tactile temporal prediction error processing Layer-specific activation in human primary somatosensory cortex during NeuroImage

ContentslistsavailableatScienceDirect

NeuroImage

journalhomepage:www.elsevier.com/locate/neuroimage

Layer-specific activation in human primary somatosensory cortex during tactile temporal prediction error processing

Yinghua Yu

, Laurentius Huber

, Jiajia Yang

, Masaki Fukunaga

, Yuhui Chai

,

David C. Jangraw

, Gang Chen

, Daniel A. Handwerker

, Peter J. Molfese

, Yoshimichi Ejima

, Norihiro Sadato

, Jinglong Wu

, Peter A. Bandettini

aGraduate School of Interdisciplinary Science and Engineering in Health Systems, Okayama University, 3-1-1 Tsushima-Naka, Kita-ku, Okayama 700-8530, Japan

bSection on Functional Imaging Methods, National Institute of Mental Health, Building 10, 10 Center Dr Bethesda, MD 20892, USA

cMR-Methods Group, MBIC, Cognitive Neuroscience Department, Faculty of Psychology and Neuroscience, University of Maastricht, Cognitive Neuroscience, Room 1.014, Oxfordlaan 55, 6229 EV Maastricht, The Netherlands

dDivision of Cerebral Research, National Institute for Physiological Sciences, 38 Nishigonaka, Myodaiji, Okazaki, Aichi, 444-8585 Japan

eScientific and Statistical Computational Core, National Institute of Mental Health, Building 10, 10 Center Dr Bethesda, MD 20892, USA

fBeijing Institute of Technology, 5 South Zhongguancun Street, Hiadian District, Beijing 100081, China

gFunctional MRI Core Facility, National Institute of Mental Health, Building 10, 10 Center Dr Bethesda, MD 20892, USA

a r t i c le i n f o

Keywords:

Layer-specific fMRI Tactile prediction

Primary somatosensory cortex Temporal prediction error High-resolution CBV–fMRI

a b s t r a ct

Thehumanbraincontinuouslygeneratespredictionsofincomingsensoryinputandcalculatescorresponding predictionerrorsfromtheperceivedinputstoupdateinternalpredictions.Inhumanprimarysomatosensory cortex(area3b),differentcorticallayersareinvolvedinreceivingthesensoryinputandgenerationoferror signals.Itremainsunknown,however,howthelayersinthehumanarea3bcontributetothetemporalprediction errorprocessing.Toinvestigatepredictionerrorrepresentationinthearea3bacrosslayers,weacquiredlayer- specificfunctionalmagneticresonanceimaging(fMRI)dataat7Tfromhumanarea3bduringataskofindex fingerpokingwithno-delay,short-delayandlong-delaytouchingsequences.Wedemonstratethatallthreetasks increasedactivityinbothsuperficialanddeeplayersofarea3bcomparedtotherandomsensoryinput.ThefMRI signalwasdifferentiallymodulatedsolelyinthedeeplayersratherthanthesuperficiallayersofarea3bbythe delaytime.Comparedwiththeno-delaystimuli,activitywasgreaterinthedeeplayersofarea3bduringtheshort- delaystimulibutlowerduringthelong-delaystimuli.Thisdifferenceactivityfeaturesinthesuperficialanddeep layerssuggestdistinctfunctionalcontributionsofarea3blayerstotactiletemporalpredictionerrorprocessing.

Thefunctionalsegregationinarea3bacrosslayersmayreflectthattheexcitatoryandinhibitoryinterplayinthe sensorycortexcontributionstoflexiblecommunicationbetweencorticallayersorbetweencorticalareas.

1. Introduction

Forsurvivalinachangingenvironment,humanslearnfromexpe- riencetopredictfutureevents.Criticaltothiscapacityistheinterpre- tationofsensoryinputandthegenerationofinternalpredictionabout futureinputs(deLangeetal.,2018;Mumford,1992;Shippetal.,2013).

Forexample,whenhumansperceivearhythmictactilesequence,they cantakeadvantageofthetemporalregularitytoformpredictionsabout thetimingoffutureinputs(Yuetal.,2019).Ifthesepredictionsmatch thetemporalrhythmsof actualsensorystimuli,this informationcan beusedtoenhancetheperceptionofsubsequentinputs.Alternatively, ifanincomingsignaldoesnotmatchtheprediction(e.g.,occurslater thanpredicted),thebrainwillgeneratepredictionerrorsignalsthatcan

∗Correspondingauthor.

E-mailaddress:[email protected](Y.Yu).

beusedtoupdatetheinternalprediction,therebyimprovingfuturesen- soryperception.Thisprocessisachievedwithinhierarchicallyorganized corticalstructuresinwhichhigher-levelareasgeneratepredictionsand transmit signalsbacktolower-levelareasthrough top-downconnec- tions.Incontrast,lower-levelareasservetocalculatepredictionerrors betweenpredictionandactualityandthentransmiterrorsignalsbackto higher-levelareasthroughbottom-upconnectionstooptimizeinternal prediction(BarrettandSimmons,2015;KellerandMrsic-Flogel,2018).

The humanprimary somatosensory cortex(S1) is essential not only forreceivingsensoryinputbutalsoforreceivingpredictivefeedback (Yuetal.,2019),buttheprecisefunctionsofindividualcorticallayers ofS1inpredictionerrorprocessingisunclear.

The neural circuitsunderlying tactile prediction processing have beenmappedtoarea3bofhumanS1.Inourpreviousstudy(Yuetal.,

https://doi.org/10.1016/j.neuroimage.2021.118867.

Received21July2021;Receivedinrevisedform27December2021;Accepted29December2021 Availableonline30December2021.

1053-8119/© 2021TheAuthors.PublishedbyElsevierInc.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense

andL5/6neuronsarecriticalforintegratinginformationrelatedtosen- soryinputwithcorticalfeedbacktoreducepredictionerrors.

