Citation
Information Propagation in In Vitro Networks of Neurons with Engineered Feedforward Structures

Material Information

Title:
Information Propagation in In Vitro Networks of Neurons with Engineered Feedforward Structures
Creator:
Alagapan, Sankaraleengam
Place of Publication:
[Gainesville, Fla.]
Florida
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (118 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Biomedical Engineering
Committee Chair:
WHEELER,BRUCE
Committee Co-Chair:
DING,MINGZHOU
Committee Members:
ORMEROD,BRANDI K
BANERJEE,ARUNAVA
Graduation Date:
8/9/2014

Subjects

Subjects / Keywords:
Action potentials ( jstor )
Brain ( jstor )
Connectivity ( jstor )
Cultured cells ( jstor )
Data transmission ( jstor )
Electrodes ( jstor )
In vitro fertilization ( jstor )
Neurons ( jstor )
Neuroscience ( jstor )
Tunnels ( jstor )
Biomedical Engineering -- Dissertations, Academic -- UF
functionalconnectivity -- informationfidelity -- microelectrodearrays -- microfluidicdevice -- neuronalnetworks -- patterning
Genre:
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Biomedical Engineering thesis, Ph.D.

Notes

Abstract:
Research in neuroscience has shown that action potentials generated by neurons underlie most brain computation and the main features of the action potentials are the rate at which they are generated and precise times at which they are generated. However the nature of propagation of these features through multiple groups of neurons is poorly understood. Feedforward networks are a well researched theoretical model for studying this propagation. The drawback of such models, though, is the loss of generalization owing to the number of assumptions. A suitable alternative for such studies is Brain-on-a-chip technology. The technology involves in vitro dissociated neuronal networks that are grown on planar multi-electrode arrays and restricted to predefined structures using different techniques. Since different regions of the network can be sampled for electrophysiological activity, it serves as a useful model to study the propagation of information through the network. The focus of this dissertation is to utilize Brain-on-a-chip technology and study how information flows through a feedforward network and what features of the neuronal network structure affect this information flow. Substrate patterning and microfabrication techniques that enable control over network structure were developed and dissociated neurons, which form random connections among themselves in the absence of any constraints were restricted to form feedforward networks. Analysis of extracellular signals recorded from these networks suggested that information is transmitted through synchronized rate changes and the reliability of transmission of information to different parts of a network is affected by structural properties of the network. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (Ph.D.)--University of Florida, 2014.
Local:
Adviser: WHEELER,BRUCE.
Local:
Co-adviser: DING,MINGZHOU.
Statement of Responsibility:
by Sankaraleengam Alagapan.

Record Information

Source Institution:
UFRGP
Rights Management:
Copyright Alagapan, Sankaraleengam. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Classification:
LD1780 2014 ( lcc )

Downloads

This item has the following downloads:


Full Text

PAGE 1

INFORMATIONPROPAGATIONININVITRONETWORKSOFNEURONSWITHENGINEEREDFEEDFORWARDSTRUCTURESBySANKARALEENGAMALAGAPANADISSERTATIONPRESENTEDTOTHEGRADUATESCHOOLOFTHEUNIVERSITYOFFLORIDAINPARTIALFULFILLMENTOFTHEREQUIREMENTSFORTHEDEGREEOFDOCTOROFPHILOSOPHYUNIVERSITYOFFLORIDA2014

PAGE 2

c2014SankaraleengamAlagapan

PAGE 3

Tomyparents,mywifeandmybrotherfortheirloveandunwaveringsupport

PAGE 4

ACKNOWLEDGMENTS IthankDr.BruceWheelerforhisvaluablesupportandguidanceduringthecourseofmystudy.AspecialthankstoDr.ThomasDeMarseforconstantlymotivatingmeandbeingasoundingboardformyideasandmylabmembersEricFrancaandDr.LiangbinPan,fortheirhelpinmyexperiments,statisticsandprovidingdataforanalysis.IalsothankmembersofmysupervisorycommitteeDr.BrandiOrmerod,Dr.MingzhouDingandDr.ArunavaBannerjeeandmembersofBrewerLabDr.GregoryBrewer,MichaelBohlerandDr.StathisLeondopulosatSouthernIllinoisUniversityfortheirvaluableinputonmywork.Lastbutnotleast,Ithankmyfriendsandfamilyforbeingsupportiveduringthecourseofmystudy. 4

PAGE 5

TABLEOFCONTENTS page ACKNOWLEDGMENTS ................................. 4 LISTOFFIGURES .................................... 7 ABSTRACT ........................................ 9 CHAPTER 1INTRODUCTIONANDBACKGROUND ..................... 11 1.1Motivation .................................... 11 1.2NeuronalAssembliesandFeedforwardNetworks ............... 12 1.2.1NeuralCoding .............................. 16 1.2.2PropagationofNeuralCodesinFeedforwardNetworks ....... 18 1.3BrainonaChip ................................. 22 1.3.1DissociatedNeuronalNetworksonMultiElectrodeArrays ...... 23 1.3.2PatterningNeurons ........................... 27 1.3.3MicrouidicDevices ........................... 29 1.4OverviewofDissertation ............................ 30 2EXPERIMENTALMETHODS ........................... 31 2.1InVitroNeuronalNetworkswithDenedTopology ............. 31 2.1.1SubstratePatterning .......................... 31 2.1.2FabricationofMicrotunnelDevices .................. 34 2.1.3CellCulture ............................... 35 2.2DataAcquisitionandPre-processing ..................... 36 2.3AnalyticalMeasures .............................. 37 2.3.1ElectrophysiologicalActivityofDissociatedNeuronalCultures ... 37 2.3.2IdentifyingRepeatingSpatio-TemporalPatternsinBursts ...... 38 2.3.3FunctionalConnectivity ........................ 39 2.3.4SpikeTrainSimilarity .......................... 42 2.3.5StatisticalAnalysis ........................... 43 3INFORMATIONTRANSMISSIONINNETWORKSWITHCONTROLLEDCONVERGENCE ..................... 44 3.1CharacterizationofNetworkActivity ..................... 48 3.2FunctionalConnectivityandNetworkConvergence ............. 48 3.3FidelityofInformationTransmission ..................... 52 3.4Discussion .................................... 57 4TRANSFEROFINFORMATIONINADIRECTIONALNETWORK ..... 61 4.1ConstructionofaTwoLayeredFeedforwardNetwork ............ 61 5

PAGE 6

4.2PropagationofBurstsinTwoLayeredNetwork ............... 66 4.3EectofNumberofConnectionsonElectrophysiologicalProperties .... 70 4.4FunctionalConnectionStrength ........................ 73 4.5FidelityofInformationTransmission ..................... 75 4.6Discussion .................................... 77 5INFORMATIONTRANSMISSIONTHROUGHMULTIPLELAYERSINANETWORK .................................. 81 5.1PropagationofBurststhroughMultipleLayers ................ 83 5.2FidelityofInformationTransmission ..................... 84 5.3EectofDisinhibition ............................. 85 5.4ConstrainingDirectionalityusingPhysicalBarrier .............. 87 5.5Discussion .................................... 89 6CONCLUSION .................................... 92 6.1GeneralDiscussion ............................... 92 6.2FutureWork ................................... 94 REFERENCES ....................................... 96 BIOGRAPHICALSKETCH ................................ 118 6

PAGE 7

LISTOFFIGURES Figure page 1-1ConceptualSchematicofaFeedforwardNetwork ................. 14 1-2Modelofsongmotorcontrol ............................. 15 1-3Illustrationofmodesofspikingactivitypropagation ................ 20 1-4PlanarMultiElectrodeArrayswithDissociatedNeuronalCulture ........ 24 1-5ElectrophysiologicActivityofDissociatedCulturesonMEAs ........... 25 2-1MicroContactPrintingProcess ........................... 33 2-2SchematicofcomputingVictor-Purpuradistancemetric ............. 43 3-1Schematicofnetworkswithdierentlevelsofconvergence ............ 45 3-2Fluorescentmicrographsofpatternednetworksoncoverslips ........... 46 3-3Planarmultielectrodearrayswithpatternedneuronalcultures. .......... 47 3-4ActivityPropertiesofNetworkswithDierentLevelsofConvergence ...... 49 3-5MeanCGCvaluesinnetworkswithdierentlevelsofconvergence. ........ 50 3-6DistributionofCGCvaluesversusdistance ..................... 51 3-7DistributionofslopesoflinesttedtodecayofCGCvalues ........... 53 3-8CGCvaluesvsdistanceinnetworkswithdierentlevelsofconvergence ..... 54 3-9Spiketrainsimilarityinnetworkswithdierentlevelsofconvergence ...... 55 3-10Distributionofpathlengths ............................. 56 3-11SpiketrainsimilarityvsPathlength ........................ 57 4-1Twochamberedmicrouidicdeviceconnectedby51tunnels ........... 62 4-2Sequentialplatingprocedure ............................. 63 4-3Delayinactionpotentialrecordedfromelectrodesundertunnel ......... 64 4-4Delayhistogramofactionpotentialsdetectedatelectrodesunderatunnel ... 65 4-5Burstinitiationintwochamberdevice ....................... 67 4-6Percentageofburstsobserved ............................ 68 4-7Exampleofidentiedclusterproles ........................ 69 7

