Paroxysmal slow waves mark ictal networks Authors: Florent J. M. Boyer‐Aymé, Hamza Imtiaz, Ofer Prager, Evyatar Swissa, Alaa Abu Ahmad, Yonatan Serlin, Refat Aboghazleh, Ilan Goldberg, Maayan Ben nun Caller, Idit Tamir, Oded Shor, Ben Whatley, Felix Benninger, Alon Friedman Published: 2026 ABSTRACT AbstractObjectiveEpilepsy diagnosis and treatment monitoring are hindered by the episodic, heterogeneous expression of seizures and by normal‐appearing scalp electroencephalography (EEG) in many patients. We previously described paroxysmal slow‐wave events (PSWEs), brief epochs of broadband slowing detectable on EEG in people with epilepsy. In the present study, we sought to further define the clinical significance of this biomarker.MethodsWe used intracerebral and epidural recordings in a paraoxon rat model of temporal lobe epilepsy, as well as long‐term video‐EEG monitoring (LTM) in patients with temporal lobe epilepsy undergoing intracerebral recordings combined with scalp EEG, or with scalp EEG alone.ResultsWe show that PSWEs arise preferentially in temporo–frontal networks, co‐occur with global slowing, and increase during both spontaneous and pharmacologically induced seizures.Epidurally recorded PSWEs were temporally coupled to deep temporal discharges and were bidirectionally modulated by γ‐aminobutyric acid (GABA)ergic agents (increased with pentylenetetrazol and decreased with pentobarbital).In patients with temporal lobe epilepsy, simultaneous intracerebral recordings and scalp EEG showed that scalp PSWEs mirrored hippocampal spike‐and‐wave activity. PSWEs were more frequent during the preictal and ictal periods than during the interictal baseline, a finding confirmed in a retrospective analysis of 137 seizures from 18 patients recorded with scalp EEG alone.SignificanceThese data indicate that surface PSWEs can index remote epileptiform activity and support their use as a quantitative, noninvasive biomarker for detecting EEG‐silent deep foci and for pharmacodynamic evaluation. Key points Paroxysmal slow‐wave events (PSWEs) are detectable on scalp electroencephalography (EEG) and reflect activity within epileptic temporal‐frontal networks.In animal models, PSWEs are temporally linked to deep temporal discharges and are bidirectionally modulated by pro‐ and anti‐convulsant γ‐aminobutyric acid (GABA)ergic drugs.In patients with temporal lobe epilepsy, scalp PSWEs mirror hippocampal scalp‐and‐wave activity and increase during the preictal and ictal periods.PSWEs may serve as a noninvasive biomarker for detecting deep epileptiform activity and monitoring pharmacodynamic effects. Key points Paroxysmal slow‐wave events (PSWEs) are detectable on scalp electroencephalography (EEG) and reflect activity within epileptic temporal‐frontal networks.In animal models, PSWEs are temporally linked to deep temporal discharges and are bidirectionally modulated by pro‐ and anti‐convulsant γ‐aminobutyric acid (GABA)ergic drugs.In patients with temporal lobe epilepsy, scalp PSWEs mirror hippocampal scalp‐and‐wave activity and increase during the preictal and ictal periods.PSWEs may serve as a noninvasive biomarker for detecting deep epileptiform activity and monitoring pharmacodynamic effects. Paroxysmal slow‐wave events (PSWEs) are detectable on scalp electroencephalography (EEG) and reflect activity within epileptic temporal‐frontal networks. Paroxysmal slow‐wave events (PSWEs) are detectable on scalp electroencephalography (EEG) and reflect activity within epileptic temporal‐frontal networks. In animal models, PSWEs are temporally linked to deep temporal discharges and are bidirectionally modulated by pro‐ and anti‐convulsant γ‐aminobutyric acid (GABA)ergic drugs. In animal models, PSWEs are temporally linked to deep temporal discharges and are bidirectionally modulated by pro‐ and anti‐convulsant γ‐aminobutyric acid (GABA)ergic drugs. In patients with temporal lobe epilepsy, scalp PSWEs mirror hippocampal scalp‐and‐wave activity and increase during the preictal and ictal periods. In patients with temporal lobe epilepsy, scalp PSWEs mirror hippocampal scalp‐and‐wave activity and increase during the preictal and ictal periods. PSWEs may serve as a noninvasive biomarker for detecting deep epileptiform activity and monitoring pharmacodynamic effects. PSWEs may serve as a noninvasive biomarker for detecting deep epileptiform activity and monitoring pharmacodynamic effects. INTRODUCTION Epilepsy stands out as one of the most prevalent chronic brain disorders, characterized by the susceptibility of cortical regions to generate seizures. 1 The diagnosis of epilepsy remains challenging due to the transient nature of seizures, and the heterogeneity in their source and clinical presentation. 2 , 3 , 4 , 5 In addition, monitoring response to pharmacological, neuromodulatory, or surgical therapies currently relies predominantly on patient reports that lack an easily accessible objective measure for their effect on brain epileptiform activity. Currently, electroencephalography (EEG) remains a gold standard for epilepsy diagnosis. 6 However, assessment relies heavily on visual examination of EEG recordings to detect abnormal activity, such as seizures and interictal epileptiform discharges (IEDs). 7 This approach is limited, as EEG studies may appear normal in up to 50% of patients with epilepsy. 8 IEDs can also be present in 10%–12% of individuals without epilepsy, 8 , 9 , 10 , 11 further complicating the accuracy of clinical diagnosis. Consequently, the development of diagnostic and pharmacodynamic biomarkers using quantitative EEG characteristics of the epileptic brain is essential. Enhanced power in slow frequency bands (delta and theta) coupled with reduced power in faster bands (alpha and beta) has been observed in various brain disorders, 12 , 13 , 14 , 15 , 16 , 17 including epilepsy. 13 We recently reported an algorithm for the detection of transient slowing of cortical activity recorded in inter‐ictal EEG recordings from patients with epilepsy, 18 , 19 which we termed “paroxysmal slow‐wave events” (or PSWEs). PSWEs are characterized by a median power frequency (MPF) of 1–6 Hz for ≥5 s (1–5 Hz for ≥10 s in rodents). PSWEs were shown to be a robust discriminator between patients (and rodents) with epilepsy and healthy individuals, with areas under the curve (AUCs) >0.8 in a receiver‐operating characteristic (ROC) analysis. 