Fast ripple band high-frequency activity associated with thalamic sleep spindles in pediatric epilepsy

Abstract

Objective:

To investigate high-frequency activities (HFA) associated with thalamic sleep spindles.

Methods:

We studied a cohort of ten pediatric patients with medication resistant epilepsy who were identified as potential candidates for thalamic neuromodulation. These patients had thalamic sampling as well as presumed epileptogenic zones, using stereotactic EEG (SEEG) with a sampling frequency of 2,000 Hz. We quantified the summated high-frequency activity (HFA) in the fast ripple band associated with sleep spindles using 20-minute scalp EEG and SEEG recordings during non-REM sleep and analyzed its correlation with spindle characteristics.

Results:

HFA, with a median peak frequency of 330 Hz, was distinctively observed in the thalamus and temporally correlated with thalamic sleep spindles. Such HFA demonstrated significant coupling with the sleep spindle range of 11–16 Hz. The duration of HFA positively correlated with higher density and longer duration of accompanying thalamic spindles. Thalamic HFA’s duration negatively correlated with the presence of cortical interictal epileptiform discharges. Thalamic spindles generated in channels with HFA often coincided with sleep spindles in various brain regions.

Conclusion:

Fast ripple band HFA associated with sleep spindles was observed exclusively in the thalamus.

Significance:

Thalamic HFA associated with thalamic spindles may represent a thalamus-specific physiological phenomenon.

1. Introduction

High-frequency oscillations (HFOs) exceeding 80 Hz have been an active area of research, being linked to both pathological contexts in epilepsy and physiological processes in the brain (Jefferys et al., 2012, Zijlmans et al., 2012). These oscillations have initially demonstrated potential as reliable neurophysiological biomarkers for delineating the epileptogenic zone (Bragin et al., 1999a, Bragin et al., 1999b, Jirsch et al., 2006, Staba et al., 2004, Worrell et al., 2004). HFOs have been the subject of extensive research, particularly in cases suitable for surgical resection in epilepsy. Numerous research groups have explored the utility of HFOs in guiding the process of surgical resection, with the goal of improving postoperative seizure outcomes (Jacobs et al., 2018, Jacobs et al., 2010, Wu et al., 2010, Zweiphenning et al., 2022). In the context of neuromodulation therapy, a recent study reported that targeting cortical areas generating HFOs with RNS was associated with improved seizure outcomes (Weiss et al., 2023). However, despite the potential, HFOs can also be recorded in healthy brain regions, which complicates the interpretation of HFOs. Physiological HFOs are frequently recorded in the somatosensory cortex, hippocampus, and occipital cortex (Frauscher et al., 2018, Guragain et al., 2018, Kuroda et al., 2021). HFOs in the hippocampus, also known as hippocampal ripples, play important cognitive roles, such as memory consolidation, by interacting with various frequency bands (Ngo, Fell and Staresina, 2020, Oyanedel et al., 2020). Additionally, high-frequency activity (HFA) is broad-band, non-oscillatory, and has been reported in healthy brain regions such as the occipital lobe, hippocampus, and language-related cortices. This activity is often linked with physiological HFOs (Frauscher et al., 2018, Melani et al., 2013, Norman et al., 2019). While the thalamus is known for its crucial role in interacting with the hippocampus and cerebral cortex through sleep spindles, (Gonzalez et al., 2018, Latchoumane et al., 2017) research specifically focusing on HFOs/HFA in the thalamus remains scarce. Thalamic and subthalamic fast ripple band activity (> 200 Hz) have been identified as potential indicators of physiological activities in patients with essential tremor (Schnitzler et al., 2018) and Parkinson’s disease, rather than reflecting the disease activity. (Wang et al., 2014) Thus, the association of HFA with sleep spindles in the thalamus presents a significant area for further research.

To bridge this knowledge gap, we analyzed HFA in the thalamus during spindles within a unique cohort of patients with pediatric-onset medication resistant epilepsy (MRE) who underwent thalamic sampling during stereotactic EEG (SEEG) evaluations for potential thalamic neuromodulation candidates. We specifically explored the relationship between thalamic HFA and thalamic sleep spindles, as well as the effect of epileptiform discharges on these thalamic HFA, to ascertain the biological characteristics of HFA detectable in the AN (anterior nucleus of the thalamus) and CM (centromedian nucleus of the thalamus) (Fig. 1).

Fig. 1. Flow chart of the study:

After EEG preprocessing, we detected spindles using an automated spindle detector with subsequent human validation. To determine the presence of high-frequency activity (HFA), we generated an averaged time–frequency plot and classified the status of HFA (+) by each channel, which revealed that HFA was observed solely in the thalamic channels. This HFA activity was correlated to thalamic spindle characteristics. Thus, we subsequently investigated the characteristics of thalamic channels based on the HFA status by the relationship of interictal epileptiform discharges, co-occurrence of thalamic spindles and cortical sleep spindles, and phase-amplitude coupling.

2. Materials and Methods

2.1. Patient cohort

This study employed a retrospective design to identify patients with pediatric-onset epilepsy who had been admitted to UCLA Mattel Children’s Hospital. These patients underwent chronic SEEG implantation with electrodes inserted into the thalamus, specifically into the anterior nucleus of the thalamus (AN) and/or centromedian nucleus of the thalamus (CM), between November 2020 and February 2023. They have MRE and were suspected to have focal seizure onset and were considered potential candidates for thalamic neuromodulation. Patients were excluded from the analysis when there was a lack of spindles or no EEG segment free from seizures for over two hours.

