Alterations in effective connectivity anchoredon the insula in major depressive disorder

Sarina J. Iwabuchia, Daihui Pengb,nn, Yiru Fangb, Kaida Jiangb,Elizabeth B. Liddlea, Peter F. Liddlea, Lena Palaniyappana,c,n

aTranslational Neuroimaging, Division of Psychiatry, Institute of Mental Health, University of Nottingham,Room-09 C Floor, Triumph Road, Nottingham, NG7 2TU England, UKbDivision of mood disorders, Shanghai Mental Health Center, Shanghai Jiao Tong University School ofMedicine, ChinacNottinghamshire Healthcare NHS Trust, Nottingham, UK

Received 18 March 2014; received in revised form 26 July 2014; accepted 10 August 2014

KEYWORDSEffective connectivity;Insula;Depression;Granger causality;Resting fMRI

AbstractRecent work has identi!ed disruption of several brain networks involving limbic and corticalregions that contribute to the generation of diverse symptoms of major depressive disorder(MDD). Of particular interest are the networks anchored on the right anterior insula, whichbinds the cortical and limbic regions to enable key functions that integrate bottom-up and top-down information in emotional and cognitive processing. Emotional appraisal has been linked toa presumed hierarchy of processing, from sensory percepts to affective states. But it is unclearwhether the network level dysfunction seen in depression relates to a breakdown of thispresumed hierarchical processing system from sensory to higher cognitive regions, mediated bycore limbic regions (e.g. insula). In 16 patients with current MDD, and 16 healthy controls, weinvestigated differences in directional in"uences between anterior insula and the rest of thebrain using resting-state functional magnetic resonance imaging (fMRI) and Granger-causalanalysis (GCA), using anterior insula as a seed region. Results showed a failure of reciprocalin"uence between insula and higher frontal regions (dorsomedial prefrontal cortex) in additionto a weakening of in"uences from sensory regions (pulvinar and visual cortex) to the insula. Thissuggests dysfunction of both sensory and putative self-processing regulatory loops centeredaround the insula in MDD. For the !rst time, we demonstrate a network-level processing defect

www.elsevier.com/locate/euroneuro

http://dx.doi.org/10.1016/j.euroneuro.2014.08.0050924-977X/& 2014 Elsevier B.V. and ECNP. All rights reserved.

Abbreviations: AAL, Automated Anatomical Labeling; ACC, anterior cingulated cortex; BOLD, blood–oxygen level dependent; CBT,Cognitive Behavioral Therapy; DMPFC, dorsomedial prefrontal cortex; DPARSF, Data Processing Assistant for resting-state fMRI; DSM,Diagnostic and Statistical Manual for Mental Disorders; EPI, echo-planar imaging; FOV, !eld of view; FWE, Family-wise Error; FWHM, full-width, half-maximum; GCA, Granger causality analysis; HAMA, Hamilton Anxiety Scale; HRF, haemodynamic response function; HDRS,Hamilton Depression Rating Scale; MNI, Montreal Neurological Institute; rAI, right anterior insula; SFG, superior frontal gyrus; SPM, StatisticalParametric Mapping; TE, echo time; TMS, transcranial magnetic stimulation; TR, repetition time

nCorresponding author at: Translational Neuroimaging, Division of Psychiatry, Institute of Mental Health, University of Nottingham,Room-09 C Floor, Triumph Road, Nottingham, NG7 2TU England, UK.

nnCo-corresponding author.E-mail address: lena.palaniyappan@nottingham.ac.uk (L. Palaniyappan).

European Neuropsychopharmacology (]]]]) ], ]]]–]]]

Please cite this article as: Iwabuchi, S.J., et al., Alterations in effective connectivity anchored on the insula in major depressive disorder.European Neuropsychopharmacology (2014), http://dx.doi.org/10.1016/j.euroneuro.2014.08.005

extending from sensory to frontal regions through insula in depression. Within limitations ofinferences drawn from GCA of resting fMRI, we offer a novel framework to advance targetednetwork modulation approaches to treat depression.& 2014 Elsevier B.V. and ECNP. All rights reserved.

1. Introduction

With a lifetime prevalence of approximately 1 in 10individuals, Major Depressive Disorder (MDD) is the mostcostly of brain disorders (Sobocki et al., 2006) and has beenhighlighted as the 2nd most disabling condition for itsimpact on quality of life (Murray and Lopez, 1997) with ahigh degree of association with suicide rates (Cavanaghet al., 2003). In the last decade, important progress hasbeen made in understanding the neural basis of depression,and key brain networks have been identi!ed to play a role ingenerating the diverse symptoms of depression.

A prominent neural network model for MDD proposed byMayberg et al. (1999) suggests that depression is characterizedby a disruption in the limbic–cortical network. Using PET, it hasbeen shown that in healthy individuals, activity is decreased inthe limbic and paralimbic structures over the cortical, while indepression, this pattern is reversed with overactivity of limbicregions and decreases in dorsal cortical regions (Mayberg et al.,1999). To offer support for this model, Hamilton et al. (2011a)have recently documented increased excitation within thelimbic and paralimbic structures (e.g. the hippocampus), andincreased inhibition of cortical structures (e.g. dorsomedialprefrontal cortex (DMPFC) and cuneus/posterior cingulate) bythe commonly implicated ventral anterior cingulate cortex(vACC). A region in the DMPFC, termed the ‘dorsal nexus’,has also shown to possess abnormally increased connectivitywith the three major networks (i.e. cognitive control, affective,default mode) in MDD (Sheline et al., 2010). This ‘hot-wiring’pattern results in widespread information processing imbalanceacross the brain, which is thought to underlie the complexsymptoms and cognitive de!cits observed in MDD.

