theta phase coordinated memory reactivation reoccurs in a ...found an early reactivation episode...

1

Thetaphasecoordinatedmemoryreactivationreoccursinaslow-oscillatoryrhythmduringNREMsleepThomas Schreiner1, Christian F. Doeller1,2, Ole Jensen3, Björn Rasch4 & TobiasStaudigl1

1Donders Institute forBrain,CognitionandBehaviour,RadboudUniversity,Nijmegen,TheNetherlands2KavliInstituteforSystemsNeuroscience,CentreforNeuralComputation,TheEgilandPauline Braathen and Fred Kavli Centre for Cortical Microcircuits, NTNU, NorwegianUniversityofScienceandTechnology,St.OlavsHospital,TrondheimUniversityHospital,Trondheim,Norway

3UniversityofBirmingham,SchoolofPsychology,Birmingham,UnitedKingdom4UniversityofFribourg,DepartmentofPsychology,Fribourg,SwitzerlandCorrespondenceshouldbeaddressedtoT.S.([email protected])orT.St.([email protected]).

not peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (which was. http://dx.doi.org/10.1101/202143doi: bioRxiv preprint first posted online Oct. 12, 2017;

http://dx.doi.org/10.1101/202143

2

It has been proposed that sleep’s contribution to memory consolidation is toreactivate prior encoded information. To elucidate the neural mechanismscarrying reactivation-related mnemonic information, we investigated whethercontent-specificmemorysignaturesassociatedwithmemoryreactivationduringwakefulness reoccur during subsequent sleep.We show that theta oscillationsorchestratethereactivationofmemories,irrespectiveofthephysiologicalstate.Reactivationpatternsduringsleepautonomouslyre-emergedata rateof1Hz,indicatingacoordinationbyslowoscillatoryactivity.The memory function of sleep relies on the reactivation of newly acquiredinformation during non rapid-eye movement (NREM) sleep1. Rodent studieshave consistently shown hippocampal reactivation of previous learningexperiences during sleep2, while studies in humans have provided first hintsindicating similar processes3,4. Furthermore, triggering reactivation processesduring sleep by re-exposure to associated memory cues (targeted memoryreactivation,‘TMR’)hasbeenshowntoimprovememoryconsolidation5.However, the neural mechanisms coordinating reactivation-related mnemonicinformation are essentially unknown. Here, we investigated whether memoryreactivationduringwakefulnessandsleepshareoscillatorypatterns thatcarrymemory-representation specific information, using electroencephalography(EEG)andmultivariateanalysismethods.Buildingonprevious findings6,wehypothesized that low-frequencyoscillatoryphase conveys a representation (i.e. content)-specific temporal code. Usingrepresentationalsimilarityanalysis(RSA)torevealthephase-relatedsimilaritybetween content-specific representations7, we provide evidence for memory-reactivationprocessesduringwakefulnessandtheirreoccurrenceduringNREMsleep.

Figure1:Experimentaldesignandbehavioralresults.(a) Participantsperformeda vocabulary-learning task in the evening. They learned to associate Dutch words (cues) withGerman words (targets). After the initial learning phase, a cued recall including feedback wasperformed (‘recall1’). The cued recall was repeated without feedback (‘recall2’). Subsequently,participantssleptfor3hours.During NREM sleep, 80Dutchwords(40cued,40cued+feedback)were repeatedly presented. Memory performance was assessed in the final retrieval phase aftersleep(b)PresentingsingleDutchwordcuesduringNREMsleepenhancedmemoryperformanceascomparedtoword-pairTMRanduncuedwords.Retrievalperformanceisindicatedaspercentageofrecalledwords,withperformancebeforesleepsetto100%.Valuesaremean±s.e.m.**P<0.01.


