Time-resolved large-scale neural coactivations in macaque monkey

Abstract

Large-scale distributed activations in various modes of spatiotemporal organizations have been extensively reported in both hemodynamic and electrical/magnetic human brain signals, which provides knowledge on how information is being hierarchically processed and integrated among functionally linked brain regions. These large-scale distributed activations have also been identified in brain signals from animals, indicating that they are preserved brain organizations in species evolution. Recent studies using human electroencephalography (EEG) and magnetoencephalography (MEG) have further revealed that large-scale distributed activations are frequency-specific and of fast dynamics (tens of milliseconds), while these phenomena have not been investigated in animals. The present study used electrocorticography (ECoG) data recorded with the coverage of nearly entire hemisphere(s) to investigate the existence of time-resolved large-scale coactivation patterns (CAPs) in monkey brains and compare them to CAPs from whole-head human EEG data both at resting states. The present results reveal brain-wide patterns of CAPs in monkey ECoG data, which share significant similarities to human EEG CAPs, both in the alpha band, on spatial and temporal patterns not only in individual CAPs but also on relative differences among different CAPs. The transition patterns among all monkey ECoG CAPs further reveal a similar superstructure as in human EEG CAPs that controls the dynamics of brain state transitions at rest and their spatial expressions. These findings suggest that large-scale brain events of fast dynamics exist in non-human primates and they are of functional importance cross species, similar as time-averaged ones that have been well reported in literature.

1. Introduction

The human brain consists of hierarchical networks that allow for the information processing between functionally linked brain regions (van den Heuvel and Hulshoff Pol, 2010). Functional connectivity (FC) between regions of the human brain has been revealed in neuroimaging data, dominantly via functional magnetic resonance imaging (fMRI), at resting (Cordes et al., 2000; Greicius et al., 2003; Hampson et al., 2002) and task conditions (Fox and Raichle, 2007; Arfanakis et al., 2000; Fox et al., 2006; Fransson, 2006). Particularly, synchronous activations of functionally correlated regions at rest indicate distributed large-scale spatiotemporal organizations across the entire brain (Leopold and Maier, 2012), which form a set of the so-called resting-state networks (RSNs) (Liu et al., 2013; van den Heuvel et al., 2009). Recently, several new large-scale patterns that provide more direct knowledge on the potential drivers of the classical correlation-based RSNs have been reported. Among them, coactivation patterns (CAPs) (Liu and Duyn, 2013) from fMRI indicate similar large-scale distributed activations as correlational RSNs, but via assuming a point process (thus transient events) (Tagliazucchi et al., 2011; Cifre et al., 2020). Global propagating waves (GPW) are another category of large-scale phenomena reported in human fMRI data (Raut et al., 2021). Their propagational dynamics along anatomic structures reveal more biologically explainable features as compared with RSNs and have been suggested critical in shaping dynamics of RSNs (Gu et al., 2021). Other research efforts have revealed more large-scale patterns, e.g., quasiperiodic patterns (QPP) (Abbas et al., 2019; Thompson et al., 2014; Yousefi et al., 2018).

Due to the indirect nature of fMRI signals, the neuronal origins of these large-scale phenomena in human brains have been investigated from various perspectives. Successful detections and characterizations of similar RSNs in electrical/magnetic signals generated from primary neuronal responses, including electroencephalography (EEG) (Custo et al., 2017; Ding et al., 2022; Yuan et al., 2016), electrocorticography (ECoG) (Betzel et al., 2019; Breshears et al., 2018; Nir et al., 2008), and magnetoencephalography (MEG) (Luckhoo et al., 2012; Stam, 2010; Brookes et al., 2011; Hipp et al., 2012), provide the first set of evidence supporting the neuronal origins of fMRI RSNs. Other large-scale phenomena, i.e., CAPs and traveling waves, have also been detected in human EEG data (Ding et al., 2022). Human MEG studies report several short-lived large-scale brain states, which show similar spatial patterns as fMRI RSNs (Vidaurre et al., 2018). The second set of evidence is from animal studies. These large-scale patterns have been observed in both lower mammals (e.g., mice and rats) and non-human primates (e.g., macaque), indicating their neuronal origins and intrinsic natures to brain functions from the perspective of preserved brain organizations in species evolution (Majeed et al., 2011; Hutchison and Everling, 2012; Gozzi and Schwarz, 2016). In particular, RSNs have been extensively studied and reported in fMRI from both mice and monkey (Hutchison and Everling, 2012; Sforazzini et al., 2014; Hutchison et al., 2011) and have been replicated with similar patterns in recent monkey ECoG studies (Hindriks et al., 2018; Vezoli et al., 2021). Large-scale propagating waves have been reported in early animal neurophysiological studies (Huang et al., 2010; Stroh et al., 2013; Steriade et al., 1993a; Luczak et al., 2007) while without the whole-brain coverage of recording. Recent calcium-based optical imaging techniques directly image whole-brain dynamics of propagating waves in mice (Matsui et al., 2016). At the same time, many fMRI studies have reported global nonspecific patterns (Fox et al., 2009) and global dynamic events like GPWs in mice (Matsui et al., 2016), rats (Majeed et al., 2009), and macaque monkey (Gu et al., 2021). Similarly, global nonspecific patterns have been identified in monkey from ECoG (Wen and Liu, 2016), while CAPs have been reported in mice from fMRI (Gutierrez-Barragan et al., 2019) and in monkey from ECoG (Liu et al., 2015). Several studies have directly compared brain-wide patterns in both human and animals, which all lead to the conclusion of strong similarities in the spatiotemporal properties of large-scale patterns across species (Wen and Liu, 2016; Gu et al., 2021).

While early EEG/MEG human studies (Brookes et al., 2011; Yuan et al., 2016) have been conducted with the primary aims to better understand and validate large-scale phenomena from fMRI, due to their inherent nature of high temporal resolution (i.e., milli-seconds, ms), recent EEG/MEG human studies have indicated novel and complementary information beyond fMRI, i.e., frequency-specific (Vidaurre et al., 2018; Becker and Hervais-Adelman, 2020) and fast dynamics (~100 ms) (Baker et al., 2014; Ding et al., 2022). However, time-resolved frequency-specific large-scale patterns have not been reported in animals considering these are relatively new developments in humans as well. At the same time, noninvasive technologies for whole-brain imaging of neuronal electrical activities in small animals have been made possible (Matsui et al., 2016) and limited whole-brain imaging datasets are also available in large non-human primates via invasive recordings (i.e., macaque Nagasaka et al., 2011). These provide the opportunity to study the correspondence of time-resolved large-scale electrical phenomena between animals and humans beyond time-averaged hemodynamic patterns, as well as to establish the homologues of human large-scale electrical patterns of frequency-specificity and fast dynamics in animal models. The use of such animal models is of significant values and represents a critical component in further understanding such large-scale phenomena and their contributions to behaviors as they could be experimentally manipulated, e.g., pharmacological treatment and neurostimulation, that are not possible in human participants.

