NeuroImage 60 (2012) 2062–2072

Contents lists available at SciVerse ScienceDirect

NeuroImage

j ourna l homepage: www.e lsev ie r .com/ locate /yn img

Spatiotemporal dynamics of the brain at rest — Exploring EEG microstates aselectrophysiological signatures of BOLD resting state networks

Han Yuan, Vadim Zotev, Raquel Phillips, Wayne C. Drevets, Jerzy Bodurka ⁎Laureate Institute for Brain Research, 6655 South Yale Avenue, Tulsa, OK 74136, USA

⁎ Corresponding author. Fax: +1 918 502 5135.E-mail address: jbodurka@laureateinstitute.org (J. B

1053-8119/$ – see front matter © 2012 Elsevier Inc. Alldoi:10.1016/j.neuroimage.2012.02.031

a b s t r a c t

a r t i c l e i n f o

Article history:Accepted 13 February 2012Available online 22 February 2012

Keywords:BOLD fMRIEEGMicrostatesResting state networksICA

Neuroimaging research suggests that the resting cerebral physiology is characterized by complex patterns ofneuronal activity in widely distributed functional networks. As studied using functional magnetic resonanceimaging (fMRI) of the blood-oxygenation-level dependent (BOLD) signal, the resting brain activity is associ-ated with slowly fluctuating hemodynamic signals (~10 s). More recently, multimodal functional imagingstudies involving simultaneous acquisition of BOLD-fMRI and electroencephalography (EEG) data have sug-gested that the relatively slow hemodynamic fluctuations of some resting state networks (RSNs) evinced inthe BOLD data are related to much faster (~100 ms) transient brain states reflected in EEG signals, that arereferred to as “microstates”.To further elucidate the relationship between microstates and RSNs, we developed a fully data-driven ap-proach that combines information from simultaneously recorded, high-density EEG and BOLD-fMRI data.Using independent component analysis (ICA) of the combined EEG and fMRI data, we identified thirteen mi-crostates and ten RSNs that are organized independently in their temporal and spatial characteristics, respec-tively. We hypothesized that the intrinsic brain networks that are active at rest would be reflected in both theEEG data and the fMRI data. To test this hypothesis, the rapid fluctuations associated with each microstatewere correlated with the BOLD-fMRI signal associated with each RSN.We found that each RSN was characterized further by a specific electrophysiological signature involving fromone to a combination of several microstates. Moreover, by comparing the time course of EEG microstates tothat of the whole-brain BOLD signal, on a multi-subject group level, we unraveled for the first time a set ofmicrostate-associated networks that correspond to a range of previously described RSNs, including visual,sensorimotor, auditory, attention, frontal, visceromotor and default mode networks. These results extendour understanding of the electrophysiological signature of BOLD RSNs and demonstrate the intrinsic connec-tion between the fast neuronal activity and slow hemodynamic fluctuations.

© 2012 Elsevier Inc. All rights reserved.

Introduction

The activity of the resting brain, as measured using blood-oxygenation-level-dependent (BOLD) functional magnetic resonanceimaging (fMRI), reveals spontaneous, large-amplitude, low-frequency(b0.1 Hz) fluctuations that are temporally correlated and spatially orga-nized into specific functional networks, collectively termed the “restingstate networks” (RSNs) (Biswal et al., 1995; Fox and Raichle, 2007).Multiple RSNs can be identified using seed-based correlation mappingto activity measured over spatial loci selected a priori (Biswal et al.,1995), or can be delineated using data-driven approaches, such as inde-pendent component analysis (ICA) (Beckmann et al., 2005; Calhoun etal., 2001; Damoiseaux et al., 2006; Mantini et al., 2007). The RSNs are

odurka).

rights reserved.

evident in the human brain during the awake resting state, as well asduring task performance, sleep and anesthesia (Fox and Raichle,2007), or at different vigilance level (Olbrich et al., 2009). In experimen-tal animals RSNs also have been demonstrated under anesthesia(Vincent et al., 2007) and wakeful rest (Liang et al., 2011). Each RSN re-sembles either the spatial pattern of BOLD responses elicited by activebehavioral tasks involving the task-relevant visual, sensorimotor, audi-tory, or attentional modality (Smith et al., 2009), or by the spatial net-work of regions that are active in a resting state but decrease activityin most tasks (i.e., the “default” mode network) (Greicius et al., 2003;Raichle et al., 2001). More recent evidence suggests that spontaneousbrain activity plays an important role in regulating behavioral responsesby providing endogenous or top-downmodulation of sensory or cogni-tive processing (Boly et al., 2007; Hesselmann et al., 2008). Meanwhile,emerging evidence shows that neurological or psychiatric diseases areassociated with alterations in RSN activity (Fornito and Bullmore,2010). Thus, the spontaneous BOLD activity of RSNs reflects a funda-mental aspect of cerebral physiology.

2063H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

The neurophysiological basis of the BOLD RSNs is not fully under-stood, however. The BOLD signal is an indirect measure of neural ac-tivity (Logothetis, 2008), and the extent to which slow fluctuations ofRSNs at the level of ~10 s reflect faster neuronal dynamics at the mil-lisecond level remains unclear. If indeed BOLD RSNs are linked to fastneuronal activity it is also not clear what are physiological relevantmechanisms establishing such link. Although substantial evidencelinks task-evoked BOLD responses to spiking activity (Logothetis etal., 2001), local field potentials (Logothetis et al., 2001; Mukamel etal., 2005), and macro-level recording (such as electroencephalogra-phy, EEG) (Yuan et al., 2011), the electrophysiological correlates ofspontaneous BOLD activity in the RSNs have not been established. Aseminal finding in this regard, however, was the report that slow fluc-tuations (also b0.1 Hz) of band-limited power measured invasivelyusing intracranial electrodes were correlated between brain areaswithin (He et al., 2008; Leopold et al., 2003) and across hemispheres(Nir et al., 2008). Moreover, to investigate potential links betweenspontaneous neural electrophysiological activity and BOLD fluctua-tions, several studies simultaneously measured EEG and BOLD-fMRIsignals to assess band-limited EEG power as electrophysiologicalcorrelates for RSNs. The EEG power in the alpha (8–13 Hz) or beta(13–30 Hz) frequency band was correlated with the spontaneousBOLD signal in brain regions that partially overlap with some RSNs,particularly in the visual/parietal cortex (Goldman et al., 2002; Laufset al., 2003; Moosmann et al., 2003; Sadaghiani et al., 2010). Anotherconcurrent EEG and BOLD fMRI study correlated the time course of sixRSNs with the time course of EEG power across a wide spectrum offrequency bands (1–50 Hz) (Mantini et al., 2007). A diverse electro-physiological profile was obtained for these six RSNs that showedeach RSN was correlated with EEG activity at all frequency bands;however, signatures specific to each RSN were not identified. Sponta-neous electrophysiological activity as reflected by EEG does not gen-erally exhibit a single frequency or even single frequency bandoscillations, and instead a wide spectrum of rhythms generally is col-lectively represented in the EEG signal (Buzsaki and Draguhn, 2004;Varela et al., 2001). The neural activity at a specific frequency bandthus is unlikely to constitute the electrophysiological correlate of anRSN.

