

Introduction:

Spatial patterns of long-range seizure propagation in epileptic networks have not been well characterized. Here, we use ictal high gamma activity as a proxy of intense neuronal population firing to map the spatial evolution of seizure recruitment.

Methods:

Ictal high gamma activity (80–150 Hz) was analyzed in 13 patients with 72 seizures recorded by stereotactic depth electrodes, using previously validated methods. Distinct spatial clusters of channels with the ictal high gamma signature were identified, and seizure hubs defined as stereotypically recruited non-overlapping clusters. Clusters were correlated with asynchronous seizure terminations to provide supportive evidence for independent seizure activity at these sites. The spatial overlap between seizure hubs and interictal ripples was compared.

Results:

Ictal high gamma activity was detected in 71% of seizures and 10% of implanted contacts, enabling tracking of contiguous and noncontiguous seizure recruitment. Multiple seizure hubs were identified in 54% of cases, including 43% of patients thought preoperatively to have unifocal epilepsy. Noncontiguous recruitment was associated with asynchronous seizure termination (Odds ratio=19.7, p=0.029). Interictal ripples demonstrated greater spatial overlap with ictal high gamma activity in cases with single seizure hubs compared to those with multiple hubs (100% vs 66% per patient, p=0.03).

Conclusions:

Ictal high gamma activity may serve as a useful adjunctive biomarker to distinguish contiguous seizure spread from propagation to remote seizure sites. High gamma sites were found to cluster in stereotyped seizure hubs rather than being broadly distributed. Multiple hubs were common even in cases that were considered unifocal.

Predictors of multifocal epileptogenic zones are needed to help address the limitations of surgical resection and laser ablation in controlling focal drug-resistant epilepsy.^1–3 Multiple seizure onset zones (SOZ) are a strong indicator of poor outcome,⁴ however the relevance of sites that are not active at onset but emerge later in the seizure is uncertain. Few studies have addressed this question, with inconclusive findings.^5–8

Previously we proposed that seizures are spatially organized into dual zones characterized by distinct neuronal firing patterns: an ictal core with synchronized paroxysmal depolarizations, and a surrounding penumbra, in which excitatory barrages from the core elicit a strong inhibitory response.⁹ These cannot be reliably differentiated using standard EEG interpretation,^9,10 but can be distinguished by ictal high gamma activity (HGA) at 80–150 Hz, demonstrating sustained activity phase-locked with the EEG, and validated using human microelectrode recordings.^11,12 Ictal HGA in clinical electrodes therefore may track the progression of locally synchronized neuronal population firing, taking advantage of the strong high gamma signature produced by paroxysmal depolarizing shifts.^13–17

We hypothesize that ictal HGA organizes into noncontiguous spatial clusters, and that these may represent independent seizure hubs, even in cases with a single seizure type and onset zone. Using ictal high gamma clustering, activation sequences, and seizure termination patterns, we characterize the dynamic spatiotemporal properties of seizure recruitment, and define patient-specific seizure hubs in consecutive patients with focal epilepsy implanted with depth electrodes. We also evaluate whether the spatiotemporal organization of ictal HGA is reproduced by interictal ripples, which have been associated with the epileptogenic zone but may be generated by different mechanisms.^18–22 This descriptive study provides a framework for implementing analysis of ictal HGA into the surgical evaluation of poorly localized and multifocal epilepsy.

Methods

Patient Description

Thirteen consecutive patients at Columbia University Medical Center (CUMC) undergoing surgical evaluation between 2014 and 2016 were included. All clinical seizures were included for each patient with two exceptions. In one patient who required electrode revision, only post-revision seizures were included. Another patient had numerous focal aware seizures, of which nine were randomly selected. All protocols for Human Subjects Research were approved by the CUMC Institutional Review Board with informed consent.

