ContentslistsavailableatScienceDirect
Neuroscience and Biobehavioral Reviews
j o u r n a l ho me p ag e :w w w . e l s e v i e r . c o m / l o c a t e / n e u b i o r e v
Review article
A systematic review of resting-state functional-MRI studies in
anorexia nervosa: Evidence for functional connectivity impairment in cognitive control and visuospatial and body-signal integration
Santino Gaudio (MD)
a,b,∗, Lyle Wiemerslage (PhD)
a, Samantha J. Brooks (PhD)
a,c, Helgi B. Schiöth (PhD)
aaDepartmentofNeuroscience,FunctionalPharmacology,UppsalaUniversity,BMC,Box593,Husargatan3,75124,Uppsala,Sweden
bCentreforIntegratedResearch(CIR),AreaofDiagnosticImaging,Universita“CampusBio-MedicodiRoma”,viaAlvarodelPortillo,200-00133-Rome, Italy
cPsychiatryNeuroimagingGroup(PNG),UniversityofCapeTown,DepartmentofPsychiatryandMentalHealth,OldGrooteSchuurHospital,AnzioRoad, Observatory,CapeTown,7925,SouthAfrica
a r t i c l e i n f o
Articlehistory:
Received14April2016
Receivedinrevisedform11August2016 Accepted30September2016
Availableonline8October2016
Keywords:
Eatingdisorders Anorexianervosa Neuroimaging Resting-state Defaultmodenetwork Insula
Cognitivecontrol Bodyimagedisturbances
a b s t r a c t
Thispapersystematicallyreviewstheliteraturepertainingtotheuseofresting-statefunctionalmag- neticresonanceimaging(rsfMRI)inanorexianervosa(AN),classifyingstudiesonthebasisofdifferent analysisapproaches.WefollowedPRISMAguidelines.Fifteenpaperswereincluded,investigatingatotal of294participantswithcurrentorpastANand285controls.Thestudiesusedseed-based,whole-brain independentcomponentanalysis(ICA),network-of-interestICAbasedandgraphanalysisapproaches.
Thestudiesshowedrelativelyconsistentoverlapinresults,yetlittleoverlapintheiranalyticalapproach and/ora-prioriassumptions.Functionalconnectivityalterationsweremainlyfoundinthecorticolim- biccircuitry,involvedincognitivecontrolandvisualandhomeostaticintegration.Someoverlapping findingswerefoundinbrainareasputativelyimportantinAN,suchastheinsula.Theseresultssuggest alteredfunctionalconnectivityinnetworks/areaslinkedtothemainsymptomdomainsofAN,suchas impairedcognitivecontrolandbodyimagedisturbances.Thesepreliminaryevidencessuggestthatmore targetedtreatmentsneedtobedevelopedthatfocusonthesetwosymptomdomains.Furtherstudies withmulti-approachanalysesandlongitudinaldesignsareneededtobetterunderstandthecomplexity ofAN.
©2016TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBYlicense (http://creativecommons.org/licenses/by/4.0/).
Contents
1. Introduction...579
2. Methods...580
2.1. Searchstrategyandinclusioncriteria...580
2.2. Qualityassessmentanddataabstraction...580
3. Results...580
3.1. Seed-basedapproachstudies(n=4)...580
3.2. WholebrainICAbasedstudies(n=2)...583
3.3. NetworkofinterestICAbasedapproachstudies(n=5)...584
3.4. Graphanalysisstudies(n=4)...584
4. Discussion...585
4.1. Mainfindings...585
4.2. ANsymptomdomainsandetiologicalandclinicalinsight:overlapinresultsbetweenthersfMRIstudies...586
5. Conclusions...587
AppendixB. PRISMAflowdiagram...587
References...588
∗ Correspondingauthorat:ViaPietroTacchini,2400197,Rome,Italy.
E-mailaddresses:santino.gaudio@gmail.com,santino.gaudio@libero.it(S.Gaudio).
http://dx.doi.org/10.1016/j.neubiorev.2016.09.032
0149-7634/©2016TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/).
1. Introduction
Anorexia nervosa(AN)is a severe mental disorderwiththe highestrateofmortalityinallpsychiatricdisorders(Arcelusetal., 2011)andittypicallyaffectsadolescentgirlsandyoungwomen.
ANischaracterizedbyself-inducedstarvationandsevereweight loss,anintensefearofweightgain orbecomingfat,adistorted bodyimageandfoodaversion(AmericanPsychiatricAssociation, 2013).Todate,theaetiologyofANisnotfullyunderstood.How- ever,agrowingconsensussuggestsamultifactorialorigin,inwhich neurobiologicalfactorscancontributetothevulnerability,onset, maintenanceandrelapseofAN(Kayeetal.,2013;Treasureetal., 2015;Zipfeletal.,2015).
Modern neuroimaging techniques have provided important insightforANpathophysiologyanditsneurobiologicalsubstrate.
Bothstructuralandfunctionalneuroimagingtechniqueswereused toexplorebrainabnormalitiesinANpatientsandtotrytoencode theneuralcircuitsinvolvedinAN.Structuralneuroimagingstud- ies,viavoxelbasedmorphometry(VBM),showedthatANpatients havedecreasesinglobalgrayandwhitematterandincreasesin cerebrospinalfluid(VandenEyndeetal.,2012;Titovaetal.,2013).
Furthermore,arecentmeta-analysisofVBMstudiesinANpatients (Titova etal., 2013)highlightedregionalgray matter decreases overlappingbetweenstudiesinspecificareas:lefthypothalamus, leftinferiorparietallobe, rightlentiformnucleusand rightcau- date−areas relatedtoappetiteand somatosensoryperception.
Ontheotherside,functional neuroimagingstudieshavemainly usedfunctionalmagneticresonance(fMRI)utilizingspecificstimuli relatedtomainANsymptoms,suchasfood-related,body-related andreward-relatedtasks,aswellasexecutivecontrol-relatedtasks (Kayeet al.,2013;GaudioandQuattrocchi,2012).Inparticular, alteredfunctionalactivationswerefoundincognitivecontrolareas in response to foodimages (Brooks et al., 2011, 2012), in the ventral anterior cingulate-striato-thalamic areas in response to cognitive-behaviouralflexibilitytasks(Zastrowetal.,2009),andin theposteriorparietalareasandprefrontalcortex-insulanetwork inresponse totasks relatedtotheperceptiveand theaffective componentsof body image distortion respectively (Gaudio and Quattrocchi,2012).Onthebasisofneuroimagingfindingsaswellas psychologicaldescriptions,acurrentneurobiologicalmodelofAN suggeststhatchildhoodtemperamentandpersonalitytraits(e.g.
