Structured spontaneous activity through delta oscillation-based discrete states in the medaka telencephalon Authors: Saori Yokoi, Ryosuke Koike, Kotaro Yamashiro, Shinichi Nakagawa, Yuji Ikegaya, Nobuyoshi Matsumoto Published: 2026 ABSTRACT AbstractSpontaneous neural activity shapes brain states, which are associated with various behaviors, yet its organization in the teleost telencephalon is poorly understood. We characterized intrinsic neural activity dynamics in medaka (Oryzias latipes) by recording 32-site local field potentials in the lateral (Dl) and medial (Dm) dorsal telencephalon under anesthesia. We segmented delta-band (1–4 Hz) signals into 3-s episodes and computed pairwise correlation matrices for each episode. We embedded them via principal component analysis and identified brain states from the correlation patterns using k-means. State occupancy was biased rather than uniform, and temporal dynamics were structured; the episode-to-episode distances were shorter than those for shuffled surrogates, and the transitions preferentially recurred within the same state. These results indicate that, even under reduced sensory drive, the medaka telencephalon self-organizes into stable and preferred network configurations with structured propagation, echoing conserved principles of spontaneous activity across vertebrates. Our framework advances understanding of intrinsic brain function in teleosts and provides a platform for future studies in conscious animals.Keywords: Teleost, Medaka, Telencephalon, Delta, Principal component analysis, k-meansSubject terms: Neuroscience, Zoology Introduction Medaka fish (Oryzias latipes) has been regarded as an invaluable model animal, owing to its ease of husbandry, well-annotated genome1, and established genetic manipulation techniques2,3. In particular, the compact brain of medaka fish offers a better system to analyze the function of individual neural nuclei and cytoarchitecture, further enabling us to understand complicated neural circuits and behaviors. Prior research has demonstrated that medaka fish exhibits a variety of behaviors including reproduction4, feeding5, fear responses6, exploration7, and social behavior8–10. Since these various behaviors are regulated by neural activity, the neural mechanisms can be addressed by physiological recordings. However, neural activity in the conscious and unconscious medaka fish has not been fully explored in terms of physiology. Even in the absence of sensory stimuli, the brain remains intrinsically active. This type of brain activity is known as spontaneous activity and was once treated as neural noise. However, recent research has shown that spontaneous activity is involved in selective responses to sensory input and in pathological conditions11–14. In this sense, spontaneous activity should be considered not as noise, but rather as a functional signal. Furthermore, spontaneous activity has been studied mostly in quiescent/anesthetized mammals, fish, and birds, through electrophysiology and imaging15–17. These studies have suggested that spontaneous activity is conserved across species. For example, slow-frequency delta-band oscillations (described below) occur in the cerebral cortex in mammals under unconsciousness (not only during sleep, but also under anesthesia). Yet, spontaneous activity in evolutionarily more ancient vertebrates, including teleosts, has been poorly understood. Spontaneous activity is often investigated using local field potentials (LFPs), reflecting collective population-level neural activity dominated by synchronous synaptic inputs. LFPs comprise various frequency band components, indicative of brain states and functional activity. In particular, delta-band (1–4 Hz) oscillations in mammals are associated with sleep and memory18,19. Additionally, delta oscillations in mammals are generated by synchronization in the thalamocortical circuit under anesthesia20. In teleost fish, neural projections from the preglomerular complex to the lateral part of the dorsal telencephalon (Dl) have been proposed to be functionally analogous to the mammalian thalamus-to-neocortex pathway21. Although such correspondence is not anatomically identical, delta oscillations have also occurred in the anesthetized trout brain22, supporting the possibility that low-frequency synchronous activity is conserved across vertebrates. The telencephalon in vertebrates, including teleosts, is involved in higher-order brain functions such as learning and memory, sensory information processing, and emotion23,24. In teleosts, the lateral (Dl) and medial (Dm) subdivisions of the dorsal telencephalon are considered potential homologues of the mammalian hippocampus/neocortex and amygdala, respectively, based on developmental, molecular, and functional similarities25. Although their anatomical organization differs due to the everted topology of the teleost pallium, the Dm and Dl exhibit distinct functional specializations; the Dl participates in multimodal sensory