Ultrafast oscillations in the human brain and their functional significance Authors: Milan Brázdil, Vojtěch Trávníček, Jonathan Curot, Martin Pail, Monika Służewska‐Niedźwiedź, Robert Roman, Filip Plešinger, Emmanuel Barbeau, Michal Kucewicz, Petr Klimeš, Jan Cimbálník, Pavel Jurák, William C Stacey, Gregory A Worrell Published: 2026 ABSTRACT AbstractObjectiveThe upper frequency limit of human brain activity remains unknown. Using ultrahigh sampling rate (≥20 kHz) intracranial microelectroencephalography, this study aimed to systematically explore and quantitatively characterize brain field oscillations beyond the established high‐frequency oscillation range (>2 kHz), and to determine their relationship to epileptogenic tissue.MethodsWe analyzed intracranial electroencephalographic recordings from 15 patients with drug‐resistant epilepsy, comprising 466 microcontacts and 434 macrocontacts implanted in mesiotemporal structures across four international epilepsy centers. A custom spectrogram‐based detector optimized for ultrahigh frequencies was developed to identify ultrafast oscillations (UFOs; >2 kHz). UFO rates were assessed in epileptic (successfully resected) and nonepileptic hippocampi and compared with established electrophysiological biomarkers, including interictal epileptiform discharges, ripples, and fast ripples. Statistical comparisons were performed using mixed‐effects models to account for intersubject and interelectrode variability.ResultsWe identified a previously undescribed class of short‐duration oscillatory events spanning 2–8 kHz. UFOs occurred at significantly higher rates in epileptic compared with nonepileptic mesiotemporal regions, with the strongest differentiation observed in the 2–3‐kHz band. Two distinct UFO phenotypes were consistently observed: (1) spindlelike, narrow‐band oscillations and (2) sharp‐onset, rapidly decaying bursts. Both forms were prevalent within epileptic hippocampi but were exceedingly rare in nonepileptic structures. UFOs exhibited pronounced temporal intermittency and spatial focality and frequently arose independently of ripples, fast ripples, and very high‐frequency oscillations, indicating that they represent a distinct electrophysiological phenomenon. Macrocontacts detected UFOs only exceptionally, highlighting the necessity of microelectrode recordings to capture activity in this ultrahigh‐frequency regime.SignificanceThese findings substantially extend the known frequency range of human brain field activity and identify ultrafast oscillations as a novel biomarker of neuronal hyperexcitability. UFOs likely reflect pathological microcircuit dynamics inaccessible to conventional clinical recordings and provide new insights into the organization and pathophysiology of epileptogenic networks.Keywords: epilepsy, high‐frequency oscillations (HFOs), intracranial EEG (iEEG), microelectrodes, neuronal hyperexcitability, ultrafast oscillations (UFOs) Key points Using intracranial microEEG recordings in a cohort of epileptic patients, novel short‐lasting oscillations in frequencies between 2 and 8 kHz were identified. Using intracranial microEEG recordings in a cohort of epileptic patients, novel short‐lasting oscillations in frequencies between 2 and 8 kHz were identified. A maximum of these ultrafast micro‐oscillations was observed here in the frequency range of 2–5 kHz, but in four subjects they reached up to 7 and 8 kHz, respectively. A maximum of these ultrafast micro‐oscillations was observed here in the frequency range of 2–5 kHz, but in four subjects they reached up to 7 and 8 kHz, respectively. UFO are more frequent in epileptic hippocampi and are rare in nonepileptic brain tissue. UFO are more frequent in epileptic hippocampi and are rare in nonepileptic brain tissue. INTRODUCTION Using direct electrical signal recordings from the human brain via intracranial electrodes, a wide range of frequencies spanning from ultraslow (<.5 Hz) to very fast oscillations (500–2000 Hz) have been repeatedly detected. 1 , 2 All these electrophysiological phenomena can be observed in both healthy and functionally impaired brain structures using standard clinical macroelectrodes (1–10 mm2) to record local field potentials that reflected the summated synaptic and action potential activity, commonly called intracranial electroencephalography (iEEG). 3 In parallel, some studies with clinical macroelectrodes suggest that the human brain can generate even higher frequencies (>2 kHz). 4 , 5 , 6 If this is true, then EEG microrecordings, which remove volume averaging, should be a more potent tool to probe for these ultrafast frequency activities, as shown previously. 