2.2. Standard protocol approvals, registrations, and patient consent

The research protocol received approval from the institutional review board (IRB) at UCLA. The study protocol that covered this study exempted the requirement for written informed consent since any retrospective EEG data used in this study were anonymized prior to data extraction and analysis. The justification of thalamic sampling with SEEG was carefully discussed in the context of clinical needs, and no additional intracranial electrodes were placed for the purpose of the research study. Informed consent was obtained for all the surgical procedures related to this study, including SEEG placement in the thalamus. A separate IRB-approved research protocol has been tracking surgical outcomes, including complications from thalamic implants. It’s important to note that this study did not constitute a clinical trial and was not registered in any public registry.

2.3. Surgical evaluation

The plan for SEEG placement was discussed at our multidisciplinary epilepsy surgery conference (consisting of epileptologists, neurosurgeons, radiologists, and neuropsychologists) and was based on the combination of data from seizure semiology, neurological examination, neuroimaging findings (MRI, PET, and magnetoencephalography), neuropsychological evaluation, and scalp EEG with emphasis primarily on seizure onset zones. Ipsilateral (to the presumed site of the presumed seizure onset) AN and CM thalamic SEEG electrodes were placed to determine whether an ictal pattern can be detected in the thalamus and provide a potential target for neuromodulation. If the patient had bilateral seizure onsets during the phase 1 monitoring with scalp EEG, thalamic SEEG electrodes were placed on the side of the most significant seizure burden, or if unclear bilateral thalamic electrodes were placed (SEEG sampling details are listed in Table 1).

Table 1.

Patient Characteristics.

Subject ID	Age	Sex	Neuroimaging (MRI/PET)	Epilepsy type/etiology	Prior surgery	Current AED use	Thalamic Sampling (number)	Non-thalamic Sampling location (number)	Recording Duration (min)
1	2.7	M	L cortical & subcortical Tubers, subependymal nodules along L FTP / L FTP	Focal / Structural (TSC)	L laser ablation, CC, L T lobectomy & L O Tub resection	CBD, CLB, FBM, LEV	RAN: AN (3) RCM: Vlp (2), CM (1)	R: F, P, T (56)L: F (22)	25.6
2	2.9	M	NL / NL	Focal / Genetic-Structural (FCD due to NRPL3 mutation)	None	OXC, VGB	RCM: CM (3)	R: F, T, LIM (54) L: F, T, LIM (18)	20.0
3	10.5	M	R F gliosis / R F	Focal /Structural (R ethmoid encephalocele)	R F pole resection	LCM, LEV, BTM	RAN: AN (4) RCM: CM (4)	R: F, T, P, LIM (56)L: F, T, P, LIM (54)	20.0
4	12.9	F	NL / R T	Focal / Unknown	VNS	FBM, CLB, KD	RAN: AN (3), MDm (1) RCM: PUL (4)	R: F,T,LIM (42) L: F,T, LIM (34)	30.3
5	16.1	M	NL / L T, P	Focal / Unknown	None	CBD, OXC, CLB	LAN: AN (3), MDm (1) LCM: CM (4)	R: F, P, O, T, LIM (74)L: F, P,O, T, LIM (66)	20.1
6	18.3	M	NL / NL	Focal / Unknown	None	LEV, LTG, RUF, CLB	RAN: MDm (2), AN (2) RCM: CM (3)	R: F, T, LIM, P (36) L: F, T, LIM (26)	26.5
7	21.9	M	L T hyperintense FLAIR Signal / NL	Focal /Structural (Encephalitis & FCD)	VNS	CNB, LCM, LEV	LAN: MDm (2), LD (2) RAN: AN (2), MDm (1), VLa (1)	R: F, T, P, LIM (60)L: F, T, P, LIM (60)	17.2
8	25.2	F	L TPO gliosis / L T,P,O	Focal / Structural (FCD)	L temporal resection, VNS	BTM, CLB, CNB	LAN: MDm(3), AN(1)LCM: VPL(4)	L: F, P, O, T, LIM, BG (30)	20.6
9	25.7	F	NL / L T	Focal / Unknown	VNS	LTG, LEV	LAN: MDm(3), AN(1)LCM: CM(2)	L: F, T, LIM (66)	20.4
10	28.1	M	L P & PV WMT2FLAIR hyperintensity / L P	Focal / Structural (FCD)	L hemicraniotomy	CNB, OXC, CZP	LAN: MDm(2), LD(2)LCM: CM(2),	R: T,P, LIM (24) L:T,P, LIM (56)	23.4

AN, anterior nucleus of thalamus; BG, basal ganglia, BTM, brivaracetam; CBD, cannabidiol; CLB, clobazam; CZP, Clorazepate; CM, centromedian nucleus of thalamus; CNB, cenobamate; F, female; F, frontal;

FBM, felbamate; FCD, focal cortical dysplasia; KD, ketogenic diet; LCM, lacosamide; L, left; LD, laterodorsal; LEV, levetiracetam, LIM, limbic system; LTG, lamotrigine, M, male; MD, mediodorsal; NL, normal; O, occipital;

OXC, oxcarbazepine; P, parietal; PUL, pulvinar; PV, periventricular; R, right; Ruf, rufinamide; T, temporal; TSC, tuberous sclerosis, complex; Tub, tuber; ZNS, zonisamide; VGB, Vigabatrin;

VLa Ventral lateral anterior nucleus; Vlp, Ventral lateral posterior nucleus, VNS, vagus nerve stimulator.

