Anticorrelated resting-state functional connectivity in ...cogns.northwestern.edu/cbmg/Liang2011.pdf · Anticorrelated resting-state functional connectivity in awake rat brain Zhifeng

NeuroImage xxx (2011) xxx–xxx

YNIMG-08578; No. of pages: 10; 4C: 3, 4, 5, 7, 8, 9

Contents lists available at SciVerse ScienceDirect

NeuroImage

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

Anticorrelated resting-state functional connectivity in awake rat brain

Zhifeng Liang, Jean King, Nanyin Zhang⁎Center for Comparative Neuroimaging, Department of Psychiatry, University of Massachusetts Medical School, Worcester, MA 01655, USA

⁎ Corresponding author at: Center for Comparative Neuof Psychiatry, University of Massachusetts Medical ScWorcester MA 01655, USA. Fax: +1 508 856 8090.

E-mail address: [email protected] (N. Z

1053-8119/$ – see front matter © 2011 Elsevier Inc. Aldoi:10.1016/j.neuroimage.2011.08.009

Please cite this article as: Liang, Z., et al.doi:10.1016/j.neuroimage.2011.08.009

a b s t r a c t

Article history:Received 8 June 2011Revised 11 July 2011Accepted 6 August 2011Available online xxxx

Keywords:Resting stateFunctional connectivityAnticorrelationRatAmygdalaInfralimbic cortex

Resting-state functional connectivity (RSFC) measured by functional magnetic resonance imaging has playedan essential role in understanding neural circuitry and brain diseases. The vast majority of RSFC studies havebeen focused on positive RSFC, whereas our understanding about its conceptual counterpart – negative RSFC(i.e. anticorrelation) – remains elusive. To date, anticorrelated RSFC has yet been observed without thecommonly used preprocessing step of global signal correction. However, this step can induce artifactualanticorrelation (Murphy et al., 2009), making it difficult to determine whether the observed anticorrelation inhumans is a processing artifact (Fox et al., 2005). In this report we demonstrated robust anticorrelated RSFC ina well characterized frontolimbic circuit between the infralimbic cortex (IL) and amygdala in the awake rat.This anticorrelation was anatomically specific, highly reproducible and independent of preprocessingmethods. Interestingly, this anticorrelated relationship was absent in anesthetized rats even with globalsignal correction, further supporting its functional significance. Establishing negative RSFC independent ofdata preprocessing methods will significantly enhance the applicability of RSFC in better understandingneural circuitries and brain networks. In addition, combining the neurobiological data of the IL-amygdalacircuit in rodents, the finding of the present studywill enable further investigation of the neurobiological basisunderlying anticorrelation.

roimaging (CCNI), Departmenthool, 55 Lake Avenue North,

hang).

l rights reserved.

, Anticorrelated resting-state functional conn

© 2011 Elsevier Inc. All rights reserved.

Introduction

Resting-state functional connectivity (RSFC) has been intensivelyand extensively studied using functional magnetic resonance imaging(fMRI) (Biswal et al., 1995). Resting-state fMRI (rsfMRI) measuresspatial patterns of functional connectivity across the brain bydetecting temporal correlations of low-frequency spontaneous fluc-tuations of the blood-oxygenation-level dependent (BOLD) signal.Using this technique, RSFC was consistently revealed in multiplenetworks in humans (Biswal et al., 1995; Greicius et al., 2003;Hampson et al., 2002; Lowe et al., 1998) and animals (Liang et al.,2011; Vincent et al., 2007; Zhang et al., 2010), and was sensitive toeffects like sleep, anesthesia and aging (Horovitz et al., 2009; Stevenset al., 2008). Additionally, altered RSFC was found in multiplepathological conditions (Greicius et al., 2007), indicating its vitalneurobiological and psychopathological relevance (Albert et al., 2009;Kennedy et al., 2006). Taken together, it has been strongly suggestedthat RSFC plays a very important role in brain function.

Conceptually, temporal correlations of spontaneous BOLD fluctu-ations between functionally connected brain regions should includeboth positive and negative values. More importantly, positive and

negative correlations in RSFC are most likely related to distinctneurophysiologic substrates underlying functional connections. Todate, predominant efforts have been spent investigating positiveRSFC, whereas negative correlation (i.e. anticorrelation) has beenmuch less studied.

Anticorrelation was first reported between the default modenetwork (DMN) and task-positive network (TPN) in the human (Foxet al., 2005). This temporally inverse correlation in spontaneous BOLDfluctuations was initially interpreted as competition or functionalsegregation for opposite goals between neural networks. Further, thestrength of this anticorrelation was associated with response time incognitive functions (Kelly et al., 2008) and performance in workingmemory tasks (Hampson et al., 2010). However, these interpretationswere complicated by one commonly used fMRI preprocessingprocedure-global signal regression (Murphy et al., 2009). Thisprocedure was used to remove global physiological noise in resting-state functional images, and thus improved the spatial specificity ofRSFC (Fox et al., 2009; Scholvinck et al., 2010). However, Murphy et al.(2009) in their recent study pointed out that global signal removal caninduce artifactual anticorrelation. This is because removal of theglobal signal ensures that the sum of correlation coefficients across allvoxels within the whole brain must approach zero, and thus thisprocedure mandated anticorrelation. Although several preprocessingmethods were proposed subsequently in hope to overcome thelimitation of global signal regression, contradicting results wereobtained (Anderson et al., 2010; Chang and Glover, 2009). For instance,

ectivity in awake rat brain, NeuroImage (2011),

2 Z. Liang et al. / NeuroImage xxx (2011) xxx–xxx

it was reported that anticorrelation between DMN and TPN waspresent with or without model-based physiological noise correction(Chang andGlover, 2009) and this anticorrelationwas not static (Changand Glover, 2010). Anderson et al. (2010), however, reported theabsence of anticorrelation between DMN and TPN using phase-shiftedsoft tissue regression.