PAGE 8

4-8Distributionofconcurrenceofactivitypatterns .................. 70 4-9Dynamicsofspikingwithinbursts ......................... 71 4-10Delayinpropagationofevokedburst ........................ 72 4-11Analysisofburstpropagationbetweentwochambers. ............... 73 4-12FunctionalconnectivityreectedbyCGCvalues .................. 74 4-13Functionalconnectivitybetweendierentregions ................. 75 4-14Spiketrainsimilaritywithinbursts ......................... 76 5-1Schematicandmicrographofinvitro4layernetwork ............... 82 5-2Burstdynamicsin4layernetwork ......................... 83 5-3Propagationofbursts ................................. 84 5-4Fidelityofinformationtransmission ......................... 85 5-5Eectofdisinhibitiononburstdynamics ...................... 87 5-6Eectofdisinhibitionondelityofinformationtransmission ........... 88 5-7Changeindelityofinformationtransmission ................... 88 5-8Burstpropagationinasequentiallyplated4layernetwork ............ 90 8

PAGE 9

AbstractofDissertationPresentedtotheGraduateSchooloftheUniversityofFloridainPartialFulllmentoftheRequirementsfortheDegreeofDoctorofPhilosophyINFORMATIONPROPAGATIONININVITRONETWORKSOFNEURONSWITHENGINEEREDFEEDFORWARDSTRUCTURESBySankaraleengamAlagapanAugust2014Chair:BruceWheelerMajor:BiomedicalEngineeringResearchinneurosciencehasshownthatactionpotentialsgeneratedbyneuronsunderliemostbraincomputationandthemainfeaturesoftheactionpotentialsaretherateatwhichtheyaregeneratedandprecisetimesatwhichtheyaregenerated.Howeverthenatureofpropagationofthesefeaturesthroughmultiplegroupsofneuronsispoorlyunderstood.Feedforwardnetworksareawellresearchedtheoreticalmodelforstudyingthispropagation.Thedrawbackofsuchmodels,though,isthelossofgeneralizationowingtothenumberofassumptions.AsuitablealternativeforsuchstudiesisBrain-on-a-chiptechnology.Thetechnologyinvolvesinvitrodissociatedneuronalnetworksthataregrownonplanarmulti-electrodearraysandrestrictedtopredenedstructuresusingdierenttechniques.Sincedierentregionsofthenetworkcanbesampledforelectrophysiologicalactivity,itservesasausefulmodeltostudythepropagationofinformationthroughthenetwork.ThefocusofthisdissertationistoutilizeBrain-on-a-chiptechnologyandstudyhowinformationowsthroughafeedforwardnetworkandwhatfeaturesoftheneuronalnetwork'sstructureaectthisinformationow.Substratepatterningandmicrofabricationtechniquesthatenablecontrolovernetworkstructureweredevelopedanddissociatedneurons,whichformrandomconnectionsamongthemselvesintheabsenceofanyconstraintswererestrictedtoformfeedforwardnetworks.Analysisofextracellularsignalsrecordedfromthesenetworkssuggestedthatinformationistransmittedthrough 9

PAGE 10

synchronizedratechangesandthereliabilityoftransmissionofinformationtodierentpartsofanetworkisaectedbystructuralpropertiesofthenetwork. 10

PAGE 11

CHAPTER1INTRODUCTIONANDBACKGROUND 1.1MotivationThebrainandcerebralcortexinparticularhasbeenasubjectofintenseinterestandresearchduringthepastfewdecades.Thecentralquestioninunderstandingbehaviorishowexternalenvironmentandcuesareencodedwithintheelectricalactivityofthebrainorinotherwordswhatistheneuralcode.Dierentmodelsandmechanismshavebeendevisedtoelucidatehowcertainenvironmentalcuesareencoded.Howeveroneaspectoftheneuralcodewhichhasn'tbeendetailedbyexperimentsisthatofpropagation-howtheinformationcontainedintheneuralcodeistransmittedtodierentregionsinnetworksofneurons.Mostoftheresearchinthisareahavebeencarriedoutusingnumericalmodels.Anissuewiththemodelingstudiesisthattheassumptionsthatunderlieamathematicalmodelseverelylimitthemodelbehaviorwithaconsequentlossofgeneralization.Insuchascenarioinvitrodissociatedculturesoflivingneuronsprovideacompellingalternative.Althoughadissociatedculturedoesnothavethesamespecicstructureastheinvivobrainarea,themannerofdevelopmentiscloselysimilarintermsofactivitypatterns( BlankenshipandFeller , 2009 ).Itcanbearguedthatinsuchnetworksthegeneticinformationthatgovernstheinitialconnectivitystructureislostandhencearandomnetworkisformed.Theadvantagehoweveristhatdierenttechniquesarenowavailablethatcanbeutilizedtobringstructuretothesenetworks.Couplingthesestructuredinvitronetworkswithaplanarmultielectrodearrayprovidesasystemwherestructurecanbecontrolledandelectricalactivityfromdierentregionsofthenetworkcanbesampled.Thissystemcanbeusedtostudytheeectaneuronalnetwork'sstructurehasonitsfunction.Thisdissertationismotivatedonsuchanidea.Usinganinvitrosystemconsistingdissociatedneuronalculturesrestrictedinstructure,thenatureofpropagationofinformationfromonesetofneuronstoanotherwherethestructural 11

PAGE 12

connectionsbetweenthesetsofneuronshavebeenpredenedandcontrolledisstudiedindetail.Havinglaidoutthemotivation,adetailedbackgroundonthesubjectunderstudyispresentedbelow. 1.2NeuronalAssembliesandFeedforwardNetworksThecerebralcortexisknowntobethecenterofdierentcomputationsincludingrepresentationsofsensoryobjects,decisionmakingandprogramsformotoracts.Ithasbeenthoughtthatthemainunderlyingfactorsthatenablethesecomputationarethehighlydistributednatureofthecorticalconnectivityandthedynamicsoftheneurons.Theconnectivityallowsdierentareastooperateinparallelandtransmitinformationbackandforththroughnumerousfeed-forwardandrecurrentconnectionswhilethedynamicsallowsforself-organizationofspatiotemporalpatternsofactivityevenwhentheunderlyinganatomicalconnectionsarexed.Theinterplayofthesetwofactorsallowsfortheintegrationofcomputationsthatoccuratspatiallysegregatedregionstogeneratecoherentperceptsandactions( Uhlhaasetal. , 2009 ).'Cellassemblies'or'neuralensembles'areconsideredtobeapossiblephysicalmanifestationofthisidea.Thisconcept,initiallyformulatedbyDonaldHebb( Hebb , 1949 ),describesacellassemblyasanetworkofmutuallyexcitingneuronsthatactivaterepeatedlywheneveraparticulartaskisperformed.Themutualexcitationallowsneuronstomaintaintheiractivityevenafterthestimulusceasesandmayserveasinputtothenextcellassembly.Thusachainofassemblieseachtriggeredbyapreviouscanbeformedandthiswasproposedtobethebasisforcomputation.Thephenomenonofsuchachainformingwastermedasphasesequence.ThesecondpartoftheideathatHebbformulatedwasthetransientstrengtheningofconnections(synapses)betweentheneuronsinacellassemblythatleadtotheaphorism"neuronsthatretogether,wiretogether".Laterthisideawasusedtoexplainthe'binding'problem.Thequestionbeingaskedwashowdoesthebrainrepresentdierentpropertiesoftheexternalenvironmentinacoherentuniedmanner.Forexampleifonewasshownawoodenredcube,howistheinformationabouttheshapeoftheobject 12