Recently, microstates of the EEG signal have been proposed as po-tential electrophysiological correlates of spontaneous BOLD activity(Britz et al., 2010; Musso et al., 2010). The topographic representationof the EEG scalp electrical field affords temporal resolution at themillisecond-level, yet does not change randomly or continuouslyover time, remaining stable over periods of ~100 ms. Such quasi-stable and unique topographic distributions of the electrical field po-tential have been termed “microstates” (Lehmann, 1980; Lehmann,1990; Lehmann et al., 1987). The EEG microstate reflects the summa-tion of concomitant neuronal activity across brain regions rather thanactivity specific to any frequency band (Koenig et al., 2002; Lehmann,1980; Lehmann et al., 1987; Wackermann et al., 1993). Microstates inspontaneous activity have been associated with abstract thoughtsarising as subjects rest in an awake state (Lehmann et al., 1998) andthe pre-stimulus states also have been shown to influence the subse-quent task-evoked BOLD response (Arieli et al., 1996; Britz et al.,2009). Moreover, EEG microstate analysis was used to differentiatesubjects with panic disorders from healthy controls (Kikuchi et al.,2011), suggesting potential applications in psychiatric research.

Two previous studies used concurrent EEG and BOLD-fMRI to in-vestigate the correlation between microstates and spontaneousBOLD activity (Britz et al., 2010; Musso et al., 2010). In both studiesmicrostates were segregated into multiple groups based upon themaximum spatial dissimilarity between groups. A microstate timecourse was derived using the spatial correlation between instanta-neous topography and each group, and then compared to the BOLDsignal over entire brain using a general linear model approach. Specif-ic brain regions in which the BOLD activity correlated to the

microstate time courses also were identified, and these regionsappeared to partially overlap with known RSNs, suggesting that mi-crostates may serve as informative electrophysiological correlates ofthe spontaneous BOLD activity. However, in both studies the spatialsimilarity was the key factor used to ascribe microstates to differentgroups and to determine the time course of each microstate, irrespec-tive of the temporal information contained in the EEG recording.Therefore, the topographies on the scalp associated with microstatesoriginating from distinct brain areas could have been partially corre-lated due to the volume conduction effect (Nunez, 1995), and yetcould have been ascribed into the same group. Moreover, the spatialdissimilarity between groups did not guarantee temporal indepen-dence (e.g., microstates 1 and 2 in Britz et al., 2010 were strongly cor-related). Thus, when comparing the time course of microstates to thatof the BOLD signal, partially correlated microstates would eliminatethe independence across networks, resulting in overlap between thenetworks associated with various microstates.

In order to overcome this limitation, and to investigate the extentto which the BOLD RSNs reflect neuronal activity, we developed anovel, data-driven approach to extract the microstates in EEG usingindependent component analysis (ICA). We applied ICA to decom-pose the EEG recording into thirteen spatially distinct microstatesthat showed maximal temporal independence from each other. Inthe meantime, ten RSNs were identified from simultaneously collect-ed BOLD fMRI data using spatial ICA. We hypothesized that the intrin-sic RSNs would be reflected in EEG activity as well as in BOLD activity.To test this hypothesis, we examined the temporal correlation be-tween the independent microstates and BOLD activity within RSNs.Each RSN was characterized by a specific electrophysiological signa-ture involving one, two, or a combination of several, microstates. Tofurther examine the spatial specificity of temporal independentmicrostates with respect to their relationship to a particular RSN,the time courses of the microstates were compared to that of thewhole-brain BOLD signal in a general linear model and, for the firsttime, electrophysiological correlates were found for a wide range ofnetworks, including the canonical visual, motor, auditory, attention,frontal, visceromotor and default mode networks.

Material and methods

Subjects and experimental protocol

The study was conducted at the Laureate Institute for Brain Re-search, and was approved by the University of Oklahoma InstitutionalReview Board (IRB). Nine healthy, right-handed subjects (meanage=33±10 years; one female) participated in the study. All partici-pants providedwritten informed consent as approved by the Universityof Oklahoma IRB. The subjects received financial compensation for theirparticipation.

Simultaneous EEG and fMRI data were recorded on each subjectover three functional imaging sessions, each lasting 6 min and 10 s.Subjects were instructed to rest with eyes closed and to remainawake. Upon debriefing none of the subjects reported sleeping duringthe study sessions and subsequent inspection of the EEG confirmedthe self-reports. The data from all nine subjects (27 functional ses-sions in total) were included in the analysis.

Simultaneous EEG/fMRI recording

High-density EEG signals from 126 channels were recorded usingMRI-compatible BrainAmpMR Plus amplifiers (Brain Products GmbH,Munich, Germany). The sintered Ag/AgCl ring electrodes weremounted into a scalp cap according to the standard 10–5 system. Allelectrodes were referenced to the FCz position, while a ground elec-trode was located at the AFz position. One additional electrode wasplaced on the subjects' back to monitor the electrocardiographic

2064 H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

signal. The impedance of all electrodes was maintained below 10 KΩthroughout the recording. The internal sampling clock of the EEG ampli-fierwas synchronizedwith theMRI scanner 10 MHzmaster clock signalusing the SyncBox device (Brain Products GmbH, Munich, Germany), inorder to prevent variant sampling of imaging artifacts and to facilitateartifact correction (Mandelkow et al., 2006). The signals were recordedat a sampling frequency of 5000 Hz with an analog filter (from 0.016 to250 Hz) and a resolution of 0.1 μV.

Imaging was performed on a General Electric (GE) DiscoveryMR750 whole-body 3 Tesla MRI scanner (GE Healthcare, Waukesha,Wisconsin, USA). The scanner is equipped with a scalable 32-channel digital MRI receiver capable of performing massively parallelfMRI in real time (Bodurka et al., 2004). A standard 8-channelreceive-only head coil array was used for MRI signal reception.Whole-brain BOLD fMRI was conducted using the echo planar imag-ing (EPI) sequence. A single-shot gradient-recalled EPI sequencewith Sensitivity Encoding (SENSE) was employed (34 axial 2.9-mmthick interleaved slices with 0.2-mmgap; TR/TE=2000 ms/30 ms;ma-trix size=64×64; FOV=220 mm×220 mm; sampling bandwidth=250 kHz, 185 volumes). At this spatial resolution (3.4×3.4×2.9 mm3),cardiac and respiratory related physiological noise dominates thenoise in fMRI time series (Bodurka et al., 2007). In a previous studywe have shown that the contribution of physiological noise can bereduced while preserving sensitivity to BOLD-related signal changesby lowering the flip angle (Gonzales-Castillo et al., 2011). Another ad-vantage of reducing the flip angle for studies involving concurrentfMRI and EEG recordings is that this technique reduces the levels ofradio-frequency (RF) energy deposition,which decreases the likelihoodof localized heating at the EEG electrodes. We thus set the flip angle at30°.

Anatomical images were acquired using a T1-weightedmagnetization-prepared rapid gradient-echo (3D MPRAGE) sequencewith SENSE (124 contiguous axial slices with 1.2 mm slice thickness;matrix size=256×256; FOV=220 mm×198 mm; TR/TE=5 ms/1.98 ms; flip angle=8°; sampling bandwidth=31.25 kHz).

A photoplethysmograph with an infra-red emitter placed under thepad of the subject's left index finger was used to obtain pulse oximetry-based recordings of the heart rate, and a pneumatic respiration belt wasused to record respiratory motion. The recording of the cardiac pulseand respiratory data was synchronized with the fMRI data acquisition.Both waveforms were sampled at a frequency of 50 Hz.