18 We also showed that PSWEs often co‐localized with blood–brain barrier (BBB) dysfunction and are more common in patients with drug‐resistant epilepsy. 20 In addition, we identified high rates of PSWEs as a predictive marker for epilepsy in patients presenting with a first seizure. 21 A higher percentage of time in PSWEs and lower MPF during PSWEs were also found as predictive biomarkers for posttraumatic epilepsy (PTE) in patients following severe traumatic brain injury (sTBI). 19 EEG slowing in patients with epilepsy was reported as far back as 1946, 22 and later described in additional studies, 23 , 24 as well as in regions adjacent to the seizure‐onset zone. 25 However, the relationship between polymorphic (in contrast to rhythmic) EEG slowing and epileptic activity remains poorly understood. Moreover, although slow waves have been postulated to represent ictal events, 24 others correlated slow activity with a reduction in cerebral blood flow (CBF) or speculated that these events represent post‐ictal surrounding inhibition 26 , 27 involving the recruitment of large populations of inhibitory neurons. 28 , 29 , 30 In the current study, using epidural and intracerebral recordings in a rat model of temporal lobe epilepsy (TLE), we aimed to explore the spatial and temporal characteristics of paroxysmal slowing and its association with epileptiform activity and seizures. To translate these preclinical findings, we analyzed a publicly available EEG database of TLE (the European Epilepsy Database) 31 and evaluated simultaneous scalp and intracerebral EEG recordings from a patient with TLE. MATERIALS AND METHODS Experimental design To examine the effect of epileptogenesis on the occurrence of PSWEs, we first recorded from rats both prior to and 8 weeks after induction of status epilepticus (SE), as well as from sham‐treated control animals (n = 4 per group; timeline shown in Figure S1A). Recordings began 1 week before SE induction and continued for 2 months (Figure 1E; Figure S4). In a separate cohort of animals (n = 10; Figure 1F–G, Figures 2, 3, 4, 5), electrodes were implanted 4–8 weeks after paraoxon exposure (for electrode coordinates see Table S1), at which time animals exhibited spontaneous seizures (see timeline in Figure S1B and 32 , 33 ). To minimize the detection of sleep‐related slowing, PSWE analysis was restricted to data acquired 3–8 h after the onset of the dark cycle. PSWEs in a rat model of temporal lobe epilepsy (TLE). (A) Cresyl‐violet–stained coronal section 4 weeks after paraoxon‐induced status epilepticus (SE) showing neuronal loss in the entorhinal cortex (arrows). (B) Electrode locations. (C) Coronal section showing the electrode tract in right lateral entorhinal cortex (LEC R.; yellow arrow). (D) Epidural recording: Voltage (upper/middle trace) and median power frequency (MPF, lower trace). A PSWE is marked between red dashed lines; inset (upper trace) shows 5‐s segment during a PSWE (asterisk). (E) Because of the low sample size, here, channels (not animals) are individually compared to themselves before and after poisoning. Timeline shown in Figure S1A. PSWE features (pre‐SE vs 3 months post‐SE; epileptic rats, n = 16 channels) vs sham controls (n = 15 channels). (F, G) Here, PSWE metrics are compared in n = 10 animals. Timeline shown in Figure S1B. (F) Left: PSWE occurrence (min−1) in epileptic animals; right: Significance heat map showing FDR‐adjusted q‐values (Benjamini–Hochberg correction of Kruskal–Wallis p‐values for each brain region). (G) Left: Percent time in PSWE in epileptic animals; right: q‐value heatmap (same correction family as in F). Red = q < 0.05 (significant); Blue = q ≥ 0.05 (not significant). PSWE, paroxysmal slow‐wave event; MPF, median power frequency; LEC, lateral entorhinal cortex; SE, status epilepticus; R, right. Inter‐regional characteristics of PSWE occurrence in epileptic rats. (A–G) Timeline shown in Figure S1B. (A, B) Changes in MPF (A) and power (V 2; B) during PSWEs across recording sites. Rows indicate the “remote” channel with PSWE detection; columns show simultaneous locally calculated values in the other channels. * indicates significance (two‐way ANOVA, qmax < 0.0228) between PSWE and non‐PSWE epochs in term of MPF or V2. No significance was found with V2, presumably due to data dispersion, although all channels displayed an increase of power during PSWE in a remote channel. (C) Correlation between MPF and power changes across sites (Pearson, p < 0.0001), both expressed as ratios to baseline. (D–F) Correlograms aligned to PSWE onset in anterior epidural right (Epi. Ant. R.; D), posterior epidural right (Epi. Post. R.; E), and lateral entorhinal cortex right (LEC R.; F). PSWEs were “locally” counted in 60‐s bins from −300 to +300 s, baseline‐corrected to the [−300, −240[s bin. The [0, 60[s bin marks PSWE onset in the “remote” channel. Colored triangles (non‐green) = peak PSWE occurrence before onset in the remote channel (see also Figure S3A–C). (G) Mean PSWE counts in LEC R. relative to baseline [−300, −240[s before [−240, 0[s and after [0, 240[s PSWEs in Epi. Ant. R (Wilcoxon, p < 0.001). PSWE, paroxysmal slow‐wave event; MPF, median power frequency; LEC, lateral entorhinal cortex; Epi., epidural; Ant., anterior; Post., posterior; R., right. Ethics and animal care All experimental procedures in animals were approved by the Animal Care and Use Ethical Committees at the Ben‐Gurion University of the Negev, and were conducted in adherence to the National Institutes of Health (NIH) Guide for the Care Use of Laboratory Animals in Research. TLE model Paraoxon‐induced status epilepticus (SE) was used to trigger TLE, as reported. 