PAGE 13

integratedwithitscolorandthematerialitismadeof.Itwashypothesizedcellswhichrepresentdistinctfeaturesaredynamicallyboundbythestimulusleadingtotransientstablecellassemblieswhichasawholeenabletorepresentacomplexstimulus( Singer , 2001 ).Theconcepthasevolvedovertheyearswithevidencefortheexistenceofsuchassembliesbeingobtainedfromnumerousexperimentsthatrecordedactivityfrommultipleneurons( Buzsaki , 2004 , 2010 ; Fujisawaetal. , 2008 ; Gersteinetal. , 1989 ; Harris , 2005 ; Ikegayaetal. , 2004 ; LaurentandDavidowitz , 1994 ; Miller , 1996 ; Pastalkovaetal. , 2008 ).Thefeedforwardnetwork(FFN)showninFigure 1-1 isasimplisticmodelthatbuildsontheneuralensembleideaandhelpstounderstandthecomputationthatmaybecarriedoutinsuchnetworks.AnFFNimpliesthenetworktopologyinwhichgroupsofneuronsareconnectedinsuchawaythatactivityfromonegroupowstothenextgroupinacascadingmannerbutnotinreverse.Basedonthisidea Abeles ( 1991 )studiedwhatwouldbethedierentpropertiesofanetworkofneuronsthatcouldcarryoutcomputation.Hepostulatedthatforneuronstotransmitinformationeciently,aseriesofconvergentdivergentconnectionsbetweendierentpoolsofneuronsisrequired.Healsoshowedthatactivitywouldpropagateinasynchronousfashionandhetermedsuchnetworks'synrechains'.Themodelhasbeenfoundtobeusefulindierentcomputations( Arnoldietal. , 1999 ; Izhikevich , 2006 ; Jacquemin , 1994 ).Evidenceforsuchstructureswithinbrainthough,hasn'tbeenfoundasthesamplingofneuronsisgenerallytoosparsetondmultipleneuronsthatarephysicallyconnected.However,evidenceforfeedforwardarchitectureswithinbrainhasbeenfoundinmanycases.Oneofthewell-studiedexamplewherefeedforwardnetworksplayaprominentroleinbehaviorisinhighvocalcenter(HVC)nucleusofsongbirdbrain( Hahnloseretal. , 2002 ; LiandGreenside , 2006 ; Longetal. , 2010 ).ThesongisrepresentedasahighlystereotypedsparsesequenceofspikeburstsbyneuronsofHVCwhichprojectontoanotherareaknownastherobustnucleusofarcopallium(RA)causingtheRAneuronstoproduce 13

PAGE 14

Figure1-1. ConceptualSchematicofaFeedForwardNetwork.Thenetworkcomprisednlayerswhereblackcirclesrepresentneuronswhilethearrowsrepresentsynapticconnectionbetweentheneurons.Eachlayerhasanumberofneuronsrepresentedbyblackdots spikebursts.EachneuroninHVCisactiveduringaparticularphaseofthesongwhiletheneuronsinRAareactivedependingonwhichHVCneuronsareactiveinthe10-20mssegmentbeforeit( Feeetal. , 2004 ).ThemechanismbehindthistypeofactivitypatterenisexplainedasanexcitatoryfeedforwardnetworkbetweentheneuronsofHVCandRAasshowninFigure 1-2 .Anotherexampleforfeedforwardarchitectureinintactbrainisthevisualcortex.Informationfromretinalganglioncellsissenttolateralgenticulatenucleus(LGN)inthalamus.TheLGNprojectsmainlyontoV1regionofvisualcortexandmoseoftheextrastriatecorticalareasareactivatedbyV1.TheV1isconsideredtobeatthelowestlevelofhierarchywhereasetheotherareasreceivinginputfromV1areconsideredtobeathigherlevelinthehierarchy.Thisentirepathwayisconsideredtobeafeedforwardnetwork.Apartfromthis,anumberofotherpathwayshavebeenfoundwithinthesamesystem( Callaway , 1998 ; Callawayetal. , 2004 ; FellemanandVanEssen , 1991 ; Serreetal. , 2007 ). 14

PAGE 15

Figure1-2. Modelofsongmotorcontrol.Top(A)showstheworkinghypothesisforvocalcontrolsignalgeneration.Anatomically,HVCneuronsprojectontoRAneuronsandeachoftheHVAneuronsdrivesdierentsetsofneuronsinRA.MotorunitinegratestheactivityinRAneuronstoproducemusclecontrol.Bottom(B)showstheactivitypatternaccordingtothemodel.Duringsongvocalization,dierentsetofneuronsinHVAarechronologicallyactiveduringdierentsequencesofthesong.ThistranslatestodierentsetofneuronsinRAbeingactiveresultinginavariablemotorcontrolsignal.Reprintedbypermissionfrom( LeonardoandFee ( 2005 ))SocietyforNeurosciencec2005 15

PAGE 16

1.2.1NeuralCodingAcentralquestioninunderstandinghowthebrainprocessesinformationisthatofneuralcoding.Neuralcodereferstotheneuronalactivityfeaturesthatrelatetotangiblefeaturesofexternalenvironment.Inotherwords,neuralcodeistherepresentationandinterpretationofexternalworldinthebrain.Inadditiontothereliablerepresentationofstimulus,agoodcandidateforneuralcodeshouldservethefunctionofbeingtransformablebythebrainandbeingtransmissibletodierentpartswithoutlosingtheinformationitservestoencode( PerkelandBullock , 1968 ).Actionpotentials(spikes)arewidelyacceptedtoformanimportantpartofneuralcoding.Howevertheexactfeaturesoftheirspikingbehaviorthatisrelevantforcodinghasbeenunderdebate.Classically,therehavebeentwostrongcandidatesfortheneuralcodebasedonthehypothesizedroleofneuronasatemporalintegratororcoincidencedetector.Intheformercaseneuronsintegrateinputsfromotherneuronstemporallyandproduceresponsesproportionaltothenumberofinputactionpotentialsreceived( ShadlenandNewsome , 1994 , 1998 ; ShadlenandMovshon , 1999 ; SoftkyandKoch , 1993 )whileinthelattercaseneuronsdetecttheorderofspikesandhenceproduceresponseonlywhentheincomingactionpotentialsarecoincidentwithinashorttimewindow( AzouzandGray , 2000 ; GautraisandThorpe , 1998 ; Gray , 1999 ; Konigetal. , 1996 ).Theargumentagainsttheideaofneuronsasintegratorsorratecodinghasbeenthatthetimerequiredtoproducearesponseinsuchacasewouldbemuchlongerthanthenormallyobservedresponsetimesinsensorytaskswhiletheargumentagainstthelatteristhatcorticalneuronsarenoisyandthereforeprecisetimingoratemporalcodeisnotplausibleforneuronstocommunicatereliably.Nevertheless,evidenceforthepresenceofbothschemesofencodinghavebeenfoundinthebrain.RatecodehasbeenidentiedearlierfromworksofAdrian( Adrian , 1928 )whofoundthattheringrateofsensoryneuronincreasedwithincreasingload(stimulusintensity)tothemusclethatisinnervatedbytheneuron.Subsequentlyothershaveshownthat 16

