Frequency‐specific alterations in brain connectivity induced by pulvinar stimulation

Abstract

Objective

Deep brain stimulation (DBS) is emerging as a promising therapy for patients with drug‐resistant epilepsy, particularly those who are either unsuitable for or unresponsive to resective surgery. The potential benefit of DBS in these patients may stem from its ability to reduce excessive brain functional connectivity (FC). Given that patients undergoing presurgical evaluation in our institution are implanted with stereoelectroencephalographic (SEEG) electrodes in the thalamus, specifically in the pulvinar medialis (PuM), our aim was to investigate the impact of different stimulation frequencies on brain FC. We sought to determine whether specific frequencies were more effective in modulating FC.

Methods

SEEG was used to investigate the effects of PuM stimulation across a broad frequency range (1–200 Hz) in a cohort of 14 patients with drug‐resistant focal epilepsy. FC was assessed using the nonlinear correlation coefficient (h ²) and node strength calculations.

Results

Our findings revealed a reduction in FC at stimulation frequencies of 10 Hz and >90 Hz, contrasting with an increase in FC in the 20–80‐Hz range. This modulation of FC extended beyond the epileptogenic zone, influencing all assessed brain lobes, with the parietal, insular, and subcortical regions particularly affected by high‐frequency stimulation. Within the epileptogenic zone, however, the observed FC changes were notably more complex.

Significance

These results underscore the potential of high‐frequency stimulation to decrease interictal FC in epilepsy patients, although intermediate frequencies may exacerbate it and warrant caution. Crucially, this study highlights the effects of PuM stimulation on FC patterns, supporting the role of high‐frequency thalamic stimulation as a promising DBS parameter for improving epilepsy management strategies.

Key points.

• Stimulation frequencies of 10 Hz and >90 Hz led to a reduction in brain FC, whereas frequencies between 20 and 80 Hz were associated with an increase in FC.

• The decrease in FC observed with high‐frequency stimulation (>90 Hz, up to 150 Hz) extended beyond the epileptogenic zone, significantly affecting all brain lobes, particularly in parietal, insular, and subcortical regions.

• High‐frequency thalamic stimulation shows promise for reducing interictal FC in epilepsy patients, but intermediate frequencies (20–80 Hz) might exacerbate FC, highlighting the need for careful selection of DBS frequency in treatment strategies.

1. INTRODUCTION

Epilepsy is known to affect approximately 1% of the population worldwide. ¹ Although it is often effectively managed with antiseizure medications, approximately 30% of patients prove drug‐resistant. ² For patients with focal drug‐resistant epilepsy, resective surgery is the best option, but only a subset of patients are eligible. ³ Alternative neuromodulation strategies have been developed for these patients, including vagal nerve stimulation, deep brain stimulation (DBS), and intracerebral responsive neurostimulation. ⁴ DBS is increasingly used for drug‐resistant epilepsy, mainly targeting the anterior nuclei of the thalamus (ANT), with 54% of patients being responders after 2 years of follow‐up and approximately 10% being seizure‐free for >2 years. ⁴ , ⁵ Among putative DBS mechanisms of action, equilibration of excitatory/inhibitory neurotransmission, changes in background power spectrum/synchrony, and reduction of interictal epileptiform activity have been hypothesized. ⁴ Because focal epilepsy has been recognized as a condition associated with prominent alterations in brain connectivity and networks, ⁶ , ⁷ the effectiveness of neuromodulation may also be linked to the stimulation‐induced changes in brain functional connectivity (FC). ⁸ , ⁹ A recent study demonstrated that DBS in epilepsy induces changes in resting‐state FC, as observed through functional magnetic resonance imaging (MRI), targeting corticosubcortical networks. ¹⁰ These alterations may partially explain the antiseizure effects of stimulation.