The ambiguity of anticorrelated RSFC has become a major obstacleto further understanding its neurophysiologic mechanism and hassignificantly limited its applicability. Therefore, validating the exis-tence of anticorrelation that is independent of preprocessing methodsis of critical importance particularly considering the possibility thatnegative RSFC might represent a group of functional connections witha distinct neurophysiologic mechanism and thus could be crucial forbetter understanding of neural circuitries and brain diseases. In orderto achieve this goal, identifying a neural circuit with robust negativeRSFC is a crucial step.

Well-documented reports of the neural circuit between infra-limbic cortex (IL) and amygdala in the rat (Pape and Pare, 2010) haveshed light on the aforementioned issue. This frontolimbic circuit hasbeen extensively studied in several aspects: Anatomically, IL andamygdala share dense reciprocal interconnections (McDonald, 1998;Russchen, 1982a, 1982b; Sesack et al., 1989). Functionally, neurobi-ological evidence indicates that IL plays a role of inhibitory regulationof the amygdala (Rosenkranz and Grace, 2001). Specifically, IL sendsglutamatergic projections to intercalated cells of the amygdala(Amano et al., 2010; Berretta et al., 2005). Intercalated interneuronsin turn send GABAergic projections to central amygdala nucleus, thusenabling IL to exert inhibitory modulation on amygdala. Additionally,electrophysiological studies indicate that the stimulation of IL areasuppresses the basolateral amygdala activity (Likhtik et al., 2005;Rosenkranz and Grace, 2001), and decreases the responsiveness ofcentral amygdala (Quirk and Gehlert, 2003; Quirk et al., 2003). Basedon these findings, we hypothesize that anticorrelated RSFC should bepresent within the IL-amygdala circuit. To test this hypothesis, in thecurrent study we have systematically examined the temporalrelationship of spontaneous BOLD fluctuations between IL andamygdala using rsfMRI in awake and isoflurane-anesthetized rats.We reported robust anticorrelated RSFC within the IL-amygdalacircuit regardless of global signal regression in awake rats. Thisanticorrelation, however, disappeared in anesthetized rats even withglobal signal regression. Time-frequency dynamics of this negativefunctional connectivity were also examined using wavelet analysis.

Methods

Animals

Twenty four adult male Long-Evans (LE) rats (300–400 g) wereobtained from Charles River Laboratories. Animals were housed inPlexiglas cages andmaintained in ambient temperature (22–24 °C) ona reversed 12-h light:12-h dark cycle. Food and water were providedad libitum. All studies were approved by IACUC Committee of theUniversity of Massachusetts Medical School.

Acclimation procedure

Rats were acclimated toMRI restraint and noise as described in ourprevious studies (King et al., 2005). Briefly, rats were anesthetizedwith isoflurane and secured in Plexiglas stereotaxic head holdersusing plastic ear-bars. EMLA cream was applied topically to minimizepain of mechanical restraint. Animals were then placed into a blackopaque tube ‘mock scanner’ with tape-recorded scanner noises.Animals were acclimated for eight days, one session per day. The timeof exposure was increased from 15 min on the first day to 90 min ondays 6, 7 and 8, with an increment of 15 min per day.

Please cite this article as: Liang, Z., et al., Anticorrelated resting-statedoi:10.1016/j.neuroimage.2011.08.009

Animal preparation

Animal was briefly anesthetized using isoflurane and fitted into ahead restrainer with a built-in coil. The head was placed into thecylindrical head-holder with the canines secured over a bite bar, thenose secured with a nose clamp, and ear bars positioned insidethe head-holder with adjustable screws fitted into lateral sleeves. Thebody of the animal was then placed into a body restrainer. After thissetup procedure was completed, the isoflurane was removed and therestraining system was positioned in the magnet for imaging underawake condition.

Rats (16 of the 24) underwent the imaging session in the anes-thetized condition at minimal 7 days after being imaged at the awakecondition. The animal preparation procedure was the same as in theawake condition. Isoflurane gas (2%) was then delivered to theanimal through a nose cone in themagnet tomaintain the anesthetizedstate. The body temperature of the animal was monitored andmaintained at 37 °C±0.5 °C using a feedback controlled heating pad.Imaging sessions started at least 15–20 min after animals were placedin the magnet.

MR experiments

All experiments were carried out on a Bruker 4.7 T/40 cmhorizontalmagnet (Oxford, UK) interfaced with a Biospec Bruker console. Adual 1H radiofrequency (RF) coil configuration (Insight NeuroImagingSystems, Worcester, MA) consisting of a volume coil for exciting thewater proton spins and a surface coil for receiving MRI signal was used.The volume and surface coils were actively tuned and detuned toprevent mutual coil coupling.

For each session, anatomical images were acquired using a fast spin-echo sequence (RARE) with the following parameters: TR=2125 ms,RARE factor=8, TE=50 ms, matrix size=256×256, FOV=3.2 cm×3.2 cm, slice number=18, slice thickness=1 mm. Gradient-echoimages covering the whole brain were then acquired using the echo-planar imaging (EPI) sequencewith the following parameters: TR=1 s,TE=30 ms, flip angle=60°, matrix size=64×64, FOV=3.2 cm×3.2 cm, slice number=18, slice thickness=1 mm. Two hundredvolumes were acquired for each run, and six runs were obtained foreach session.

Pre-processing of imaging data

Part of the raw rsfMRI data (16 out of 24 rats) were from a previousstudy (Liang et al., 2011) and reprocessed for the purpose of the presentstudy. Data from the other 8 (out of 24) rats were acquired for thepresent study.