anxiety,cognitiveinflexibility,obsessionalityandperfectionism) mightbeasignofneurobiologicalriskfactorsforthedevelopment ofAN,andthatalteredeatingpatternsinthosewithANcouldbe ameansofdecreasingnegativemoodbroughtaboutbyaltered interactions betweentheserotoninergic system(i.e.aversive or inhibitory)andthedopaminergic(i.e.reward)system(Kayeetal., 2013;Zipfeletal.,2015;Treasureetal.,2015).Particularlyconsid- eringneuroimagingfindings,thisneurobiologicalmodelsuggests thatANpatientsmayhaveanimbalanceininformationprocessing, linkedtoalterationsoftheventrallimbiccircuits(whichcomprises amygdala,anteriorinsula,anteriorventralstriatum,anteriorcin- gulatecortex,andtheorbito-frontalcortex),aswellasthedorsal executivecircuits(whichparticularlyincludesdorsalregionsofthe caudate,dorso-lateralprefrontalcortex,andparietalcortex).These twobraincircuitsareprimarilyimplicatedininhibitorydecision makingprocessesandreward-relatedbehavioursandtheiralter- ationmightsustainANsymptomatology(Kayeetal.,2013).Other neurobiologicallyinformedmodelsofANhavepointedouttherole ofanxiety,stress,andfear,thegratifyingnatureofANsymptoms, andtheconsequentshifttohabitualorcompulsivebehavioursas possiblekeyfactorsinthepersistenceofAN(Zipfeletal.,2015;
Treasureetal.,2015).
During the last decade, several research groups have used a stimulus-free fMRI approach, defined as resting-state fMRI (rsfMRI)(e.g.FoxandGreicius,2010).In thesestudies,thepar-
ticipantsarepositionedinthescannerinanawake-statewithout performing anyparticulartask. This methodallows for making temporal correlations between brain areas, based on sponta- neous task-independent fluctuations of the blood-oxygenation leveldependent(BOLD)signalinthebrain(Biswaletal.,2010;
Greiciusetal.,2009).RsfMRIconsidersthefunctionalconnectivity ofbrainareas,whereastraditional,task-dependentfMRIconcerns activityofparticularbrainareasinresponsetoastimulus.RsfMRI isasimpleprocedure,hasashortscantime−ontheorderofmin- utes,andthereproducibilityofresultsisrobustwithinthesame participant(Zuoetal.,2010).However,whilersfMRIisthoughtto reflectfundamentaltraitsinpersonalityorpsychologicalfunction (HarmelechandMalach,2013),somehavecriticizedtheultimate usefulnessofrsfMRIresults(MorcomandFletcher,2007).More- over,thedebateontheroleofunwantedthoughtsinrestingstate functional connectivity is still open (Kühn et al., 2014) and it couldbeamethodologicalconcerninpsychiatricdisorders,due tothefactthatseveralpsychiatricdisordershaveintenseforms ofunwantedthoughts(e.g.obsessivecompulsivedisorder(Julien etal.,2007)andruminationonweightandbodyshapeispresentin AN(AmericanPsychiatricAssociation,2013).Whilegeneralitiesof thebasicmethodologyinrsfMRIwouldbenefitfromfurtherstan- dardizationandoptimization(e.g.scanwhileeyesclosed/open,or awake/asleep),rsfMRIhasregardlessemergedasastapletoolin neuroimagingstudies−perhapsequalinutilitytotask-basedfMRI (FoxandGreicius,2010).
AnumberofstudieshaveusedrsfMRItoinvestigatefunctional connectivityin patientswitheating disorders.Several different approachescanbeusedinrsfMRIanalysis.Inparticular,themost commonarenetwork-basedandseed-basedapproaches(Foxand Raichle,2007;VanDenHeuvelandHulshoffPol,2010).Regarding thefirstapproach,independentcomponentanalysis(ICA)isthe mostcommonanalysistoinvestigateintrinsicneuralnetworks,and itdoesnotnecessarilyrequireanapriorihypothesis.ICAanalyses theentireBOLDsignalandisolatesdifferentindependentcompo- nentsputativelyreflectingseparateresting-statenetworks(RSNs) (FoxandRaichle,2007;VanDenHeuvelandHulshoffPol,2010).
RSNsaredefinedasthesetsofbrainareasthatshowstrongtempo- ralcoherenceintherestingbrain,andarethoughttorepresent specificframeworksofbrain function(Damoiseauxetal.,2006;
Smithetal.,2009).Anumber ofRSNshavebeenidentifiedand investigated,andamongthese,themoststudiednetwork isthe default-modenetwork(DMN)(eg.Raichleetal.,2001;Smithetal., 2009).TheDMNinvolvesareassuchasthemedialprefrontalcortex, theposteriorcingulate/precuneus,hippocampusandtheinferior parietal cortex (Raichleet al., 2001). It is alsoknownby alter- nativenomenclature, suchasthetask-negativenetwork (TNN), orthehippocampal-cortexmemorysystem(Vincentetal.,2008).
TheDMNis recruitedduring non-goalorientedcognitiveactiv- ity,introspectiverumination,low-levelarousal,homeostatic-and self-regulation, suchas during “day-dreaming,” and supporting autobiographical,internallyfocused,declarative,episodicmemory retrieval(Smithetal.,2009).TheDMNisdeactivatedantagonisti- callyandanti-correlatedwiththeexecutivecontrolnetwork,which is a medial-frontal systemincluding theanterior cingulateand para-cingulate cortex (Smith et al., 2009;Barkhof etal., 2014).
Theexecutivecontrolnetwork,ortask-positivenetwork,under- liesexecutivefunctioning,suchasworkingmemory,goal-oriented cognition,impulse-control,andemotionalprocessing(Smithetal., 2009;Barkhofet al.,2014).In additiontotheDMNand execu- tivecontrolnetwork,therearemanyothermajorcanonicalresting statenetworksthatarefrequentlyidentifiedintheliterature(for detailsseeSmithetal.,2009;Barkhofetal.,2014).Inparticular, thesenetworksincludethesensorimotornetwork(i.e.striataland parietalcortex),visualnetwork(i.e.occipitalcortex),cerebellum network (i.e. cerebellum),salience network (SN:frontalcortex,
anteriorcingulateandanteriorinsularcortexcircuitry),anddor- salattentionnetwork(DAN:insularcortexandposteriorparietal).
Theseed-basedapproachisbasedonanapriorihypothesisandit assessesfunctionalconnectivitybetweenapreselectedseedregion andotherbrainregions(VanDenHeuvelandHulshoffPol,2010).
Inbrief,afunctionalconnectivitymapisextractedfromthetempo- ralcorrelationsbetweentheregionofinterestandallotherbrain regions(e.g.Leeetal.,2014).
OtherrsfMRIapproachesincludegraphanalysisandeffective connectivity.Ingraphanalysis,thebrainisconsideredacomplex networkconsistingnodes(i.e.designatedbrainareasdistributed acrossthewholebrain)connectedbyedges(i.e.thefunctionalrela- tionshipbetweennodes),whichtheunderlyingactivityovertime dividingandorganizingbrainnetworkswithinthefieldofnodes (eg.RubinovandSporns,2010;Geisleretal.,2015).Graphanal- ysisassessestheglobalandlocalpropertiesofbrainregionsand analysesanumberofglobalandnodalmetrics,suchastheaverage pathlengthbetweenallpairsofnodes,andthesumoftheedges’
weightsthatconnectstoagivennode,calleddegreecentrality(DC).