EEG Analysis

All patients were implanted with depth electrodes containing 8–16 contacts with 3.5mm center-to-center spacing (PMT Corporation, Chanhassan, MN). Intracranial EEG was recorded using the Xltek EEG system (Natus Medical Incorporated, Pleasanton, CA), sampled at 500 Hz, with an epidural mini-depth electrode reference. SOZs were determined through visual EEG inspection (typically 1–70 Hz) by the treating epileptologist as the channel(s) with the first sustained ictal electrographic change from interictal baseline, and reviewed by at least two authors (ST, LMB, CAS) with disagreements resolved by consensus. Epilepsy syndromes were considered unifocal if a single SOZ was present across all seizures. As a proxy for the potential of a given site to maintain seizure activity independently,^5,7,23 we assessed correlation with asynchronous seizure termination. Asynchronous termination was identified by a difference in termination of the ictal rhythm in multiple channels of at least 1s, or a difference of at least 500ms if the rhythm across channels had different morphology and frequency at seizure termination. Channels were then accordingly divided into non-overlapping termination groups.

Ictal High Gamma Analysis

Ictal HGA was visually identified from filtered EEG data (FIR, order 90, bandpass 80–150 Hz) in Persyst Insight (Persyst Development Corp, Solana Beach, CA), and validated by comparison to the MATLAB FIR1 filter with the same parameters (Mathworks, Natick, MA).^11,24 Ictal HGA was defined as previously described, with delayed activation after seizure onset of repetitive high gamma bursts for at least 10s, standing out from the background,²⁵ and visually correlated with the low-frequency (1–70 Hz) EEG rhythm (Fig. 1B). The requirement for high gamma bursts to correspond to ictal EEG discharges was intended to mirror the quantitative phase-locked high gamma method developed to track the path of seizure recruitment,^11,24 and allow for simple translation to clinical use. Simultaneous HGA activation was defined by onset within 100ms between channels. HGA was visually assessed to exclude artifact, including non-physiologic high frequency activity in a single channel and bursts occurring simultaneously across all channels. Additionally, channels were excluded if they demonstrated significant environmental and/or instrumental artifacts. HGA was identified blinded to electrode location. In seizures with secondary generalization, ictal HGA was analyzed up to the time of clinical transition to bilateral tonic-clonic activity with associated generalized ictal EEG rhythm. For all other seizure types, analysis was performed from electrographic onset to termination.

Figure 1. Local propagation within a seizure hub.

(A) Schematic model of multi-site seizure propagation with recruited territory distinguished from non-recruited territory by the presence of high frequency (HF) activity phase-locked to the low frequency (LF) ictal EEG rhythm. (B-E) Representative example of local contiguous recruitment. Seizure onset (B) in standard EEG (top trace, black arrowhead) precedes the activation of phase-locked ictal HGA (bottom trace, green arrowhead). Sequential activation of ictal HGA occurs from deep to superficial contacts along the depth array (C). Standard EEG reveals early activity across multiple contacts (D), compared to the ordered recruitment demonstrated by ictal HGA activation (E). Artifact (gray bars) is characterized by identical high frequency events in all channels.

Ripple Analysis

Intracranial EEG segments with durations of 10 minutes of NREM sleep and free of extracerebral artifact were visually identified. Segments were chosen at least 24 hours after implantation to minimize anesthetic effect and four hours from the most recent seizure if possible. As the majority (>99%) of ripples (80–250 Hz) were detected in association with epileptiform discharges in the same or any other channel, we aimed to minimize bias due to variable representation of multifocal discharge populations. Multifocal discharge populations were defined as epileptiform discharges with non-overlapping fields. Each discharge population was visually identified, and in three cases, an extended segment beyond 10 minutes was selected to include 20 representative discharges from each population.

Automated ripple detection was performed using the short time energy (RMS) method in RIPPLELAB.²⁶ A minimum cutoff of eight oscillations was used to decrease false positive identification.²⁷ Detections were reviewed using simultaneous views of the unfiltered event, filtered event, and spectral density plots for the given channel. The full clinical montage was visually reviewed for the presence of a co-occurring epileptiform discharge in any other channel. After visual inspection of detections and exclusion of false-positives, ripple rates were determined for each 10-minute segment, as well as for the extended segments.