Imaging data was preprocessed using Medical Image Visualizationand Analysis (MIVA, http://ccni.wpi.edu/), Statistical ParametricMapping (SPM8) software (Wellcome Department of CognitiveNeurology, London, UK) and MATLAB (Mathworks, Inc., Sherborn,MA). All images were first aligned and co-registered to a fully seg-mented standard rat atlas in MIVA (Liang et al., 2011; Zhang et al.,2010). The registration procedure provided the coordinates of eachseed ROI in the image space. In this study two seed regions of interest,bilateral IL and bilateral amygdala, as well as a control seed region,unilateral (right) motor cortex, were selected (as shown in Fig. 1).After registration, all functional images were pre-processed withsteps of motion correction, spatial smoothing (FWHM=1 mm),voxel-wise linear detrending and 0.002–0.1 Hz band-pass filtering.Data sets with excessive motion (N0.5 mm, 17 runs in total) werediscarded. The time course for each individual voxel was furthercorrected for head movement by regression on the six motionparameters (translations and rotations) estimated in the procedureof motion correction. The global signal was estimated by averagingthe time courses of all voxels inside the whole-brain mask. The

functional connectivity in awake rat brain, NeuroImage (2011),

Bil t l ILBilateral IL

+2.8 +1.8

Bilateral Amygdala

-2.2 -1.2

Right Motor Cortex

R L+1.8-0.2

R L

Fig. 1. Seed ROI definitions. Three seed ROIs were used in the present study: bilateralinfralimbic cortex (IL), bilateral amygdala and unilateral (right) motor cortex. All ROIswere defined based on a fully segmented standard rat atlas in MIVA and overlaid onanatomical images (Liang et al., 2011; Zhang et al., 2010). Distances to Bregma (mm)are labeled at the bottom of each image.

3Z. Liang et al. / NeuroImage xxx (2011) xxx–xxx

ventricle and white matter signal was estimated by averaging thetime courses of all voxels inside the ventricle and white matter.

Functional connectivity analysis

Functional connectivity was evaluated using seed-based correla-tional analysis on a voxel-by-voxel basis (Zhang et al., 2010).Regionally averaged time courses from all voxels inside the seed regionswere used as reference time courses. Pearson cross-correlation co-efficients between reference time courses and the time course of eachindividual voxel were calculated. This correlational analysis was carriedout for each run. Correlation coefficients were transformed usingFisher's z transformation and then averaged across runs and animals.Subsequently, the averaged z values were transformed back to r values,yielding a mean correlation map for each seed. RSFC maps weredisplayed by thresholding the correlation coefficient at 0.21 and acluster size of 10 voxels (equivalent to uncorrected pb0.001) (Formanet al., 1995).

The reliability of functional connectivity was examined throughinter-subject reproducibility. Animals were randomly divided intotwo subgroups and functional connectivity maps were separatelycreated for each group. The strength of functional connectivity wasquantitatively compared on the voxel-by-voxel basis between thetwo subgroups.

Wavelet analysis

Wavelet transform coherence (WTC) was previously utilized foranalyzing dynamic changes between rsfMRI time series (Chang andGlover, 2010). This approach was used in the present study toinvestigate the dynamics of the anticorrelated relationship betweentime courses of IL and amygdala. Briefly, the continuous wavelettransform of a time series (xn, n=1,2…N) with equal time step Δtwas defined as:

WX n;sð Þ =ffiffiffiffiffiffiΔts

r∑N

n′=1

xnψ0 n′−n� � Δt

s

� �� �ð1Þ

Please cite this article as: Liang, Z., et al., Anticorrelated resting-statedoi:10.1016/j.neuroimage.2011.08.009

where n is the time index, s is the time scale, and ψ0 is the Morletwavelet as follows:

ψ0 ηð Þ = π−1=4eiω0ηe−12η

2

ð2Þ

where ω0 is dimensionless time and set at 6, η is dimensionlessfrequency. The wavelet power was defined as WX n; sð Þ

2. Similarly,the cross wavelet transform (XWT) for two time series is defined as

WXY n; sð Þ = WX n; sð ÞWY� n; sð Þ ð3Þ

Where * denotes complex conjugation. XWT evaluates the com-mon power of two time series in time frequency space. To evaluate thecoherence of cross wavelet transform, cross wavelet coherence wascalculated as follows:

R2n sð Þ =

S s−1WXYn sð Þ

� 2S s−1 WX

n sð Þ 2·S s−1 WYn sð Þ 2 ð4Þ

Cross wavelet coherence can be seen as local “correlation coef-ficients” in time frequency space. The statistical significance wasdetermined using Monte Carlo methods. Wavelet transform coher-ence and cross-wavelet transform were implemented with a matlabtoolbox provided by Grinsted et al. (http://www.pol.ac.uk/home/research/waveletcoherence/), and detailed information could befound in Ref. (Grinsted et al., 2004).

Results

Anticorrelated relationship between amygdala and infralimbic cortex

Fig. 2 showed the RSFC maps from the seed of IL. Anticorrelatedfunctional connectivity between IL and amygdala was evidentwithout any global signal correction (referred to as “uncorrected”hereafter) (Fig. 2a). Interestingly, negative RSFC from IL was onlyobserved in amygdala while positive RSFC was widely spread acrosscortical and subcortical areas. With the correction of the global signal(Fig. 2c), the spatial location of anticorrelation remained in amygdala.In addition, anticorrelation was also observed in some other regionssuch as hypothalamus (HT) after global signal regression. By contrast,the wide spread positive RSFC seen in the uncorrected map wasgreatly confined to more anatomically specific regions includinganterior cingulate cortex (ACC), septum, caudate-putamen (CPU),neuclus accumbens (NAcc), and dorsal lateral prefrontal cortex(dlPFC). These results were consistent with the previous literaturesuggesting that global signal regression significantly improved thespatial specificity of positive RSFC (Fox et al., 2009). The RSFC mapobtained after removing the white matter and ventricle signal(Fig. 2b) showed an intermediate pattern between the uncorrectedmap (Fig. 2a) and themap corrected for the global signal (Fig. 2c), alsoconsistent with the results in human studies (Fox et al., 2009).

