NeuroImage 42 (2008) 1047–1055

Contents lists available at ScienceDirect

NeuroImage

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

Frequency specificity of functional connectivity in brain networks

Changwei W. Wu a,b, Hong Gu a, Hanbing Lu a, Elliot A. Stein a, Jyh-Horng Chen b,⁎, Yihong Yang a,⁎a Neuroimaging Research Branch, National Institute on Drug Abuse, NIH, Baltimore, MD, USAb Interdisciplinary MRI/MRS Lab, Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan

⁎ Corresponding authors. Y. Yang is to be contactedNational Institute on Drug Abuse, NIH, Room 7A709,Baltimore, MD 21224, USA. Fax: +1 443 740 2816. J.-H. CNational Taiwan University, Sec. 4, No. 1, Roosevelt Road,3366 3517.

E-mail addresses: YihongYang@intra.nida.nih.gov (Y.(J.-H. Chen).

1053-8119/$ – see front matter. Published by Elsevier Indoi:10.1016/j.neuroimage.2008.05.035

a b s t r a c t

a r t i c l e i n f o

Article history:

Synchronized low-frequenc Received 14 February 2008Revised 16 April 2008Accepted 17 May 2008Available online 15 July 2008

y spontaneous fluctuations of the functional MRI (fMRI) signal have been shownto be associated with electroencephalography (EEG) power fluctuations in multiple brain networks withinpredefined frequency bands. However, it remains unclear whether frequency-specific characteristics exist inthe resting-state fMRI signal. In this study, fMRI signals in five functional brain networks (sensorimotor,‘default mode’, visual, amygdala, and hippocampus) were decomposed into various frequency bands within alow-frequency range (0–0.24 Hz). Results show that the correlations in cortical networks concentrate withinultra-low frequencies (0.01–0.06 Hz) while connections within limbic networks distribute over a widerfrequency range (0.01–0.14 Hz), suggesting distinct frequency-specific features in the resting-state fMRIsignal within these functional networks. Moreover, the connectivity decay rates along the frequency bandsare positively correlated with the physical distances between connected brain regions and seed points. Thisdistance-frequency relationship might be attributed to a larger attenuation of synchrony of brain regionsseparated with longer distance and/or connected with more synaptic steps.

Published by Elsevier Inc.

Introduction

Specific cortical and subcortical regions, where neural ac-tivity unfolds, couple to other brain regions forming func-tional networks associated with cognition and action (Buzsakiand Draguhn, 2004; Varela et al., 2001). Even under a restcondition in the absence of entrained task performance, thesebrain regions are thought to possess dynamic, synchronizedoscillations, upon which the interactions between brain re-gions or so-called “functional connectivity” can be observedusing resting-state functional magnetic resonance imaging(fMRI) techniques (Biswal et al., 1995; Greicius et al., 2003;Hampson et al., 2002; Lowe et al., 1998). Several studiessupport the assumption that the low-frequency resting-statefMRI fluctuations have a neuronal basis rather than physio-logical artifacts induced by cardiac pulsations and/or respira-tion (Cordes et al., 2001; De Luca et al., 2006; Leopold et al.,2003; Lowe et al., 1998), although the exact neurophysiolo-gical mechanism of functional connectivity is still under active

at the Neuroimaging Branch,Suite 200, 251 Bayview Blvd.,hen, Room 706, Ming-Da Bldg.,Taipei 106, Taiwan. Fax: +886 2

Yang), jhchen@ntu.edu.tw

c.

investigations (Lu et al., 2007; Mantini et al., 2007). Using atemporal cross-correlation method, Biswal et al. first demon-strated functional connectivity within a frequency range (0–0.08 Hz) in sensorimotor networks (Biswal et al., 1995).Subsequently, resting-state fMRI has been widely applied toinvestigate spontaneous activity in cognitive and sensory net-works, as well as pathological alterations in these networks(Greicius et al., 2004; Hampson et al., 2002; Li et al., 2002;Lowe et al., 1998, 2000).