1 , 7 This study aimed to analyze resting‐state iEEG data from intrahippocampal hybrid (macro/micro) depth electrodes in a cohort of patients with epilepsy using high sampling frequencies (25, 30, or 32 kHz). Here, we focused on EEG activities between 2 and 8 kHz. Hypothetical ultrafast electrical events within the human brain might bear distinct functional significance when compared to lower frequencies and possibly could substantially contribute to better insight into the pathophysiology of epileptic disorders on the microscopic level. MATERIALS AND METHODS Data recorded in 15 subjects investigated in four independent epilepsy monitoring units were analyzed. For each patient, a 10‐min artifact‐free segment of iEEG recorded during resting state was manually identified and selected. See Table 1 for detailed information for each patient. TABLE 1Patients' demographic and clinical characteristics.Article subject numberSexAge at seizure onset, yearsAge at SEEGSeizure frequency per monthMRI (signs of)Implanted hybrid electrodesType and side of epilepsySOZSurgery/histopathologyEngel outcome (years)Total microcontactsMicrocontacts in epileptic/nonepileptic hippocampus1F20492NormalR TT/RRight hippocampusR AMTR/gliosisIA (5)66/epileptic2M222730+Posttraumatic changes, left lateral TL TT/LL lateral T neocortexL lateral T lesionectomy/posttraumatic changesID (4)66/nonepileptic3M33534L HSL TT/LL hippocampusL AMTR/HSID (3)66/epileptic4F3281Bilateral HSR T, L TT/bilateralBilateral; L and R hippocampusNA34NA5M20398NormalL TT/LL hippocampusL AMTR/gliosisIA (7)2416/epileptic6M39474NormalR T, L TT/bilateralBilateral; L and R hippocampusNA28NA7F1224L HSL T, R TT/LL hippocampusL AMTR/HSIC (8)348/nonepileptic8F16356NormalR TT/R + E/LR hippocampus + L extratemporalNA63NA9F91915Postoperative changes, R TR T, L TT/RR hippocampus, R T pole + R entorhinal cortexR AMTR/HSIB (5)35NA10M26464–7NormalR T, L TT/LL anterior and medial T regionsL AMTR/HS + gliosisIA (2)36NA11F33495–7R HS + posttraumatic changes, R TR T, L TT/RR hippocampus and amygdalaR AMTR/HSIA (4)3212/epileptic12M10276Subtle corticosubcortical blurring within T poles bilaterallyL T, R TFr/RR anterior cingulateCortical resection/FCDIA (2)3216/nonepileptic13F112415Corticosubcortical blurring, R temporoparietalR TT/RR amygdala + R entorhinal ctxCortical resection/FCD IIBIB (1)268/epileptic14F63230+Corticosubcortical blurring, L cingulate gyrusL TFr/LLeft midcingulate ctxCortical resection/FCD IIBIA (1)48NA15M112810–12Corticosubcortical blurring, L temporoparietalL TT/LL hippocampus and amygdalaAMTR/FCD IIAIB (1)5624/epilepticAbbreviations: AMTR, anteromesial temporal resection; ctx, cortex; E, extratemporal; F, female; FCD, focal cortical dysplasia; Fr, frontal; HS, hippocampal sclerosis; L, left; M, male; MRI, magnetic resonance imaging; NA, not applicable; R, right; SEEG, stereoelectroencephalography; SOZ, seizure onset zone; T, temporal. Patients' demographic and clinical characteristics. Abbreviations: AMTR, anteromesial temporal resection; ctx, cortex; E, extratemporal; F, female; FCD, focal cortical dysplasia; Fr, frontal; HS, hippocampal sclerosis; L, left; M, male; MRI, magnetic resonance imaging; NA, not applicable; R, right; SEEG, stereoelectroencephalography; SOZ, seizure onset zone; T, temporal. Patients Fifteen consecutive patients (eight females, seven males) with drug‐resistant focal temporal or frontal lobe epilepsy (Commission on Classification and Terminology of ILAE, 1989) who underwent presurgical evaluation using intracranial stereo‐EEG (SEEG) recordings with at least one hybrid depth electrode gave consent and were included in the study. All of them suffered from focal impaired awareness seizures, and two of them additionally from focal to bilateral tonic–clonic seizures. Five patients were recorded at Mayo Clinic, Rochester, Minnesota, United States; four patients at University Hospital Center of Toulouse (CHU Toulouse), Toulouse, France; three patients at University Hospital, Wrocław Medical University, Wrocław, Poland; and three patients at Brno Epilepsy Center, Brno, Czech Republic. Subjects ranged in age from 19 to 53 years (mean age = 35 years, SD = 11). The main demographic and clinical characteristics of the involved subjects are shown in Table 1. For the comparison between epileptic and nonepileptic hippocampi, inclusion criteria were as follows: (1) seizure‐free outcome after surgery (Engel class I) and (2) the presence of a microelectrode located either within the resected (epileptic) or the nonresected (nonepileptic) hippocampus. All patients underwent a comprehensive presurgical evaluation, including a detailed history and neurological examination, brain magnetic resonance imaging (MRI), neuropsychological testing, and scalp and invasive video‐EEG monitoring. The duration of clinical monitoring and the location and number of implanted electrodes were determined per clinical considerations. After invasive video‐EEG monitoring, 12 patients