Twoothercommontechniquesincludedynamiccausalmodelling (DCM)andGrangercausality,whichbothcalculateeffectivecon- nectivity(Friston,2011).Inbrief,theeffectiveconnectivityassesses thecausalanddynamicinfluenceofoneregiononanother,aswell asthedirectionalityofinformationflowbetweenbrainregions(e.g.
Kullmannetal.,2014).
Currently,theliteratureonrsfMRIstudiesinANseemstopoint toa lack ofconsistency in theanalytic approaches and, subse- quently,alsothefindings.Theaimofthispaper,therefore,isto systematicallyreview studiesusingrsfMRI inpatientssuffering fromANandexplorewhetherrsfMRIaddsusefulinsightinunder- standingof ANpathophysiology. Inpursuitof thisaim,wealso comparetheresultsandsubsequentinterpretationsfromthedif- ferentrsfMRIapproachestodetermineifthereisconsensusinthe brainareas/networksaffected.Lastly,weshalldiscusstherelevant themesandfutureconsiderationsregardingtheultimateuseful- nessofrsfMRIinANresearchandtreatment.
2. Methods
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines (Moher et al., 2009).Thestatementconsistsofachecklistofrecommendeditems tobereported(AppendixA:PRISMAchecklist)andafour-stepflow diagram(AppendixB:PRISMAflowdiagram).
2.1. Searchstrategyandinclusioncriteria
Databasesusedforthesearchwere:PubMed(frominceptionto May2016)andScopus(frominceptiontoMay2016).Wesearched usingthe terms: “anorexianervosa” or “eating disorders” AND
“fMRI”,“restingstate”,or“functionalconnectivity”.Thereference listsofexaminedfull-textpaperswerescrutinisedforadditional relevantpublications. Inaddition, expertcolleaguesin thefield werecontactedforsuggestiononfurtherstudiesnotconsidered inoursearch.
Tobeincludedinthereview,studieswererequiredto:1)be writtenin English,2)investigate a sampleof participantswith currentorrecoveredfromANincross-sectionalcase-controlorlon- gitudinaldesign,3)investigatebrainfunctionalityatrest.Studies usingtasksorstimuli(e.g.hungerstate)duringtheMRscanning sessions were excluded. Due to the limited number of peer- reviewedrsfMRIstudiesonAN,wedidnotconsiderconfounding factorssuchassampleinhomogeneity(e.g.ANsubtypes),presence ofpsychiatriccomorbidityorpharmacologicalhistory,whichmay limitsomeofthestudiesincluded.TheheterogeneityofrsfMRI
approachesandlackofsufficientnumberofpapersusingsimilar approachesorsimilarseedornetworksofinterestdonotallowus toperformameta-analysis.
2.2. Qualityassessmentanddataabstraction
Toreduceariskofbias,PRISMArecommendationsforsystem- aticliteratureanalysishave beenstrictly followed.Twoauthors (S.G.andL.W.)independentlyselectedpaperabstractsandtitles, andanalysedthefullpapersthatmettheinclusioncriteria,resolv- ingdisagreementsthroughconsensus.
Dataextractedfromeachstudywere:sampletype,studydesign, samplesize,resting-statefMRIapproaches,andselectedfindings.
3. Results
Fifteenresting-statefMRIstudiesofparticipantswithcurrentor pastANwereincludedinthereview(AppendixB:PRISMAflowdia- gram).Overall,thestudiesincludedatotalof294participantswith currentorpastANand285controlparticipants.Table1reports thesamplecharacteristicsofeachANstudy.Ofthefifteenstudies includedinthisreview,fourstudiesusedaseed-basedapproach andelevenadata-drivenapproach(Table2).Oftheelevendata- drivenapproachstudies,twostudiesadoptedawholebrainICA basedapproach,fivestudiesusedanetworkofinterestICAbased approach,andfourstudiesadoptedgraphanalysis(Table2).Inthe following,studiesthatappliedseed-based,wholebrainICAbased, networkofinterest ICAbased,andgraph analyseswillbesum- marized.Fig.1reportsthemainfindingsofthersfMRIstudiesin AN.
3.1. Seed-basedapproachstudies(n=4)
Theseed-basedapproachisthemostcommonmethodtoexam- inefunctionalconnectionsofanaprioriselectedbrainregionwith allotherregions(VanDen Heuveland HulshoffPol,2010).The regionofinterestcanbeselectedfromprevious,task-dependent fMRIresultsoritcanbeanaprioridefinedregion.FourrsfMRIstud- iesadoptedaseed-basedapproach(Leeetal.,2014;Favaroetal., 2013;Biezonskietal.,2015;Collantonietal.,2016)eachselect- ingdifferentseeds(Table2).Leeetal.(2014)enrolledasampleof adultANoutpatients(18AN,20BNand20controls)utilizingaclus- terwithinthedorsalanteriorcingulatecortex(dACC)astheseed region.ThisareawasselectedbasedontheirpreviousfMRIstudy ofANpatientsshowingalteredfunctionalactivationinresponse tohigh-calorie foodimages(Kim et al.,2012).The seedanaly- sisshowedthatANpatientshadastrongersynchronousactivity betweenthedACCandretrosplenialcortexandbetweenthedACC andprecuneuscomparedtocontrols.Furthermore,bothgroupsof eatingdisorderedpatients(AN+BN)demonstratedstrongersyn- chronousactivitybetweenthedACCandprecuneuscomparedto controls.Inasecondstudy,Favaroetal.(2013)scannedanadult subsampleoftheirparticipants,whichcomprised33ANpatients and30controls.Threebilateralseedswereselected:dorsolateral prefrontalcortex(DLPFC),ventromedialPFC(vmPFC),andventro- lateralPFC(vlPFC).Nosignificantdifferenceswerefoundbetween AN patientsand controls.However,when analyzingdifferences withintheANgrouponthebasisofgenotypesforfunctionalpoly- morphismsofthecatechol-O-methyltransferase(COMT)protein, functionalconnectivitydifferenceswerefoundintheprefrontal seedsbetweentheMet-andVal-carriersubsamples. Thethird study byBiezonski et al. (2015) examineda clinical sample of 16lateadolescentandadultANinpatientsand15 controls,and testedsevenbilateralthalamicseeds.Eachseedcorrespondedtoa regionofthethalamuswithpreferentialconnectionstoaspecific lobeorcorticalregion(Johansen-Bergetal.,2005)andfunctional
Table1
Samplecharacteristicsofresting-state-fMRIstudies.