In order to evaluate the quality of EEG data recorded inside thescanner, the EEG also was recorded separately from three of thenine subjects in a dark and quiet environment outside the scanners'magnetic field using the same amplifier and electrode configurations.Subjects were instructed to lie in a supine position and rest with theireyes closed. Three sessions were recorded, each lasting 6 min and30 s. 126-channel EEG and one channel ECG signals were acquiredat a sampling frequency of 1000 Hz using an analog filter (from0.016 to 250 Hz) and at a resolution of 0.1 μV.

fMRI preprocessing

The data processing steps for the concurrently acquired fMRI andEEG data are schematically illustrated in Fig. 1. The EPI image prepro-cessing was performed using Analysis of Functional NeuroImagessoftware (AFNI, http://afni.nimh.nih.gov/) (Cox, 1996). The first fivevolumes of each fMRI run were excluded from analysis to allow theBOLD signal to reach steady state. Fluctuations in the time seriesdata at the respiration and cardiac frequencies and their first har-monics were removed using the RETROICOR method (Glover et al.,2000). fMRI time series fluctuations related to the respiration volumeper time (RVT) were also removed with delays of 0, 5, 10, 15 and 20 susing the method of Birn et al. (Birn et al., 2006; Birn et al., 2008). ThefMRI data then were corrected for slice timing and rigid-body motion,co-registered to anatomical images, spatially transformed into the

stereotaxic array of Talairach and Tournoux (1988) and smoothedusing a Gaussian kernel of FWHM=4 mm.

Spatial ICA of fMRI

Spatial ICA was used to decompose the EPI time series into indepen-dent brain activity patterns originating from the spatial covariance of themeasured signals. The preprocessed fMRI data were subjected to groupspatial ICA using the GIFT toolbox (http://mialab.mrn.org/software/gift/index.html). The order of independent components (ICs)was estimatedto be 30, which was determined from all individual data sets using theMinimum Description Length (MDL) criterion (Li et al., 2006;Rissanen, 1983). The MDL criterion selected order will help to avoidusing too few independent components due to possible merging of rel-evant ICs into one, and will help to avoid spurious ICs due to too manycomponents. All twenty-seven scanning sessions from nine subjectswere concatenated and reduced to 30 temporal dimensions using prin-ciple component analysis (PCA), from which independent componentswere estimated using the infomax algorithm (Bell and Sejnowski,1995). A back-reconstruction was subsequently performed to obtainthe waveforms of ICs for each session (across TR). Thus for each IC, thewaveform corresponded to the time course of a specific pattern ofbrain activity and the intensity of this pattern was represented by thespatial map of the IC. The intensity values in each map were convertedto Z-scores and thresholded at Z=2.0 (Damoiseaux et al., 2006). Theresulting components and their time courses were screened and thoseassociated with nuisance processes (i.e. motion, large vessels, ventriclesand susceptibility) were excluded. From the remaining components tenwere selected as anatomically relevant areas, which notably appearedconsistent with previously reported resting networks (Damoiseaux etal., 2006). These components represent endogenous patterns of brainactivity that aremaximally independent from each other and coherentlyconnected within each pattern, which are referred to as resting statenetworks (Damoiseaux et al., 2006; Mantini et al., 2007).

EEG preprocessing

The EEG data acquired simultaneously with the fMRI data were cor-rected for artifacts related to the gradient switching and cardiac ballisticeffect using the template subtraction method (Allen et al., 1998; Allenet al., 2000) implemented in BrainVision Analyzer software (Brain Prod-ucts GmbH, Munich, Germany). Residual ballistocardiac artifacts in theEEG signals were removed using the independent component analysismethod. The de-noised data were subsequently band-pass filtered from1 Hz to 70 Hz, downsampled to 250 Hz, and re-referenced to the com-mon average reference. For the EEG signals recorded outside the scanner,data were similarly band-pass filtered from 1 Hz to 70 Hz, downsampledto 250 Hz, and re-referenced to the common average reference.

Microstate analysis of EEG

Detailed steps of the microstate analysis are illustrated in Supple-mental Fig. 2. The original microstates were derived as EEG topologiesat the local peaks of global field power (GFP) of preprocessed andartifact-corrected EEG data. These microstates consist of arrays ofvoltage values from all electrodes extracted at the time points oflocal GFP peaks. In order to identify the most dominant and distinc-tive topologies, namely the main microstates, we developed a data-driven approach based on temporal ICA. Rather than clustering mi-crostates based on their topographical similarity, we applied ICA todecompose the microstates into several sources whose time courseswere maximally independent from each other. After the original mi-crostates were derived from multi-channel EEG data, their topologieswere concatenated across sessions and subjects, and then subjectedto a single group ICA analysis using the EEGLAB toolbox (http://sccn.ucsd.edu/eeglab/). The total variance of all channels was

Fig. 1. Schematic illustration of data analysis steps conducted on concurrently acquired high-density EEG and BOLD fMRI data. The lower panel shows examples of the main EEGmicrostates (MS) and BOLD resting state networks (RSNs); their corresponding time courses from a single session also are shown.

2065H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

balanced across subjects before concatenation. The number of ICs inmicrostates was matched with that of the spatial ICs in fMRI, i.e. 30,since we hypothesized that independent brain networks are com-monly reflected in EEG as well as in fMRI. A back-projection wasperformed to obtain maps associated with each IC, i.e. the main mi-crostate. Thus for each IC, the map corresponds to the spatial topologyof a specific microstate and the intensity of the microstate is reflectedin the temporal trace of the IC, which is maximally independent fromothers. The same decomposition matrix was applied to the continu-ous EEG, which yielded an intensity value for each IC at each timepoint. The ICs and their waveforms were examined and those associ-ated with nuisance processes (i.e. eye movements, residual ballisto-cardiac artifact and muscle activity) were excluded, resulting inthirteen ICs, i.e. thirteen temporal independent microstates. Due tothe oscillating nature of EEG signals, only the absolute intensityvalues of ICs are considered as the time course of microstates. As sug-gested in previous studies, the critical characterization of a microstatein terms of the scale-free property depends upon the microstate's rela-tive duration (Van de Ville et al., 2010).We thus labeled themicrostatesbased upon their intensity values and then neutralized their absolutevalues. Specifically, a microstate was assigned to a time point based onwhich microstate has the maximal absolute intensity value. This

approach yielded normalized time courses for the microstates, com-posed of series of 0′s and 1′s, with the value of 1 being assigned to ami-crostate only at the time point when the absolute intensity value of thatmicrostate exceeded that of every other microstate. In order to accom-modate the different temporal scales of EEG and fMRI, the normalizedtime courses of the microstates were convolved with a double-gammahemodynamic response function (HRF) (Friston et al., 1998) and thendown-sampled to TR.

In a secondary analysis, we instead extracted the microstates usingthe spatial clustering method described in Britz et al. (2010). Usingthe EEG topologies at the local GFP peaks the original microstateswere subjected to clustering analysis, andmicrostates of similar topolo-gy were grouped together, resulting in four main microstates. The spa-tial correlation was calculated between the continuous EEG and eachof the main microstates to obtain the time courses of the microstates.As described above, the time courses of the microstates then were con-volved with the HRF and down-sampled to TR.

Combined microstate and fMRI analysis

We hypothesized that the intrinsic network activity of the brain iscommonly reflected in the EEG as well as in the fMRI data. To test this

2066 H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

hypothesis, we first examined the temporal correlation between thetime courses of the EEG microstates and the fMRI RSNs extractedusing ICA. As schematically illustrated in Fig. 1, a pair-wise correlationwas calculated between thirteen temporal independent microstatesand ten RSNs across all sessions, and the correlation was thresholdedat the significance level of pb0.001 (uncorrected).