Cluster Analysis

Electrode localization was performed by coregistration of the pre-implant T1 volumetric MRI with the post-implant volumetric CT using FSL (FMRIB Software Library, FMRIB, Oxford, UK). For each seizure, ictal HGA channels were grouped into non-overlapping spatial clusters using the following separation criteria: 1) at least 2cm between contacts in different clusters; and 2) activation in different sub-lobar anatomic structures according to the Harvard-Oxford cortical/subcortical structural atlas.²⁸ For each cluster, an index contact was identified as the channel with earliest HGA within that cluster or with the highest amplitude if there were multiple channels with simultaneous onset. Propagation between clusters was defined by sequentially activated index contacts with inter-cluster distances calculated by the Euclidean distance between index contacts. For each patient, overlapping ictal HGA clusters were combined into a single seizure hub. Distinct seizure hubs were then defined using the same separation criteria as for clusters.

Statistical Analysis

Comparison between continuous data were evaluated using the Mann-Whitney U-test. Associations between categorical data were evaluated using Fisher’s exact test. Correlation was calculating using the Spearman correlation coefficient. All statistical tests were two-sided with a significance threshold of 0.05. All analyses were performed using R 3.5.3 (R Foundation, Vienna, Austria).

Data Availability

Data will be shared with other investigators upon request and subject to limitations imposed by the CUMC Institutional Review Board, Information Technology policies, and federal medical data privacy laws.

Results

Seventy-two seizures from 13 patients (mean 5.5 seizures per patient, range 2–20) were analyzed. These included 45 focal impaired awareness seizures (FIAS), 15 focal aware seizures (FAS), and 12 focal to bilateral tonic-clonic (FBTC) seizures. Presurgical characteristics are shown in Table 1. Ictal HGA was identified in 12/13 (92%) patients and 51/72 (71%) seizures. Among seizures without ictal HGA (16 FIAS, two FAS, three FBTC), there were 15 brief seizures with durations less than 45s (13 FIAS, two FAS), one FIAS with a limited field of ictal discharges, and five limited by excessive artifact (two FIAS, three FBTC).

Table 1.

Patient demographics and presurgical characteristics

ID	SEX	AGE	SEIZURE TYPE	NUMBER OF SEIZURES	MRI	PET	PRE-OP SUSPECTED MULTIFOCAL
1	F	53	FIA	3	Nonlesional	R mesial temporal	No
2	F	50	FIA; FBTC	3	L MTS; multiple bilateral cavernomas	L mesial and temporal pole	Yes
3	F	30	FA	5	R choroidal fissure cyst	R mesial and temporal pole	No
4	M	25	FIA	4	L frontal cortical encephalomalacia	L frontal cortical and basal ganglia/thalamus	No
5	M	23	FIA; FBTC	4	L choroidal fissure cyst	Bilateral anterior temporal	Yes
6	M	30	FIA	4	Nonlesional	Bilateral temporal	Yes
7	F	52	FIA	2	L temporoparietal signal and possible cortical thickening; R aneurysm clipping	L mesial/anterior temporal	Yes
8	F	29	FBTC	2	Punctate L medial frontal and R posterior temporal susceptibility	R>L temporal	No
9	F	31	FIA	4	L temporo-parieto-occipital band heterotopia; possible L hippocampal atrophy; dandy-walker variant	L mesial and temporal pole	Yes
10	F	29	FA	9	L MTS	L anterior hippocampus	No
11	F	55	FIA	20	Possible R frontal subependymal signal abnormality	R anteromedial temporal	No
12	M	25	FA; FBTC	3	Bilateral posterior periventricular heterotopias	R temporal	Yes
13	M	19	FIA; FBTC	9	Postsurgical R temporal encephalomalacia	R hemispheric	No

FA, focal aware; FIA, focal impaired awareness; FBTC, focal to bilateral tonic-clonic; MTS, mesial temporal sclerosis