Study Participants Age(years) BMI EDduration(years)
Mean (SD) Mean (SD) Mean (SD)
Amiantoetal.(2012) AN:n=12 20.00 (4.00) 16.27 (0.99) 0.91 (0.41)
BN:n=12 23.00 (5.00) 21.57 (2.38) – –
HC:n=10 24.00 (3.00) 21.35 (3.16) – –
Biezonskietal.(2015) AN:n=15 19.00 (0.50) 17.37 (0.29) 4.73 (0.83)
HC:n=16 20.31 (0.77) 21.57 (0.45) – –
Boehmetal.(2014)* AN:n=35 16.10 (2.56) 14.78 (1.26) 1.57 (2.25)
HC:n=35 16.16 (2.64) 20.81 (2.72) – –
Boehmetal.(2016) rec-AN:n=31 22.27 (3.08) 20.69 (1.62) 3.72 (2.66)
HC:n=31 21.73 (2.99) 21.37 (2.10) – –
Collantonietal.(2016) AN:n=35 25.40 (6.90) 15.80 (1.80) 6.70 (6.90)
HC:n=34 25.00 (6.20) – – – –
Cowdreyetal.(2014) rec-AN:n=16 23.06 (3.55) 21.33 (2.17) 3.50 (2.38)
HC:n=14 24.11 (2.85) 21.01 (1.56) – –
Ehrlichetal.(2015)* Geisleretal.(2015)* Lordetal.(2016)*
AN:n=35 16.10 (2.56) 14.78 (1.26) – –
HC:n=35 16.16 (2.64) 20.81 (2.72) – –
Favaroetal.(2012) AN:n=29 25.80 (6.90) 14.50 (2.30) 6.20 (6.90)
rec-AN:n=16 23.80 (4.80) 19.20 (1.00) 2.30 (1.70)
HC:n=26 26.70 (6.70) 21.80 (3.20) – –
Favaroetal.(2013) AN:n=33 26.90 (7.30) 15.80 (1.90) – –
HC:n=30 25.80 (6.70) – – – –
Gaudioetal.(2015) AN:n=16 15.80 (1.70) 16.20 (1.20) 0.33 (0.15)
HC:n=16 16.30 (1.40) 21.10 (1.90) – –
Kullmannetal.(2014) AN:n=12 23.30 (4.70) 15.50 (1.50) – –
HC:n=14 24.60 (2.90) 21.40 (1.50) – –
HCA:n=12 24.10 (3.20) 22.00 (1.90) – –
Leeetal.(2014) AN:n=18 25.20 (4.20) 16.00 (1.70) 3.80 (2.60)
BN:n=20 22.90 (3.90) 21.60 (2.30) 3.80 (4.70)
HC:n=20 23.30 (1.80) 19.90 (1.90) – –
Phillipouetal.,(2016) AN:n=26 22.81 (6.67) 16.63 (1.19) 6.42 (7.43)
HC:n=27 22.46 (3.16) 22.60 (3.53) – –
Note:BMI=Bodymassindex.ED=eatingdisorder.AN=anorexianervosa.BN=BulimiaNervosa.HC=healthycontrol.rec-AN=recoveredAN.HCA=healthycontrolathletes.
*thesefoursamplesarecomposedbythesameparticipants.
Fig1. Mainbrainareasimplicatedinresting-statefunctionalMRI.Diagramofthemainfindingsfromeachresting-statefMRIstudyofANpatientsclassifiedonthebasis ofthedifferentrestingstateapproaches.EachcirclerepresenttheregionswhereANpatientsexhibitedfunctionalconnectivityalterations:yellowreferstotheseedbased approachstudies(seedsarereported);greenreferstowhole-brainICA-basedstudies(alteredareaswithinthenetworksarereported);redreferstonetwork-of-interest ICA-basedstudies(alteredareaswithinthenetworksarereported);bluereferstographanalysisstudies(alteredareasarereported).LabelsarereportedinTable2.See Table2forspecificdetailsoftheareasimplicatedbyeachmethod.
connectivityanalyseswerefocusedonthalamicconnectivitywith thefrontallobes.AN patientshad greaterconnectivitybetween the centralmedial thalamus and thebilateral DLPFC and lower functional connectivitybetween theanterior thalamus and the leftanteriorPFCcompared tocontrols.Whilethealterations in thalamo-frontalconnectivitywerenotassociatedwithattention and visuospatialprocessing, theywere associated withimpair- mentsin performance ontasks assessing cognitivecontrol and
workingmemory.Mostrecently,Collantonietal.(2016)examined anadultsubsamplethatcomprised35ANpatientsand34controls.
Theyselectedthreedifferentseeds:superiorparietallobule,right inferiorfrontalgyrusandpre-supplementarymotorarea,consid- eringthemaspartofthedorsalattention,ventralattention,and pre-supplementarymotorareanetworksrespectively.Theauthors foundthatANpatientsshowedalteredfunctionalconnectivityin therightinferiorfrontalgyruscomparedtocontrols.
Table2
Descriptionofresting-statefMRIstudiesinAN.
Label Study Brainregionsand/ornetworks analysed
Mainfindings Mainclinicalinterpretation
Seed-basedstudies
A Leeetal.(2014) Seed:dorsalanteriorcingulatecortex (dACC)
ANpatientsshowedstronger synchronousactivitybetweenthe dACCandretrosplenialcortexand betweenthedACCandprecuneus.ED groupdemonstratedstronger synchronousactivitybetweenthe dACCandprecuneus.
ThealtereddACC-precuneussynchrony mightbeassociatedwiththe disorder-specificruminationoneating, weightandbodyshapeinpatientswith eatingdisorders.
B Favaroetal.(2013) Seed:dorsolateral,ventrolateraland ventromedialprefrontalcortex
Analysesdidnotrevealanysignificant differencesbetweenANpatientsand healthywomen.WithinANsample,FC differencesweredetectedbetween patientswithdifferentvariantsforthe COMTprotein.
ANpatientswithaparticularvariantfor theCOMTproteinhadincreasedprefrontal cortexFCcomparedtocarriersofa differentvariant.
C Biezonskietal.(2015) Seed:sevenbilateralthalamicseeds.
Onlyconnectivitywiththefrontal lobeswasinvestigated
ANpatientsshowedgreater connectivitybetweenthecentral medialthalamusandthebilateral dorsolateralprefrontalcortexand lowerFCbetweentheanterior thalamusandtheleftanterior prefrontalcortex.
Alterationsinthalamo-frontalcircuitsmay havearoleinmediatingaspectsof cognitivedysfunctioninANpatients.
D Collantonietal.(2016) Seed:superiorparietallobule,right inferiorfrontalgyrusand pre-supplementarymotorarea
ANpatientsshowedalteredFC(inboth thepositiveandnegativeconnectivity) oftherightinferiorfrontalgyrus, consideredaspartoftheventral attentionnetwork
Impairedventralattentionnetwork connectivity(i.e.therightinferiorfrontal gyrusalteration)mayaffectcognitive controlprocessesandexogenousstimuli filtering.
