Functional connectivity in self‐limited epilepsy with centrotemporal spikes increases with epilepsy duration and interictal spike exposure

Abstract

Objective

To determine whether persistent exposure to interictal spikes is associated with progressive changes in functional connectivity in children with self‐limited epilepsy with centrotemporal spikes (SeLECTS).

Methods

Connectivity was calculated from electroencephalograms (EEGs) of 68 children with SeLECTS and 65 age‐ and sex‐matched controls using the weighted phase lag index. First, we assessed whether connectivity increased with longer epilepsy duration, categorizing duration based on time since first seizure (less vs. greater than 6 months). Second, in a subset of 19 SeLECTS children with repeated EEGs, we assessed whether trajectory of connectivity over time differed based on persistence versus resolution of spikes. Connectivity during sleep and wakefulness was examined separately.

Results

During sleep, connectivity was lowest in controls, intermediate in patients with recent onset epilepsy, and highest in those with longer epilepsy duration. Elevations in connectivity were greatest within the right occipital region at onset and became more widespread with longer epilepsy duration, especially increasing in the left temporal region. During wakefulness, connectivity in recent onset epilepsy trended lower than controls, whereas longer duration epilepsy trended higher, such that the two epilepsy groups differed significantly from each other but not from controls. In the cohort of children with repeated EEGs, patients with persistent spikes showed increasing connectivity over time in sleep, whereas those with spike resolution demonstrated decreasing connectivity over time. There were no significant changes in awake connectivity over time regardless of spike persistence or resolution.

Significance

Functional connectivity in SeLECTS increases progressively with longer duration of spike exposure, suggesting that ongoing spikes drive neural network alterations. Spikes are a potential treatment target to prevent progressive brain network disruption and preserve cognitive outcomes.

Key points.

• Cross‐sectional study showed a dose‐dependent relationship between SeLECTS duration and functional connectivity.

• Early focal occipital connectivity changes spread to involve broader networks with increasing duration of disease.

• Longitudinal analysis revealed an association between hyperconnectivity and the persistence of spikes over time.

• These findings implicate spikes in neural network abnormalities and offer rationale for exploring their potential as a treatment target.

1. INTRODUCTION

Self‐limited epilepsy with centrotemporal spikes (SeLECTS) is the most common focal epilepsy syndrome of childhood, comprising 15% of pediatric epilepsy cases. SeLECTS is characterized by sleep‐potentiated focal seizures and interictal spikes from the sensorimotor cortices, typically resolving by adolescence. ¹ , ² Despite this mild clinical course, more than half of SeLECTS patients suffer from cognitive impairments; two thirds have attentional deficits and half experience difficulties with language, reading, or both. These challenges affect quality of life. ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ Even after seizures are controlled, cognitive impairments can persist or even emerge. ⁹ Although seizures are infrequent, interictal spikes occur hundreds to thousands of times per night and may contribute to cognitive impairment. ⁶ , ⁹

Evidence suggests spikes disrupt brain networks supporting cognitive function in SeLECTS by altering structural and functional connectivity. ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ Increased frontal and central connectivity is associated with cognitive deficits. ¹⁵ , ¹⁶ Studies pairing functional magnetic resonance imaging (fMRI) with electroencephalography (EEG) show increased connectivity between spike‐generating motor regions and frontal regions immediately after spikes, which correlates with poorer language function. ¹⁷ , ¹⁸ EEG functional connectivity analyses demonstrate that spikes cause immediate increases in connectivity between motor and frontal regions, and show that connectivity remains elevated compared to typically developing children even during spike‐free periods of sleep. ¹⁹ SeLECTS children with frequent spikes have higher connectivity compared to those with rare spikes and to typically developing children. ²⁰ Cross‐sectional fMRI and EEG studies demonstrate that functional connectivity in children with SeLECTS diverges from that seen in typically developing children with increased epilepsy duration ²¹ , ²² , ²³ during wakefulness. Studies of functional connectivity during sleep, the behavioral state most affected by SeLECTS and one critical for memory consolidation and cognition, are limited, as are longitudinal studies ²⁴ , ²⁵ assessing connectivity over time.

We hypothesize that spikes drive functional connectivity changes in SeLECTS, with longer spike exposure leading to greater connectivity deviations. To test this, we conducted a retrospective cohort study analyzing EEGs from children with SeLECTS and age‐/sex‐matched controls. We measured connectivity during spike‐free periods, focusing our analyses on the sleeping state, given prior evidence of hyperconnectivity in SeLECTS during sleep ¹⁹ ; we also present supplementary analyses on connectivity during wakefulness. We conducted two complementary analyses: (1) cross‐sectional analysis comparing connectivity between controls and SeLECTS patients grouped by epilepsy duration to assess whether connectivity deviations increase with disease duration; and (2) longitudinal analysis of SeLECTS patients with two EEGs, some of whom have experienced resolution of spikes, to evaluate whether continued spike exposure drives connectivity over time. To test the hypothesis that connectivity changes begin focally, initially involving spike‐generating motor regions, and then spread to involve more brain regions over time, we analyzed both connectivity between specific electrodes and averaged values across all scalp electrodes. ²⁶ We explored whether age at epilepsy onset, age‐related developmental changes, and seizure frequency moderated our findings. We studied beta band connectivity, as it consistently appears abnormal in SeLECTS, ²⁷ , ²⁸ , ²⁹ , ³⁰ and because it can be modified with repetitive transcranial magnetic stimulation, thus making it a potential target for noninvasive neuromodulation. ³¹