Electroencephalography (EEG) studies have provided sig-nificant evidence of a tight coupling between regional cerebralhemodynamic responses (HDR) and local field potentials (LFP)(Lauritzen and Gold, 2003; Logothetis et al., 2001). Recentstudies demonstrated that the HDR corresponding to taskstimulation in primate visual cortex correlates specifically withLFP oscillations in the gamma (γ) band (Niessing et al., 2005;Wilke et al., 2006), while the resting-state HDR in the rat so-matosensory cortex correlates with LFPs in the lower frequencyrange, particularly in the δ band (Lu et al., 2007), suggestingspectral differences under distinct physiological conditions.Mantini et al. further demonstrated that each functional net-work corresponds to a specific frequency-specific rhythm (Man-tini et al., 2007). These observations imply that functionalconnectivity may be frequency specific and that frequencycharacteristics may be distinct for different brain networks.Several previous studies have explored the frequency domain offunctional connectivity; some observed frequency relationships

1048 C.W. Wu et al. / NeuroImage 42 (2008) 1047–1055

on a large scale (Cordes et al., 2001; Salvador et al., 2005, 2008),while others aimed at topologically reciprocal connections be-tween regions instead of between/within known brain net-works (Achard et al., 2006; Salvador et al., 2007).

Our goal, therefore,was to investigate the spectral features ofresting-state spontaneous fMRI fluctuations between multiplefunctional networkswith finer frequency intervals. To do so, weassessed the frequency characteristics of BOLD functional con-nectivity signal in five brain networks (sensorimotor, ‘defaultmode’, visual, amygdala, and hippocampus). Using multipleband-pass filters with 0.01–0.04 Hz bandwidth, the resting-state fMRI signal was decomposed into 12 frequency bands andanalyzed from each of the five networks between 0–0.24 Hz.Both between- and within-network frequency-specific dispa-rities were seen in cortical and limbic subcortical networks.

Methods

Participants

Twenty healthy participants were included in this study (13males, 36±9year-old). Theywere screenedwith aquestionnaireto ensure that they had no history of neurological illness, psy-chiatric disorders or past drug use. Present drug use and preg-nancy were assessed with urine testing. Informed consent wasobtained from all participants prior to the experiments inaccordance with the protocol approved by the InstitutionalReview Board of the National Institute on Drug Abuse.

Data acquisition

Six minutes of resting fMRI data were acquired on a 3.0 TSiemens Allegra scanner (Siemens, Erlangen, Germany) using asingle-shot, gradient-recalled echo planar imaging (EPI) se-

Table 1Description of the selected seed points and ROIs for each brain network

Network Names of seed or regionof interest

Abbreviation

Visual Primary visual area (right, seed)Primary visual area BA17Secondary visual area BA18Third visual area BA19Lateral geniculate nucleus LGN

Default mode Posterior cingulate cortex (seed)Posterior cingulate cortex PCCAnterior cingulate cortex ACCInferior parietal cortex (bilateral) IPCMedial temporal gyrus (bilateral) MTG

Sensorimotor Primary motor area (right, seed)Primary motor area (right) M1-rPrimary motor area (left) M1-lSupplementary motor area SMASecondary sensory area (bilateral) S2

Amygdala Amygdala (right, seed)Amygdala (right) AMG-rAmygdala (left) AMG-lMedial prefrontal cortex MPCInsula (bilateral) ISL

Hippocampus Hippocampus (right, seed)Hippocampus (right) HPC-rHippocampus (left) HPC-lMedial prefrontal cortex MPCParahippocampal gyrus (bilateral) PHG

a The distances of bilateral ROIs were average values of left and right for facilitating the p

quence (2000ms repetition time (TR), 27ms echo time (TE), and77° flip angle and a head volume coil. Head motion was mini-mized using individually custom-made foam padding. Thirty-nine axial slices (220×220 mm2

field of view, 64×64 in-planematrix size, and 4mmslice thickness), aligned along the anteriorcommissure – posterior commissure plane allowed for wholebrain coverage. Prior to the resting scan, subjectswere instructedto rest with their eyes closed, not to think of anything in par-ticular, and not to fall asleep. For spatial normalization and loca-lization, a set of high-resolution T1-weighted anatomical images(3D-MPRAGE with 256×192×160 matrix size; 1×1×1 mm3 in-plane resolution; 1000ms inversion time; TR/TE=2500/4.38ms;flip angle=8°) was acquired on each subject.