WholebrainICAbasedStudies
E Cowdreyetal.(2014) Twelveresting-statenetworks identifiedandinvestigated
Increasedtemporalcoherence betweentheDMNandtheprecuneus andthedorsolateralprefrontal cortex/inferiorfrontalgyrusinsubjects recoveredfromAN.
Thefindingsarecompatiblewiththecore symptomsofANincludingruminative preoccupationoneating,weightand shape,excessiveplanningandimpaired cognitiveflexibility.
F Gaudioetal.(2015) Eightresting-statenetworksidentified andinvestigated
Decreasedtemporalcorrelation betweentheexecutivecontrol networkFCmapsandtheanterior cingulatecortex(ACC)inANpatients.
ThedecreasedFCbetweenexecutive controlnetworkandtheACCcouldexplain theimpairedcognitiveflexibilityin relationtobodyimageandappetiteinAN patients.
NetworkofinterestICAbasedstudies
G Favaroetal.(2012) Networksofinterest:medial,lateral, andventralvisualnetworksand somatosensory
DecreasedFCintheleft occipitotemporal
junctionwithintheventralvisual networkandincreasedFCintheleft superiorparietalcortexFCwithinthe somatosensorynetworkinANpatients.
DecreasedFCintherightmiddle frontalgyruswithintheventralvisual networkinrecoveredANsubjects.
Thefindingsmayexplainthefailureofthe integrationprocessbetweenvisualand somatosensoryperceptualinformation thatsustainsbodyimagedisturbance.
H Amiantoetal.(2013) Networksofinterest:cerebellar Alterationswithinthecerebellar networkbetweenEDpatients comparedtocontrols(e.g.,increased connectivitywithinsulaeandvermis decreasedconnectivitywithparietal lobe).Increasedconnectivitywiththe insulaeinANcomparedtoBN, increasedconnectivitywithanterior cingulatecortexinBNcomparedtoAN
Thefindingssupporttheroleofthe cerebellumasakeyareaofintegrationof differentfunctionsimportantinED psychopathology.
I Boehmetal.(2014) Networksofinterest:fronto-parietal, defaultmode,salience,visualand sensory-motor
IncreasedFCbetweentheangular gyrusandthefronto-parietalnetwork andbetweentheanteriorinsulaand thedefaultmodenetworkinpatients withAN.
IncreasedFCwithinthefronto-parietal networkmayberelatedtoexcessive cognitivecontrolinANpatients.The anteriorinsula/DMNalterationsmaybe relatedtoruminationsaboutfoodand bodilyappearance.
J Boehmetal.(2016) Networksofinterest:fronto-parietal, defaultmode,salience
RecoveredANsubjectsshowed reducedFCbetweenthedorsolateral prefrontalcortexandthe
fronto-parietalnetwork.Additionala priorianalysisfoundincreasedFC betweenangulargyrusandFPNin recoveredsubjects.Nodifferences werefoundintheDMN.
SomealteredFCpatternsfoundinAN patientsarestillpresentafterlong-term ANrecovery.Inparticular,DMNalterations seemtonormalize,whilefronto-parietal networkalterations,alsorelatedto cognitivecontrolfunctions,seemto persist.Theauthorssuggestedthis interpretationsonthebasisoftheir previousstudy[i.e.Boehmetal.(2014) (labelI)].
Table2(Continued)
Label Study Brainregionsand/ornetworks analysed
Mainfindings Mainclinicalinterpretation
K Phillipouetal.(2016) Networksofinterest:ROIsofdefault mode,sensory-motorandvisual networks.aROItoROIapproachwas used.(seethepaperfordetails)
DecreasedFCwerefoundbetweenthe sensory-motorandvisualnetworksin ANpatientscomparedtocontrols.In particular,ANpatientsshowed reducedFCbetweenprimary somatosensory,andsecondaryvisual andassociativevisualareas;and betweenprimarymotor,and secondaryvisualandassociativevisual areas.
NodifferenceswerefoundintheDMN.
ReducedFCbetweensensorimotorand visualnetworksmaysuggestanaltered visuospatialprocessinginANpatients, relatedtobodyimagedisturbances.
Graphanalysisstudies
L Kullmannetal.(2014) degreecentralityandeffective connectivity
ReducedFCoftheinferiorfrontalgyrus (IFG)bilaterallyandalteredeffective connectivity(i.e.fromtherightIFGto themiddlecingulatecortex,fromthe bilateralorbitofrontalgyrustothe rightIFGandfromthebilateralinsula totheleftIFG)inANpatients
Reducedconnectivitywithinthecognitive controlsystemofthebrainandincreased connectivitywithinregionsimportantfor salienceprocessing.
M Ehrlichetal.(2015) Network-basedstatisticapproachand regionalhomogeneity
Asubnetworkofconnectionshad decreasedconnectivityinANpatients, areasincludedtheamygdala, thalamus,fusiformgyrus,putamenand theposteriorinsula.
Thefindingsmightreflectchangesinthe propagationofalteringsensationsto urgenthomeostaticimbalancesand pain-processes,whichareknowntobe severelydisturbedinANandmightexplain thestrikingdiscrepancybetweenpatients’
actualandperceivedinternalbodystate.
N Geisleretal.(2015) 7globaland7nodalgraphmetrics(see thepaperfordetails)
Decreasedconnectivitystrengthand increasedpathlengthintheposterior insulaandthalamusinANpatients.
Reducedlocalnetworkefficiencyinthe thalamusandposteriorinsulamayreflecta mechanismexplainingtheimpaired integrationofvisuospatialand homeostaticsignalsinANpatients.
O Lordetal.(2016) Globalandlocalnetworkproperties, includingnetwork-basedstatistics, comparingtwocommonparcellation approaches(i.e.anatomicaland literaturebasedanalyses)(seethe paperfordetails)
FCalterationswerefoundinAN patientscomparedtocontrols includingtheinsulaandthalamus.
Thesedifferenceswereconsistent acrosstheparcellationapproaches.
ReducedFCinasubnetworkincludingthe insulaand
thalamus(asfoundinthisstudy independentoftypeofatlas)couldbe relatedtoanalteredprocessingofsignals suchasbodysize,hungerandpain.
Note:Dorsalanteriorcingulatecortex=dACC.Anorexianervosa=AN.Eatingdisorder=ED.Functionalconnectivity=FC.Defaultmodenetwork=DMN.Anteriorcingulate cortex=ACC.Bulimianervosa=BN.Inferiorfrontalgyrus=IFG.
Insum,giventhattheseed-basedrsfMRIstudiesusedifferent brain region seeds, comparisons between the studies are lim- ited.Nevertheless,functionalconnectivityalterationswerefound betweendACCandtheprecuneus,inthalamus-frontalcircuits,and intherightinferiorfrontalgyrus−regionsassociatedwithcogni- tivecontrolprocessesandruminationonweightandbodyshape.