In the present study, we investigated the presence of CAPs in macaque monkeys, representing large-scale transient neuronal events that have been identified in human EEG (Ding et al., 2022). To test this hypothesis, we used ECoG data of four macaque monkeys, available from the public domain (Yanagawa et al., 2013; Nagasaka et al., 2011), with the high-density electrode coverage of single/both hemisphere(s) and high-density EEG data from 34 human participants recorded in our own lab. Our results identified a set of frequency-specific (i.e., alpha band) transient CAPs in monkey ECoG data, which shared the similar spatial, temporal, and transitional properties as the set of CAPs identified from human EEG data. These results provide the first set of evidence, suggesting that transient large-scale phenomena and their time-frequency signatures are preserved across species beyond those time-averaged large-scale phenomena from fMRI.

2. Methods and materials

2.1. Monkey data acquisition and preprocessing

ECoG data of four macaque monkeys (C, G, K, and S) were obtained from a public source (neurotycho.org). Detailed methods and protocols of experimental, surgical, and recording procedures could be found in literatures (Yanagawa et al., 2013; Nagasaka et al., 2011). All studies were approved by the RIKEN ethics committee and in accordance with the recommendations of the Weatherall report, “The use of non-human primates in research”.

Briefly, a customized 128-channel electrode array (Unique Medical, Japan) was implanted into the subdural space of the left hemisphere of each monkey, except for Monkey S having one experiment of 256 channels available with a 128-channel electrode array on both the left and right hemisphere of the brain. Array coverage was contiguous over the parietal, temporal, occipital, and frontal lobes with electrode spacing at 5 mm. There was also middle wall coverage on Monkey C and G. The electrodes were 3 mm in diameter and consisted of platinum discs. Reference electrodes were implanted in the subdural space and ground electrodes in the epidural space. A total of 32 experiments were conducted over different days and monkeys. Each experiment included the eyes-closed waking condition and one of the following three conditions: ketamine/medetomidine anesthesia, propofol anesthesia, or natural sleep except for the resting experiment for Monkey C and S consisting solely of the eyes-closed waking condition. The monkey was sitting calmly, and the ECoG signals were recorded for up to 20 min during this eyes-closed condition before the administration of anesthesia drugs or natural sleep. For the experiments of anesthesia, the loss of consciousness (LOC) was defined behaviorally (see details in Nagasaka et al., 2011) and after that, neural activity, which was characterized by slow-wave activity, was recorded for ~25 min with heart rate and breathing monitored carefully. For the experiments that included the natural sleep condition, electrooculography (EOG) and electromyography (EMG) signals were monitored and used together with ECoG to determine the onset of sleep (see details in Nagasaka et al., 2011). All ECoG and other electrical recordings, i.e., EMG and EOC, were sampled at 1 kHz by the Cerebus data acquisition system (Blackrock, UT, USA).

There were 32 recordings (Table 1) from four monkeys. In the present study, we focused our analysis on ECoG data from awake conditions and, therefore, resting data (awake with eyes covered) before the administration of anesthesia drugs or natural sleep were extracted firstly, which resulted in ECoG data of length from 5 to 20 min, and then independently preprocessed using EEGLAB (Delorme and Makeig, 2004). Each extracted dataset was filtered using a notch filter with the band-stop frequency at 50 Hz to remove powerline noises. Zero to four channels with excessive motion artifacts were removed in individual datasets and then interpolated with data from their surrounding channels. Each dataset was then bandpass-filtered from 0.5 Hz to 150 Hz and re-referenced to the common average. It is noted that no ECoG segments were rejected to maintain the continuity of data.

Table 1.

Available Monkey ECoG Recordings with resting conditions, including the number, length of recordings, and the number of electrodes.

Monkey ID	# of Recordings	Length of Recordings (mins) (mean±std)	Electrode Coverage	# of Electrodes
S	3	20.8 ± 1.1	Left hemisphere	128
	1	5.0 ± 0.0	Both hemispheres	256
C	13	11.9 ± 5.3	Left hemisphere	128
G	12	12.3 ± 4.0	Left hemisphere	128
K	3	12.2 ± 0.1	Left hemisphere	128

Note Monkey S has the only one both-hemisphere recording from all four monkeys.

2.2. Human data acquisition, preprocesssing and electrophysiological source imaging

2.2.1. Dataset

The human EEG data used in the present analysis was acquired in our previous reported study (Shou et al., 2022). Briefly, resting-state high-density EEG data was obtained from 34 healthy participants (age: 24±5 years, range 18–38 years, 9 females) with informed consents. The study was approved by the Institutional Review Board at the University of Oklahoma Health Science Center (OUHSC). Each dataset had a duration of 10 min and was sampled at 1 kHz using a 128-channel Amps 300 amplifier (Electrical Geodesics Inc., OR, USA). Structural MRI was gathered from each participant as well using a GE MR750 scanner with GE’s “BRAVO” sequence: FO=240 mm, axial slices per slab=180, slice thickness=1 mm, image matrix=256×256, TR/TE=8.45/3.24 ms.

2.2.2. EEG preprocessing

All datasets were preprocessed using the exact same precedure on individual participant’s dataset using EEGLAB (Delorme and Makeig, 2004). Each individual dataset was filtered by a band-pass filter of 0.5–100 Hz, and a notch filter of 58–62 Hz. Noisy channels and independent components (ICs) of artifacts related to ocular, muscular and cardiac activities, were identified by the FASTER plugin (Nolan et al., 2010) with the support of visual inspections as well. Artifactual ICs of ocular/muscular artifacts and obsessive movements were then removed and data on the identified noisy channels were interpolated with data from their surrounding channels. Finally, each dataset was down-sampled to 250 Hz and re-referenced to the common average. It is noted that no EEG segments were rejected to maintain the continuity of data.