RecentcomputationalmodelingstudieshaveproposedthatL2/3and L5/6neurons make distinct functional contributions tosensory pre- dictionandpredictionerror processing (BarrettandSimmons,2015; Bastosetal.,2012).L2/3pyramidalneuronsarethoughttocomputethe errorbetweenpredictedandactualsensoryinputsandthensenderror signalstohigher-levelareas(e.g.,middlecingulatecortex(Yangetal., 2021a))andtoL5/6neuronsinthesamecolumn.Thecortico-cortical connectionsfromL2/3neuronstohigher-levelareasservetotransmit predictionerror signals sothat these areas cangenerate more accu- ratefuturepredictions.Incontrast,theintracolumnarprojectionsfrom L2/3toL5/6modulate thegainof incoming sensoryinput strength (BarrettandSimmons,2015).These distinctfunctionalroles ofL2/3 andL5/6neuronsareconsistentwiththosereportedinanimal stud- ies(JordanandKeller,2020;O’Connoretal.,2010;Plutaetal.,2015; Quiquempoixetal., 2018).Inparticular, arecentstudy(Jordanand Keller, 2020)suggeststhat L2/3neuronsof themouse visualcortex computepredictionerrorsbysubtracting predictedandactual visual flowinputs,whereasL5/6neuronsplayaroleinintegratingvisualflow andlocomotion.Todate,however,thereisnoempiricalevidencefor suchprecisecontributionsofL2/3andL5/6topredictionandprediction errorinhumans.Thislackofknowledgeislargelyduetothetechnical challengesofcapturinglayer-specificbrainactivityintherelativelythin (about2mm)humansensorycortex(Dumoulinetal.,2018;Finnetal., 2020;Selfetal.,2019;TurnerandDeHaan,2017;Uludağ andBlin- der,2018;Yangetal.,2021b).

In the present study, we aim to investigate the cortical layer- specificactivity thatis related toprediction error processing in hu- manarea3b.Weemployedlayer-specificfunctionalmagneticresonance imaging (fMRI) at 7T using concurrent measures of vascular-space- occupancy (VASO) and blood oxygenation level-dependent (BOLD) imaging(Huberetal., 2020,2017, 2016).Followingtheperspective ofthepreviousfindings(JordanandKeller,2020),wehypothesizethat bothL2/3andL5/6ofhumanarea3bwillbeinvolvedintheprediction errorprocessing,whileeachlayermayhavedistinctfunctions.Inpar- ticular,L2/3mightreceivethetop-downfeedbackandfeatureforerror calculation,whiletheL5/6mightspecializeintheintegrationandmod- ulationofmultipleinputssuchasincomingsensoryinputandtop-down feedback.Totestthishypothesis,wedesignedthreetactiletemporal prediction(TP)tasks.Allparticipantsreceivedrhythmictactilefinger pokinginthreeTPtasksandwereaskedtopredictatargetpoketothe indexfinger.Wemanipulatedthetargettimingsuchthatindexfinger pokingoccurredeitheron-beat(TPontask– nodelay)oroff-beat(TPoff task– delayed).Theviolationofthetemporalrhythmin theoff-beat taskisreferredtoaspredictionerrorcondition,whichallowedustoin- vestigatehowthepredictionerrorprocessingmodulatestheactivityof L2/3andL5/6withinarea3b.Fortheoff-beatTPoff task,wefurther manipulatedthelengthofthetemporalintervalbeforetheindexfinger

protectinghumansubjects.Additionalfiveparticipantsgavewrittenin- formedconsentunderthelocalmedicalethicscommitteeattheNational InstituteforPhysiologicalSciences,Japan.Onefemalewasre-invited toparticipateinanadditionalsession(onadifferentday)toconfirm reproducibility;therefore,twelveexperimentalsessions(n=12)were conductedintotal.

2.2. Experimentalsessionsetupandimageacquisition

Each sessionconsistedof one fingersomatotopicmappingrun of 9.9mindurationandone ortwopredictiontaskrunsof16 mindu- ration.Noparticipantwasinthescannerforlongerthan120minper session.

The same fMRI sequence andimage reconstruction pipeline was used asin our previous study(Yu et al., 2019).Slice-selective slab- inversionconcurrentmeasuresofVASO(Luetal.,2003)andBOLDsig- nalswereacquiredusinga7Tscanner(SiemensHealthineers,Erlangen, Germany)equippedwitha32-channelRFcoil(NovaMedical,Wilm- ington,MA,USA)andanSC72bodygradientcoil.TheVASO-relevant TR-loop acquisition settings were as follows: TI1_nulled = 1100 ms, TI2_BOLD=2845ms,andTR_pair=3490ms.Thecoil-combineddatacom- prisedinterleavedBOLDandVASOcontrastsobtainedasconcomitant timeseries(Huberetal.,2017).Thesetimeseriesarecorrectedforrigid volumemotionandareseparatedbycontrastwiththeeffectivetempo- ralresolutionofTR=3490msforeachindividualcontrast.Thenominal resolutionwas0.71mmacrosscorticaldepthswith1.8-mmthickslices perpendiculartothepostcentralbankoftherightcentralsulcuswith 3D-EPI(Poseretal.,2010).VASOcontrastiscorrectedforBOLDcon- taminationsbythedivisionofbloodnulledMR-signalandnot-nulled MR-signal acrossconsecutiveTRs.This wasperformedbasedon 1–4 shortEPItestrunswith5measurementsandtheironlinedepictioninthe vendor-provided3D-viewer.Thehigherdetectionsensitivityof BOLD (withlowlayerspecificity)wasusedtodeterminethesingle-voxelac- tivityscoresforsubsequenceregionsofinterest(ROI)definition.The higherlayerspecificityofVASO(withlowersensitivity)was usedto extract layer-profilesofvoxel-averagesforeachlayerwithoutvenous biases.