An additional structure that has garnered increased attentionas a key contributor to the pathophysiology of MDD is theanterior insula. The anterior insula shares extensive and wide-spread anatomical connections to cortical and limbic regionsand plays a crucial role in the coordination between thecognitive control and default mode network (Menon andUddin, 2010). The rich connectivity of the anterior insula toboth sensory processing regions (visual, auditory cortices andthalamus) and the so-called self-processing multimodal regions(prefrontal and anterior temporal) suggests a unique role foranterior insula in the ordered or hierarchical informationprocessing, presumed to include bottom-up (sensory to multi-modal) and top-down (multimodal to sensory) reciprocal in"u-ences. Given the insula's posited role in functions related toemotional processing, in addition to its connections to limbicregions, it seems reasonable that the diverse symptoms of MDDmay be associated with impaired functioning of this region.Indeed, the functional activity and connectivity of the insulahave been shown to be perturbed in several studies of patientswith depression (Busatto, 2013; Hamilton et al., 2012, 2011b;Sliz and Hayley, 2012; Sprengelmeyer et al., 2011). For

example, the activity of the right fronto-insular network wasattenuated in patients with depression, and this was associatedwith greater levels of maladaptive rumination (Hamilton et al.,2011b). Studies of regional homogeneity, which measures localconnectivity, have also shown reductions in the right insula,which correlated positively with anxiety and hopelessness (Yaoet al., 2009). Similar patterns were also observed in !rst-degreerelatives of MDD patients, suggesting that insular dysfunctionrepresents a crucial vulnerability marker for depression (Liuet al., 2010).

Functional changes have also been reported in the insulafollowing various treatments for MDD. Regional cerebralblood "ow of the insula was reduced following repetitivetranscranial magnetic stimulation (rTMS) treatment, whichalso correlated with treatment ef!cacy (Kito et al., 2011).Other studies have also revealed reductions in insulaactivity with antidepressant treatment (Delaveau et al.,2011; Fitzgerald et al., 2008). In a recent PET study, theright anterior insula (rAI) was localized as a crucial regionfor differentiating MDD patients who are responsive to CBTor antidepressants (McGrath et al., 2013). Hypometabolismof the rAI predicted response to CBT, while hypermetabo-lism predicted a patient to respond to antidepressants.Therefore, the insula offers great promise as a target regionfor treatment of MDD, where its modulation will havewidespread effects on the connectivity of networks thatare disrupted in MDD. Studying the directed in"uence(effective connectivity) of the insula on other structuresimplicated in depression is crucial for our understanding ofthe neural circuitry of depression.

We therefore sought to investigate the effect of depres-sion on the effective connectivity of the rAI with the rest ofthe brain. We employed Granger causality analysis (GCA) onresting-state functional magnetic resonance imaging (fMRI)data to investigate whether positive or negative in"uencesof the rAI on the rest of the brain differ in patients withcurrent major depression. We anticipated a reduction in theconnectivity of the fronto-insula network in MDD patients,as well as more widespread abnormalities in connectivitybetween the rAI and limbic and cortical regions. It isimportant to note that while Granger-causality allowsdirectional inferences to be made on the basis of temporalprecedence observed within the limits of the sparse resolu-tion offered by fMRI, the inferences do not provide con-clusive evidence on underlying neural state shifts.

2. Experimental procedures

2.1. Participants

Sixteen patients and 16 healthy controls were recruited fromShanghai Mental Health Centre and Shanghai Huashan Hospital.The MDD group consisted of 9 females and 7 males, aged 26–45 years

S.J. Iwabuchi et al.2

Please cite this article as: Iwabuchi, S.J., et al., Alterations in effective connectivity anchored on the insula in major depressive disorder.European Neuropsychopharmacology (2014), http://dx.doi.org/10.1016/j.euroneuro.2014.08.005

(mean 34.4376.72 years) and received 6–17 years (mean15.6371.99 years) education. All patients were !rst-onset MDDand drug-naïve (no current or past exposure to antidepressants)with illness duration between 3.6 and 14.57 weeks (mean8.0174.19). Patients were included in the study if they met DSM-IV diagnosis criteria of MDD, were aged between 20 and 50 years,had no history of brain injury, primary psychotic ideations, and nosubstance abuse within the past six months, no behavioral indica-tions of possible impaired mental status, and scored Z20 on the 24item-Hamilton Depression Rating Scale (HDRS) and o7 on theHamilton Anxiety Scale (HAMA). Participants were excluded if theyhad any mental disorder related with organic disease, or metcriteria of any current/past Axis I disorder of DSM-IV (e.g. schizo-phrenia, bipolar disorder, anxiety disorders including social phobia,panic disorder, drug addiction or substance dependence), and anycontraindication to MRI (e.g. current pregnancy or breastfeeding,metallic implants).

The control group contained 9 females and 7 males with age of25–42 years (mean 33.7576.36 years) and they received 10–17 years(mean 15.8172.48 years) education. None of the controls metcriteria for any current or past Axis I disorder, using the StructuredClinical Interview for DSM Axis I and scored o7 on both the HDRSand HAMA. Control participants were excluded if they had anypositive psychiatric history or family psychiatric history and anycontraindication aforementioned to MRI.

The two groups showed no differences in sex, age or educationlevel (p40.05). Those individuals who were currently pregnant orbreastfeeding, or had physical limitations that prohibited themfrom undergoing an fMRI scan were also excluded. Clinical anddemographic features of the sample are presented in Table 1.

All participants gave informed consent and the study wasapproved by the Institutional Review Board of Shanghai MentalHealth Centre.

2.2. Data acquisition

Structural and functional images were acquired on a 3.0-T GeneralElectric Signa scanner, using a standard whole-head coil. High-resolution T1-weighted spoiled grass gradient recalled 3D MRIsequence was used with the following parameters: 146 coronalslices, 1 mm thickness, slice gap=0, TR=500 ms, TE=14 ms, "ipangle=151. Whole-brain functional images were acquired usingmulti-slice echo planar imaging (EPI) sequence using the followingparameters: 22 axial slices, TR=3 s, TE=30 ms, "ip angle=901,FOV=24 cm! 24 cm, resolution=3.75! 3.75! 5 mm3, slice gap=0.Resting state was de!ned as no prescribed cognitive tasks during anfMRI scan. Subjects were instructed to remain motionless, with eyesclosed but not to fall asleep. The total acquisition time of theresting state scan was 5 min. The image acquisition protocol gavegood coverage of most of the ventral frontal areas except fora portion of basal orbitofrontal cortex where signal loss due to

suscepitibility effects was notable. The data of two control subjectswere excluded for further processing due to acquistion errors,resulting in a total of 14 subjects in the healthy control group.