To further assess the spatial specificity of a particular microstaterelative to that of the RSNs, we examined the extent to which thetime course of a microstate coincides with that of the BOLD signal atthe voxel-level using a general linear model (GLM) analysis. Thetime course of a microstate was used as a regressor in the GLM. Themotion parameters and the regressors related to RETROICOR andRVT were modeled as nuisance variables in the GLM to further re-move the motion and physiological noise. A set of Legendre polyno-mials (up to three-order polynomials per session) also was includedas regressors of no interest to remove nonspecific temporal drifts inthe BOLD signal. All subjects' data were pooled together in the GLMto assess the effect of each microstate at a group level using thefixed-effect model. Separate GLMs were formed for different micro-states. Statistical images were obtained using voxel-wise t-tests. Theresulting statistics were corrected for multiple comparisons usingthe false discovery rate (FDR) and thresholded at voxel-levelqb0.05. While the temporal correlation calculated above indicatedwhich microstates were related to a particular RSN, the statistical im-ages indicated the extent to which this correlation existed across thewhole brain. Toward this end the statistical images associated witheach microstate were compared to the fMRI RSNs extracted by ICA.

Results

Microstates and resting state networks identified from simultaneousrecordings

After preprocessing, the artifact-free EEG signals from 126 chan-nels (Supplemental Fig. 1) were subjected to microstate analysis. Asample segment of the GFP is shown in Supplemental Fig. 2, togetherwith the time courses of three sample independent componentsdecomposed from the original microstate topographies using ICA.The decomposed components constituted the main temporal inde-pendent microstates and at each time point, the type of microstatewas determined based on the time courses of the ICs. As shown inSupplemental Figs. 2B and D, the topographies of the ascribed mainmicrostates to a large extent approximate the original topographies,while maintaining maximum temporal independence. At the grouplevel thirteen main temporal independent microstates (Fig. 2A)were extracted from ICA (labeled as MSk, k=1…13). The thirteenmi-crostates were identified after excluding those components related to

Fig. 2. Thirteen temporal independent EEG microstates (labeled as MSk

nuisance processes (i.e., eye movement, residual ballistocardiac arti-fact and muscle activity). The microstates MS1, MS6, MS7, MS8, andMS9 appeared to originate from parietal and/or occipital cortex. Themicrostates MS2, MS3, and MS4 had contributing sources from thesensorimotor cortex, while microstates MS5 and MS13 were domi-nated by temporo-parietal sources. The other microstates (MS10,MS11 and MS12) appeared to originate from the medial or lateralfrontal cortex. An example of single-session time courses of threemicrostates after convolution with HRF is shown in Fig. 1 (and inSupplemental Fig. 2E). All main microstates time courses featurelow frequency temporal fluctuations.

The RSNs were identified from the BOLD signal using spatial ICA(exemplified in Fig. 1). The RSNs consisted of maximally independentpatterns and associated individual time courses. Ten RSNs were iden-tified across subjects (Fig. 2B), which are consistent with those previ-ously reported (Damoiseaux et al., 2006; Mantini et al., 2007). RSN1corresponded to portions of the canonical default-model network(Greicius et al., 2003; Raichle et al., 2001), and involved the bilateralangular/middle temporal gyrus, posterior cingulate gyrus, medialprefrontal cortex (PFC) and bilateral superior frontal gyri. RSN2 andRSN3 corresponded to sensorimotor networks involving the primarymotor and sensory cortices and the supplemental motor area, togeth-er with foci in the post-central (RSN2) and pre-central (RSN3) gyri,respectively. RSN4 and RSN5 resembled the previously described “at-tention networks” (Corbetta and Shulman, 2002) that predominantlyinvolved the left and right hemispheres, respectively, and includedthe dorsolateral PFC, dorsomedial PFC/dorsal anterior cingulate cor-tex, middle temporal gyrus, posterior cingulate gyrus and superiorparietal cortex. RSN6 involved the temporal and insular cortex, andpart of the extended visceromotor network (Ongür and Price,2000). RSN7 and RSN8 corresponded to two visual networks, withRSN7 involving more lateral visual areas and RSN8 involving moremedial visual areas. RSN9 mainly involved the superior temporalarea along with bilateral superior frontal and middle frontal areas,corresponding to the auditory-phonological network (Mantini et al.,2007). RSN10 included the medial PFC, and the dorsal anterior andposterior cingulate cortices. These areas formed a frontal networkand have been found to be involved in executive control and workingmemory function (Miller and Cohen, 2001). The ten RSNs (1–10)identified here to a large extent correspond to the RSNs reported byDamoiseaux et al. (2006; i.e., RSNs ‘B’, ‘F’, ‘F’, ‘C’, ‘D’, ‘I’, ‘A’, ‘E’, ‘I’, ‘J’,respectively).

The temporal independent microstates identified at the grouplevel generally were evident also at the single subject level. Supple-mental Fig. 3A shows data from a representative subject in whichnine of the main microstates were evident in the EEG recordings ac-quired simultaneously with fMRI (Ak, k=1…9). The nine temporal

, k=1..13) (A) and ten BOLD RSNs (B) obtained at the group level.

2067H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

independent microstates (1–9) closely approximated the topogra-phies of the microstates identified at the group level (MS1, MS2,MS3, MS5, MS7, MS10, MS11, MS12, and MS13, respectively). In thesame subject, nine main microstates also were identified from EEG re-cordings acquired outside scanner (Supplemental Fig. 3B). The topog-raphies of the microstates obtained inside the scanner appearedsimilar to those obtained outside the scanner for the three subjectsstudied in both environments. Identified in all individual subjects,temporal independent EEG microstates are shown in supplementalFig. 5. These single-subject-level microstates are to a large extent sim-ilar with the topologies of the thirteen group-level microstates (as inFig. 2).

Correlation between microstates and resting state networks

To test the hypothesis that common RSNs identified in the BOLDfMRI data also are reflected in concurrently acquired EEG data, we in-vestigated whether the identified temporal independent microstatesserved as electrophysiological correlates of the BOLD RSNs. We per-formed correlational analyses to assess the relationships betweenthe time courses of the RSNs and those of the microstates, and theirpair-wise correlation coefficients are shown in Fig. 3 (thresholded atpb0.001, uncorrected). These results revealed that each BOLD RSNhad a characteristic electrophysiological signature that involved oneor two microstates, or a combination of several microstates. The re-sults conversely showed that each microstate was related to one orup to a few RSNs. A sub-group of themicrostates (MS 1–6) correlatedto only one or two RSNs, whereas microstate MS13 correlated to sev-eral RSNs. Interestingly, a sub-group of the microstates (MS 7–12),which originated from occipito-temporo-parietal (MS 7–9) or fron-tal areas (MS 10–12), shared a similar profile in regard to the RSNs,corresponding largely to visual and sensorimotor networks. Two mi-crostates (MS5 and MS12) showed time courses that reciprocallycorrelated to different networks (for MS5, attention (L) versus vis-ceromotor; for MS12, attention (R) versus sensorimotor/visual/auditory).