Ictal High Gamma Spatial Clustering

Ictal HGA was detected in an average of 10% (range 0–20%) of the total implanted contacts per patient, and were organized into distinct clusters. Ictal HGA was recorded exclusively from gray matter structures, with the exception of one case in which two contacts located extra-axially in CSF adjacent to epileptogenic cortex consistently recorded HGA. A total of 88 ictal HGA clusters were identified from the 51 seizures, with a median of 4.5 (range 1–21) contacts per cluster. Ictal HGA was confined to a single cluster in 25 seizures from 10 patients, of which 21 were mesial temporal and four were neocortical (Table 2). Among the five patients with bilateral implantation, ictal HGA clusters were found bilaterally in five seizures from two patients, with all involving both mesial temporal and lateral temporal sites. Ictal HGA clusters overlapped with the clinically-defined SOZ in 36/51 (71%) seizures; seven patients had all SOZs within clusters, two patients had all SOZs outside clusters, and three patients had SOZs both within and outside clusters.

Table 2.

Ictal high gamma localization and surgical characteristics

ID	# ARRAYS (CONTACTS)	SEEG SOZ	ICTAL HIGH GAMMA CLUSTER LOCALIZATION	SEIZURE HUBS	SURGERY	ENGEL OUTCOME	FOLLOW-UP (MO)
1	6 (54)	R MT	R amygdala, hippocampus	Single	Inferior hippocampus + parahippocampal gyrus laser ablation	II-A	49
2	8 (100)	L MT	L hippocampus	Single	Hippocampal head/body + amygdala laser ablation	I-D	29
3	8 (124)	R MT	R hippocampus	Single	Hippocampal head/body laser ablation	I-B	18
4	8 (86)	L frontal perilesional (clinical); L MT (subclinical)	L superior frontal gyrus; L inferior parietal lobule; L posterior cingulate; L hippocampus	Multiple	NA	NA	_
5	16 (190)	Bilateral MT	L hippocampus; R hippocampus; R inferior temporal gyrus	Multiple	NA	NA	_
6	14 (190)	Bilateral MT; L orbitofrontal	L entorhinal cortex; L parahippocampal gyrus; R middle temporal gyrus	Multiple	NA	NA	_
7	11 (130)	L MT	L hippocampus	Single	Hippocampal head/body + amygdala laser ablation	II-B	34
8	11 (110)	L frontal neocortical	NA	NA	Medial frontal cortical lesionectomy	I-A	18
9	12 (122)	L temporoparietal neocortical	L parietal heterotopia; L temporal heterotopia	Multiple	Temporoparietal corticectomy	III-A	25
10	10 (112)	L MT	L hippocampus; L parahippocampal gyrus	Multiple	Hippocampal head/body + amygdala laser ablation	I-D	24
11	9 (104)	R temporal neocortical	R hippocampus	Single	Extended temporal lobectomy	I-B	6*
12	12 (154)	R posterior perilesional; R MT	R hippocampus; R parahippocampal gyrus; R middle temporal gyrus; R superior temporal gyrus	Multiple	Hippocampal head/body + medial occipital laser ablation	IV-A	22
13	11 (146)	R temporal neocortical	R superior temporal gyrus; R inferior parietal lobule; R fusiform gyrus	Multiple	Anterior temporal lobectomy + superior temporal gyrus and posterior middle temporal gyrus resection	I-D	21

^*Long-term follow-up data not available

SOZ, seizure onset zone; MT, mesial temporal

Ictal High Gamma Activation Sequences

To evaluate seizure propagation and the implied expansion of recruited territory, we assessed activation sequences of ictal HGA. Sequential activation of ictal HGA was observed within single clusters (Fig. 1), as well as between distinct clusters (Fig. 2). In six seizures from three patients, there was simultaneous activation of clusters, indicating nearly instantaneous recruitment of sites separated by 2.0–4.7 cm. Excluding these simultaneously activated clusters, the mean activation delays between the first-to-second and second-to-third clusters were similar (37.2s [6–182s] vs 35.0s [7–68s], U=48, p=0.59), with evidence of both remote cluster activation at short delays, and adjacent cluster activation at long delays (Fig. 3A).

