Statistical mapping of ictal high-frequency oscillations in epileptic spasms

Authors


Address correspondence to Eishi Asano, M.D., Ph.D., M.S. (CRDSA), Division of Pediatric Neurology, Children’s Hospital of Michigan, Wayne State University, 3901 Beaubien St., Detroit, MI 48201, U.S.A. E-mail: eishi@pet.wayne.edu

Summary

Purpose:  We assessed 636 epileptic spasms seen in 11 children (median 44 spasms per child) and determined the spatial and temporal characteristics of ictal high-frequency oscillations (HFOs) in relation to the onset of spasms.

Methods:  Electrocorticography (ECoG) signals were sampled from 104–148 cortical sites per child, and the dynamic changes of ictal HFOs were animated on each individual’s three-dimensional (3D) magnetic resonance (MR) image surface.

Key Findings:  Visual assessment of ictal ECoG recordings revealed that each spasm event was characterized by augmentation of HFOs. Time-frequency analysis demonstrated that ictal augmentation of HFOs at 80–200 Hz was most prominent and generally preceded those at 210–300 Hz and at 70 Hz and slower. Recruitment of HFOs in the rolandic cortex preceded the clinical onset objectively visualized as electromyographic deflection. The presence or absence of ictal motor symptoms was related more to the amplitude of HFOs in the Rolandic cortex than in the seizure-onset zone. In a substantial proportion of epileptic spasms, seizure termination began at the seizure-onset zone and propagated to the surrounding areas; we referred to this observation as the “ictal doughnut phenomenon.” Univariate analysis suggested that complete resection of the sites showing the earliest augmentation of ictal HFOs was associated with a good surgical outcome.

Significance:  Recruitment of HFOs at 80–200 Hz in the rolandic area may play a role in determining seizure semiology in epileptic spasms. Our study using macroelectrodes demonstrated that ictal HFOs at 80–200 Hz preceded those at 210–300 Hz.

Observations in previous studies of humans and animal models of epilepsy have led to the hypothesis that high-frequency oscillations (HFOs) at 80 Hz and above may reflect fields of hypersynchronized action potentials and may serve as a biomarker of epileptogenicity (reviewed in Gotman, 2010). Microelectrode recordings of local field potentials from human and rodent brains have demonstrated the presence of interictal paroxysmal HFOs at 80–200 Hz in seizure-onset and non–seizure-onset zones, whereas interictal HFOs at 200–500 Hz were primarily identified in the seizure-onset zones (Bragin et al., 1999b; Staba et al., 2004; Engel et al., 2009). Studies of patients with focal epilepsy using macroelectrodes also showed that the sites showing interictal paroxysmal HFOs at 80–500 Hz were often observed in the seizure-onset zones on electrocorticography (ECoG) (Urrestarazu et al., 2007; Worrell et al., 2008; Jacobs et al., 2009; Schevon et al., 2009; Crépon et al., 2010). Furthermore, ictal ECoG recordings have shown that ictal discharges at the seizure-onset zone often contained HFOs at 80 Hz and above (Allen et al., 1992; Fisher et al., 1992; Jirsch et al., 2006; Ochi et al., 2007). A recent study reported that a good surgical outcome was associated with resection of cortical sites showing interictal HFOs at 80–500 Hz (Jacobs et al., 2010).

It still remains to be statistically determined what range of HFOs is most consistently observed earliest at the ictal onset. The spatiotemporal modulations of ictal HFOs can be statistically determined by time-frequency ECoG analysis, as often employed in studies of sensorimotor-related modulations of HFOs (Crone et al., 1998; Fukuda et al., 2008). Enrolling a large number of ictal events into analysis can ideally increase the statistical power to address this question. In order to address the above-mentioned question with a large statistical power, we decided to study children with epileptic spasms, which are uniquely characterized by clusters of multiple brief seizures and clinical manifestations of seizure events resembling each other (Koehn & Duchowny, 2002). We specifically tested the hypothesis that ictal HFOs with faster frequencies occur prior to those with slower frequencies. Furthermore, the dynamic changes of intracranially recorded HFOs were animated on each individual’s three-dimensional (3D) cortical surface reconstructed from magnetic resonance (MR) images.

Do ictal HFOs drive or are they driven by seizure manifestations? In order to prove that ictal HFOs drive seizure manifestations, recruitment of HFOs in the symptomatogenic zone (Rosenow & Lüders, 2001) must precede the onset of seizure manifestations. Here, we assessed the spatiotemporal modulations of ictal HFOs on ECoG with a spatial resolution of 1 cm and a temporal resolution of 5 ms, relative to the onset of seizure manifestation. Specifically, we tested the hypothesis that recruitment of HFOs in the primary sensorimotor area for the upper extremity precedes the onset of seizure manifestation objectively visualized on electromyography (EMG) from the upper extremities (Fusco & Vigevano, 1993; Panzica et al., 1999; Bisulli et al., 2002). We also determined whether the presence or absence of ictal motor manifestations was related to ictal HFOs in the sensorimotor area or seizure-onset zone. Our prediction was that large augmentation of HFOs not in the seizure-onset zone but in the sensorimotor area would drive prominent ictal motor manifestations visualized on EMG, whereas small HFO augmentation would not.