32 , 33 In short, Sprague Dawley rats (9–10 weeks old, 300–350 g body weight, provided by Harlan in Israel) were treated with paraoxon (Merck, USA): intramuscular (IM) injections of 0.45 mg/kg, equivalent to a 1.4 times the median lethal dose (LD50) (lethal dose, 50% mortality when no further treatment is provided) dissolved in propylene glycol/saline at a ratio of 1:27.25 (v/v). One minute after paraoxon injection, animals were injected with atropine (Fluka, USA) (IM, 3 mg/kg, in saline) and toxogonin (Merck Serono, Israel) (IM, 20 mg/kg, in saline) to reduce the peripheral effects of the paraoxon and to prevent mortality. Midazolam (Rafa Laboratories, Israel) was administered (IM, 1 mg/kg) 30 min following paraoxon exposure. Electrode implantation in animals Rats were deeply anesthetized using a constant flow of isoflurane (2%) and oxygen; body temperature was maintained throughout the procedure at 37°C ± 0.2°C using a heating pad (Mirom Medical & Research Equipment LTD, Israel). Thin epidural (screw diameter, 1000 μm, Palboreg, Israel) and intracerebral silver (coated diameter, 203.2 μm, A‐M system, USA)/tungsten (coated diameter, 50.8 μm, WireTronic Inc., USA) wire electrodes were implanted using a digital stereotaxic instrument (RWD Life Science). In a subgroup of animals, tungsten electrodes were used. Intracerebral electrodes included the dorsal peduncular (DP) cortex, the most ventral part of the medial prefrontal cortex (mPFC) bilaterally, CA1 region of the right hippocampus (Hipp), right basolateral amygdala (BLA), bilateral lateral entorhinal cortex (LEC), and visual cortex (V1). One anterior and one posterior epidural electrode were also implanted on the right side of the skull. The anterior electrode (Epi. Ant. R.) was inserted above the primary (M1) and secondary (M2) motor areas, whereas the posterior electrode (Epi. Post. R.) recorded signals in the retrosplenial cortex (RSA), as well as in the posteriomedial (PM) and primary (V1) visual areas. A reference was placed over the cerebellum and a ground electrode was inserted in the neck. (For the coordinates used, see Table S1.) Following implantation, animals were returned to their cage, and allowed to recover for 7–14 days before recordings started. Emphasis was given to temporal lobe regions (Hipp, BLA, LEC), considering their association with the ictal‐onset zone and the described histopathology. 34 Histology At the end of the recording period, rats were anesthetized with ketamine/xylazine (i.p., 79.2 and 59.4 mg/kg), followed by intracardiac perfusion with phosphate‐buffered saline (PBS) and 4% paraformaldehyde (PFA). Brains were removed, post‐fixed, and transferred sequentially into 10%, 20%, and 30% sucrose‐PBS (24–72 h each) before freezing and sectioning (20–40 μm using a cryostat). Cresylviolet staining 35 was used to assess damage and verify electrode placement under a light microscope (Axioskope 2plus, Zeiss, Germany). Signal recording and analysis in animals Data were acquired using rodentPACK (Emka Technologies, France) or PowerLab 8/35 combined with a Bio Amp ML138 (ADInstruments, New Zealand) at 1000 Hz and band‐pass filtered (1–45 Hz). Admittedly, the 1 Hz high‐pass filter, originally implemented to reduce sweat‐related EEG artifacts, may affect the low‐frequency signal but was proven efficient for PSWE detection in multiple publications involving more than 1000 EEG recordings. 18 , 19 , 20 , 36 Outputs of the input amplifiers were digitized at 10‐bit resolution for rodentPACK and a 16‐bit resolution for the PowerLab 8/35 (electronic gain: ~41.9 dB and 26.02 dB, respectively). PSWEs were detected using custom MATLAB scripts when the MPF was 1–5 Hz for ≥10 s, as described. 18 , 37 Hence, non‐PSWE epochs included an MPF >5 Hz, but also activities with a MPF of 1–5 Hz, as long as the 10 consecutive second threshold within these frequencies was not reached. Metrics included PSWE occurrence (min−1), time in PSWE (% of recorded time with PSWEs), mean duration, and MPF. Once 10 consecutive seconds with an MPF within 1–5 Hz had been detected, any additional consecutive second within this range was considered part of the same PSWE (e.g., 60 consecutive seconds were counted as a single event). Hence, besides PSWE rate, this method of calculation highlights the importance of additional metrics such as the % time in PSWE to evaluate the impact of these activities on the recording. Seizures and interictal activity were identified by visual inspection (LabChart 8, ADInstruments). Median power frequency MPF was calculated as follows: power spectrum analysis was performed—using the discrete Fourier transform “fft” function in MATLAB—on the band‐pass filtered recording (see Signal recording and analysis in animals and Human recording and analysis sections for more filtering information). The MPF was defined as the frequency at which the total spectral power is equally distributed, with 50% below and 50% above this value. Hence, MPF calculation is a technique of dimensionality reduction providing a global picture of the spectrogram with a single metric. As described, 18 , 37 MPF was calculated in 2‐s‐long signal epochs with 1‐s overlap. This MPF calculation then allowed further characterization of both PSWE and non‐PSWE epochs. Median power frequency was defined as the frequency bin at which the cumulative spectral power reached 50% of the total power within the analyzed epoch; thus, it represents the 50th percentile of the spectral power distribution. MPF and power variation Local changes in channel activity during PSWE in remote brain regions were investigated. For each “local” channel, MPF was calculated and then averaged, either during PSWE and non‐PSWE epochs in specific remote brain regions. MPF variation was then defined as: (1) ΔMPFlocalHz=MPFlocalt=remotePSWE−MPFlocalt=remotenoPSWE where MPFlocalt=remotePSWE and MPFlocalt=remotenoPSWE are the mean MPFs in the “local” channel during PSWE and non‐PSWE activity in the “remote” channel, respectively. Δ power, estimated as mean squared amplitude [V2], was calculated similarly. First, mean‐squared amplitude was calculated by squaring the voltage values of the 1–45 Hz band‐pass filtered recording. Then, within each segment, percentage change was calculated as: (2) ΔVlocal2%=Vlocal2t=remotePSWE−Vlocal2t=remotenoPSWEVlocal2t=remotenoPSWE×100 For correlation analysis, MPF and power variation were calculated as ratios: (3) ΔrMPFlocal=MPFlocalt=remotePSWE−MPFlocalt=remotenoPSWEMPFlocalt=remotenoPSWE for the MPF and (4) ΔrVlocal2=Vlocal2t=remotePSWE−Vlocal2t=remotenoPSWEVlocal2t=remotenoPSWE for the power. To conduct the Pearson analysis of Figure 2C, mean ΔrMPF as well as mean ΔrV2 were both calculated for each combination of remote channel and local brain region (in which MPF and voltage were measured in parallel with PSWE detection or when no PSWE was recorded in the remote channel) for each of the 10 animals (timeline shown in Figure S1B) giving a total of more than 500 mean ΔrMPF and mean ΔrV2 coupled values. Because those calculations rely on means rather than events, no animal was overrepresented (e.g., with a higher number of events than average). Correlogram analysis Correlograms were computed with custom MATLAB scripts to examine temporal relationships between PSWEs in different regions, and between seizures and PSWEs. For each PSWE occurring in a remote channel, PSWEs in other channels were counted in 60‐s bins, and rates were normalized to baseline (−300 s and −240 s). For PSWE‐seizure correlograms, seizures with a clear onset (>60 s duration) were identified visually (LabChart), and PSWEs were counted within 10‐s bins around onset. Low‐frequency correlation analysis To assess coupling between surface and deep electrodes during PSWEs, recordings from epidural (Epi. Ant. R.) and entorhinal cortex (LEC R.) electrodes were band‐pass filtered (1–5 Hz). Correlations were computed (MATLAB corrcoef) during epidural PSWE vs non‐PSWE periods. Pharmacological experiments Additional pharmacological experiments were performed in paraoxon‐treated rats with established epilepsy (8–12 weeks post‐injection). Here, montage and recording settings were identical to those shown in Figure S1A but the experimental timeline differed (not shown). Animals received intraperitoneal injections of the pro‐convulsant GABAA receptor antagonist, pentylenetetrazol (PTZ, 10 mg/kg, n = 3) or the anticonvulsant barbiturate pentobarbital (PENTO, 4 mg/kg, n = 3), five times at 30‐min intervals. Intracerebral activity was recorded before and during treatment to assess effects of GABAergic modulation on PSWEs. Animals were killed 20 min after the last injection. To analyze treatment effects, we first constructed correlograms aligned to injection time and quantified PSWE occurrence (Figure 3E,F). For each animal, the pre‐injection baseline interval −1200, 0s was compared with the 0, 1200s interval following each of the five injections. Correlograms were constructed using 60‐s bins. Correlogram data were analyzed using the Friedman test followed by false discovery rate (FDR) correction with the two‐stage step‐up procedure of Benjamini, Krieger, and Yekutieli, allowing comparison of each time bin with the others. PSWE occurrence increases during seizures. (A–D) Timeline shown in Figure S1B. (A) Correlogram of PSWEs and seizures within the same channel. PSWEs were counted in 10‐s bins from −60 s to +60 s; t = 0 marks seizure onset (green line). (B) Mean PSWE counts before [−60s, 0 s[ and after [0 s, +60s[ seizure onset across channels. (C) Percent time in PSWE during entire recording vs during seizures. (D) MPF during PSWEs, seizures, and seizure/PSWE‐free segments. (E, F) Additional group of poisoned epileptic rats with identical montage as in Figure S1A and recording settings but with a different timeline. Here, only epidural electrode results are being shown. Note that these were repeated injections. For each animal (n = 3 rats), the five injection bins are aligned with their unique initial baseline (see Methods). Effect of pentylenetetrazol (E) and pentobarbital (F) on PSWE occurrence. MPF, median power frequency; PSWE, paroxysmal slow‐wave event. For the summed analysis, all baseline bins from −1200, 0s were pooled and compared with the pooled post‐injection bins from each injection window separately (∑ baseline vs ∑ injection 1, ∑ baseline vs ∑ injection 2, etc.). This yielded five comparisons per animal (yielding 15 observations per group). Statistical differences in summed data were assessed using the two‐tailed non‐parametric Wilcoxon test. Human recordings and analysis The retrospective study of anonymized patient data was approved by the Rabin Medical Center institutional review board (IRB). Stereo‐encephalography (sEEG) with simultaneous scalp EEG was performed during routine pre‐surgical evaluation. Five depth macroelectrodes (8–10 contacts, 5‐mm spacing; Ad‐Tech, USA) were stereotactically implanted (CRW frame, Integra, USA), targeting the amygdala, anterior and posterior hippocampus, and posterior temporal and occipitotemporal cortex (for electrode coordinates see Table S2). Planning was based on pre‐operative 3 T magnetic resonance imaging (MRI) (Philips, Netherlands) co‐registered with intraoperative computerized tomography (CT) (O‐arm, Medtronic, Ireland) using Stealth 8 navigation (Medtronic); post‐operative CT confirmed electrode placement. Recordings began 24 h post‐surgery, sampled at 2 kHz, and downsampled to 250 Hz. Scalp EEG (10–20 system, 19 electrodes) was sampled at 250 Hz. Patients pressed a trigger button at seizure onset. Data were preprocessed in EEGLAB 38 as reported previously 18 , 19 , 21 : band‐pass filtered (1–45 Hz), decomposed with independent component analysis (ICA), and artifact components (muscle, eye, heart) removed using ICLabel. 