Figure 2. Remote propagation between seizure hubs.

(A-E) Representative example of multi-site recruitment. In a bilateral implant (A), ictal HGA occurs in three distinct clusters (green, blue, maroon), representing a minority of the contacts with standard ictal EEG rhythms (gray). Standard EEG (1–70 Hz) demonstrates asynchronous ictal rhythms leading to asynchronous terminations in compressed (B) and expanded (C) views. Ictal HGA reveals delayed activation (D) with an ordered pattern of recruitment across clusters. Non-sustained HGA is seen in Channel #2 prior to development of sustained activity, and distinct from artifact (gray bar). Localization of ictal HGA reveals a recruited cluster in entorhinal cortex followed by parahippocampal gyrus and then contralateral temporal neocortex.

Figure 3. Ictal high gamma clustering and seizure termination.

(A) Inter-cluster activation delay versus distance. Euclidean distances are measured between the first-to-second activated clusters (blue), and in later clusters from the most proximally located earlier cluster (yellow). Dashed lines indicate estimated propagation speed, including nearly simultaneous activation between distant clusters (**). (B) Ictal HGA clusters plotted against asynchronous seizure termination groups, stratified by seizure type, and showing lack of asynchronous terminations in FAS. (C) Correlation between total number of seizure hubs and termination groups per patient. Gray shaded area represents the 95% confidence interval.

Ictal High Gamma Clusters and Asynchronous Termination

To evaluate the potential for secondarily activated ictal HGA clusters to maintain independent seizure activity, we assessed seizure termination patterns. Asynchronous termination was observed in 22 seizures, including 20/51 (39%) with ictal HGA and 2/21 (9.5%) without ictal HGA. The final termination group overlapped with the SOZ in 51/72 (71%) of seizures, whereas final termination was isolated to spread sites in 21/72 (29%) of seizures (which included the independent SOZs of other seizures in four instances from two patients with multiple SOZs). Asynchronous termination was more commonly observed in seizures with multiple clusters compared to seizures with single or no clusters (n=13, Odds Ratio=19.7, 95% CI 1.0–1617, p=0.029). The correlation between number of ictal HGA clusters per seizure and number of channel groups terminating asynchronously varied by seizure type; a positive correlation was observed for FIAS and no correlation observed for FAS due to absence of asynchronous termination in this group (Fig. 3B). Pairs of contacts within the same cluster were more likely to share the same termination group than pairs from different clusters (mean proportion per patient 0.91 vs 0.42, U=74, p=0.0046). In 13 seizures from five patients, channels from clusters activated later in the seizure consistently demonstrated delayed termination compared to clusters activated earlier; in contrast there was only one seizure in which channels from clusters activated later demonstrated earlier termination. In addition, there were 13 termination groups without corresponding ictal HGA.

Patient-level Seizure Hubs

The contacts comprising ictal HGA clusters showed consistent seizure-to-seizure activation, indicative of stereotyped seizure hubs per patient. There were 26 seizure hubs of unique non-overlapping ictal HGA clusters. The mean and median proportion of seizures per patient in which ictal HGA contacts were activated when the corresponding seizure hub was recruited were 0.84 and 0.83, respectively. The index contact, representing the channel activated first within a cluster, remained invariant in 25/26 (96%) of seizure hubs.

There were a median of 2 (range 0–4) seizure hubs per patient. The number of seizure hubs per patient correlated with the total unique asynchronous termination groups (r_s=0.56, p=0.047, Fig. 3C). Preoperatively, 6/13 (46%) cases were suspected to have multifocal onsets (including two cases of bitemporal epilepsy), and multiple seizure hubs were found in 7/13 (54%) cases. However, there was limited overlap between these groups, with multiple seizure hubs in 3/7 (43%) cases suspected preoperatively to have unifocal epilepsy and in 4/6 (67%) cases suspected to have multifocal involvement, including both bitemporal cases (Table 2).