integration, particularly of visual information21, whereas the Dm is linked to emotional and social processing26. In addition, the Dm receives visual21,27, auditory21,28, and gustatory information29. In this light, characterizing spontaneous activity dynamics of the Dl and Dm could provide a new framework for exploring how intrinsic network states contribute to the behaviors in medaka fish. However, to date, spontaneous neural activity in the medaka telencephalon has not yet been systematically profiled. To this end, using multi-channel silicon probes, we recorded LFPs from 32 sites in the Dl and Dm in anesthetized medaka fish and extracted delta-band oscillations as an index of spontaneous activity. For each 3-s segment (called an episode), we computed Pearson’s correlation coefficients between all possible channel pairs, generating a series of correlation matrices representing episode-wise activity patterns. Each matrix was converted into a feature vector (32 × 32 values), projected into a three-dimensional space using principal component analysis, and clustered into discrete brain states using k-means based on Euclidean distances to identify the episode-to-episode transitions and characterize the state dynamics in terms of neural activity in the medaka telencephalon. Materials and methods Ethical approval All procedures and experimental protocols involving live medaka fish (Oryzias latipes) were conducted in strict accordance with the ARRIVE guidelines as well as the guidelines approved by the Animal Care and Use Committee of Hokkaido University (permission no. 23-0115). To ensure the well-being of the animals, surgeries and recordings were carried out under MS-222 anesthesia, adhering to the principles outlined in the NIH Guide for the Care and Use of Laboratory Animals. Diligent efforts were taken to minimize animal discomfort and distress throughout the study. Animals Medaka fish (n = 6; Oryzias latipes; d-rR/Tokyo strain) were kept in groups in plastic aquariums (16 cm × 25 cm × 14 cm [height]). All fish were hatched and bred within our laboratory facilities at Hokkaido University. For the experiments, we randomly selected male and female medaka fish aged between 4 to 5 months. The water temperature was maintained at approximately 28 °C, and a consistent photoperiod was provided, with 14 h of light from 8 a.m. to 10 p.m. daily, using standard fluorescent lighting. Drug MS-222 (C₁₀H₁₅NO₅S; T0941, Tokyo Chemical Industry, Japan)30 was dissolved in distilled water to prepare a 0.8% (w/v) stock solution, adjusted to pH 7.0 with NaOH, and stored at − 20°C until use. For experiments, the stock solution was diluted with system water to a final concentration of 0.02% (w/v) and used as an anesthetic for medaka fish. Electrophysiology Based on previous literature31, a stereotaxic frame and head plate were custom-fabricated using a 3-D printer. The stereotaxic frame was fixed on a chamber. The chamber was filled with anesthetic solution so that the medaka fish was fully submerged and rapidly anesthetized. The anesthetized fish was fixed to the custom-built head plate with UV curing adhesive and insect pins, after which the head plate was attached to the stereotaxic frame with plastic screws (Fig. 1a). Spontaneous neural activity in the dorsal telencephalon in medaka fish recorded by a 32-channel silicon probe. (a) 3D rendering of the custom-fabricated stereotaxic frame and silicon probe setup for medaka (left) with a simplified atlas of the medaka telencephalon including the Dl and Dm (inset). The 3D scale bars indicate 1 cm in each of the three dimensions. Channel assignment of the 32-channel probe (right): channels 1–16 and 17–32 correspond to the right and left hemispheres, respectively. (b) Representative traces of LFPs. Channels 1–8 (blue), 9–16 (red), 17–24 (red), and 25–32 (blue) correspond to the right Dl, right Dm, left Dm, and left Dl, respectively. (c) Lateral view of the adult medaka brain (top), with a vertical line (purple) indicating the rostrocaudal position of the coronal section used for histological verification of electrode tracks (bottom). The back side of the silicon probe was precoated with DiI (magenta) before implantation. Coronal sections were counterstained with DAPI (blue). Note that at this rostral level, the dorsal telencephalon is primarily composed of the Dl and Dm and that bilateral implantation of silicon probes was confirmed using adjacent serial sections shown in d. (d) Same as c, but coronal sections adjacent to the one shown in c, demonstrating the electrode tracks in the contralateral hemisphere. Note that left and right hemispheres were separated during sectioning at this rostral level. (e) Cumulative distributions of Pearson’s correlation coefficients of delta-band oscillations. Pairwise correlations were calculated between all possible channel pairs throughout the entire recording period. The coefficients are compared between channel pairs located within the same region (intra-region, green) and those across different