39 Preprocessed EEG was segmented into 2‐s windows (50% overlap) and MPF calculated. PSWEs were defined as an MPF of 1–6 Hz for ≥5 s in ≥2 electrodes. For scalp EEG during long‐term monitoring, data were stored in 1‐h blocks (University of Freiburg SQL servers). The block containing a seizure and the preceding hour were retrieved, merged, and aligned to seizure onset, yielding 60 min preictal and 20 min postictal data. Recordings with ictal activity in the 1‐h period preceding the seizure onset or data gaps >3 s were excluded. Statistical analysis Analyses were performed in GraphPad Prism 9. When PSWE features (occurrence, duration, MPF) failed normality, data were analyzed with the Kruskal–Wallis test. Normally distributed measures (% time in PSWE, MPF during PSWE) were tested with mixed‐effects analysis of variance (ANOVA; Geisser–Greenhouse correction). Correlogram data were analyzed with non‐parametric tests. FDR correction (two‐stage step‐up, Benjamini, Krieger, and Yekutieli) was applied to ANOVA, Kruskal–Wallis, Friedman, and two‐way ANOVA. Q values (q) represent FDR‐adjusted significance with q min/p min = strongest significance and q max/p max = weakest. Because several analyses involved numerous and partially interdependent comparisons, particularly in temporally structured correlogram data, family‐wise correction methods such as Bonferroni were considered overly conservative. We therefore used FDR correction, which better balances sensitivity and false‐positive control in datasets with dependent comparisons, such as neurophysiological recordings. RESULTS PSWE occurrence in epileptic animals differs across brain regions Consistent with prior results, 18 PSWE occurrence, duration, and time in PSWEs increased in epileptic rats vs pre‐SE baseline (Figure 1E; Figure S2A), with no change in controls (PSWE/min, p = 0.0052; percent time in PSWE, p = 0.0006, Wilcoxon two‐tailed tests). PSWEs were detected across all electrodes but occurred more frequently in the right medial prefrontal cortex (mPFC R.) and lateral entorhinal cortex (LEC R.) than in the visual cortex (V1 R.) (1.37 ± 0.16, 1.16 ± 0.43 and 0.25 ± 0.25 PSWE/min, respectively, Kruskal–Wallis p < 0.0001, statistic = 47.37; Figure 1F). Likewise, the percent time in PSWEs was higher in mPFC R. and LEC R. than in V1 R. (73.14% ± 10%, 47.6% ± 22.5%, and 8.01% ± 10.17%, respectively, mixed‐effects analysis p < 0.0005; overall p < 0.0001, F(2.922, 25.46) = 31.30, Geisser–Greenhouse ε = 0.4174; Figure 1G). Similarly, PSWE duration was longer in mPFC R. (32.7 ± 6.7 s) and LEC R. (24.1 ± 8.3 s) than V1 R. (15.02 ± 5.17 s; p < 0.03; Figure S2C). MPF was lower in mPFC R. (2.85 ± 0.3) but not in LEC R. (3.18 ± 0.31), compared to V1 R. (3.55 ± 0.4; p = 0.005; Figure S2D). These findings indicate region‐specific PSWE patterns. For details of the Kruskal–Wallis and mixed‐effects analysis, see the corresponding matrices (#Q, ##Q; Figure 1F,G). In contrast, analysis of the relative change in MPF showed that the drop during PSWE (expressed as percent baseline) was greatest in the epidural electrodes (Epi. Ant. R. and Epi. Post. R., Figure S2E). The largest MPF decrease was observed in the Epi. Ant. R., whereas the smallest decrease was in the LEC L. (−49.23% ± 3.74% vs −29.84% ± 3.19%, respectively; p < 0.0001, mixed‐effects analysis; Figure S2E). Overall, the magnitude of MPF reduction during PSWEs, relative to baseline, differed across electrodes (Figure S2E) and the corresponding z‐scores are shown in Figure S2F. PSWEs are associated with global slowing To test whether PSWE in one region was linked to remote changes, we measured MPF across channels during PSWEs recorded in any given channel. PSWEs in one channel were associated with a slowing (MPF reduction) in others (−1.22 ± 0.8 Hz; Figure 2A; two‐way ANOVA, q max = 0.0228, q min < 0.0001; corresponding normalized z‐score can be found in Figure S2G), and with increased power (94.32% ± 102.5%; Figure 2B). MPF reduction correlated with power increase (r = −0.514, p < 0.0001, Pearson correlation; Figure 2C), suggesting greater network synchronization during PSWEs. Correlogram analysis showed that PSWE rate rose in mPFC R. and LEC R. 60–1 s before detection in the anterior epidural electrode (Figure 2D,G; q max = 0.0040, q min < 0.0001, and q max = 0.063 and q min < 0.0001, Friedman test for the mPFC R. and LEC R., respectively). PSWEs in the posterior epidural electrode appeared with a delay relative to the anterior epidural electrode (Figure 2E), indicating propagation. Across channels, PSWE peaks coincided with or followed onset in LEC R. (Figure 2F,G). These findings suggest that PSWEs are associated with ictogenic regions and induce slowing (see Discussion). Occurrence of PSWE increases during seizures Spontaneous seizures were recorded in all paraoxon‐treated rats. Video‐EEG showed behavioral arrest with or without jerks or tonic seizures. Correlogram analysis revealed that PSWE occurrence was higher during seizures than in preictal period (−60 to 0 s; Figure 3A,B; two‐way ANOVA, q max = 0.0315, q min < 0.0001). Across channels, time in PSWE was greater during seizures compared to overall baseline (Figure 3C, q max = 0.0067, q min < 0.0001). MPF was the lowest during a PSWE, higher during seizures, and highest outside PSWEs (Figure 3D). To test whether PSWEs reflect an hyperexcitable state or more inhibitory drive, epileptic rats received the pro‐convulsant GABAergic antagonist PTZ or positive GABAA modulator PENTO. PTZ application concurrently increased PSWE occurrence and triggered epileptiform activity and seizures (2/3 rats after one injection; one rat developed SE after the fifth; Figure 3E). PTZ Friedman test's overall p‐value was < 0.0001 for the correlograms. Summed pre‐injection −1200, 0 time bins had significantly lower total number of PSWEs compared to post‐injection 0, 1200 interval (1 ± 1.46 vs. 7.26 ± 11.37, p < 0.0156 two‐tailed Wilcoxon). PENTO reduced PSWE rates and spontaneous movement (Figure 3F). Friedman test's overall p‐value was < 0.0001 for the full correlograms. Summed pre‐injection −1200, 0 time bins had significantly higher total number of PSWE compared to post‐injection 0, 1200 interval (3.66 ± 4.65 vs 0.86 ± 1.59, p < 0.0078 two‐tailed Wilcoxon). These findings suggest that PSWEs reflect hyperexcitable, hypersynchronized, network states suppressed by GABAA transmission. PSWEs in human EEG recordings To examine PSWEs in human patients, we analyzed retrospective EEG data from 18 individuals with TLE (137 seizures; European database, see Methods). PSWEs were detected during interictal, preictal (~3 min) and post‐onset periods (Figure 4A). The percent time in PSWEs was lower in the interictal interval compared to the preictal interval (61.29% ± 29.98%, 66.03% ± 29.31%, respectively; p = 0.0006, Mann–Whitney U test) and lower in the interictal interval compared to the postictal interval (61.29% ± 29.98%, 66.81% ± 28.94%, respectively; p < 0.0001, Mann–Whitney U test). PSWEs during long‐term scalp EEG in temporal lobe epilepsy. (A) Example from T5 (left posterior