2.4. SEEG placement and confirmation

BrainLab Elements software was used for planning the electrodes to the intended targets using T1-weighted sequences, and the trajectories were guided by a gadolinium-enhanced T1-weighted MRI. Target subcortical structures (including AN and CM) were identified and outlined by an experienced neuroradiologist (NS) on a case-by-case basis prior to the trajectory planning.(Edmonds et al., 2023) The targets and trajectories that were planned using the MRI MPRAGE2 sequence were then co-registered to a volumetric computed tomography (CT) scan acquired after placing the patient’s head into the Leksell frame. Each electrode was placed using the Leksell coordinates obtained from the BrainLab elements software. Four contact Spencer Depth Electrodes with 2.5 mm spacing were used exclusively for thalamic targets. An intraoperative or immediate postoperative CT scan was used to rule out intracranial hemorrhage and confirm the final position and trajectory of each electrode placed. The location of electrode placement was verified post-operatively with CT co-registered with the pre-op T1-weighted MRI using the BrainLab Elements software. Anatomical segmentation was conducted by using preoperative MRI scans to generate a patient-specific reconstruction with FreeSurfer (Fischl, 2012). Subsequently, this reconstruction facilitated the co-registration process in Brainstorm, allowing for the precise localization of each depth electrode’s contacts. The positions of these contacts were classified according to the Desikan-Killiany-Tourville atlas.(Klein and Tourville, 2012) The location of thalamic electrodes was confirmed using automated segmentation with subsequent visual verification.(Iglesias et al., 2018) For subsequent analysis, electrodes situated in white matter or associated with lesions (such as tubers) were excluded.

2.5. EEG data acquisition

SEEG and scalp EEG recordings were obtained using Nihon Kohden amplifier JE-120A (Irvine, California, USA) with a digital sampling frequency of 2,000 Hz, which enabled a bandpass frequency of 0.01–600 Hz. All EEGs were part of the clinical EEG recording. The EEG data were preprocessed using a notch filter at 60 Hz intervals to remove power-line noise. We identified 20-minute segments of EEG data during sleep stage 2 (N2) based on scalp EEG and SEEG findings that occurred more than 2 h apart from seizures. These EEG segments were obtained from the initial night of implantation. Bipolar montages were utilized for SEEGs, while averaged montages were applied for scalp EEGs.

2.6. Detection and characterization of sleep spindles

Prior to analysis, EEG data underwent preprocessing, which included bandpass filtering to extract the characteristic oscillatory frequency range associated with sleep spindles (sigma band: 11–16 Hz). We then employed the YASA spindle detector’s short-term energy (STE) spindle detector (Vallat and Walker, 2021) to identify spindles within the recorded EEG data (https://github.com/raphaelvallat/yasa). Specifically, sigma-band EEG activities with durations between 0.5 and 3 s, exceeding 1.5 standard deviations from the mean amplitude, were selected as candidate spindles (Schreiner et al., 2022). This preprocessing step was applied to all scalp EEGs (Fz, Cz, Pz) and SEEG channels targeting cortex and the thalamus. Subsequently, the detected spindle events underwent visual inspection for validation. Of note, spindles associated with interictal spikes were removed from the analysis since spikes can increase the power in the sigma band (Kramer et al., 2021). A board-certified epileptologist (AD), who has had favorable consistency with HN with a kappa-score of 0.62, performed the validation. This spindle detector has been validated against visual inspection of thalamic spindles (N = 1740) with an F1 score of 0.63, Recall of 0.66, and precision of 0.63, which is compatible with the previous reports (Warby et al., 2014).