2.3. Data preprocessing

The !rst four time points of the resting state data were discardeddue to instability of the initial MRI signal, leaving 100 time pointsremaining for further processing. The data were !rst preprocessed(slice timing, realignment, spatially normalization and re-sampledat 3 mm3) using SPM8 (www.!l.ion.ucl.ac.uk/spm). Movement parameters were assessed for each participant, and participants wereexcluded if movement exceeded 3 mm or 31. Then, using the DataProcessing Assistant for resting-state fMRI (DPARSF) package(Chao-Gan and Yu-Feng, 2010), linear drift was removed and thedata were spatially smoothed using a 6 mm FWHM Gaussian kernel.A band-pass frequency !lter (0.01ofo0.08 Hz) was then applied toreduce physiological high frequency noise (Biswal et al., 1995). Thedata was then blind-deconvolved using the methods described indetail by Wu et al. (2013), which uses the region-speci!c haemodynamic response function (HRF) modeled from large amplitudeBOLD-spikes based on the assumption that such spikes representspontaneous neural events. As a powerful approach, GCA may beconfounded by varied HRF across the brain. Resting-state fMRI datain particular provides further challenges relative to task-basedfMRI, wherein there is a lack of explicit inputs for modeling signaldynamics and estimating the underlying neural signal (Friston et al.,2008; Wu et al., 2013). It has been argued that this makes theinterpretation of BOLD-level causal in"uences as being related toneural-level causal in"uences more dif!cult (Smith et al., 2011).The blind deconvolution approach addresses these problems bydeconvolving the hemodynamic response with the observed fMRIsignal, which allows the extraction of neuronal variable from thedata; thus bringing us closer to neural-level causal models. Inaccordance with Wu et al. (2013), We used a threshold of 1.2 standard deviations for determining an event occurrence, and 15 sfor maximum lag between a neural event and BOLD response. Headmotion parameters were included as a covariate when computingthe Granger-causal maps for each subject. Finally, the root meansquare values of the movement were also calculated in accordancewith Power et al. (2012), and compared between the two groups.The two groups did not differ in the degree of head motion (mean(SD) in patients=0.0165 (0.0059), controls=0.0196 (0.0082), t=1.18, p=0.25) ruling out a systematic confounding effect ofhead-motion in the observed results.

2.4. Granger causality analysis

Our primary hypothesis was to investigate the in"uence of the rAIon the rest of the brain. We therefore used the coordinatesobtained in the study by McGrath et al. (2013) that showed

Table 1 Clinical and demographic features.

Depressed patients (n=16) Healthy subjects (n=16) t p

Age (years) 34.4376.72 33.7576.36 0.09 0.77Gender (male/female) 7/9 7/9Education level (years) 15.6371.99 15.8172.48 "0.27 0.82Age of onset (years) 33.2576.98Duration of illness (weeks) 8.0174.1924-item HRSD score 30.8877.69 3.8171.05 32.26 0.0014-item HAMA score 5.6270.72 3.3171.25 8.18 0.00

Abbreviations: HDRS, Hamilton Depression Rating Scale; and HAMA, Hamilton Anxiety Scale.Two sample t-test.

3Alterations in effective connectivity anchored on the insula in major depressive disorder

Please cite this article as: Iwabuchi, S.J., et al., Alterations in effective connectivity anchored on the insula in major depressive disorder.European Neuropsychopharmacology (2014), http://dx.doi.org/10.1016/j.euroneuro.2014.08.005

differential response dependent on treatment type (30 24 "14).A 6 mm radius sphere centered on this coordinate was used as therAI seed region for the GCA. Bivariate !rst-order coef!cient-basedvoxelwise GCA was performed using the REST software (http://www.restfmri.net). Granger causality estimates the causal effect ofthe seed region on every other voxel in the brain (X to Y effect), aswell as the Y to X effect, the causal effect of every voxel in thebrain (Y) on the seed region (X). A positive coef!cient from X to Yindicates that activity in region X exerts a causal in"uence on theactivity region Y in the same direction (i.e. positive in"uence).Similarly, a negative coef!cient from X to Y suggests that theactivity of region X exerts an opposing directional in"uence on theactivity of region Y (i.e. negative in"uence). Using this approach,we are able to build a Granger-causal model based on the temporalelements of regional BOLD activity. The GCA maps were enteredinto a one-sample t test at an uncorrected threshold of po0.0005with a cluster extent of 100 voxels to identify large clusters ofpositive and negative coef!cients of the entire sample (patientsand controls). We then used these clusters as orthogonal masks forsmall volume correction (SVC) for subsequent group analyses usinga one-way analysis of variance (ANOVA) (FWE corrected po0.05). Toestablish that the whole-sample one-sample test is inherentlyindependent, in addition to being orthogonal to the two-samplecontrast, we repeated the one-sample t test covarying for group(patients and controls) which showed a very high degree ofsimilarity (results shown in Supplementary Table 1 and Figure 1).We have therefore based our contraints using the originalnon-covaried results for the group analyses. The mean GCAcoef!cients of the signi!cant clusters from the group analyses wereextracted and subjected to a one-sample t test to determine thedirection (positive or negative) of causal in"uence for each group.The signi!cant clusters (discussed in Section 3 – DMPFC andpulvinar) identi!ed from the group comparison analyses were thenused as secondary seed regions and entered into another GCA usingthe same method described above. This approach enabled us tocross-validate GCA results using a reverse inference approach asemployed in a previous study (Palaniyappan et al., 2013). Allstatistical analyses were performed in SPM8 and SPSS 21.0 (SPSSInc., Chicago, Illinois, USA) and used an ! level of po0.05.

3. Results

3.1. Right anterior insula seed

In the entire sample of both patients and healthy controls,the one-sample t test revealed four large clusters to bepositively in"uenced by the rAI (uncorrected po0.0005,k=100). The rAI was also negatively in"uenced by two largeclusters (uncorrected po0.0005, k=100). One-sample t testresults are presented in Table 2. Two-sample t testsrevealed a signi!cant difference in the causal out"ow fromthe rAI to left DMPFC ("24 20 58) showing a signi!cantpositive in"uence in controls, while patients exerted nosigni!cant connectivity. There was also a signi!cant differ-ence in the causal in"ow from the pulvinar and caudate("12 "25 19) to the rAI where controls showed a negativein"uence, while patients showed no signi!cant connectivity.Two-sample group comparison t tests are illustrated inTable 3.