2. MATERIALS AND METHODS

2.1. Inclusion and exclusion criteria

The Stanford University Institutional Review Board approved this study. Patient and control EEGs were identified using Stanford Research Repository tools. We included routine EEGs from Lucile Packard Children's Hospital between 2005 and 2023 in children aged 3–15 years at time of EEG. Some EEGs were included in a previous study. ¹⁹ Exclusion criteria for both groups included prematurity (<35 weeks), abnormal brain MRI, other epilepsy syndromes, neurosurgery, severe brain injury, known genetic disorders, documented intellectual disability, or significant medical conditions (e.g., congenital heart disease, cancer, chronic immunomodulation).

2.1.1. SeLECTS group

Eligible patients met diagnostic criteria for SeLECTS as specified in the 2022 International League Against Epilepsy guidelines, ² with epilepsy onset at between 3 and 15 years of age, typical SeLECTS seizure semiology (hypersalivation, facial or hemibody twitching, or nocturnal tonic–clonic seizures), and evidence of centrotemporal spikes on EEG. We included all initial diagnostic EEGs and follow‐up EEGs, recording both wakefulness and sleep.

2.1.2. Control group

Controls were children who underwent clinical EEGs and who, after evaluation, had no concerns of epilepsy or serious neurological problems (indications included syncope, headache, altered consciousness). Controls were also excluded if they were prescribed antiseizure medications (ASMs) or psychoactive medications.

2.2. EEG acquisition

EEGs were recorded using the Nihon‐Kohden or Natus acquisition systems in a standard 10–20 montage with 19 scalp electrodes at 200 or 500 Hz. All were routine clinical studies of ≤1 h.

2.3. EEG annotation

EEGs were manually annotated by trained students (B.S.G., M.E.V.) and reviewed by a board‐certified neurologist (F.M.B.) to identify spike‐free epochs in wakefulness and sleep (Figure 1). Spike‐free epochs were defined as 500‐ms epochs ≥2 s away from artifacts or spikes, to ensure that raised connectivity associated with spikes did not influence results. For sleep annotations, we avoided periods with vertex waves or spindles. Sleep was defined by loss of the posterior dominant rhythm and the presence of sleep architecture (vertex waves, spindles, K‐complexes). As routine EEGs were used, only stage II sleep was annotated and analyzed. To ensure representative sampling, we chose epochs ≥10 s apart from each other, evenly distributed across the record. Connectivity measures are sensitive to number of epochs included, ³² with reasonable stabilization at 20 epochs. ³³ Spike‐free epochs in sleep in SeLECTS are limited, so to ensure equal matching across conditions, we gathered 20–25 epochs each in wakefulness and sleep.

FIGURE 1.

Creating electroencephalographic epochs for analysis. Annotation of a spike‐free period is shown.

2.4. Clinical factors

We recorded sex and age at EEG and for SeLECTS participants, ASM use at EEG, epilepsy duration (time since initial seizure), and duration of follow‐up.

2.5. Connectivity measures

We measured connectivity using the weighted phase‐lag index (wPLI), a phase‐based connectivity metric robust against the confounding effects of volume conduction in EEG ³⁴ and sensitive to connectivity differences in SeLECTS. ¹⁹ wPLI decomposes EEG signals into the frequency band of interest and then compares the phase of waveforms across brain regions; nonzero phase lags are considered true connectivity. ³⁴ The beta band has been consistently reported as abnormal in SeLECTS ¹⁹ , ²³ , ²⁷ , ²⁸ , ²⁹ , ³⁰ and can be reliably estimated from short epochs. EEG was band pass filtered between 12 and 30 Hz using a zero‐phase, acausal filter (firwin design, Hamming window from the MNE‐Python library) ³⁵ to extract the beta band, downsampled to 200 Hz, and Hilbert transformed. We calculated wPLI between each of the 171 unique electrode pairs for every epoch using custom Python code and averaged across epochs. ³² , ³⁴ , ³⁶ Separate connectivity values were calculated for the awake and asleep states. Data were prepared for statistical analysis using the Polars ³⁷ Python software package facilitated by the PyArrow library ³⁸ , ³⁹ for efficient organization and storage of connectivity matrices.

We considered connectivity at three levels of spatial specificity, as averaging values improves stability of connectivity estimates in pediatric data, and use of multiple measures allows for evaluation of both local and widespread patterns of connectivity. ²⁶ Pairwise connectivity is wPLI between each of the 171 unique electrode pairs, offering the most topographic precision. Average connectivity considers each electrode as a node, averaging pairwise connectivity values between one electrode and the other 18 electrodes, yielding 19 values per subject. Global connectivity is the average of all pairwise connectivity values, giving one connectivity value per subject and thereby enabling comparison of widespread connectivity changes.