Although the ANT is the only US Food and Drug Administration‐approved target for DBS in epilepsy, several alternative targets have been identified, such as various thalamic subnuclei (e.g., the centromedian for generalized epilepsy, the pulvinar medialis for posterior quadrant focal epilepsy ⁹ , ¹¹ , ¹² ) and the hippocampus. ¹³ Among these, the pulvinar medialis is the largest thalamic nucleus and has extensive neural connections throughout the brain, notably with the amygdala, hippocampus, temporal neocortex, cingulate cortex, and orbitofrontal cortex. ¹⁴ , ¹⁵ The medial pulvinar has been shown to be involved in the propagation of focal seizures of multiple origins, establishing strong corticocortical connectivity during seizures that participates in clinical presentation (e.g., awareness alteration ¹⁶ ) and plays a role in seizure termination. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ The pulvinar medialis could serve as a promising target for DBS in epilepsy, with preliminary studies suggesting beneficial effects from both acute ²² and chronic stimulation. ²³

Although there is consensus among experts to set DBS within the 130–145‐Hz range, ²⁴ the optimal parameters for DBS in epilepsy have not yet been established ⁴ , ²⁵ and are based on empirical data. The impact of stimulation frequency on efficacy is still unclear. Building on studies linking neuromodulation efficacy to changes in FC, we postulated that these alterations, detected through stereoelectroencephalographic (SEEG) electrodes following pulvinar medialis stimulation, could act as markers for determining the optimal stimulation frequency. This approach could help identify the most effective frequency for seizure suppression, paving the way for personalized treatment parameters tailored to individual patients.

To this end, we investigated the influence of pulvinar medialis stimulation frequency on brain FC. The impact on FC was explored across a broad frequency range (1–200 Hz) while keeping the output current and pulse width constant. We hypothesized that certain frequencies could selectively reduce FC, positioning them as potential candidates for optimal parameter settings in chronic DBS for patients with drug‐resistant focal epilepsy.

2. MATERIALS AND METHODS

2.1. Patients

Fourteen consecutive patients who underwent SEEG recordings for the presurgical investigation of drug‐resistant focal epilepsy (Alcis, Besançon, France) and had at least two contacts within the pulvinar medialis were included. Electrodes were stereotactically implanted with robotic assistance (ROSA, Medtech, Zimmer Biomet), and their positions were intraoperatively verified using computed tomography (CT; Moebus, Airo, Stryker). The electrodes used for invasive recordings consisted of 10–18 contacts (2 mm long, .8 mm in diameter, and spaced 1.5 mm apart). Anatomical targets were determined individually for each patient, regardless of this study. The implantation scheme was based on information from the noninvasive evaluation and clinical assumptions regarding the localization of the epileptogenic zone (EZ). In our department, we routinely investigate the thalamus, especially the pulvinar medialis, during SEEG when there is an indication to implant a posterior temporal electrode to explore the planum temporale and Heschl gyrus. Inserting an orthogonally implanted electrode slightly deeper at the planum temporale level to record the pulvinar medialis with the most medial contacts is technically straightforward and has demonstrated safety. ²¹ , ²⁶ , ²⁷ The data utilized in this research were collected after obtaining informed consent from all participants and receiving approval from the local institutional review board Programme d'Accés aux données de Santé (PADS).

An identical stimulation protocol was implemented an average of 7.5 days after implantation.

2.2. Electrical stimulations and recordings

SEEG signals were recorded on a 256‐channel Natus system, sampled at 512 Hz, and captured without digital filtering. For each patient, a series of bipolar biphasic stimulations were performed using a CE‐marked microstimulator (Micromed) between the two deepest contacts of the electrode located in the pulvinar, with these contacts disconnected from the recording system during stimulation.

We used Gardel (https://meg.univ‐amu.fr/doku.php?id=epitools:gardel) ²⁸ to coregister each subject's preimplantation T1‐weighted MRI and postimplantation CT scans. The segmentation and reconstruction of each electrode contact's position were performed automatically. Each SEEG contact was then assigned to a brain region using an automated atlas (Virtual Epileptic Patient [VEP atlas]) for parcellation and gray or white matter classification.²⁹ The neurosurgeon (R.C.) visually examined the coregistration and confirmed that all contacts were correctly located within the pulvinar medialis (Figure 1).

FIGURE 1.