regions (inter-region, black). P = 1.34 × 10−7, D = 0.30, Kolmogorov–Smirnov test. (f) Correlation of delta oscillations between channel pairs is compared between the left (magenta) and right (cyan) hemispheres. No significant differences are found between hemispheres (P = 0.22, D = 0.19, Kolmogorov–Smirnov test). (g) Correlation of delta oscillations between channel pairs is compared between the Dl (blue) and Dm (red). No significant differences are found between the two regions (P = 0.32, D = 0.17, Kolmogorov–Smirnov test). (h) Pearson’s correlation coefficients of delta-band LFPs during the entire recording period A representative correlation matrix based on Pearson’s correlation coefficients of delta-band LFPs. Each element (i, j) indicates the correlation between channels i and j. Hot and cold colors indicate higher and lower correlation coefficients, respectively. This heat map is created using MATLAB R2024a. LFP, local field potential; Dl and Dm, the lateral and medial subdivisions of the dorsal telencephalon; DiI, 1,1'-dioctadecyl-3,3,3',3'-tetramethylindocarbocyanine perchlorate. Through microscopy, the skull was gently removed using syringe needles. A 32-channel silicon probe (Buzsaki32, NeuroNexus Technologies, USA) was connected to a headstage (C3324, Intan Technologies, USA) via an adaptor (Adpt-A32-OM32, NeuroNexus), which was supported by a micromanipulator (MM-3, Narishige, Japan) and stand (Z-1, Narishige). The headstage was further connected to the Open Ephys acquisition board through multiplexed cables (C3213, Intan)32,33. The back side of four shanks was gently precoated with 10 μg/ml 1,1'-dioctadecyl-3,3,3',3'-tetramethylindocarbocyanine perchlorate (DiI) (in acetone) for histological verification of electrode tracks (see the next section); note that the back side of the shanks does not have recording channels, while the front side does. The silicon probe was carefully implanted into the medaka brain under a microscope and bilaterally aimed at the rostral level of the telencephalon (Fig. 1c, d), where the Dl and Dm were the predominant dorsal pallial structures, thereby avoiding the inclusion of intermediate or caudal regions. Unlike rodent studies18,34–36, the electrode was not precisely localized, but the electrode track was identified in post-hoc histological verification to confirm the coordinates for target regions. LFPs were amplified and digitized at a sampling frequency of 30 kHz by the headstage and transmitted to the data acquisition system. Signals were monitored using the Open Ephys GUI. During the recording, the chamber was filled with 0.02% MS-222 solution without active oxygenation, and the temperature was maintained at ~ 20°C. All neurophysiological recordings were performed between 11 a.m. and 3 p.m., within the light phase of the light/dark cycle. LFP data were excluded from the subsequent analysis if post-hoc histological verification (see below) confirmed that the electrode track was out of the target regions. Histology After recording, the anesthesia was maintained on the medaka fish, and its brain was carefully dissected. The brain was then immersed in 4% paraformaldehyde (PFA) in 0.01 M phosphate-buffered saline (PBS; pH 7.4) overnight for fixation. The fixed brain was coronally sectioned at a thickness of 100 μm using a vibratome. Serial slices were mounted on glass slides and coverslipped with a mounting medium containing DAPI (19,178-91, Nakalai Tesque, Japan). Fluorescent images were captured using a stereomicroscope (BX51-34-FL, Olympus, Japan). The positions of electrodes were confirmed by identifying tracks in the histological tissue. Data were excluded from the subsequent analysis if the electrode position was outside the target brain region. Data analysis All data analyses were performed using custom-made MATLAB R2024a routines (MathWorks, USA). Summarized data are reported as the mean ± the standard error of the mean (SEM). The null hypothesis was statistically rejected when P < 0.05 unless otherwise specified. When multiple comparisons were required, we corrected the original significance level (i.e., 0.05) with Bonferroni correction and compared the original P values with the corrected significance level. Although the delta-band (1–4 Hz) oscillations were inherently outside the frequency range of common electrical noise (e.g., 50/60 Hz hum), a digital notch filter (50 Hz) was applied offline to the raw LFP signals. In addition, since all recordings were performed under MS-222 anesthesia, the medaka were completely immobilized; consequently, muscular twitches or movement-related artifacts were negligible throughout the recording period. Any segments containing high-amplitude transient noise (e.g., occasional sudden jumps) were manually identified. After such jamming noise (e.g., hum noise and unexpected sudden electrical artifacts) was removed from raw LFP signals originally sampled at 30 kHz, the purified signals were downsampled to 2 kHz. Downsampled signals were further bandpass-filtered