Methods

Patients

The inclusion criteria of the present study consisted of (1) a two-stage epilepsy surgery using extraoperative subdural ECoG recording in Children’s Hospital of Michigan, Detroit, between April 2006 and December 2009; (2) epileptic spasms captured during ECoG recording; and (3) subdural electrodes chronically implanted on “the rolandic area of interest” defined as the lateral surface of precentral and postcentral gyri 4 cm or above from the Sylvian fissure (Fig. S1), an area comprising the primary sensorimotor area for the upper extremity (Haseeb et al., 2007; Fukuda et al., 2008). The exclusion criteria consisted of (1) the presence of massive brain malformations (such as large porencephaly, perisylvian polymicrogyria, or hemimegalencephaly), which are known to confound the anatomic landmarks for the central sulcus; and (2) history of previous brain surgery. We studied a consecutive series of 11 children with a diagnosis of epileptic spasms (age 1.3–8.8 years; seven female; Table 1) who satisfied the inclusion and exclusion criteria. The study has been approved by the Wayne State University Institutional Review Board, and written informed consent was obtained from the guardians of all subjects.

Table 1.   Clinical data
Patient/genderAge at surgeryAntiepileptic medicationsElectrode placementNumber of sampled sitesMRISurgeryHistologySurgical outcome (follow-up)
  1. f, female; m, male; CLB, clobazam; LEV, levetiracetam; LTG, lamotrigine; OXC, oxcarbazepine; TPM, topiramate; VGB, vigabatrin; VPA, valproic acid; VB6, vitamin B6; Lt, left; Rt, right; F, frontal; P, parietal; T, temporal; O, occipital; Sz, seizure; Subtotal hemispherectomy, resection of the frontal-parietal-occipital-temporal lobes while preserving the rolandic cortex (i.e., pre- and postcentral gyri).

  2. aMultiple subpial transections employed on the rolandic cortex.

1/f1 year 4 monthsVGBRt FPTO112NormalRt subtotal hemispherectomyGliosis>90% reduction (20 months)
2/m1 year 5 monthsVPA, LEVRt FPTO112NormalRt TPOaDysplasia>90% reduction (19 months)
3/f1 year 8 monthsCLB, LEVLt FPTO104Slightly reduced volume of Lt hemisphereLt hemispherectomyPolymicrogyriasz-free (14 months)
4/m2 years 6 monthsTPM, VB6Lt FPTO120NormalLt subtotal hemispherectomyaDysplasia>90% reduction (8 months)
5/f3 years 7 monthsVGBLt FPTO104Subtle blurring of the gray-white matter junction in Lt TLt TPOaGliosis>90% reduction (8 months)
6/f3 years 10 monthsVPA, TPMLt FPTO120NormalLt TPOaGliosis>90% reduction (13 months)
7/f4 years 7 monthsVB6Lt FPTO; Rt FP148Subtle blurring of the gray-white matter junction in Lt FLt subtotal hemispherectomyGliosissz-free (12 months)
8/f4 years 9 monthsVPARt FPTO120Cortical tubersRt FTOCortical tuberssz-free (43 months)
9/f6 years 3 monthsCLBRt FPTO124NormalRt FGliosissz-free (9 months)
10/m7 years 0 monthsLEV, VPA, OXCRt FPTO116NormalRt FPDysplasiasz-free (42 months)
11/m8 years 9 monthsOXC, LTGRt FPTO128Blurring of the gray-white matter junction in Rt TRt subtotal hemispherectomyDysplasiaOne postoperative sz (9 months)

Subdural electrode placement

For subdural ECoG recording and subsequent functional cortical mapping, platinum grid macroelectrodes (intercontact distance: 10 mm; diameter: 4 mm; Ad-tech, Racine, WI, U.S.A.) were surgically implanted as described previously (Asano et al., 2009a). ECoG signals were sampled from a total of 1,308 cortical sites, with electrode contacts in each subject ranging from 104–148 (Fig. S1). The placement of subdural electrodes was guided by the results of scalp video–electroencephalography (EEG) recording, magnetic resonance imaging (MRI), and interictal glucose metabolism on positron emission tomography (PET). All electrode plates were stitched to adjacent plates and/or the edge of dura mater, to avoid movement of subdural electrodes after placement. In addition, intraoperative pictures were taken with a digital camera before dural closure to confirm the spatial accuracy of electrode display on the 3D brain surface reconstructed from MRI.

Coregistration of subdural electrodes on the individual 3D MRI

MRI including a T1-weighted volumetric spoiled gradient echo image as well as fluid-attenuated inversion recovery (FLAIR) image was obtained preoperatively (Asano et al., 2009b). Planar x-ray images (lateral and anteroposterior) were acquired with the subdural electrodes in place for electrode localization on the brain surface; three metallic fiducial markers were placed at anatomically well-defined locations on the patient’s head for coregistration of the x-ray image with the MRI. A 3D surface image was created with the location of electrodes directly defined on the brain surface, as described previously (von Stockhausen et al., 1997; Muzik et al., 2007). The central sulcus, the precentral gyrus, and the postcentral gyrus were identified according to anatomic MRI landmarks, as described previously (Fukuda et al., 2008).