underwent surgical intervention, details of which are shown in Table 1. The follow‐up interval after epilepsy surgery was at least 1 year. The study was approved by the ethics committees of all participating institutions. All patients signed an informed consent form. SEEG recordings At Mayo Clinic, custom hybrid depth electrode designs (AD‐Tech Medical Instrument Corporation) were used. These electrodes are based on standard four‐ and eight‐contact clinical depth electrodes. The depth electrodes consist of a 1.3‐mm‐diameter polyurethane shaft with platinum/iridium (Pt/Ir) clinical macroelectrode contacts; each clinical contact is 2.3 mm long with either a 5‐ or 10‐mm center‐to‐center spacing (surface area = 9.4 mm2). In addition to the standard clinical contacts, the eight‐contact depth electrode has 8–24 microcontacts (1–4 macrocontacts), and the four‐contact depth electrode has 16 microcontacts, positioned radially along the shaft of the depth electrode. Each electrode has a bundle of nine Pt/Ir microwires that protrude 1–3 mm from the shaft tip. The microwires are made of insulated Pt/Ir wire, 40 μm in diameter. The surface of the microelectrodes is 10−3 mm2 with .5–1‐mm spacing. The patients received one hybrid SEEG electrode in the temporal lobe bilaterally, and in one patient subdural strips and other intracerebral electrodes were implanted for monitoring of the parietal and occipital lobe. The electrode position within the brain was verified using computed tomography (CT) with electrodes in situ. A clinical acquisition system (Neuralynx Inc.) was used to acquire iEEG at Mayo Clinic. The signals were acquired using a wideband direct current (DC)‐coupled analog front end (passband approximately DC to 9 kHz) and digitized at 32 kHz. Macro‐EEG was subsequently filtered by Barlett–Hanning 1‐kHz low‐pass filter and downsampled to 5‐kHz sampling frequency. The stainless steel suture placed in the vertex region of the scalp midline between the international 10–20 Cz and Fz electrode positions was used as a common reference. In Brno Epilepsy Center, aside standard depth multicontact platinum electrodes with a diameter of .8 mm, a macrocontact length of 2 mm, and an intercontact distance of 1.5 mm (Microdeep, DIXI Medical or Depth Electrodes Range, ALCIS), one custom hybrid depth electrode (2069‐ECP‐8C6‐35 T06‐2G‐26, ALCIS) were used in each subject. This electrode has eight macrocontacts (each with surface area of 5 mm2) and six microcontacts positioned radially along the shaft (surface area of 1075 μm2) interspersed between the first and the third macrocontacts (Figure 3). Each patient received 3–14 depth electrodes, usually orthogonally in the temporal and facultatively in the frontal lobes. Their position within the brain was verified using MRI and CT with electrodes in situ. Distribution of ultrafast ripples (1–2 kHz) and ultrafast oscillations (UFOs; 2–8 kHz) across microcontact recordings in four investigated patients. For each subject, data from only one electrode are shown. (A, C) Epileptic hippocampus, (B) nonepileptic hippocampus, and (D) epileptic entorhinal cortex. Event detections for each microcontact in time are given in color strips reflecting the event type (I, II, or both). Each strip represents a 1‐s time interval with at least one micro‐oscillation. Please note that the depiction of the hippocampi in the center of the figure is not anatomically precise but intentionally stylized. Its purpose is to illustrate the approximate location of the electrodes rather than to provide an exact anatomical representation. A 192‐channel research EEG acquisition system (M&I, BrainScope) was used for the recording of 30 min of awake resting interictal SEEG recordings with a sampling rate of 25 kHz and dynamic range of ±25 mV. The analog‐to‐digital conversion is performed by an oversampling analog‐to‐digital convertor with an internal low‐pass decimation filter, which provides effective antialiasing before the signal is downsampled to the final output sampling rate (25 kHz). The EEG acquisition unit was battery‐powered to eliminate line noise, and microelectrodes were connected to specialized high‐impedance amplifiers (>1 GΩ). We used standard epilepsy monitoring unit protocols, but emphasis was given to eliminating power sources of electromagnetic radiation and 50‐Hz power grid influence. No special shielded environment was used. A ground electrode served as a reference for microcontacts; mean from all macrocontacts was used as a reference for macrocontacts. At CHU Toulouse, the hybrid electrodes (DIXI Medical) consisted of a standard macroelectrode (diameter = .8 mm) equipped with two or three tetrodes that protruded up to 2 mm from the shaft between the first and second most distal macrocontacts. A micrometer screw was used to extend the tetrodes from