2.2.3. Cortical source imaging

To map human EEG signals to the similar recording space (i.e., epicortical surface) and spatial resolution as monkey ECoG data, electrophysiological source imaging was performed to reconstruct brain current sources over the cortical surface from scalp EEG, which deconvoluted the effects of volume conductors of the skull (Lai et al., 2005) and therefore led to cortically distributed current sources in the comparable space to ECoG. Via registering the cortical surface from each individual participant to the standard template for human cortex, reconstructed cortical current source data from different human individuals could be converted into the same spatial domain to perform the group-level analysis. These mapping steps transformed data into the unified space (i.e., the cortex) with anatomical definitions of various human brain functional regions, which facilitated the interpretation of obtained CAP results within human participants as well as across the species of human and monkey. In details, individual structural MRI data were segmented using FREESURFER (Fischl, 2012) to extract the surfaces of the scalp, skull, and brain for building individual volume conduction models, and the interface between white and gray matters for the cortical current density (CCD) source model. These surfaces were tessellated into triangular elements (volume conduction model: 10,242 nodes and 20,484 triangles, CCD model: 20,484 nodes and 40,960 triangles). For the CCD model, the nodes on the medial wall adjoining the corpus callosum, basal forebrain, and hippocampus, were excluded from the source space, resulting in a total number of 18,715 sources. The electrical conductivities of the scalp, skull, and brain for the volume conduction models were assigned as 0.33/Ωm, 0.0165/Ωm, and 0.33/Ωm, respectively (Lai et al., 2005). Electrode locations were registered on the scalp surface by aligning three landmark fiducial points digitized at EEG sessions to those identified from anatomical MRI for each individual participant. Based on these models, the boundary element method (Hamalainen and Sarvas, 1989) was used to compute the forward relationship between the scalp EEGs and current source amplitudes of all nodes on individual CCD models. The minimum-norm estimate (Hamalainen and Ilmoniemi, 1994) was then used to reconstruct current source amplitudes of all nodes on individual CCD models from recorded EEG signals at the resolution of individual timeframes, leading to a time series of cortical tomography at the same temporal resolution of preprocessed EEG data (i.e., 250 Hz). See more details in (Shou et al., 2022).

2.3. CAP analysis via clustering

To obtain CAPs from monkey preprocessed ECoG data and human cortical tomographic current source (cTCS) data, a series of processing steps were performed for the preparation of input data to the clustering analysis. Firstly, each dataset from individual recording sessions of individual monkeys or individual human participants was filtered using the FIR filter from EEGLAB into data for different frequency bands, i.e., delta (1–4 Hz), theta (5–7 Hz), alpha (8–12 Hz), beta (13–30 Hz), and gamma (31–100 Hz). The following analysis only focused on the alpha-band data from each dataset as the alpha rhythm is the dominant components in human resting-state EEG (Kirschfeld, 2005) and its existence in monkeys has been reported (Popov and Szyszka, 2020). Secondly, instantaneous amplitudes of time courses of alpha-band ECoG data on individual electrodes or alpha-band cTCS data on individual CCD nodes were calculated individually using the Hilbert transform (Hipp et al., 2012; Baker et al., 2014). Thirdly, while the human cTCS data and monkey ECoG data are in the similar spatial domain (cortical surface vs epicortical surface), the spatial dimension of the human cTCS data was much larger than the spatial dimension of the monkey ECoG data (18,715 CCD nodes vs 128/256 ECoG electrodes), which might have a biased impact on the performance of the clustering analysis. Therefore, a dimension reduction approach was adopted from the previous CAP analysis (Shou et al., 2022), in which 100 regions of interest (ROIs) were defined over the cortex for each hemisphere based on an atlas and each ROI represented a parcel of cortical units of functional similarities (Schaefer et al., 2018). ROI-level instantaneous amplitude data for human were calculated as the average of the cTCS instantaneous amplitude data from all cortical sources within the same ROI, which reduced the spatial dimension of human cTCS data to 200 (left and right hemispheres together) that were close to the spatial dimension of monkey ECoG data. Fourthly, both ROI-level instantaneous amplitude data for human and electrode-level instantaneous amplitude data for monkey were then converted to z-scores via a within-session normalization (i.e., subtracting the mean and dividing the standard deviation across all time points of each recording session per ROI/electrode). Finally, the K-means clustering was performed using the L1-norm distance as the measure on these normalized data with the number of clusters varying from 2 to 20 and the optimal number of clusters (i.e., optimal k-value) for each analysis was selected based on the metric of percentage of explained variance being larger than 90 % (Schaefer et al., 2018). The K-means clustering analysis was performed at the group level in human, in which human ROI-level instantaneous amplitude z-scores were concatenated across individual participants. On the contrary, to evaluate the reproducibility of CAPs from different recording sessions of the same monkey, the K-means clustering analysis was performed at individual sessions for each monkey and statistics for quantitative measures of CAPs (see below) were calculated separately for different monkeys. This was designed because the ECoG electrode coverage and positions were different in different monkeys and the group-level analysis was thus not able to be conducted across monkeys.

2.4. Spatial and temporal metrics for evaluating similarities/differences between ECoG and EEG CAPs

The outputs of the clustering analysis labeled each timeframe data from either human or monkey to one of identified CAPs. The brain-wide tomography of each CAP was defined as the averaged map across all timeframes (z-scores) belonging to the CAP separately for individual recording sessions in monkeys and altogether for data from all human participants (the group level). To assess the hemispheric symmetry of CAPs, the hemispheric symmetry index was calculated as the Pearson correlation between values of matched parcels on left and right hemispheres from the above used cortical surface atlas (Schaefer et al., 2018). Multiple temporal metrics calculated for individual CAPs for indivdidual recording sessions based on the cluster labels included occurrence, lifetime, fractional occupancy (FO) rate, and direct transition rate. It was noted that there were multiple recording sessions per monkey but only one recording session per human participant and, therefore, calcuated temporal metric values could also be considered as samples per participant in human. An occurrence of a CAP was defined as multiple consecutive time frames that were labeled toward the same CAP, and the occurrence rate of a CAP was the ratio between the total number of occurrences of a specific CAP and the total number of occurrences of all CAPs in each recording session. Lifetime of a CAP was calculated for each occurrence as the total time of consecutive timeframes belonging to one occurrence of the CAP, and the mean lifetime of a CAP was firstly averaged over all occurrences in individual recording sessions, and then averaged over all sessions for individual monkeys or all human participants. FO was defined as the total number of timeframes belonging to a CAP divided by the total number of timeframes in each recording session. These session-level measures were then statistically compared between different CAPs or different CAP groups (see the section of Results) using paired t-tests (unless otherwise noted) with the Bonferroni correction in individual monkeys. In humans, participant-level measures were similarly statistically compared between different CAPs or different CAP groups. Direct transition among CAPs were defined as the transition from a CAP occurrence to a different CAP occurrence. The direct transition rate of a CAP (at time t) to other CAPs (at time t + 1) defined as the numbers of transitions from the CAP to other CAPs, divided by the total number of occurrences of the CAP. The transition rates among a set of CAPs from one analysis were plotted as a net using igraph in R (Csárdi et al., 2025). Each net consists of a number of nodes (each for one CAP) and arrowed lines for directional direct transition rates among two CAPs. Other spatial and temporal measures were coded in node properties (e.g., color and size). The layout of a net was determined using the force-directed layout algorithms (the Kamada Kwai algorithm) in igraph.