Pre-processing

Functional data were processed using the Analysis of Func-tional Neuroimaging (AFNI) software package (Cox,1996). Mo-tion correction was performed by volume registering each 3Dvolume to a base volume. Linear detrending was applied toeliminate signal drift induced by system instability. The imageswere then converted into Talairach space and linearly resam-pled to an isotropic resolution (3×3×3 mm3). Subsequently,spatial smoothing was applied using a Gaussian isotropic ker-nel (full width at half maximum of 6 mm) to minimize indi-vidual variances and enhance the signal-to-noise ratio. The T1anatomical image was segmented into gray matter, whitematter, and cerebrospinal fluid (CSF) maps using the SPM5package (Wellcome Institute of Cognitive Neurology, UK).

After these processing procedures, 12 datasets with spe-cific frequency bands (0–0.01, 0.01–0.02, 0.02–0.04, 0.04–0.06,0.06–0.08, 0.08–0.10, 0.10–0.12, 0.12–0.14, 0.14–0.16, 0.16–0.20,0.20–0.24 Hz and the typical range of 0–0.1 Hz) were generatedusing Chebyshev type II low-pass/band-pass filters in MATLAB

Talairach coordinate Volume(mm3)

Euclidean distanceto seed (mm)

X Y Z

3 −81 8 1083 −87 11 3591 14.24 −79 11 13,689 20.6

10 −72 16 11,421 36.3±22 −32 3 2916 57.7

3 −54 24 1622 −57 18 12,609 16.73 41 32 36,315 101.1

±43 −64 33 11,826 46.1a

±53 −14 −9 12,231 78.1a

36 −28 53 16235 −28 54 12,339 13

−44 −27 53 7803 77.8−3 −29 59 5427 37.1

±53 −23 14 22,086 71.2a

23 −5 −15 10820 −8 −10 945 5.9

−25 −7 −11 1296 45.8−2 44 13 12,717 63.2

±39 −9 13 13,905 51a

30 −24 −9 10828 −27 −5 999 10.7

−32 −26 −6 1161 60.1−2 44 13 12,717 80.2

±26 −34 −7 19,224 36a

resentation.

1049C.W. Wu et al. / NeuroImage 42 (2008) 1047–1055

(MathWorks, Inc., CA). The edges of the bandswere constrainedwithin 6 dB of decay, while at least 30 dB of attenuation wasguaranteed in the stop-bands (pass-band edges ±0.003 Hz).

To correct for the potential influence of physiological noise,estimations of cardiac and respiratory information were per-formed using a temporal independent component analysis(ICA) (Beall and Lowe, 2007). Specifically, twelve componentswere identified by temporal ICA from our resting-state data-sets. Among the spatial patterns of the 12 components, theone with highest spatial correlation with a predefined cardiacsource map was considered as the major source of cardiacresponse. The time courses from the cardiac-correlated mapwere averaged to generate a cardiac estimator. The same pro-cedure was done to generate a respiratory estimator. Thesecardiac and respiratory estimators were then used to regressout potential physiological influence in the calculation of cor-relation coefficients.

Statistical analysis

Spherical seeds (6 mm in diameter) were prescribed foreach of the five networks based on standard Talairachcoordinates: 1) posterior cingulate cortex (PCC) [3, −54, 24];

Fig. 1. Frequency-specific functional connectivity maps (pb0.05, corrected) of the three cortimost columns denote the functional connectivity maps within the typical frequency range

2) primary motor cortex (M1) [36, −28, 53]; 3) primary visualcortex (V1 or BA17) [3, −81, 8]; 4) amygdala (AMG) [23, −5,−15]; and 5) hippocampus (HPC) [30, −24, −9] (Table 1). Allspherical seeds were chosen from the right side of the brain.Individual analyses of the 12 frequency bands were carried outon the five networks for every participant using linear regres-sion. The average time series from each spherical seed wastaken as the major predictor in the regression model. Othernuisance covariates were also considered (the six motion pa-rameters and the time series retrieved from the segmentedwhite matter mask).