2.7. Definition of High-Frequency activity (HFA) and classification of detected electrode

As an initial exploratory analysis, the detection of HFO-like activity in the thalamus was first performed with the STE algorithm, with our built-in deep learning- based classification module integrated within the package (Zhang et al., 2024). Specifically, EEG signals were processed using 80 and 500 Hz bandpass filters. We identified HFO candidates by applying a root mean square threshold, optimized for a duration criterion of more than 6 ms. Among three patients with an RMS threshold of 5SD, no HFO candidates were detected. Therefore, we set a lower RMS threshold to 2SD, which occasionally detected repetitive HFO-like activity overriding thalamic sleep spindles (Fig. 2B). Among these patients, 71.5 % of the HFO-like activities were detected within spindles at thalamic SEEG channels, with a peak frequency of 297.36 Hz (208.51–334.46 Hz). Furthermore, the cross-correlation histogram of the HFO-like activity centering the trough of the thalamic spindle showed these HFOs-like activity peaking at the center of the spindle trough (median 0.25 s; Fig. 2C). Because the power of HFOs in the subcortical area tends to be weaker than the cortical HFOs (Schnitzler et al., 2018, Wang et al., 2014), and difficulty in separating with artifacts, we decided to detect and analyze HFOs by summating spindles which potentially contain HFOs similarly to event-related potential. We justified this approach also because such low-amplitude HFOs seen in the sleep spindles seemed different from our traditional pathological HFOs seen in epileptic tissues. After confirming the presence of detected spindles in each channel, we extracted surrounding signal segments, creating 5-second intervals for each detected spindle by centering the trough time point of the sigma band of the spindle at 2.5 s. Subsequently, time–frequency analysis was conducted within the frequency ranges of 3–80 Hz and 80–500 Hz for each segment. The time–frequency plot was generated by continuous Gabor Wavelets ranging from 3–80 Hz and 80–500 Hz with a frequency resolution of 128 and 420, respectively (Zhang et al., 2022a, Zhang et al., 2024, Zhang et al., 2022b). Averaging time–frequency plots of the detected spindles, time-locked to the trough point of the sigma band, across each channel between 80–500 Hz provided a quantitative assessment of HFA associated with sleep spindles for subsequent analysis. The status of the presence of HFA was defined by utilizing the averaged time–frequency plots obtained in the previous step. The 5-second time–frequency data were vectorized by calculating the average within the frequency band of 200–500 Hz, followed by a smoothing by convolution by 200 msec time window. A baseline activity period was established using the first 0.5 s of each 5-second interval, and instances of HFA were identified when the smoothed averaged vector exceeded 5 SD from this baseline for more than 0.5 s. This power threshold was based on the definition of traditional HFO detection, and duration was based on the length of spindle detection, which allowed us to define 5-second intervals containing HFA and to investigate their association with spindle activity. The peaks of HFA events were determined by identifying the time and frequency coordinates corresponding to the point achieving the maximum power within the defined time–frequency window. A similar process was performed in the sigma band (11–16 Hz) for comparison. For visualization purposes, time–frequency plots of the centered 3-second interval (0.5––4.5 s in the time–frequency window, centering the trough of sigma band activity) were generated.

Fig. 2. SEEG sampling and HFOs-like activity detected in the thalamus.

(A) Left: A top and front view of a 3D reconstruction of the SEEG implantation from all the patients (n = 10) is shown. Right: The trajectory of the thalamic electrodes is superimposed on the coronal view, around the anterior nucleus of the thalamus at the upper end and the centromedian nucleus of the thalamus at the lower end. (B) Examples of the detected HFOs superimposed on sleep spindles in thalamic channels, using the Short-Term Energy (STE) method with a 2 SD threshold of the root mean square, are presented (5SD threshold did not detect any candidates). The top panel displays the raw EEG signal for 10 s. Green dots denote detected HFOs-like activity during spindle activity. The middle row shows the expanded raw EEG and filtered EEG signal. Note that HFO-like activity occurs during the upgoing phase of spindle activity (black and gray arrows show HFO-like activity that was detected and undetected by the HFO detectors, respectively). The bottom panel shows the time–frequency plot between 10–80 and 80–500 Hz of expanded EEG signal. Note that apparent fast ripple band activities are detected during sleep spindles. (C) Cross-correlation histogram among all subjects centering the trough of spindles revealed that the HFOs-like activity peaks around spindles. SEEG: stereotactic EEG; HFOs: high-frequency oscillations.

2.8. Thalamic spindle characteristics based on HFA status

To conduct a comprehensive analysis of spindle characteristics within the thalamic electrodes depending on the state of HFA, we selected channels that detected the highest number of spindles for each thalamic electrode. These thalamic channels were further categorized into two groups: HFA (+) and HFA (−) channels. We defined HFA(+) channels as thalamic channels with confirmed HFA using the above method. We quantified the frequency, power, duration, and density of the thalamic spindles based on their HFA status. Within the HFA (+) group, we conducted a comparative analysis of the peak latencies between HFA and sigma bands derived from the channel-wise averaged time–frequency plot. Finally, we explored the correlation between HFA duration and thalamic spindle duration/density.

2.9. Spindle subcategorization by the presence of interictal epileptiform discharges (IEDs)

In this study, we defined two types of interictal epileptiform discharges (IEDs): (1) HFOs associated with spikes (spk-HFOs) and (2) spikes. For spk-HFOs detection, we followed our prior work. (Zhang et al., 2024, Zhang et al., 2022b) Briefly, after using the STE-based HFO candidate detection of RMS thresholding at 5SD (detailed described at 2.7), a convolutional neural networks-based deep learning pipeline was used to further distinguish between artifacts, HFOs without spikes, and real spk-HFOs. For spike detection, we followed the previously published methodology (Dahal et al., 2019). Initially, we applied a bandpass filter to isolate the frequency range of 25–80 Hz, removing power line interference (at 60 Hz) and its harmonic frequencies. We then rectified the filtered signal to enhance spike visibility. Afterward, the signal was divided into 30-second segments. For each segment, a threshold was calculated as three standard deviations from the segment’s signal envelope. Signals exceeding this threshold were marked as potential spikes and subsequently validated through visual inspection. Such IEDs (both spk-HFOs and spikes) were evaluated in all cortical channels and in specific subcortical channels, namely the hippocampus and amygdala, on the ipsilateral side of the thalamic channels. Contacts that sampled white matter, lesions (such as tubers), and outside of the brain tissue were removed. Since IEDs might affect the characteristics of spindles and HFA, we categorized spindles into three groups, similar to a previous study (Sákovics et al., 2022): (1) Pure Spindle: These thalamic spindles show no apparent relationship with IEDs. (2) Spindles Preceded by IEDs: These thalamic spindles were preceded by ipsilateral cortical IEDs within one second. (3) Spindle Co-occurring with IEDs: These spindles occurred simultaneously with ipsilateral cortical spikes during the thalamic spindle activity. The characteristics of these spindles and HFA were derived for each HFA (+) thalamic nuclei based on the summarized time–frequency spectra characteristics of HFA and spindles in each category.