temporal) electrode. Upper: Voltage; lower: MPF. PSWEs were automatically detected when MPF was 1–6 Hz for ≥5 s (gray dashed line = 6 Hz threshold; shaded = PSWEs). Left: 7‐min segment beginning 24 min before seizure onset. Right: 6‐min segment spanning ~3 min before and after seizure onset. (B) Quantification across seizures. Time in PSWE and PSWE duration were greater pre‐ and post‐onset vs interictal. The number of simultaneously active channels was highest post‐onset. MPF within PSWEs was lower post‐onset than inter‐ and pre‐onset. Red line: Median. Error bars represent interquartile range from 25% to 75%. **p < 0.001; ***p < 0.0001. PSWE duration in the interictal interval was lower compared to the preictal interval (34.61 ± 20.37, 38.49 ± 19.65 s, respectively; p = 0.0002, Mann–Whitney U test) and in the interictal interval compared to the postictal interval (34.61 ± 20.37, 38.49 ± 19.67 s, respectively; p < 0.0001, Mann–Whitney U test). There were fewer channels detecting PSWEs in the interictal interval compared to the preictal interval (65.72% ± 32.22%, 71.39% ± 30.66%, respectively; p = 0.0004, Mann–Whitney U test) and in the interictal interval compared to the postictal interval (65.72% ± 32.22%, 73.54% ± 30.35%, respectively; p < 0.0001, Mann–Whitney U test). MPF during PSWEs was higher in the interictal interval compared to the postictal interval (2.99 ± 0.71 Hz, 2.67 ± 0.78 Hz, respectively; p < 0.0001, Mann–Whitney U test) and in the preictal interval compared to the postictal interval (2.87 ± 0.85 Hz, 2.67 ± 0.78 Hz, respectively; p < 0.0001, Mann–Whitney U test). PSWEs reflect deep epileptiform activity Rodent and human data indicated that PSWEs are linked to network hyperexcitability and seizures. To test whether surface PSWEs reflect remote epileptiform activity, we averaged spike–wave (SW) discharges during seizures in rats (Figure 5A). SWs consistently originated in the LEC followed by hippocampus and mPFC R. (Figure 5B). Wave components were simultaneously observed in epidural electrodes with inverted polarity, but was absent in V1 R. electrode (Figure S3D). Correlation between LEC R. and ipsilateral epidural signals was stronger and more negative during PSWEs vs background activity (no PSWEs) (r = −0.35 vs r = −0.21, n = 10, p < 0.0001, paired t test; Figure 5B). These findings suggest that PSWEs reflect cortical slowing driven by deep epileptiform discharges. Epidural slowing accompanies remote deep epileptiform activity. (A) Rat example: Seizure in entorhinal cortex with corresponding epidural MPF; red band marks PSWE range (1–5 Hz). (B) Average of 22 slow‐wave patterns: Spike peak appears in LEC R., then hippocampus (Hipp. R.), then mPFC R. Right: Correlation between Epi. Ant. R. and LEC R. shows stronger negative coupling during PSWEs vs non‐PSWE periods (paired t test, p < 0.0001). (C) Patient with right TLE (long‐term monitoring): Hippocampal low‐voltage fast activity precedes clinical onset (“Trigger”, blue); surface EEG shows simultaneous slowing. Seizure evolved to bilateral tonic–clonic with post‐ictal suppression. (D) PSWE occurrence (min−1) and percent time binned around seizure onset for two seizures (s1, s2) show pre‐ictal increases. (E) Scalp topography across four seizures shows higher PSWE occurrence in right temporal electrodes −60 to −20 s before onset. (F) Average of five hippocampal slow‐wave events aligned with slowing in scalp electrodes (F4, Fp2, P10, P4). PSWE, paroxysmal slow‐wave event; MPF, median power frequency; LEC, lateral entorhinal cortex; Hipp, hippocampus; mPFC, medial prefrontal cortex; Epi. Ant. R., right anterior epidural; EEG, electroencephalography; LTM, long‐term monitoring; TLE, temporal lobe epilepsy. We next analyzed simultaneous surface and sEEG recordings from a right‐handed man in his 30s with drug‐resistant TLE (tumor in the right temporal lobe, extending to the temporal horn of the right lateral ventricle, involving the head and body of the hippocampus; Figure S5). Because seizure onset could not be localized using video‐EEG long‐term monitoring, five sEEG electrodes were implanted in the right temporal lobe, specifically in the amygdala, anterior and posterior hippocampus, and posterior temporal cortex and temporo‐occipital junction. sEEG showed seizure onset in the posterior hippocampal electrode (Contacts 1 and 2) with spread to the posterior temporal cortex (Contact 1, Figure 5C). Clinical onset (color distortion, staring, oral automatism) occurred 9 s later (indicated by “Trigger 2” in Figure 5C.2), whereas EEG remained silent until spread. Across four seizures, PSWE occurrence, percentage time in PSWEs, and duration rose in the right temporal electrodes 20–60 s before clinical onset and the appearance of the electrographic seizure on the scalp EEG (Figure 5D,E). As in rats, hippocampal SWs were associated with surface slowing (Figure 5F). DISCUSSION Recent work suggests that paroxysmal slow‐wave events (or PSWEs) may mark epileptic brain states. 18 , 21 , 37 We show that PSWEs preferentially arise in temporo‐frontal networks, increase pre‐ictally, correlate with deep temporal epileptiform activity, and are bidirectionally modulated by GABAergic drugs. Together, these findings support PSWEs as a noninvasive index of ictogenic activity. In rats, PSWEs were most prominent in entorhinal cortex and ventral mPFC, two regions strongly implicated in TLE. Occurrence in one site coincided with global slowing and power increases, consistent with widespread network synchronization. Correlogram analyses further suggested initiation in anterior temporal structures with subsequent neocortical spread (Figure 2D–G, Figure S3A–C). Notably, q values were lowest around time 0 across several channel comparisons, indicating tight temporal coupling between regional events. However, some apparent backward extension of PSWE peaks into earlier pre‐ictal bins may reflect volume conduction or remote contamination of local signals rather than true early onset. This limitation should be considered in future applications of PSWEs for seizure prediction and presurgical localization. In addition, the smaller MPF decrease observed in the LEC and mPFC relative to