the electrode shaft after implantation. The tetrodes protruded from the shaft at a 30° angle. Each tetrode was made of four tungsten wires, 20 μm in diameter. The theoretical surface of each microcontact was 6.28 × 10−4 mm2. Each tetrode was 70–80 μm in diameter. For hybrid electrodes equipped with two tetrodes, tetrodes were at a 180° angle in relation to each other. They also had nine 2‐mm‐long macrocontacts regularly spaced at intervals along the shaft according to the overall length of the electrode. Three models with two tetrodes were available in a total exploration length space between contacts (in mm) of 33.2/1.9, 40.4/2.8, 50.8/4.1. The length of the electrodes was chosen according to the depth of the cerebral target. For hybrid electrodes equipped with three tetrodes, the tetrodes were at a 120° angle in relation to each other. They had six 2‐mm‐long macrocontacts, three distal and three lateral, all spaced 2 mm apart (three models were available with exploration lengths of 34, 42, and 50 mm). These three tetrodes cover a theoretical triangular surface of 2 mm2. The three‐tetrode configuration was generally chosen when the electrode passed through white matter between the lateral and distal cerebral targets, where it is usually not useful to record EEG signal. Microelectrode recordings were obtained using a 64‐channel Cerebus system (Blackrock Microsystems) with a 30‐kHz sampling rate in two patients, and Atlas 16SX 256‐channel system (Neuralynx) hardware filtered between .1 and 8192 Hz and sampled at 32 768 Hz, subsequently downsampled to 16 kHz, in two other patients. At Wrocław Medical University Hospital, custom hybrid depth electrode designs (AD‐Tech Medical Instrument Corporation) were used. These electrodes are based on standard 4–8‐contact clinical depth electrodes. The depth electrodes consist of a 1.3‐mm diameter polyurethane shaft with Pt/Ir clinical macroelectrode contacts; each clinical contact is 2.3 mm long with either 5‐ or 10‐mm center‐to‐center spacing (surface area = 9.4 mm2). In addition to the standard clinical contacts, the six‐contact depth has 10 microcontacts (between macrocontacts 1–3) positioned radially along the shaft of the depth electrode. The eight‐contact depth electrode, called the Benhke‐Fried electrode, has a bundle of eight Pt/Ir microwires that protrude 1–3 mm from the tip of the electrode lead. The microwires are made of insulated Pt/Ir wire, 40 μm in diameter. The patients were implanted with one or two of the six‐contact hybrid SEEG electrodes in the temporal or the frontal lobe and 3–4 of the eight‐contact Benhke–Fried electrodes in the temporal cortex unilaterally in any one patient with other standard clinical depth electrodes implanted for monitoring of the other lobes. The electrode position within the brain was verified using CT with electrodes in situ. A clinical acquisition system (Neuralynx Inc.) was used to acquire iEEG at high sampling rates for research purposes. The microcontact signal acquired using a wideband DC‐coupled analog front end (passband approximately DC to 9 kHz) and digitized at 32 kHz, whereas the macrocontact signals were downsampled to 5‐kHz sampling frequency. A subgaleal strip of four macrocontacts was placed in the vertex region of the scalp midline between the international 10–20 Cz and Fz electrode positions for a common reference signal for the macrocontacts. Microcontact signals were referenced locally to one of the microchannels that showed no neuronal spiking or other neural activities. EEGWave was used for interactive graphical analysis of large, multichannel data, custom Python scripts were employed for batch processing of data and event detection, and R software was used for statistical analysis. Detection methods Oscillations > 2 kHz Detection was performed in a 30‐s sliding window. First, a spectrogram of the input signal was computed with a Blackman–Harris fast Fourier transform (FFT) window of .015 s with 97% overlap and detrending of the windowed signal. Every row of the spectrogram was divided by its median (normalized) in a 1‐s sliding window (Figure 1B). Only frequencies of 2–8 kHz were taken into account to keep maximal analyzed frequencies under the practical limit of sampling frequency (fs)/3 and to respect hardware filters of recording devices. Events from lower frequencies are out of the scope of this analysis. One‐dimensional detection signal was created by summing all rows of the filtered and frequency‐bounded (2000–8000 Hz) spectrogram. Ninety‐ninth percentile multiplied by a factor of 5 was used for peak detection on the one‐dimensional signal (Figure 1C). Peak areas bounded by its inflex points were marked as high‐frequency activity. For each putative oscillation detection, two things were saved for further analysis: Original raw signal of the