PAGE 17

ratecodingisthepreferredencodingmodeindierentregionsofthebrain.Theneuronsinvisualcortexencodeorientationandfacialfeatureswithringrates( Aggelopoulosetal. , 2005 ; Albright , 1984 ; Baddeleyetal. , 1997 ; HubelandWiesel , 1962 , 1968 ; Hubeletal. , 1977 ; Rollsetal. , 2006 ),whilethoseinmotorcortexencodedirectionofmovement( ChapinandNicolelis , 1999 ; Georgopoulosetal. , 1986 , 1988 ; Nicolelisetal. , 1998 ),andthoseinnucleusaccumbensencodetheexpectationofareward( CarelliandDeadwyler , 1994 ; GermanandFields , 2007 ).Themostprominentexampleoftemporalcodinginthesenseofprecisetimingofspikesisintheauditorysystem(nucleuslaminaris)ofbarnowlswheretherelativedierenceinthearrivaltimeofspikesconveyinformationofthelocationoftheprey( CarrandKonishi , 1990 , 1988 ).Anotherwellstudiedsystemthatinvolvestemporalordercodingisthevisualsysteminblowieswherethedierenceinarrivaltimesofspikesfromtheperipheryandcenteroftheretinaprovidesinformationaboutthedirectionofvisualow( Higginsetal. , 2004 ; SingleandBorst , 1998 ).Temporalordercodinghasbeenthoughttobeunderlyingprocessincertainareasofmammaliancortex( Larkumetal. , 1999 ).Howeverthepreciseinformationthetemporalordercarriesaswasevidentfromthebarnowlandblowyexamplesisnotclearinthesecases.(Foradetailedreviewsee Stiefeletal. ( 2013 ))Analternateideaoftemporalcodingthatismoreprevalentisthatofstimulusencodedinthesynchronousringofneuronsanddirectlyderivesfromthecoincidencedetectorhypothesisofneuronalfunction.Theideaalsolendsitselfwelltotheconceptofcellassembliesasneuronsencodingthesamefeatureinatimevaryingstimulus(whichisgenerallythecaseinrealworldsituations)shouldreatthesametimeandwasrstproposedby Milner ( 1974 )and VonDerMalsburg ( 1985 , 1994 ).Evidencehasbeenfoundinvisualcortexattributedtobinding( Grayetal. , 1989 ; GrayandSinger , 1989 ; KreiterandSinger , 1996 ; SingerandGray , 1995 ),attentionandstimulusselection( BorgersandKopell , 2008 ; Friesetal. , 2001 , 2008 ),inolfactorysystem 17

PAGE 18

( Perez-Oriveetal. , 2002 ; Stopferetal. , 1997 ),motorcortex( Hatsopoulosetal. , 1998 ; Shmieletal. , 2006 )andinhippocampus.Synchronizationisnotrestrictedtojustspikes.Itispossiblethatspikesmaybesynchronouswithlowfrequencyoscillationsfoundinthebrain.Theplacecellsinrathippocampusprovideaclassicalexampleofthisidea( O'Keefe , 1979 ; O'KeefeandDostrovsky , 1971 ; O'KeefeandRecce , 1993 ).Thespikingoftheseneuronsislockedtothethetaoscillationinsuchawaythatthetimingofthespikerelativetothephaseencodesthelocationoftheanimalinspace.Otherstudieshaveshownthattimelockedspikingrelativetogammaoscillationwasinvolvedinencodingfeaturesofvisualstimulus( Havenithetal. , 2011 ; Vincketal. , 2010 ),workingmemory( Pesaranetal. , 2002 )aswellasmemoryformation( Jutrasetal. , 2009 ; Sederbergetal. , 2007 ).Althoughspikerateandsynchronyappeartobecontradictory,thereisaconsiderableoverlapwhentemporalcodingdictatedbycorticalrhythmsisconsidered( Ainsworthetal. , 2012 ).Therehavebeenbothexperimentswhereconcurrentchangesinrateandcorrelationsareobserved( Ahissaretal. , 2000 ; Biederlacketal. , 2006 )andtheoreticalstudieswhereratecodingandsynchronyareexplainedbyasinglemodel( MasudaandAihara , 2002 ). 1.2.2PropagationofNeuralCodesinFeedforwardNetworksMostofthestudiesmentionedintheprevioussectionhaveelucidatedhowinformationisencodedintheringpatternofneurons.Anotheressentialpropertyofaneuralcodeasdenedby PerkelandBullock ( 1968 )isthatitbetransmittedtofromsensory/motorareastosubsequentneuronswithoutlossofinformation.Therehavebeenstudiesthathavelookedathowthetwomaincodespropagatefromoneregiontoanotherwithasimplefeedforwardnetworkasthemodelofneuronalconnectivity.Activitypropagationinafeedforwardnetworkcanbecategorizedintotwomodesofpropagation-asynchronousratemodeandsynchronousspikingmode.Theformercorrespondstothemodewherespikeswithoutanycorrelationsbetweenthetimings 18

PAGE 19

arepropagatedthroughdierentlayerswithoutlosingtheoverallrateofspiking.Thelattercorrespondstothemodeinwhichspikespropagatingthroughthelayersbecomeincreasinglysynchronousi.e.,developstrongcorrelationsbetweenthetimings.ThetwomodesareillustratedinFigure 1-3 .Ithasbeenshownthatsynchronycanbetransmittedalongfeedforwardnetworks( Aertsenetal. , 1996 ; Diesmannetal. , 1999 ; Gewaltigetal. , 2001 ).Inallthestudies,pulsepackets(synchronizedspikes)wereinjectedintotherstlayerofasimulatedmodelofthefeedforwardnetworkandthepropagationwasstudiedbyvaryingparametersofthepulsepacket(numberofspikesinapacketandthevariationoftimingofthespikeswithinapacket).Itwasfoundthattherewerebothaparticularrangeofparameterswherethepulsepacketpropagatedsuccessfullyaswellasarangewherepropagationdiedafterafewlayers.Statespaceanalysisprovidedastablexedpointandasaddletheformerdenotingsuccessfulpropagationandthelatterdenotingfailedpropagation. ShadlenandNewsome ( 1998 )arguedforthetransmissionofrateinfeedforwardnetworksbasedonthevariabilityinneuronspiketimingsinmonkeyvisualcortex.TheyalsoobservedthattheratewasroughlyconstantandISIhistogramwassimilartoaPoissonprocess.Theyproposedamodelwherethebalanceofexcitationandinhibitionleadtotransmissionofrateswithoutanyinducedsynchronicitybetweenneuronsinsubsequentlayersofthefeedforwardnetwork.However,usingthesamemodelproposedbyShadlenandNewsome, Litvaketal. ( 2003 )showedthatratescouldtravelonlythroughacertainnumberoflayersbeyondwhichdierentinputratesconvergedtoavalueindependentoftheinitialringrate.Thepresenceofsharedinputsateachlayerwashypothesizedtobethereasonthateventuallyledtocorrelatedring.Inanotherstudy, vanRossumetal. ( 2002 )showedthelevelsofbackgroundnoiseinanFFNaectedthetypeofactivitythatispropagated.Atnoiseless/lowlevelbackgroundnoise,pulsepacketspropagated(onlystrongandsynchronizedstimulus)whileatmediumlevelbackgroundnoise,thestimuluspropagatedasaratewithoutanycorrelationsbetweenneuronsin 19

PAGE 20

Figure1-3. Illustrationofmodesofspikingactivitypropagationinfeedforwardnetworks.ashowsanasynchronousringrate(20spikes/second)thatisinputtherstlayerofafeedforwardnetwork.Thetoppartshowstherasterplotwhereeachdotcorrespondstoaspikeandeachlineofdotscorrespondstospikesfromasingleneuron.Inbandc,theringratepropagatesinastableaynschronousmannerthroughlayers3and6withoutanydistortion.Thisistheratepropagationmode.Indandetheringratebecomessynchronousandringpatternbecomesuniform.Thisisthesynchronypropagationmode.fandgshowhowthespikerateisaectedwhenvaryinglevelsofasynchronousspikerateisfedintothefeedforwardnetwork.Inratepropagationmode(f)thedistinctivenessoftherateismaintainedwhileinthesynchronousmode(g)thedierenceislostandallinputratesconvergetoasinglerate.ReprintedbypermissionfromMacmillanPublishersLtd[NatureReviewsNeuroscience]( Kumaretal. , 2010 ),c2010 20