Implantation in the pulvinar medialis. (A) Coregistration of the postimplantation computed tomography with the preimplantation magnetic resonance imaging (MRI) scan allowed determination of the position of the stereoelectroencephalographic (SEEG) electrodes. Both thalami are labeled in purple on the axial MRI view. The cross depicts the third contact of the SEEG electrode (Heschl's gyrus 's electrode) [H] within the lateral part of pulvinar nucleus, the first and second contact being in the pulvinar medialis. (B) Schematic representation of the thalamic nuclei. We show here the location of the pulvinar medialis as compared with other targets of deep brain stimulation in epilepsy (anterior Nuclei [pink], centromedian nucleus [green]). G(Left), B(Low), D(Right), P(Posterior).

The following stimulation protocol was applied to all patients: 20 s of prestimulation activity recording, followed by 20 s of stimulation at n Hz (n ∊ [1, 2, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 130, 150, 200]), and then 20 s of poststimulation activity recording, totaling 1 min (20 s prestimulation—20 s stimulation—20 s poststimulation). A 20‐s "washout" period was applied between each cycle to allow connectivity to return to baseline. This duration was chosen to ensure stable connectivity estimation with h ² and maintain equal periods of interest. ²⁸ , ²⁹ The overall time of stimulation was <25 min, without the setup. We aimed to collect enough data to perform the FC analysis while keeping the experiment as short as possible for the patients. Other stimulation parameters included a fixed output current of 2 mA and a pulse width of 450 μs, similar to Filipescu et al. ²²

AnyWave in‐house software (https://meg.univ‐amu.fr/wiki/AnyWave) ³⁰ was used to visualize and process the SEEG data. Only the contacts within the gray matter were included in the analysis. After visually identifying and removing channels with artifacts, all remaining contacts from both hemispheres were retained for analysis.

2.3. FC analysis

We conducted an FC analysis to study the interdependencies between SEEG bipolar channels using the h ² nonlinear correlation coefficient on broadband SEEG signals (1‐45 Hz). ³¹ , ³² The h ² quantifies the level of dependence of a signal Y on a signal X, with more flexibility than a simple linear regression. This technique is particularly suitable for analyzing electroencephalographic (EEG)/SEEG signals in epilepsy. ⁷ A piecewise linear regression (dividing the data of each window into several segments and fitting a linear regression to each segment) is performed between each pair of signals in a sliding window. The h ² is the coefficient of determination, which measures the goodness of fit of the regression (equivalent to the r ² used in linear regression). The h ² is bounded between 0 (no correlation) and 1 (maximal correlation; Figure 2). The h ² values were computed with a sliding window of 2 s, an overlap of 1 s, and a maximum shift of 100 ms between signals. Stimulated contacts were removed from the analysis. The resulting h ² values were summarized into a connectivity matrix (symmetrizing the matrix using the highest h ² value between two channels), providing a comprehensive view of the connectivity levels between distinct brain areas. We computed the node strength from the connectivity matrix (h ² values) for each anatomical brain region (defined by the VEP atlas), called a node. Strength was calculated for each time window for each connection between nodes and averaged through time to having one value by pairs: (nC: number of connection; nT: number of sliding windows). Then, to determine the node strength, we summed the strength of pairs in the same anatomical region (excluding the connection between nodes corresponding to the same channels) and divided it by the number of connections. Then, we calculated an overall FC metric by averaging all node strengths.

FIGURE 2.

Modification of the baseline functional connectivity due to the pulvinar medialis (PuM) stimulation. (A) X and Y represent signals recorded from two parts of the brain. h ² measures the reduction in variance of the signal Y obtained by considering its amplitude as a function of the amplitude of the signal X. The variance of this perturbation can be estimated by a regression curve. h ² = 0 if there is no relation between X and Y, and if signal Y is completely determined by signal X, then h ² = 1. Here, we wanted to analyze the impact of PuM stimulation across frequencies (electrode shown in orange, contacts 1–2) with h ² as indicator of FC. (B) Paradigm of stimulation. Stimulation of the PuM lasted 20 sm and the h ² analysis was performed 20 s before and after. Different frequencies were applied, keeping the other parameters (i.e., amplitude and pulse width) constant. f, frequency.

In the context of our study, the node strength metrics carry specific significance, evaluating the importance of a particular region within the entire network. The FC analysis was done during the different periods (prestimulation and poststimulation). The stimulation artifact was occasionally visible on multiple channels, even those distant from the stimulated one. We attempted several strategies to remove it, including filtering, subtracting the average artifact, and using independent component analysis, but none was sufficiently effective (see Figure S4). Therefore, to rule out any artificial changes in connectivity due to the artifact, we chose not to analyze the period during stimulation.