putative detection. Original raw signal of the putative detection. Spectral profile created by summing all columns of normalized spectrogram within the time‐bounded detection area (Figure 1). Spectral profile created by summing all columns of normalized spectrogram within the time‐bounded detection area (Figure 1). Ultrafast oscillation detection method. Primary detection of every high‐frequency activity is depicted in the left panel. (A) Raw signal. (B) Normalized spectrogram of raw signal. (C) Summing all rows of the spectrogram creates a one‐dimensional detection signal, which is thresholded. (D) Primary detections. (E) Spectral profiles of primary detections. Right panel shows filtering of primary oscillations. (F) First, oscillations are assessed regarding whether their spectral profile is clear or not clear; the latter are removed from further analysis. (G) Other oscillations are sorted into two groups. EEG, electroencephalographic; Rel., relative; std, SD of amplitude. Every spectral profile of a detected high‐frequency event was subsequently classified as either rhythmic or nonrhythmic. Event was marked as rhythmic if the ratio of maximum to mean of the spectral profile was greater than 10. Other profiles were marked as events with a low spectral peak, because there was no dominant frequency and therefore they could be considered as widespread noise. Maximum to mean ratio served as the tuning parameter of the detector and was chosen heuristically. Lowering this number led to more detections; however, their main frequency was less dominant. By setting this parameter to the value of 10, we chose 34.8% ± 22.1% of primary detections, depending on the record (Figure 1F). The detection pipeline was implemented within our published EPYCOM (Electrophysiology Computational Module), and the full source code is publicly available online. 8 We observed that there were two major shapes of oscillations; type I smoothly increased and decreased its amplitude and oscillated in one frequency band, which was represented by one circular or ellipsoidal area in the spectrogram, whereas a type II oscillation started with a sudden increase followed by a slow decrease of amplitude often accompanied by a shift of mean amplitude (see Figure 1G, bottom). This behavior led into one vertical strip in the spectrogram followed by high activity in one frequency area. To distinguish between these two shapes, the raw signal of every detection bounded by detection start and end was analyzed. The point where the absolute value of the derivative reached half of its maximum was marked as the true detection start. One fourth of detection length to the left from the true detection start was marked as section 1, and one fourth of detection length to the right was marked as section 2. If the true start was not placed between one fourth and three fourths of detection length, it was automatically categorized as type I. The nature of the detector (length of FFT window, using inflex points for detection start) ensured that type II oscillations were always detected prior to their typical amplitude rise. Type II oscillation was defined by either of the following: Equation 1 represents sudden change in SD of amplitude (std) within an oscillation. So does Equation 2, but with one condition added, which represents a shift of means between two sections. Clear oscillations that did not fulfill any of these conditions were marked as type I. Constants 10 and 4.5 can be considered as tuning parameters and were chosen heuristically. Using these constants we categorized cca 35% of oscillations as type II and 65% oscillations as type I. In some cases, the ultrafast oscillations (UFOs) were detected simultaneously in more than three microcontacts at the same time. We declared these detections to be suspected artifacts and omitted them from statistics to minimize false positives. Forty‐two percent of all detections were declared artifacts. To statistically compare the absolute count of type I and type II oscillations, we used nonparametric paired Wilcoxon signed‐rank test, where oscillation count on each patient channel served as one sample. Ripples, fast ripples, and very fast ripples Oscillations in ripple (80–250 Hz), fast ripple (250–500 Hz), and very fast ripple (500–1000 Hz) bands were detected using line‐length detector. 