To compute statistical significance across subjects, thecorrelation maps from individual subjects were converted tonormal-distributed z-scoremaps accounting for the degrees offreedom (DOF) of each frequency band. The effective DOF wasestimated by dividing the number of time points in the timeseries by a correction factor, which was calculated from thearea under the curve of the autocorrelation function (Fox et al.,2005). Group-level analyses were performed in AFNI usingmixed-effects ANOVA on the z-scores for each frequency band.Corrections for multiple comparisons were executed at thecluster level using Gaussian random field theory (voxel-wisethreshold of pb10−4 was applied to the connectivity maps,

cal networks: (a) visual system, (b) default mode, and (c) sensorimotor system. The left-(0–0.1 Hz) and the green dots denote the position of the seed points.

1050 C.W. Wu et al. / NeuroImage 42 (2008) 1047–1055

with a minimum cluster volume threshold of 243 mm3,yielding an overall false positive pb0.05 as determined byMonte Carlo simulation).

ROI generation

To compare the strength of functional connectivity in thefive brain networks, a total of twenty regions of interest (ROIs)were generated based on the statistical group results of thetypically used low-frequency band (0–0.1 Hz) (see Table 1).The ROIs in the sensorimotor network consisted of right M1(M1-r), left M1 (M1-l), supplementary motor area (SMA), andsecondary somatosensory cortex (S2). In the default-modenetwork, the ROIs included PCC, anterior cingulate cortex(ACC), bilateral inferior parietal cortex (IPC), and bilateral me-dial temporal gyrus (MTG). BA17, 18 and 19 (roughly cor-responding to V1, V2 and V3), and LGNwere selected from theAFNI atlas in the visual system. In the amygdala system, rightAMG (AMG-r), left AMG (AMG-l), medial prefrontal cortex(MPC), and bilateral insula (ISL) were included, while the rightand left hippocampus (HPC-r and HPC-l), MPC and para-hippocampal gyrus (PHG) were included as ROIs of thehippocampus seed. Most ROIs in the limbic system, such asAMG-r, AMG-l, HPC-r, HPC-l, and PHG, were chosen based onAFNI atlas structures combined with functional connectivitymaps (0–0.1 Hz). The seed voxels embedded in a ROI wereexcluded in the calculation of average correlation coefficientfrom the ROI in order to avoid artificial enhancement of thecoefficient due to autocorrelation. Lastly, a region of CSF insidethe ventricles was manually chosen as a reference region, dueto the presumption of its irrelevance with brain functionalconnectivity.

Results

Figs. 1 and 2 show the connectivity maps of the cortical andsubcortical limbicnetworks, respectively, in each frequencyband(themiddle row represents the slice location of the seed points).In the cortical networks (visual, ‘defaultmode’ and sensorimotorsystems; see Fig. 1), the ‘long-distance’ connections to distal

Fig. 2. Frequency-specific functional connectivity maps (pb0.05, corrected) of the two limfunctional connectivity maps within the typical frequency range (0–0.1 Hz) and the green d

brain regions are generally most apparent at the lower fre-quency ranges (0.01–0.06 Hz); see e.g. the BA17 to LGN area in(a), the PCC to ACC in (b), and ipsilateral to contralateral M1 in(c). At higher frequencies, the long-distance connectionsbecome less synchronous, while the ‘short-distance’ connec-tions around the seed points spread over wider frequencyranges. In contrast, the long-distance connections in the twolimbic networks (Fig. 2) were strongest at relatively highfrequency bands (0–0.14 Hz in amygdala and 0.01–0.10 inhippocampus).

Fig. 3 illustrates the average correlation coefficients be-tween the selected ROIs and corresponding seed points in thefive networks over frequency bands. The correlations betweenseed points and CSF are also shown as references, since theCSF signal is considered to be uncorrelated with signals inbrain regions. Regarding the visual and sensorimotor net-works (Figs. 3a and c), the spectral distributions peak withinthe 0.01–0.02 Hz range and decrease substantially at higherfrequency bands. Similar frequency distributions were ob-served in the default mode except that the peak exhibited aslightly wider frequency range (0.01–0.04 Hz). In contrast, thefrequency dependency pattern in the two limbic system re-gions (Figs. 3d–e) either remained relatively constant or de-creased slightly and monotonically as a function of frequency.The frequency dependence between cortical and limbic net-works can also be demonstrated in F-score maps from theANOVA (pb0.05, corrected), with frequency as themain factor.As shown in Fig. 4, the locations of the frequency-dependentareas precisely match the spatial patterns of the functionalconnectivity cortical network maps. However, no significantfrequency-dependent area in limbic networks remains aroundthe seeds (green dots in Fig. 4) or around the within-networkROIs, indicating that the resting fMRI signal in limbic net-works is independent of the 0–0.24 Hz frequency range. Insummary, there was a qualitative frequency-specific dis-crepancy between distinct cortical and subcortical brainnetworks.