We examined differential changes in FC based on the side (ipsilateral or contralateral to the stimulated channel) and the relationship to epileptogenicity (within or outside the EZ). As all seizure onset patterns included discharge > 12 Hz, this latter was defined using the Epileptogenicity Index, which quantifies the ability of a brain region to generate fast activity (>12 Hz) early during the seizure course. It is normalized by its maximal value per seizure across channels and ranges from 0 (no epileptogenicity) to 1 (maximal epileptogenicity). Channels with a value ≥ .4 were defined as belonging to the EZ. ³³

To determine which regions had a significant connectivity variation, we thoroughly compared pre‐ and poststimulation periods on each node strength with a Wilcoxon test using the ranksum function of MATLAB, and the approximate method to compute the z‐value. Given the multiple comparisons, a false discovery rate correction was applied to establish a new threshold corresponding to a corrected p‐value of <.05 to identify nodes with significant changes. The terms connection and disconnection are defined based on the sign of the z‐value; a connection occurs when the poststimulation strength exceeds the prestimulation strength, whereas a disconnection occurs when the poststimulation strength is lower than the prestimulation strength. We analyzed only the regions for which the data from at least two patients were available and sampled by a minimum of 10 contacts. To ensure statistical power, we summarized the regions into five anatomical subdivisions (parietal, temporal, frontal, insula, and subcortical structures). The occipital lobe was not analyzed because of insufficient sampling. All the regions and their corresponding lobes are summarized in Supporting Information (Table S1). Finally, we determined the percentage within a lobe (how many nodes out of the total number) that had a significant connection or disconnection.

To address the limited data available for certain lobes and anatomical subdivisions (specifically the parietal and subcortical structures), we grouped stimulation frequencies into ranges, similar to the approach used by Mina et al. for analyzing the impact of the centromedian nucleus of the thalamus. ³⁴ Additionally, we defined frequency ranges based on our initial analysis of their impact on overall FC. The objective was to explore all frequencies in the range of 1–200 Hz. For the analyses, to ensure statistical power, we opted for defining three distinct frequency groups, as follows:

• Frequencies > 80 Hz: high frequencies;

• Frequencies between 20 and 80 Hz: medium frequencies; and

• Frequencies < 20 Hz: low frequencies.

Finally, we used Brainstorm, an open‐source application for analyzing brain recordings, to visualize the regions, utilizing the VEP atlas for segmentation. ³⁵

2.4. Statistical analyses

To assess the impact of stimulation frequencies on overall FC, we analyzed the variation in node strengths across all frequencies. Changes in FC due to stimulation frequency were calculated using linear mixed models with the "lme4" package ³⁶ of the statistical software R. These calculations were performed for each stimulation frequency and each epoch (pre‐ and poststimulation periods). We included stimulation frequency and period as fixed effects, with patients as a random effect. Post hoc analyses were conducted using Satterthwaite's method with Bonferroni correction. To analyze the impact of specific lobes on FC, we employed a similar mixed model, incorporating frequency range and lobe as fixed effects and patients as a random effect. All statistical analyses were performed in R Studio (v2022.12.0+35).

3. RESULTS

Fourteen adult patients (10 females, four males) with a mean age of 36.7 years at the time of SEEG (SD = 10.9, range = 23–52) were studied. In most patients, the EZ exhibited a distributed network organization, with eight of 14 having epileptogenic regions located within two distinct lobes. Eight patients subsequently underwent resective surgery, resulting in seizure freedom for three. Table 1 provides the main characteristics of the patients. Figure S3 provides the number of contacts/subjects per region.

TABLE 1.

Main characteristics of patients.