2.10. Analysis of spindle co-occurrence

Spindles were categorized as co-occurring when the maximum power of the spindle fell within the detected thalamic spindle window of either HFA (+) or HFA (−) channels. The co-occurrence percentage was calculated using the formula: (number of co-occurring spindles in target location with thalamic spindles) / (total number of spindles in the target location channel) × 100 by each channel. Subsequently, the numbers from channels at the same location were averaged. Channels associated with the seizure onset zone were excluded. Channels other than the target thalamic channels were further classified into frontal, parietal, temporal, cingulate/insula, hippocampus (including the hippocampus, parahippocampus, and fusiform gyrus), adjacent thalamic, and other thalamic channels and the co-occurrence percentage was calculated. The adjacent thalamic channel was defined as channels located within the same thalamic electrode, while other thalamic channels referred to channels on the ipsilateral side but were different from the targeted thalamic nucleus channel.

2.11. Phase-Amplitude coupling (PAC) analysis

To quantify the strength of local neurophysiological coupling between HFA and spindle phase at the thalamus, we employed PAC analysis to HFA (+) channels. We adopted the modulation index reported by Tort et al., which is a widely recognized metric in the field (Tort et al., 2010). Our analysis focused on specific frequency bands: a low-frequency band spanning 11–16 Hz for spindle activity and a high-frequency range of 200–500 Hz for HFA. We analyzed PAC for the time segment from −1.5 to 0.5 s, centering the time-zero to the trough of the sigma band to include both spindle and HFA bands on each spindle event, and then summarized the results. By shuffling the pairs of trials for phase and amplitude signal, 200 hundred permutations using maximum statistics were used to define the significance of the modulation index. The PAC analysis was performed using the “Tensorpac” package, a reputable open-source tool extensively utilized for PAC analysis in neural data research (Combrisson et al., 2020).

2.12. Statistical analysis

All statistical analyses were performed in Python. For the bivariate analysis of continuous variables, the Mann-Whitney test was performed. When comparing three groups, a one-way ANOVA was employed, followed by the Tukey-Kramer test for multiple comparisons. The correlation was analyzed using Pearson correlation. All of the multiple comparisons were adjusted using Bonferroni correction unless otherwise specified.

2.13. Data and code availability

The data and codes are available from the corresponding author upon request.

3. Results

3.1. Study cohort

Initially, twelve patients were identified for inclusion. Of those, ten subjects underwent subsequent analysis because two subjects did not have sleep spindles during the recording, and there was no availability of a seizure-free period (at least 2 h away from a seizure). The details of the study cohort are described in Table 1. Each subject had between 38 and 148 depth electrode contacts with a median of 83 (Fig. 2A). These contacts provided unilateral coverage of AN/CM in eight patients, bilateral AN coverage in one patient, and unilateral CM coverage in one patient. Each thalamic electrode contained four contacts. Scalp EEG recordings were conducted in six out of the ten patients. Out of the 40 AN contacts, 39 were in the thalamus, with 19 in AN, 15 in mediodorsal magnocellular, four in laterodorsal, and one in ventral lateral anterior nuclei. Among the 36 CM contacts, 29 were situated within the thalamus, including 19 at CM, four at ventral posterolateral, four at pulvinar, and two at the ventral lateral posterior. There were no complications associated with the placement of SEEG electrodes, including those in the thalamus, and specifically, there were no incidents of intracranial hemorrhage.

3.2. HFA evaluation during spindles

The spindle detector identified 14,071 spindles, including 3,120 from the thalamic channel, 1,411 from the scalp, and 9,540 from the cortical channels. The presence of HFA, the summated activity of HFA associated with spindles, was confirmed exclusively in thalamic channels (Fig. 3). Notably, the HFA activity exhibited fluctuations in power that were temporally associated with spindle band activity. Out of 57 designated thalamic bipolar channels, four channels were outside of the thalamus. Out of 53 channels within the thalamus, 22 met the definition of HFA (+) channels. None of the four channels outside the thalamus were HFA (+). Conversely, none of the cortical channels (465 bipolar channels) or scalp channels (18 monopolar channels) exhibited HFA. The median peak frequency of HFA was in the fast ripple range, which was 330.88 Hz (IQR: 318.82–340.39 Hz). For further investigation, we selected a single bipolar channel with the most frequent spindle detection from each thalamic electrode (each electrode contains two to three bipolar channels), which resulted in ten HFA (+) channels and nine HFA (−) channels.

Fig. 3. Examples of Spindle Detection and Summation Analysis of Time-Frequency Plot.