http://dx.doi.org/10.1101/202143

3

In a first stepwe evaluated,whetherwe could identify content-specific phasepatterns of memory retrieval during wakefulness, indicating recall-relatedmemory reactivation. The degree of phase similarity for retrieving the samememory content (i.e., word) during consecutive recall instances (recall1 /recall2) was assessed using the pairwise phase consistency and contrastedbetween rememberedandnon-rememberedwords (SupplementaryFig.1+2)forfrequenciesbetween3and16Hz.We found significantly higher phase similarity for remembered words ascomparedtonon-rememberedwordsinthethetarange(cluster-randomization:P=0.006,correctedformultiplecomparisons),peakingat5Hz(Fig.2a+b).Thetime-course of the phase similarity at 5 Hz displayed an early significantdifference between remembered and non-remembered words (cluster-randomization: P = 0.008, corrected for multiple comparisons; Fig. 2c).Additional analyses indicated that thephase similarity resultswerenotbiasedby spectral power (see Supplementary Results). Given our design, it seemsunlikely that those resultswere driven by similarities in auditory stimulation.Still, we tested this possibility by assessing phase-similarity between learningand both recall instances, with the learning data being segmented around theonset of the Dutch words (thus before any association was learned; seeMethods).Nosignificantclusterwasobserved(bothP’s>0.3).

Figure2:Wordspecificphase-similarityduringwakeretrieval(a) Significantly enhanced phase similarity during successful subsequent retrieval was observedearly after cue onset (t = 0 sec) in the theta range (b) peaking at 5Hz. (c) time-course andtopographyofphasesimilarityat5Hz,indicatingarapidreactivationofmemorycontent.Theonesecondtime-windowaroundthecenterofthestrongestclusterishighlighted.


http://dx.doi.org/10.1101/202143

4

Thenextcrucialstepwastotestwhetherthesecontentspecificfeaturestracked by phase-similarity at 5 Hz would be shared between reactivationprocesses during wakefulness (‘recall2’) and sleep (TMR). Because memoryreactivationduringsleepcouldemergeatanypointafterTMRcuepresentation,phase-similaritybetweenrecall2andTMRwasexaminedwithaslidingwindowapproach7,usingthesingle-trialphaselockingvalue.Targetwordsrememberedafter sleep were paired with their equivalent during ‘recall2’ and contrastedagainst non-remembered words. A one second time-window exhibiting thestrongestcontent-specificityfromthepre-sleepretrieval(center:0.193ms)wasused as sliding window. Test statistics on the averaged difference betweenremembered and non-remembered words revealed the reactivation of recall-relatedphase-patternsat5HzduringTMRduringsleep(cluster-randomization:P = 0.02, corrected for multiple comparisons) over right temporal electrodes(Fig. 3a). There was no significant difference in spectral power biasing theresults (see SupplementaryResults). To assess the frequency specificity of theobtained results, the same analysis was performed for 3Hz and 8Hz. Nosignificanteffectswerefound(bothP>0.16).

Toexaminethetime-courseofthereactivationeffect,similarity-measureswere averagedacross significant electrodes and t-statisticswere computed foreverytime-point.Fourdistinctreactivationepisodesemerged,peakingat390ms(t16=4.49,P=0.0003),1990ms(t16=4.59,P=0.0002),2760ms(t16=3.08,P=0.007) and3310ms (t16 = 3.31,P = 0.004). Importantly, this pattern of resultssuggests that presenting a memory cue during sleep triggered a repetitivecascade of memory-reactivation, with reactivation processes fluctuating at afrequency of ~1 Hz (Fig.3d). Testing all combinations of recall and TMR timewindows revealed that no additional recall episodes were reactivated duringTMR(Fig.3b).Toevaluatethesourcesofthescalp-leveleffects,phasesimilaritywasassessedon virtual sensors by applying a DICS beamformer. Source level contrastsrevealed differences in right (para)hippocampal regions as well as morewidespreaddifferenceinleftfrontalareas,includingthefrontalgyrusandinsula(Fig.3c).


http://dx.doi.org/10.1101/202143

5

Figure 3: Retrieval-TMR similarity. (a)Re-currentreactivationofrecall-relatedphase-patternsat5HzduringTMRemergedoverrighttemporalelectrodes(b)Assessingphase-similaritybetween every time-point of retrieval and TMR confirmed the cycling result pattern (c) Sourcereconstructionoftheobtainedeffectsindicatingdifferencesinright(para)hippocampalregionsandleft-frontalareas(d)FrequencyspectrumoftheTMRsimilaritymeasuresshoweda1Hzperiodicityofreactivationprocesses.(e) Inlinewithbehavioralpredictions,providingatargetstimulusaftertheTMR-cueblockedassociatedreactivationprocesses.