3.2. DMPFC seed

The one-sample t test showed the DMPFC to exert a positivein"uence on the precuneus (uncorrected po0.0005,

k=100). Eight clusters showed positive causal in"ow tothe DMPFC and one cluster exerted negative in"uence onthe DMPFC. In the two-sample t test, signi!cant differenceswere found in the ACC, orbitofrontal, inferior frontal,putamen and insular regions. Controls exerted signi!cantpositive in"uence from bilateral ACC and orbitofrontal/insula regions to the DMPFC, while patients showeddecreased positive in"uence. The rAI insula seed from theinitial GCA analysis was also used as a small volumecorrection mask to determine the reverse in"uence (DMPFCto rAI in"uence), which showed signi!cant negative in"u-ence in controls but signi!cant positive in"uence inpatients.

3.3. Pulvinar seed

The one sample t test for the entire sample revealednegative causal out"ow from the left pulvinar to six clusters(uncorrected po0.0005, k=100). Four additional clustersshowed positive in"uence to the pulvinar. Group compar-isons revealed that in controls, the pulvinar exerted sig-ni!cant negative in"uence on the striatum/insula region,while patients showed decreased negative in"uence. Sig-ni!cant positive in"uence was exerted toward the pulvinarfrom bilateral calcarine/cuneus, bilateral striatrum/insula,and cerebellum. However, patients showed weaker positivein"uence for all these regions.

A summary of the directed in"uences inferred in patientsand controls is presented in Figure 1.

4. Discussion

Using an effective connectivity approach for the !rst timeto study the interaction of anterior insula with rest of thebrain in depression, we have identi!ed an extended fronto-limbic circuitry with aberrant connectivity. This circuitryextends from sensory processing regions (cuneus), throughsubcortical centers (thalamus, caudate), to paralimbicsalience processing regions (insula, ACC, hippocampus andstriatum) and heteromodal prefrontal sites (DMPFC). Sev-eral regions from this extended processing system havebeen identi!ed to be dysfunctional in previous studies ofdepression (Drevets et al., 1992; Hamilton et al., 2011a;Kennedy et al., 2001; Sheline et al., 2010).

We present a presumed hierarchical processing modelbased on these !ndings: in healthy controls, a putative‘ascending’ system of in"uence can be traced from thevisual cortex (cuneus) to prefrontal regions. The visualcortex positively in"uences pulvinar of the thalamus, whichin turn (at resting state), negatively in"uences the insula. Infact, at rest, there exists a reciprocal feedback loopbetween the insula and the thalamus. The insula alsoreceives a putative top-down feedback with negative in"u-ence from the DMPFC. Together, these negative in"uencespreclude an insula-centered limbic overdrive in healthycontrols. Based on this model, in depression, it appearsthat both the putative bottom-up and top-down regulatoryloops centered on the insula are defective. The thalamus–insula loop is still notable but weakened, while the DMPFCshows an abnormal positive in"uence on the insula. Inaddition, we also noted a reduction in the cuneus to

S.J. Iwabuchi et al.4

Please cite this article as: Iwabuchi, S.J., et al., Alterations in effective connectivity anchored on the insula in major depressive disorder.European Neuropsychopharmacology (2014), http://dx.doi.org/10.1016/j.euroneuro.2014.08.005

thalamic in"ow in depression. Taken together, the insula-centered limbic system appears to ‘disintegrate’ in depres-sion. Although speculative, based on the assumption thatthe prefrontal cortex executes top-down cognitive control(Miller and Cohen, 2001), this model brings together thedisparate observations concerning the neural basis ofdepression and offers a framework that can be furthertested experimentally.

In our sample of treatment-naïve patients with MDD, wesaw a general attenuation in connectivity throughout thecircuitry (irrespective of positive or negative effect), with anumber of regions showing no signi!cant connectivity at allin patients. Of particular interest is the observation that theDMPFC has a positive in"uence on the rAI in MDD, which may

be crucial in creating a dynamic imbalance within theextended corticolimbic circuitry. This !ts well with thenotion that a primary corticolimbic imbalance leads toinef!cient emotional regulation in MDD (Mayberg, 1997;Mayberg et al., 1999). What is quite prominent from thecurrent results is the lack of bidirectional communicationbetween regions, which suggests that the ‘resting brain’ indepression displays a signi!cantly aberrant connectivitypro!le.

Our data are also strikingly similar to that of a recentstudy by Guo et al. (2013) who used functional (rather thaneffective) connectivity analysis and demonstrated thegreatest degree of dysconnectivity in depression involvedthe left superior frontal gyrus (SFG), right putamen, and

Table 2 One sample t-test (po0.0005, k=100) of all subjects for GCA for directional in"uence to and from the rightanterior insula (McGrath's region – MNI coordinates: 30, 24, "14), left DMPFC ("24 20 58) and pulvinar ("12 "25 19) seeds.Regions listed are at peak maxima of clusters.

Regions MNI coordinates t value Cluster size (voxels)

x y z

Positive causal out!ow from right anterior insula (X to Y)Right thalamus 3 "19 10 9.65 3635Right dorsomedial prefrontal 3 35 55 7.28 399Precuneus 0 "76 49 7.2 513Right anterior cingulate cortex 12 38 22 5.91 224

Negative causal in!ow to right anterior insula from rest of brain (Y to X)Right hippocampus 36 "31 "5 11.97 1175Precuneus 6 "73 55 5.56 220

Positive causal out!ow from left DMPFC (X to Y)Precuneus 3 "49 70 8.18 464

Positive causal in!ow to left DMPFC from rest of brain (Y to X)Right orbitofrontal 30 26 "11 7.97 729Right middle occipital 36 "85 16 7.47 152Left hippocampus "18 "4 "11 6.9 663Left superior temporal "48 "28 19 6.86 196Right cerebellum 33 "46 "23 6.29 123Right anterior cingulate cortex 15 44 13 6.17 119Left anterior cingulate cortex "9 23 22 6.09 201Left precentral "48 5 28 5.63 164