Examples of sleep spindles and HFAs at each sampling site (AN, CM, frontal cortex, and scalp EEG) are shown. At each site, the upper panel displays a spindle waveform, filtered sigma band signal, and its corresponding time–frequency plot. The first and the second rows of the bottom panel show summated time–frequency plots of HFA and sleep spindles. The third row represents the averaged power changes in the 200–500 Hz band as a function of time in blue color (orange represents the smoothed vector). Red dashed lines indicate the 5 SD line from the baseline. N represents the number of spindles from a single channel. It should be noted that the average power in the middle row exceeded 5 SD lines over 0.5 s, which classified these channels as HFA(+). AN: anterior nucleus of the thalamus; CM: centromedian nucleus of the thalamus; HFA: high-frequency activity.

3.3. Characteristics of thalamic spindles based on HFA status

Within each thalamic electrode, we identified the thalamic channel that detected the highest number of spindles (n = 19). Each channel was classified into two groups based on the aforementioned criteria: HFA (+) (n = 10) and HFA (−) (n = 9) channels. Our analysis revealed several notable distinctions between these two groups. Specifically, HFA (+) channels had lower peak frequencies (median 12.75 [IQR: 12.02–13.22] Hz versus median 14.85 [IQR: 14.00–15.30] Hz; p = 0.014, Cohen’s d = 0.15), longer durations (median 1.16 [IQR: 1.04–1.38] seconds versus 0.91 [IQR: 0.72–1.20] seconds; p = 0.055, Cohens’d = 0.71), and higher densities (median 5.57 [IQR: 3.17–7.23]/min versus median 0.70 [IQR: 0.31–3.33]/min; p = 0.013, Cohen’s d = 0.78) than HFA (−) channels. However, none of these p-values remained significant after multiple comparisons due to the limited number of samples (p > 0.05). Within the HFA (+) channels, the peak latency (median −0.31 [IQR: −0.60 to −0.14] seconds; p = 0.037) of the HFA band was earlier compared to those of the sigma band. Additionally, we identified a positive correlation between HFA duration and thalamic spindle density (correlation coefficient of 0.61; p < 0.01), as well as between HFA duration and thalamic spindle duration (correlation coefficient of 0.58; p < 0.01) (Fig. 4).

Fig. 4. Characteristics of Spindles and HFA:

Within each thalamic electrode, we identified the thalamic channels that detected the highest number of spindles. Each channel was classified into two groups based on the aforementioned criteria: HFA(+) (n = 10) and HFA(−) (n = 9) channels. Panels in (A) denote characteristics of spindles detected in the HFA(+) and HFA(−) groups. Note that HFA(+) channels had higher spindle density and a lower frequency band of spindles. Panel (B) denotes the peak latency of the averaged HFA (200–500 Hz) and sigma band activities (11–16 Hz) among HFA(+) channels. Peak latencies were earlier in HFA than in the sigma band. Scatterplots in panel (C) show a positive correlation between HFA duration and spindle density, and in panel (D) show a positive correlation between HFA duration and spindle duration. HFA: high-frequency activity; *: p < 0.05.

3.4. The effect of IEDs on thalamic spindles and HFA

The predominance of the fast ripple band power in HFA led us to investigate the impact of IEDs, such as spk-HFO and spikes, on spindles and HFA. We first investigated the effect of spk-HFO observed on the side of thalamic spindles in the HFA(+) channels. We detected 1131 spindles in the selected thalamic channels, comprising 1001 pure spindles (n = 10 thalamic channels), 54 spindles preceded by IEDs (n = 5 thalamic channels), and 61 spindles co-occurring with IEDs (n = 7 thalamic channels). The duration of HFA was significantly longer in pure spindles compared to spindle preceded by IEDs (F = 5.30, p = 0.015 adjusted p-value = 0.04; Tukey-Kramer Pot Hoc test PS vs SPI: p = 0.02, Cohen’s d 1.47) (Fig. 5). We had similar observations based on the effect of spikes on thalamic spindles and HFA (see Supplementary Fig.1).

Fig. 5. The Effect of Spike-HFOs on Spindles and HFA:

All thalamic spindles were categorized into PS (Pure Spindles), SPI (Spindles Preceded by IEDs), and SOI (Spindle Co-occurring with IEDs) based on the relationship between cortical IEDs and thalamic spindles (A). Cortical or subcortical (hippocampus and amygdala) IEDs ipsilateral to thalamic spindles were used. EEG traces in blue color are thalamic spindles, while in orange represents ipsilateral cortex EEG. Panel (B) shows a comparison of spindle (top) and HFA (bottom) between the three groups, which did not show any statistically significant differences. However, we observed HFA duration was significantly longer in PS than SPI and SOI (F = 6.68, p = 0.047; Tukey-Kramer Pot Hoc test PS vs SPI: p = 0.02; PS vs. SOI p = 0.01). Ctx: cortex; HFA: high-frequency activity; Hipp: hippocampus; IEDs: interictal epileptiform discharges; PS: pure spindles; SPI: spindles preceded by interctal epileptiform discharges; SOI: spindles co-occuring with interictal epileptiform discharges; *: p < 0.05.