epidural electrodes (Figure S2E) may reflect lower baseline MPF in these structures, due either to ongoing subthreshold epileptiform activity or to physiological slow oscillations that reduce the dynamic range for further slowing during PSWEs. PSWEs increased during seizures, consistent with a transition to a hyperexcitable state (Figure 3, Figure S4). Several caveats should be acknowledged. In the animal recordings, seizure onset and offset were identified by visual inspection, which may have limited the precision of interictal, preictal, and ictal boundaries. Moreover, spectral leakage during ictal activity could influence estimation of the PSWE frequency band. Nevertheless, the pharmacological experiments provide convergent support for a mechanistic link between PSWEs and network excitability: PTZ increased PSWE occurrence while also provoking epileptiform activity and seizures, whereas pentobarbital reduced PSWEs together with behavioral activity. These data strongly support an association between PSWEs and seizure‐promoting network states. Mechanistically, several observations link PSWEs to excitatory drive from deep foci: (i) PSWEs increased during seizures and SE; (ii) PTZ (GABAA antagonist) increased PSWEs, whereas pentobarbital (positive GABAA modulator) decreased them; (iii) in rats, hippocampal/entorhinal/mPFC SW activity co‐occurred with epidural slowing and stronger negative correlations between deep and surface signals during PSWEs; and (iv) in the sEEG/scalp case, hippocampal fast activity preceded clinical onset, whereas scalp EEG showed synchronized slowing, with PSWEs increasing minutes before scalp ictal patterns emerged. Together, these findings suggest that at least a subset of surface PSWEs reflects the influence of deep epileptiform activity on distributed cortical networks (Figure 6). PSWEs index remote epileptiform activity (schematic). In temporal lobe epilepsy, scalp‐recorded PSWEs reflect hypersynchronous discharges in deep/anterior temporal cortex, remote from recording electrodes. The PSWE algorithm captures epileptiform‐related slowing, including OIRDA and TIRDA. PSWE, paroxysmal slow‐wave event; EEG, electroencephalography; OIRDA, occipital intermittent rhythmic delta activity; TIRDA, temporal intermittent rhythmic delta activity. At the same time, an inhibitory contribution to PSWE generation cannot be excluded. 40 GABAA‐mediated transmission can promote seizure‐like events in the rat hippocampus via back‐propagation from the CA3/hilus to the dentate gyrus. 41 Hypersynchronous GABAergic‐mediated transmission was also described in the BBB breakdown model of epileptogenesis, in which low‐dose application of the GABAergic antagonist bicuculline led to short‐term inhibition of both spontaneous and evoked epileptic paroxysmal activity. 42 Thus, PSWEs may not simply reflect a linear shift toward excitation, but rather a network state in which excitation, inhibition, and synchronization interact. In this framework, GABAergic phase‐coupled networks, such as medial septum‐driven hippocampal theta rhythms, 43 or broader coordination systems such as the claustrum, 44 could contribute to PSWE generation and inter‐regional synchronization. In addition, surround inhibition via interneurons activation and GABAB‐mediated K+ conductance 28 , 29 , 30 , 45 , 46 may also generate slow waves of inhibition, as seen during SW discharges. 40 As such, the remote opening of GABAB channels might be synchronized with increased Ca2+‐dependent K+ conductance within the epileptic foci. 40 Such surround inhibition could explain the remote hypersynchronized slowing we observed. Hence, PSWEs may arise from hypersynchronized network interactions rather than from excitation alone. Far‐field effects also remain a plausible explanation for some PSWEs recorded simultaneously with remote epileptiform activity. In some cases, PSWEs and deep epileptiform discharges may represent the same underlying event recorded at different sites and distorted by the filtering properties of brain tissue. However, several observations argue against a purely passive far‐field explanation. In rats, epidural PSWEs showed inverted polarity relative to distant SWs (Figure 5B) and lacked the spike component. In addition, synchronized activity was absent in V1 R., despite its proximity to the entorhinal electrode (Figure S3B), although differences in electrode geometry may also contribute to this finding. These data suggest that at least some PSWEs are generated through active network processes rather than simple passive conduction. Still, dipole effects, far‐field conduction, and surround inhibition are not mutually exclusive, and all may contribute to PSWE detection at the cortical surface during deep epileptiform activity. Overall, PSWEs likely capture a heterogeneous spectrum of pathological slowing that includes—but is not limited to—intermittent rhythmic delta activity (IRDA, 2–4 Hz). 47 , 48 , 49 , 50 By design, our detector targets prolonged (≥5 s in humans; ≥10 s in rodents) 1–6 Hz events with notable theta content. This profile previously distinguished patients with epilepsy from controls with good diagnostic performance (AUC >0.8). Together with the observation that localization of PSWEs may predict BBB dysfunction 18 and focal structural abnormalities, 36 the present findings support PSWEs as a biomarker for atypical epileptiform activity generated within deep brain structures, specifically within the ictogenic zone. Compared with previous studies, 18 , 37 control rats showed relatively high PSWE rates, possibly reflecting electrode‐related injury or known susceptibility of Sprague Dawley rats to SW activity. 