Negative causal in!ow to left DMPFC from rest of brain (Y to X)Precuneus "3 "46 64 6.77 319

Negative causal out!ow from left pulvinar (X to Y)Right thalamus 12 "19 1 9.51 2332Left insula "30 "25 16 8.89 1233Left middle occipital "48 "76 10 6.4 378Left calcarine "12 "73 7 5.59 155Right orbitofrontal 12 53 "5 5.51 100Left inferior parietal "33 "34 43 5.51 203

Positive causal in!ow to left pulvinar from rest of brain (Y to X)Right rolandic operculum 36 "19 19 10.93 1474Left insula "30 "25 16 10.37 3339Left orbitofrontal "33 50 "14 7.32 183Left calcarine "12 "64 10 6.07 333

Abbreviations: DMPFC, dorsomedial prefrontal cortex; GCA, Granger causality analysis; and MNI, Montreal Neurological Institute.

5Alterations in effective connectivity anchored on the insula in major depressive disorder

Please cite this article as: Iwabuchi, S.J., et al., Alterations in effective connectivity anchored on the insula in major depressive disorder.European Neuropsychopharmacology (2014), http://dx.doi.org/10.1016/j.euroneuro.2014.08.005

right insula. Speci!cally, there was abnormal connectivityfrom the SFG to the insula, and from the insula to theputamen. In particular, the greatest overlap of the changesbetween the SFG and putamen with the insula was localizedto the right insula region, suggesting this region to be a keystructure in the disrupted connectivity in depression. To addfurther support, the rAI has also been revealed to behavephysiologically differently depending on a patient's responseto antidepressants or CBT, where hyperactivity predictedresponse to antidepressants and hypoactivity to CBT(McGrath et al., 2013). With this in mind, the current studyadds to the mounting evidence for the rAI as a core regionin the pathophysiology of MDD, and encourages future workto incorporate the rAI in network models of affectivedisorders.

The connectivity of the thalamic circuitry was signi!-cantly reduced in patients with depression compared to

controls. Given its widespread connections with the cortexand limbic regions, including the insula, orbitofrontalcortex, amygdala and ACC (Mufson and Mesulam, 1984),disruption of this region will consequently impact a dis-tributed circuitry as observed in the current study. In arecent meta-analysis, the pulvinar in particular was shownto have abnormally high baseline activity in MDD (Hamiltonet al., 2012). It is thought that the pulvinar plays a vital rolein emotional attention and awareness (Pessoa and Adolphs,2010), which may suggest that hyperactivity of this regionleads to an excessively heightened sense of awareness andattention to emotional information. In fact, an fMRI studyreported that the pulvinar activates exclusively to detectedaffective stimuli (Padmala et al., 2010), so it is possible thatsuch a response is more sensitive in patients with MDD.Moreover, as suggested by Hamilton et al. (2012), thethalamic–insular disconnectivity could lead to a failure of

Table 3 Two sample t-test (po0.05, peak level FWE corrected using small volume correction from one sample t-test) ofdifference between patients and controls for causal in"uence to and from right anterior insula, left DMPFC and left pulvinarseeds (AAL regions are included in corrected cluster).

Regions MNIcoordinates

Mean (SD) pathcoef"cients incontrolsa

Mean (SD) pathcoef"cients inpatients

tvalue

Clustersize(voxels)

x y z

Causal out!ow from right anterior insula (X to Y) to:Left middle frontal and superior frontal (DMPFC)

(control4depression)"24 20 58 1.6 (0.71) 0.32 (0.69) 4.34 44

Causal in!ow to right anterior insula (Y to X) from:Bilateral thalamus including pulvinar, left caudate

nucleus (depression4control)"12 "25 19 "2.23 (1.33) "0.54 (1.34) 5.18 98

Causal out!ow from left DMPFC (X to Y) to:Right orbitofrontal and insula (depression4control) 33 26 "14 "0.78 (0.82) 0.9 (1.59)b 4.11 26Causal in!ow to left DMPFC (Y to X) from:Left ACC and superior frontal (control4depression) "6 32 19 1.78 (0.92) 0.45 (1.1) 4.01 70Right ACC and orbitofrontal (control4depression) 9 38 7 1.78 (1.27) 0.49 (0.77)b 3.74 86Right orbitofrontal, insula, putamen, inferior

frontal (control4depression)33 23 "14 1.9 (0.87) 0.49 (0.81)b 4.79 133

Causal in!ow from left pulvinar (X to Y) to:Left putamen, insula, globus pallidus, caudate

nucleus, orbitofrontal, amygdala, inferiorfrontal, hippocampus (depression4control)

"24 17 "5 "2.15 (0.7) "0.59 (0.9)b 5.66 449

Causal in!ow to left pulvinar (Y to X) from:Bilateral calcarine sulcus, left lingual gyrus,

bilateral cuneus (control4depression)"9 "64 19 1.74 (0.67) 0.36 (0.9) 4.12 211

Left putamen, insula, globus pallidus, caudatenucleus, amygdala, hippocampus, orbitofrontal(control4depression)

"24 20 "5 2.36 (0.82) 0.69 (0.95)b 5.29 596

Right putamen, caudate nucleus, globus pallidus,amygdala, hippocampus, orbitofrontal, insula(control4depression)

27 14 1 2.33 (1.08) 0.58 (1)b 5.41 337

Cerebellum 6, Crus 1, 4/5, right inferior occipital,bilateral fusiform (control4depression)

"24 "43 "29 2.15 (0.74) 0.58 (0.94)b 5.11 600

Abbreviations: AAL, Automated Anatomical Labeling; ACC, anterior cingulate cortex; FWE, Family-wise Error; DMPFC, dorsomedialprefrontal cortex; MNI, Montreal Neurological Institute; and SD, standard deviation.

aSigni!cantly different from zero at po0.005.bSigni!cantly different from zero at po0.05.