3.5. Analysis of spindle co-occurrence in various brain regions

Most of the thalamic spindles co-occurred with at least one cortical or scalp spindle, regardless of the status of HFA (66.7 % [IQR: 41.2–94.0 %] Supplementary Fig.2). Thus, to inspect the hubness of each channel, we computed the percentage of co-occurrence when cortical or scalp spindling (count of co-occurring spindle with the thalamic spindle / cortical or scalp spindle in each location) × 100. In HFA (+) channels, spindles showed a tendency to co-occur in other brain regions compared to HFA (−) channels. Notable differences were observed in the cingulate/insula (median 24.6 % [IQR: 13.6–59.8 %] versus median 4.0 % [IQR: 1.85–8.96 %]; p = 0.016, Cohens’ d = 1.21), parietal lobe (median 39.9 % [IQR: 22.0–44.4 %] versus median 0 % [IQR: 0.0–16.4 %]; p = 0.04. Cohen’s d 1.31), and scalp (median 29.5 % [IQR: 18.4–42.8 %] versus median 1.16 % [IQR: 0.39–14.8 %]; p = 0.048, Cohen’s d 1.07) channels. Furthermore, HFA (+) channels demonstrated a tendency to co-occur with adjacent thalamic channels within the same electrode (median 69.0 % [IQR: 41.1–82.8 %] versus median 25.0 % [IQR: 0.0–58.7 %], p = 0.024, Cohen’s d = 0.74) (Fig. 6). Again, however, none of these p-values remained significant after multiple comparisons due to the limited number of samples (p > 0.05). Additionally, as a post-hoc analysis, we considered anatomical and functional connections of the brain regions to AN and CM. All of the thalamic channels with HFA regardless of the rank of spindle density within the electrode, were included in this study. Specifically, thalamic spindles generated within HFA (+) AN channels showed higher co-occurrence rate than CM in cingulate/insula (median 20.1 % [IQR: 8.0– 48.0 %; n = 4] versus median 14.1 % [IQR: 4.0 % – 27.9 %; n = 4]; p = 0.88; Cohen’s d 0.80), frontal (median 26.7 % [IQR: 6.6–32.4 %; n = 7] versus median 11.8 % [IQR: 3.4–24.5 %; n = 4]; p = 0.41; Cohen’s d 0.85), and parietal area (median 28.1 % [IQR: 0.5–44.4 %; n = 6] versus median 17.5 % [IQR: 2.7–31.0 %; n = 5]; p = 0.58, Cohen’s d 0.75).

Fig. 6. Analysis of spindle co-occurrence in various brain regions:

Panel (A) presents an example of co-occurring and non-co-occurring spindles in different brain regions. The co-occurring spindles were defined as those with the maximum power of the sigma band located within the time window of the spindle seen in the HFA (+) or (−) channel. The upper panels display the HFA(+) channel, while the middle panel shows the cortex channels, which are co-occurring. The bottom panel illustrates two scalp spindles that are co-occurring (left side) and not co-occurring (right side), respectively. Panel (B) shows the co-occurrence ratios in different brain regions. *: p < 0.05 HFA: high-frequency activity.

3.6. PAC between thalamic fast ripple and sigma bands at the thalamus

We examined PAC measured as modulation index between the sigma band phase and HFA using pure spindle in HFA (+) channels (n = 10; Fig. 7, upper panel). There was significant coupling between the sigma band phase and HFA measured as modulation index in 6 out of 10 (60.0 %) HFA (+) channels. Among the significant channels, the peak of modulation index was observed at the median on 11.10 Hz phase (IQR: 11.00–11.42 Hz) and 347.8 Hz (IQR: 334.3–403.7 Hz) power. There were differences in the preferred phase between channels. There were nine subjects that had HFA in the thalamic channels. Within those subjects, five out of nine subjects had at least one channel with significant PAC.

Fig. 7. HFA coupled with sigma band:

Examples of PAC between the HFO (200–500 Hz) amplitude and sigma band (11–16 Hz) phase in the three thalamic channels from different subjects. The upper row are comodulograms, the middle row are results from permutation analysis, and the bottom row are polar plots. Statistical differences were defined based on two hundred permutations using maximum statistics. The polar plots show variable preferred phase in HFA(+) channels. PAC: phase-amplitude coupling; HFA: high-frequency activity.

4. Discussion

Leveraging the unique cohort of ten patients with MRE who had thalamic SEEG sampling within AN and CM, we characterized thalamic HFA associated with thalamic sleep spindles as a summated event of HFO-like activity. Such HFA within the fast ripple range, were exclusively observed in the thalamus and not in other areas. These HFA within the fast ripple band demonstrated a strong and consistent coupling with the sigma band, which falls within the sleep spindle frequency range. This observation provides compelling evidence of locally generated synchronized neurophysiological interactions between HFA and sleep spindles in the thalamus. In fact, thalamic channels associated with HFA during spindles presented higher density and duration of thalamic spindles, and co-occurring in multiple brain areas than those without HFA. Additionally, the presence of accompanying IED had a negative effect on the HFA duration. These findings support thalamic HFA during spindles as a physiological activity. However, we recognize that physiological HFOs or HFA investigated in multiple brain areas were mainly reported in the ripple band range (Frauscher et al., 2018, Norman et al., 2019, Skelin et al., 2021). Thus, further investigations are needed to clarify the role of the different peak frequencies in HFA in the thalamus.