Extraoperative video-ECoG recording

Extraoperative video-ECoG recordings were obtained for 3–5 days, using a 192-channel Nihon Kohden Neurofax 1100A Digital System (Nihon Kohden America Inc, Foothill Ranch, CA, U.S.A.), which has an input impedance of 200 MOhm, a common mode rejection ratio >110 dB, an A/D conversion of 16 bits, and a sampling frequency selectable from 200–10,000 Hz. The sampling rate was set at 1,000 Hz with the amplifier band pass at 0.08–300 Hz. This clinical recording system has adequate specifications for recording low-voltage HFOs (Crone et al., 2006; Fukuda et al., 2008; Kobayashi et al., 2010). The averaged voltage of ECoG signals derived from the fifth and sixth subdural electrodes of the ECoG amplifier (system reference potential) was used as the original reference. ECoG signals were then remontaged to a common average reference. The advantages and limitations of using a common average reference for measurement of event-related gamma oscillations were discussed previously (Crone et al., 2001; Asano et al., 2009b). Channels contaminated with large interictal epileptiform discharges or artifacts were excluded from the average reference. No notch filter was used for further analysis in any of the subjects. As part of our routine clinical procedure, surface EMG electrodes were placed on the left and right deltoid muscles, and electrooculography electrodes were placed 2.5 cm below and 2.5 cm lateral to the left and right outer canthi. Antiepileptic medications were discontinued or reduced during ECoG monitoring until a sufficient number of habitual seizures were captured.

Visual assessment of ictal video-ECoG recording

In this intracranial study, the seizure onset associated with each epileptic spasm event was visually defined as the initial appearance of a sustained-widespread burst of HFOs >30 Hz lasting 0.5 s or longer (Fusco & Vigevano, 1993; Panzica et al., 1999; Asano et al., 2005; Ramachandrannair et al., 2008), which were clearly distinguished from the background rhythms and could not be explained by artifacts or physiologic state changes (such as arousal) (Gotman et al., 1993). Spasm events were visually classified into types based on the seizure-onset zones on ictal ECoG, regardless of ictal manifestations. Independently, spasm events were classified as (1) those in which motor symptoms were visualized as EMG deflections (Fig. 1) and (2) those accompanied by no EMG deflections (Fig. 2). Visual assessment was conducted by H.N. and E.A. while being blinded to the results of time-frequency analysis. To differentiate artifacts from cortical signals, ECoG traces were visually inspected with a low-frequency filter at 53 Hz and a sensitivity of 20 μV/mm; thereby, broadband irregular deflections containing very high-frequency signals synchronized with facial muscle activities seen on electrooculography electrodes were treated as movement artifacts (Otsubo et al., 2008; Nagasawa et al., 2010). ECoG traces affected by artifacts were not included into further analysis.

Figure 1.


Ictal ECoG discharges associated with spasms. (A) Ictal ECoG traces in Patient 4 are shown (type 1 spasms; Table 2). Low-frequency filter: 53 Hz. High-frequency filter: 300 Hz. Ictal augmentation of ripple-band HFOs occurred at channel 1 and gradually involved the surrounding channels. The offset of ripple-augmentation occurred at channel 1 and gradually involved the surrounding channels. The trigger point for time-frequency analysis was placed at the EMG onset (right deltoid muscle). (B) Time-frequency plots derived from 62 spasms are shown. Augmentation of ripple-band HFOs preceded the EMG onset (denoted as ±0 ms). (C) The amplitudes of ripple-band HFOs associated with spasms are shown (see also Video S1).

Figure 2.


Ictal ECoG discharges associated with spasms. (A) Ictal ECoG traces in Patient 9 are shown (type 1 spasms; Table 2). Low-frequency filter: 53 Hz. High-frequency filter: 300 Hz. Ictal augmentation of ripple-band HFOs occurred at channel 1 and gradually involved the surrounding channels. Augmentation of ripple-band HFOs lasted longer at channel 1 compared to the surrounding channels. The trigger point for time-frequency analysis was placed at the ictal ECoG onset. (B) Time-frequency plots derived from 110 spasms are shown. No significant augmentation of ripple-band HFOs was noted at channel 5 in the rolandic area. No evidence of EMG deflections was noted at left deltoid EMG channel. (C) The amplitudes of ripple-band HFOs associated with spasms are shown (see also Video S2).

Measurement of ECoG amplitude modulations associated with spasms

Analysis of ictal HFOs relative to “ the onset of EMG deflection”

This time-frequency analysis was employed in spasm events with the clinical onset visualized as EMG deflections (Table 2), and was designed to evaluate the spatiotemporal relationship between the onset of augmentation of HFOs and the onset of arm jerking. The analysis was performed separately for each spasm type. Ictal ECoG amplitude modulations were evaluated using the trigger point set at the onset of EMG deflection. The principal methods have been validated previously (Brown et al., 2008; Nagasawa et al., 2010).