2.6. Data analyses

We conducted two analyses: (1) cross‐sectional analysis comparing connectivity between controls and SeLECTS patients grouped by epilepsy duration (<6 months vs. >6 months since first seizure), including only SeLECTS EEGs with spikes; and (2) longitudinal analysis of SeLECTS patients with two EEGs, categorized as Spikes‐Persist (spikes on both EEGs) or Spikes‐Resolve (spikes on first EEG only).

2.6.1. Cross‐sectional analysis of epilepsy duration

To assess whether connectivity changed with disease duration, we split EEGs into three groups: controls, SeLECTS‐Short (EEG recorded <6 months after first seizure), and SeLECTS‐Long (EEG >6 months after first seizure). Prior studies use different cutoffs for new onset epilepsy, ²² , ²³ , ⁴⁰ ranging from a few days to 12 months. ²³ The duration of seizures in SeLECTS is typically 24–48 months, ⁴¹ but children also have spikes preceding seizure onset by months to years. ⁴¹ , ⁴² We chose a 6‐month cutoff as it represents a clinically identifiable window for early intervention after epilepsy diagnosis, while acknowledging that it does not account for total duration of exposure to spikes, as onset of this is unknown for most patients.

For SeLECTS, only EEGs with spikes were included to ensure the epilepsy had not yet resolved. We compared controls, SeLECTS‐Short, and SeLECTS‐Long in pairwise, average, and global connectivity, with primary analyses focusing on sleep and supplementary analyses focusing on wakefulness. To account for ASM effects, we conducted sensitivity analyses comparing controls to unmedicated SeLECTS patients.

2.6.2. Longitudinal analysis of spike persistence

To investigate whether spikes contributed to connectivity, we conducted a longitudinal analysis of children with SeLECTS who had two EEGs. All included children had spikes on their first EEG. Children with spikes on their second EEG were categorized as Spikes‐Persist, whereas those without spikes at second EEG were categorized as Spikes‐Resolve. Within‐group comparisons assessed whether connectivity (pairwise, average, and global) changed over time, whereas between‐group comparisons assessed whether groups differed from each other at either EEG. Primary analyses focused on sleep and supplementary analyses on wakefulness.

2.7. Statistical analysis

Statistics were calculated using Statistical Analysis System OnDemand for Academics. We compared the number of epochs included and EEG quality control measures for the cross‐sectional (analysis of variance [ANOVA]) and longitudinal (repeated measures ANOVA) analyses. Cross‐sectional and longitudinal connectivity analyses were performed using generalized estimating equations (GEEs) with independent correlation matrix to account for repeated measures of SeLECTS subjects who contributed both awake and asleep values, some of whom contributed multiple EEGs. For average and pairwise analyses, we ran separate models for each of the 19 electrodes and 171 electrode pairs respectively and adjusted significance thresholds using Bonferroni correction (p < .0026 for average, p < .0003 for pairwise analyses).

2.7.1. Cross‐sectional analysis

We compared the age and sex of the three groups (Control, SeLECTS‐Short, and SeLECTS‐Long) using ANOVA and chi‐squared, respectively. We compared age at epilepsy onset and ASM use in the two SeLECTS groups using an independent sample t‐test and chi‐squared. To test whether connectivity changed with epilepsy duration, we fit GEE models with connectivity as the dependent variable and group (Control, SeLECTS‐Short, and SeLECTS‐Long), behavioral state (awake, asleep), and group by behavioral state interaction as independent variables, adjusting for age, sex, and ASM use. We included both awake and asleep connectivity data in our GEE models to enhance estimates and stratified analyses by behavioral state. ⁴³ , ⁴⁴ Our primary outcome was group differences in connectivity during sleep, and supplementary analyses assessed global and average connectivity differences during wakefulness. We conducted additional supplementary analyses, focused on global connectivity in the SeLECTS group only, to evaluate whether age at EEG, age at epilepsy onset, or seizure frequency influenced the relationship between epilepsy duration and connectivity (Tables S9–S11).

2.7.2. Longitudinal analysis

We compared Spikes‐Persist and Spikes‐Resolve groups on continuous variables (age at EEG1 and EEG2 and time between two EEGs) using t‐tests and on categorical values (sex, ASM use at EEG1 and EEG2, and ASM changes between the EEGs) using chi‐squared analyses. To test whether spikes were associated with connectivity change, we fit a GEE model with connectivity as the dependent variable and group (Spikes‐Persist, Spikes‐Resolve), EEG (EEG1, EEG2), and the group by EEG interaction as independent variables, adjusting for age, sex, and ASM use. We stratified analyses by group and by EEG, assessing whether there were within‐group connectivity changes between EEG1 and EEG2 or between‐group connectivity differences at EEG1 or EEG2.

3. RESULTS

3.1. Subjects

We reviewed 1019 control charts to identify 65 control children meeting inclusion and exclusion criteria, all with one EEG. All 65 controls had wakefulness on EEG, and 41 also had sleep. We reviewed 397 charts to identify 68 children meeting inclusion and exclusion criteria for SeLECTS, who contributed a total of 91 EEGs. For the cross‐sectional analysis, 14 EEGs without spikes were excluded. Of the 77 remaining EEGs, two captured only sleep, 14 captured only wakefulness, and 61 had both, yielding 75 awake and 63 asleep SeLECTS EEGs for analysis. For longitudinal analyses, we included SeLECTS subjects with two EEGs showing both asleep and awake data, which yielded 19 subjects with 28 EEGs.