3.2. WholebrainICAbasedstudies(n=2)
Theapproachmostutilisedtoexaminewhole-brainconnectiv- itypatternsisICA(Foxand Raichle,2007;Van DenHeuveland HulshoffPol, 2010).This approach doesnot requireana priori seedregion,hence,isdata-driven,andallowsgeneralpatternsof connectivitybetweenseveralbrainregionstoemergebased on temporalsynchronicity(VanDenHeuvelandHulshoffPol,2010).
ICAthencomparesthesepatternsbetweengroups.Sinceitsdevel- opment,rsfMRIstudies viaICAhave robustlydescribed several differentandwidelyacceptedresting-statenetworks(Damoiseaux etal.,2006andSmithetal.,2009),whichhaveledtothenetwork- of-interestbasedapproach(seeSectionbelow).
TworsfMRIstudiesinvestigatedwholebrainfunctionalconnec- tivityviaICA,focusingonthemostwidelyacceptedresting-state networks(Cowdreyetal.,2014;Gaudioetal.,2015)(Table2).The firststudywasconductedonadultparticipantsrecoveredfromAN (16pastAN,15controls)(Cowdreyetal.,2014).TheANpatientshad recoveredbyatleastoneyear.Twelveresting-statenetworkswere identified, corresponding to previously-described RSNs (Smith etal.,2009).Alteredfunctionalconnectivitywasfoundwithinthe
default modenetwork (DMN). In particular, AN recoveredpar- ticipantsshowedincreasedresting-statefunctional connectivity betweentheDMNandtheprecuneusandtheDLPFC/inferiorfrontal gyruscomparedtocontrols.Thesecondstudywasconductedon asampleofmedication-naïveadolescentoutpatientsattheearli- eststagesofAN(lessthan6monthsofANduration)andwithno psychiatriccomorbidity(Gaudioetal.,2015).Eightresting-state networkswereindentifiedandallofthemcorrespondedtoprevi- ouslyreportedsignificantRSNs(Smithetal.,2009).Thefunctional connectivitymapsshowedalterationswithintheexecutivecontrol network.Specifically,decreasedfunctionalconnectivitywasfound inANpatientscomparedtocontrolsbetweentheexecutivecon- trolnetworkandinaregionoftheACCclosetotheborderofhe paracingulategyri.
Tosumup,thesetwostudies,whileusingthesameapproach (whole-brainICA),investigatedsamplesthatdifferedinage(i.e.
adults vsadolescents)and stageof AN(recovery statevs earli- eststagesofAN).Thefindingsofthetwostudiesimplicatedtwo differentnetworks:theexecutivecontrolnetwork andDMN.In particular,decreasedfunctionalconnectivitybetweentheexecu- tivecontrolnetworkandtheACCwasseenintheearlieststage ofAN,brainregionsassociatedwithcognitivecontrolandemo- tionalprocessingwhichareperhapsalteredneuralcircuitsinearly stages ofthedisorder.Conversely,increasedfunctional connec- tivitybetweentheDMNandtheprecuneusandtheDMNtothe DLPFC/inferiorfrontalgyruswasseeninadultswhomhadrecov- eredfromAN,apossibleresidualneuralpatternassociatedwith alteredattentionalandcognitivecontrolresources.
3.3. NetworkofinterestICAbasedapproachstudies(n=5)
Asmentionedabove,anumberofrobustandwidelyaccepted RSNshave been identified using ICA(Damoiseauxet al., 2006;
Smithetal.,2009).Buildingonthis,anotherrsfMRIapproachuses establishedRSNsasapriorinetworks-of-interestsimilarly asin theseed-based approach (e.g. Favaro et al., 2012). Five rsfMRI studiesadoptedthisapproachinvestigatingoneormoreresting- statenetworksofinterestandfocusingonANpatients(Amianto etal.,2013;Boehmetal.,2014;Phillipouetal.,2016),participants withcurrentorpastAN(Favaroetal.,2012),oronlyparticipants recoveredform AN (Boehm et al.,2016)(Table2).Thefirst by Favaroetal.(2012),focusedonfournetwork ofinterest:visual networks(i.e.medial,lateral,andventral)andonesomatosensory network,hypothesizinganimpairmentofmultisensoryintegration processinginAN.TheyscannedasamplethatwascomposedofAN patients(N=29),ANrecoveredparticipants(diseasefreeforatleast 6months.n=16)andcontrols(n=26).TheauthorsfoundthatAN patientsshowedalteredfunctionalconnectivitywithintheven- tralvisualnetworkandthesomatosensorynetworkandthatAN recoveredparticipantsonlyshowedfunctionalalterationswithin theventralvisualnetwork.Specifically,therewasdecreasedfunc- tionalconnectivitybetweentheventralvisualnetworkandtheleft occipitotemporaljunctionand increasedfunctionalconnectivity betweenthesomatosensorynetworkandtheleftsuperiorparietal cortexinANpatientscomparedtocontrols.Decreasedfunctional connectivitybetween the ventralvisual network and theright middlefrontalgyruswasalsofoundinANrecoveredparticipants comparedtocontrols.Inthesecondstudy,Amiantoetal.(2013) scannedasamplecomposedof12adultANoutpatients,12adultBN patientsand10controls,investigatingthecerebellarnetwork.The authorsfoundanumberofdifferencesbetweenthegroups.Inpar- ticular,ANoutpatientsshowedincreasedfunctionalconnectivity betweenthecerebellarnetworkandthevermis,leftinsula,bilat- eraltemporalpole,andposteriorcingulatecortexanddecreased functionalconnectivitybetweenthecerebellarnetworkandpari- etallobecomparedtocontrols.BNpatientsandthemixedgroup ofpatients(AN+BN)showedpartiallysimilarresultscomparedto controls.Inaddition,BNpatientsshowedincreasedfunctionalcon- nectivitybetweenthecerebellarnetworkandlateralhemispheric areasofthecerebellum,ACC,andprecuneusaswellasdecreased functionalconnectivitybetweenthecerebellarnetworkandthe rightinferiorfrontalgyrus.AcomparisonbetweentheANandBN groupsrevealedANpatientshadincreasedfunctionalconnectivity betweenthecerebellarnetworkandthebilateralanteriorinsula, precuneus,andrightinferiorfrontalgyrus,aswellasdecreased functionalconnectivityinthecerebellarhemispheresandACC.The thirdstudybyBoehmetal.(2014),focusedonfiverestingstatenet- works:theDMN,thesaliencenetwork,thefronto-parietalnetwork, thevisualnetwork, and somatosensory network, toinvestigate someimportantdomainsthatmaybeofrelevancetoANclinical symptoms(e.g. self-referentialprocessing, rewarding and emo- tionalstimuliprocessing,visualandsomatosensoryinformation processing).Theauthorsstudiedasampleofadolescentandyoung- adultpatientswithAN(n=35)andacontrolsample(N=35).Differ- enceswerefoundwithintheDMNandthefrontal-parietalnetwork.