Table 2.   Ictal ECoG finding
PatientSpasm type (N)Ictal doughnut phenomenonEMG deflectionSz onset zone (Surgery)Surgical outcome (follow-up)Maximum augmentation in sz onset site (%)Maximum augmentation in rolandic area (%)
BetaGammaRippleFast rippleBetaGammaRippleFast ripple
  1. Patients 5, 7, and 11 had two distinct spatial patterns on ictal ECoG discharges. Patients 4 and 6 had three distinct spatial patterns on ictal ECoG discharges. Due to the presence of artifacts associated with spasms, some spasm events were excluded from time-frequency analysis (mean: 1.8 spasms per type; range: 0–12 per type). The initial significant augmentation in the seizure onset zone involved the ripple-band in 14 spasm types, the gamma-band in three types (Patient 2, Patient 10 and Type 1 of Patient 11), the fast-ripple band in one type (Patient 8) and the beta-band in none.

  2. N, number of spasm events included into time-frequency analysis; Lt, left; Rt, right; F, frontal; P, parietal; T, temporal; sz, seizure; MSTs, multiple subpial transections.

 1Type 1 (32)++Rt F (Not resected)>90% reduction (20 months)49775122157496422499218
 2Type 1 (42)++Rt F (MSTs)>90% reduction (19 months)18972979787320092114951009
 3Type 1 (45)+Lt P (Resected)Sz-free (14 months)1411002199215202106801270820
 4Type 1 (62)++Lt F (MSTs)>90% reduction (8 months)375854137939127510321237430
Type 2 (15)++Lt F (Resected)183896693114396276251199
Type 3 (19)++Lt T (Resected)246637344166200409453409
 5Type 1 (25)++Lt F (MSTs)>90% reduction (8 months)163329646220136193272148
Type 2 (54)++Lt P (MSTs)182261325230292273406242
 6Type 1 (23)+Lt F (MSTs)>90% reduction (13 months)1921371006622218121695
Type 2 (43)+Lt P (Resected)7856082034522619520183
Type 3 (46)+Lt T (Resected)3588215004472021078447
 7Type 1 (36)++Lt F (Not resected)Sz-free (12 months)1345627711847824623193
Type 2 (14)+Lt F (Not resected)70273315738219319777
 8Type 1 (29)+Rt F (Resected)Sz-free (43 months)13955473531450403237
 9Type 1 (110)Rt F (Resected)Sz-free (9 months)7731240118326211614
10Type 1 (11)+Rt F (Resected)Sz-free (42 months)1975482729611939825094
11Type 1 (5)+Rt T (Resected)One postoperative sz (9 months)100135112092117015820553
Type 2 (25)Rt T (Resected)5544559612210410114050

Time-frequency analysis was performed using BESA EEG software (MEGIS Software GmbH, Gräfelfing, Germany); each suitable ECoG trial was transformed into the time-frequency domain using a complex demodulation technique (Papp & Ktonas, 1977; Hoechstetter et al., 2004). In that technique, the time-frequency transform was obtained by multiplication of the time-domain signal with a complex exponential, followed by a low-pass filter. The low-pass filter used here was a finite impulse response filter of Gaussian shape, making the complex demodulation effectively equivalent to a Gabor transform. As a result of this transformation, the signal was assigned a specific amplitude and phase as a function of frequency and time (relative to the onset of EMG deflection). In this study, the amplitude (also known as “square root of power”) was used for further analysis. Time-frequency transformation was performed for frequencies between 20 and 300 Hz and latencies between −1,500 and +1,000 ms relative to the onset of EMG deflection, in steps of 10 Hz and 5 ms (Fukuda et al., 2010a). This corresponded to a time-frequency resolution of ±19.9 Hz and ±11.1 ms (defined as the 50% amplitude drop of the finite impulse response filter).

At each time-frequency bin we analyzed the percent change in amplitude (averaged across spasm events of each type) relative to the mean amplitude during the reference period. A total of 50 epochs of 2,500 ms, derived from the interictal traces with the fewest spikes at least 30 s prior to epileptic spasms, were treated as the reference period in the present study. The percent change in amplitude is commonly termed “event-related synchronization and desynchronization” (Pfurtscheller & Lopes da Silva, 1999) or “temporal spectral evolution” (TSE) (Salmelin & Hari, 1994). The percent change in amplitude (unit: %) was calculated without a frequency band at 60 Hz (and its harmonic frequencies) if visual inspection revealed a 60-Hz (and its harmonic frequencies) artifact peak on the amplitude spectral curve for all subdural electrodes (Nishida et al., 2008; Fukuda et al., 2010b).

To test for statistical significance for each obtained TSE value, two-step statistics was performed using the BESA software (Hoechstetter et al., 2004; Asano et al., 2009b). First, statistics based on bootstrapping approach were applied to obtain an uncorrected p-value at each time-frequency bin. In a second step, correction for multiple testing for multiple neighboring bins was performed using the approach developed by Simes (1986). In all figures, blue color indicates attenuation of amplitude and red color indicates augmentation of amplitude in the corresponding time-frequency bin relative to the reference period.

TSE values in a given electrode were declared to be statistically significant only if amplitude modulations involved at least 20 Hz in width and lasted for a period containing at least 10 oscillations (Asano et al., 2004). Such an analytic approach is justified, since a very small probability of type-I error can be expected in determination of amplitude modulations, and ictal discharges are generally defined as “sustained” rhythmic activities (Gotman et al., 1993; Modur & Scherg, 2009). A previous ECoG study reported that HFOs of six oscillations were frequently seen during the interictal state (Worrell et al., 2008). Both interictal and ictal HFOs reported in previous ECoG studies commonly involved wide-range frequency bands ranging at least 20-Hz in width (Ochi et al., 2007; Crépon et al., 2010). Nonetheless, we still recognize that our approach may underestimate amplitude-modulations with a restricted frequency band (<20-Hz in width) or those with a short duration (<10 oscillations).