3. Results

The CAPs results reported below from the number of clusters (i.e., optimal k-value) that explained at least 90 % of the variance of data in each analysis (note that the CAPs obtained under other k-values indicate similar patterns). The mean optimal k-values were consistent across four monkeys ranging between 12 and 13 (Fig. 1B) and the k-values of individual sessions in each monkey also indicated small changes (Fig. 1C). The optimal k-value at the same explained variance (90 %) on grouped data from human is 8 (Fig. 1A).

Fig. 1.

Explained variance as a function of the number of clusters. The optimal k-value in each dataset (marked as a red ‘x’) was selected as the smallest number of clusters that provides >90 % of explained variance in data. (A) The optimal k-value determination in human data from the group of 34 participants. (B) Means and standard deviations of optimal k-values for four monkeys obtained in individual recording sessions (Table 1). (C) Explained variances as a function of the number of clusters in four monkeys (green shaded regions: standard deviations).

3.1. Spatial patterns of alpha-band ECoG and EEG CAPs

Fig. 2 shows the spatial patterns of the group-level CAPs on inflated cortical surfaces for human and the individual-session CAPs on epicortical surfaces from the both-hemisphere ECoG recording (Table 1) in Monkey S while the representative examples of individual-session CAPs from single-hemisphere ECoG recordings in all four monkeys (all left hemispheres, Table 1) are shown in Fig. 3. To verify the clustering on single-hemisphere data should generate similar CAPs as both-hemisphere data, independent clustering analyses on left/right-hemisphere only ECoG data separated from the only both-hemisphere recording in Monkey S were performed. Our results indicate most CAPs identified from the both-hemisphere data could be identified using the left/right-hemisphere only data (Supplementary Figure 1). It is therefore assumed that the CAPs from single-hemisphere data in all four monkeys are representative to CAPs that would be identified if both-hemisphere data are available.

Fig. 2.

Spatial patterns of human and monkey CAPs from alpha-band electrical data covering both hemispheres at their optimal k-values. Human CAPs were obtained from the group of 34 participants and monkey CAPs were obtained from the only both-hemisphere ECoG data in Monkey S. The CAPs were sorted out according to their activation strengths from top to bottom (for human CAPs, sorted CAPs are from left to right first and then from top to bottom). hCAP: the first CAP; lCAP: the last CAP; slCAP: the second last CAP; numerical values on each CAP: the range of activation strengths in human and the absolution maximal activation strength in monkey.

Fig. 3.

Spatial patterns of representative alpha-band CAPs from data covering left hemisphere only in all four monkeys at their optimal k-values. The CAPs were sorted according the same criteria as in Fig. 2.

The CAPs identified from human and monkey show the patterns of large-scale distributed activations in both species, which could be visually conveniently grouped into three different categories. CAPs 5, 6, and 7 in humans (Fig. 2A) have all ROI-level z-scores higher than their corresponding means (zero) of individual ROIs and CAPs 3, 4, 6, 10, 12, and 13 in Monkey S (Fig. 2B) have all electrode-level z-scores higher than their corresponding means across all electrodes, indicating global high-activations (the GHA group). Meanwhile, CAPs 2, 3, and 8 in humans and CAPs 2, 5, and 8 in Monkey S exhibit z-scores less than zero, indicating global low-activations (the GLA group). The rest CAPs from human EEGs and Monkey S ECoGs have both positive and negative z-scores, indicating non-global activations (the NGA group). These three groups could be equally identified in the CAPs from the left-hemisphere only ECoG recordings of all monkeys (Fig. 3). Their consistent detections in different recording sessions of different monkeys and humans suggest the reproducibility of such transient large-scale events both within and across species. Moreover, the decreasing magnitudes of CAPs, reflected similarly in both the global activations (i.e., sum of ROI/electrode-level z-scores over the whole brain) and maximal ROI/electrode-level activation, are observed in the GHA and GLA groups in both humans and monkeys. The activation magnitudes of the NGA CAPs are in general between those from the GHA and GLA groups in terms of the maximal ROI/electrode-level activation. Here three distinct CAPs, i.e., the highest global-activation CAP (CAP 5 from humans and CAP 12 from Monkey S, termed as hCAP, Fig. 2) from the GHA group, the lowest global-activation CAP (CAP 8 from humans and CAP 5 from Monkey S, termed as lCAP), and the secondary lowest global-activation CAP (CAP 2 from both humans and Monkey S, termed as slCAP), are further noted since they exhibit distinct temporal and transitional characteristics (see discussions below). Similar hCAPs, lCAPs, and slCAPs are identified in the single-hemisphere only ECoG data from all four monkeys (Fig. 3). These CAPs in general show much broader brain-wide co-(de)activations and the highest/lowest maximal ROI/electrode-level activations, respectively, and are consistently detected in different sessions from monkeys, in different monkeys, and in different species.

Both EEG and ECoG CAPs show the distinct patterns of symmetry between two hemispheres. The symmetric pattern is dominant in human CAPs (Fig. 2A) with the symmetric index over all CAPs as 0.785±0.136. It is also largely present in CAPs from Monkey S (Fig. 2B), especially for the CAP 12, 4, 13, and 3 from the GHA group, the CAP 8, 2, and 5 from the GLA group, and the CAP 1 and 9 from the NGA group (the symmetric index over these CAPs was 0.576±0.221). Some monkey CAPs seem less symmetric due to different strengths of co-activations (areas of positive z-scores), but still symmetric in terms of the relative spatial distributions of co-activations and co-deactivations (areas of negative z-scores), e.g., weakened co-activations within the left visual cortex in CAP 6. The hemispheric symmetry of some CAPs is preserved in a pair of CAPs (the CAP 7 and 11 pair). It was not possible to verify the hemispheric symmetry of CAPs in other monkeys due to lack of right-hemisphere recordings, while the symmetric CAP pairs could be similarly identified when both-hemisphere data from Monkey S were analyzed separately for left and right hemispheres (Supplementary Figure 1).