Patient	Age at epilepsy onset, years	Age at SEEG, years	Sex	Stimulation timing, days after implantation	PuM stimulation side	EZ localization	Etiology	Surgery	Outcome	Electrodes/contacts, n	Ipsilateral/contralateral contacts, n	EZs/NIs, n
1	10	38	F	7	L	Left temporal plus (temporal lateral + insula)	Hippocampal sclerosis	Yes (tailored resection)	Engel I (3‐year FU)	13/101	88/13	2/99
2	12	31	F	3	L	Left frontobasal	NDT	Yes (lesionectomy)	Engel I (3‐year FU)	15/90	75/15	1/89
3	5	48	F	9	L	Left temporal plus (temporal anterior + insula)	Unknown	No	NA	16/112	98/14	21/91
4	3	31	M	7	L	Left mesiotemporal	Genetic (DEPDC5)	Yes (ATL + VNS)	Engel III (3‐year FU)	12/115	98/17	5/110
5	23	55	F	7	L	Left parietal + posterior insula	Unknown	Yes (DBS pulvinar)	Engel II (2‐year FU)	20/136	104/32	16/120
6	12	25	M	12	L	Left temporal anterior	Unknown	Yes (tailored temporal)	Engel I (3‐year FU)	15/142	120/22	26/116
7	40	50	F	10	R	Right temporal anterior	Vascular	Yes (ATL)	Engel III (3‐year FU)	16/158	129/29	13/145
8	21	26	F	8	R	Left temporal plus (with insula)	Unknown	No	NA	18/150	120/30	21/129
9	4	23	F	3	R	Right temporal anterior	Unknown	Yes (tailored right temporal lobectomy)	Engel IV (2‐year FU)	17/147	121/26	12/135
10	30	52	F	9	R	Bitemporal	Neurocysticercosis	VNS	Engel IV (3‐year FU)	17/112	80/32	15/97
11	21	36	F	1	L/R	Bitemporal	Unknown	No	NA	16/128	90/38 38/90	29/99
12	3	30	M	3	L/R	Left mesiotemporal	Unknown	No	NA	19/195	34/161 161/34	23/172
13	17	27	F	9	L/R	Bitemporal	Unknown	No	NA	19/174	64/110 110/64	16/158
14	12	42	M	13	L/R	Right frontoinsular	Unknown	No	NA	19/184	100/84	10/174

Note: For those with bilateral thalamic implantation, the left and right pulvinar were stimulated in separate sessions.

Abbreviations: ATL, anterior temporal lobectomy; DBS, deep brain stimulation; EZ, epileptogenic zone; F, female; FU, follow‐up; L, left; M, male; NA, not available; NDT, neurodevelopmental tumor; NI, Not Involved (Zone); PuM, pulvinar medialis; R, right; SEEG, stereoelectroencephalography; VNS, vagus nerve stimulation.

3.1. High‐frequency stimulation decreases overall FC

First, we evaluated the effects of different stimulation frequencies on overall FC (Figure 3A). Previous studies have shown that a decrease in FC during temporal lobe seizures is associated with better ictal awareness. ³⁷ Therefore, we tested which frequency of pulvinar stimulation produces this effect, indicated by a reduction in node strength after stimulation. Figure 3B presents overall FC before and after stimulation across the various frequencies applied to the pulvinar. A general decrease in FC was observed at frequencies between 90 and 200 Hz, with significant reductions at 130 and 150 Hz (p < .001). A similar significant decrease was also seen at 10 Hz (p < .001). In contrast, stimulation at medium frequencies (20–80 Hz) led to a consistent increase in FC, with significant changes at 20, 40, 60, and 80 Hz (p < .001). Finally, a significant increase was also observed following 1‐Hz stimulation.

FIGURE 3.

Overall functional connectivity (FC) before (PRE) and after (POST) pulvinar medialis stimulation according to the frequency of stimulation. (A) Illustration of the three distinct analyses of the effects of pulvinar medialis stimulation on the brain FC (overall, ipsilateral, and contralateral). (B) Difference of mean with SD range of overall FC before (PRE) and after (POST) stimulation. Note the increase for low and medium frequencies (1, 20, 40, 60, and 80 Hz) and the decrease for high frequencies (130 and 150 Hz). (C) The same analysis is presented, focusing on electrodes ipsilateral and contralateral to the stimulation, providing insight into the differential effects of the stimulation on each side. Significance levels are denoted as follows: *p < .05, **p < .01, ***p < .001. SEEG, stereoelectroencephalographic.