Our analysis focused on similaritymeasures between recall and single-cueTMR,becausepresentingDutch-Germanword-pairsduringsleepabolishedthe beneficial effects of TMR on later memory performance8. Based on thisbehavioraloutcome,wepredicted thatprovidingboth thecueandtargetwordduringTMRshouldblock functionally relevantmemory-reactivationprocesses.Inlinewithourhypothesis,therewasnosignificanteffectwhencomparingtheaveraged difference between subsequently remembered and non-rememberedwords(P>0.17).Asthetopographicaldistributionresembledourmainresults(Fig.3e),thesameelectrodeclusterwasusedtocharacterizethetime-course.Wefound an early reactivation episodepeaking at 270ms (t16=3.2,P = 0.005) forword-pairTMR,thusbeforetheonsetofthesecondword.Nolaterepisodewasobservable, indicating that the presentation of a second stimulus may haveblockedfurthermemoryreactivation.

We show for the first time thatmemory-related reactivation processesduring wakefulness and sleep share the same neural signatures in humans.Specifically,thetaoscillationsat5Hzorchestratedthereactivationofmemories,irrespective of the physiological state. This result thereby extends previousfindings, indicating a role for theta oscillations in mediating communicationbetween the medial temporal lobe and neocortical regions9 and by that thecortical reactivation of memories during wake retrieval10. Importantly,hippocampal reactivations are also thought to drive consolidation processesduringsleep,leadingtotheintegrationofnewlyacquiredmemoriesintocorticalnetworks1.Consistently,oursourcelevelresultscorroboratethisassumptionas


http://dx.doi.org/10.1101/202143

6

not only right (para)hippocampal areas, previously associated with successfulTMR inhumans11, but also language-related regions in the left frontal cortex12showedthephasesimilarityeffects.

Moreover, presentingmemory cues during sleep triggered a cascade ofreactivations.AfterprovidingaTMRcue,memory-relatedreactivationpatternsre-emerged autonomously at a frequency of ~1 Hz, which is in line with theassumption that slow oscillatory activity during sleep guides reactivationprocesses1,5. This result also tightly parallels previouswork in rodents, whichdemonstrated thatpresenting auditory cuesduring sleepbiases the contentofassociatedmemoryreactivationswithmaintainingthebiasingeffectformultipleseconds13,14.Importantly,presentingasecondstimulusabolishedthisfluctuationanddiminishedthebeneficialeffectsofTMR.Insumourresultsdemonstratethesimilarityofmemoryreactivationduringwakefulnessandsleep,withacyclingand spontaneous re-processing of memories during sleep when triggered bycueing.

References1. Rasch,B.&Born,J.Physiol.Rev.93,681–766(2013).