between 1–4 Hz to extract delta-band components of neural signals. Pearson’s correlation coefficients of delta-band oscillations for the entire recording period were calculated for all 1024 (i.e., 32 × 32) possible pairs of LFP signals. The delta-band signals were then segmented into 3-s windows, which produced approximately 1200 segments for about 1-h recording. Each 3-s segment was referred to as an “episode”, hereafter. For each 3-s episode, Pearson’s correlation coefficients were calculated for all 1024 possible pairs of delta oscillations and displayed in a form of 32 × 32 pseudocolored matrix; that is, the component (m, n) had the correlation coefficient of the delta-band signals between channels #m and #n, and higher (i.e., closer to 1) and lower (i.e., closer to -1) values were indicated by hot and cold colors. Two algorithms (i.e., principal component analysis (PCA) and k-means clustering) were used to better understand the transition in neural activity states. PCA was applied to flatten each 32 × 32 matrix (i.e., 1024 elements) into a vector and embed it in a three-dimensional space. The k-means clustering algorithm was used to detect states in the PCA space, which was subsequently repeated for all medaka fish. The optimal number of clusters for the k-means algorithm was determined based on the Elbow method37,38. In short, the k-means clustering algorithm was run with the number of clusters (k) ranging from 2 to 20. The sum of squared errors (SSE) was computed for each value of k to measure the isolation quality of the resulting clusters. Here, the number of delta oscillation-based states in medaka brain was determined as the optimal number of clusters, the point where the rate of change in the SSE with respect to the number of clusters was minimal. A cluster containing segment j was defined as state j. Occurrence probability of transitions from state j to segment j + 1 was visualized in the digraph chord chart, with high probabilities represented by thick bands. To compile statistics by testing the significance of state transitions, surrogate datasets were generated by randomly permuting the temporal sequence of states 999,999 times (using the Mersenne Twister algorithm in MATLAB for randomness and reproducibility). This procedure disrupted the temporal structure of transitions while preserving the overall occupancy of each state, providing an expected distribution under the null hypothesis (i.e., a null distribution) that state transitions occur independently and without temporal structure for subsequent statistical comparisons. Results We extracellularly recorded LFPs from 32 sites bilaterally in the dorsal telencephalon (i.e., the left and right Dl and Dm) in anesthetized medaka fish for approximately 1 h (Fig. 1a–d). Since the telencephalon of the medaka used here was approximately 1.0–1.2 mm in width (Fig. 1c, d), the 32-channel silicon probe (i.e., Buzsaki32) with a shank-to-shank distance of 200 μm (Fig. 1a) allowed for the simultaneous and independent placement of shanks into the Dl and Dm regions without overlapping. Moreover, each shank features 8 recording sites spaced 20 μm apart, covering a 140 μm vertical span. This scale is smaller than the Dl and Dm, ensuring that recorded LFPs represent localized population activity. As delta oscillations are often indicative of spontaneous activity39, we band-pass filtered the raw LFP signals at 1–4 Hz (i.e., delta frequency band) and segmented into episodes. We calculated Pearson’s correlation coefficients of filtered LFPs (i.e., delta oscillations) during the whole recording period between all possible channel pairs (Fig. 1e) and found distinct but partially similar correlation patterns of delta oscillations during the recording period (Fig. 1e). Correlation of delta oscillations between channels was significantly larger within regions than across regions (P = 1.34 × 10–7, D = 0.30, Kolmogorov–Smirnov test; Fig. 1e); however, intra-regional correlation of delta oscillations was not significantly different between the left and right hemispheres (P = 0.22, D = 0.19, Kolmogorov–Smirnov test; Fig. 1f), or between the Dl and Dm (P = 0.32, D = 0.17, Kolmogorov–Smirnov test; Fig. 1g). These results indicate that the Dl and Dm are not independent but participate in partially overlapping network dynamics. To capture these local dynamics more directly, we created a correlation matrix for a given episode (Fig. 1h) and used this correlation matrix as an index of spontaneous neural activity. For each fish, we calculated a correlation matrix for each episode. To investigate how many distinct states emerged in ongoing activity in the telencephalon even in the absence of sensory attributes, we flattened each 32 × 32 correlation matrix into a 1024-dimensional vector. This vector was further embedded into a three-dimensional PCA space, where each point represented the correlation structure of a single episode. We then classified the points into discrete states using k-means, an unsupervised