PAGE 21

alayer.Moreimportantlytheyshowedthatinformationtransmissionispossibleinacombinationofratemodepropagationwithpopulationcoding.Therehasbeenevidencefortransmissionofbothrateandsynchronywhileinotherstudies,presenceofcorrelationamongspiketimestendedtoaectthepropagationofrates. Kumaretal. ( 2008 )simulatedalocallyconnectedrandomnetworkcontaining4000excitatoryand1000inhibitoryneuronstolearnwhichspikingconditionsleadtosuccesfulpropagationofapulsepacketinafeedforwardnetworkembeddedwithinsuchanetwork.Theycompared4dierentspikingbehaviors(combinationofsynchronousorasynchronousandregularorirregular)determinedbyexternalinputandrecurrentinhibition/excitationbalanceandshowedasynchronousirregularactivityinrecurrentnetworksfacilitatespropagationofbothsynchronousspikingandasynchronousringratetransferthroughsubsequentlayersinFFnetwork.Inarelatedwork( Kumaretal. , 2010 ),theystudiedtheeectofsynapticstrengthandprobabilityofformationofasynapsebetweenneuronsinsuccessivegroupsonthepropagationofrateandsynchrony.Itwasfoundthataspecicregioninparameterspaceexistedwhererateandsynchronypropagationcantakeplacereliablyinthenetworks.Incontrastwhen Mazureketal. ( 2002 )systematicallystudiedhowanensembleofneuronstransmitratevaryingsignalsinthepresenceorabsenceofweakcorrelationitwasfoundthatcorrelationstendtodestroythediscernibilityofratevariations.ExperimentalstudiesinstudyinginformationtransferinFFNshavebeenfewandfarbetween.InastudybyAlexReyes( Reyesetal. , 2003 ),afeedforwardnetworkwasimplicitlyconstructedbysendingsummedEPSCsfromsimulatedneuronstopyramidalneuronsinasliceanditerativelyfeedingtheoutputEPSCsfromthislayertosubsequentlayersofneurons.Itwasobservedthatastheactivitypropagatedthroughthelayersofthenetwork,synchronywasestablishedeventually.Whenthepropagationofaratevaryingsignalwastested,theratebecamedistortedwhenpropagatingthroughthelayerssuggestingthatthereisalimittoratepropagationandeventuallysynchronytakesover. 21

PAGE 22

Inanotherwork FeinermanandMoses ( 2006 )showedthatpopulationratecodespropagatedalongafeedforwardnetworkconstructedfromdissociatedneuronalcultures.Howeverthisstudyusedcalciumimaginginsteadofdirectspikingactivityfromneurons.Itisapparentthatalmosttheentiretyofanalysisofpropagationofneuronalcodesinfeedforwardnetworkshasbeeninnumericalmodels.Howeverthenumberofassumptionsthatunderlieamathematicalmodelseverelylimitthemodelbehaviorwithaconsequentlossofgeneralization.Insuchascenario,amodelsystemwhichcapturesallthecomplexitiesofbiologicalneuronsandnetworksofneuronswhilebeingamenabletomanipulationsinstructurewouldservetoadvanceourunderstandingofpropagationofinformationinneuronalnetworks.'Brainonachip'issuchsystemwhichmightaddresstheneedforabiologicalmodelsystemwithcontrolonstructuralfeaturesandisdesribedindetailinthenextsection. 1.3BrainonaChipResearchintoinformationprocessinginthebrainusuallyinvolvesthestudyofelectrophysiologicalsignals.Thesesignalscanberecordedfromliveanimalbrainsinvivobyimplantingmicro-wires,tetrodesandmulti-electrodearrays.Patch-clampingtechniqueshaveenabledresearcherstounderstandtheprocessesinvolovedatthecellularlevel.Howeverthereremainedaneedforanintermediatelevelofinvestigationwheretheeectsofcellularprocessescanbestudiedatanetworklevelinsmallfunctionalcellularstructures.Insuchascenarioinvitrotechniqueswhereslicesofbraintissuesweresustainedoutsidetheanimal,controlledenvironmentswerepossibledespitethemodellosinglotofcomplexityinherentintheliveanimal.Also,withtheadventoftechnologieslikemicroelectrodearrays,substratemodicationandadvancedanalysistechniques,asimplistic,smallermodelofactualbrain'Brainonachip'hasbeendeveloped.Thismodelisusefulinansweringbasicquestionsinneuroscience. 22

PAGE 23

1.3.1DissociatedNeuronalNetworksonMultiElectrodeArraysPlanarMicroelectrodeArrays(MEAs;Fig 1-4 )oerasuitableplatformtostudytheelectrophysiologicalbehaviorofnetworksofneuronsinvitro.Microelectrodearraysconsistofmetallicelectrodesplatedonanonmetallicsubstrate(glassorplastic)bylithographicprocess.Theyprovideaninterfacetomeasuretheextracellularpotentialsofneurons.Thespatialdimensionanddistributionoftheseelectrodespermitsonetomeasurefrommanyneurons,enablingtostudysmallscaleneuronalnetworks.TheearliestMEAwasdevelopedby Thomasetal. ( 1972 )andusedforrecordingfromheartcells.Later, Gross ( 1979 )and Pine ( 1980 )independentlydevelopedanMEAsystemtorecordfromasingleneuroninaneuronalculture.FromthenonMEAshavebeenvastlyusedinstudiesinvolvingbrainslicesanddissociatedprimaryculturesandprovideasuitabletechnologytorecordfrommultipleneuronsinanetworkandstudyvariousphenomenalikeplasticity,learning,andactivitypatterns.Dissociatedneuronalculturesofvirtuallyeveryneuronaltypeprocuceelectricalactivityspontaneouslywhichisaectedbyanumberoffactorslikeageoftheculture,celldensity( Kamiokaetal. , 1996 ; Wagenaaretal. , 2006 ),compositionofmediaandpresenceofotherpharmacologicalagents( CornerandRamakers , 1992 ; Espostietal. , 2007 ; VanDenPoletal. , 1996 ).TheactivityasdetectedbyaplanarMEAisintheformofisolatedspikes(actionpotentials)occuringrandomlyatdierentpartsofthenetworkintherstfewdaysafterplating.Thisisfollowedbyoccurenceofburstswhicharealsoisolatedandasynchronous.Byday10-14,theseburstssynchronizeandnetworkwideburstsareobserved(Fig 1-5 ).Theseareepisodesduringwhichmostneuronsexhibitarapid,transientincreaseinringratefollowedbyaquiescentperiodduringwhichafewneuronsresporadicallyasshowninFig 1-5 .Bythreeweekstheactivitydevelopsintopatternscharacterizedbynon-periodicsynchronizedandrepeatedactivitywhichdoesnotchangeandthusindicatesthematurationofthenetwork( Chiappaloneetal. , 2006 ; MaromandShahaf , 2002 ; Rolstonetal. , 2007 ; Sunetal. , 2010 ; VanPeltetal. , 23

PAGE 24

Figure1-4. PlanarMultiElectrodeArrayswithDissociatedNeuronalCulture.Themiddleinsetshowsan8x8arrayofelectrodeswithanetworkofneuronsontop.ThemicrographwastakenonDIV14.Thelowerinsetshowsasmallsectionofthemiddleinset. 2004 ).Theunderlyingreasonforthisevolutioninactivitypropertiesisbelievedtobetheformationofnetworkconnectionsthatarethenprunedleadingtoastablematureculturebytheendof3weeksinvitro( Ichikawaetal. , 1993 ; Muramotoetal. , 1993 ; VanHuizenetal. , 1985 ).However,theexactmechanismbehindtheburstingbehaviorisunclear.Explanationsrangefromanimbalanceinintrinsicexcitationandinhibitioninthenetwork(inhibitiondepressedandexcitationincreased);( Darbonetal. , 2002 ; Streitetal. , 2001 )andionchanneldynamics( Jimboetal. , 1993 ; Robinsonetal. , 1993 )topathologicalmanifestationofabsenceofsensoryinputs( Lathametal. , 2000 ; Martinoiaetal. , 2004 ; Wagenaaretal. , 2005b )orincompletedevelopmentduetocultureconditions( Corneretal. , 2002 , 2005 )asmanyregionsininvivonervoussystemshowsimilarburstingbehaviorduringdevelopmentwhichlaterdisappearoncethesensory 24

PAGE 25

inputsbegintoarrive( BlankenshipandFeller , 2009 ).Therehavealsobeenanumberofreportsfortheinuenceofnetworktopologyonthedynamicsofthebursts( EytanandMarom , 2006 ; Takahashietal. , 2010 ).Alternativehypothesesaboutburstsarethattheyaremanifestationsofmemoryinthenetwork( Raichmanetal. , 2006 )orcarriersofinformationinthenetworks( BeggsandPlenz , 2003 ; FeinermanandMoses , 2006 ; Pasqualeetal. , 2008 ). Figure1-5. ElectrophysiologicActivityofDissociatedCulturesonMEAs.ThetracesshowactivityrecordedfrommultipleelectrodesoftheMEA.Therectangularboxshowsasynchronousburstthatpropagatesthroughthenetwork.Thepropagationdelayisnotobviousatthistemporalresolution.Theinsetshowsisolatedactionpotentials. MEAshavebeenusefulnotonlyinrecordingthespontaneousactivitybutalsoinevokingactivitybyelectricalstimulation( Grossetal. , 1993 ; JimboandKawana , 1992 ; Maheretal. , 1999 ).Typicallythestimulusconsistsofabiphasicpulseofpre-determinedvoltageorcurrent.Theresponseofthesenetworksistypicallycharacterizedbyspikestime-lockedtostimulusthatarerepeatablewhichextendupto20msafterthe 25