3.2. Effect of pulvinar stimulation on ipsilateral and contralateral FC

Second, we investigated whether variations in connectivity were more prominent on the ipsilateral or contralateral side of the stimulated pulvinar (Figure 3C). We observed a consistent pattern in both contralateral and ipsilateral regions, mirroring the global effect, although the effects were more pronounced ipsilaterally. Statistically significant changes were limited to the ipsilateral side, with increases following 1‐, 20‐, 30‐, and 60‐Hz stimulation and decreases following 10‐ and 130‐Hz stimulation (p < .001 and p < .05).

3.3. Effect of stimulation on FC within and outside the EZ

Third, we specifically examined the influence of pulvinar medialis stimulation inside the EZ and the non‐EZ. As previously established, the EZ exhibited higher FC during the prestimulation period than the non‐EZ. ⁷ Regarding the EZ, we observed that the patterns of FC changes varied with stimulation frequencies that differed from those seen in the overall FC analysis. In particular, we observed a significant decrease in FC after 10‐, 100‐, and 150‐Hz stimulation and a significant increase after 1‐, 5‐, 20‐, 60‐, and 200‐Hz stimulation (Figure 4A; p < .001). Conversely, the non‐EZ showed FC changes similar to the overall FC analysis. We found significant decreases after 10‐, 90‐, 100‐, 130‐, and 150‐Hz stimulation and significant increases after 1‐, 20‐, 30‐, 40‐, 50‐, 70‐, and 80‐Hz stimulation (Figure 4B; p < .001). FC changes did not differ whether or not the propagation zone included the insuloparietal lobe (see Figure S2).

FIGURE 4.

Impact of pulvinar medialis stimulation on functional connectivity (FC) within the epileptogenic zone (EZ) and non‐EZ (NEZ) networks according to the frequency of stimulation. (A) Difference of mean FC before (PRE) and after (POST) stimulation within the EZ network. (B) Difference of mean FC before and after stimulation within the NEZ network. Significance levels are denoted as follows: *p < .05, ***p < .001. For simplicity, the illustration represents the case of contralateral stimulation. In most cases, stimulation was performed ipsilateral to the EZ, with one patient receiving contralateral stimulation (Patient 8) and four patients undergoing bilateral stimulation.

3.4. Effect on FC according to lobe and stimulation frequency

Lastly, we examined whether pulvinar medialis stimulation had varying effects on FC across different brain lobes (Figure 5A). Within individual lobes, we observed both increases and decreases in connectivity strength, suggesting a differential effect of pulvinar medialis stimulation across brain regions. In terms of disconnection (decreases in node strength), low‐frequency stimulation had a pronounced impact on subcortical structures and the parietal lobe. More than 50% of nodes in the parietal lobe and >30% in subcortical structures showed significant disconnection. Conversely, medium‐frequency stimulation caused a more pronounced decrease in FC within subcortical structures than in the temporal and frontal lobes. High‐frequency stimulation resulted in a significant reduction in connectivity, particularly in subcortical structures (with >50% disconnection, the highest level compared to other regions) and the parietal lobe, mirroring the pattern observed with 10‐Hz stimulation (Figure 5B). Regarding the percentage of connections (increases in node strength), low‐frequency stimulation resulted in a higher percentage of nodes with significant increases in both subcortical structures and the parietal lobes. With medium‐frequency stimulation, subcortical structures were the most responsive, exhibiting the highest percentage of nodes with significant increases, with a median of 75%. Finally, the response of subcortical structures to high‐frequency stimulation was less pronounced than in other regions, with the parietal lobes showing the highest percentage of increase.

FIGURE 5.

Change in connectivity status of brain regions following stimulation. (A) The five brain lobe categories (subcortical structures including thalamus and caudate nucleus). (B) Percentage of nodes exhibiting significant disconnection after stimulation across three frequency ranges: low frequencies (1–10 Hz), medium frequencies (20–80 Hz), and high frequencies (90–200 Hz). (C) Percentage of nodes demonstrating significant connectivity after stimulation across the three frequency ranges. Significance levels are denoted as follows: *p < .05, **p < .01, ***p < .001.

In summary, our findings indicate that deep structures, insula, and parietal regions are particularly sensitive to the frequency of stimulation in the pulvinar medialis (see Supporting Information Table S2).