The reciprocal anticorrelated relationship between the amygdalaand IL can be observed in the RSFC maps from the amygdala as shownin Fig. 3. Negative RSFC was clearly observed in IL in the uncorrectedmap (Fig. 3a). Similarly, corrections of the ventricle and white mattersignal (Fig. 3b) as well as the global signal (Fig. 3c) significantlyimproved the spatial specificity of positive RSFC between amygdalaand HT, as well as between amygdala and hippocampus. Theanticorrelation between amygdala and IL remained largely thesame. Additionally, some other regions such like CPU also showedan anticorrelated relationship with amygdala after global signalremoval. Figs. 2 and 3 collectively showed high anatomical specificityof the reciprocal anticorrelated relationship between the amygdalaand IL.

functional connectivity in awake rat brain, NeuroImage (2011),

(a)

Amygdala

(c)HT Amygdala

2.2-2.3--4.2 2.2-2.3--4.2

Amygdala ACC Septum

CPU+0.82.0-2.1-

0

0.5

IL+0.82.0-2.1-

0

0.5

CPU

dlPFCACC

IL8.3+8.2+8.1+

-0.5R L8.3+8.2+8.1+

-0.5NAccIL

R L

(d)(b)HT

2.2-2.3--4.2-3.2-4.2Amygdala

ACC

-2.2

Septum

+0.82.0-2.1-

0.5

2.0-2.1-

0.5

CPU+0.8

dlPFCACC

Amygdala

0

-0.58.3+8.2+8.1+

R L8.2+8.1+

0

-0.5+3.8

NAccIL

R L

Fig. 2. IL RSFC maps. (a) The IL RSFC map in the awake condition without correction of any global signal. (b) The IL RSFC map in the awake condition with correction of the ventricleand white matter signal. (c) The IL RSFC map in the awake condition with correction of the global signal. (d) The IL RSFC map in the anesthetized condition with correction of theglobal signal. Data from all maps were corrected for six movement parameters. All maps were overlaid on anatomical images. Distances to Bregma (mm) are labeled at the bottom ofeach image.

4 Z. Liang et al. / NeuroImage xxx (2011) xxx–xxx

Absence of anticorrelation in anesthetized rats

Considering that one major function of the IL-amygdala circuitry isregulating affective behaviors, it can be expected that anesthesia willdisrupt the functional connectivity within the IL-amygdala circuit.Indeed, our data showed that the anticorrelated relationship betweenIL and amygdala observed in awake rats was completely abolished inisoflurane-anesthetized rats even with the global signal correction asshown in both Figs. 2d and 3d. This remarkable difference indicatedthat: i) the anticorrelated relationship observed in awake rats was notinduced by preprocessing methods because the same preprocessingmethods were applied to both awake and anesthetized rats data, andii) the anticorrelation between amygdala and IL has importantfunctional relevance that is impacted by anesthesia.

RSFC maps of unilateral motor cortex

In order to examine the specificity of the anticorrelation betweenamygdala and IL, a control seed region unilateral motor cortex wasalso selected. Fig. 4 demonstrated the RSFC maps of unilateral (right)motor cortex in awake (Fig. 4a) and anesthetized (Fig. 4b) rats,respectively. Both maps were obtained after the global signalregression. In awake rats, we observed strong functional connectionsbetween right and the left motor cortices, whereas this bilateral

Please cite this article as: Liang, Z., et al., Anticorrelated resting-statedoi:10.1016/j.neuroimage.2011.08.009

connection was less apparent in anesthetized rats. This result isconsistent with the notion that anesthesia reduced the strength ofRSFC (Liu et al., 2010). More importantly, no anticorrelated RSFC wasobserved in either awake or anesthetized group, suggesting thatthe anticorrelated relationship observed between IL and amygdalawas specific to the frontolimbic circuit as oppose to a general effect.

Distributions of correlation coefficients of RSFC between IL and amygdala

It was previously reported that global signal regression dramat-ically changed the distribution of computed correlation coefficientsin RSFC maps: (i) artifactual negative correlations were induced, and(ii) the distribution became approximately normal with a meancorrelation value close to zero (Fox et al., 2009). In the present study,we independently extracted regional mean time courses fromamygdala and IL for each run, with or without global signal regression,and calculated their temporal correlation coefficients. Fig. 5 showedthe histograms of correlation coefficients before (Fig. 5a) and after(Fig. 5b) global signal regression. The distribution of correlationcoefficients indeed shifted toward a Gaussian shape with global signalregression (Fig. 5b). Nevertheless, the majority of correlation coef-ficients was in the negative range regardless of global signal regression(Fig. 5a,mean correlation coefficient=−0.20, Fig. 5b, mean correlationcoefficient=−0.37).

functional connectivity in awake rat brain, NeuroImage (2011),

(a) (c)HT Amygdala

2.2-2.3--4.2 2.2-2.3--4.2

Amygdala

+0.8 2.0-2.1-

IL0

0.5

+0.8 2.0-2.1-

0

0.5

CPU IL8.3+8.2+8.1+

-0.5 R L8.3+8.2+8.1+

-0.5 CPU IL

R L(b)

HT Amygdala(d)

2.2-2.3--4.2

Amygdala

2.2-2.3--4.2

+0.8 2.0-2.1-

0.5

+0.8 2.0-2.1-

0.5

8.3+8.2+8.1+

IL0

-0.5 R L0

-0.5 8.3+8.2+8.1+ R L

Fig. 3. Amygdala RSFC maps. (a) The Amygdala RSFC map in the awake condition without correction of any global signal. (b) The Amygdala RSFC map in the awake condition withcorrectionof the ventricle andwhitematter signal. (c) TheAmygdalaRSFCmap in the awake conditionwith correction of the global signal. (d) TheAmygdala RSFCmap in the anesthetizedcondition with correction of the global signal. Data from all mapswere corrected for six movement parameters. All mapswere overlaid on anatomical images. Distances to Bregma (mm)are labeled at the bottom of each image.