Specifically,ANpatientsshowedincreasedfunctionalconnectivity betweentheDMNandtheleftanteriorinsula/frontaloperculum andincreasedfunctionalconnectivitybetweenthefrontal-parietal networkandtheleftangulargyruscomparedtocontrols.Alater studybyBoehm etal.,2016; tested 31women recoveredfrom ANagainst31age/sex-matchedcontrols.Threenetworks(i.e.the defaultmode,frontal-parietal,andsaliencenetwroks)weretested, investigatingwithinandbetweenrestingstatefunctionalconnec- tivityand functionalnetwork connectivity.No differenceswere foundintheDMN,butthefrontal-parietalnetworkhadreduced
connectivitywiththedorsolateralprefrontalcortexintherecov- eredANparticipantscomparedtocontrol.Aswasalsoreportedin patientswithactiveAN(Boehmetal.,2014),afurtherhypothesis- drivenanalysisagainfoundincreasedconnectivitybetweenthe frontal-parietalnetworkandtheleftangulargyrusinthepartic- ipantsrecoveredfromAN.Finally,nodifferenceswerefoundin functionalnetworkconnectivity.ThefifthstudybyPhillipouetal.
(2016)investigatedtheDMN,sensorimotor network,and visual networkin26femaleswithANand27age/sex-matchedcontrols.
Whilethis studytechnicallywasnot ICA-based(butROI-to-ROI seed-basedtype),wehaveincludedithere,asitlimiteditsanalysis tofunctionalnetworks.NodifferenceswerefoundintheDMN,but reducedfunctionalconnectivitywasfoundbetweenthesensory- motorandvisualnetworksinANpatientscomparedtocontrols.
Insum,andsimilarlyasintheseed-basedapproachmentioned above,thenetwork-of-interestICAbasedstudiesdidnotalways usethesameapriorinetworksofinterest,limitinganaggregation offindings, andalsocomparison betweenstudies.Inparticular, alteredfunctionalconnectivitywasfoundinANpatientswithin andbetweenthesomatosensoryandvisualnetworksandwithin thedefaultmode,fronto-parietalandcerebellarnetworks.Onthe otherhand,altered functionalconnectivitywasfoundinpartic- ipants recovered from AN in the default mode, fronto-parietal and visualnetworks.However,theseresultswerepartiallydis- cordant between both the rsfMRI studies on AN patients and participantsrecoveredfromAN.Inaddition,differencesbetween ANandbulimianervosapatientswerefoundinthecerebellarnet- workfunctionalconnectivityintheonlystudywhichalsorecruited bulimianervosapatients.Combined,thissuggeststhatnetworks implicatedincognitivecontrolandvisualandsomatosensoryinte- gration maybeaffectedin AN and thatthese alterations could partiallypersistinparticipantsrecoveredfromAN.Furthermore, itcanbepreliminarysuggestedthatrsfMRImaybeusefulfordif- ferentialEDtypes.
3.4. Graphanalysisstudies(n=4)
Graphanalysisissimilartowhole-brainICAanalysisinthatboth areemergent,data-drivenmethods,whichdonotrelyonapriori information.Graphanalysisbuildsanetworkofnodes(i.e.desig- natedbrainareasdistributedacrossthewholebrain),suchasagrid ofarrangednodes,andcreatesweightededgesbetweennodesto describefunctionalconnectivitybetweenbrainregions.Here,we examinefourstudiesutilizing suchtechniques(Kullmannetal., 2014;Ehrlichetal.,2015;Geisleretal.,2015;Lordetal.,2016) (Table2).ThemethodemployedbyKullmannetal.(2014)built anetworkbasedonnodeswiththemostnumerous,statistically- significantedges(i.e.,degreecentrality).Thesehighly-connected regionswerethencomparedbetweenthreedifferentgroups(all youngadultfemales): 12 ANpatients, 14 controls, and 12ath- letes.Reducedconnectivitywasfoundwithintheinferiorfrontal gyrus(IFG)bilaterallyforANpatientscomparedtocontrols.There werenodifferencesbetweencontrolsandathletes.Theauthors alsoperformedaseed-basedanalysisusingtheIFGastheapri- oriregion.Theyfurtherfounddecreasedconnectivitybetweenthe rightIFGandbilateralorbitalfrontalgyrus,andincreasedconnec- tivitybetweentheleftIFGtothebilateralinsulainAN patients comparedtocontrols.Again,nodifferenceswerefoundbetween controlsandathletes.ThenextstudybyEhrlichetal.(2015)tested 35ANpatientsand35age-matchedcontrols(allfemale,ranging fromearlyteenagertoyoungadult).Networkswerebuiltbypick- ingtheclusterwiththemostnumerous,statistically-significant nodes/ROIbasedonresamplingstatistics.Comparedtocontrols, theyfounddecreasedconnectivityintheANgroupforanetwork ofsevennodeswithregionsincludingtheleftamygdala,leftthala- mus,rightfusiformgyrus,bilateralputamen,andbilateralposterior
insula.Geisleret al.(2015)testedthesamecohort usinganini- tialgraph of 160nodes (arranged by meta-analysis from fMRI (Dosenbachetal.,2010),andsearchedfornetworksineachpartic- ipantbyrecursivelyremovingedgesuntilalledgesmetstatistical thresholds.Quantitative measurementsofthenetworks,aswell asofindividualnodes,werethencomparedbetweengroups.In termsofthenetwork,theANgrouphadincreasedaveragepath length(sumofedges)comparedtothecontrols,indicatinganinef- ficientnetwork.Morespecifically,atthenodallevel,ANpatients hadincreasedpathlengthintheleftmiddleinsula,rightposterior insula,andbilateralthalamus.ANpatientsalsohadlowlocaleffi- ciencyinarightposterioroccipitalnode,andincreasedefficiencyin anodewithintherightanteriorprefrontalcortex,however,these differencesinlocalefficiencyseemedunstable.Afinalgraph-based rsfMRIstudyagainexaminedthesamecohortfromEhrlichetal., 2015andGeisleretal.(2015),comparingmethodologicalstrategies forgraphanalysis(Lordetal.,2016).Inthismostrecentwork,Lord etal.comparedgraphanalysisbetweentheANpatientsandcon- trolswithtwodifferentbasesforanalysis:1)ananatomical-based analysiswhereactivityinregionswasassessedbasedontraditional, anatomically-definedregions(e.g.,insula,cingulate,etc.),and2) acalculation-basedanalysiswhereactivitywasmeasuredinthe same160sphericalROInodesassignedasmentionedabove(and establishedbyDosenbachetal.,2010).Thetwoschemesconverged ononly5outof20regionswhendetectingdifferencesbetweenAN patientsandhealthycontrols,withtheanatomicalanalysisfinding differencesinparietalandcingulateregionsandthenode-analysis findingdifferencesintheinsulaandthalamus.Theauthorscon- cludethatthemostprominentdifferencesinactivitybetweenAN patientsandcontrolsarerobust−regardlessoftheschemeusedfor analysis.However,theauthorsrecommendthatanaprioriselec- tionofregions,specificforadisease,arelikelymoreusefulthan arbitraryorrandomassignments.