3.2. Epoch counts

Counts of epochs extracted for analysis did not differ between groups, or within groups over time, ensuring balanced data for connectivity analyses (Table S1).

3.3. Clinical information

3.3.1. Cross‐sectional analysis of epilepsy duration

Forty‐four SeLECTS EEGs were recorded ≤6 months from first seizure (SeLECTS‐Short), and 33 were recorded >6 months from first seizure (SeLECTS‐Long; Tables 1, 2). Groups were evenly balanced in sex and age at EEG; groups differed in ASM use and age at epilepsy onset. From the time of their first EEGs, SeLECTS patients had a mean duration of follow‐up with neurology of 2.9 ± 2.4 years (range = 0.0–8.8 years).

TABLE 1.

Counts of subjects corresponding to different demographic groups, excluding repeat EEGs without spikes.

Characteristic	Control, n = 65	SeLECTS‐Short, n = 44	SeLECTS‐Long, n = 33	Statistic	p
Sex, n (%)
Female	22 (34%)	14 (32%)	10 (30%)	χ2 = .14	.93
Male	43 (66%)	30 (68%)	23 (70%)
Age at EEG, years, mean ± SD	8.80 ± 2.41	8.32 ± 2.04	9.34 ± 2.03	F = 2.02	.14
Age at onset, years, mean ± SD	–	8.26 ± 2.07	6.44 ± 2.43	F = 12.54	<.01*
Epilepsy duration, years, mean ± SD	–	0.06 ± 0.06	2.90 ± 1.80	F = 110.7	<.01*
Antiseizure medication use, n (%)	–	6 (14%)	21 (64%)	χ2 = 20.7	<.01*

Note: Statistics represent independent t‐test, analysis of variance, and chi‐squared contingency tests between groups.

Abbreviations: EEG, electroencephalogram; SeLECTS, self‐limited epilepsy with centrotemporal spikes.

^* p < .05 (significant difference).

TABLE 2.

Clinical information of Spikes‐Resolve and Spikes‐Persist groups.

Characteristic	Spikes‐Resolve, n = 9	Spikes‐Persist, n = 10	Statistics	p
Sex
Female	4 (44%)	4 (40%)	χ2 = .04	.85
Male	5 (55%)	6 (60%)
Age at EEG1, years, mean ± SD	8.5 ± 2.2	8.1 ± 1.7	t = .41	.69
Age at EEG2, years, mean ± SD	11.3 ± 2.7	9.9 ± 1.7	t = 1.36	.19
Change in age, years, mean ± SD	2.8 ± 1.8	1.8 ± .9	t = −1.5	.14
ASM use at EEG1, n (%)	4 (36%)	3 (30%)	χ2 = .421	.51
ASM use at EEG2, n (%)	7 (64%)	7 (70%)	χ2 = .15	.70
Change in ASM use, n (%)	4 (21%)	3 (15%)	χ2 = .09	.76

Note: Statistics represent independent t‐tests and chi‐squared contingency tests. Change in age is the time between EEG1 and EEG2. Change in ASM use represents the number of children who started using an ASM between EEG1 and EEG2; no child discontinued ASM use.

Abbreviations: ASM, antiseizure medication; EEG, electroencephalography.

Twenty‐four SeLECTS subjects were taking prophylactic daily ASMs at the time of their 27 corresponding EEGs. All were prescribed monotherapy (11 [46%] levetiracetam, 10 [42%] oxcarbazepine, 2 [8%] valproic acid, 1 [4%] sulthiame). ASMs were initiated to achieve seizure control. Children taking ASMs had earlier age at epilepsy onset (6.2 ± 2.2 vs. 8.2 ± 2.2 years, p < .001), suggesting ASM use may mark earlier disease onset rather than greater severity.

3.3.2. Longitudinal analysis of spike persistence

Nineteen SeLECTS subjects were included: nine “Spikes‐Resolve” and 10 “Spikes‐Persist.” Groups were evenly balanced in sex, age at each EEG, change in age between EEGs, ASM use at each EEG, and change in ASM use between EEGs (Table 2). In both groups, some children started and no children stopped ASM use between the EEGs.

3.4. Cross‐sectional analysis of SeLECTS duration

3.4.1. Global connectivity during sleep

Global connectivity was lowest in controls, higher in SeLECTS‐Short, and highest in SeLECTS‐Long groups (Figure 2A, Table S2). Significant differences occurred between controls and both SeLECTS‐Short and SeLECTS‐Long, and between SeLECTS‐Short and SeLECTS‐Long.

FIGURE 2.