2. Chen,Z.&Wilson,M.A.TrendsNeurosci.40,260–275(2017).

3. Peigneux,P.etal.Neuron44,535–45(2004).

4. Schönauer,M.etal.Nat.Commun.8,15404(2017).

5. Oudiette,D.&Paller,K.A.TrendsCogn.Sci.(2013).

6. Schyns,P.G.,Thut,G.&Gross,J.PLoSBiol.9,(2011).

7. Michelmann,S.,Bowman,H.&Hanslmayr,S.PLoSBiol.14,e1002528(2016).

8. Schreiner,T.,Lehmann,M.&Rasch,B.Nat.Commun.6,8729(2015).

9. Fuentemilla,L.,Barnes,G.R.,Düzel,E.&Levine,B.Neuroimage85Pt2,730–7(2014).

10. Nyhus,E.&Curran,T.Neurosci.Biobehav.Rev.34,1023–1035(2010).

11. Dongen,E.V.Vanetal.(2012).doi:10.1073/pnas.1201072109

12. Binder,J.R.,Desai,R.H.,Graves,W.W.&Conant,L.L.Cereb.Cortex19,2767–2796(2009).

13. Bendor,D.&Wilson,M.aNat.Neurosci.15,1439–44(2012).

14. Rothschild,G.,Eban,E.&Frank,L.M.Nat.Neurosci.(2016).doi:10.1038/nn.4457


http://dx.doi.org/10.1101/202143

7

DataandcodeavailabilityAll data and analysis code are available on reasonable request from thecorrespondingauthor.AcknowledgementsT.S. is supported by a grant of the Swiss National Science Foundation (SNSF;P2ZHP1_164994). T.St. received funding from the European Union's Horizon2020 research and innovationprogrammeunder grant agreementNo661373.C.F.D.’s research is funded by the Kavli Foundation, the Centre of ExcellenceschemeoftheResearchCouncilofNorway–CentreforBiologyofMemoryandCentre forNeural Computation, The Egil and Pauline Braathen and Fred KavliCentre for Cortical Microcircuits, the National Infrastructure scheme of theResearch Council of Norway –NORBRAIN, the Netherlands Organisation forScientific Research (NWO-Vidi 452-12-009; NWO-Gravitation 024-001-006;NWO-MaGW406-14-114;NWO-MaGW406-15-291)andtheEuropeanResearchCouncil(ERC-StGRECONTEXT261177;ERC-CoGGEOCOG724836).B.R. is supported by the SNSF (100014_162388) and the Clinical ResearchPriorityProgram(CRPP)“SleepandHealth”fromtheUniversityofZurich.ContributionsT.SandB.R.conceivedthedesign.T.S.collectedthedata.T.S.andT.St.analyzedthe data. T.S. and T.St. wrote themanuscript. All of the authors discussed theanalysesandresultsandfinalizedthemanuscript.

CompetingfinancialinterestsTheauthorsdeclarenocompetingfinancialinterests.CorrespondingauthorCorrespondenceto:ThomasSchreiner([email protected])orTobiasStaudigl([email protected])


http://dx.doi.org/10.1101/202143

8

OnlineMethods ParticipantsThedataweretaken fromSchreiner,Lehmann&Rasch(2015)1.Thus,detailedinformation about participants, stimuli, task, data acquisition and behavioralresultscanbefoundintheoriginalarticle.Fromthetotalof20participants(13female; age: 22.45 ± 2.39)who entered themain EEG analyses in the originalstudy1,3datasetshadtobeexcludedduetoextensiveartifactsinthepre-sleepEEGdata(recall1and2,foradetaileddescriptionpleaseseebelow).TaskandProcedureAllparticipantsperformedavocabulary-learning task in theevening (~10pm).Thetaskconsistedof120DutchwordsandtheirGermantranslation,randomlypresented in three rounds.With regards to the first learning round, each trialconsistedofaDutchword,whichwassucceededbytheGermantranslation.Allwordswerepresentedvialoudspeaker.Thetrialsofthesecondround(referredtoas‘recall1’)startedwiththepresentationoftheDutchword(cue),followedbyaquestionmarkforupto7seconds.Theparticipantswereaskedtovocalizethecorrect German translation (target) within the 7 seconds (if possible) or toindicate if they were not able to do so. In any case, the correct Germantranslation was presented afterwards. The same cued recall procedure wasaccomplished in the third round (‘recall2’), except that performance feedbackwas omitted (for a detailed scheme for recall1 and recall2 see SupplementaryFig.1).Thelearningphasewasfollowedbya3hoursretentionintervalofsleep.DuringNREMsleepsubsetsoftheDutchcuewordslearnedbeforetheretentioninterval were repeatedly replayed for 90 minutes via loudspeaker, either assingle cues (only the Dutchwords), 40 asword pair cues (Dutch and Germanwords)and40werenotreplayedatall.Duringwordpaircueing,thepresentedword-pairsconsistedofcorrectword-pairs justas learnedbeforetheretentioninterval for 8 participants, while 9 participants were presented with newlyformed Dutch-German word-pairs during sleep. Importantly opposed toreplayingsinglecues,cueingofwordpairs,irrespectiveofthecategory(correct,wrong), was associated with a suppression of the beneficial effects of cueing.Thus,we capitalizedourmainanalysison the single-cueTMRcondition,whilethe word-pair condition served as important control analysis. In all threecategories,therelationofrememberedandnon-rememberedwordpairsofthelast learningtrialbeforesleepwasmaintained.Hence,allcategoriescomprisedthesamenumberofrememberedandnon-rememberedwordsbeforesleep.Allwords were individually and randomly chosen for each participant using anautomaticMATLABalgorithm.Aftersleep,recallofthevocabularywastestedinafinalretrievalphaseusingacuedrecallprocedure.