S.J. Iwabuchi et al.6

Please cite this article as: Iwabuchi, S.J., et al., Alterations in effective connectivity anchored on the insula in major depressive disorder.European Neuropsychopharmacology (2014), http://dx.doi.org/10.1016/j.euroneuro.2014.08.005

successful sensory information transfer up the cortical–striatal–pallidal–thalamic circuit to dorsal frontal corticalregions, owing to low dopamine activity in the striatum(Bowden et al., 1997; Meyer et al., 2001).

In the current study, we had a limited sample size tostudy the relationship between various symptom clusters indepression and the dysconnectivity within the hierarchicalsalience processing system. Nevertheless, one can speculatethat the dysconnectivity involving visual salience processingcan be related to symptoms such as anhedonia, while thehyperconnectivity between the DMPFC, a key node in theself-processing system and the salience system can berelated to self-centered ruminations in the depressed state.

Though speculative at present, our data leads us toconsider the pathway between the DMPFC and the fronto-insular region as a putative network target that can bemodulated for therapeutic bene!t in depression. FromDMPFC to frontoinsular cortex, controls display a negativein"uence, while patients contrastingly exhibited a signi!-cant positive in"uence. The work by Sheline et al. (2010)has demonstrated that the DMPFC, or so-called “dorsalnexus” region has increased functional connectivity to thethree major networks (cognitive control, affective, defaultmode) in depression. Our study of effective connectivitypinpoints the DMPFC to insula (a cognitive control region)link as the most affected in depression, while aberrantconnectivity centered on the insula affects various keynodes in a distributed sensory and limbic system throughthe thalamic pulvinar region. Therefore, targeting this area,for example through Transcranial Magnetic Stimulation,could restore the physiological state of the putative moodcircuitry. In fact, in a recent comprehensive review, theDMPFC along with other frontal regions have been offered asmore effective alternatives for TMS compared to theconventional DLPFC target, suggesting its greater involve-ment in emotion regulation in depression (Downar andDaskalakis, 2013). In support of this, Salomons et al.(2013) have demonstrated the effective use of the DMPFCas a TMS target. Better outcomes were associated witha TMS-induced decrease in connectivity between theDMPFC, bilateral insula and the parahippocampal gyrus/amygdala and connectivity increase between the DMPFC

and bilateral thalamus. Further exploration of these alter-native targets for TMS treatment would be extremelyvaluable in achieving targeted neuro-modulation of keybrain networks, and thus contribute to its wider use as atreatment option for depression.

One limitation of our study is that we have based ournetwork around the rAI seed with speci!c focus of connec-tions to and from this region. Therefore, we have notincluded consistently implicated regions such as the ventralACC (Hamilton et al., 2011a) in the circuitry. We anticipateabnormalities in network-level interactions between insulaand ventral ACC in MDD that requires further clari!cation infuture work. In terms of methodology, it has been notedthat variation in hemodynamic delay between brain regionsmay confound the identi!cation of reliable causal in"uences(Smith et al., 2011), as was demonstrated using simulatedfMRI data. However, Schippers et al. (2011) demonstratedusing real fMRI data, GCA can identify causal in"uences witha high level of accuracy. Furthermore, we have identi!edvery similar circuitry to that found in the study by Hamiltonet al. (2011a), providing support for the existence of theseputative causal connections. In addition, we employed ablind deconvolution procedure to our data in order toovercome some of the issues relating to inhomogeneity inhemodynamic processes across the whole brain and toenable neuronal-level interpretation of fMRI based causalmodels. We also note that the differences observed herecan re"ect the state of low mood and distress in patients,rather than the diagnosis of depressive disorder per se. Thisis a limitation inherent to the observational nature of thestudy that can be addressed by studying effective connec-tivity changes using mood induction paradigms in thefuture. Lastly, we acknowledge that our sample size islimited in the current study. However, we emphasize theimportance of our sample as !rst-episode, drug-naïve; thuseliminating the confounding medication effects.

5. Conclusion

In summary, the present study highlights a key role played bythe insula in the large-scale network-level abnormalities seen in

Figure 1 Simpli!ed schematic of effective connectivity in controls and patients with MDD. Solid red lines indicate signi!cantpositive connectivity, solid blue lines represent signi!cant negative connectivity, and dashed lines indicate signi!cantly reducedconnectivity in patients compared to controls. Missing arrows in MDD signify no signi!cant connectivity in either direction. (Forinterpretation of the references to color in this !gure legend, the reader is referred to the web version of this article.)

7Alterations in effective connectivity anchored on the insula in major depressive disorder

Please cite this article as: Iwabuchi, S.J., et al., Alterations in effective connectivity anchored on the insula in major depressive disorder.European Neuropsychopharmacology (2014), http://dx.doi.org/10.1016/j.euroneuro.2014.08.005

depression. By demonstrating the importance of both sensoryand prefrontal in"uences on limbic circuits in depression, ourresults shift the focus from spatially restricted, sparsely con-nected brain network models to an abnormality in an extended(presumably hierarchical) insula-centered processing de!cit indepression. This provides a novel hypothesis for experimentalveri!cation and raises the question of translational utilitythrough targeted brain network modulation.

Role of funding source

Funding for this study was provided by National Natural ScienceFoundation of China (91232719), the “12th Five-year Plan” ofNational Key Technologies R&D Program (2012BAI01B04), NationalKey Clinical Disciplines at Shanghai Mental Health Center (OMA-MH2011-873), Grants YG2012MS11, STCSM134119a6200, GWHW201208.These grants had no further role in study design; in the collection,analysis and interpretation of data; in the writing of the report; andin the decision to submit the paper for publication.

Contributors

Author SJI undertook data analyses and wrote the !rst draft of themanuscript. Author YRF designed and supervised the study. AuthorKJ supervised the study. Author EBL prepared the data for proces-sing. Author PFL supervised parts of the data analysis and con-tributed to interpretation of results. Author DHP collected all of thedata and analyzed the clinical characteristics. Author LP developedthe hypothesis, supervised the data analysis, interpreted the resultsand helped draft the manuscript. All authors contributed to andhave approved the !nal manuscript.

Con!ict of interest

SJI, YF, KJ and EBL report no potential con"icts of interest inrelation to this work.