Time-frequency analysis of HFOs relative to “the visually determined seizure onset”

This time-frequency analysis was employed in spasm events with the clinical onset not visualized as EMG deflections because of no motor symptoms. Ictal ECoG amplitude modulations were evaluated using the trigger point set at the visually determined seizure onset. Otherwise, the aforementioned analytic approach was employed to evaluate the spatiotemporal modulations of ictal HFOs.

Delineation of ECoG data on 3D MRI

ECoG data for each electrode channel can be exported to the given electrode site on the individual 3D MRI-derived brain surface in two different ways (Fukuda et al., 2010b; Nagasawa et al., 2010). To delineate “when, ”“where,” and “at what frequency band” significant alteration of spectral amplitude occurred, time-frequency plot matrixes created above were placed onto a 3D MRI at the cortical sites corresponding to their respective subdural electrode positions. To animate “when,”“where,” and “how many fold” HFOs were increased or decreased, the percent change in amplitude was sequentially delineated on the individual 3D MRI. Thereby, oscillations were classified into those at beta band (20–30 Hz), gamma band (40–70 Hz), ripple band (80–200 Hz), and fast-ripple band (210–300 Hz). This classification was applied not based on the results of this study but on the analytic approaches employed in previous human ECoG studies (Schevon et al., 2009; Fukuda et al., 2010a). We are aware that the distinction between ripples and fast-ripples has been arbitrarily defined (Zelmann et al., 2009).

Statistical analysis

We determined whether augmentation of fast-ripple band HFOs occurred prior to that of ripple band HFOs and whether augmentation of ripple band HFOs occurred prior to that of gamma oscillations. The Friedman test was applied to determine whether any of these onset latencies differed from others. If the p-value was <0.05 on the Friedman test, the Wilcoxon signed-rank test was applied as a post hoc test to compare the pairwise onset latencies (Fukuda et al., 2010a). Bonferroni correction was applied to the post hoc test, and the p-value <0.017 was regarded as significant.

Subsequently, we determined if augmentation of HFOs in the seizure-onset zone and rolandic area of interest both preceded the onset of deltoid EMG deflections in spasms showing motor manifestations, using one-sample t-tests. The seizure-onset zone was defined as the electrode site showing the initial amplitude augmentation of HFOs with significance according to the time-frequency analysis. Finally, we determined whether the amplitudes of HFOs in the seizure-onset zone and in the rolandic area could predict the presence of ictal motor symptoms; the Mann-Whitney U-test was applied to determine whether “the maximum amplitude of ripple-band HFOs in the seizure-onset zone” as well as “that in the rolandic area of interest (averaged across electrodes)” was larger in spasms showing EMG deflections compared to in those without.

Results

Visual ECoG assessment

Each spasm was associated with widespread HFOs lasting 0.5–4 s; such ictal HFOs subsided without evolving into another form of ictal discharges such as repetitive spike-and-wave discharges (Figs. 1–3). Based on the seizure-onset zone, a total of 18 types of epileptic spasms were observed among 11 patients (Table 2). Among these 18 types, 11 showed motor manifestations clearly visualized as deltoid EMG deflections. The remaining 7 types were not accompanied by deltoid EMG deflections; these spasms were characterized by a brief behavioral arrest and/or wide opening of the eyes. In this cohort of patients, no independent focal seizures were captured; none of the patients showed epileptic spasms beginning with repetitive spike bursts on ECoG.

Figure 3.


Ictal ECoG discharges associated with spasms. (A) Ictal ECoG traces in Patient 5 are shown (type 1 spasms; Table 2). Low-frequency filter: 53 Hz. High-frequency filter: 300 Hz. Ictal augmentation of ripple-band HFOs occurred at channel 1 and gradually involved the surrounding channels. The offset of ripple-augmentation occurred at channel 1 and gradually involved the surrounding channels. The trigger point for time-frequency analysis was placed at the EMG onset (right deltoid muscle). (B) Time-frequency plots derived from 25 spasms are shown. Augmentation of ripple-band HFOs preceded the EMG onset (denoted as ±0 ms). (C) The amplitudes of ripple-band HFOs associated with spasms are shown (see also Video S3).