9 For every oscillation type, signal was first processed by a third order band‐pass filter in the respective band, and detection was performed by line‐length detector with the following parameters: statistical window of 10 s, 6 SD was set as the detection threshold, window size was calculated as fs/50 samples, and window overlap was set to 25% of the sliding window length. Spikes Barkmeier detector 10 was used for spike detection with the following parameters: scale factor = 70, threshold coefficient of 4 SD, left and right slope ratios of 700, left and right spike duration of .1 s. Statistical analyses All statistical analyses were performed using R software version 4.2.1. The significance threshold was set to α = .05 (5%) for all tests. To compare the number of oscillations between epileptic and nonepileptic hippocampi, we employed a generalized linear mixed model (GLMM) implemented in the function glmmTMB() from the glmmTMB library (https://www.rdocumentation.org/packages/glmmTMB). The model accounted for repeated measurements from multiple electrodes within each patient by including patient identity as a random effect. Because the oscillation counts displayed overdispersion, the negative binomial distribution was used. For each analysis, the dependent variable was the number of detected oscillations per electrode in a 10‐min resting segment. The fixed effect was the group factor (epileptic vs. nonepileptic tissue). Separate GLMMs were fitted for: All UFOs combined, representing the total number of UFOs (2–8 kHz) detected across all channels. All UFOs combined, representing the total number of UFOs (2–8 kHz) detected across all channels. Type I UFOs, characterized by smooth spindlelike morphology and a single‐band oscillation pattern. Type I UFOs, characterized by smooth spindlelike morphology and a single‐band oscillation pattern. Type II UFOs, characterized by an initial sharp deflection followed by a decaying oscillatory burst. Type II UFOs, characterized by an initial sharp deflection followed by a decaying oscillatory burst. Each model was evaluated for six 1‐kHz‐wide frequency bands (2–3, 3–4, 4–5, 5–6, 6–7, 7–8 kHz) and for the broad band 2–8 kHz. For each band, the model estimated coefficients, 95% confidence intervals (CIs), and p‐values for the group effect. Data adequacy was verified prior to testing. Bands with insufficient observations (typically >5 kHz for type II) were described qualitatively rather than modeled. Descriptive statistics were derived from model‐based estimates and summarized as means with 95% CIs. The analytical design followed the same protocol across all datasets to ensure comparability. The limited sample size (n = 9 patients: n = 6 epileptic, n = 3 nonepileptic) was recognized as a limitation of inferential power, particularly at higher frequencies. RESULTS As expected, in both macro‐ and microcontacts of the depth EEG electrodes implanted into the mesiotemporal structures, and mainly into the epileptic hippocampus, we repeatedly detected high‐frequency oscillations (HFOs) in the frequency range of ripples (80–250 Hz) and fast ripples (250–500 Hz). We also detected very high‐frequency oscillations (VHFOs) in the range of very fast ripples (500–1000 Hz) and ultrafast ripples (1–2 kHz). 1 , 11 , 12 In addition, short‐lasting oscillations in frequencies between 2 and 8 kHz were occasionally detected in the microrecordings from all of the investigated subjects. Two distinct types of these UFOs could be differentiated, and both types were observed in recordings across four independent EEG labs (Figure 2). Type I UFOs had a pattern of typical spindlelike ripples, they were not accompanied by synchronous lower frequency activities, and their median amplitude was 85.2 μV (interquartile range [IQR] = 45.4–216.9). Type II UFOs started with an initial “high‐amplitude” sharp deflection followed by an oscillation rapidly decreasing in its voltage. Also, type II UFOs mostly occurred independently of lower frequency activities (ripples, fast ripples, very fast ripples, and ultrafast ripples), and the median amplitude was 323.5 μV (IQR = 138.9–650.3). They were less frequent when compared to type I UFOs (330 vs. 2085, p < .001). Both types were found in 12 of 15 patients, whereas the remaining three only had type I UFOs detected. Notably, UFOs fluctuate in character, time, and location; different frequencies and types can occur in the same channel and both independently or synchronously in neighboring channels (Figure 2D). In the time domain, 8.16% (IQR = 3.88–13.74) of UFOs overlapped with spikes, and more rarely they overlapped with ripples (0%, IQR = 0–10.78), fast ripples (0%, IQR = 0–10.09), and very fast ripples (0%, IQR = 0–13.05). Examples of ultrafast oscillations (UFOs). (A) Visualization of raw (unfiltered) signals with UFOs in different spectrogram settings (median normalization is used in all the time frequency maps in the paper). (B) Examples of types I and II (time–frequency maps and correlated raw electroencephalographic data). (C) Relative rates of spikes, ripples (R), fast ripples (FR), very fast ripples (VFR), ultrafast ripples (UFR), and UFOs across all multisubject microrecordings. (D) Example of mutual temporal relationships of distinct oscillations in two selected adjacent and