2.3. Experimentalparadigmandprocedures

2.3.1. Fingersomatotopicmappingrun

TodelineatetheprecisecorticalROIforindividualparticipants,we mappedthesomatotopicrepresentationoftheleftfourfingers(D2:in- dex,D3:middle,D4:ring,andD5:pinky)inthecontralateralarea3b usinganon–off blockdesign.Acustom-designedfingerstimulationde- vice(Yuetal.,2019)wasusedtopokeeachfingerthroughtheplastic sticksundereachfingertip.Duringon-phase(17.5s),eachofthefour fingerswasrandomlyandindependentlypokedat4–5Hzandeachfin- gerwaspokedfivetimesinonerun.Participantswereinstructedwith

Fig.1. Illustrationoftheexperimentaltasksandtimechart.(A)Tactilestimulationsequencesofthethreetemporalprediction(TP)tasksandtherandom sensory(RS)inputcontroltask.InthethreeTPtasks,thefingerswerestimulatedinsetorder(pinkytoindex),withtheindexfingerreceivingstimulationwith thesameintervalastheotherfingers(on-beat).IntheTPoff_shortandTPoff_longtasks,indexfingerstimulationwasdelayedbyonebeat(approximately0.37s) orbytwobeats(approximately0.74s),respectively.IntheRStask,fingerswerestimulatedinarandomorder.(B)Thefourtaskblockswerepresentedfor34sat 20sintervals,andeachtaskblockwasrepeatedfourtimesineachrun.Theorderoftheblockswascounterbalanced.TPon:TemporalPredictionon-beat,nodelay;

TPoff_short:TemporalPredictionoff-beat,shortdelay;TPoff_long:TemporalPredictionoff-beat,longdelay;RS:Randomsensoryinputacrossfourfingers.

therequest:“Keepyourattentionontheleft pokedfingertipsduring on-phase.” Duetotimelimitations,oneoftheelevenparticiptantsper- formedsomatotopicmappingrunonlyforD2andD3.

2.3.2. Tactiletemporalprediction(TP)taskrun

Inthefollowingexperiments,weinvestigatedpredictionandpredic- tionerror-inducedsignalsacrosscorticallayersduringarhythmictactile stimulationtask.AsshowninFig.1A,participantsreceivedsequences oftactilepokestothefourfingertipsintwogeneralpatterns:(a)&(b) sequentiallyfromthepinkytoindex(D5toD2)inthethreeTPtasks;

(c)randomlypokingacrossthefourfingersintherandomsensory(RS) inputcontroltask.IntheTPtasks,theintervalbetweenstimulationof thepinky,ring,andmiddlefingerswasheldconstant(atapproximately 0.37s),whereastheintervalbetweenthemiddleandindexfingerswas varied.

(a)Tactiletemporalpredictionon-beat(TPon)task:Thepartici- pantswereinstructedwiththefollowingtextonthescreen:"Payatten- tiontothepokingoneachfingerandthenpredictwhenyourleftindex fingerwillbepokedbasedonthetemporalrhythm."Theactualsensory stimuliinvolvedtheexperimenterpokingtheparticipants’fourfingers inanorderedfashionfromD5toD4toD3toD2.Thetemporalrhythm betweeneachpokingwasapproximately0.37s.Inthiscase,thetem- poralrhythmofD2pokingmatchedthepredictedrhythmlearnedfrom isochronouspoking ofthefirstthreefingers. Weexpectedthatthese learnedrhythmicmodulationswouldgeneratepredictivefeedbacksig- nalstotheD2regionofarea3b.Thus,predictionoftheD2stimulus wouldimprovebyaconsistenttemporalrelation.

(b) Tactile temporal prediction off-beat (TPoff_short and TPoff_long)tasks:Theexperimentalinstructionwasthesameasthe TPontask.Theactualsensorystimuliinvolvedtheexperimenterpoking thefourfingersoftheparticipantinanorderedfashionfromD5toD4 toD3toD2.ThetemporalrhythmbetweeneachpokingfromD5toD3 wasapproximately0.37s.However,thelastpokingD2wasdelayedby aonebeatinterval(additional0.37s)forTPoff_shorttaskandtwo-beats interval(additional0.74s)forTPoff_longtask.Inthesecases,thetem- poralrhythmofD2pokingmismatchedthepredictedrhythmlearned

fromisochronouspokingofthefirstthreefingers.Thus,weexpected thatalongerD3-to-D2stimulusintervalwouldinducepredictionerrors manifestedbydistinctlayer-specificactivitypatternsintheD2region ofarea3b.

(c) Randomsensory (RS) input control task: The participants wereinstructedwiththefollowingtextonthescreen:“Payattention tothepokingoneachfinger, butdonot trytopredictany pattern.” Evenwecannotrulefullyoutthepredictioneffectssincehumansare keepingpredictthefutureinputsbasedonthepredictivecodingprinci- ple(Bastosetal.,2012;deLangeetal.,2018),itwasexpectedtoreduce thepredictionofthenextpokingpositionbypresentingthefingerpok- inginrandomorder.Thus,theRStaskwasusedtoinducethalamicinput totheD2regioninarea3bwithreducedstimulus-drivenprediction.

Asshownin Fig.1B,allfourtask blocksweredeliveredfor34-s separatedbya20-soff period,andeachtaskblockwasusuallyrepeated fourtimes.Theorderoftheblockswascounterbalanced.Furthermore, topreventadaptation,TPoff_shortandTPoff_longblocksincluded80%

delayedpokingtrials,and20%oftrialsincludedrhythmicpoking as TPontask.Byadding20%no-delay(TPon)trialstoTPoff blocks,the amount oferror componentsintheTPoff blockswas variedbutstill moresignificantthanTPonblocks.Thus,theTPoff vs.TPoncontrast wasexpectedtoreflectthechangeinactivityinlayersthataremore responsivetopredictionerrorprocessing.