PAGE 26

stimulusfollowedbyspikingpatternsthatarehighlyvariableandinsomecasesevennetworkwidebursts.Theearlyspikescomprisethedirectresponseanddenotedirectactivationofneuronsbythestimuluswhilethelaterspikesaretheindirectresponseanddenotetheactivationbysynaptictransmissionfromdirectlyactivatedneurons( Jimboetal. , 2000 ; MaromandShahaf , 2002 ; Wagenaaretal. , 2004 ).Aprominentareainwhichdissociatedculturesareusedforstudyisinlearningandmemory.Experimentsinvivoandinslicesshowedlongtermpotentiationinducedbyatetanicstimulationofneuronsinthehippocampus( BlissandLmo , 1973 ; GustafssonandWigstrom , 1986 ; SchwartzkroinandWester , 1975 ).Thestimulustypicallyconsistedofhighfrequency(20Hz)pulsesadministeredforabout10minutes.Similarresponseswereobservedincaseofdissociatedcultures( Chiappaloneetal. , 2008 ; Jimboetal. , 1999 ; TatenoandJimbo , 1999 )suggestingthemolecularmechanismsunderlyingthephenomenonarepreservedinthesecultures.Manyotherstudieshaveshownthatdissociatednetworksareaectedbylowfrequency(0.1-1Hz)electricalstimulationalso. ShahafandMarom ( 2001 )showeddissociatednetworkscanbeconditionedtoproduceadesiredresponsebystimulatingat0.3-1Hz.Othershaveshownthatthedynamicsofringisalteredafterlowfrequencystimulation( Bolognaetal. , 2010 ; Vajdaetal. , 2008 ).Amajordrawbackofdissociatedculturesistheseeminglyrandomnatureofconnectionsinthenetwork.Thoughtherehavebeendierentstudiesthathaveshowedthefunctionalconnectivityinthesenetworksaresmall-world( Downesetal. , 2012 ; Srinivasetal. , 2007 )andscale-free( Pasqualeetal. , 2008 ),thereislittlecontroloverthestructuralconnections.Thislimitstheuseoftheseculturesforthestudyofstructure-functionrelationshipsandtransferofinformation.Anumberoftechniquesthathavebeendevelopedconcurrentlyhelpovercomethislimitationandenablethecreationofnetworkswithdesiredstrucuturalconnections( GuillotinandGuillemot , 2011 ; 26

PAGE 27

WheelerandBrewer , 2010 ).UsingthesetechniquesinconjunctionwithMEAsallowsstudyingnetworkwideeectsofimposedstructureinthesecultures. 1.3.2PatterningNeuronsPatterningisabiotechniquedevelopedtocontrolthepositionofcellsinculturetherebyconstrainingtheconnectionsamongthecells.Itinvolvesthecreationofacytophilicsurface(generallyaprotein)ofdesiredpatternonacytophobicbackgroundtherebyrestrictingcelladhesiontothepattern.Thetechniquehasitsoriginsindevelopmentofbiologicallyintegrateddevices( HaddonandLamola , 1985 ; Hig-gins , 1985 ; Wadhwa , 1990 )andhasbeendevelopedwidelyoveradecade.Therearetwoaspectsofpatterning-thecreationofpatternsonsubstratesandtheprocessbywhichtheproteinsadheretothesubstrate.Thesimplestmethodofproteinadherenceisphysicaladsorptionofproteinwhichmightbedrivenbyionic,hydrophobicorvanderWaalsforces( AndradeandHlady , 1986 ; SoderquistandWalton , 1980 ).Comparatively,aproteincovalentlylinkedtothesurfaceisconsideredtobemorestableandisachievedbyusingabifunctionalcrosslinkerssuchassilanes,silica-basedlinkersthatbindtothesubstrateviaasilanolbondandtheotherfunctionalendservesasabindingsiteforaprotein( Bhatiaetal. , 1989 ; Linetal. , 1988 ; Shriver-Lakeetal. , 1997 ).Cellspreferattachingtotheseproteinsandwhenthebackgroundismadeofadierentsubstancethatpreventscellattachment,networkswithwelldenedtopologiescanbeconstructed.Theconventionalmethodforcreatingpatternshasbeenphotolithography.Aphotoresistiscoatedonthesubstrateandusingamaskwiththedesiredpattern,thephotoresistisexposedtoUVlightcausingtheremovalofphotoresistintheexposedregion.Thentheproteinisappliedtothesubstrateandtheremainingphotoresistisliftedotoleavebehindtheproteininthepattern( Coreyetal. , 1996 ; Kleinfeldetal. , 1988 ; Lometal. , 1993 ; Sorribasetal. , 2002 ).PhotochemicalmethodsinwhichUVlightinteractswithchemicalspeciesdirectlytocreateproteinattractiveregionsonthesubstratehavealsobeenused( Fodoretal. , 1991 ; Matsudaetal. , 1990 ; Sigristetal. , 27

PAGE 28

1995 ; Yanetal. , 1993 ).Alternatively,chemicalssuchasalkylsilanesoralkanethiolswhichassembleintoorganizedlayershavebeenused.Thesearetermed'selfassembledmonolayers'andpatterningisachievedbyUVirradiationofthechemicalsresultinginregionsthatcanbelinkedtofunctionallyactivemacromoleculres( Edwardsetal. , 2013 ; Hickmanetal. , 1994 ; Kiraetal. , 2009 ; Kumaretal. , 1995 ; Lopezetal. , 1993 ; Matsuzawaetal. , 2000 ; Mooneyetal. , 1996 ).Mostofthemethodsmentionedaboverequirespecializedequipmentthatcanbetediousandtimeconsumingtoproducepatternedsurfances.Microcontactprintingisanothertechniquethathasbeendevelopedtoavoidtheseissues.Asthenamesuggests,itinvolvesprintingthepatternonthesubstratebycontact.Stampscontainingthedesiredpatternsmadeofpolymerisusedtotransfertheproteinactingasinkontothesubstrate.Thetechniquewasrstdevelopedby KumarandWhitesides ( 1993 )tocreatealkanethiolpatternsongold.Adaptationsofthistechniquehasenabledtocreatepatternsonglasssurface( Branchetal. , 1998 ; Changetal. , 2003 ; Nametal. , 2006 ; Wheeleretal. , 1999 )andtherebyenablingtobeusedwithMicroelectrodeArrayswhicharegenerallymadeofglass.MicrocontactPrintinghasbecomeawidelyusedtechniquetocreatepatternednetworks( Cornishetal. , 2002 ; Nametal. , 2004 ; Oenhausseretal. , 2007 ; Ruizetal. , 2008 , 2007 ; Zengetal. , 2007 )andisusedinthisworktocreatenetworkswithcontrolledconvergence.Studieshaveshownthatpatterningaltersthestructuralpropertiescomparedtoanonpatternedrandomneuralnetwork. Wyartetal. ( 2002 )showedthatconnectivityisrestrictedtonearestneighborbypatterning. Vogtetal. ( 2005a )systematicallystudiedbranchinginnetworkspatternedintheformofagridandfoundneuritespreferredcertaindirectionscomparedtoanunrestrainednetwork. Changetal. ( 2001 )showedthatpatternednetworkshadahighernumberofactiveelectrodesaswellashigherringratecomparedtoarandomnetwork.Functionally,thesenetworksaresimilartosliceculturesintermsofactivitypatternsandsynapticplasticity( Vogtetal. , 2005b ).Mostofthe 28