Connectivity differences between Control, SeLECTS‐Short, and SeLECTS‐Long groups during sleep. Models of global (A), average (B), and pairwise (C, D) connectivity, adjusted for age, sex, and antiseizure medication use. (A) Least square means of global connectivity and 95% confidence intervals. Significant differences are denoted by an asterisk (p < .05). (B) Forest plots of group differences in average connectivity when comparing SeLECTS‐Short (blue) and SeLECTS‐Long (red) groups to controls. Significant differences between SeLECTS‐Long and Control groups are indicated by an asterisk, and between SeLECTS‐Short and Controls by a dagger (p < .0026). There are no significant connectivity differences between SeLECTS‐Short and SeLECTS‐Long groups. (C) Heat maps of group differences in pairwise connectivity. Red indicates higher connectivity in SeLECTS than controls, whereas blue indicates lower connectivity in SeLECTS. The diagonal divides Control comparison to SeLECTS‐Long (top) and SeLECTS‐Short (bottom). Significant differences are denoted by an asterisk (p < .0003). (D) Topographic map illustrating significant differences in pairwise connectivity between SeLECTS‐Long and controls. Each line represents a significant difference in pairwise connectivity between the two connected electrodes. SeLECTS, self‐limited epilepsy with centrotemporal spikes; wPLI, weighted phase‐lag index.

3.4.2. Average connectivity during sleep

Average connectivity was lowest in controls, higher in SeLECTS‐Short, and highest in SeLECTS‐Long at each electrode (Figure 2B, Table S3). After multiple comparisons correction, SeLECTS‐Long showed higher connectivity than controls at 13 of 19 electrodes (Fp1, F3, F4, F8, C3, C4, P3, P4, Pz, T3, T4, T5, O2). Connectivity in SeLECTS‐Short was significantly higher than in controls at only one electrode (O2) and was significantly lower than SeLECTS‐Long at T3.

3.4.3. Pairwise connectivity during sleep

Pairwise connectivity was significantly higher in SeLECTS‐Long compared to controls between 11 electrode pairs (Figure 2C,D). Electrodes most frequently involved in these pairs were T3, P4, O2, F3, T5, Pz, and F4. Connectivity between left frontal and left temporal regions (F3–T5) was significantly higher in SeLECTS‐Long versus SeLECTS‐Short (mean difference = .056, SE = .015, p = .0003). There were no significant pairwise differences between SeLECTS‐Short and controls.

3.4.4. Awake connectivity

During wakefulness, global connectivity was lowest within SeLECTS‐Short, higher in controls, and highest in SeLECTS‐Long, although only SeLECTS‐Short and SeLECTS‐Long differed significantly. SeLECTS‐Long average connectivity was higher than SeLECTS‐Short at F3, F8, and T5 (Figure 3, Tables S4, S5).

FIGURE 3.

Connectivity differences in Control, SeLECTS‐Short, and SeLECTS‐Long groups during wakefulness. Models of global (A) and average (B) connectivity during wakefulness, adjusted for age, sex, and antiseizure medication use are shown. (A) Least square means of connectivity and 95% confidence intervals. Significant differences are denoted by an asterisk (p < .05). (B) Forest plots of group differences in average connectivity when comparing SeLECTS‐Short (blue) and SeLECTS‐Long (red) groups to controls. Significant differences between SeLECTS‐Short and SeLECTS‐Long are indicated by a double dagger (p < .0026). SeLECTS, self‐limited epilepsy with centrotemporal spikes; wPLI, weighted phase‐lag index.

3.4.5. Connectivity excluding ASM use

When including only SeLECTS children who did not use ASMs, findings were largely stable (Tables S6–S8). Global connectivity was significantly higher in both SeLECTS‐Short and SeLECTS‐Long groups compared to controls in sleep (Tables S6–S8). Average connectivity was significantly higher between SeLECTS‐Short and controls at one electrode (O2), and significantly higher between SeLECTS‐Long and controls at two electrodes (C4, T3). In wakefulness, connectivity was significantly higher in SeLECTS‐Long compared to SeLECTS‐Short at one electrode (F3).

3.4.6. Effects of age, seizure count, and age at onset

Figure S1 demonstrates global connectivity distribution by subject age, and supplementary analyses formally examined whether age, seizure frequency, or age at onset might confound our findings (Tables S9–S11). In linear regression models restricted to the epilepsy group, epilepsy duration remained the strongest predictor, with moderate to strong standardized effects (β = .35–.60 during wakefulness), whereas other potential confounders showed minimal to no effects (β = .12–.20 during wakefulness). When modeled together with duration, neither age (awake: p = .07, asleep: p = .93), seizure count (awake: p = .37, asleep: p = .61), nor age of onset (awake: p = .07, asleep: p = .93) contributed meaningful variance.

3.5. Longitudinal analysis of spike persistence

3.5.1. Global connectivity during sleep

Connectivity in the Spikes‐Persist group increased over time, whereas it decreased in the Spikes‐Resolve group, but these within‐group changes did not reach statistical significance (Figure 4A, Table S12). These divergent connectivity trajectories led to significant group differences over time, however, with the Spikes‐Persist group showing significantly higher connectivity than the Spikes‐Resolve group at EEG2.

FIGURE 4.