2.4. Dataanalysis

2.4.1. Motioncorrection

AllfMRIdatawerecorrectedforheadmotionusingStatisticalPara- metricMappingVersion12software(FunctionalImagingLaboratory, UniversityCollegeLondon,UK)(Fristonetal.,2007).Tominimizeer- rorsonthemotionestimationduetonon-linearmotionatair-tissuein- terfaces,themotionparameterestimationwasrestrictedtoamanually drawnROIofthecentralsulcus.

2.4.2. Anatomicalreferencemethods

Toavoidadditionalresolutionlossduetorepeatedresamplingsteps andtoavoidanyerrorsofthedistortioncorrectionandregistration,we

didnotregisterthefunctionaldatatoananatomicalreferencedataset.

Instead,weusedthefunctionaldatadirectlyasananatomicalreference aswasperformedpreviously(Yuetal.,2019).

2.4.3. Generallinearmodel(GLM)analysis

GLMwasconductedusingFSL5.0.9(FMRIBSoftwareLibrary,Uni- versityofOxford,UK)(Jenkinsonetal.,2012).VASOandBOLDsig- nalsforallrunsweremodeledwithaBLOCKfunctionconvolvedwith the canonical hemodynamic response function using the FEAT tool of FSL.Furthermore, wealso usedseveral AFNIcommands(Version ID=AFNI_18.1.08)forfMRIdataprocessing(Cox,1996).

2.4.4. Layeringmethodsandprofileextraction

Layer-specificanalyseswereconductedusingtheopensoftwaresuite LAYNII (https://github.com/layerfMRI/LAYNII) (Huberetal., 2021).

Theborderlinesbetweencerebrospinalfluid(CSF),graymatter(GM), andwhitematter(WM)wereusedasthebasistodefinecorticaldepths (a.k.a.layers).Toavoidsingularitiesattheedgesinangularvoxelspace, thecorticaldepthsweredefinedonafive-foldfinergridthantheorig- inalEPIresolution.Then,wefirstcreatethefingerROIimagescontain thesegmentationofGManditsborders,whichconsistsoffourinteger values(0=irrelevantvoxels,1=innerGMsurfacevoxels,2=outerGM surfacevoxels,3=pureGMvoxels).Byapplyingthelayerclassification commandtotheseROIimages,11equi-volumelines(Waehnertetal., 2014)werecalculatedacrossthecorticaldepthineachROI.TheseROIs wereusedtoextractthecorticaldepth-dependentprofilesofallexperi- mentaltasks.Pleasenotethatwithanominal0.71mmresolutionandan approximatecorticalthicknessof2mminarea3b,theeffectiveresolu- tionallowsthedetectionofonly3independentdatapoints.Hence,the defined11corticaldepthsdonotrepresenttheMRIeffectiveresolution.

Forvisualization,corticallayer-specificsmoothingwasapplied.How- ever,allcorticalactivityprofileswereevaluatedfromtheunsmoothed data.

2.4.5. Statisticalanalysis

Thedifferencebetweenanypairoftaskconditionswerestatistically assessedthroughalinearmixed-effects(LME)modelingapproachusing theRpackagenlme(RCoreTeam,2013).Withthepair-wisedifference ateachlayerfromeachparticipantasthedatafortheresponsevariable, theLMEmodelwasformulatedwithnointercept,withlayersasafixed- effectsfactorandwitharandominterceptforcross-sessionsvariability.

3. Results

3.1. Fingersomatotopicmappingrun

Thefingerregionsinarea3bweresomatotopicallymappedinthis run,inwhichfingerswereseparatelypoked.Theexamplesofindexfin- ger(D2,red)andmiddlefinger(D3,blue)BOLDactivationmapofone participantispresentedinFig.2A.AclearrepresentationofD2andD3 fingersfromthemedialtolateralsidewasidentifiedalongarea3bin allparticipants(Fig.2B).These mapsweresubsequentlyused tode- terminetheROIsofeachfingerinrelevantpartsofarea3bduringthe TPandRStaskruns.Toreduce theoverlapeffectofadjacentfingers representationinarea3b,wedelineatethepreciseD2ROIbyavoiding overlappingvoxelswithotherfingers.

3.2. Temporalprediction(TP)andrandomsensory(RS)inputcontroltask runs

Weexaminedthefunctionalactivityinarea3bduringthreeTPtasks andcomparedthemtothoseofaRStask(Fig.3AB).Spatialmapsofre- sults(shownforarepresentativeparticipantinFig.3B)indicatethatall fourtasksexcitearea3b,whichisconsistentwithourpreviousfinding (Yuetal.,2019).

Averagedlayer-specificVASOandBOLDresponseprofilesintheD2 ROIareshownforallfourtasksinFig.4A.TheseVASOactivitypro- fileswerehighlyconsistentwiththeactivitymapforasingleparticipant showninFig.3B.Despiteresidualinter-participantvariability,there- sponse profilesforall participantsarealmostconsistentlymodulated forthedifferenttasks. Specifically,theTPtasksincreasedactivityin thesuperficiallayersregardlessof predictionerror(VASOprofilesin Fig.4B).However,activityinthedeeplayersofarea3bdifferedacross thethreeTPtasks.Specifically,activityindeeplayersincreasedduring theTPoff_shorttask(greendashedline)comparedwithduringtheTPon task,butduringtheTPonandTPoff_longtasksdidnotdiffersignificantly (bluedashedline).IncontrasttoallTPtasks,thenon-predictionRStask evokedstrongactivityinthemiddlelayersbutnotinsuperficialordeep layers(blackdashedlineinFig.4A).Sincethemodulationofpredic- tionerroractivitywasdesignedforD2,theactivityofotherfingerROIs (i.e.,D3,D4,D5)showedmostlysimilarpatternsacrosstaskswhichall tasksevokedthemostrobustactivityinthemiddlelayers.Incontrast, wecanconfirmtask-relateddifferencesfromtheBOLDresponses,but thedistinctionbetweenlayerswaslessclear.ThisisbecauseBOLDhas limitedlayerspecificityandisbiasedtowardthesuperficiallayersand