General linear model analyses based on microstates

While the above correlations assessed temporal relationships be-tween the time courses of the RSNs and temporal independent micro-states, we further examined the extent to which each microstaterelated spatially to the BOLD RSNs. The time course of each microstatewas entered into the GLM as a regressor to map correlations with thelocal BOLD signal across the entire brain. The resulting statisticalmaps (Figs. 4 and 5) revealed the brain areas where the BOLD signalcorrelated significantly with the relevant microstate's time course.The microstate-informed statistical maps also were compared to thespatial patterns of the BOLD RSNs, identified independently from

Fig. 3. Pair-wise correlation coefficient (CC) between the time courses of thirteen tem-poral independent EEG microstates and ten BOLD RSNs. Values were threshold at thevoxel level of pb0.001 (uncorrected). Note that the relationship between microstatesand RSNs falls into three categories. MS 1–6 each correlated with only one or twoRSNs. In contrast, MS 7–12 are similarly correlated to sensorimotor, visual and auditorynetworks. MS 13 was correlated to several RSNs.

the fMRI data alone. As shown in Fig. 4A, MS1 correlated to theBOLD signal in the bilateral angular/middle temporal gyrus, anteriorand posterior cingulate gyrus, and medial prefrontal cortex, whichcorresponded to regions implicated in the default mode networkand the frontal network. MS2 identified regions of the sensorimotorcortex, supplemental motor area, and superior temporal gyrus, repre-senting the sensorimotor network and the auditory network. MS4 in-volved regions of the middle frontal gyrus, superior parietal lobule,middle temporal gyrus, and posterior cingulate gyrus, resemblingthe functional anatomical pattern of the attention network. MS4also involved the bilateral pre-central gyrus, as part of the sensorimo-tor network. Similarly, MS5 was related to the BOLD signal from themiddle frontal, middle temporal and posterior cingulate gyri, represent-ing an attention network. Interestingly, the attention network identi-fied by MS5 localized predominantly to the left hemisphere, whichcorresponded to the left-lateralized attention network extracted fromthe fMRI data. MS5 also involved the superior temporal gyrus, as partof the visceromotor network. MS6 andMS3 correlated to regions impli-cated in the default mode network and the sensorimotor network,respectively.

In contrast to the microstates (MS 1–6) that only were associatedwith one or two RSNs, the other microstates (MS 7–13) shown corre-lated with more than two RSNs. Moreover, an overlapping set of re-gions was identified from MS 7–12 (exemplified by representativestatistical maps from MS7 in Fig. 5A). The regions in which thesemicrostate spatial topographies overlapped included the lateral andmedial visual cortices, the superior temporal gyrus, the pre- andpost-central gyri, and the supplemental motor area, which corre-sponded to the sensorimotor, lateral/medial visual and auditory net-works. Another microstate, MS13, identified regions that resemblethe default mode network, the frontal network, the attention net-works and the medial visual network (Fig. 5B).

In order to assess whether the RSNs can be identified by temporalindependent microstates at the single-subject level, the microstatesderived from the EEG data of a representative subject were used asregressors in the GLM and compared to the fMRI data acquired simul-taneously from the same subject. As shown in Supplemental Fig. 4,distributed networks were found to correlate with the microstatetime course. These networks correspond to several RSNs identified in-dependently from the fMRI data of the same subject, including thesensorimotor, visual, attention, auditory and default mode networks.

Between-microstates correlation

In order to examine the relationship between temporal indepen-dent microstates, we calculated the pairwise correlation betweenthe regressors derived from the microstate time courses, as shownin Fig. 6A. The time courses for MS 7–12 were correlated, consistentwith their similar profile in the statistical maps. The other microstatesshowed less inter-correlation. To clarify whether the observed corre-lational results from the original microstate time courses might havebeen introduced by the process of convolving the data with a hemo-dynamic response function, we calculated the pairwise correlationbetween microstate time courses before convolution, i.e. the absoluteintensity values of IC waveforms. As shown in Fig. 6B, this procedurerevealed correlations between MS 7–12 that remained significant butwere smaller in magnitude.

In other post hoc comparisons we examined relationships similarto those assessed above with the microstates extracted using the spa-tial clustering method of Britz et al. (2010). Four main clustered mi-crostates (MS 1′–4′) were identified from the group data, whichappeared consistent to the topologies identified in previous reportsthat applied the spatial clustering method (Britz et al., 2010; Koeniget al., 2002). Fig. 6C shows the pairwise correlation between the re-gressors derived from the time courses of these microstates, asreflected by the spatial correlation between the continuous EEG and

Fig. 4. Statistical maps identified by the time courses of temporal independent microstates (MS 1–6) at the group level (corrected for multiple comparison by thresholding atqb0.05). The BOLD data are overlaid on a structural T1-weighted image of a subject. The statistical maps also are compared to the areas where BOLD activity was associatedwith distinct RSNs. Blue arrows indicate that the statistical maps were derived from the GLM analysis of the relevant microstate. Brown arrows indicate a comparison betweenmicrostates-associated maps and RSN activations at the same slice. MS1 (A), MS2 (B), MS4 (C) and MS5 (D) corresponded to two RSNs. MS3 (F) and MS6 (E) only correspondedto one RSN. Collectively RSNs identified from MS 1–6 include sensorimotor, auditory, attention, frontal, visceromotor and default mode networks.

2068 H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

each of the main microstates. The correlation between the microstatetime courses without convolution is shown in Fig. 6D. Compared withthe data depicted in Figs. 6A–B, nominally greater correlations wereobserved for the microstates extracted using the spatial clusteringmethod demonstrated in Figs. 6C–D. The convolution procedure alsoresulted in stronger correlations in Fig. 6C than those shown inFig. 6D.

Discussion

In this work we investigated the extent to which BOLD RSNs re-flect spontaneous, resting cerebral electrophysiological activity mea-sured using high-density extracranial EEG recordings. We developeda novel and fully data-driven approach to extract independent micro-states from the high-density EEG data and then compared the tempo-ral dynamics of these microstates to the fluctuations of the RSNsidentified from simultaneously acquired fMRI data. Our resultsshowed for the first time that RSNs identified in the BOLD fMRI dataare associated with specific electrophysiological signatures involvingone or more microstates identified in the EEG data. Our results haveshown that a wide range of RSNs, including visual, sensorimotor,

auditory, attention, frontal, viseromotor and default model networksare correlated with corresponding microstates. These results supportthe hypothesis that during the eyes-closed at rest behavioral condi-tion, RSNs characterized by temporally slow fluctuations in BOLDactivity are correlated with spontaneous, temporally fast, neuronalelectrical activity microstates in the EEG recording.

Previous studies that explored microstates as the electrophysio-logical correlates of RSNs focused largely on the spatial topographiesof microstates (Britz et al., 2010; Musso et al., 2010). Using clusteringanalysis methods, these studies segregated microstates into groupsbased upon the maximum spatial dissimilarities between groups,irrespective of their temporal characteristics. The time courses of mi-crostates also were derived on the basis of spatial correlation. Never-theless, the spatial dissimilarity between groups does not guaranteetemporal independence. Instead, the microstate time courses com-monly were correlated, as reported in the previous study (Fig. 3 inBritz et al., 2010) as well as replicated in our study (Fig. 6D). Sincethe microstate time courses were entered as regressors into theGLM used to delineate associated RSNs, significant correlation be-tween regressors would result in the identification of networks thatwere highly overlapping (e.g., as seen in Supplemental Fig. 4 of Britz

Fig. 5. Statistical maps identified by the time courses of MS7 (A) and MS13 (B) at the group level (corrected for multiple comparison by thresholding at qb0.05) and compared tothe RSNs identified within the BOLD fMRI data. All maps were overlaid on a structural T1-weighted image of the same subject as in Fig. 4. The statistical maps derived fromMS 7–12appeared similar to each other, so only those from MS7 are shown here. Blue arrows indicate that the statistic maps were derived from GLM analysis using microstate time courses.Brown arrows indicate a comparison between microstates-associated maps and RSN BOLD data at the same slice. MS7 corresponded to visual, sensorimotor and auditory. MS13corresponded to a wide range of RSNs, including visual, attention, frontal and default mode networks.