simultaneous microrecording channels. TFA, time frequency analysis. Importantly, when restricting the analysis to standard (macro) depth electrodes that had a sufficiently high sampling rate, HFOs of >2 kHz were not detectable except in a single contact in one of the three available patients. We verified that these were UFOs by comparing them with neighboring microelectrodes. However, the rate of UFOs on the macrocontact was nearly 50 times smaller than on adjacent microcontacts (.08/min vs. 3.8/min). In our study, UFOs were consistently and locally detected in microrecordings from obviously epileptic hippocampi (in all five available intrahippocampal recordings) and once from a neocortex microelectrode adjacent to the epileptic hippocampus. A maximum of micro‐oscillations was observed here in the frequency range of 2–5 kHz, but in four subjects they reached up to 7 and 8 kHz, respectively (Figure 3 and 4). Ultrafast oscillation (UFO) rates in epileptic versus nonepileptic mesiotemporal regions across nine subjects. (A) Overall oscillation rates in the broad frequency range (2–8 kHz). (B) UFO rates by frequency band, comparing epileptic (red) and nonepileptic (black) regions. (C, D) UFO rates separated by type I (C) and type II (D) across frequency bands. Each point represents an individual microchannel. *p < .05. n/a, not enough data for statistical testing; ns, p > .05. Goodness of fit statistics are available in Supplementary Note 2. The UFO rates varied interindividually, across time, and across recording microcontacts, with a maximum rate of 1.2–15.9/min for most active channels within six definitely epileptic structures (successfully operated). The UFOs were present both widely across several recording microcontacts (generally in lower frequency ranges, 2–3 kHz) as well as more locally, and repeatedly even in a single contact (in higher frequency ranges). Importantly, type I and type II of UFOs did not represent the identical phenomenon, even if both of them predominantly affected similar channels (Figure 3). Finally, parallel recordings from adjacent macrocontacts consistently revealed traditional HFOs and VHFOs. When comparing rates of UFOs (in the broad frequency range 2–8 kHz) within epileptic (n = 6) versus nonepileptic (n = 3) mesiotemporal areas (hippocampi and entorhinal cortex), a mixed‐effects model with patient as a random factor revealed significantly higher UFO rates in epileptic tissue (p < .05; Figure 4). The same effect was present in the 2–3‐kHz subband (p < .05), and a similar—but statistically nonsignificant—trend was observed in the 3–4‐kHz range (estimate = 2.16 vs. 45, p = .125). In contrast, no convincing or only very rare UFOs were detected in nonepileptic hippocampi (n = 3). Detailed parameters of the GLMMs are provided in Tables S1–S3 of the Supplementary Material. DISCUSSION Despite a prevailing assumption that ultrafast EEG activity has its limitation at approximately 1000 Hz and faster frequencies might be nothing but “neuronal chatter,” 13 several papers published in the past decades are telling us the opposite. A novel ultrafast electrophysiological component of somatosensory‐evoked potentials (1235–2632 Hz) was directly recorded from the human brain by Sakura et al. 4 and even with higher frequencies by Cao et al. 6 Three other independent research groups provided convincing proof of nonevent VHFOs up to 2000 Hz in human epileptic brains. 1 , 5 , 11 , 12 , 14 By analogy, high‐frequency activities at 2000 Hz were identified in vivo intrahippocampal electrophysiological recordings from a rat using a novel pseudowavelet approach, and these activities increased in prominence with epileptogenesis. 15 Finally, even a simple specialized electric organ of the weakly electric fish Apteronotus leptorhynchus comprising neuronlike cells is able to produce physiological and intentional discharges up to 2.2 kHz. 16 In this special case, however, extreme neuronal firing seems to be in large part a result of the evolution of voltage‐gated ion channel genes. 17 Our observation of UFOs in the frequencies between 2 and 8 kHz, recorded using microelectrodes mostly from the human epileptic hippocampi and their surroundings, can be seen as the next breakthrough step moving the limits of EEG and our understanding of electrical events in the human brain. Microelectrodes have much higher impedance than macroelectrodes and are highly prone to artifacts. 