Connectivity differences over time between Spikes‐Resolve and Spikes‐Persist groups during sleep (A–F) and wakefulness (G–I). Models of global (A, G), average (B, C, H, I), and pairwise (D–F) connectivity are adjusted for age, sex, and antiseizure medication use. (A, G) Least square means of global connectivity. Error bars represent 95% confidence intervals; significant differences are denoted by an asterisk (p < .05). (B, H) Average connectivity change within groups (EEG2 – EEG1), with 95% confidence intervals. There were no significant changes. (C, I) Average connectivity differences between groups at EEG1 and EEG2. Negative values represent higher connectivity in the Spikes‐Resolve group, whereas a positive value represents higher connectivity in the Spikes‐Persist group. Significant differences at EEG2 are denoted by an asterisk (p < .0026). There were no significant differences at EEG1. (D) Topographic map showing significant pairwise connectivity changes between groups at EEG 1 (green), EEG 2 (red), and within the Spikes‐Persist group from EEG 1 to EEG 2 (blue). (E) Pairwise connectivity changes within groups (EEG2 – EEG1). Blue represents a reduction in connectivity between EEGs, whereas red represents an increase. Significant differences are denoted by an asterisk (p < .0003). The diagonal divides comparisons within the Spikes_Resolve (top) and Spikes‐Persist (bottom) groups. (F) Pairwise connectivity differences between groups at EEG1 and EEG2. Red represents higher connectivity in the Spikes‐Persist group, whereas blue represents higher connectivity in Spikes‐Resolve. Significant differences are denoted by an asterisk (p < .0003). The diagonal divides comparisons between groups at EEG1 (top) and EEG2 (bottom). EEG, electroencephalography; wPLI, weighted phase‐lag index.

3.5.2. Average connectivity during sleep

Connectivity trended up across all but one electrode in Spikes‐Persist and down in all but one electrode in Spikes‐Resolve over time, although differences were not significant (Figure 4B,C, Table S13). There were no significant connectivity differences between Spikes‐Resolve and Spikes‐Persist groups at EEG1, but by EEG2, Spikes‐Persist showed higher connectivity at seven electrodes (F4, F7, Cz, P3, P4, T5, and O2).

3.5.3. Pairwise connectivity during sleep

Pairwise connectivity decreased significantly over time in the Spikes‐Resolve group between Fp2 and C4 (Figure 4D–F). There were no significant changes over time within Spikes‐Persist.

A general pattern of connectivity decrease in Spikes‐Resolve and connectivity increase in Spikes‐Persist was again noted. Left parietal to occipital (P3–O1) connectivity was significantly higher in Spikes‐Persist than Spikes‐Resolve at EEG1, whereas connectivity was significantly higher in four electrode pairs by EEG2 (F4–P3, T3–P4, T4–Cz, and T6–O2).

3.5.4. Awake connectivity

When comparing global connectivity during wakefulness, there were no significant differences within or between Spikes‐Persist and Spikes‐Resolve groups, although generally Spikes‐Persist connectivity was higher. There were no significant differences within or between groups in average connectivity during wakefulness (Figure 4G–I, Tables S14, S15).

4. DISCUSSION

This study provides compelling evidence that functional brain connectivity in children with SeLECTS increases progressively with disease duration and suggests that spikes drive divergent connectivity trajectories over time. Using both a cross‐sectional and longitudinal approach, we demonstrate that children with longer epilepsy duration have greater connectivity compared to controls, and that connectivity changes likely depend on whether spikes persist. These findings support the hypothesis that interictal spikes, traditionally viewed as benign markers of SeLECTS, are a driving force behind network changes that may underlie the cognitive difficulties within SeLECTS.

4.1. Connectivity increases with greater duration of SeLECTS

Our cross‐sectional analyses corroborate a clear dose‐dependent increase in connectivity with longer epilepsy duration ²³ and newly suggest that the most pronounced progression occurs in sleep. This pattern is most obvious when considering global connectivity although also apparent in the channel average and pairwise data. Prior studies using various imaging and neurophysiologic tools have also found progressive changes in connectivity with longer SeLECTS duration, although these changes vary by brain region and modality of study. ²² , ²⁵ , ³⁰ fMRI studies show variable connectivity patterns across brain regions with disease duration, ²¹ , ²² whereas structural connectivity measured by diffusion tensor imaging consistently decreases. ⁴⁰ , ⁴⁵ Consistent with our results, a prior EEG study also found that beta‐band connectivity progressively increases with disease duration during wakefulness. ²³ Our study additionally highlights much stronger connectivity changes during sleep. Notably, in SeLECTS of short duration, we see consistent (though mostly nonsignificant) elevations in sleep connectivity but unchanged or slightly reduced awake connectivity. In children with SeLECTS of longer duration, connectivity is increased in both states, and increases in sleep are of higher magnitude than those in wakefulness. The earlier and larger changes noted in sleep may be due to progressive, excessive thalamocortical connectivity that develops over time in children with SeLECTS. ⁴⁶ Our work provides additional evidence that epilepsy disrupts typical developmental trajectories of connectivity ⁴⁶ and demonstrates that this effect occurs even in SeLECTS, a condition whose perception as benign has historically discouraged longitudinal investigation.