2069H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

et al., 2010); the ability to identify RSNs associated with individualmicrostates at the group level thus would be limited (Britz et al.,2010; Musso et al., 2010).

In order to overcome this potential confound, we developed anovel approach to analyze microstates via ICA. The spatiotemporaldynamics of microstates were decomposed into maximally indepen-dent components, such that each component represented a main mi-crostate. The topography of the IC reflected the spatial origin of themicrostate. The time course of the IC indicated the projected activityof the microstate, and this parameter was entered as a regressor inthe GLM used to identify RSNs associated with the microstate. Byusing ICA to extract the microstate information, the temporal correla-tion between microstates was maximally suppressed, thereby en-abling the identification of multiple, distinct RSNs that otherwisemight not have been delineated from correlated microstates.

Notably, the time course of temporal independent microstatesidentified by the ICA analysis showed lower correlation between dif-ferent microstates (Fig. 6B) compared to those identified using thespatial clustering analysis (Fig. 6D). Similarly, after convolving with

a hemodynamic response function (also a low-pass filtering), thetime courses also showed relatively low levels of inter-correlation(Fig. 6A). When comparing the time course of each microstate tothat of the RSNs on a pair-wise basis (Fig. 3), few examples werefound in which the microstates and RSNs were significantly correlat-ed. Particularly in MS 1–6, each microstate showed significant corre-lations with only one or two RSNs. In contrast, MS 7–12 were largelycorrelated to the sensorimotor and visual RSNs, each of which wascomposed of two sub-networks. In the spatial aspect, the networkidentified from the time course of microstate corresponded to theRSNs as indicated by the temporal correlation.

In relation to the ten RSNs identified from the fMRI data, the tem-poral independent microstates we identified appeared to show one ofthe following three characteristics: 1) microstates that specificallycorrelated to one or two RSN (MS 1–6); 2) microstates that correlatedto more than two RSNs and showed overlap with other microstates intheir relationships to a set of RSNs (MS 7–12); or 3) microstates thatcorrelated nonspecifically to several RSNs (MS 13). The MS 1–6 eachcorrelated with one or two RSNs and the RSNs associated with these

Fig. 6. Pair-wise correlations coefficient (CC) between the thirteen temporal independent microstate time courses (MS 1–13) obtained using the method outlined in Fig. 1 before(B) versus after (A) convolution with the BOLD hemodynamic response function (HRF). Pair-wise correlations between the four clustered microstates derived using the spatial clus-tering analysis method (MS 1′–4′) before (D) versus after (C) convolution with HRF. All non-zero values reported were thresholded at pb0.001 (uncorrected).

2070 H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

microstates varied from one to another, indicating high specificity ofthese microstates as electrophysiological correlates of distinct RSNs.Collectively MS 1–6 identified sensorimotor, auditory, attention, fron-tal, viseromotor and default mode networks (i.e., all RSNs except thevisual networks). In contrast, MS 7–12 were all related to the senso-rimotor, visual and auditory RSNs. The fact that overlapping networkswere correlated with MS 7–12 activity (Fig. 5A) is consistent with thecorrelation extant between the regressors associated with MS 7–12(Fig. 6A). Notably, the time courses of MS 7–12 demonstratedstrengthened correlation coefficients after convolution with the HRF(Fig. 6B). In contrast, other microstates of correlation at similar levelwith MS 7–12 showed elimination of correlation after convolution.Such variation in maintaining long-range dependency suggests thatthe microstates may be of different scale-free properties (Van deVille et al., 2010).

A wide range of RSNs, including the visual, sensorimotor, auditory,attention, frontal, visceromotor and default mode networks, were as-sociated with the EEG microstates delineated in the present study. Toour knowledge, this is the first simultaneous EEG and fMRI study thathas systematically revealed specific electrophysiological correlates(EEG microstates) of such a wide range of BOLD RSNs (i.e., the micro-states correlated with a total of ten RSNs identified in the fMRI data).Previous studies had explored band-limited power of the EEG record-ing as the electrophysiological correlates of spontaneous BOLD RSNactivity. A limited number of RSNs, such as the visual networks(Goldman et al., 2002; Moosmann et al., 2003) and the defaultmode network (Laufs et al., 2003), had been identified using the elec-trophysiological activity limited to a single EEG frequency band.Moreover, the electrophysiological activity from single frequencybands generally correlated with BOLD activity in several RSNs, possi-bly because the power variations associated with distinct frequencybands usually exhibit coherence, as demonstrated in the EEG

(Mantini et al., 2007) and intracranial recordings obtained fromhumans (Manning et al., 2009) or experimental animals (Leopold etal., 2003). A comparison between the time courses of multiple RSNsand EEG power revealed that each RSN is widely associated withthe EEG signal in all frequency bands, and that the EEG activity withineach frequency band is correlated to most of the RSNs (Mantini et al.,2007). These results suggest that the band-limited power of the EEGrecording correlates with spontaneous BOLD activity, but affords lim-ited sensitivity for delineating specific electrophysiological signaturesfor particular RSNs.

Other previous studies thus explored the extent to which EEG mi-crostates serve as electrophysiological correlates of BOLD RSNs. Nev-ertheless, these studies identified a relatively limited number ofRSNs at the group level (Britz et al., 2010; Musso et al., 2010). In thepresent study, using a novel ICA approach to extract the microstates,we delineated thirteen microstates based upon maximal indepen-dence. Our results demonstrated the temporal and spatial correspon-dence between the networks identified from temporal independentmicrostates and a wide range of previously characterized BOLDRSNs. Furthermore, most of the EEG microstates revealed using ourapproach proved relatively specific with respect to their relationshipto individual RSNs, supporting the hypothesis that EEG microstatesexist as reliable electrophysiological signatures of BOLD RSNs.

Our findings hold important implications for studying resting statefunction in neuropsychiatric disorders. Increasing evidence indicatesthat the properties of RSNs are altered in a variety of brain disorders(Fornito and Bullmore, 2010). In addition, recent studies of EEGmicrostates have reported that the fast dynamics of microstates (atdurations of ~100 ms) are critical for linking information across rela-tively long time ranges (Van de Ville et al., 2010). Moreover, alter-ations in the temporal dynamics of EEG microstates are evident inpanic disorder (Kikuchi et al., 2011) and schizophrenia, with the

2071H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

abnormalities in the latter condition being associated with auditory hal-lucinations (Kindler et al., 2011). Although conducted in a moderatesample size, this study reveals a tight correlation between the slow fluc-tuations of RSN BOLD activity and the fast spatiotemporal dynamics ofEEG microstates. Our results also demonstrate that such characteristicEEG microstates and their associated RSN correlates can be identifiedat the single-subject level. Temporal independent EEGmicrostates (sup-plemental Fig. 5) are consistent with the thirteen group-level micro-states, which demonstrated the feasibility of this method in retrievingtemporal independent microstates at the single-subject level. Such re-sults also suggest great potential of the proposed method for studyingspatiotemporal brain dynamics, by exploring ICA-derived temporal in-dependent EEG microstates as signatures of BOLD RSNs in individualswith various neuropsychiatric disorders. Thus the relationships betweentemporal independent EEG microstates and their associated BOLD RSN,unraveled intrinsic connections between neuronal activity andhemody-namicsfluctuations and provide parameters thatmay prove informativein future studies of cerebral functions in healthy brain aswell as cerebraldysfunction in brain disorders.