Ictal ECoG changes initially involved a ripple band (80–200 Hz)

The initial significant augmentation in the seizure-onset zone involved the ripple-band in 14 spasm types, the gamma-band in three types (Patients 2, 10, and 11), the fast-ripple band in one type (Patient 8), and the beta-band in none (Table 2). All patients had seizure-onset zones involving the neocortex. Augmentation of beta oscillations during spasms was generally mild and reached the predefined significance only in 4 of the 18 types. The Friedman test demonstrated that at least one of the onsets of augmentation differed among fast-ripple, ripple, and gamma bands (p < 0.001). The Wilcoxon signed-rank test demonstrated that the onset of ictal augmentation of ripple-band HFOs occurred somewhat (but not significantly) earlier than that of gamma-band oscillations (p = 0.02; mean difference 48 ms; median difference 63 ms) and significantly earlier than that of fast-ripple HFOs (p = 0.001; mean difference 136 ms; median difference 105 ms). The dynamic changes of ictal augmentation of distinct oscillations are delineated in Videos S1–S3. In 13 of the 18 spasm types (Table 2), ictal HFOs initially subsided in the seizure onset, whereas ictal HFOs continued to be augmented at the surrounding sites; seizure termination began at the seizure-onset zone and was propagated to the surrounding areas (Figs. 1 and 3). When ictal HFOs subsided in the seizure-onset zone, augmentation of HFOs still remained in the surrounding area and exhibited a doughnut shape in the topographic image. Therefore, we referred to this electrocorticographic observation as the “ictal doughnut phenomenon.”

The maximum change in amplitude in the seizure-onset zone was 145% on average for beta oscillations, 617% for gamma oscillations, 784% for ripple-band HFOs, and 398% for fast-ripple band HFOs (Table 2). The Friedman test demonstrated that at least one of the maximum amplitude change differed among fast-ripple, ripple, and gamma and beta bands (p < 0.001). The maximum amplitude of ictal augmentation of ripple-band HFOs was somewhat (but not significantly) larger than that of gamma oscillations (p = 0.05 on Wilcoxon signed-rank test) and significantly larger than that of fast-ripple band HFOs (p = 0.001) as well as that of beta oscillations (p < 0.001).

Ictal augmentation of HFOs preceded ictal motor manifestations

In all of the 11 spasm types showing EMG deflections, significant augmentation of HFOs at the seizure-onset zone and rolandic area of interest occurred prior to the onset of EMG deflections. The initial augmentation of ripple-band HFOs at the seizure-onset zone occurred on average 470 ms prior to the EMG onset [95% confidence interval (95% CI) 562–379 ms]. Similarly, the initial augmentation of ripple-band HFOs at the rolandic area of interest occurred on average 402 ms prior to the EMG onset (95% CI 536–267 ms).

The amplitude of ictal HFOs in the rolandic area determined the presence or absence of motor symptoms

In spasms showing EMG deflections, the maximum percent change of ripple-band HFOs across the rolandic area of interest was 597% on average, which was larger than in spasms without EMG deflections (average change 127%; p < 0.001 on Mann-Whitney U-test). Again in spasms showing EMG deflections, the maximum percent change of ripple-band HFOs at the seizure onset zone was 877% on average, which was not significantly larger than in spasms without EMG deflections (638%; p = 0.4). Ictal ripple-band HFOs in the primary sensorimotor area predicted the presence or absence of ictal motor symptoms better than those in the seizure-onset zone.

Surgical outcome

The mean postoperative follow-up period was 18 months. Five patients have been seizure-free and one patient had only one postoperative seizure (Table 1); the seizure-onset zones were surgically removed in five of these six patients who obtained such excellent outcomes. The remaining five patients had 90% reduction in seizure frequency; at least one of the seizure-onset zones was not removed in these five patients, and multiple subpial transections were employed in the preserved seizure-onset zone in four of the five patients (Table 2). Patients whose seizure-onset zones were completely resected had a larger chance of obtaining an excellent outcome compared to those with an incomplete resection (p = 0.015 on the Fisher’s exact probability test). Multivariate analysis (Asano et al., 2009a) was not feasible due to the lack of sufficient sample size.

Discussion

Mechanism of ictal ripple-band HFOs followed by oscillations at other frequency bands

The present study using time-frequency analysis with statistics demonstrated that ictal augmentation of ripple-band HFOs was most prominent and generally preceded that of slower oscillations in the seizure-onset zone. Previous ECoG studies have shown that focal seizures are very often characterized by focal augmentation of ictal ripple-band HFOs followed by gradual slowing of the frequency of ongoing oscillations (Akiyama et al., 2006; Khosravani et al., 2009; Nishida et al., 2009). Previous ECoG studies also showed that ripple-band HFOs driven by sensory and cognitive events gradually evolved into slower oscillations (Axmacher et al., 2008; Fukuda et al., 2008, 2010a). Augmentation of HFOs followed by augmentation of slower oscillations was also reported in previous studies of in vitro rat somatosensory cortex (Roopun et al., 2006; Kramer et al., 2008) and in vitro rat hippocampus (Whittington et al., 1997; Traub et al., 1999).

The mechanism of fast-to-slow transition of ripple-band HFOs in human cerebral cortex remains to be elucidated. It was demonstrated that in vitro kainate application to the rat somatosensory cortex augmented gamma oscillations in the superficial cortical layers II–III, beta oscillations in the deep cortical layer V, and subsequently slower oscillations in all cortical layers (Roopun et al., 2006; Kramer et al., 2008). Such in vitro gamma oscillations were suppressed by the blockage of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors mediating fast synaptic transmission or a blockade of γ-aminobutyric acid (GABAA) receptors (Roopun et al., 2006). Alternatively, such in vitro beta oscillations were blocked by reducing gap junction conductance with carbenoxolone but were not affected by blockage of synaptic transmission; the period of such beta oscillations was set by an outward potassium current in cortical layer V (Roopun et al., 2006). Subsequently appearing slower oscillations were facilitated upon reduction of glutamatergic excitation using an AMPA receptor antagonist (Kramer et al., 2008).