While the brain-wide co-(de)activations prevail in CAPs, some CAPs (particularly the NGA CAPs) also reveal regional focuses if only co-activations are concerned (Figs. 2 and 3). High bilateral co-activations within the frontal cortex are detected in human EEGs (i.e., CAP 1) and monkey ECoGs (CAP 1 for both-hemisphere and CAP 10 for left-hemisphere in Monkey S; CAP 8 in Monkey C; CAP 12 in Monkey G; and CAP 6 in Monkey K). High bilateral co-activations within the posterior-parietal cortex are detected in human CAP 4 and CAP 11 for left-hemisphere in Monkey S; CAP 5 in Monkeys C and G; and CAP 3 in Monkey K. Furthermore, high bilateral co-activations within the anterior temporal cortex are shown in human CAP 7 and corresponding high bilateral precentral temporal co-activations are shown in CAP 3 for both both-hemisphere and left-hemisphere in Monkey S; CAP 6 in Monkey G; and CAP 7 in Monkey K. More monkey CAPs indicating distributed regional co-activations could be consistently identified across different sessions and monkeys, including high co-activations in the temporal and occipital cortices (CAP 1 in Monkey S; CAP 11 in Monkey G and K) and in the frontal and parietal cortices (CAP 12 in Monkey C; CAP 7 in Monkey G; and CAP 2 in Monkey K).

3.2. Temporal patterns of alpha-band ECoG and EEG CAPs

Three temporal metrics were calculated for both group-level (across all participants) human EEG CAPs and group-level (across all datasets for each monkey) monkey ECoG CAPs i.e., FO, occurrence, and lifetime (Fig. 4). In general, the GHA CAPs have relatively low occurrences and FO rates and the GLA CAPs have relatively high FO rates. Slightly higher lifetimes are observed in the GLA and GHA CAPs as compared with the NGA CAPs. The differences of these temporal metrics were statistically compared between individual CAPs (i.e., hCAP, lCAP, or slCAP) or between CAP groups (i.e., the GHA and GLA groups) of distinct spatial patterns, as well as between individual CAPs or CAP groups to a reference, which was calculated from their corresponding NGA CAPs.

Fig. 4.

Temporal metrics (i.e., occurrence rate, mean lifetime, and FO) of both human and monkey alpha-band CAPs. In human, each of these temporal metrics was calculated on individual participants and then summarized with mean and standard deviation values for the CAPs belonging to the GHA, GLA, and NGA groups (left three columns). In monkey, each metric was calculated on individual recording sessions and then summarized separately for each individual monkeys. The right three columns present the temporal metric data for hCAP, lCAP, and slCAP. The statistical comparisons were performed between all possible pairs of right three columns in each plot and one of right three columns to the NGA group the left (serving as the reference) and the corresponding p-values were labeled (Bonferroni corrected).

The first key observation is that both human and monkey hCAPs always show significantly lower occurrences than their corresponding lCAPs (Human and Monkey C, G: p < 0.005, K: p < 0.01, S: p < 0.05) and slCAPs (Human and Monkey C, G, S: p < 0.005, K: p < 0.01). When compared with the session/participant reference means, all human and monkey hCAPs (Human and Monkey C, G, S: p < 0.005, K: p < 0.05) and most lCAPs (Human and Monkey C, G: p < 0.005, S: p < 0.05, K: lower but not significant) have significantly lower occurrences. Meanwhile, the occurrences of slCAPs in both species are always significantly higher than those of the lCAPs (Human and Monkey C, G, S: p < 0.005, K: p < 0.01), hCAPs (Human and Monkey C, G, S: p < 0.005, K: p < 0.01), and reference means (Human and Monkey C, G, K: p < 0.005, S: p < 0.01). Secondly, the hCAPs (Human and Monkey C, G: p < 0.005, K, S: p < 0.01) and lCAPs (all p < 0.005) have significantly higher mean lifetimes than the reference means while the lifetimes of the most slCAPs from both species show no significant differences. Notably, the overall mean lifetimes for human and monkey CAPs are both at the scale of 100 ms. Finally, as the FO rates are determined by both their occurrences and lifetimes, the differences between hCAPs/lCAPs/slCAPs and their respective reference means are largely enhanced. The FO rates of all hCAPs from both species are significantly lower than the reference means (Human and Monkey C, G, S: p < 0.005, K: p < 0.05) and the FO rates of all lCAPs and slCAPs from all monkeys are significantly higher than the reference means (all p < 0.005).

3.3. Transitional patterns of alpha-band ECoG and EEG CAPs

Alpha-band CAP direct transition rates in both human and monkey data indicate similar structured patterns that reveal moment-to-moment dynamics between brain states coded differently in the CAPs from the GHA, GLA, and NGA groups in individual sessions, which are highly supported by the spatial and temporal patterns of CAPs discussed above. Firstly, the transitions between the GHA CAPs and the GLA CAPs are largely mediated by the NGA CAPs (no direct transitions between the GHA CAPs and the GLA CAPs in both Figs. 5 and 6). Secondly, the transitions between all hCAPs and their corresponding lCAPs (e.g., between CAP 3 and CAP 5 in human and between CAP 5 and CAP 12 in Monkey S in Fig. 5) indicate the longest paths that usually visit multiple other brain states coded by CAPs from all three groups. Similar transition patterns are also typical in single-hemisphere data from four monkeys (Fig. 6 and Supplementary Figure 2). Lastly, it is noted that there are no direct transitions between all hCAPs and their corresponding lCAPs in data from all sessions (as listed in Table 1).

Fig. 5.

Transitional patterns of alpha-band human CAPs (A) and monkey CAPs (B) from the same datasets in Fig. 2. Each transition map is plotted (after removing the transitions of rate < 0.05 for clarity) with the size of circles coded with occurrence rate (left) and mean lifetime (right), to illustrate the relationship of occurrence/lifetime of CAPs with their relative positions in the transition map. In particular, both hCAP and lCAP (insets) have low occurrence rates, long mean lifetimes, and are located on the boundary of transition maps. Red circles: GHA CAPs; blue circles: GLA CAPs; gray circles: NGA CAPs. Arrows: directed transitions; color of arrows: the start CAP for the transition; thickness of arrow lines: transition rates from the start CAP. L: left; R: right.

Fig. 6.

Transitional patterns of alpha-band monkey CAPs from the same four left-hemisphere only ECoG datasets in Fig. 3 for Monkey C, K, G, and S. Illustrations are same as Fig. 5 with the size of circles coded with occurrence rates. For illustrations with the size of circles coded with mean lifetimes, see Supplementary Figure 2.