The physical distance between ROIs and seed points alsoexerted aneffect on the functional connectivitynetworks. Fig. 5ashows the connectivitycorrelationswithin the typical frequency

bic networks: (a) amygdala and (b) hippocampus. The left-most columns denote theots denote the position of the seed points.

Fig. 3. Average frequency-specific connectivity strengths between predefined ROIs and corresponding seeds for the five brain networks—a: visual system, b: default mode, c:sensorimotor, d: amygdala, and e: hippocampus.

1051C.W. Wu et al. / NeuroImage 42 (2008) 1047–1055

range (0–0.1 Hz) versus direct seed-ROI Euclidean distancesfor the five brain networks. The connectivity strength (asmeasured by cross correlation) is negatively associated withthe physical distance, suggesting that the strength of func-tional connectivity is attenuated with long-distance propaga-tion. The correlation between the connectivity strength and thedistance was −0.74 (pb0.002) in the three cortical systems(black) and −0.63 (pb0.05) in the two limbic systems (red). Tofurther investigate the relationship between the frequencyspecificity in these networks, a gamma function betweencorrelation coefficient and frequency, CC(f)=a · (f−b)c ·e− (f−b) / d,was adopted to fit the frequency-distributed correlations ineach ROI. In this equation, the parameters a, b and c were usedto describe the offset and rising features of the fitted curve,while the parameter d represented the flatness of the curve inthe frequency domain. A low flatness value indicates a highdecay rate along frequency, i.e., the connectivity is attributed

mainly from low-frequency components. Fig. 5b illustratesthe fitted flatness versus the physical Euclidean distances foreach ROI. The cortical systems have evident negativecorrelation between flatness and distances (black, R=−0.76,pb0.002), suggesting that the connectivity between seedpoints and distant ROIs is primarily attributed to ultra-low-frequency components. A similar trend was also seen inlimbic system networks (Fig. 5b) with higher flatness but anon-significant correlation (red, R=−0.48, p=0.14).

To evaluate the stability of functional connectivity networksacross subjects, the absolute coefficients of variation (CV, stan-dard deviation/average correlations) of the 20 participants arepresented in Fig. 6 over individual frequency bands for the fivebrain networks. CVs of low-frequency components (0–0.12 Hz)are approximately two times larger (except for LGN in visualsystem), while CVs of higher frequency bands (0.12–0.24 Hz)are much larger, than the CVs of the typically used frequency

Fig. 4. F-maps (pb0.05, corrected) of cortical and limbic networks, representing frequency-dependent areas within the brain networks. Each column denotes the F-maps of a networkand the green dots denote the position of the seed points.

1052 C.W. Wu et al. / NeuroImage 42 (2008) 1047–1055

range (0–0.1 Hz). This suggests that functional connectivityacross individuals is more stable at lower frequency bandsthan higher frequency bands.

Discussion

In the current study, the frequency response of resting-state functional MRI signals was assessed and compared infive brain networks in twenty normal participants. Resultsshow that both similarities and differences exist in thespontaneous fluctuations spectrograms between and withinthe examined brain networks. Cross-subject variations in thespectral domain provide evidence that connectivity is gen-erally stable among individuals, particularly at low-frequencybands.