The present study also demonstrated that ictal augmentation of ripple-band HFOs consistently preceded that of fast-ripple band HFOs. The temporal relationship between ripple-band and fast-ripple HFOs at the seizure onset has not been well studied. A recent case report of medial temporal lobe epilepsy showed that the ictal discharges recorded with a sampling frequency of 10,000 Hz began with augmentation of ripple-band HFOs followed by augmentation of fast-ripple band HFOs as well as very fast oscillations up to 800 Hz (Kobayashi et al., 2010). Studies of patients with focal seizures using microelectrodes and macroelectrodes demonstrated the presence of interictal HFOs with frequency initially ranging at a ripple-band and gradually evolving into a fast-ripple band, whereas some interictal HFOs had a frequency staying either at a ripple-band or fast-ripple band (Le Van Quyen et al., 2008; Worrell et al., 2008; Crépon et al., 2010). Taken together, we failed to find convincing evidence that ictal fast-ripple HFOs drive or generate ripple-band HFOs. A possible explanation for our observation is that fast-ripple band HFOs may represent a secondary local response (Schevon et al., 2009).

Causal relationship between ripple-band oscillations and seizure manifestations

Ictal recruitment of ripple band HFOs in the rolandic area of interest preceded the onset of contraction of the upper extremity associated with epileptic spasms. Compared to epileptic spasms accompanied by no EMG deflection, those with EMG deflection had a larger amplitude of ripple-band HFOs in the rolandic area of interest. More importantly, the presence or absence of ictal motor symptoms was explained by the amplitude of ictal HFOs in the rolandic area rather than that in the seizure-onset zone. These novel observations suggest that augmentation of HFOs in the rolandic area is the causative predictor of motor manifestations during spasms and further support the hypothesis that not (or not exclusively) brainstem (Hrachovy & Frost, 1989) but cortical activation is primarily responsible for the semiology of epileptic spasms (Gaily et al., 1995; Asano et al., 2005). Previous ECoG studies reported that voluntary movement of the hand and passive movement of the hand by electrical stimulation can augment HFOs in the rolandic area (Crone et al., 1998; Fukuda et al., 2008; Nagasawa et al., 2010), but it is unlikely that augmentation of ictal HFOs seen in the present study purely represented the results from motor manifestations, since significant augmentation of HFOs occurred approximately 400 ms prior to the EMG onset. The lag between initial augmentation of ripple-band HFOs and the onset of EMG deflection in our patients with epileptic spasms was far longer than that between the peak of giant evoked potentials on scalp EEG and the peak of EMG deflection in patients with cortical myoclonus (400 ms in spasms vs. 14–35 ms in cortical myoclonus; Brown et al., 1999). Such a large time difference could be partly attributed to the methodologic difference between studies. It is also possible that the effect of ripple-band HFOs on ictal motor symptoms could be smaller than that of giant evoked potentials seen in cortical myoclonus.

Relationship between ictal ripple-band HFOs and surgical outcome

Univariate analysis suggested that complete resection of the site showing earliest amplitude augmentation (i.e., seizure-onset zone) was associated with a good surgical outcome. We are aware that the sample size was small and a multivariate analysis was not feasible in this retrospective study. Nonetheless, this finding is consistent with the widely accepted notion that complete resection of the seizure-onset zone is a key factor in obtaining a good outcome (Jeha et al., 2007; Jayakar et al., 2008; Asano et al., 2009a). Surgical resection generally includes the seizure-onset zone and epileptogenic lesions seen on neuroimaging. Multiple subpial transections, instead of resection, were employed on the rolandic area showing the seizure onset in four patients, all of whom experienced postoperative seizures, although the frequency of seizures was significantly reduced. This finding is consistent with the results of a meta-analysis of 76 studies suggesting that only 16% of the patients who had multiple subpial transections on the eloquent cortices obtained a long-term seizure-free outcome (Téllez-Zenteno et al., 2005). In our institute, the extent of cortical resection is determined after the epilepsy surgery team and the family of the patient have extensive discussions regarding the pros and cons of surgical resection of rolandic area.

Propagation of ictal HFOs

In a substantial proportion of epileptic spasm events, seizure termination began at the seizure-onset zone and was propagated to the surrounding areas; we referred to this novel electrocorticographic observation as the “ictal doughnut phenomenon” (videos S1 and S3). Previous studies of patients with focal seizures provided representative examples of ictal ECoG recordings, which failed to suggest the presence of “ictal doughnut phenomenon” (Allen et al., 1992; Fisher et al., 1992). We did not describe this phenomenon in our previous study (Asano et al., 2005), where ECoG signals were sampled with a sampling frequency of 200 Hz and time-frequency analysis was conducted using Fast Fourier Transformation, which has a good frequency resolution but does not have a good temporal resolution. Further clinical studies are warranted to determine whether the “ictal doughnut phenomenon” is specific to epileptic spasms and whether there are other clinical predictors of this phenomenon. The present study suggested that candidate clinical predictors of “ictal doughnut phenomenon” include younger age of patients, as patients showing “ictal doughnut phenomenon” were somewhat younger than those without (Table 2; p = 0.06 on t-test).