Fig.3. IllustrationoftheexperimentaltasksandthecorrespondingBOLD activationmapsofarepresentativeparticipant.(A)Tactilestimulationse- quencesofthethreetemporalprediction(TP)tasksandtherandomsensory (RS)inputcontroltask.(B)Activationmaps(FSLzstatisticmaps,clustersde- terminedbyz>1.6)duringthefourtasks.Toprow:rawdata,Bottomrow:with smoothingineachlayer.Thewhitedashedlinedemarcatestheregionofinterest fortheindexfingerrepresentationinarea3b.AllthreeTPtasksevokedstrong activityinbothsuperficialanddeeplayersregardlessofwhetheraprediction errorwaspresent.Incontrast,theRStaskevokedmorerobustactivityinthe middlelayers.

largedrainingveins.Furthermore,weperformedanadditionalanalysis toverifywhetherthedifferentfingerstimulationsequences(TPonand TPoff tasksversusRStask)affectthesteady-statefunctionalresponse overtheblocks.Inshort,wedidnotfindanyspecificfunctionalresponse patternfortheRStaskcomparedtootherpredictiontasks(Figuress1 ands2).TheanalysisprocessingandresultsareprovidedintheSupple- mentarymaterial.Notably,themorefinelydefined11corticaldepths (datapoints)donotrepresenttheeffectiveresolutioninourMRIdata.

Fig.5. AveragedVASOsignalchangesinsuperficialanddeeplayersofhu- mancorticalarea3bandalayer-specificcircuitmodel.(A)Thebargraphs representtheaverageactivitychangesinsuperficiallayers(corticaldepths2–4 inFig.3)anddeeplayers(corticaldepths7–10)forallsessions.(B)Alayer- specificcircuitmodeloftheprimarysomatosensorycortex(area3b).^∗:p<0.05,

∗∗∗:p<0.001.

Toquantifythelayer-specificpredictionandpredictionerroractiv- ityinsuperficialanddeeplayersofarea3b,wecollecteddatapoints fromsuperficiallayers(datapoints2–4)anddeeplayers(datapoints 7–10)andcomparedVASOsignalchangesamongthethreeTPtasks andRStask(Fig.5A).ThesecontrastsrevealedinvertedV-shapedac- tivityinthedeeplayersthatwasdependentonthelengthof thede- lay between middle and index finger poking, with enhanced activ- ity in thedeeplayers duringthe shortdelayperiod[TPoff_short vs.

TPon,effectmagnitude=0.32±0.09,p=0.01]butreducedagaindur- ingthelongerdelayperiod[TPoff_longvs.TPoff_short,effectmagni- tude = 0.43±0.09, p<0.001].Again, activityin deeplayers during

Fig.4. Layer-specificVASOandBOLDactiv- ityprofilesoftheindexfingerregionincon- tralateralarea3b.(A)Fourtasksdifferentially modulatedevenbothVASOandBOLDactivity profiles,thedistinctionbetweenlayersofBOLD activitywaslessclear.(B)TheVASOactivity profilesshowthatforallthreeTPtaskscom- paredtotheRStask,theactivationincreased tothe same levelin superficial layers (left) butdifferentially in the deep layers (right).

TheTPoff_shorttask(greenline)inducedthe strongestactivationinthedeeplayersthanthe othertwoTPtasks.Here,n=12representsthe numberofindividuallyconductedexperimen- talsession(elevenparticipantswithoneretest).

Errorbarsindicate thestandarderrorofthe meanacrosssessions.

TPoff_long– delayedbyatwobeatsinterval).Weshoweddistinctlayer- specificactivityinhumanarea3b(Fig.4AB)duringthreeprediction tasksbyusinghigh-resolutionfMRIat7T.Bycomparinglayer-specific activityinarea3bduringthethreeTPtasks,weshowedthatactivity indeeplayers(L5/6)isselectivelymodulatedbytheerrorprocessing whichoccurredduringthedelayedperiodoftheTPoff_shorttask,while superficiallayers(L2/3)showedcomparableactivityduringallTPtasks (Fig.5A).Ourfindingsofthedistinctlayerprofilesareroughlyconsis- tentwithourhypothesisthatbothsuperficialanddeeplayersofhuman area3bwillbeinvolvedinthepredictionerrorprocessingwhileeach layercontributetodistinctpartsofthisprocessing.

ThedistinctactivityprofilesacrosscorticallayersinFig.4Bshowed thatpredictionandpredictionerrorprocessingproducespecificlami- narpatternsofneuralactivityinthesensorycortex.Weincludedthe RStaskasacontroltocharacterizethelayer-specificactivityofsensory inputwithoutpredictionandpredictionerror.Weidentifiedonepeak ofactivityinthemiddlelayers(L4)ofarea3bfortheRStask(VASO signals,dashedblackline,Fig.4A),whichissupportedbytheunder- standingthatthemiddlelayeroftheprimarysensorycortexreceives thalamicsensoryinputs(DouglasandMartin,2004).Nevertheless,one wouldexpecttoobservemorerobustactivitythanwefoundinboth superficialanddeeplayersduringtheRStask,whichmightbecaused bythebasicsensoryprocessing.Thatis,eventhoughparticipantswere askednottopredictanypatternintheRStask,itislikelythatthisac- tivitycannotbeconsciouslysuppressed:thepredictivecodingprinciple (Bastosetal.,2012;deLangeetal.,2018)suggeststhatthisinstinctis deeplyingrained.Thesepredictivesignalsarethoughttoincreasethe activityinbothsuperficialanddeeplayers.Onepossibleexplanationof thedecreasedactivityinthesuperficialanddeeplayersinthepresent studyisthatcross-fingersuppressioninarea3boccurredasallpartici- pants’fourfingerswerepokedduringtheRStask.Thiscross-fingersup- pressioneffectinarea3bhasbeenconfirmedinnon-humanprimates (Reedetal.,2011,2010)bystimulatingtwohandlocationssimultane- ouslyorasynchronously.Theyfoundthattheresponsesuppressionon theadjacentfingersinarea3bneuronsoccurredforasynchronyadja- centfingersstimulation,butthemaximumsuppressioneffectoccurred atthestimulusintervalof30msandalmostequallyforallotherlonger intervals(upto500ms).Inthepresentstudy,givenalltheintervals betweentheD3andD2stimulationaregreaterthan375ms,webelieve thatthiscross-fingersuppressionplaysanegligiblecontributiontoour temporalpredictionfindings.