In summary, partially consistentresults were foundfor the graphanalysisstudies.Functionalconnectivityalterationsmainly involvedtheinsula andthalamus aswellastheinferiorfrontal gyrus,amygdala,fusiformgyrus,orputamen.Inparticular,thelast studybyLordetal.,whichcomparedtwo typesofgraphanaly- sesbasedoneitherananatomicaloracalculatedbasis,generally echoesthedifferencesfoundin ourcomparisons oftheapriori andemergent,data-drivenanalyses:whileeach typeofanalysis indeeddetectsdifferencesbetweenANandcontrolsatthemost basiclevel,thedetailscanbevariable.Fromallthis,morestan- dardizationofresting-statemethods,particularlygraphanalysis, isrequired.Regardless,theareasfoundtoberelatedtoANinthe abovestudiescanallreasonablybelinkedtothesymptomology ofANregardingsomatosensory/visualintegrationprocessesand cognitivecontrol.
4. Discussion 4.1. Mainfindings
Thepresentpaperisthefirst,toourknowledge,tosystemati- callyreviewresting-statefunctionalmagneticresonanceimaging (rsfMRI)studiesconductedwithanorexianervosa(AN)patients.
AlthoughtheuseofrsfMRIinANresearchisrathernew,thenum- berof publishedstudiesindicatesanincreasinginterest in this neuroimagingtechniqueamonggroupsofresearch.Thegrowing interestamongresearchersisrelatedtothefactthatrsfMRIisa simpleprocedurethatcanimprovetheunderstandingofpsychi- atricdisorders(Barkhofetal.,2014).Inthissystematicreview,we included15papersusing differentresting-stateapproaches(i.e.
seed-based,wholebrainICA-bases,networkofinterestICA-based, andgraphanalysisstudies),withatotalof294participantswith
current or past AN and 285control participants.Overall, stud- iesshowed relativelyconsistent overlapin results(see Table2 andFig.1),specificallyinvolvingthecorticolimbiccircuitry,which helpstoformaclinicalperspectiveontheunderlyingneurobiolog- icallinkstoANsymptoms.Ontheotherhand,thestudieshadlittle overlapintheiranalyticalapproachand/ora-prioriassumptions.
Theseed-basedstudiesfocusedondifferentseeds(see Table2) andshowedfunctionalconnectivitydifferencesbetweenANand HCinthedACC(Leeetal.,2014),theThalamus(Biezonskietal., 2015),andtherightinferiorfrontalgyrus(Collantonietal.,2016).
Moreover,thenetwork-of-interestICA-basedstudiesfocusedon differentnetworks(e.g.,visual,somatosensory,cerebellar,fronto- parietal,defaultmode,andsaliencenetworks)inbothANpatients (Favaroetal.,2012;Amiantoetal.,2013;Boehmetal.,2014)and participantsrecoveredfromAN(Favaroetal.,2012;Boehmetal., 2016);allofwhichfoundalteredfunctionalconnectivitywithindif- ferentnetworks(seeTable2).Inparticular,arecentstudy,viaaROI toROIapproach,focusedondefaultmode,visual,somatosensory networksandfounddecreasedfunctionalconnectivitybetweenthe sensory-motorandvisualnetworksin ANpatientscompared to controls(Phillipouetal.,2016).However,resultswereinconsistent betweenstudiesinvestigatingthesamenetworks(i.e.DMNand visualnetworks)(Favaroetal.,2012;Boehmetal.,2014;Phillipou etal., 2016).Regardingthevisualnetworks,while Favaroetal.
(2012)founddecreasedfunctionalconnectivitywithintheventral visualnetworkinbothANpatientsandrecoveredANparticipants, Boehmetal.(2014)didnotfinddifferencesinthisnetworkand Pilippouandcolleaguesonlyfoundalterationsbetweenthevisual andsomatosensorynetworksinANpatients.RegardingtheDMN, increasedfunctionalconnectivitywasfoundinANpatients(Boehm etal.,2014),butnodifferenceswerefoundinparticipantsrecov- eredfromAN(Boehmetal.,2016)orwhenadoptingaROItoROI approachinANpatients(Phillipouetal.,2016)).Furthermore,the twodata-drivenapproaches,whichdonothaveapriorihypotheses (whole-brain,ICA-basedandgraphanalysis),alsoshoweddiffer- entresults. Thetwo studiesthat useda whole-brainICA-based approachshowedfunctionalconnectivityalterationsintheDMN inparticipantsrecoveredformAN(Cowdreyetal.,2014)andexec- utivecontrolnetworkinadolescentpatientswithAN(Gaudioetal., 2015),respectively.Andthefourstudiesthatadoptedgraphanal- ysisfoundalterationsinvolvingeithertheinferiorfrontalgyrus (Kullmannetal.,2014),insula(Ehrlichetal.,2015;Lordetal.,2016), orthalamus(Geisleretal.,2015;Lordetal.,2016).
Thediscordantresultsbetweenstudiesmaypartlybedueto practicallimitationssuchasdifferencesbetweencohorts:e.g.age, eatingdisorderduration,current orpastpharmacologicaltreat- ment,psychiatriccomorbidity,prandialstate.Inparticular,severity anddurationofANarewell-knownconfoundingfactorsinANneu- roimagingstudies due totheimpact of malnutritionon neural processing(Kayeetal.,2013).Thus,sampleheterogeneitycould explainsomedifferencesinthisanalysis.Forexample,thesam- plesintwowhole-brainICA-basedstudieswerecomposedofeither adultparticipantsrecovered fromAN(Cowdrey etal., 2014)or adolescentpatientsintheearlystagesofAN(Gaudioetal.,2015), whilethesamplesofthefourgraph-basedstudiesincludedadults andamixedsampleofadultsandadolescentpatientswithunre- porteddurationsofAN(Kullmannetal.,2014;Ehrlichetal.,2015;
Geisleretal.,2015;Lordetal.,2016).Thedifferenceinresting- stateapproachisanotherconfoundingfactor,asevenstudiesusing thesamecohort founddifferentresultsusingdifferentmethods (Boehmetal.,2014;Ehrlichetal.,2015;Geisleretal.,2015).Thisis expectedtosomeextent,assomersfMRIstudies(suchasnetwork- of-interest,ICA-based)arelimitedtospecificareas/networksand haveveryspecifictestinghypotheses(e.g.Amiantoetal.,2013).
Resultsofseed-basedstudies,aswell,arelimitedtotheapriori regionschosenforanalysis,anddifferentseedregionsmayproduce