Ripple Overlap with Ictal High Gamma

Interictal ripples were identified in 9/13 (69%) cases. The mean rate per patient was 30 (1–69) per 10 minutes, with a mean individual channel rate of 4.2 (1–31) per 10 minutes. There were 70 unique channels (4.3% of total implanted) demonstrating ripples, with 5 channels identified solely from extended segments used to ensure adequate sampling of all discharge populations. In patients with both ictal HGA and interictal ripples, ripples were observed within 8/19 (42%) of seizure hubs. Among the 20 channels with ripples located outside of seizure hubs, seven (35%) were from immediately adjacent contacts. The proportion of ripple channels within seizure hubs was dependent on the number of hubs per patient (Fig. 4). In cases with single seizure hubs, 100% of ripples were found within the hub, compared to cases with multiple seizure hubs, in which an average 66% of ripple channels per patient were found within hubs (n=9, U=0, p=0.03).

Figure 4. Association between interictal ripples and seizure hubs.

Overlap of interictal ripples (80–250 Hz) in cases with at least one seizure hub defined by ictal HGA. All interictal ripples overlap with ictal HGA in cases with a single seizure hub, whereas in cases with multiple seizure hubs, ripples are found in 8/19 (42%) of seizure hubs, and 29% of ripple channels are located outside of all hubs.

Surgical Outcomes

Of the 13 patients, 10 underwent resection or laser ablation, of which 8/10 (80%) had Engel Class I/II outcomes (Table 2). All patients with single seizure hubs had Engel Class I/II outcomes. The two patients with Engel Class III/IV outcomes had multiple seizure hubs outside the SOZ. Three of the five remaining patients with multiple hubs were not considered candidates for resection or laser ablation. Good outcomes were seen in 5/5 (100%) patients in whom all seizure hubs were entirely removed, compared to 2/4 (50%) with at least one hub not entirely removed.

Discussion

In this series of patients with implanted depth electrodes, we investigated the spatial clustering and temporal activation sequence of ictal HGA, and the association with asynchronous seizure termination. We found that even in patients with unifocal epilepsy, ictal HGA revealed recruitment of multiple noncontiguous seizure hubs during seizure propagation, which were associated with asynchronous seizure terminations.

Ictal HGA as described here, with sustained bursting correlated to the low-frequency EEG rhythm has been previously validated as a biomarker of synchronized neuronal population firing using human microelectrode data,¹¹ and in a subsequent surgical outcome study.²⁴ This definition differs from other reports of ictal high frequency activity,^25,29–34 which include events such as non-sustained gamma bursts at the ictal-interictal transition. We found frequent instances (as shown in Fig. 1–2) of a characteristic ictal EEG rhythm without accompanying HGA. This behavior is predicted by our prior work distinguishing the recruited ictal core from the surrounding penumbra with disorganized multiunit firing and preserved inhibition,⁹ which may show high-amplitude EEG signals but is less likely to generate HGA.^11,13

Over half of the cases here were found to have multiple distinct seizure hubs. This is likely to be true of many focal epilepsies,⁴ and may be associated with poor post-operative seizure control. The method described here may facilitate identification of independent seizure foci more rapidly than other currently available clinical methods.³⁵ We propose that ictal HGA be used as an adjunct to clinical intracranial EEG interpretation, by tracking the path of recruited territory as the seizure advances from a possibly unsampled origin through sampled brain sites. The association between ictal HGA clustering and termination asynchrony, in which ictal HGA channels in different clusters and separated by greater distance were more likely to terminate asynchronously, may provide a method of detecting undersampling. For example, a group of channels terminating synchronously without HGA, as seen in a minority of cases here, may imply synaptic spread from an unsampled seizure hub. The absence of ictal HGA entirely or appearing only late in the seizure may also suggest recruitment in an unsampled region. Indeed, given the inherent sampling limitations of intracranial EEG, it is unlikely that all sites generating ictal HGA were identified in this study. On the contrary, the 2cm separation threshold between clusters was chosen empirically and may have falsely separated adjacent clusters, when in fact there was contiguous spread through unsampled tissue between clusters. The criteria requiring clusters to be in different anatomic structures was used to minimize this possibility, although we chose to bias towards separation of clusters so as to avoid obscuring multiple propagation mechanisms.