5Z. Liang et al. / NeuroImage xxx (2011) xxx–xxx

Reliability of anticorrelation between amygdala and infralimbic cortex

To test the reliability of the anticorrelated relationship betweenspontaneous BOLD fluctuations in amygdala and IL, data from all

(a) (

2.2-2.3--4.2

+0.8 2.0-2.1-

0

-0.5

0.5

8.3+8.2+8.1+R L

Motor

Fig. 4. RSFC maps from unilateral (right) motor cortex in (a) the awake condition and (bDistances to Bregma (mm) are labeled at the bottom of each image.

Please cite this article as: Liang, Z., et al., Anticorrelated resting-statedoi:10.1016/j.neuroimage.2011.08.009

animals were randomly divided into two subgroups and a RSFC map,with the seed of IL, was individually obtained for each subgroup.Figs. 6a and b showed the RSFC maps from the two subgroups,demonstrating excellent consistency. Quantitatively, the computed

b)

2.2-2.3--4.2

+0.8 2.0-2.1-

0

-0.5

0.5

8.3+8.2+8.1+R L

) the anesthetized condition. Both maps were obtained after global signal regression.

functional connectivity in awake rat brain, NeuroImage (2011),

Nu

mb

er o

f V

oxe

lsN

um

ber

of

Vo

xels

Correlation Coefficient

Correlation Coefficient

(a)

(b)

Fig. 5. Histograms of correlation coefficients between regional mean time courses of ILand amygdala (a) without and (b) with global signal regression.

6 Z. Liang et al. / NeuroImage xxx (2011) xxx–xxx

correlation coefficients between the two RSFC maps well agree witheach other on a voxel-by-voxel basis (Fig. 6c, r=0.58, pb10−5).Similar results can be obtained from the seed of amygdala (datanot shown). These results suggest that the anticorrelated relation-ship between amygdala and IL observed in awake rats was highlyreliable.

Time-frequency dynamics of anticorrelation between amygdala andinfralimbic cortex

WTC was utilized to investigate time-frequency dynamics of theanticorrelated relationship between amygdala and IL, with andwithout global signal regression. We observed a strong anti-phaserelationship in cross-wavelet power andwavelet transform coherencebetween the time courses of IL and amygdala (Fig. 7 showed oneexample), and this anti-phase relationship was relatively consistentthroughout the whole scan. In addition, the anti-phase relationshipwas evident without global signal regression.

Please cite this article as: Liang, Z., et al., Anticorrelated resting-statedoi:10.1016/j.neuroimage.2011.08.009

Discussion

In the present study we have characterized the anticorrelatedtemporal relationship between spontaneous BOLD fluctuations in ILand amygdala in awake rats. To the best of our knowledge, this is thefirst study investigating negative RSFC in animals. Independent ofpreprocessing methods, we observed robust anticorrelation withinthis anatomically well-defined frontolimbic circuit. In addition, thisanticorrelation was highly reliable as reflected from its highreproducibility between two randomly divided subgroups. Moreover,this anticorrelation was between two distinct and distant anatomicalregions, and contained high anatomical specificity. Furthermore, theanticorrelated relationship between the two regions was absent inanesthetized rats even with global signal regression. Taken together,data of the present study have provided strong evidence validatingthe existence of anticorrelated RSFC.

The influence of global signal regression

Although the presence of anticorrelation was independent ofglobal signal regression, it was noticeable that global signal regressionindeed affected the spatial pattern of RSFC maps and the distributionof correlation coefficients. Consistent with previous reports (Fox et al.,2009), global signal removal significantly improved the spatialspecificity of positive RSFC. Interestingly, global signal regressionenlarged areas of negative RSFC. Compared to uncorrected maps,additional regions such as hypothalamus (HT) showed anticorrelatedrelationship with IL (Fig. 2c), and CPU showed anticorrelatedrelationship with amygdala (Fig. 3c). Although the origin of theseenlarged anticorrelated areas after global signal removal was notclear, we speculate that it could result from the propagation effect ofindirect connectivity. It is well known that amygdala and HT aretightly connected as part of the amygdala–hypothalamic–pituitary–adrenal axis which is responsible for the autonomic stress/fear bodyresponse. This functional connection resulted in positive correlationsbetween BOLD fluctuations in amygdala and HT as shown in Figs. 3band c. Since IL and amygdala contained an anticorrelated relationshipin their BOLD fluctuations, it can be predicted that this anticorrelatedrelationship would propagate to areas that were positively correlatedto amygdala such as HT. This propagation effect, being masked by theglobal signal in uncorrected maps, became detectable after the globalsignal was removed. Similar argument can be used to explainthe anticorrelation between amygdala and CPU appeared after globalsignal regression. However, it has to be noted that we cannot ruleout the possibility that IL and HT are directly connected with ananticorrelated relationship. Further experiments are needed to resolvethis issue.

The distribution of correlation coefficients between IL andamygdala was also altered by global signal regression, shifting to anapproximately normal distribution centered at about CC=−0.37.However, it did not change the sign of the majority of correlationvalues. Taken together, although global signal regression indeedaffected the resultant RSFC maps as expected (Murphy et al., 2009),it was clearly not attributing to the anticorrelated RSFC observedbetween IL and amygdala.