PFL has received institutional grant support from the MedicalResearch Council (G0601442 and MR/J01186X/1) and the Dr. HadwenTrust (PI: Peter Morriss); and receives book royalties from the RoyalCollege of Psychiatrists. DHP is supported by the Science Fund ofShanghai Jiao Tong University (YG2012MS11), Fund of Science andTechnology Commission of Shanghai Municipality (134119a6200), Over-seas Talent Project of Shanghai Health Bureau (GWHW201208). LP issupported by a Wellcome Training Fellowship (WT/Z/11/096002).

Acknowledgments

Dr. Peng and Dr. Palaniyappan contributed equally to this work.

Appendix A. Supporting information

Supplementary data associated with this article can befound in the online version at http://dx.doi.org/10.1016/j.euroneuro.2014.08.005.

References

Biswal, B., Zerrin Yetkin, F., Haughton, V.M., Hyde, J.S., 1995.Functional connectivity in the motor cortex of resting humanbrain using echo-planar mri. Magn. Reson. Med. 34, 537–541.http://dx.doi.org/10.1002/mrm.1910340409.

Bowden, C., Cheetham, S.C., Lowther, S., Katona, C.L., Crompton,M.R., Horton, R.W., 1997. Reduced dopamine turnover in thebasal ganglia of depressed suicides. Brain Res. 769, 135–140.

Busatto, G.F., 2013. Structural and functional neuroimaging studiesin major depressive disorder with psychotic features: a criticalreview. Schizophr. Bull. 39, 776–786. http://dx.doi.org/10.1093/schbul/sbt054.

Cavanagh, J.T.O., Carson, A.J., Sharpe, M., Lawrie, S.M., 2003.Psychological autopsy studies of suicide: a systematic review.Psychol. Med. 33, 395–405. http://dx.doi.org/10.1017/S0033291702006943.

Chao-Gan, Y., Yu-Feng, Z., 2010. DPARSF: a MATLAB toolbox for“Pipeline” data analysis of resting-state fMRI. Front. Syst.Neurosci. 4, 13. http://dx.doi.org/10.3389/fnsys.2010.00013.

Delaveau, P., Jabourian, M., Lemogne, C., Guionnet, S.,Bergouignan, L., Fossati, P., 2011. Brain effects of antidepres-sants in major depression: a meta-analysis of emotional proces-sing studies. J. Affect. Disord. 130, 66–74. http://dx.doi.org/10.1016/j.jad.2010.09.032.

Downar, J., Daskalakis, Z.J., 2013. New targets for rTMS indepression: a review of convergent evidence. Brain Stimul.:Basic Transl. Clin. Res. Neuromodul. 6, 231–240. http://dx.doi.org/10.1016/j.brs.2012.08.006.

Drevets, W.C., Videen, T.O., Price, J.L., Preskorn, S.H., Carmi-chael, S.T., Raichle, M.E., 1992. A functional anatomical studyof unipolar depression. J. Neurosci. 12, 3628–3641.

Fitzgerald, P.B., Laird, A.R., Maller, J., Daskalakis, Z.J., 2008.A meta-analytic study of changes in brain activation in depres-sion. Hum. Brain Mapp. 29, 683–695. http://dx.doi.org/10.1002/hbm.20426.

Friston, K.J., Trujillo-Barreto, N., Daunizeau, J., 2008. DEM: a varia-tional treatment of dynamic systems. NeuroImage 41, 849–885.http://dx.doi.org/10.1016/j.neuroimage.2008.02.054.

Guo, S., Yu, Y., Zhang, J., Feng, J., 2013. A reversal coarse-grainedanalysis with application to an altered functional circuit indepression. Brain Behav. 3, 637–648. http://dx.doi.org/10.1002/brb3.173.

Hamilton, J.P., Chen, G., Thomason, M.E., Schwartz, M.E., Gotlib,I.H., 2011a. Investigating neural primacy in major depressivedisorder: multivariate Granger causality analysis of resting-statefMRI time-series data. Mol. Psychiatry 16, 763–772. http://dx.doi.org/10.1038/mp.2010.46.

Hamilton, J.P., Etkin, A., Furman, D.J., Lemus, M.G., Johnson, R.F.,Gotlib, I.H., 2012. Functional neuroimaging of major depressivedisorder: a meta-analysis and new integration of baselineactivation and neural response data. Am. J. Psychiatry 169,693–703. http://dx.doi.org/10.1176/appi.ajp.2012.11071105.

Hamilton, J.P., Furman, D.J., Chang, C., Thomason, M.E., Dennis, E.,Gotlib, I.H., 2011b. Default-mode and task-positive network activityin major depressive disorder: implications for adaptive and mala-daptive rumination. Biol. Psychiatry 70, 327–333. http://dx.doi.org/10.1016/j.biopsych.2011.02.003.

Kennedy, S.H., Evans, K.R., Krüger, S., Mayberg, H.S., Meyer, J.H.,McCann, S., Arifuzzman, A.I., Houle, S., Vaccarino, F.J., 2001.Changes in regional brain glucose metabolism measured withpositron emission tomography after paroxetine treatment ofmajor depression. Am. J. Psychiatry 158, 899–905.

Kito, S., Hasegawa, T., Koga, Y., 2011. Neuroanatomical correlatesof therapeutic ef!cacy of low-frequency right prefrontal tran-scranial magnetic stimulation in treatment-resistant depression.Psychiatry Clin. Neurosci. 65, 175–182. http://dx.doi.org/10.1111/j.1440-1819.2010.02183.x.

Liu, Z., Xu, C., Xu, Y., Wang, Y., Zhao, B., Lv, Y., Cao, X., Zhang, K.,Du, C., 2010. Decreased regional homogeneity in insula andcerebellum: a resting-state fMRI study in patients with majordepression and subjects at high risk for major depression.Psychiatry Res. 182, 211–215. http://dx.doi.org/10.1016/j.pscychresns.2010.03.004.

S.J. Iwabuchi et al.8

Please cite this article as: Iwabuchi, S.J., et al., Alterations in effective connectivity anchored on the insula in major depressive disorder.European Neuropsychopharmacology (2014), http://dx.doi.org/10.1016/j.euroneuro.2014.08.005

Mayberg, H.S., 1997. Limbic-cortical dysregulation: a proposed modelof depression. J. Neuropsychiatry Clin. Neurosci. 9, 471–481.