Conclusions

We have shown that spontaneous, temporally fast, electrophysio-logical activity as reflected in EEG microstates is correlated with theslower hemodynamic fluctuations of the BOLD RSNs. We identifiedthirteen independent EEG microstates and ten BOLD RSNs that wereorganized independently in their temporal and spatial characteristics,respectively. Our results revealed that each of the ten BOLD RSNsidentified in the resting state fMRI data was characterized by a rela-tively specific electrophysiological signature involving from one mi-crostate to a combination of several microstates. The concurrentlyacquired fMRI and EEG data thus reveals that complex spatial andtemporal dynamics of neuronal activity are reflected by the interrela-tionships between neuroimaging measures obtained using modalitiesthat vastly differ in their spatial and temporal properties. These find-ings support the potential of multimodal fMRI and EEG approaches toelucidate normal and pathological interactions between cerebralfunction and behavior, cognition or emotion.

Acknowledgments

This research was supported by the Laureate Institute for BrainResearch and The William K. Warren Foundation.

Appendix A. Supplementary data

Supplementary data to this article can be found online at doi:10.1016/j.neuroimage.2012.02.031.

References

Allen, P.J., Polizzi, G., Krakow, K., Fish, D.R., Lemieux, L., 1998. Identification of EEGevents in the MR scanner: the problem of pulse artifact and a method for its sub-traction. Neuroimage 8, 229–239.

Allen, P.J., Josephs, O., Turner, R., 2000. A method for removing imaging artifact fromcontinuous EEG recorded during functional MRI. Neuroimage 12, 230–239.

Arieli, A., Sterkin, A., Grinvald, A., Aertsen, A., 1996. Dynamics of ongoing activity:explanation of the large variability in evoked cortical responses. Science 273,1868–1871.

Beckmann, C.F., DeLuca, M., Devlin, J.T., Smith, S.M., 2005. Investigations into resting-state connectivity using independent component analysis. Philos. Trans. R. Soc.Lond. B Biol. Sci. 360, 1001–1013.

Bell, A.J., Sejnowski, T.J., 1995. An information-maximization approach to blind separationand blind deconvolution. Neural Comput. 7, 1129–1159.

Birn, R.M., Diamond, J.B., Smith, M.A., Bandettini, P.A., 2006. Separating respiratory-variation-related fluctuations from neuronal-activity-related fluctuations in fMRI.Neuroimage 31, 1536–1548.

Birn, R.M., Smith,M.A., Jones, T.B., Bandettini, P.A., 2008. The respiration response function:the temporal dynamics of fMRI signal fluctuations related to changes in respiration.Neuroimage 40, 644–654.

Biswal, B., Yetkin, F.Z., Haughton, V.M., Hyde, J.S., 1995. Functional connectivity in themotor cortex of resting human brain using echo-planar MRI. Magn. Reson. Med.34, 537–541.

Bodurka, J., Ledden, P.J., van Gelderen, P., Chu, R., de Zwart, J.A., Morris, D., Duyn, J.H., 2004.Scalable multichannel MRI data acquisition system. Magn. Reson. Med. 51, 165–171.

Bodurka, J., Ye, F., Petridou, N., Murphy, K., Bandettini, P.A., 2007. Mapping the MRIvoxel volume in which thermal noise matches physiological noise — implicationsfor fMRI. Neuroimage 34, 542–549.

Boly, M., Balteau, E., Schnakers, C., Degueldre, C., Moonen, G., Luxen, A., Phillips, C.,Peigneux, P., Maquet, P., Laureys, S., 2007. Baseline brain activity fluctuations predictsomatosensory perception in humans. Proc. Natl. Acad. Sci. 104, 12187–12192.

Britz, J., Landis, T., Michel, C.M., 2009. Right parietal brain activity precedes perceptualalternation of bistable stimuli. Cereb. Cortex 19, 55–65.

Britz, J., Van De Ville, D., Michel, C.M., 2010. BOLD correlates of EEG topography revealrapid resting-state network dynamics. Neuroimage 52, 1162–1170.

Buzsaki, G., Draguhn, A., 2004. Neuronal oscillations in cortical networks. Science 304,1926–1929.

Calhoun, V.D., Adali, T., Pearlson, G.D., Pekar, J.J., 2001. A method for making group in-ferences from functional MRI data using independent component analysis. Hum.Brain Mapp. 14, 140–151.

Corbetta, M., Shulman, G.L., 2002. Control of goal-directed and stimulus-driven attentionin the brain. Nat. Rev. Neurosci. 3, 201–215.

Cox, R.W., 1996. AFNI: software for analysis and visualization of functional magneticresonance neuroimages. Comput. Biomed. Res. 29, 162–173.

Damoiseaux, J.S., Rombouts, S.A., Barkhof, F., Scheltens, P., Stam, C.J., Smith, S.M.,Beckmann, C.F., 2006. Consistent resting-state networks across healthy subjects.Proc. Natl. Acad. Sci. 103, 13848–13853.

Fornito, A., Bullmore, E.T., 2010. What can spontaneous fluctuations of the bloodoxygenation-level-dependent signal tell us about psychiatric disorders? Curr.Opin. Psychiatry 23, 239–249.

Fox, M.D., Raichle, M.E., 2007. Spontaneous fluctuations in brain activity observed withfunctional magnetic resonance imaging. Nat. Rev. Neurosci. 8, 700–711.

Friston, K.J., Fletcher, P., Josephs, O., Holmes, A., Rugg, M.D., Turner, R., 1998. Event-relatedfMRI: characterizing differential responses. Neuroimage 7, 30–40.

Glover, G.H., Li, T.Q., Ress, D., 2000. Image-based method for retrospective correction ofphysiological motion effects in fMRI: RETROICOR. Magn. Reson. Med. 44, 162–167.

Goldman, R.I., Stern, J.M., Engel Jr., J., Cohen, M.S., 2002. Simultaneous EEG and fMRI ofthe alpha rhythm. Neuroreport 13, 2487–2492.

Gonzales-Castillo, J., Roopchasnsingh, V., Bandettini, P.A., Bodurka, J., 2011. Physiologicalnoise effects on the flip angle selection in BOLD fMRI. Neuroimage 54, 2764–2778.

Greicius, M.D., Krasnow, B., Reiss, A.L., Menon, V., 2003. Functional connectivity in theresting brain: a network analysis of the default mode hypothesis. Proc. Natl. Acad.Sci. 100, 253–258.

He, B.J., Snyder, A.Z., Zempel, J.M., Smyth, M.D., Raichle, M.E., 2008. Electrophysiologicalcorrelates of the brain's intrinsic large-scale functional architecture. Proc. Natl.Acad. Sci. 105, 16039–16044.

Hesselmann, G., Kell, C.A., Eger, E., Kleinschmidt, A., 2008. Spontaneous local variationsin ongoing neural activity bias perceptual decisions. Proc. Natl. Acad. Sci. 105,10984–10989.

Kikuchi, M., Koenig, T., Munesue, T., Hanaoka, A., Strik, W., Dierks, T., Koshino, Y., Minabe,Y., 2011. EEGmicrostate analysis in drug-naive patients with panic disorder. PLoSOne6, e22912.