http://dx.doi.org/10.1101/202143

9

EEGrecordingandpreprocessingEEG was recorded using a high-density 128-channel Geodesic Sensor Net(ElectricalGeodesics,Eugene,OR).Impedanceswerekeptbelow50kΩ.Voltagewas sampled at 500Hz and initially referenced to electrode Cz. Offline EEGpreprocessing was realized using BrainVision Analyzer software (version 2.0;Brain Products, Gilching, Germany). Data were offline re-referenced to anaverage-reference. The continuous EEG was epoched into intervals from1,000msbeforeuntil3,000msafterwordonset.Trialswithartifacts(e.g.,muscleand movement artifacts) were manually removed after visual inspection. Eyeblinks and movements of the pre-sleep EEG recordings (recall1 & 2) werecorrected using independent component analysis2. For each phase (recall1,recall2 and TMR), segments were categorized based on the subjects’ memoryperformance in the final retrieval phase into later remembered and non-rememberedwords.All succeedinganalysesstepswererealizedwithMATLAB(theMathWorks)usingtheopen-sourceFieldTriptoolbox3.WordspecificphasesimilarityduringawakerecallTodetectcontent(i.e.word)-specificphasepatternsofsuccessfulrecallduringwakefulness, a modified version of the pairwise phase consistency (PPC) wasapplied4,5. Ina first step,oscillatoryphasewasextractedusingcomplexMorletwaveletsof6cyclesforallfrequenciesbetween1and20Hzinstepsof0.5Hzand1ms,rangingfrom1000mspre-stimulusto3000msafterstimulusonset.To compute the pairwise phase consistency,words remembered after ‘recall2’werepairedwiththeirequivalentduring‘recall1’andcontrastedagainstwordsnot rememberedduring ‘recall2’and theirequivalentduring ‘recall1’.Thereby,thedegreeofphasesimilaritybetweenidenticalwordsandassociatedmemorycontentwasassessedduringconsecutiverecallinstances(recall1andrecall2;foran illustration see supplementary Fig.1). The rationale hereby is that recallprocesses associatedwith remembered items should exhibit a higher content-relatedsimilaritytoeachotherascomparedtonon-rememberedones,giventhepotentiallyweak/non-existingreactivationofamemorytraceinthelattercase.For each pair of trials, the cosine of the absolute angular distance was thencomputed and finally averaged across all (same or different) combinations4,5.Therebyavalue, representing theaveragesimilarityspecifically foreachsetofcombinations, was derived for every electrode, frequency and time-bin andsubsequently used for statistics. We assessed phase similarity for the samewords(identicalauditorycuewordpresentation)presentedduring‘recall1’and‘recall2’ and contrasted remembered pairs against non-remembered pairs.Therefore,potentialconfoundinginfluencesofsimilarityinauditorystimulationormereperceptualprocessesshouldbeequalandbothconditions(rememberedandnon-rememberedpairs)andthuscontrolledfor.Tofurtherstrengthenthispoint, we assessed the phase-similarity between learning and both recallinstances. Importantly,data from learningwasalsosegmentedwithregards to