clustering method; note that the number of states was automatically determined by the Elbow method (Fig. 2a, b). We demonstrated that the activity patterns were almost discrete and subclustered in the PCA space. To ask whether the neural activity patterns in the telencephalon arose homogeneously or more preferentially to certain states, we counted the number of episodes in each state, indicating the biased distribution of the episodes (Fig. 2c). The estimated number of states ranged between 3 and 18 for all fish tested (Fig. 2d); note that deep anesthesia may reduce the variety of neural activity and hence lead to a small number of neural states. Diversity of brain states defined by spontaneous neural activity in the medaka Dl and Dm. (a) A representative 3D map of spontaneous activity states derived from 3-s episodes of delta-band oscillations in the Dl and Dm. Each episode was embedded as a point in a three-dimensional PCA space (first (leftmost)). The 3D PCA space is projected onto PCA1-PCA2 (second), PCA1-PCA3 (third), and PCA2-PCA3 (fourth) spaces. The episodes were clustered using k-means, after which each cluster was defined as a distinct state. Colors represent different states. (b) Representative correlation matrices for states 1 (left) and 2 (right), calculated from Pearson’s correlation coefficients of delta-band LFPs. These heat maps are created using MATLAB R2024a. (c) The number of episodes assigned to each state shown in a, demonstrating biased state occupancy. (d) Estimated number of states across all fish tested in this study (n = 6 fish). Dl and Dm, the lateral and medial subdivisions of the dorsal telencephalon; PCA, principal component analysis. We next scrutinized the time-series dynamics of the correlated neural activity using the Euclidian distance between a given episode and the next in the PCA space (Fig. 3a). We subsequently characterized whether and how the propagation pattern remained temporally stable or dynamically toggled to another. To this end, we examined the state-to-state dynamics by quantifying the probability of episode-to-episode transitions (Fig. 3b, c). The distances between consecutive episodes for the real data were significantly smaller than those for shuffled datasets (P < 5.98 × 10–306, D = 0.56, Kolmogorov–Smirnov test; Fig. 3b), indicating that episode transitions were not random. Instead, the episode transitions exhibited a form of “biased” patterns as specific episode transitions were significantly more frequent. Moreover, certain episodes tend to remain within the same state. The median of the episode-to-episode distances in the PCA space for the original data was significantly lower than for the surrogate data (P = 0.03, t5 = 3.12, n = 6 fish, paired t-test; Fig. 3d). The ratio of biased transitions within the same state was significantly larger than that between different states (P = 0.02, t5 = 3.20, n = 6 fish, paired t-test; Fig. 3e). Dynamics of transition patterns of delta oscillation-based states. (a) A representative chord diagram showing transitions between states. Each link represents a transition from the state of Nth episode to that of the subsequent episode (i.e., the stateepisode(N) and stateepisode(N+1)). Note that the state to which the episode(N) belonged was represented as the stateepisode(N). (b) Euclidean distances between a given episode and the next in the PCA space for real (blue) and surrogate (black) datasets from a representative fish. (c) Transition matrix corresponding to a. Transition from the Nth episode (i.e., episode(N)) to the next is transformed into a state-to-state transition matrix. Warmer colors indicate higher transition probabilities. Asterisks mark statistically significant transitions (P < 0.05/182, permutation tests with post-hoc Bonferroni correction), defined as biased transitions. This heat map is created using MATLAB R2024a. (d) Median Euclidean distances of episode-to-episode transitions in the PCA space for real (blue) and surrogate (black) datasets across all 6 fish. (e) The ratio of biased transitions occurring within the identical state was significantly higher than that between different states, indicating temporal recurrence and stability of specific activity patterns (n = 6 fish). PCA, principal component analysis. Discussion In this study, we investigated the spontaneous neural activity patterns within the dorsal telencephalon of anesthetized medaka fish by analyzing delta-band LFPs from 32 recording sites. Our findings reveal that the spontaneous activity in the medaka telencephalon is not random; instead, it exhibits subclustered patterns in the PCA space with biased distributions of activity states. Furthermore, we showed the temporal dynamics of these activity patterns do not resemble Brownian motion or random walks but display structured patterns with preferential transitions. These insights into the intrinsic brain dynamics of teleost fish under unconsciousness provide a new platform for understanding fundamental principles of brain