One important consideration is that the children with longer disease duration also had significantly earlier epilepsy onset and were more likely to be treated with ASMs (Table 1). Our supplementary analyses, including sensitivity analyses focused only on unmedicated children and regressions assessing the relative importance of epilepsy onset versus duration, suggest that epilepsy duration is more critical for determining connectivity than these other variables. Together, this suggests that network reorganization accumulates progressively with spike exposure, rather than reflecting an inherently more severe phenotype. Early onset epilepsy does, however, result in both spike and ASM exposure during vulnerable developmental windows, and it is challenging to disentangle the effects of each or to determine whether either causes the cognitive deficits more common in this population. ¹¹ , ⁴⁷ In SeLECTS, spikes and language deficits are likely to precede seizures, ²⁵ , ⁴¹ , ⁴⁸ and future studies should test whether connectivity changes mediate these effects.

4.2. Connectivity changes evolve from focal to global with increasing duration of SeLECTS

Together, the global, average, and pairwise connectivity metrics tell a story of focal changes early in disease that spread over time and topography. We find that the most robust average connectivity differences are initially seen in the right occipital region and then grow over time to involve broader networks. The early occipital changes were unexpected, as we anticipated initial alterations would localize to the centrotemporal regions where spikes originate, although another recent study also reported this pattern of posterior to frontotemporal spatial spread. ²³ These findings may suggest that the occipital cortex is more susceptible to early network disruptions in epilepsy, consistent with prior fMRI evidence of increased activity in the occipital (calcarine) cortices in children with SeLECTS. ²¹ Alternatively, this may represent deviation from an expected developmental trajectory; typically developing children show large age‐related increases in occipitotemporal connectivity over time, whereas children with SeLECTS have a flatter trajectory with increasing age. ⁴⁹ Children with newly diagnosed SeLECTS have also been shown to display localized disruption of cognitive networks, predominantly in the frontal and posterior cingulate cortices, that later become widespread. ¹⁴ Our findings suggest that although initial connectivity alterations may not occur directly at spike foci, they rapidly propagate to distributed networks, consistent with SeLECTS being a network disorder rather than purely focal pathology.

4.3. Hyperconnectivity is associated with persistence of spikes

Spikes are associated with large transient connectivity increases in children with SeLECTS, ¹⁹ , ³³ and fMRI studies have offered evidence that these spike‐associated increases impact broader networks, some of which are involved in language functions, during wakefulness. ¹⁷ , ¹⁸ Whether spikes drive chronic network changes, especially in sleep, however, remains unclear. Our longitudinal analyses show for the first time that hyperconnectivity both increases as spikes persist and decreases once spikes resolve, providing compelling evidence for a mechanistic link between spike activity and network reorganization. Two hypotheses could explain our findings: (1) cumulative exposure to spikes progressively drives hyperconnectivity via activity‐dependent plasticity ⁵⁰ ; or (2) high connectivity represents an inherently permissive brain state facilitating spike generation, with connectivity normalization a prerequisite for spike resolution. Our longitudinal data show that connectivity decreases if spikes resolve and increases if they persist, which is more consistent with the permissive model. A recent fMRI study also suggests reversible network states enable spike generation, demonstrating that increased thalamocortical functional connectivity between the ventrolateral thalamus and inferior motor cortex is present in children with SeLECTS when epilepsy is active (defined by recent seizures) and normalizes after seizure resolution. ⁴⁶ In contrast, our cross‐sectional data support the spike‐driven model, as connectivity increases with disease duration whereas spike frequency decreases with age. ⁵¹ Several other studies also find persistent changes in structural ⁴⁵ and functional connectivity ⁵² after spike resolution, suggesting that permanent changes have occurred. It is possible that both of these explanations are at play, with spikes initially creating reversible functional alterations that over time lead to persistent structural and functional changes. This transition may represent a critical threshold where reversible network states become entrenched developmental alterations, with implications for timing and targets of therapeutic intervention.

4.4. Clinical implications

Our findings provide a neurobiological rationale for using spikes as a treatment target in SeLECTS to offer potential cognitive benefits. Taken together, the evidence linking spikes, connectivity, and cognition supports a clear theoretical progression; spikes cause acute cognitive disruption ¹² and immediate increases in connectivity between epileptic and cognitive networks, ¹⁹ with stronger spike‐induced connectivity correlating with worse language function. ¹⁸ , ⁵³ Our study adds the critical finding that connectivity changes progress with epilepsy duration, specifically in association with persistent spike exposure, creating network alterations even during spike‐free periods. The dose‐dependent relationship between spikes and connectivity supports a causal role for spikes in inducing network abnormalities in SeLECTS. Although the cognitive consequences of these spike‐associated network disruptions remain incompletely understood, evidence demonstrates an association between chronic connectivity changes and lasting cognitive deficits. In children with SeLECTS, disrupted connectivity correlates with poorer language function, ¹⁰ baseline network measures of connectivity predict cognitive outcomes years later, ²⁴ and children with persistent spikes show continued cognitive impairments even after seizure control. ⁹ , ⁵⁴ Whether chronic spike exposure directly interferes with cognition, however, cannot be answered using retrospective data lacking cognitive assessments and instead will require causal methodologies, careful prospective neuropsychological testing, and longitudinal follow‐up.