http://dx.doi.org/10.1101/202143

10

theonsetoftheDutchwords.SincethepresentationoftheGermantranslationsduringlearningwasadelayedby3seconds,nomemoryreactivationcouldhavebeen made at this point and the recorded EEG activity primarily mirrors theauditoryperceptionandprocessingofaparticularword.Furthermore,aspowerdifferencescanbiasphaseestimation,wefurthertestedwhethertherewasasignificantdifferenceinpowerbetweenconditions.Initially,we estimated oscillatory power for the very same contrasts as utilized in thephaseanalyses(i.e.oscillatorypowerforwordsrememberedafter‘recall2’wassubtractedfrompowervaluesof‘recall1’andcontrastedagainstthedifferenceofpower-values of words not remembered during ‘recall2’ and their equivalentduring‘recall1’).Inaddition,wetestedpotentialdifferencesinoscillatorypowerfor remembered versus non-remembered items, independently for recall1 andrecall2 (for adetailedoverview see SupplementaryFig. 3). Powervalueswereextractedspecificallyfor5HzusingacomplexMorletwaveletof6cycles.WordspecificphasesimilaritybetweenrecallandTMRAs described above, we initially focused our analysis to the single-cue TMRcondition (theword-pairs conditionwas used as a control analysis, describedbelow). Thus, we tested in a first step whether content specific features ofmemorieswouldbesharedbetweenrecallbeforeandsingle-wordTMRduringsleep, indicating that TMR during sleep leads to the reactivation of thoseproperties. Target words remembered after cueing during sleep were pairedwith their equivalent during ‘recall2’ and contrasted against target words notremembered after sleep-cueing and their equivalent during ‘recall2’. Thus,between those pairs of successfully acquired memories a phase similaritymeasurewascalculatedandcontrastedagainstwords lackingastablememorytrace.FollowingMichelmannetal.(2016)5,severalrestrictionswereusedwhenapplyingthephase-basedRSAtospecifyelectrodesandfrequencyofinterestandthe utilized time windows. We used 5Hz as frequency of interest, given thatcontent-specific phase similarity peaked at this frequency during wake-recall.Hence, oscillatory phase at 5Hzwasmaximally content-sensitive during recallbeforesleep.Inaddition,thetime-windowat‘recall2’wasconfinedto1second,evenly distributed (±500ms) around the peak of the recall-associated phase-similarity measure of 5Hz. Similarly, electrodes were adopted from showingcontent-specificphasesimilarityduringwake-recall(83electrodesintotal).Toaccount for the fact thatreactivationprocessesassociatedwithTMRduringsleep could happen at any time after cue presentation, a sliding windowapproachwasutilized.Thereby,phase similarity for all trial combinationswasdeterminedbetweeneachelectrodeandtimepointassociatedwithTMRandthe1secondtimewindowderivedfromtherecallbeforesleep.We determined phase similarity with the Single-trial Phase Locking Value (S-PLV)6,hencethesimilaritywithregardstophaseangledifferencesovertime.Again,phasevalueswereextractedusingacomplexMorletwaveletof6cycles