Previous studies have documented the occurrence of HFO band power (150–300 Hz) in patients with essential tremor in the ventral intermediate nucleus (Schnitzler et al., 2018) and with Parkinson’s disease in the subthalamic nucleus (Wang et al., 2014). These studies observed that HFO band power was not linked to tremor syndromes and that HFO power inversely correlated with akinesia/rigidity scores. Consequently, they concluded that thalamic HFO band power was more indicative of physiological rather than pathological processes. These also align with the low-amplitude, high-frequency activity (600 Hz) seen during somatosensory-evoked EEGs, which proposed that the source originated from the thalamus (Curio, 2000). We demonstrated the presence of HFA in the fast ripple range accompanying thalamic spindles by employing time–frequency analysis. This method differs from previous approaches that used the Fourier transform to identify peaks. Our approach enabled us to elucidate the temporal relationship between HFA and thalamic spindles’ peak latency, revealing that HFA might occur earlier than thalamic spindles based on the peak latency difference. Additionally, channels accompanying HFA had higher density and longer duration of thalamic spindles, which is in line with the recent report showing that thalamic spindles in the AN and MD, accompanied by ripples, had a longer duration (Szalárdy et al., 2024). Moreover, the HFA tended to couple with thalamic spindles. Spindles are often generated during the transition from the downstate to the upstate (Mak-McCully et al., 2017) by repetitive activation of low-threshold Ca²⁺ channels by the thalamoreticular nucleus (Bartho et al., 2014). Potentially, these HFA might be physiological activities related to these inhibitory or excitatory postsynaptic potentials at the thalamocortical loop.

Cortical fast ripple band activity is known to be associated with epileptic activity; thus, our study investigated HFA in relation to both IEDs and thalamic sleep spindles, utilizing objective, open-source detectors for HFOs (Zhang et al., 2024) and spindles (Vallat and Walker, 2021). We found that HFA appeared independently from IEDs. HFA’s duration was relatively shorter with the presence of IEDs, suggesting that the robustness of thalamic spindles might be impacted by IEDs, which aligns with previous studies (McLaren et al., 2023, Wodeyar et al., 2024). This phenomenon might be different from the relationship in the cortex where cortical spindles can be induced by IEDs seen in the prefrontal cortex and hippocampus (Dahal et al., 2019, Gelinas et al., 2016, Sákovics et al., 2022). Together, we speculate that thalamic HFA with a fast ripple band may be a physiological activity.

Our findings hold the potential for significant clinical applications. Recognizing HFA associated with thalamic sleep spindles as markers of healthy thalamic activity could aid in distinguishing non-pathological thalamic regions. This, in turn, may guide the identification of alternative targets for therapeutic interventions in addition to the radiological information.

Interpretation of the current results should consider the following aspects of the study. Regarding the study cohort, the sample size was small (n = 10) and involved pediatric patients with heterogeneous epilepsy etiologies. Additionally, our investigation mainly focused on the AN and CM thalamic nuclei with a limited number of samples. As previous studies have emphasized the significance of the pulvinar nucleus in limbic and temporal lobe epilepsy (Wu et al., 2023), it would be of interest to extend the analysis to incorporate the pulvinar nucleus to study the physiological characteristics in each thalamic nucleus. Additionally, our thalamic sampling was predominantly unilateral. Gaining insights from patients with bilateral, symmetric thalamic sampling in the future would be enlightening, particularly in comparing findings based on the side exhibiting greater epileptic activities. Given that our subjects have epilepsy, there’s a possibility of an abnormal thalamocortical network influencing our analysis results. However, conducting this type of analysis on healthy subjects poses significant practical challenges. The recording via macroelectrodes is also a limitation. In general, microelectrodes are considered more robust against volume conduction and more sensitive to fast band activity, such as fast ripples. In the current study, we interpreted the results of averaged time–frequency analysis to eliminate the influence of artifacts. We also attempted to establish the presence of significant sigma-HFA PAC in the thalamus as evidence for the local origin of the fast ripples, which revealed significant coupling in 60 % of the thalamic channels. Further studies using microelectrodes may confirm our findings by increasing the signal-to-noise ratio and assessing its local activity. Finally, thalamic HFA coupled with spindles manifests as repetitive, low-power activities in the fast ripple band, which are not the same as traditional physiological HFOs/HFA in the cortex and thus require careful interpretation.

Our findings suggest the presence of thalamic HFA related to thalamic spindles, which requires further investigation. In future studies, we plan to include more individuals with comprehensive thalamic sampling, as clinically indicated, including the sampling of the pulvinar nucleus (in cases such as temporal, parietal, and occipital epilepsy). Furthermore, we are keen to explore the effect of neuro-stimulation on the thalamic HFA and sleep spindles in the thalamus and scalp. As technology continues to advance and awareness of neuro-modulatory treatment options grows, we anticipate an expanded pool of candidates eligible for thalamic stimulation, which will further our understanding of HFA in the thalamus (Dalic et al., 2022, Nagahama et al., 2021).

Abbreviations:

AN

anterior nucleus of the thalamus

CM

centromedian nucleus of the thalamus

IEDs

interictal epileptiform discharges

IRB

institutional review board

HFOs

High-frequency oscillations

MRE

medication resistant epilepsy

PAC

Phase-Amplitude Coupling

SEEG

stereotactic EEG

HFA

High-Frequency Activity

STE

short-term energy

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.clinph.2025.01.011.