Impact of anesthesia

The anticorrelated relationship between IL and amygdala wasabsent in the anesthetized condition. Accumulating evidence hassuggested that anesthesia profoundly affects RSFC. For instance, Luet al. (2007) demonstrated a dose–dependent decrease of cross-hemispheric functional connectivity in α-chloralose-anesthetizedrats. Similarly, Liu et al. (2010) found that intrinsic BOLD fluctuationsand functional connectivity in the resting rat were strongly dependenton anesthesia depth. In human subjects, functional connectivity in the

functional connectivity in awake rat brain, NeuroImage (2011),

(a)

-7.2 -6.2 -5.2 -4.2

(b)

-7.2 -6.2 -5.2 -4.2

-3.2 -2.2 -1.2 -0.2

7.2 6.2 5.2 4.2

-3.2 -2.2 -1.2 -0.2

7.2 6.2 5.2 4.2

+0.8 +1.8 +2.8 +3.8 0

0 5

0.5

+0.8 +1.8 +2.8 +3.8 0

0.5

0.5

R = 0.58

1st half+4.8 +5.8 +6.8 +7.8

-0.5

2nd half+4.8 +5.8 +6.8 +7.8

-

(c)

RSFC (1st half of Rats)

RS

FC

(2n

d h

alf

of

Rat

s)

Fig. 6. Reproducibility of anticorrelation between IL and amygdala. Data from all animals were randomly divided into two subgroups. (a) The IL RSFC map generated from onesubgroup with global signal regression. (b) The IL RSFCmap generated from the other subgroup with global signal regression. Distances to Bregma (mm) are labeled at the bottom ofeach image. (c) The voxel-to-voxel correlation of the RSFC strength between the two subgroups.

7Z. Liang et al. / NeuroImage xxx (2011) xxx–xxx

motor cortices was completely ablated with deep anesthesia (Peltieret al., 2005). Taken together, these results suggest that anesthesiasignificantly weakens RSFC relative to the awake condition. In thepresent study, anesthesia weakened the positive RSFC between leftand right motor cortex (Fig. 4), and completely abolished the negativeRSFC between IL and amygdala regardless of preprocessing pro-cedures (Figs. 2d and 3d). These results were in line with previousanimal imaging studies indicating that anesthesia reduces the ampli-tude of RSFC (Liu et al., 2010). More importantly, distinct differencebetween awake and anesthetized rats further ruled out the possibilitythat the anticorrelation observed at the awake condition was aprocessing artifact because the same preprocessing procedures wereapplied to both conditions. Furthermore, our data demonstrated thatRSFC might serve as a sensitive marker for the functionality of braincircuitry given the fact that the IL-amygdala circuit is critically involvedin affective behaviors that are impacted by anesthesia. This result alsoprovided important evidence supporting the advantage of measuringRSFC in awake animals particularly in studies of neural circuitriessubserving cognitive and emotional functions (Liang et al., 2011; Zhanget al., 2010).

Possible neural mechanism

The anatomy and function of the IL-amygdala circuit have beenwell studied using various methods. These studies may shed light onunderstanding the neural mechanism underlying the negative RSFCwithin this circuit. It is well known that the IL-amygdala circuitry

Please cite this article as: Liang, Z., et al., Anticorrelated resting-statedoi:10.1016/j.neuroimage.2011.08.009

is implicated in affective behaviors such as fear conditioning andextinction in rodents (LeDoux, 2000), as well as in emotion regulationin humans and nonhuman primates (Phelps and LeDoux, 2005;Phelps et al., 2004). In addition, malfunction in this circuit has beenfound to be tightly linked to mood and anxiety disorders (Shin et al.,2004). Anatomically, IL and amygdala share dense reciprocalconnections (McDonald, 1998; Russchen, 1982a, 1982b; Sesacket al., 1989). These physical connections provide the anatomicalbasis of the observed anticorrelated RSFC between the two regions.More importantly, there is substantial evidence suggesting IL couldexert inhibitory regulation on amygdala. For instance, electricalstimulation of IL reduces responsiveness of central nucleus outputneurons in the amygdala to basolateral amygdala (BLA) stimulation(Quirk et al., 2003), and chemical stimulation of IL activate cFos in theITC neurons which are known to inhibit central nucleus outputneurons (Pare and Smith, 1993). These results collectively suggestthat the anticorrelated relationship between amygdala and IL ob-served in the present study could arise from an inhibitory interactionbetween them.

Time-frequency dynamics

One recent human study examined the anticorrelation betweenDMN and TPN by employing WTC and found that the anticorrelationbetween these two networks was not static (Chang and Glover, 2010).In the present study, wavelet analysis revealed a more stable anti-phase relationship between spontaneous BOLD fluctuations from IL

functional connectivity in awake rat brain, NeuroImage (2011),

Without correction

(a)

With correction

(d)(c)

(b)

Fig. 7. Wavelet transform coherence analysis revealed anti-phase relationship between IL and amygdala. (a) Cross wavelet power of IL and amygdala time series from onerepresentative RSFC run. (b) Cross wavelet coherence of IL and amygdala time series from the same RSFC run. Time series in (a) and (b) were not corrected for the global signal.(c) Cross wavelet power of IL and amygdala time series from the same RSFC run after global signal regression. (d) Cross wavelet coherence of IL and amygdala time series after globalsignal regression. X axis represents time (s) and Y axis represents period.

8 Z. Liang et al. / NeuroImage xxx (2011) xxx–xxx

and amygdala in awake rats. This difference may suggest a strongeranticorrelated relationship over time in rats andmay also explain whyit can be observed even in the mask of the global signal.