4. DISCUSSION

In this study, we leveraged the unique opportunity presented by SEEG patients with electrodes implanted within the pulvinar medialis to examine the effects of its stimulation on brain FC. Specifically, we investigated these effects across different stimulation frequencies and analyzed their anatomical distribution. Our key findings are as follows. First, acute stimulation of the pulvinar medialis significantly alters brain FC. Second, these alterations are frequency‐dependent, with frequencies of 90–150 Hz and 10 Hz reducing overall connectivity, whereas frequencies of 20–80 Hz and 200 Hz increased it. Lastly, these changes were observed in both EZs and non‐EZs, with a more pronounced impact in the nonepileptogenic regions, particularly in the parietal lobe and subcortical structures. Within the EZ, however, the observed FC changes are notably more complex.

Our findings of decreased connectivity following high‐frequency stimulation of the pulvinar medialis are consistent with other studies on ANT DBS, which also demonstrated a significant reduction in large‐scale FC. This aligns with the decrease in FC observed during ictal pulvinar medialis stimulation in patients with hippocampal stimulation‐induced seizures. Moreover, the decrease in Fc was shown to be correlated with better ictal awareness. ³⁷ It suggests that closed‐loop DBS may exert a similar desynchronization effect; however, further studies specifically investigating this neuromodulation paradigm are required to confirm this hypothesis. ANT DBS has been shown to desynchronize hippocampal background activity and reduce large‐scale brain FC, as evidenced by SEEG recordings. ¹⁰ Additionally, scalp EEG investigations indicate that stimulation leads to widespread desynchronization, particularly over temporal regions, in patients responding to ANT DBS. ³⁸ Similarly, it has been shown that during the ON period of vagus nerve stimulation, brain FC decreases in responsive patients. ⁸ Thus, it appears that a potential common mechanism of neuromodulation is the decrease or normalization of pathological hyperconnectivity observed in drug‐refractory epilepsy. ⁶

Our study also demonstrated that stimulation frequency critically influences the effects of pulvinar medialis stimulation on brain FC, highlighting the importance of carefully selecting specific frequencies for therapeutic interventions. The limited studies that have examined this aspect of stimulation frequency have consistently reported differential effects across frequencies. For instance, Yu and colleagues found increased spectral coherence between the ipsilateral hippocampus and the ANT at frequencies between 15 and 45 Hz, whereas frequencies > 45 Hz resulted in a decrease. However, frequencies > 60 Hz, commonly used in DBS, were not investigated. ¹⁰ A single‐case study of DBS targeting the centromedian nucleus found that a low frequency (2 Hz) and high frequencies (70, 100, and 150 Hz) desynchronized SEEG interictal epileptic activity in cases of premotor focal cortical dysplasia. In contrast, stimulation at 50 Hz had no effect. ³⁹ Our data further demonstrate that low (1 Hz) and medium frequencies (20–80 Hz) can increase brain FC, whereas 10 Hz and frequencies > 80 Hz lead to a decrease, except for 200 Hz, which seems to increase it. A potential mechanism for this is that medium frequencies enhance thalamic output (increased glutamatergic thalamic cell firing), resulting in a higher average excitatory postsynaptic potential in cortical pyramidal cells, as suggested by a modeling study. Mina et al. ³⁴

Desynchronization at 10 Hz could be explained by an interference between stimulation and the genesis of the alpha rhythm, which is closely linked to oscillations between the pulvinar and the cortex. Stimulation at this frequency possibly disturbs the intrinsic alpha rhythm, disrupting its role in maintaining a synchronized, inhibitory state. ⁴⁰ , ⁴¹

High‐frequency stimulation in the range of 100–150 Hz likely interacts with fast‐spiking interneurons and other networks involved in generating high‐frequency oscillations. ³⁴ These frequencies are known to modulate cortical excitability and may override the intrinsic rhythms of neural populations, leading to a breakdown of synchronized activity. This desynchronization can also be associated with increased network complexity and may reflect a shift toward more dynamic and less predictable neural states.