The highest connectivity coherence was generally seen in aspecific frequency period of the spontaneous fluctuationsamong these brain networks, especially in the ultra-low-frequency range (0.01–0.06 Hz), which agrees well with pre-vious reports (Achard et al., 2006; Salvador et al., 2008).Nevertheless, spectral discrepancies between andwithin brainnetworks were also seen. A prominent between-network dis-crepancy revealed in this study is that the functional con-

Fig. 5. (a) Correlation coefficients within the typical frequency range (0–0.1 Hz) versus the Euobserved between cortical (black) and limbic (red) networks. (b) The fitted flatness of ediscrepancy in panel b indicates that the resting-state fMRI signal in cortical and subcortica

nectivity in the three cortical networks is dominated by ultra-low frequencies, whereas that in the two limbic system regionsis found in frequency ranges up to 0.14 Hz. Such dissimilarityin the frequency domain implies that either the functionalconnectivity characteristics in different brain networks canreflect distinct frequency and/or firing pattern, or the under-lying electrophysiological signal and/or the signal transductionmechanisms from electrophysiology to hemodynamics mightvary in different brain regions. Further studies of simultaneousfMRI and electrophysiological measurements in various brainsystems may provide further insight to understand the exactmechanism for this observation. It is also worth mentioningthat even though the spectral features of two networks (e.g.visual and sensorimotor systems) were quite similar, theirinter-network connectivity seems relatively weak, suggestingan asynchrony between the two fluctuation patterns betweenthese two systems (Fox et al., 2005). Based on these obser-vations, it can be speculated that multiple independent spon-taneous oscillations co-exist during the resting state, whichmay have specific frequency feature characteristics within dif-ferent brain networks.

A discrepancy of within-network connections in differentfrequency bands was also revealed. Our data showed that long-

clidean distance between ROIs to their corresponding seeds. Different distributions areach ROI versus the physical distance to their corresponding seed points. The salientl limbic networks has distinct frequency distributions.

Fig. 6. Absolute coefficients of variations (CV) of ROIs from five brain networks—a: visual system, b: default mode, c: sensorimotor, d: amygdala, and e: hippocampus.

1053C.W. Wu et al. / NeuroImage 42 (2008) 1047–1055

distance connections (e.g. contralateral M1) seem to be morefrequency specific (peaking at low-frequency bands), whereasshort-distance connections (e.g. ipsilateral M1) are distributedin a relatively wider frequency range, especially in cortical net-works (as shown in Figs. 1 and 3). We posit that the neu-robiological basis of this phenomenon may originate fromelectrophysiological signals linking the two hemispheres. Insupport of this conjecture, intracortical LFPs recorded from thelunate sulcus using a 15-electrode array, each at least 2.5 mm

apart, showed that inter-electrode coherence decreased asfrequency and/or distance increased, suggesting that LFPs inlower frequency bands travel a longer distance (Leopold et al.,2003). However, the maximum separation of electrode pairswas 10.6 mm, which is a shorter distance than the within-network connecting regions in the currentwork. Animal studiesrecording simultaneous electrophysiological and fMRI signalswould be beneficial to validate this hypothesis. Another spe-culation to the frequency-distance disparity may be attributed

1054 C.W. Wu et al. / NeuroImage 42 (2008) 1047–1055

to a larger attenuation of synchrony for brain regions separatedwith longer distance and/or connected by more synaptic steps.Taking the visual systems as anexample, synaptic relays followadirect transmission route (LGN-V1–V2–V3 for simplicity), with-out taking into account any alternative route from other brainregions. Fig. 3a shows gradual reduction of the connectivitystrength from the seed to BA 17,18 and 19 (corresponding to V1,V2 and V3, and the seed being at BA 17) over all frequencyranges, suggesting that the functional connectivity strengthmight be dependent on synaptic steps. The weak connection toLGN might be due to low ‘top-down’ modulation during the“closed-eye” state. Similar phenomenon was also observed inthe sensorimotor network (Fig. 3c), where the contralateral S1/M1 and SMA (one synaptic step) had stronger connections thanS2 (two synaptic steps). Of course, differences inhemodynamics,extent of anatomical divergence and convergence, firingpatterns and rates, density of vasculatization and couplingmechanics could also explain our observation. The quantitativedependence of functional connectivity on synaptic steps andphysical distance, as well as the associated physiological andhemodynamic mechanisms, may be better revealed using ani-mal models with combined anatomical, electrophysiologicaland fMRI studies.