The local contiguous seizure propagation in this study corresponds to the slow invasion of the ictal wavefront at less than 1 mm/s that has been previously demonstrated in animal models,³⁶ as well as in human microelectrode recordings.⁹ However, there were instances of rapid recruitment across large distances, implying much faster propagation speeds similar to those of epileptic discharges emanating from the slow-moving ictal wavefront (0.1–1 m/s).^37,38 While there was no direct evidence of seizure activity spreading through functional connections of large-scale brain networks,^39,40 this type of distant recruitment indicates that a long-range network mechanism is involved. This situation is distinct from visual EEG observations of simultaneous fields extending to remote sites, which can occur via synaptic distribution without necessarily reflecting seizure recruitment at each site.³⁷ Instead, rapid recruitment at noncontiguous sites must proceed by a different mechanism. One possibility is impaired local inhibition at the secondary site permitting near-instantaneous “jumps” as demonstrated in an animal model using focal bicuculline microinjections at sites remote to 4-aminopyridine seizure foci.⁴¹ This suggests a potential mechanism involving rapid glutamatergic current distribution combined with impaired feedforward inhibition, however other mechanisms yet to be described may lead to the appearance of distal recruitment.

This analysis included multiple seizures per patient to determine ictal high gamma stereotypy, which was demonstrated by consistent localization, activation sequence, and channel composition of ictal HGA. The stereotyped appearance of ictal HFA will allow further investigation of how these seizure hubs interact with each other and to distinguish which sites contribute to the epileptogenic zone. While resection of early appearing ictal HGA was a stronger predictor of postoperative seizure freedom than later-onset HGA in prior work,²⁴ the prognostic significance of contiguous vs non-contiguous recruitment is uncertain. A larger study is required to determine whether the presence of delayed seizure hubs are negative outcome predictors, and whether intervention at these sites may improve outcomes.

Compared to interictal ripples, which have also generated significant interest as a biomarker of epileptogenicity,^{18–22,42,43} ictal HGA provides necessary temporal information to identify areas recruited earlier versus later in the seizure.^24,25,31,32 Ictal and interictal high frequency oscillations have been previously detected in overlapping channels at early spread sites,⁴⁴ but the correlation with delayed spread sites remains poorly characterized. While we found concordant localization between ictal HGA and ripples in cases with a single seizure hub, ripples in cases with multiple hubs failed to identify the majority of these, while also appearing outside of hubs. These findings align with the widespread detection of interictal high frequency oscillations in prolonged recordings.⁴⁵ Nevertheless, we did find evidence for spatial clustering of ripples with ictal HGA, and therefore future research may be aimed at evaluating the predictive value of both biomarkers in combination on surgical outcomes. Ripple rates in this study were lower than in some prior studies,^19,43 and within the ranges reported by others,^21,46 likely reflecting differences in the individual parameters used. Similarly, the close association seen in this study between ripples and interictal epileptiform discharges has been previously reported,^47,48 but is higher than in other studies,^18,49 which may be due to methodological differences. A conservative approach to ripple measurement was important for this study to avoid false positives at the expense of reduced sensitivity. However, this analysis was confined to the ripple band, and therefore limited by inability to assess fast ripples, which have a higher reported specificity for epileptogenicity.^21,43

In summary, ictal HGA may track seizure recruitment within and across multiple distinct seizure hubs. We speculate that focal seizures spread broadly not only by synaptic distribution of excitatory currents but also by seeding new hubs that amplify seizure distribution and create the appearance of large-scale coordinated network behavior.