Regarding the spatial distribution of high‐frequency pulvinar medialis stimulation, we demonstrated its ability to significantly reduce FC in most brain regions, particularly in the parietal and subcortical structures. This spatial pattern may be linked to the pulvinar medialis's extensive structural connectivity profile and its frequent involvement in various EZs. ¹⁴ , ²¹ Interestingly, the changes in FC are more pronounced in areas less affected by ANT DBS, suggesting that pulvinar DBS could be an attractive alternative target to ANT due to its distinct connectivity profile. ⁹ Overall, recent studies suggest that the choice of DBS target should be informed by the specific connectivity profile of the targeted nuclei and the individual patient's epileptogenic network. Notably, pulvinar stimulation induced significant FC changes in both EZs and non‐EZs, with effects being more pronounced in the nonepileptogenic regions. This indicates that DBS may also modulate seizure propagation networks and areas not directly involved in seizure generation. ⁹

Furthermore, the extent to which the degree of focality of the epileptogenic network influences FC patterns remains unclear. Evidence suggests that FC is higher in the EZ than outside, and a widespread FC decrease is observed even in focal epilepsy. ⁶ , ⁴² Therefore, the pattern of FC likely depends on both the extent and location of the EZ. To our knowledge, no studies have analyzed FC during brain stimulation according to the spatial organization of the EZ yet, making this an area ripe for further investigation.

Our study acknowledges several limitations, including the relatively small number of patients, which resulted in undersampling of some regions and limited statistical power to examine the effects on specific brain areas (precluding for instance the analysis of the occipital lobe). A potential approach to overcoming this limitation in future studies on the topic could be to focus on canonical large‐scale functional networks. Additionally, to ensure a tolerable stimulation protocol for patients, we only explored the influence of stimulation frequency without modifying other parameters such as pulse width and amplitude. It was also necessary to control for factors other than stimulation frequency. Despite efforts to mitigate stimulation artifacts using various techniques (filtering, averaging, and independent component analysis), we did not achieve satisfactory results, leading to the exclusion of the stimulation period from the analysis (see Figure S4). Further methodological research is needed to address this issue and validate the analysis of brain FC during stimulation. Moreover, our study focused solely on the immediate poststimulation period; therefore, we did not assess long‐term effects or neuroplasticity induced by DBS, which may differ and contribute to its efficacy. Additionally, our protocol is limited in time (20‐s poststimulation washout), which might introduce cross‐contamination between trials of stimulation. The variations we observed in the effects of pulvinar medialis stimulation from one frequency to another (desynchronization vs. synchronization; Figure 2) suggest limited contamination. Finally, we did not correlate the observed connectivity changes with clinical variables. Future studies examining the link between clinical response and brain connectivity modulation due to DBS would be of great interest.

5. CONCLUSIONS

Our analysis reveals the nuanced effects of pulvinar medialis stimulation on brain FC, which vary with stimulation frequency. High‐frequency stimulation, up to 150 Hz, decreases excessive interictal FC in patients with drug‐resistant focal epilepsy, whereas specific intermediate frequencies enhance connectivity. This effect extends beyond the EZ, influencing all brain lobes, particularly the parietal lobe, in line with the extensive projections of the pulvinar medialis. Importantly, these findings underscore the substantial impact of pulvinar medialis stimulation on FC, supporting the use of high‐frequency thalamic stimulation as an effective DBS parameter for improving epilepsy management strategies.

AUTHOR CONTRIBUTIONS

Romain Carron, Stanislas Lagarde, Fabrice Bartolomei, and Christian‐George Bénar designed the stimulation and analysis protocols. Romain Carron, Emma Acerbo, and Stanislas Lagarde conducted the experiments and collected the data. Romain Carron operated on the patient for depth electrode implantation. Emma Acerbo and Aude Jegou carried out the analysis. Emma Acerbo, Aude Jegou, Stanislas Lagarde, and Romain Carron drafted the manuscript, with all authors contributing to its finalization.

CONFLICT OF INTEREST STATEMENT

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.

Supporting information

Figure S1.

Figure S2.

Figure S3.

Figure S4.

Table S1.

Table S2.

Data S1.

Acerbo E, Jegou A, Lagarde S, Pizzo F, Makhalova J, Trébuchon A, et al. Frequency‐specific alterations in brain connectivity induced by pulvinar stimulation. Epilepsia. 2025;66:2690–2702. 10.1111/epi.18405

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.