Multiple cerebral networks during the resting state have alsobeen consistently identified using data-driven approaches suchas independent component analysis (ICA) (Damoiseaux et al.,2006; De Luca et al., 2006), implying that independentphysiological mechanisms may co-exist in these brain regionsto sustain the network functions at rest. Effort has beenmade tounderstand the linkage between the fMRI fluctuations and theirpossible neuronal bases. Recently Mantini et al. provided pre-liminary neurobiological evidence using simultaneous BOLD-EEG acquisition in human subjects (Mantini et al., 2007). Theirresults showed that each brain network is differentiallyweighted to different rhythms in the EEG power spectrum,which might be consistent with our frequency analysis results.However, due to poor spatial resolution in EEG recordings, itmay be premature to compare the EEG frequency responsedirectly to the fMRI response. Furthermore, the signal transduc-tion from electrophysiology to hemodynamics might be non-linear in terms of frequency response.

Physiological noise, including the cardiac/respiratory-induced signal changes and variations in respiration depth,can reduce the power in functional connectivity analysis (Birnet al., 2006; Lowe et al., 1998; Raj et al., 2001). Cordes et al.suggested that respiratory (0.1–0.5 Hz) and cardiac (0.6–1.2 Hz)fluctuations contribute less than 10% of the connectivity in vi-sual and auditory cortices in humans, using a directly-sampledacquisition and a larger frequency binning compared to thecurrent study (Cordes et al., 2001). Nevertheless, variations inrespiration depth are particularly problematic in resting-statefMRI data because it is associated with the ultra-low-frequencyrange (~0.03 Hz). To minimize this artifact, regressing outthe global signal has been suggested to provide comparableresults to removal of low-frequency respiratory fluctuations(Birnet al., 2006). In this study, the average signal retrieved fromwhite matter was used as a nuisance covariate regressor foreach subject, which preserves the involvement of cortical signaland removes the global variation. We have also used a newlydeveloped method to estimate the cardiac and respiratoryrelated time courses by temporal ICA (Beall and Lowe, 2007),and then corrected for the potential influence of physiologicalnoise.

Conclusion

Wehave examined the spectral response of the resting-statefMRI signal inmultiple functional networks of the human brain.The connections in three different cortical networks concen-trate within the ultra-low frequencies (0.01–0.06 Hz) while theconnections in two subcortical limbicnetworks distributeover awider frequency range (0.01–0.14 Hz), suggesting distinctfrequency-specific features in the resting-state fMRI signalacross these functional networks. Discrepancies between thesenetworks suggest that distinct physiological mechanisms,perhaps related to differences in ‘spontaneous’ neuronal firingrates or patterns, may co-exist in these brain regions to sustainthe network functions at rest. It has also been shown that thespectral flatness along the frequency range is negativelycorrelatedwith the physical distance from brain ‘target’ regionsto the seed. We suggest that connectivity strength may berelated to the number of synaptic steps or distance betweenbrain regions. These frequency-specific and anatomic-depen-dent properties may help interpret the sensitivity and specifi-city of the resting-state fMRI signal, and should be taken intoaccount in analyzing and interpreting resting-state fMRI data.

Acknowledgments

This work was supported by the Intramural ResearchProgram of the National Institute on Drug Abuse (NIDA),National Institute of Health (NIH). We would like to thankMark Lowe and Erik Beall of the Cleveland Clinic Foundationfor providing software to correct for physiological noise,Thomas Ross of the National Institute on Drug Abuse andMichael Fox of Washington University for their helpfuldiscussions on data analysis.

References

Achard, S., Salvador, R., Whitcher, B., Suckling, J., Bullmore, E., 2006. A resilient, low-frequency, small-world human brain functional network with highly connectedassociation cortical hubs. J. Neurosci. 26, 63–72.

Beall, E.B., Lowe, M.J., 2007. Isolating physiologic noise sources with independentlydetermined spatial measures. Neuroimage 37, 1286–1300.

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.

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.

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

Cordes, D., Haughton, V.M., Arfanakis, K., Carew, J.D., Turski, P.A., Moritz, C.H., Quigley,M.A., Meyerand, M.E., 2001. Frequencies contributing to functional connectivity inthe cerebral cortex in qresting-stateq data. AJNR Am. J. Neuroradiol. 22, 1326–1333.

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. U. S. A. 103, 13848–13853.