The influence of motion

In the present study, effects of movement in rats wasminimized byusing (i) motion correction; (ii) discarding data sets with excessivemovement (N0.5 mm, 17 sessions in total); and (iii) regressing outmotion correction covariates. However, we did notice that movementof awake rats was significantly larger than that of anesthetized rats.There is the possibility that the difference in anticorrelation betweenawake and anesthetized rats was due to different levels of movementduring data acquisition. To rule out this possibility, data from awakerats with minimal movement (b0.125 mm, i.e. 1/4 voxel size) wereselected (33 sessions in total, 27.3% of the whole data set). In thissubgroup, movement in awake rats was not significantly differentfrom anesthetized rats (two-sample t-test, p=0.19). Fig. 8 showedthat strong anticorrelated RSFC between amygdala and IL waspersistent in this subgroup. This result suggests that the antic-orrelated RSFC observed in the present study cannot be attributed tothe factor of movement.

Please cite this article as: Liang, Z., et al., Anticorrelated resting-statedoi:10.1016/j.neuroimage.2011.08.009

Limitations and future implications

Although the current study has shown for the first time robustfMRI anticorrelation in a systemwith known inhibitory connections, itcannot resolve the debate on the origin of anticorrelations in humans.There are important differences between the rat and human resultswhich prohibit this extension. First, the rat anticorrelations arepresent even prior to global regression, but the human anticorrela-tions are not. Second, the rat anticorrelations are between two specificanatomic regions with known strong anatomical connections, thehuman anticorrelations are between two wide-spread networks.Nevertheless, the finding of this study makes it possible to uncoverthe neurophysiologic basis of anticorrelated RSFC when combiningwith other techniques such as neuron recordings. Since anticorrelatedRSFC represents a group of functional connections with distinctneurophysiologic features, it can tremendously contribute to studiesof neural circuitries and brain networks. More importantly, given thevital role that RSFC plays in regulating brain function at normal andpathological conditions, the results of the present study can help testthe hypothesis that negative RSFC might serve as an importantbiomarker to evaluate the functionality of neural circuits at normaland pathological conditions. Therefore, the present study has openeda new avenue to further expanding the applicability of rsfMRI.

functional connectivity in awake rat brain, NeuroImage (2011),

(b)(a)

2.2-2.3--4.2 -3.2 -4.2 -2.2

+0.8 2.0-2.1- 2.0-2.1- +0.8

8.3+8.2+8.1+

0

-0.5

0.5

8.2+8.1+

0

-0.5

0.5

+3.8 R L R L

Fig. 8. IL RSFCmaps from a subgroup of awake rats with movement smaller than 0.125 mm (a) without and (b) with global signal regression. Movement in this subgroup was similarto that in anesthetized rats (two sample t-test, p=0.19).

9Z. Liang et al. / NeuroImage xxx (2011) xxx–xxx

Acknowledgment

We thank Dr Wei Huang and Ms Meghan Heffernan for theirtechnical assistance. We also would like to thank anonymousreviewers for their insightful comments. This publication was madepossible by the NIH Grant Number 1R01MH067096-02 (PI: Jean King,PhD) and 5R01DA021846-02 (PI: Jean King, PhD) from the NationalInstitute of Health, and the institutional fund from the University ofMassachusetts Medical School. Its contents are solely the responsi-bility of the authors and do not necessarily represent the official viewsof the NIH.

References

Albert, N.B., Robertson, E.M., Miall, R.C., 2009. The resting human brain and motorlearning. Curr. Biol. 19, 1023–1027.

Amano, T., Unal, C.T., Pare, D., 2010. Synaptic correlates of fear extinction in theamygdala. Nat. Neurosci. 13, 489–494.

Anderson, J.S., Druzgal, T.J., Lopez-Larson, M., Jeong, E.K., Desai, K., Yurgelun-Todd, D.Network anticorrelations, global regression, and phase-shifted soft tissue correc-tion. Hum. Brain Mapp. 32, 919–934.

Berretta, S., Pantazopoulos, H., Caldera, M., Pantazopoulos, P., Pare, D., 2005. Infralimbiccortex activation increases c-Fos expression in intercalated neurons of theamygdala. Neuroscience 132, 943–953.

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

Chang, C., Glover, G.H., 2009. Effects of model-based physiological noise correction ondefaultmodenetwork anti-correlations and correlations. Neuroimage 47, 1448–1459.

Chang, C., Glover, G.H., 2010. Time-frequency dynamics of resting-state brainconnectivity measured with fMRI. Neuroimage 50, 81–98.

Forman, S.D., Cohen, J.D., Fitzgerald, M., Eddy, W.F., Mintun, M.A., Noll, D.C., 1995.Improved assessment of significant activation in functional magnetic resonanceimaging (fMRI): use of a cluster-size threshold. Magn. Reson. Med. 33, 636–647.

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

Fox, M.D., Zhang, D., Snyder, A.Z., Raichle, M.E., 2009. The global signal and observedanticorrelated resting state brain networks. J. Neurophysiol. 101, 3270–3283.

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., Flores, B.H., Menon, V., Glover, G.H., Solvason, H.B., Kenna, H., Reiss, A.L.,Schatzberg, A.F., 2007. Resting-state functional connectivity in major depression:abnormally increased contributions from subgenual cingulate cortex and thalamus.Biol. Psychiatry 62, 429–437.

Grinsted, A., Moore, J.C., Jevrejeva, S., 2004. Application of the cross wavelet transformand wavelet coherence to geophysical time series. Nonlinear Processes Geophys.11, 561–566.

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.

Please cite this article as: Liang, Z., et al., Anticorrelated resting-statedoi:10.1016/j.neuroimage.2011.08.009

Hampson, M., Driesen, N., Roth, J.K., Gore, J.C., Constable, R.T., 2010. Functionalconnectivity between task-positive and task-negative brain areas and its relation toworking memory performance. Magn. Reson. Imaging 28, 1051–1057.

Horovitz, S.G., Braun, A.R., Carr, W.S., Picchioni, D., Balkin, T.J., Fukunaga, M., Duyn, J.H.,2009. Decoupling of the brain's default mode network during deep sleep. Proc. Natl.Acad. Sci. U.S.A. 106, 11376–11381.