http://dx.doi.org/10.1101/202143

11

for frequencies between 1 and 20Hz in steps of 0.5Hz. For computationalefficiency phase values were down-sampled to 100Hz. Phase similarity wasassessedbetweeneverypairofremembereditemsandcontrastedagainstphasesimilarity between pairs of non-remembered items. The pre-stimulus intervalbetween -500msand0mswasusedaspadding toslide therecallwindow intoTMRepisodes.Next,weexploredthetime-courseofreactivationprocessesduringsleep.Todoso the electrodes derived from the 5Hz cluster displaying the significantdifferencewereaveragedandsubjected toa seriesofpost-hoc t-testsbetweenremembered andnon-remembered combinations for every time-point of sleepcueing. Subsequently,we repeated the sliding timewindow analysis using thesame electrodes but varying time windows from the pre-sleep recall. Thisallowed us to evaluate similarity between every time-point of recall andTMR,givenanuncertaintyof±500ms.Afterwardsthedifferencesofallcombinationswere averaged across electrodes. Furthermore, two control frequencies weretested toestimate the frequencyspecificity.Oscillationsat3Hzand8Hz,whichare maximally different in phase to 5Hz (i.e. golden ratio), were utilized ascontrolfrequencies.As TMR during sleep seemed to have triggered reactivation processes in arecurrentfashion,weevaluatedwhetherthesimilaritymeasureswouldfluctuateat a certain frequency (‘TMR-spectrum’). To this end,weperformeda spectralanalysisofthetime-courseofthephasesimilaritydifferences.Weestimatedthespectralpowerforfrequenciesbetween0.25and16HzbymultiplyingahanningtapertotheFouriertransformationof thewholetrial(-0.5to3.5seconds)andevaluatedpotentialpeakfrequencies.Furthermore,thesamecontrolanalyseswithregardstolearningandoscillatorypowerasperformedfortherecallpartwereconducted.Toevaluatewhetheroursimilaritymeasuresbetweenrecall2andTMRwereinfluencedbythesimilarityinperceptualprocessing,weassessedthephase-similaritybetweenlearningandTMR,againwiththelearningdatabeingsegmentedwithregardstotheonsetofthe Dutch words (thus, before the onset of the translation). To test whetherpowerdifferencesmight bias thephase estimation,we investigated oscillatorypower for the very same contrasts as utilized in the phase analyses (i.e.oscillatory power for words remembered after ‘recall2’ was subtracted frompowervaluesof ‘TMR’andcontrastedagainstthedifferenceofpower-valuesofwords not remembered during ‘recall2’ and their equivalent during ‘TMR’). Inaddition, we tested potential differences in oscillatory power for rememberedversus non-remembered items for TMR (for a detailed overview seeSupplementaryFig.3).Powervalueswereextractedspecificallyfor5HzusingacomplexMorletwaveletof 6 cycles. For all control analyses the same time-window and electrodeselection was utilized as in the main analysis. Finally, we tested whetherpresenting an additional stimulus in theword-pair TMR conditionmight have


http://dx.doi.org/10.1101/202143

12

interferedwithon-goingreactivationprocessesasthelackofbehaviouraleffectsfor thisTMRcondition suggested suchan interpretation1.Phase-similaritywasassessed between recall andword-pair TMR in the sameway as for themainanalysis.SourceestimationTo estimate the sources of the obtained effects a virtual electrode approach,applyingtheDynamicImagingofCoherentSource(DICS)beamformingmethod7,asimplementedinFieldTrip,wasused.Aspatialfilterforeachspecifiedlocation(each grid point; 10mm3 grid) was computed based on the cross -spectraldensity, calculated for the frequency of interest (5Hz), using a complexMorletwavelet(seeabove),foralltrials(commonfilterapproach). Astimewindowsofinterest served the 1 second time-window derived from ‘recall2’ and the firstsecond of TMR, given the most extended pattern of phase similarity effectsduringthisperiod.Electrodelocationsforthe128-channelGeodesicSensorNetEEGsystemwereco-registeredtothetothesurfaceofastandardMRItemplateinMNI(MontrealNeurologicalInstitute)spaceusingthenasionandtheleftandright preauricular as fiducial landmarks without individual digitization. Astandardleadfieldwascomputedusingthestandardboundaryelementmodel8.Theforwardmodelwascreatedusingacommondipolegrid(10mm3grid)ofthegreymattervolume(derived fromtheanatomicalautomatic labelingatlas9) inMNI space, warped onto standard MRI template, leading to 1457 virtualsensors. Subsequently,dataanalysiswasaccomplishedinthesamewayonthevirtualdataasbeforeonsensorlevel. StatisticsRecallspecificphasesimilarityStatistical testing of differences in phase similarity between remembered andnon-remembered words of ‘recall1’ and ‘recall2’ was accomplished using acluster-based nonparametric permutation approach10. This approach controlstheTypeIerrorratewithregardtomultiplecomparisons(here:time,frequencyand space) by clustering neighboring sensor pairs (minimum 2 electrodes),exceedingacriticalt-valueinthesamedirection.Forallincludedfrequencybins(3-16Hz),pairedsampledt-testswerecomputedforanygivenelectrodeandforeachtime-point(rangingfrom-0.5to2.5secondswithregardtostimulusonset).Thereby,clustersofcontiguoussensorsacrossparticipantswereidentified(P<0.05, two-tailed). The cluster-level statisticwas defined from the sum of the tvaluesofthesensorsinagivencluster.Onlytheclusterwiththelargestsummedvaluewas considered and tested against the permutation distribution (Monte-Carlomethod,P <0.05, two-tailed t-test).Toestimate the time-courseand thetopographical distribution of the peak frequency, effects at 5Hz, were tested