18 Thus, our recordings were obtained using reliable high‐impedance amplifiers, analyses were done with great care, and it is therefore highly unlikely we faced some kind of “false” HFOs. The origin of SEEG data in four independent labs across the world and distinct EEG recording systems, together with intraindividually selected observations across all investigated subjects, characteristics of the events, and significant predilection of UFOs to epileptic tissue make the artificial provenance of UFOs hardly possible, at least for characteristic “spindlelike” type I UFOs. Skepticism on the authentic origin of UFOs might nevertheless be augmented with our knowledge of an absolute inactive phase of neurons at approximately 2 ms. A single synchronized neuronal group can generate discharges with the upper limit at approximately 300 Hz. On the other hand, the network mechanisms underpinning fast ripple and faster oscillations could be explained by phase delay of different subpopulations of synchronized neurons, that is, out‐of‐phase firing. 19 Such desynchronized subpopulations can fire with frequencies up to 500 Hz or even higher. The computational model presented recently by Fink et al. 20 showed a similar effect on cellular level with a simulated cluster of asynchronous, uncoupled networks of pyramidal cells. When the inhibitory mechanisms in the neural tissue are impaired and the pyramidal cells become very active due to intense synaptic input, multiunit out‐of‐phase firing can occur and produce very high frequencies, very likely far more than 1000 Hz. When the pyramidal cells are desynchronized, the aggregate electrophysiological signal can result in UFOs of any frequency which detection is then only limited with an advancement of recording devices (see Supplementary Note 1 and Figure S1 for an illustration of spike superposition leading to apparent UFOs). We used a computational model to simulate whether such high frequencies can be generated by physiological mechanisms. Using the methods described by Fink et al., 20 we simulated the output generated when a network discharge was generated by multiple clusters of synchronized pyramidal cells firing action potentials. Each individual cell in a cluster only fired once, and different clusters fired with a delay of 1/F (F = 1000, 2000, 5000, 10 000). Under these conditions, in which the overall burst is comprised of a chain of individual synchronized clusters firing with a short delay, the model was able to generate clear oscillations up to 10 000 Hz. This mechanism of UFOs' origin is further strongly supported by our recently published study, establishing a mathematical framework based on bifurcation theory and examining how a reduced neuronal coupling within an epileptic focus could lead to VHFOs and UFOs in depth EEG signals. 21 CONCLUSIONS In this paper, we describe for the first time ultrafast electric field oscillations far above 2 kHz in the human brain. These electrophysiological events can be reliably detected using advanced depth microEEG recordings, and they likely mirror neuronal hyperexcitability and reflect the hypersynchronous firing of neurons and a collective oscillation created by summation of the crests of single unit neuronal firing activity. As such, their future studies may provide a deeper insight into the pathomechanisms of epileptogenic brain. AUTHOR CONTRIBUTIONS Milan Brázdil, Vojtěch Trávníček, and Pavel Jurák: Study conception and design. Vojtěch Trávníček, Pavel Jurák, and William C. Stacey: Data analysis. Jonathan Curot, Jan Cimbálník, Martin Pail, Monika Służewska‐Niedźwiedź, Robert Roman, and Emmanuel Barbeau: Data acquisition and metadata collection. Vojtěch Trávníček and Filip Plešinger: Figure design. Petr Klimeš, Michal Kucewicz, Gregory A. Worrell, Emmanuel Barbeau, Milan Brázdil, and Vojtěch Trávníček: Writing—original draft. FUNDING INFORMATION The research was supported by Czech Science Foundation Grant No. 21‐44843L; by project LX22NPO5107 (Ministry of Education, Youth, and Sports of the Czech Republic), financed by European Union–Next Generation EU; and by the Czech Academy of Sciences‐Project RVO: 68081731. CONFLICT OF INTEREST STATEMENT The authors declare no conflict of interest. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines. Supporting information FIGURE S1. Emergence of ultrafast oscillations through progressive spike superposition. TABLE S1. All ultrafast oscillations. TABLE S2. Ultrafast oscillation type I. TABLE S3. Ultrafast oscillation type II. ACKNOWLEDGMENTS The author has nothing to report. Open access publishing facilitated by Masarykova univerzita, as part of the Wiley ‐ CzechELib agreement. DATA AVAILABILITY STATEMENT The data that support the findings of this study are available from the corresponding author, upon reasonable request. REFERENCES Associated Data Supplementary Materials FIGURE S1. Emergence of ultrafast oscillations through progressive spike superposition. TABLE S1. All ultrafast oscillations. TABLE S2. Ultrafast oscillation type I. TABLE S3. Ultrafast oscillation type II. Data Availability Statement The data that support the findings of this study are available from the corresponding author, upon reasonable request.