Another key clinical question is whether interventions targeting spikes and connectivity can prevent or reverse progressive changes and improve cognitive outcomes. Manipulation of spikes is possible with ASMs and noninvasive neurostimulation; levetiracetam reduces spike frequency ⁵⁵ and modifies spike spread to other brain regions ⁵⁶ with potential language benefits, ⁵⁷ and repetitive transcranial magnetic stimulation can reduce both connectivity and spike frequency, ³¹ , ⁵⁷ , ⁵⁸ increase sleep spindle frequency, and improve cognition by some measures. ⁵⁸ These interventional approaches represent important next steps to better characterize how modifying connectivity affects cognitive function and to determine whether spike suppression strategies should specifically focus on influencing network dynamics rather than simply reducing spike frequency.

4.5. Limitations

Several limitations should be considered. First, our use of clinical EEGs meant children's activity was not standardized during recordings, which may have limited findings particularly during wakefulness. ⁵⁹ Second, we limited our analysis to sensor space, which has limited spatial specificity, as we did not have participant MRIs and EEGs were low‐density recordings. Although more nuanced differences may emerge with source space analysis, we still saw large group differences. Third, manual selection of “artifact‐free” EEG epochs without use of automated artifact rejection tools can result in retention of subtle myogenic, cardiac, or ocular artifacts that inflate connectivity values. While epochs were visually artifact‐free and separated by at least 2 s from spikes or transient events, reducing the likelihood of residual artifact or spike‐related volume‐conduction effects on beta‐band connectivity, such influences cannot be entirely excluded. Accordingly, conclusions were drawn based on the most rigorous epoch selection possible with the available clinical EEG data. Future studies employing automated artifact rejection or correction techniques, such as independent component analysis, and unsupervised epoch selection will be important to address this concern and confirm our findings. Fourth, our study may be susceptible to selection bias, as children with longer duration of SeLECTS or those who underwent more than one EEG may have had more severe epilepsy. Similar results in sensitivity analyses excluding children on ASMs is reassuring, as ASM use generally correlates with seizure burden. Fifth, our analysis focused on spike presence versus absence rather than quantifying spike frequency or characteristics. Although this binary approach may be oversimplified, it reflects practical limitations of clinical EEGs, which are of limited duration. Direct comparison of connectivity during sleep and wakefulness, particularly combined with longitudinal quantitative assessment of spike burden, will determine whether connectivity is directly driven by spikes and whether altered sleep connectivity eventually influences wakefulness. Sixth, the 6‐month cutoff, although based on prior literature, may not represent a biologically critical time point. Seventh, in the longitudinal analysis, our modest sample size limits power to detect significant within‐group changes over time, despite clear trends toward divergent trajectories, as well as our ability to explore additional factors that may influence spike persistence and connectivity trajectories. Replication in larger prospective longitudinal cohorts would strengthen confidence in these results and allow for more detailed subgroup analyses.

5. CONCLUSIONS

We demonstrate that functional connectivity in SeLECTS increases progressively with epilepsy duration and spike persistence, with children showing divergent trajectories based on whether spikes resolve or persist over time. These results support the need for future studies assessing the cognitive impact of targeted interventions—including pharmacologic and/or noninvasive neuromodulation—that manipulate spikes and connectivity. This approach would fundamentally shift our therapeutic focus from seizure control alone to actively preventing brain network disruptions, preserving long‐term cognitive outcomes, and enhancing quality of life in children with epilepsy.

AUTHOR CONTRIBUTIONS

Marie E. Vasitas: Investigation (lead); data curation (equal); formal analysis (equal); visualization (supporting); funding acquisition (supporting); writing—original draft preparation (equal); writing—review and editing (equal). Miguel S. Menchaca: Investigation (supporting); formal analysis (equal); visualization (lead); writing—original draft preparation (equal); writing—review and editing (equal). Beatrice S. Goad: Data curation (equal); writing—review and editing (supporting). Xiwei She: Software (supporting); writing—review and editing (supporting). Christopher Lee‐Messer: Software (lead); writing—review and editing (supporting). Zihuai He: Methodology (equal); formal analysis (supporting); writing—review and editing (supporting). Fiona M. Baumer: Conceptualization (lead); supervision (lead); methodology (equal); data curation (supporting); resources (lead); funding acquisition (lead); writing—review and editing (lead); validation (lead).

CONFLICT OF INTEREST STATEMENT

None of the authors have potential conflicts of interest to be disclosed. 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.

DECLARATION OF GENERATIVE AI AND AI‐ASSISTED TECHNOLOGIES IN THE WRITING PROCESS

During the preparation of this work the authors used ChatGPT and Claude.AI to check grammar and provide readability suggestions. After using these tools, the authors reviewed and edited the suggestions as needed and take full responsibility for the content of the publication.

Supporting information

Figure S1.

Tables S1–S15.

DATA AVAILABILITY STATEMENT

The data are clinical data that have not been deidentified and hence are not publicly available, but the code used for the analysis is publicly available and cited in the article. The study was approved by the Stanford University IRB with waiver of patient consent given the retrospective nature of the research; there is no clinical trial registration. There are no reproduced materials.