51 Nevertheless, PSWEs increased during epileptogenesis and remained higher in epileptic animals than in controls, supporting their association with disease‐related network dysfunction. The human data mirrored the animal findings. In 18 TLE patients (137 seizures), time in PSWE, duration, and spread were greater pre‐ and post‐onset vs interictal, with MPF reduction post‐onset. In the simultaneous sEEG/scalp case, hippocampal epileptiform activity aligned with scalp slowing, and right temporal scalp PSWEs increased 20–60 s before clinical onset and before clear scalp ictal activity emerged. These findings suggest that PSWEs may provide clinically useful information even when conventional EEG appears relatively unrevealing. This study has several limitations. The simultaneous EEG/scalp analysis was limited to a single patient case, possible drug/state confounds, lack of coverage of extra‐temporal brain regions that may be important for propagation of epileptiform activity (e.g., the thalamus), 52 , 53 and uncertain generalizability to extratemporal epilepsies. Prospective, multi‐center studies with source‐localized PSWEs, thalamic recordings, and standardized pharmacological challenges are warranted, as is testing whether PSWEs can guide imaging in remote areas (specifically in low‐ and middle‐income countries), 36 inform presurgical planning or quantify pharmacodynamic response. Because PSWEs are also elevated in subgroup of patients with Alzheimer's disease, 18 they may also serve as markers of subclinical seizures in neurodegeneration. 54 In conclusion, PSWEs provide a scalable, quantitative, noninvasive marker of deep epileptiform activity and seizure propensity. Used alongside conventional IED detection, PSWE metrics may improve the identification of deep or scalp‐silent epileptic foci, and offer a practical readout for monitoring network excitability in both clinical and research settings. AUTHOR CONTRIBUTIONS A.F. conceived the study and secured funding. F.B.A. and H.I. developed the detection/analysis pipeline. F.B.A., R.A., O.P., E.S., and A.A.A. performed animal experiments. A.F., F.B.A., and H.I. conducted formal analyses and created figures. I.G., M.B.C., I.T., O.S., and F.B. coordinated clinical recruitment and collected and analyzed EEG/sEEG data. A.F. provided supervision. F.B.A., Y.S., B.W., and A.F. drafted the manuscript. All authors revised the manuscript critically for important intellectual content and approved the final version. FUNDING INFORMATION This work was supported by a European Research Area Network Neuron Award (CIHR Award no. NDD 168164) (A.F.), Canadian Institute of Health Research Project Grant (grant number PJT‐180636) (A.F.), Israel Science Foundation (grant No. 2254/20) (A.F. and O.P.), and USA‐Israel Binational Science Foundation grant (A.F. and F.B.). CONFLICT OF INTEREST STATEMENT The authors declare no conflict of interest. ETHICS STATEMENT 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. SOCIAL MEDIA AND ARTICLE PROMOTION Brief paroxysmal slow‐wave events on scalp EEG may provide a noninvasive window into deep epileptic brain activity, helping identify seizure networks and monitor treatment effects in temporal lobe epilepsy. Supporting information Figure S1. Experimental timelines, electrode montages, and histology. (A) Montage and timeline for the epilepsy validation experiment (related to Figure 1): right lateral entorhinal cortex (LEC R), right basolateral amygdala (BLA R), right medial prefrontal cortex (mPFC R), and right anterior epidural electrode (Epi. Ant. R). Status epilepticus (SE) was induced after electrode implantation to assess within‐animal PSWE changes following epileptogenesis. (B) Montage and timeline used in most experiments (related to Figure 1F–G, Figure 2–5): bilateral LEC, visual cortex (V1), mPFC, right anterior and posterior epidural electrodes (Epi. Ant. R, Epi. Post. R), and right hippocampus (Hipp. R). (C) Coronal sections (bright‐field) illustrating representative implant tracks in ventral mPFC, hippocampus, and V1; yellow arrow marks electrode tip. Figure S2. Brain regions differ in PSWE rate and spectral features. (A, B) Validation experiment with channels pooled for statistics: PSWE duration increased only in paraoxon‐poisoned animals (white squares) versus controls (black dots). (C, D) Unipolar recordings as in timeline Figure S1B: heatmaps show significance (#Q, ##Q) from Kruskal–Wallis tests with Benjamini–Krieger–Yekutieli FDR correction; red, q < 0.05; blue, q > 0.05. (C) PSWE duration (s) varies across regions in poisoned epileptic rats. (D) Median power frequency (MPF) during PSWEs varies across regions in the same animals. (E) Percent change of the MPF during PSWE. Heatmaps show significance (###Q) from mixed‐effects analysis; red, q < 0.05; blue, q > 0.05. (F) Z‐score calculated from data in E. (G) Z‐score from data found in Figure 2A. Here, z‐scores were computed as such: data from Figure 2A (MPF variation) was expressed as a percentage compared to baseline activity. Z‐scores were then calculated for each remote/local channel couples. Asterisks (*) represent significativity found in Figure 2A. Figure S3. Statistics for PSWE–PSWE correlograms and lack of coupling with V1 activity. (A) Epi. Ant. R (channel of interest) vs LEC R: the interval − 60 to 0 s differs significantly from all other bins, indicating a surge of PSWEs in LEC R preceding Epi. Ant. R PSWEs (qmin < 0.0001; qmax = 0.0063). (B) Epi. Ant. R vs mPFC R shows a similar pre‐event surge (qmin < 0.0001; qmax = 0.015). (C) Epi. Post. R vs Epi. Ant. R: a pre‐event surge in Epi. Ant. R precedes Epi. Post. R PSWEs (qmin < 0.0001; qmax = 0.0007). Friedman test with Benjamini–Krieger–Yekutieli two‐stage step‐up FDR correction. (D) Full dataset for Figure 5B including V1 shows that V1 activity does not correlate with slow‐wave events in LEC R, Hipp. R, or mPFC R. Figure S4. PSWE occurrence increases during status epilepticus. Rats recorded during paraoxon‐induced SE (n = 5); channels pooled owing to sample size (as in Figure 1E). Depending on normality, paired parametric or nonparametric tests were applied (see Methods). (A) PSWEs per minute increased during SE. (B) Percentage of time in PSWE increased during SE. (C) PSWE duration increased during SE. (D) No. significant change in MPF during SE. Figure S5. Intracerebral sEEG electrode locations in a patient with temporal lobe epilepsy. Fusion of post‐operative CT with pre‐operative MRI showing contacts (colored lines) over tumor, amygdala, and hippocampus. For Surgical procedures, clinical management, and recordings–see Methods. Table S1. Implantation coordinates in animals. Table S2. SEEG implantation coordinates in patient in Figure 5.