ThedirectcomparisonoftheTPonandTPoff_shorttaskstotheRS taskprovidedtheevidencetosupportourhypothesisthatbothsuper- ficialanddeeplayersofhumanarea3bareinvolvedinpredictioner- rorprocessing.Specifically,whilebothTPonandTPoff_shorttasksin- creasedactivityinsuperficialanddeeplayerscomparedtotheRStask, activityindeep layerswasfurtherstrengthenedbyprediction errors intheTPoff_shorttask.Thisenhancedactivitywithinthedeeplayers mayreflecttheoccurrenceofpredictivefeedbackandpredictionerror

havioralcostsandincreasedbrainactivityforviolationofprediction (Kanaietal.,2015b;LeeandMumford,2003;RaoandBallard,1999).

Onepossibleinterpretationisthatinhibitoryprocessingmaydominate duringthelongerdelayperiod(TPoff_longtask)whenpresentingtargets outside theexpectedtemporalwindow,thereby mitigatingenhanced deep layersactivityfromthepredictionerror.Inthis context,in the presentstudy,suchinvertedV-shapedactivitycharacteristics(Fig.5A) mightreflecttherecruitmentofdeeplayers’excitatoryneuronsbypre- dictionerrorencodingdependingonthetemporalwindowduringthis dynamicprocess.This“temporaltuning” ofdeeplayersactivitymayen- hancethefidelityofsensoryprediction,allowingflexibleregulationof sensoryexpectationsatdifferenttemporalscales.

Themoststraightforwardfunctionalinterpretationofthisinhibition ofdeeplayers’activityispreparationforastimulusthatthenfailstoar- rivewithintheexpectedtemporalwindow(Adesnik,2018;Plutaetal., 2015, 2019; Slater etal., 2019).Given thenature of theTPoff_long task,feedforward connectionsratherthanfeedback connectionsmay predominate.Inotherwords,therearenotop-downpredictionstoex- plainresponsesinlowersensoryareas,butthereareinhibitorymecha- nismstopreparefornewbottom-upinputs.Theseshiftsleadtodisinhi- bitionofsuperficiallayersandinhibitionofdeeplayers.Previousnon- humananimalstudieshavesuggestedthatexcitatoryandinhibitoryin- terplayindeeplayerscontributetoflexiblecommunicationbetweencor- ticallayersorbetweencorticalareas(Adesnik,2018;HarrisandMrsic- Flogel,2013).Consistentwiththesestudies,ourfindingssuggestdom- inatinginhibitoryactivityindeeplayersofthehumansensorycortex duringlongerpredictionperiods.Further,thesefindingsdemonstratea simpleandgeneralmechanisminwhichpredictioninteractswithtem- poralpreparativeprocessing(atleastwithinacertaintemporalrange) toinfluenceperception.Wespeculatethattime-dependentexcitatory andinhibitorymodulationsindeeplayersarecriticalfortemporalpre- diction.

Ourfindingsalsoprovideinsightintothetemporaldynamicsofin- hibitory connectivityacrossareas and/orlayers.Deepcorticallayers communicatewidelywithtargetareasviaboththalamocorticalandin- tracorticalconnections(Slateretal.,2019).Thecanonicallaminaror- ganization includesexcitatoryfeedforward inputsfrom thethalamus that projectdominantlyinto L4butalso intoL5andL2/3(Fig.5B) (ConstantinopleandBruno,2013).Further,L2/3providesfeedforward inputtobothhigh-levelareasandtoL5/6,whichinturnprovidesfeed- backtothethalamusandprojectstoL2/3, formingacompleteloop (AdesnikandNaka,2018).Incontrast,otherdistinctinhibitorycircuits maysuppressactivationindeeplayersforprocessessuchasadaptation (Measeetal.,2014).Onepotentialneuralmechanismforsuppression ofdeeplayers’activityinarea3bisthroughthalamocorticalconnec- tions(Slateretal.,2019)duringpreparationforapredictedstimulus.

Briefly,thalamocorticalprojectionscancarryexcitatorysignalstofeed- forwardinputlayerswhenactualsensorystimulationoccurs,whereas feedbackinhibitoryconnectionsmayprogressivelydominatewhenac-

tualsensoryinputisdelayed.Suchreciprocalinhibitoryconnectivity maydifferentiallymodulateup-anddownstreamcommunicationinre- sponsetotemporalinputpatterns.

Insummary,weusedawell-establishedpredictiontasktoinvesti- gatehowamismatchbetweenexpectedandactualtemporalsensory input(anerrorsignal)modulateslayer-specificactivityintheprimary somatosensorycortex.Bymanipulatingtheintervalbeforeanexpected rhythmicstimulustoproduceatime-dependentpredictionerror,were- vealeddistinctlaminaractivationpatternsinthehumancorticalarea 3bforpredictionandpredictionerrortasks.Theseobservationssuggest layer-specificcontributionstosensorypredictionandpredictionerror processesaswellasprovidenewinsightsintohowthebraingenerates sensory-guidedpredictions.Moreover,weshouldpointoutthatwetook advantageofthehighlayerspecificityofVASOtomeasurelayer-specific activityinarea3b;however,thesensitivityofVASOisstilllow.This technicallimitationresultedintheon-off blockdesign,whichmayin- creasetheblock-wisebiasandweakenthetaskeffect.Thecurrentlayer fMRItechniqueisbecomingeasiertouse(Bandettinietal.,2021);fu- turestudieswillfocuson variationoftemporalpredictionprocessing acrossareasandlayerswithanevent-relateddesign.