Kelly, A.M., Uddin, L.Q., Biswal, B.B., Castellanos, F.X., Milham, M.P., 2008. Competitionbetween functional brain networks mediates behavioral variability. Neuroimage39, 527–537.

Kennedy, D.P., Redcay, E., Courchesne, E., 2006. Failing to deactivate: resting functionalabnormalities in autism. Proc. Natl. Acad. Sci. U.S.A. 103, 8275–8280.

King, J.A., Garelick, T.S., Brevard, M.E., Chen, W., Messenger, T.L., Duong, T.Q., Ferris, C.F.,2005. Procedure forminimizing stress for fMRI studies in conscious rats. J. Neurosci.Methods 148, 154–160.

LeDoux, J.E., 2000. Emotion circuits in the brain. Annu. Rev. Neurosci. 23, 155–184.Liang, Z., King, J., Zhang, N., 2011. Uncovering intrinsic connectional architecture of

functional networks in awake rat brain. J. Neurosci. 31, 3776–3783.Likhtik, E., Pelletier, J.G., Paz, R., Pare, D., 2005. Prefrontal control of the amygdala.

J. Neurosci. 25, 7429–7437.Liu, X., Zhu, X.H., Zhang, Y., Chen, W., 2010. Neural origin of spontaneous hemodynamic

fluctuations in rats under burst-suppression anesthesia condition. Cereb. Cortex 21,374–384.

Lowe, M.J., Mock, B.J., Sorenson, J.A., 1998. Functional connectivity in single andmultisliceechoplanar 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.

McDonald, A.J., 1998. Cortical pathways to the mammalian amygdala. Prog. Neurobiol.55, 257–332.

Murphy, K., Birn, R.M., Handwerker, D.A., Jones, T.B., Bandettini, P.A., 2009. The impactof global signal regression on resting state correlations: are anti-correlatednetworks introduced? Neuroimage 44, 893–905.

Pape, H.C., Pare, D., 2010. Plastic synaptic networks of the amygdala for the acquisition,expression, and extinction of conditioned fear. Physiol. Rev. 90, 419–463.

Pare, D., Smith, Y., 1993. The intercalated cell masses project to the central and medialnuclei of the amygdala in cats. Neuroscience 57, 1077–1090.

Peltier, S.J., Kerssens, C., Hamann, S.B., Sebel, P.S., Byas-Smith, M., Hu, X., 2005.Functional connectivity changes with concentration of sevoflurane anesthesia.Neuroreport 16, 285–288.

Phelps, E.A., LeDoux, J.E., 2005. Contributions of the amygdala to emotion processing:from animal models to human behavior. Neuron 48, 175–187.

Phelps, E.A., Delgado, M.R., Nearing, K.I., LeDoux, J.E., 2004. Extinction learning inhumans: role of the amygdala and vmPFC. Neuron 43, 897–905.

Quirk, G.J., Gehlert, D.R., 2003. Inhibition of the amygdala: key to pathological states?Ann. N. Y. Acad. Sci. 985, 263–272.

Quirk, G.J., Likhtik, E., Pelletier, J.G., Pare, D., 2003. Stimulation of medial prefrontalcortex decreases the responsiveness of central amygdala output neurons. J. Neurosci.23, 8800–8807.

Rosenkranz, J.A., Grace, A.A., 2001. Dopamine attenuates prefrontal cortical suppressionof sensory inputs to the basolateral amygdala of rats. J. Neurosci. 21, 4090–4103.

Russchen, F.T., 1982a. Amygdalopetal projections in the cat. I. Cortical afferentconnections. A study with retrograde and anterograde tracing techniques. J. Comp.Neurol. 206, 159–179.

Russchen, F.T., 1982b. Amygdalopetal projections in the cat. II. Subcortical afferentconnections. A study with retrograde tracing techniques. J. Comp. Neurol. 207,157–176.

functional connectivity in awake rat brain, NeuroImage (2011),

10 Z. Liang et al. / NeuroImage xxx (2011) xxx–xxx

Scholvinck, M.L., Maier, A., Ye, F.Q., Duyn, J.H., Leopold, D.A., 2010. Neural basis of globalresting-state fMRI activity. Proc. Natl. Acad. Sci. U.S.A. 107, 10238–10243.

Sesack, S.R., Deutch, A.Y., Roth, R.H., Bunney, B.S., 1989. Topographical organization ofthe efferent projections of the medial prefrontal cortex in the rat: an anterogradetract-tracing study with Phaseolus vulgaris leucoagglutinin. J. Comp. Neurol. 290,213–242.

Shin, L.M., Orr, S.P., Carson, M.A., Rauch, S.L., Macklin, M.L., Lasko, N.B., Peters, P.M.,Metzger, L.J., Dougherty, D.D., Cannistraro, P.A., Alpert, N.M., Fischman, A.J., Pitman,R.K., 2004. Regional cerebral blood flow in the amygdala and medial prefrontal

Please cite this article as: Liang, Z., et al., Anticorrelated resting-statedoi:10.1016/j.neuroimage.2011.08.009

cortex during traumatic imagery in male and female Vietnam veterans with PTSD.Arch. Gen. Psychiatry 61, 168–176.

Stevens, W.D., Hasher, L., Chiew, K.S., Grady, C.L., 2008. A neural mechanism underlyingmemory failure in older adults. J. Neurosci. 28, 12820–12824.

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.

Zhang, N., Rane, P., Huang, W., Liang, Z., Kennedy, D., Frazier, J.A., King, J., 2010. Mappingresting-state brain networks in conscious animals. J. Neurosci.Methods 189, 186–196.

functional connectivity in awake rat brain, NeuroImage (2011),