De Luca, M., Beckmann, C.F., De, S.N., Matthews, P.M., Smith, S.M., 2006. fMRI restingstate networks define distinct modes of long-distance interactions in the humanbrain. Neuroimage 29, 1359–1367.

Fox, M.D., Snyder, A.Z., Vincent, J.L., Corbetta, M., Van, E., Raichle, M.E., 2005. The humanbrain is intrinsically organized into dynamic, anticorrelated functional networks.Proc. Natl. Acad. Sci. U. S. A. 102, 9673–9678.

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. U. S. A. 100, 253–258.

Greicius, M.D., Srivastava, G., Reiss, A.L., Menon, V., 2004. Default-mode networkactivity distinguishes Alzheimer's disease from healthy aging: evidence fromfunctional MRI. Proc. Natl. Acad. Sci. U. S. A. 101, 4637–4642.

Hampson, M., Peterson, B.S., Skudlarski, P., Gatenby, J.C., Gore, J.C., 2002. Detection offunctional connectivity using temporal correlations in MR images. Hum. BrainMapp. 15, 247–262.

1055C.W. Wu et al. / NeuroImage 42 (2008) 1047–1055

Lauritzen, M., Gold, L., 2003. Brain function and neurophysiological correlates of signalsused in functional neuroimaging. J. Neurosci. 23, 3972–3980.

Leopold, D.A., Murayama, Y., Logothetis, N.K., 2003. Very slow activity fluctuations inmonkey visual cortex: implications for functional brain imaging. Cereb. Cortex 13,422–433.

Li, S.J., Li, Z., Wu, G., Zhang, M.J., Franczak, M., Antuono, P.G., 2002. Alzheimer Disease:evaluation of a functional MR imaging index as a marker. Radiology 225, 253–259.

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

Lowe, M.J., Dzemidzic, M., Lurito, J.T., Mathews, V.P., Phillips, M.D., 2000. Correlations inlow-frequency BOLD fluctuations reflect cortico-cortical connections. Neuroimage12, 582–587.

Lowe, M.J., Mock, B.J., Sorenson, J.A., 1998. Functional connectivity in single and multi-slice echoplanar imaging using resting-state fluctuations. Neuroimage 7, 119–132.

Lu, H., Zuo, Y., Gu, H., Waltz, J.A., Zhan, W., Scholl, C.A., Rea, W., Yang, Y., Stein, E.A., 2007.Synchronized delta oscillations correlate with the resting-state functional MRIsignal. Proc. Natl. Acad. Sci. U. S. A. 104, 18265–18269.

Mantini, D., Perrucci, M.G., Del, G.C., Romani, G.L., Corbetta, M., 2007. Electrophysio-logical signatures of resting state networks in the human brain. Proc. Natl. Acad. Sci.U. S. A. 104, 13170–13175.

Niessing, J., Ebisch, B., Schmidt, K.E., Niessing, M., Singer, W., Galuske, R.A., 2005. He-modynamic signals correlate tightlywith synchronized gamma oscillations. Science309, 948–951.

Raj, D., Anderson, A.W., Gore, J.C., 2001. Respiratory effects in human functionalmagnetic resonance imaging due to bulk susceptibility changes. Phys. Med. Biol. 46,3331–3340.

Salvador, R., Suckling, J., Schwarzbauer, C., Bullmore, E., 2005. Undirected graphs offrequency-dependent functional connectivity in whole brain networks. Philos.Trans. R. Soc. Lond. B Biol. Sci. 360, 937–946.

Salvador, R., Martinez, A., Pomarol-Clotet, E., Gomar, J., Vila, F., Sarro, S., Capdevila, A.,Bullmore, E., 2008. A simple view of the brain through a frequency-specificfunctional connectivity measure. Neuroimage 39, 279–289.

Salvador, R., Martinez, A., Pomarol-Clotet, E., Sarro, S., Suckling, J., Bullmore, E., 2007.Frequency based mutual information measures between clusters of brain regions infunctional magnetic resonance imaging. Neuroimage 35, 83–88.

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

Wilke, M., Logothetis, N.K., Leopold, D.A., 2006. Local field potential reflectsperceptual suppression in monkey visual cortex. Proc. Natl. Acad. Sci. U. S. A.103, 17507–17512.