DataandCodeAvailability

Thedatapresentedhereareavailablefromthecorrespondingauthor, Y.Y.,uponreasonablerequest.

Creditauthorshipcontributionstatement

Yinghua Yu: Conceptualization, Methodology, Formal analysis, Investigation, Writing – original draft, Writing – review & editing.

LaurentiusHuber:Methodology,Formalanalysis,Investigation,Writ- ing– review&editing.JiajiaYang:Conceptualization,Methodology, Formalanalysis,Writing– review&editing.MasakiFukunaga:Inves- tigation.YuhuiChai:Investigation.DavidC.Jangraw:Writing– re- view&editing.GangChen:Formalanalysis.DanielA.Handwerker:

Writing– review&editing.PeterJ.Molfese:Writing– review&edit- ing.YoshimichiEjima:Writing– review&editing.NorihiroSadato:

Writing– review&editing.JinglongWu:Writing– review&editing.

PeterA.Bandettini:Conceptualization,Supervision,Writing– review

&editing.

Acknowledgments

We thank ArmanKhojandi andKenny Chung for support in hu- manvolunteer scanning. Theresearch was conductedaspartof the NIMHIntramuralResearchProgram(#ZIA-MH002783).Funding:This work was supported by JSPS KAKENHI (JP18K15339, JP18H01411 andJP20K07722) andJST FOREST Program (JPMJFR2041) as well as Japan–U.S. Science and Technology Cooperation Program (Brain Research). Laurentius Huber was fundedby the NWO VENIproject 016.Veni.198.032.

Supplementarymaterials

Supplementarymaterialassociatedwiththisarticlecanbefound,in theonlineversion,atdoi:10.1016/j.neuroimage.2021.118867. References

Adesnik, H., 2018. Layer-specific excitation/inhibition balances during neuronal synchro- nization in the visual cortex. J. Physiol. 596, 1639–1657. doi: 10.1113/JP274986 . Adesnik, H., Naka, A., 2018. Cracking the function of layers in the sensory cortex. Neuron

100, 1028–1043. doi: 10.1016/j.neuron.2018.10.032 .

Bandettini, P.A. , Huber, L. , Finn, E.S. , 2021. Challenges and opportunities of mesoscopic brain mapping with fMRI. Curr. Opin. Behav. Sci. 40, 189–200 .

Barrett, L.F., Simmons, W.K., 2015. Interoceptive predictions in the brain. Nat. Rev. Neu- rosci. 16, 419–429. doi: 10.1038/nrn3950 .

Bastos, A.M., Usrey, W.M., Adams, R.A., Mangun, G.R., Fries, P., Friston, K.J., 2012. Canonical microcircuits for predictive coding. Neuron 76, 695–711.

doi: 10.1016/j.neuron.2012.10.038 .

Constantinople, C.M., Bruno, R.M., 2013. Deep Cortical Layers Are Activated Directly by Thalamus. Science (80-.). 340, 1591–1594. 10.1126/science.1236425

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

doi: 10.1006/cbmr.1996.0014 .

de Lange, F.P., Heilbron, M., Kok, P., 2018. How do expectations shape perception? Trends Cogn. Sci. 22, 764–779. doi: 10.1016/j.tics.2018.06.002 .

Douglas, R.J., Martin, K.A.C., 2004. Neuronal circuits of the neocortex. Annu. Rev. Neu- rosci. 27, 419–451. doi: 10.1146/annurev.neuro.27.070203.144152 .

Dumoulin, S.O., Fracasso, A., van der Zwaag, W., Siero, J.C.W., Petridou, N., 2018. Ultra- high field MRI: advancing systems neuroscience towards mesoscopic human brain function. Neuroimage 168, 345–357. doi: 10.1016/j.neuroimage.2017.01.028 . Finn, E.S., Huber, L., Bandettini, P.A., 2020. Higher and deeper: bringing layer fMRI to

association cortex. Prog. Neurobiol. 101930. doi: 10.1016/j.pneurobio.2020.101930 . Fracasso, A., Luijten, P.R., Dumoulin, S.O., Petridou, N., 2018. Laminar imaging of pos- itive and negative BOLD in human visual cortex at 7 T. Neuroimage 164, 100–111.

doi: 10.1016/j.neuroimage.2017.02.038 .

Friston, K.J., Ashburner, J.T., Kiebel, S.J., Nichols, T.E., Penny, W.D., 2007. Statistical parametric mapping: the analysis of functional brain images, statistical parametric mapping the analysis of functional brain images.

Harris, K.D., Mrsic-Flogel, T.D., 2013. Cortical connectivity and sensory coding. Nature 503, 51–58. doi: 10.1038/nature12654 .

Huber, L.(Renzo), Poser, B.A., Bandettini, P.A., Arora, K., Wagstyl, K., Cho, S., Goense, J., Nothnagel, N., Morgan, A.T., van den Hurk, J., Müller, A.K., Reynolds, R.C., Glen, D.R., Goebel, R., Gulban, O.F., 2021. LayNii: a software suite for layer-fMRI.

Neuroimage 237, 118091. doi: 10.1016/j.neuroimage.2021.118091 .

Huber, L., Finn, E.S., Chai, Y., Goebel, R., Stirnberg, R., Stöcker, T., Mar- rett, S., Uludag, K., Kim, S.G., Han, S., Bandettini, P.A., Poser, B.A., 2020.

Layer-dependent functional connectivity methods. Prog. Neurobiol. 101835.

doi: 10.1016/j.pneurobio.2020.101835 .