Kindler, J., Hubl, D., Strik,W.K., Dierks, T., Koenig, T., 2011. Resting-state EEG in schizophrenia:auditory verbal hallucinations are related to shortening of specific microstates. Clin.Neurophysiol. 122, 1179–1182.

Koenig, T., Prichep, L., Lehmann, D., Sosa, P.V., Braeker, E., Kleinlogel, H., Isenhart, R.,John, E.R., 2002. Millisecond by millisecond, year by year: normative EEG micro-states and developmental stages. Neuroimage 16, 41–48.

Laufs, H., Krakow, K., Sterzer, P., Eger, E., Beyerle, A., Salek-Haddadi, A., Kleinschmidt,A., 2003. Electroencephalographic signatures of attentional and cognitive defaultmodes in spontaneous brain activity fluctuations at rest. Proc. Natl. Acad. Sci.100, 11053–11058.

Lehmann, D., 1980. In: Koukkou, M., Lehmann, D., Angst, J. (Eds.), Functional States of theBrain: Their Determinants. Elsevier/North-Holland Biomedical Press, Amsterdam, pp.189–202.

Lehmann, D., 1990. Past, present and future of topographic mapping. Brain Topogr. 3,191–202.

Lehmann, D., Ozaki, H., Pal, I., 1987. EEG alpha map series: brain micro-states by space-oriented adaptive segmentation. Electroencephalogr. Clin. Neurophysiol. 67, 271–288.

Lehmann, D., Strik, W.K., Henggeler, B., Koenig, T., Koukkou, M., 1998. Brain electric mi-crostates and momentary conscious mind states as building blocks of spontaneousthinking: I. visual imagery and abstract thoughts. Int. J. Psychophysiol. 29, 1–11.

Leopold, D.A., Murayama, Y., Logothetis, N.K., 2003. Very slow activity fluctuations inmonkey visual cortex: implications for functional brain imaging. Cereb. Cortex13, 422–433.

Li, Y.O., Adali, T., Calhoun, V.D., 2006. Sample Dependence Correction for Order Selection inFMRI Analysis. Proc. ISBI.

Liang, Z., King, J., Zhang, N., 2011. Uncovering intrinsic connectional architecture offunctional networks in awake rat brain. J. Neurosci. 31, 3776–3783.

Logothetis, N.K., 2008. What we can do and what we cannot do with fMRI. Nature 453,869–878.

Logothetis, N.K., Pauls, J., Augath, M., Trinath, T., Oeltermann, A., 2001. Neurophysiologicalinvestigation of the basis of the fMRI signal. Nature 412, 150–157.

Mandelkow, H., Halder, P., Boesiger, P., Brandeis, D., 2006. Synchronization facilitatesremoval of MRI artefacts from concurrent EEG recordings and increases usablebandwidth. Neuroimage 32, 1120–1126.

2072 H. Yuan et al. / NeuroImage 60 (2012) 2062–2072

Manning, J.R., Jacobs, J., Fried, I., Kahana, M.J., 2009. Broadband shifts in local field potentialpower spectra are correlated with single-neuron spiking in humans. J. Neurosci. 29,13613–13620.

Mantini, D., Perrucci,M.G., DelGratta, C., Romani, G.L., Corbetta,M., 2007. Electrophysiologicalsignatures of resting state networks in the human brain. Proc. Natl. Acad. Sci. 104,13170–13175.

Miller, E.K., Cohen, J.D., 2001. An integrative theory of prefrontal cortex function. Annu.Rev. Neurosci. 24, 167–202.

Moosmann, M., Ritter, P., Krastel, I., Brink, A., Thees, S., Blankenburg, F., Taskin, B.,Obrig, H., Villringer, A., 2003. Correlates of alpha rhythm in functional magneticresonance imaging and near infrared spectroscopy. Neuroimage 20, 145–158.

Mukamel, R., Gelbard, H., Arieli, A., Hasson, U., Fried, I., Malach, R., 2005. Coupling betweenneuronal firing, field potentials, and FMRI in human auditory cortex. Science 309,951–954.

Musso, F., Brinkmeyer, J., Mobascher, A., Warbrick, T., Winterer, G., 2010. Spontaneousbrain activity and EEG microstates. A novel EEG/fMRI analysis approach to exploreresting-state networks. Neuroimage 52, 1149–1161.

Nir, Y., Mukamel, R., Dinstein, I., Privman, E., Harel, M., Fisch, L., Gelbard-Sagiv, H.,Kipervasser, S., Andelman, F., Neufeld, M.Y., et al., 2008. Interhemispheric correlationsof slow spontaneous neuronal fluctuations revealed in human sensory cortex. Nat.Neurosci. 11, 1100–1108.

Nunez, P.L., 1995. Neocortical Dynamics and Human EEG Rhythms. Oxford UniversityPress, New York.

Olbrich, S., Mulert, C., Karch, S., Trenner, M., Leicht, G., Pogarell, O., Hegerl, U., 2009.EEG-vigilance and BOLD effect during simultaneous EEG/fMRI measurement. Neu-roimage 45 (2), 319–332.

Ongür, D., Price, J.L., 2000. The organization of networks within the orbital and medialprefrontal cortex of rats, monkeys and humans. Cereb. Cortex 10, 206–219.

Raichle, M.E., MacLeod, A.M., Snyder, A.Z., Powers, W.J., Gusnard, D.A., Shulman, G.L.,2001. A default mode of brain function. Proc. Natl. Acad. Sci. 98, 676–682.

Rissanen, J., 1983. A universal prior for integers and estimation by minimum descriptionlength. Ann. Stat. 11, 416–431.

Sadaghiani, S., Scheeringa, R., Lehongre, K., Morillon, B., Giraud, A.L., Kleinschmidt, A.,2010. Intrinsic connectivity networks, alpha oscillations, and tonic alertness: asimultaneous electroencephalography/functional magnetic resonance imagingstudy. J. Neurosci. 30, 10243–10250.

Smith, S.M., Fox, P.T., Miller, K.L., Glahn, D.C., Fox, P.M., Mackay, C.E., Filippini, N.,Watkins, K.E., Toro, R., Laird, A.R., et al., 2009. Correspondence of the brain's func-tional architecture during activation and rest. Proc. Natl. Acad. Sci. 106,13040–13045.

Talairach, J., Tournoux, P., 1988. Co-planar Stereotaxic Atlas of the Human Brain. Thieme,New York.

Van de Ville, D., Britz, J., Michel, C.M., 2010. EEG microstate sequences in healthyhumans at rest reveal scale-free dynamics. Proc. Natl. Acad. Sci. 107,18179–18184.

Varela, F., Lachaux, J.P., Rodriguez, E., Martinerie, J., 2001. The brainweb: phase syn-chronization and large-scale integration. Nat. Rev. Neurosci. 2, 229–239.

Vincent, J.L., Patel, G.H., Fox, M.D., Snyder, A.Z., Baker, J.T., Van Essen, D.C., Zempel, J.M.,Snyder, L.H., Corbetta, M., Raichle, M.E., 2007. Intrinsic functional architecture inthe anaesthetized monkey brain. Nature 447, 83–86.

Wackermann, J., Lehmann, D., Michel, C.M., Strik, W.K., 1993. Adaptive segmenta-tion of spontaneous EEG map series into spatially defined microstates. Int. J.Psychophysiol. 14, 269–283.

Yuan, H., Perdoni, C., Yang, L., He, B., 2011. Differential electrophysiological coupling forpositive and negative BOLD responses during unilateral handmovements. J. Neurosci.31, 9585–9593.