Different roles of the GLA, GHA, and NGA CAPs in brain-state transitions are co-incident with the differences in occurrence and lifetime metrics of the CAPs from these groups. In general, the CAPs with relatively high occurrences and short lifetimes appear in the middle of the transition maps while the CAPs with relatively low occurrences and long lifetimes show on the boundary of the transition maps. Both lCAPs and hCAPs represent the extreme brain states on both ends of the transition maps, which both have the lowest occurrences and the longest lifetimes (Fig. 4). Furthermore, the slCAPs that have the highest occurrences (Fig. 4) are usually in the gateway positions of the corresponding lCAPs to other brain states, e.g., CAP 2 (slCAP) for CAP 3 (lCAP) in human data (Fig. 5A) and CAP 10 (slCAP) for CAP 9 (lCAP) in Monkey K data (Fig. 6).

Beyond the roles played by individual CAPs in transitions, statistical properties of transitions were further investigated in two aspects. Firstly, the transition probability as a function of the distance between CAPs was calculated as the percentages of direct transitions at certain distances out of total direct transitions pooled from all CAPs of all human participants or monkeys. Results from both human and monkey CAPs indicate power-law decay distributions (Fig. 7A). With sufficient data available from individual monkeys, the same investigations repeated in individual monkeys indicate the consistent power-law decays. It is noted the large fluctuations observed at large distances in human data were due to low number of CAPs (i.e., 8) and same CAPs for all participants, which led to few number of distinct distances and few number of transitions at large distance values. Secondly, due to the special roles of both lCAPs and hCAPs in the transition maps and there were no direct transitions among them, it was of interests to study statistical properties of such long-path transitions via other CAPs. Fig. 7B shows the percentages of transitions from lCAP to hCAP (lCAP->hCAP, blue curves) as a function of transition time in both human participants and monkeys in comparison to a reference, i.e., from lCAP ending with another lCAP without reaching hCAP (lCAP->lCAP, black curves). In both human and monkey, the patterns from both transitions could be split around 300 ms for analysis. Within the short transition times (< 300 ms), lCAP->hCAP exhibits an early peak with the narrowed width in comparison to lCAP->lCAP in both human and monkey, which indicates that lCAP->hCAP might be driven by more temporally organized neuronal events. Within the long transition times (> 300 ms), both transitions follow the similar exponential decays in both human and monkey. The t-tests on the lifetimes of lCAPs at the start of lCAP->hCAP indicate that lCAPs have statistically significantly shorter lifetimes than those at the start of lCAP->lCAP in both human (125.01 ms vs 166.85 ms, p < 0.001, corrected) and monkeys (S: 83.02 ms vs 223.18 ms, p < 5e-13, corrected; C: 103.32 ms vs 159.34 ms, p < 5e-12, corrected; G: 81.29 ms vs 165.83 ms, p < 5e-41, corrected; K: 87.05 ms vs 118.74 ms, p < 5e-6, corrected). Furthermore, the long-path transition started with hCAPs were also similarly analyzed, but no such a difference observed in the lifetimes of hCAPs between hCAP->lCAP and hCAP->hCAP. The fact adds the evidence that lCAP->hCAP might be more temporally organized.

Fig. 7.

Statistical properties of transitions. (A) the power-law decay relationship between the probability ( %) of direct transitions of pairs of all CAPs and their distances pooled over all human participants (left) or all monkeys (right, including plots for individual monkeys as well). (B) Probabilities ( %) of the transition of lCAP->hCAP and its reference transition of lCAP->lCAP as a function of transition time pooled over all human participants (left) or all monkeys (right).

3.4. Similarities between EEG CAPs and ECoG CAPs

Table 2 summarizes the key similarities between human EEG CAPs and monkey ECoG CAPs, in which five spatial features, six temporal features, and eight transitional features were revealed. These similarities appear at two different levels: direct observations on identified patterns, e.g., brain-wide spatial patterns and similar lifetimes (~100 ms); and differences among multiple members of identified patterns, e.g., hCAP and lCAP defined in each session, and low occurrence of hCAP and high occurrence of slCAP in each session. It is noted that the majority of similarities between human CAPs and monkey CAPs are observed at the second level. Moreover, the similarities from both temporal and transitional measures show relatively higher consistencies than the similarities from spatial patterns, where some of spatial patterns, e.g., the hemispheric symmetry and regional focuses of co-activations, only show moderate levels of similarities. Nevertheless, the fact that large number of similar features were identified spanning different domains and consistencies of differences among different CAPs in these features cross three domains all indicate the strong correspondence between EEG CAPs from human and ECoG CAPs from monkeys.

Table 2.

Summary of the key similarities between the alpha-band human EEG and monkey ECoG CAPs.

Domain of Patterns	Description of Features	Human EEG	Monkey ECoG
Spatial patterns	Three groups: GHA, GLA, and NGA	Yes	Yes for all monkeys
	hCAP and lCAP	Yes	Yes for all monkeys
	Ordered changes of activation strengths across and within three groups	Yes	Yes for all monkeys
	Bilateral symmetry	Yes for all CAPs	Yes for the majority of CAPs
	Regional focuses of coactivations	Several	Many CAPs; CAPs
		CAPs	showing similar regional focuses as in human CAPs
Temporal measures	Low occurrence of hCAP	Yes	Yes
	High occurrence of slCAP	Yes	Yes
	Long lifetime of hCAP and lCAP	Yes	Yes
	Low FO for hCAP	Yes	Yes
	High FO for slCAP	Yes	Yes
	Overall mean lifetime	~100 ms	~100 ms
Transitional patterns	hCAPs and lCAPs at the two ends of the transition maps	Yes	Yes
	No direct transition between the GHA CAPs and the GLA CAPs	Yes	Yes
	Transitions between hCAP and lCAP mediated by other GHA and GLA CAPs	Yes	Yes for the majority of session data in all monkeys
	Low/high occurrences for CAPs at the end/middle of transition maps	Yes	Yes
	Long/short lifetime for CAPs at the end/middle of transition maps	Yes	Yes
	Power-decay distribution with distance	Yes	Yes
	An early peak of narrow width of tranistioin times in lCAP-> hCAP	Yes	Yes
	Significant lifetime difference of lCAPs between lCAP->hCAP and lCAP->lCAP; but not in hCAPs between hCAP->lCAP and hCAP-> hCAP	Yes	Yes

4. Discussion

The present study was set to investigate transient large-scale coactivation patterns in monkey neuronal electrical data, which was motivated by the similar recent findings in human EEG data (Ding et al., 2022; Shou et al., 2022). Using high-density ECoG data with single/both-hemispherical coverage on the epicortical surfaces of four macaque monkeys, a set of transient large-scale neuronal electrical patterns in the alpha band was identified. The direct comparisons between monkey ECoG CAPs and human EEG CAPs indicated significant correspondence in multiple feature domains. These findings suggest that transient large-scale coactivations exist in non-human primates (i.e., macaque monkeys) and their multi-domain features are preserved across species, which support that fast brain-wide activations are of functional relevance in both monkey and human brains.