http://dx.doi.org/10.1101/202143

13

specificallyusingthesameprocedureasaboveagainstaone-sideddistribution(controllingformultiplecomparisonsintimeandspace).PhasesimilaritybetweenrecallandTMRStatistical quantification of the phase similarity between ‘recall2’ and ‘TMR’contrasting remembered andnon-rememberedwordswas again accomplishedusingacluster-basednonparametricpermutationapproach.Initially,theaveragedifferencebetween0and3.5secondswastestedusingpairedsampledt-tests(P<0.05,two-tailed,controllingformultiplecomparisonsacrossspace).Tocorrectformultiplecomparisons500permutationsweredrawnandtheclusterwiththelargest summed t-value was tested against the permutation distribution. Toquantify the temporal characteristics of the obtained effects, phase similaritymeasures were averaged across the electrodes within the significant cluster.Paired sampled t-tests were computed for every time-point (P < 0.01, two-tailed).ClusterspecificestimationofmemoryreactivationbetweenrecallandTMRTo statistically test similarity differences between varying timewindows fromthe pre-sleep recall2 and TMR a series of post-hoc t-tests was accomplishedwithin the cluster of significant electrodes. T-values, thresholded against a P-valueof0.01(one-sided)weresummedupand500permutationsweredrawn.FollowingMichelmann and colleagues5 a distribution comprising the strongestcluster,thesecondstrongestclusteretc.wasformed.Subsequently,theobtainedclusterswere compared against a randomcluster distribution. Specifically, theclustershowingthehighestsumoft-valueswascomparedwiththedistributionof the maximum cluster, while the next cluster was compared to the secondstrongest cluster etc. Finally,P-valueswere divided by the number of clusters(bonferroni-correction).Onlinemethodsreferences1. Schreiner,T.,Lehmann,M.&Rasch,B.Nat.Commun.6,8729(2015).2. Jung,T.-P.etal.Adv.NeuralInf.Process.Syst.10,894–900(1998).3. Oostenveld,R.,Fries,P.,Maris,E.&Schoffelen,J.M.Comput.Intell.Neurosci.

2011,(2011).4. Vinck,M.,vanWingerden,M.,Womelsdorf,T.,Fries,P.&Pennartz,C.M.A.

Neuroimage51,112–122(2010).5. Michelmann,S.,Bowman,H.&Hanslmayr,S.PLoSBiol.14,e1002528

(2016).6. LACHAUX,J.-P.etal.Int.J.Bifurc.Chaos10,2429–2439(2000).7. Gross,J.etal.Proc.Natl.Acad.Sci.U.S.A.98,694–9(2001).8. Oostenveld,R.,Stegeman,D.F.,Praamstra,P.&VanOosterom,A.Clin.

Neurophysiol.114,1194–1202(2003).9. Tzourio-Mazoyer,N.etal.Neuroimage15,273–289(2002).10. Maris,E.&Oostenveld,R.J.Neurosci.Methods164,177–190(2007).


http://dx.doi.org/10.1101/202143

theta phase coordinated memory reactivation reoccurs in a ...found an early reactivation episode...

Documents