“Ictal doughnut phenomenon” can be explained by corticocortical propagation of ictal HFOs followed by termination of HFOs earlier in the seizure-onset zone. Termination of ictal HFOs was not systematically assessed in previous ECoG studies of patients with focal seizures or epileptic spasms. A previous study of a rat model of epilepsy showed that the duration of ictal discharges characterized by HFOs alone was significantly shorter than those showing HFOs subsequently evolving into another form of ictal discharges such as repetitive spike-and-wave discharges (Bragin et al., 1999a). In the present study, we found that ictal HFOs associated with spasms failed to evolve into another form of ictal discharges; we still do not know whether an active termination mechanism worked in the seizure-onset zone or if seizures simply terminated in the same order as they started due to the failure to sustain the ictal HFOs in a given site.

The cellular mechanisms for seizure termination are not well understood in focal seizures or epileptic spasms. It has been suggested that generation of cortical HFOs seems to involve several neurotransmitter systems including glutamatergic and GABAergic synaptic transmission (Traub et al., 2004; Le Van Quyen et al., 2006). Studies of rat models of epilepsy showed that blockade of the gap junctions with carbenoxolone facilitated termination of ictal HFOs (Carlen et al., 2000; Gajda et al., 2003). A number of investigators have developed animal models of epileptic spasms using variable methods (reviewed in Galanopoulou & Moshé, 2009). Further studies using such models may be useful to understand the mechanism of “ictal doughnut phenomenon.”

Methodologic issues

The benefits of ECoG recording include a better signal-to-noise ratio compared to scalp EEG and MEG, which record cortical signals from outside of the scalp. Inevitable limitations of ECoG recording include sampling limitation and antiepileptic drugs. In our study, the majority of patients had subdural electrodes placed only on the cortical surface of the presumed epileptogenic hemisphere; preoperative assessment using scalp video-EEG recording and neuroimaging failed to suggest the presence of independent epileptogenic zones involving the contralateral hemisphere. Therefore, we were not able to evaluate the other hemisphere or subcortical structures. Furthermore, we do not know how epileptogenic the symptomatogenic zone on the contralateral side was. It is still uncertain whether the maximal HFOs were obtained from one of the active electrodes placed at every 1 cm distance, or such activities occurred in some brain regions between subdural electrodes or the deeply situated cortex along a sulcus. Antiepileptic drugs may affect the findings of time-frequency ECoG analysis. It was reported that phenytoin, a sodium-channel blocker, elevated motor thresholds to transcranial magnetic stimulation but had no effect on motor-evoked potential amplitudes (Chen et al., 1997). A human study of interictal HFOs using macroelectrodes showed that reduction of antiepileptic drugs was followed by a 3% increase in duration of ripple-band HFOs (Zijlmans et al., 2009). The size of subdural electrodes may have affected the results of present study. A previous study of a rat model of epilepsy using macroelectrodes demonstrated that 0.85 mm2 macrocontacts detected interictal HFOs as well as contacts almost 50 times smaller (0.018 mm2), and that the duration of interictal HFOs differed minimally between the different sized contacts (Chatillon et al., 2009). The band-pass width was set at 0.08–300 Hz, and HFOs with frequency of >300 Hz were not assessed in the present study. The degree of augmentation of ripple-band HFOs was larger than that of slower oscillations, and ripple-band HFOs were visualized on ECoG traces with a low-frequency filter of 1.6 Hz in the present study (Fig. S2). Therefore, it is unlikely that ictal augmentation of ripple-band HFOs resulted from the effects of harmonics of sharp transients or slower oscillations (Bénar et al., 2010).

The interictal traces within spasm-free periods were treated as the reference period in the present study. We are aware that interictal HFOs of epileptogenic nature during the reference period may significantly affect the results of time-frequency analyses. To avoid yielding misleading results, the interictal periods with fewest spikes were chosen as the reference period, and visual inspection of raw ECoG traces was employed to validate the results of time-frequency analysis as shown in Figs 1–3. The rolandic area of interest was defined not based on the results of cortical mapping using electrical neurostimulation but based on the anatomic landmark. It has been reported that sensitivity of electrical neurostimulation is not as good in children as in adults and that failure to elicit a clinical symptom using neurostimulation does not prove the absence of eloquent function in the stimulated site (Haseeb et al., 2007).

Only 2 of the 11 patients had two different spasms with and without EMG deflections. Therefore, we were not able to compare the maximum amplitude of ripple-band HFOs between spasms with and without EMG deflections, using intra-individual paired statistics. This is another methodologic limitation in the present study.

Acknowledgments

This work was supported by NIH grants NS47550 and NS64033 (to E. Asano). We are grateful to Lunliya Thampratankul, M.D., Masaaki Nishida, M.D., Ph.D, Miho Fukuda, M.D., Ph.D., Carol Pawlak, R.EEG/EP.T, Ruth Roeder, R.N., M.S., Sarah Minarik, R.N., and the staff of the Division of Electroneurodiagnostics at Children’s Hospital of Michigan, Wayne State University for the collaboration and assistance in performing the studies described herein.

Disclosure

None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

Ancillary