CAPs from monkey ECoG and human EEG share a long list of similarities. Spatially, both ECoG CAPs and EEG CAPs reveal a set of brain-wide patterns that largely exhibit coactivations symmetric between two hemispheres (Fig. 2). Strong hemispheric symmetries have been widely observed in large-scale networked activations of slow dynamics (e.g., RSNs) from fMRI (Smith et al., 2009), functional near-infrared spectroscopy (Eggebrecht et al., 2014), and EEG/MEG (Brookes et al., 2011; Yuan et al., 2016) in resting human brains, which are time-averaged due to either sluggish response time of signals, e.g., 6–10 s for hemodynamic signals, or long time windows used in estimating connectivity metrics, e.g., several seconds to minutes. While large-scale patterns time-resolved to sub-seconds have only been scarcely reported in humans, several recent studies indicate that these transient coactivations are also of hemisphere-mirrored symmetries (Vidaurre et al., 2018; Ding et al., 2022). Hemispherical symmetries have been observed in large-scale networked activations from idling animal brains as well, including monkeys (Hutchison et al., 2011; Gu et al., 2021), in both hemodynamic (Matsui et al., 2016) and neural electrical signals (Stroh et al., 2013). Among the entire sets of CAPs identified in both monkey ECoG and human EEG data in the present study, many of them indicate either global co-activations (i.e., the GHA CAPs) or global co-deactivations (i.e., the GLA CAPs), which are consistent with the previous reported resting human EEG CAPs (Ding et al., 2022) and global nonspecific patterns in monkey ECoG (Wen and Liu, 2016). Moreover, global nonspecific patterns exist in fMRI (Fox et al., 2009), which need to be regressed out as physiological noises (Desjardins et al., 2001; Aguirre et al., 1998) to reveal classical RSNs of distinct anatomical distributions, while recent studies indicate that global nonspecific patterns contains neuronally relevant information (Murphy and Fox, 2017; Schölvinck et al., 2010). The global-high activations of the GHA CAPs and their low occurrences characterize them similar to the transient brain states of high dynamic functional connectivity from fMRI occurred at resting states (Kupis et al., 2021). Such a correspondence has been enhanced due to their similar age-dependent changes (Shou et al., 2022; Tian et al., 2018; Kupis et al., 2021).

Temporally, both ECoG CAPs and EEG CAPs are short-lived patterns at ~100 ms and ECoG and EEG envelope data variances over time could be largely explained by a relatively small set of CAPs in both monkeys (i.e., 11–13 CAPs) and humans (i.e., 8 CAPs). While at first glance, this lifetime seems related to alpha-band signals that oscillate around 10 Hz (i.e., 1/10 Hz = 100 ms), it is important to note that CAPs were obtained on the envelopes of alpha-band signals, which typically have a spectral power distribution peaked at much lower frequencies than original signal (e.g., ~1 Hz). Moreover, the lifetimes have been observed in a relatively large range for different human and monkey CAPs, i.e., roughly from 50 to 200 ms (Fig. 4) that correspond a range of frequency from 5 Hz to 20 Hz, which is wider than the frequency band used to obtain alpha-band EEG/ECoG (i.e., 8 – 12 Hz). The lifetime in the range of 50 to 200 ms has been similarly reported in the studies of brain states using microstate (Michel and Koenig, 2018; Coquelet et al., 2022) and hidden Markov model (HMM) (Baker et al., 2014; Vidaurre et al., 2018) in human EEG/MEG data. Among them, HMM-based brain states further share some spatial similarities with EEG CAPs (e.g., hemispherical symmetry) although they seem much closer to traditional fMRI RSNs (Vidaurre et al., 2018). In idling animal brains, the UP and DOWN states of slow-wave oscillations of membrane potentials (Jercog et al., 2017) show atomic wavelet events at the timescale of 10 to 100 milliseconds (Steriade et al., 1993b). Action potential firings monitored by Ca²⁺ recordings in animals have revealed waves with a latency of ~80 milliseconds between the visual and frontal cortices (Stroh et al., 2013). However, the direct brain-wide correspondence has not been demonstrated between these large-scale cellular dynamics (e.g., oscillations and waves) and macroscopic spatial patterns detectable in functional neuroimaging data. Enabled by novel wide-field optical imaging systems for monitoring Ca²⁺ signals over the entire cortical surface in small animals (Matsui et al., 2016), similar transient large-scale co-activations expressed in propagational waves have been reported. Another recent study (Gu et al., 2021) has analyzed the same monkey ECoG data as the present study and revealed propagational patterns in non-human primates. However, both studies have focused on slow dynamics (< 0.1 Hz) to directly compare with those from low-frequency hemodynamic signals. Other studies on monkey local field potential data have also revealed different large-scale patterns, e.g., multiple spectral and temporal gradients over cortical areas, that exhibit task-dependent changes (Hoffman et al., 2024) but are however not of transient nature. Similarly, the correlation-based analysis on this same ECoG data has revealed time-averaged brain-wide patterns similar to fMRI RSNs from monkeys (Liu et al., 2014). The present study reveals, for the first time to our knowledge, fast dynamics of large-scale coactivations in monkeys, which are at a similar timescale as previously reported in human EEG CAPs (Ding et al., 2022; Shou et al., 2022). Meanwhile, the relatively small numbers of brain states revealed in both ECoG CAPs and EEG CAPs in awake resting conditions are consistent with the number of brain states (typically 4–8) discovered in other neuroimaging data, e.g., resting-state fMRI, in both humans (Calhoun et al., 2014) and animals (Gutierrez-Barragan et al., 2019). It is important to note that these two temporal characteristics (i.e., lifetime and number of brain states) are reproduced with similar values independently from each individual session of four macaques, which attest to their authenticity and reliability.