function. The prominent delta-band oscillations in the medaka telencephalon under MS-222 anesthesia align with those in other vertebrate species including fish and mammals under anesthesia22,41,42. In mammals, delta oscillations are a hallmark of unconscious states, including deep sleep and general anesthesia, originating from synchronized activity within the thalamocortical circuit43,44. As higher-frequency components such as theta, beta, gamma, or ripple bands are also associated with cognitive functions in mammals34,35,45–52, such components can be mathematically extracted from the medaka LFP signals. However, the mechanisms underlying the generation of higher-frequency (> 6 Hz) oscillations and relevant functions in the fish telencephalon remain less established compared to those in rodents. Analyzing these bands without a clear biological foundation carries the risk of interpreting mathematical artifacts rather than functional physiological signals. While the precise anatomical correspondence between the medaka telencephalon and the mammalian thalamocortical system remains debated, several studies have suggested that the projection from the preglomerular complex to the dorsal telencephalon (particularly the Dl and Dm) in teleost may serve a function analogous to the mammalian thalamus-to-neocortex pathway21,25,40. Consistent with this proposed functional analogy, our results showing delta oscillations in anesthetized medaka fish further support the idea that this type of low-frequency synchronous activity is conserved across vertebrate brains, particularly under reduced sensory drive. The compact size of the medaka brain allowed us to comprehensively record neural activity in both the left and right Dl and Dm within a single coronal section. The 8 recording channels on the tip of each shank in the silicon probe were dorsoventrally clustered across a mere 140-μm span with 20-μm spacing between adjacent channels and the shanks were laterally separated by 200 μm. Despite this compact arrangement, the probe captured neural activity from a relatively wide spatial extent of the telencephalon in medaka fish. This configuration is advantageous in identification of discrete brain states, using PCA and k-means clustering. We found that the activity patterns formed subclusters in the PCA space and exhibited biased occupancy of specific states, indicating that even without external sensory input, the medaka telencephalon organizes into preferred and stable functional configurations. Since such similar intrinsic organization of spontaneous activity has been documented in mammals and birds, where it contributes to resting-state networks and sensory processing53–57, our findings extend this concept to evolutionarily more ancient vertebrates. Delta oscillations exhibit a clear spatial organization of spontaneous activity in the medaka telencephalon. The intra-regional correlation (within the Dl or Dm) is significantly higher than the inter-regional correlation (across regions) (Fig. 1e), indicating that while the telencephalon organizes into a structured global state, the Dl and Dm maintain their distinct local network identities. However, when comparing the intra-regional dynamics themselves, we did not find any significant differences between the left and right hemispheres, nor between the Dl and Dm subdivisions (Fig. 1f, g). This lack of regional difference in the degree of synchrony likely reflects the biological nature of the anesthetized brain rather than a limitation in measurement. Under MS-222 anesthesia, the reduced sensory drive may lead the medaka telencephalon to function as a single, highly integrated network, characterized by conserved resting-state configurations. It should be noteworthy that the Dl and Dm exhibit similar delta-oscillatory dynamics, given their proposed homologies to different mammalian structures (e.g., the Dl to the hippocampus/neocortex and the Dm to the amygdala). While these regions participate in distinct functional specializations during conscious behaviors, their partial synchronization under anesthesia likely represents the emergence of conserved resting-state networks that encompass multiple pallial subdivisions. Thus, the organizational principles of spontaneous neural oscillations appear to be evolutionarily conserved across these vertebrate functional modules, extending from teleosts to mammals. We found that the episode-to-episode transitions exhibited a non-random and temporally static pattern. The lower Euclidean distance in the PCA space for original data compared to surrogate data, along with the higher ratio of biased transition patterns within identical states (Fig. 3e), suggests that the propagation pattern in the fish telencephalic activity is differentiated from Brownian motion or random walks. Instead, the brain is predisposed to maintain specific activity patterns or favor certain transitions between them. This temporal stability and biased transition dynamic could reflect an underlying intrinsic