Huber, L., Goense, J., Kennerley, A.J., Ivanov, D., Krieger, S.N., Lepsien, J., Trampel, R., Turner, R., Möller, H.E., 2014. Investigation of the neurovascular coupling in posi- tive and negative BOLD responses in human brain at 7T. Neuroimage 97, 349–362.

doi: 10.1016/j.neuroimage.2014.04.022 .

Huber, L., Handwerker, D.A., Jangraw, D.C., Chen, G., Hall, A., Stüber, C., Gonzalez- Castillo, J., Ivanov, D., Marrett, S., Guidi, M., Goense, J., Poser, B.A., Bandet- tini, P.A., 2017. High-resolution CBV-fMRI allows mapping of laminar activity and connectivity of cortical input and output in human M1. Neuron 96, 1253–1263.

doi: 10.1016/j.neuron.2017.11.005 .

Huber, L., Ivanov, D., Guidi, M., Turner, R., Uluda ğ, K., Möller, H.E., Poser, B.A., 2016.

Functional cerebral blood volume mapping with simultaneous multi-slice acquisition.

Neuroimage 125, 1159–1168. doi: 10.1016/j.neuroimage.2015.10.082 .

Jenkinson, M., Beckmann, C.F., Behrens, T.E.J., Woolrich, M.W., Smith, S.M., 2012. FSL–

Review. Neuroimage 62, 782–790. doi: 10.1016/j.neuroimage.2011.09.015 . Jordan, R., Keller, G.B., 2020. Opposing Influence of top-down and bottom-up in-

put on excitatory layer 2/3 neurons in mouse primary visual cortex. Neuron 108.

doi: 10.1016/j.neuron.2020.09.024 , 1194-1206.e5 .

Kachergis, G., Wyatte, D., O’Reilly, R.C., de Kleijn, R., Hommel, B., 2014. A con- tinuous time neural model for sequential action. Philos. Trans. R. Soc. B 1–8.

doi: 10.1098/rstb.2013.0623 .

Kanai, R., Komura, Y., Shipp, S., Friston, K., 2015a. Cerebral hierarchies: predictive processing, precision and the pulvinar. Philos. Trans. R. Soc. B Biol. Sci. 370.

doi: 10.1098/rstb.2014.0169 .

Kanai, R., Komura, Y., Shipp, S., Friston, K., 2015b. Cerebral hierarchies: predictive processing, precision and the pulvinar. Philos. Trans. R. Soc. B Biol. Sci. 370.

doi: 10.1098/rstb.2014.0169 .

Keller, G.B., Mrsic-Flogel, T.D., 2018. Predictive processing: a canonical cortical compu- tation. Neuron 100, 424–435. doi: 10.1016/j.neuron.2018.10.003 .

Klein, R., Ivanoff, J., 2000. Inhibition of return. Trends Cogn. Sci. 4, 138–147.

doi: 10.4249/scholarpedia.3650 .

Lawrence, S.J.D., van Mourik, T., Kok, P., Koopmans, P.J., Norris, D.G., de Lange, F.P., 2018. Laminar organization of working memory signals in human visual cortex. Curr.

Biol. 28, 3435–3440. doi: 10.1016/j.cub.2018.08.043 .

Lee, T.S. , Mumford, D. , 2003. Hierarchical Bayesian inference in the visual cortex. J. Opt.

Soc. Am. A 20, 1434–1448 .

Lu, H., Golay, X., Pekar, J.J., Van Zijl, P.C.M., 2003. Functional magnetic resonance imag- ing based on changes in vascular space occupancy. Magn. Reson. Med. 50, 263–274.

doi: 10.1002/mrm.10519 .

Manita, S., Suzuki, T., Homma, C., Matsumoto, T., Odagawa, M., Yamada, K., Ota, K., Mat- subara, C., Inutsuka, A., Sato, M., Ohkura, M., Yamanaka, A., Yanagawa, Y., Nakai, J., Hayashi, Y., Larkum, M.E., Murayama, M., 2015. A top-down cortical circuit for accu- rate sensory perception. Neuron 86, 1304–1316. doi: 10.1016/j.neuron.2015.05.006 . Mease, R.A., Krieger, P., Groh, A., 2014. Cortical control of adaptation and sen- sory relay mode in the thalamus. Proc. Natl. Acad. Sci. USA 111, 6798–6803.

doi: 10.1073/pnas.1318665111 .

Mumford, D., 1992. On the computational architecture of the neocortex. Biol. Cybern. 66, 241–251. doi: 10.1007/BF00198477 .

Nobre, A.C., Van Ede, F., 2018. Anticipated moments: temporal structure in attention.

Nat. Rev. Neurosci. doi: 10.1038/nrn.2017.141 .

O’Connor, D.H., Peron, S.P., Huber, D., Svoboda, K., 2010. Neural activity in barrel cor- tex underlying vibrissa-based object localization in mice. Neuron 67, 1048–1061.

doi: 10.1016/j.neuron.2010.08.026 .

reflect widespread spatiotemporal integration. J. Neurophysiol. 103, 2139–2157.

doi: 10.1152/jn.00709.2009 .

Roelfsema, P.R., Holtmaat, A., 2018. Control of synaptic plasticity in deep cortical net- works. Nat. Rev. Neurosci. 19, 166–180. doi: 10.1038/nrn.2018.6 .

Self, M.W., van Kerkoerle, T., Goebel, R., Roelfsema, P.R., 2019. Benchmarking lam- inar fMRI: neuronal spiking and synaptic activity during top-down and bottom- up processing in the different layers of cortex. Neuroimage 197, 806–817.

doi: 10.1016/j.neuroimage.2017.06.045 .

doi: 10.1016/j.neubiorev.2021.07.005 .

Yu, Y., Huber, L., Yang, J., Jangraw, D.C., Handwerker, D.A., Molfese, P.J., Chen, G., Ejima, Y., Wu, J., Bandettini, P.A., 2019. Layer-specific activation of sensory input and predictive feedback in the human primary somatosensory cortex. Sci. Adv. 5, eaav9053. doi: 10.1126/sciadv.aav9053 .