Both alpha-band ECoG CAPs and EEG CAPs further reveal a similar superstructure (Ding et al., 2022) indicating a dynamic relationship among the entire set of identified CAPs. The superstructure consists of brain states at different global energy levels (i.e., the GHA, GLA, and NGA groups) and of layered activation strengths especially within the GHA group. The global high-energy brain states usually are stable (i.e., long lifetime) but have low chances to be visited (i.e., low occurrence) and the opposites for the global low-energy brain states. There are usually no direct transitions between the global high-energy and global low-energy brain states, which are of large spatial distance in the high dimension space defining CAPs (i.e., 200 parcels). Visits between the two need to be mediated via non-global-activation brain states (creating short spatial distances). This is supported by the power-law decay of direct transitions as a function of distance when all data from human or monkeys are pooled together for statistics (Fig. 7A). The transitions from the weakest global low-energy brain states to the strongest global high-energy brain states take the longest time with the need of visiting more intermediate brain states (including the CAPs from the same group). These multi-domain patterns show congruent evidence for different roles played by individual CAPs during brain state transitions in such a superstructure. While the underling mechanism of such a superstructure is unknown, the transitional characteristics among its composing CAPs suggest that it might be caused by brain-wide propagations of neuronal activations similar to those reported in animals (Ferezou et al., 2007; Stroh et al., 2013) and humans (Massimini et al., 2004). Certain brain-wide propagational patterns of neuronal origins are reserved in both animal (Gutierrez-Barragan et al., 2019) and human (Gu et al., 2021; Khan et al., 2022) hemodynamic data even after a significant low-pass filtering process during neurovascular coupling (Logothetis et al., 2001). The investigations on the long-path transitions between lCAPs and hCAPs in both human and monkey data in the present study further reveal that the transitions from lCAPs to hCAPs usually takes shorter but more uniform durations when such transitions finish within 300 ms (Fig. 7B) and the durations of starting brain states (i.e., lCAPs) need to be significantly low for the transitions from lCAPs to hCAPs to happen (not ends with another lCAP) while this is not needed in the transitions from hCAPs to lCAPs. Altogether, the underlying mechanism behind the superstructure in monkey ECoG data might be well structured propagations and their detections not only indicates the fact that such patterns are preserved during evolution, but also suggest their similar functional roles in different species. However, the exact neurophysiological function of these structured transitions among identified brain states remains an open question. Propagations of slow-wave oscillations on the time scale of 100 s of milliseconds have been largely reported in sleep studies (Sheroziya and Timofeev, 2014) indicating their importance in memory consolidation. Cellular-level phenomena at the similar time scales (Mitra et al., 2018) as discussed above suggest the possible mechanism of cortical excitability control (Raut et al., 2020) or for long-range information integration (Stroh et al., 2013). Future works are needed to study these structured transitions in the contexts of behaviors and mental states.

It is noted that both ECoG CAPs and EEG CAPs are obtained from band-limited data, i.e., the alpha band (8–12 Hz). Their frequency-specific nature has also been observed in other large-scale patterns identified from monkey electrical data at both resting (Hindriks et al., 2018) and task conditions (Chao et al., 2018; Giraud and Arnal, 2018; Vezoli et al., 2021). The resting monkey studies have further demonstrated the differences of time-averaged large-scale patterns among different frequency ranges, which attests the need of selecting a band for investigating time-resolved large-scale patterns (i.e., CAPs). In the present study, only the alpha band was selected as it is the dominant idling rhythm in resting human and monkey brains. Future studies will be needed to understand if such time-resolved large-scale patterns are also frequency dependent. Moreover, several differences were observed between ECoG CAPs and EEG CAPs. The number of discovered monkey CAPs ranged between 11 and 13, but in humans, 8 were found. This could be likely due to the higher signal-to-noise ratio (Ball et al., 2009) and spatial resolution (Engel et al., 2005) of ECoG over EEG. Increased spatial resolution might also be the reason that more monkey CAPs show regional coactivation patterns. Meanwhile, given that there are more observed similarities than differences between ECoG CAPs and EEG CAPs, the findings suggest a capacity of noninvasive high-density EEG for studying transient large-scale brain events with rooms for further improvement, e.g., via integrating EEG and MEG (Ding and Yuan, 2013; Boto et al., 2019). The preservation of these transient large-scale coactivations through evolution indicate the availability of plausible animal models to further study their roles in brain functions and potentials for clinical diagnosis and treatment evaluations (Auer, 2008; van den Heuvel and Hulshoff Pol, 2010) using more invasive approaches (Hutchison and Everling, 2012). The animal models of both lower mammals (e.g., mice and rats) and non-human primates (e.g., macaque) have been popular in investigating large-scale patterns, including RSNs (Hutchison and Everling, 2012; Sforazzini et al., 2014; Hutchison et al., 2011), propagating waves (Huang et al., 2010; Stroh et al., 2013; Steriade et al., 1993a; Luczak et al., 2007; Matsui et al., 2016; Gu et al., 2021), and CAPs (Gutierrez-Barragan et al., 2019; Liu et al., 2015). These brain-wide patterns have been found of abnormal alterations in a broad spectrum of brain diseases, e.g., Alzheimer’s disease (Sorg et al., 2007; Supekar et al., 2008), schizophrenia (Looijestijn et al., 2015), autism (Yamasaki et al., 2017), and epilepsy (Blumenfeld, 2012). Complementary to excellent spatial resolutions provided by fMRI on revealing spatial structures of these brain-wide phenomena, time-resolved frequency-specific EEG and ECoG CAPs are expected to provide opportunities to obtain more knowledge about their dynamic structures.

The present study is limited in the following aspects. Firstly, due to the invasiveness and costs of large animal studies, the amount of monkey ECoG data available is insufficient for the group-level analysis. We compromised the direct group-level comparison to the comparison between group-level human data and individual-level monkey data, which was not optimal. Secondly, most of ECoG data available in four monkeys were from the left hemisphere only. We addressed this issue by building the equivalence of CAP results from the both-hemisphere data in Monkey S and its left-hemisphere only and right-hemisphere only data (Supplementary Figure 1). The bilateral symmetry of these transient large-scale coactivations also reduced such a concern. Potential future solutions to address the above two limitations are to establish the links of these phenomena in lower mammals (e.g., mice and rats) (Jonak et al., 2020), which has been the case in studies of other brain-wide patterns (Huang et al., 2010; Stroh et al., 2013; Steriade et al., 1993a; Luczak et al., 2007; Matsui et al., 2016; Gutierrez-Barragan et al., 2019). Lastly, to build full confidence in the correspondence of transient large-scale coactivations between humans and non-human primates, use of the exact same recording modality in both species may be beneficial, such as EEG in monkeys or ECoG in selected human participants (e.g., those with epilepsy) (Ding et al., 2007; Betzel et al., 2019).

Supplementary Material

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.neuroimage.2025.121408.

Data availability

Data will be shared in the public domain.