network architecture or a tendency for the anesthetized brain to settle into specific energy-efficient network states. In the context of the medaka telencephalon, where the Dl and Dm are involved in higher-order functions such as sensory integration, social processing, and emotion regulation, these stable patterns of spontaneous activity might represent conserved computational motifs that persist even in unconsciousness and may influence subsequent recovery or baseline functional organization. We indeed focused on the Dl and Dm as they represent major subdivisions of the teleost dorsal telencephalon. Although our post-hoc histological verification confirmed electrode placement within these regions, the potential influence of volume conduction from neighboring areas, such as the subpallium, anterior/posterior pallium, or olfactory bulb, cannot be entirely ruled out. However, the use of high-density silicon probes and the anatomical selection of the rostral recording plane were intended to maximize the contribution of local Dl/Dm dynamics. To overcome these issues, cell-type-specific imaging or focal pharmacological manipulations could further dissociate the precise spatial origins of these spontaneous oscillations. While we provide new insights into the dynamics of telencephalic neural activity in medaka fish, a key limitation here is the recordings under anesthesia, which inherently differs from normal brain states and even could alter naïve neural activity patterns. Our recordings were performed at approximately 20 °C under MS-222 anesthesia without active oxygenation. While these environmental factors and the reduced sensory drive potentially influence absolute neural firing rates, they are unlikely to account for the relative neural activity patterns such as structured, non-random state transitions found in this study. Although anesthesia allows for controlled measurement of intrinsic activity without external confounding factors, it does not fully replicate the dynamics of neural activity in naturally sleeping or conscious animals. Application of our electrophysiological methodology to freely swimming small fish using lightweight, waterproof and wireless devices should be the next challenge58–62. Alternatively, combining electrophysiological approaches with virtual reality paradigms, as demonstrated in zebrafish for two-photon calcium imaging31, would enable controlled manipulation of sensory contexts and further widen our understanding of neural activity dynamics in awake fish. Extending such approaches to medaka fish, a species that relies heavily on visual cues for social recognition and decision-making63,64, would enable the presentation of virtual environments, such as encounters with familiar or unfamiliar conspecifics, to reveal visual-dependent neural activity. Moreover, integrating physiological recordings with behavioral profiling could elucidate the neural underpinnings of individual differences, such as orthodox versus contrarian behavioral tendencies, in medaka fish. Altogether, our study underscores the conserved nature of spontaneous brain activity across vertebrates and offers a robust, quantifiable framework for exploring intrinsic neural dynamics in a genetically tractable teleost model, paving the way for further investigations into how these fundamental patterns of spontaneous activity patterns support visually guided and socially driven behaviors in small fish. Acknowledgements The authors thank the National BioResource Project Medaka (https://shigen.nig.ac.jp/medaka) for providing the d-rR/Tokyo strain (strain ID: MT837). Author contributions S.Y. and N.M. conceived the project and designed the study; S.Y. performed the experiments; R.K. and N.M. analyzed the data; S.Y., K.Y. and N.M. developed methodology; S.Y., S.N., Y.I. and N.M. oversaw and managed the project; S.Y. and N.M. wrote and revised the manuscript. All approved the final version of the manuscript. Funding This work was supported by JST ERATO (JPMJER1801), the Institute for AI and Beyond of the University of Tokyo, JSPS KAKENHI (22K21353), and AMED Brain/MINDS 2.0 (24wm0625207s0101; 24wm0625401h0001; 24wm0625502s0501) to Y. Ikegaya; JSPS KAKENHI (25K18705) and grants from Yamazaki Spice Promotion Foundation, the Salt Science Research Foundation, the Public Foundation of Elizabeth Arnold-Fuji, Kobayashi Foundation and Smoking Research Foundation to N. Matsumoto; NIBB Collaborative Research Program (23NIBB336), JST FOREST Program (JPMJFR241Y), JSPS KAKENHI (21H05708; 23K05841; 23H03839), grants from Astellas Foundation for Research on Metabolic Disorders, Takeda Science Foundation and the Naito Foundation to S. Yokoi. Data availability The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Saori Yokoi and Ryosuke Koike contributed equally to this work. Contributor Information Saori Yokoi, Email: yokois@pharm.hokudai.ac.jp. Nobuyoshi Matsumoto, Email: nobuyoshi@matsumoto.ac. References Associated Data Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.