PAGE 29

abovementionedstudiesusingpatternednetworksinvolvestudyingthepropertiesofsingleneuronsinsuchnetworks.Couplingthesenetworkstoplanarmicroelectrodearraysprovideaccesstoneuronalactivityfrommultipleneuronsfromdierentregionsofthenetwork( Changetal. , 2006 ; Junetal. , 2007 ; Nametal. , 2004 ).Thisinturnprovidesinsightintothefunctionalpropertiesofsuchnetworks.Recentstudiesbyourcolleagues( Boehleretal. , 2011 )andothers( Jungblutetal. , 2009 ; Marconietal. , 2012 )haveshownthatnetworkactivitypatternsvarywhenthetopologyofthenetworkisvaried.Thusthesenetworksserveasausefulplatformtostudytheeectstructureofanetworkhasonitsfunctionalproperties. 1.3.3MicrouidicDevicesAnalternatemethodforconstrainingneuronstoparticulargeometrieshasbeentheuseofmicrouidicdevices.Thesesizeofthesedevicesareatthescaleofthoseofneuronsandareconstructedusingvariousmethodssuchasphotolithography( Jimboetal. , 1993 ),softlithography( Bani-Yaghoubetal. , 2005 ; Morinetal. , 2006 ; Tayloretal. , 2003 )andothers( Ericksonetal. , 2008 ; Suzukietal. , 2004 ).Amongthese,thesoftlithographymethodhasbeenthemostpopularoneduetotheeaseandlowcostofconstructingthesedevices.IttypicallyinvolvescreatingmoldsmadeofSU-8photoresistonasiliconsubstrateusingphotolithographyandcuringpolydimethylsiloxane(PDMS)onthemoldtocreatetheactualdevices.Thesedevicescontainregionsforconstrainingcellbodiesandchannelswhosedimensionsallowonlyaxonstopassthroughthem.Microuidicdeviceshavebeenusedwidelyinanumberofapplicationswhereconstrainingneuronsisrequired( Tayloretal. , 2010 )aswellasothers(see SiaandWhitesides ( 2003 )).Aninitialareaofusewasinneuriteguidancestudieswheretheeectsofdierentpharmacologicalagentsonneuritegrowthwerestudied( Hengstetal. , 2009 ; Rivieccioetal. , 2009 ; Tayloretal. , 2005 ; Yangetal. , 2009 ).Thesedeviceshavebeenalsousedtoprobesynapsesbetweentheneurites( Botzolakisetal. , 2009 ; Tayloretal. , 2010 ; Tourovskaiaetal. , 2008 ).Itispossibletocreatesystems 29

PAGE 30

withdierentcelltypeswherecellbodiesareisolatedbutconnectedtoeachotherthroughtunnelstostudyinteractions( Dinhetal. , 2013 ; Hosmaneetal. , 2010 ; Kana-gasabapathietal. , 2011 ).Inadditiontobeingusedforisolatingneuronsindissociatedneuronalcultures,microuidicdeviceshavebeenusedtoseparateorganotypicslicesaswell( Berdichevskyetal. , 2010 ; Parketal. , 2009 ).WhensuchdevicesareusedinconjunctionwithMEAs,activitycanbesampledfrommultipleregionsallowingstudiesoffunctionalconnectivityinanetwork( Claverol-tinturetal. , 2007 ; Claverol-Tintureetal. , 2005 ; Kanagasabapathietal. , 2012 ; Panetal. , 2011 ).Thesedevicescanbeutilizedforconstrainingconnectionsinneuronalnetworksandhencecanbeutilizedtocreateamodelsystemwithrequiredcontrol. 1.4OverviewofDissertationChapter 2 providesanoverviewofthemethodsusedincreatingnetworkswithpredeterminedstructuralconnectivityandanalyzingfunctionalpropertiesofsuchnetworks.Chapter 3 detailsresultsfromexperimentsandanalysisconductedtostudytheeectofconvergenceoninformationpropagationininvitronetworkswhileChapter 4 detailsresultsfromexperimentsconductedtostudyinformationpropagationinatwolayerinvitrofeedforwardnetworkofcorticalneurons.Chapter 5 buildsonthefeedforwardnetworkmodeldevelopedinChapter 4 toa4layerfeedforwardnetworkanddetailsfeaturesofinformationpropagationfromonelayertoanotherinsuchamodel.Chapter 6 discussesthesignicanceoftheresultspresentedinthisdissertationandpossibleresearchscenariosthatcanbepursuedbasedontheresults. 30

PAGE 31

CHAPTER2EXPERIMENTALMETHODS 2.1InVitroNeuronalNetworkswithDenedTopologyThissectiondetailstheprocessesbywhichdissociatedculturesofneuronsaremadetofollowaspecicstructureofphysicalconnections.Themainmethodsthatareusedinconjunctionwithcellculturearesubstratepatterningandmicrouidicdevices. 2.1.1SubstratePatterningInbrief,PDMSstamps,castfromaSU-8moldonasiliconsubstrate,areusedtoprintapatternofpoly-d-lysineagainst3-GPSbackgroundonMEAsurfaceusingamicroaligner.Thisprocessisdiscussedindetailinthefollowingparagraphs.Planarmulti-electrodearrays(MEAs)werepurchasedfromMultiChannelSystemandconsistedof59TiN3electrodesandonegroundelectrodeembeddedinaglasssubstrate.Electrodesof30mdiameterarearrangedin10rowsof6electrodeseach,spacedatadistanceof500m.MEAsaresoakedovernightintergazymebeforethedayofmicro-contactprintingtoremoveanycellularresiduefrompreviousexperimentsandarethenwashedwithdeionizedwateroncetoremovetergazyme.Theyarethendriedandtreatedwithoxygenplasmafor5minutes.Afterplasmatreatment,theMEAsaresilanizedwith3-glycidoxypropyl-trimethoxysilane(3-GPS,Sigma-Aldrich)bysoakinginasolutionof3-GPSintoluenefor20minutesandbakedinanovenat110Cfor40minutes.3-GPSactsasabackgroundthatinhibitsadhesionorgrowthofneurons.ApatternoftheadhesionpromotingmoleculePoly-D-Lysine(PDL)(Sigma-Aldrich)ismicro-stampedontothis3-GPStreatedsurface.Forunpatternednetworks3-GPSisnotappliedandMEAsaretreatedwithoxygenplasmaandcoatedwithPDLovernightat37C.Siliconwafersarersttreatedwithhexamethyldisilazane(HMDS)whichactsasanadhesionpromoterforphotoresistfor1minute.Thena10mlayerofSU-82010photoresist(MicrochemInc.,Newton,MA)iscoatedbyspinningat500rpmfor5sfollowedby4000rpmfor40sfollowedbybakingat95Cfor4minutes.Next,using 31

PAGE 32

analignerthethinlmmaskcontainingthepatternsismadetocontactthewaferandexposedtoultra-violet(UV)lightfor28s.ThiscausespolymerizationofSU-8intheregionsexposedtoUVlight.Followingthis,thewaferisagainbakedat95Cfor5minutes.TheSU-8isthendevelopedusingSU-8developerwhichwashesawayunpolymerisedSU-8leavingbehindthepatternasamold.Themoldissilanizedwith(tridecauoro-1,1,2,2-tetrahydroocytl)-1-trichlorosilanetoenableeasiercastingofPDMS.Asinglemoldcontainsmultiplereplicatesofthesamepatternsothatmultiplestampscanbecastfromasinglemold.Sylgard184siliconeelastomerismixedwithcuringagentintheweightratioof10:1,degassedtoremoveanyairbubblesandpouredontotheSU-8moldandcuredat40Covernight.Thisenablestheelastomertopolymeriseandbecomeasolid.ThecuredPDMSispeeledoasaslabwhichisthencutintopiecescontainingasinglepattern.Toassisteasierhandling,thepiecesareattachedtoa12mmdiametercircularcoverslip(Fisher)andstoredatroomtemperature.Theprocessisshownschematicallyin 2-1 .Thestampsarerinsedrstwithacetonetoremoveanywaterinsolubleimpuritiesonthesurfacefollowedbyethanolwhichhelpswashingoanyacetonethatisleftbehind.Thisisfollowedby18Mdeionized(DI)watertowashoanyethanolfollowingwhichtheyareblown-drywithnitrogen.Thedriedstampsarethensoakedin10%sodiumdodecylsulphate(SDS)solutionfor15minutes( Changetal. , 2003 )tobindhydrophobicsitesonthePDMSandcreateathinanioniclmtobindthecationicpolylysinethroughelectrostaticforce;theyarethenrinsedwithDIwateragaintoremoveanyexcessSDSsolution.ThisstepiscrucialasanyexcessSDScausesprecipitationofPDLinthesubsequentstepsleadingtonon-uniformtransferfromstampstosurface.Thestampisthendriedagainwithnitrogenandsoakedina1:1mixtureofPDLandFITCconjugatedPoly-L-Lysine(PLL)for1hour.TheconcentrationofPDLsolutionis100g/mlwhilethatofPLLis50g/ml.FITCconjugatedPLLensuresthepatternistransferredproperlyunderaourescentmicroscope. 32