Mayberg, H.S., Liotti, M., Brannan, S.K., McGinnis, S., Mahurin,R.K., Jerabek, P.A., Silva, J.A., Tekell, J.L., Martin, C.C.,Lancaster, J.L., Fox, P.T., 1999. Reciprocal limbic-corticalfunction and negative mood: converging PET !ndings in depres-sion and normal sadness. Am. J. Psychiatry 156, 675–682.

McGrath, C.L., Kelley, M.E., Holtzheimer, P.E., Dunlop, B.W.,Craighead, W.E., Franco, A.R., Craddock, R.C., Mayberg, H.S.,2013. Toward a neuroimaging treatment selection biomarker formajor depressive disorder. JAMA Psychiatry 70, 821–829. http://dx.doi.org/10.1001/jamapsychiatry.2013.143.

Menon, V., Uddin, L.Q., 2010. Saliency, switching, attention andcontrol: a network model of insula function. Brain Struct. Funct.214, 655–667. http://dx.doi.org/10.1007/s00429-010-0262-0.

Meyer, J.H., Krüger, S., Wilson, A.A., Christensen, B.K., Goulding,V.S., Schaffer, A., Mini!e, C., Houle, S., Hussey, D., Kennedy,S.H., 2001. Lower dopamine transporter binding potential instriatum during depression. Neuroreport 12, 4121–4125.

Miller, E.K., Cohen, J.D., 2001. An Integrative theory of prefrontalcortex function. Annu. Rev. Neurosci. 24, 167–202. http://dx.doi.org/10.1146/annurev.neuro.24.1.167.

Mufson, E.J., Mesulam, M.M., 1984. Thalamic connections of theinsula in the rhesus monkey and comments on the paralimbicconnectivity of the medial pulvinar nucleus. J. Comp. Neurol.227, 109–120. http://dx.doi.org/10.1002/cne.902270112.

Murray, C.J., Lopez, A.D., 1997. Alternative projections of mortal-ity and disability by cause 1990–2020: global burden of diseasestudy. Lancet 349, 1498–1504. http://dx.doi.org/10.1016/S0140-6736(96)07492-2.

Padmala, S., Lim, S.-L., Pessoa, L., 2010. Pulvinar and affectivesigni!cance: responses track moment-to-moment stimulus visi-bility. Front. Hum. Neurosci. 4, 4. http://dx.doi.org/10.3389/fnhum.2010.00064.

Palaniyappan, L., Simmonite, M., White, T.P., Liddle, E.B., Liddle,P.F., 2013. Neural primacy of the salience processing system inschizophrenia. Neuron 79, 814–828. http://dx.doi.org/10.1016/j.neuron.2013.06.027.

Pessoa, L., Adolphs, R., 2010. Emotion processing and the amyg-dala: from a “low road” to “many roads” of evaluating biologicalsigni!cance. Nat. Rev. Neurosci. 11, 773–783. http://dx.doi.org/10.1038/nrn2920.

Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L., Petersen,S.E., 2012. Spurious but systematic correlations in functionalconnectivity MRI networks arise from subject motion. Neuro-Image 59, 2142–2154. http://dx.doi.org/10.1016/j.neuroimage.2011.10.018.

Salomons, TV, Dunlop, K, Kennedy, SH, Flint, A, Geraci, J,Giacobbe, P, et al., 2013. Resting-state cortico-thalamic-striatal connectivity predicts response to dorsomedial prefrontalrTMS in major depressive disorder. Neuropsychopharmacology39, 488–498.

Schippers, M.B., Renken, R., Keysers, C., 2011. The effect of intra-and inter-subject variability of hemodynamic responses on grouplevel Granger causality analyses. NeuroImage 57, 22–36. http://dx.doi.org/10.1016/j.neuroimage.2011.02.008.

Sheline, Y.I., Price, J.L., Yan, Z., Mintun, M.A., 2010. Resting-statefunctional MRI in depression unmasks increased connectivitybetween networks via the dorsal nexus. Proc. Natl. Acad. Sci.USA 107, 11020–11025. http://dx.doi.org/10.1073/pnas.1000446107.

Sliz, D, Hayley, S., 2012. Major Depressive disorder and alterationsin insular cortical activity: a review of current functionalmagnetic imaging research. Front. Hum. Neurosci. 6.

Smith, S.M., Miller, K.L., Salimi-Khorshidi, G., Webster, M., Beck-mann, C.F., Nichols, T.E., Ramsey, J.D., Woolrich, M.W., 2011.Network modelling methods for FMRI. NeuroImage 54, 875–891.http://dx.doi.org/10.1016/j.neuroimage.2010.08.063.

Sobocki, P., Jönsson, B., Angst, J., Rehnberg, C., 2006. Cost ofdepression in Europe. J. Ment. Health Policy Econ. 9, 87–98.

Sprengelmeyer, R., Steele, J.D., Mwangi, B., Kumar, P., Christmas, D.,Milders, M., Matthews, K., 2011. The insular cortex and theneuroanatomy of major depression. J. Affect. Disord. 133,120–127. http://dx.doi.org/10.1016/j.jad.2011.04.004.

Wu, G.-R., Liao, W., Stramaglia, S., Ding, J.-R., Chen, H., Mar-inazzo, D., 2013. A blind deconvolution approach to recovereffective connectivity brain networks from resting state fMRIdata. Med. Image Anal. 17, 365–374. http://dx.doi.org/10.1016/j.media.2013.01.003.

Yao, Z., Wang, L., Lu, Q., Liu, H., Teng, G., 2009. Regionalhomogeneity in depression and its relationship with separatedepressive symptom clusters: a resting-state fMRI study.J. Affect. Disord. 115, 430–438, http://dx.doi.org/10.1016/j.jad.2008.10.013.

9Alterations in effective connectivity anchored on the insula in major depressive disorder

Please cite this article as: Iwabuchi, S.J., et al., Alterations in effective connectivity anchored on the insula in major depressive disorder.European Neuropsychopharmacology (2014), http://dx.doi.org/10.1016/j.euroneuro.2014.08.005