PAGE 33

Figure2-1. MicroContactPrintingProcess.TherststepinvolvesmoldcastingofPDMSstamps.Sylgardelastomermixedwithcuringagentispouredontothesiliconmoldandlefttocureat40C.Onceitiscured,thePDMSisremovedandcutintostampswithasinglepattern.ThesecondstepshowsthetreatmentofPDMSwithPoly-D-Lysine(PDL)solution.Afterthisthestampisdried(thirdstep)andmadetocontactwithMEASurfacetotransferPDLpatternontoit(stepfour).Thestampisthenremovedtoleavebehindapatternedsurface. 33

PAGE 34

MicrocontactprintinginvolvesthetransferofPDL-PLLfromthemicro-stampstotheMEAsurfaceusingacustombuiltmechanicalaligner.Silanizationwith3GPScausestheMEAstohaveahydrophobicsurface(cellspreferhydrophilicsurfacesforattachment).However3GPSisabletocrosslinkwithPDLandhenceenablesstrongadhesionofPDLtothesurfacecreatingacytophilicpatterninacytophobicbackground.ThestampswhicharecoatedwithPDL-PLLmixtureisblown-drywithnitrogenandplacedonaplatformwithaprovisionforvacuumholddownthatpreventsthestampfrommoving.TheMEAisthenmountedonavacuumchuckandplacedfacedown.Patternsgenerallyconsistedofcircularnodesconnectedbystraightlinestoenablecellbodiestoattachatnodesandneuriteattachmentinthelines.ThenodesinthestampsarealignedwiththeelectrodesintheMEAandthesurfaceofthestampismadecoplanartotheMEAsurface.TheplatformcontainingthestampisthengraduallymovedtowardstheMEAusingthealigneruntilthestampcontactstheMEAsurfacewhichenablesthetransferofPDL-PLLfromthestamptotheMEA.Thecontactismaintainedfor3minutestoensuremaximaltransfer.ArraysareimagedunderuorescencemicroscopewithaFITCdicroicforevidenceofpolylysinetransfer.Beforecellculture,sterilizationiscarriedoutbyrinsingthearrayswith70%ethanolfollowedbyrinsingwithDIwater. 2.1.2FabricationofMicrotunnelDevicesMicrotunnelDevicesaremadeofPDMSandarecastfromSU-8molds(asinthecasewithstampsinsubstratepatterning)andattachedtoMEAsurfacedirectly.TheSU-8moldfabricationfollowsasimilarprocessasthefabricationofmoldforstamps.Themaindierenceisthatitisatwolayermoldonelayerforthetunnelsandtheotherforthewells.Thefabricationprocessisasfollows.SiliconwaferistreatedwithHMDSfor1minutefollowingwhicha3mlayerofSU-82002(MicrochemInc.)isspin-coated.ThewaferwithSU-8coatingisthenbaked,exposedtoUVlightalignedwithamaskcontainingthepatternsformicrotunnels,bakedagainanddevelopedtoyieldamoldoftunnels.NextalayerofSU-82050is 34

PAGE 35

spin-coatedtoathicknessofapproximately120mandtheentireprocessofbaking,UVexposureanddevelopingisrepeatedwithasecondmaskcontainingthepatternforwells.Thewaferwithmoldissilanizedinavacuumchamberfor3hoursfromanevaporated(tridecauoro-1,1,2,2-tetrahydroocytl)-1-trichlorosilanesolutiontoalloweasierreleaseofPDMSfromthemoldduringcasting.Sylgardsiliconeelastomerbaseismixedwiththecuringagentinaratioof10:1byweight,degassedandpouredslowlyontheSU-8moldtilltheentirewaferiscovered.Itislefttocureat70Cfor2hoursonahotplate.OncethePDMSiscuredcompletely,itispeeledcarefullyfromthemold.ThepeeledPDMShasmicrotunnelstructuresinthebottomsurfacewhilethetopsurfaceisat.Holesarepunchedintheregionsmarkedbywellsusingamodiedbiopsypunch.AcircularPDMSringisattachedtothetopsurfaceandindividualdevicesarecut.Thecircularringsserveasreservoirforholdingcellculturemedia. 2.1.3CellCultureEmbryonicratcorticalneuronsharvestedatday18(E18)areplatedonpatternedMEAsormicrotunneldevicesaxedtoMEAsinmediasolution.TheintactcorticaltissueispurchasedfromBrainBitsLLCwhichshipsthetissueinamixtureofHibernateEandB27(50:1).Duringculture,thetissueistransferredtomixtureofHibernateE(BrainBitsLLC),B27(Lifetechnologies)andpapain(WorthingtonBiochemicalCorp.)ina15mltubeandusingawaterbath,shakengentlyatatemperatureof30C.Thepapainhelpsinseparationofcellsfromtheextracellularmatrix.Nextthetissueisallowedtosettleatthebottomandthesolutionisremoved.HibernateE/B27isaddedagainandthetissueistriturated10timeswithaglasspasteurpipettetodissociatethecells.Themixtureisallowedtostandfor1minutetoallowintactpiecesoftissuetosettletothebottom.Thesupernatantisthentransferredtoanother15mltube.Theprocessisrepeatedonemoretimewiththeremainingpiecesoftissueandthesupernatantiscollected,spuninacentrifugeat1000rpmfor1minute.Thecellsarenowattachedtothesidesofthetube, 35

PAGE 36

theliquidisremovedand1mlmediaisaddedandtrituratedagaintosuspendallthecellsinthissolution.Thecelldensityinthesolutioniscalculatedusingahemocytometer.Inthecaseofpatternedneuronalnetworks,NBActiv4wasusedasthemedia.InthecaseofmicrotunneldevicesamixtureofNeurobasal,Glutamax(Lifetechnologies)andB27wasused(50:0.125:1).Embryonicratcorticalneuronsharvestedatday18areplatedonpatternedMEAsormicrotunneldevicesaxedMEAsinmediasolution.TheintactcorticaltissueispurchasedfromBrainBitsLLCandstoredinamixtureofHibernateEandB27(50:1).Duringculture,thetissueistransferredtomixtureofHibernateE(BrainBitsLLC),B27(Lifetechnologies)andpapain(WorthingtonBiochemicalCorp.)ina15mltubeandusingawaterbath,shakengentlyatatemperatureof30C.Thepapainhelpsinseparationofcellsfromtheextracellularmatrix.Nextthetissueisallowedtosettleatthebottomandthesolutionisremoved.HibernateE/B27isaddedagainandthetissueistriturated10timeswithaglasspasteurpipettetodissociatethecells.Themixtureisallowedtostandfor1minutetoallowintactpiecesoftissuetosettletothebottom.Thesupernatantisthentransferredtoanother15mltube.Theprocessisrepeatedonemoretimewiththeremainingpiecesoftissueandthesupernatantiscollected,spuninacentrifugeat1000rpmfor1minute.Thecellsarenowattachedtothesidesofthetube,theliquidisremovedand1mlmediaisaddedandtrituratedagaintosuspendallthecellsinthissolution.Thecelldensityinthesolutioniscalculatedusingahemocytometer.Inthecaseofpatternedneuronalnetworks,NBActiv4wasusedasthemedia.InthecaseofmicrotunneldevicesamixtureofNeurobasal,Glutamax(Lifetechnologies)andB27wasused(50:0.125:1). 2.2DataAcquisitionandPre-processingThespontaneousactivityisacquiredusingacommercial60channelamplier-MEA1060BC(MultiChannelSystems).Thegainofthesystemis1100andthebandwidthis1Hzto3kHz.Signalsaresampledatarateof25kHzwitha12bitA/Dconverter 36

PAGE 37

(MCCard).Thedataisviewedandrecordedusingthesoftware(MCRack)providedbythehardwaremanufacturer.Stimulationiscarriedoutusingacommercialstimulusgenerator(STG2004,MultiChannelSystems,Inc.).Theblankingcircuitpresentintheamplierenablesthesuppressionofstimulationartifactintherecording(non-stimulated)electrodes.However,itintroducesaswitchingartifactwhichisremovedbyblanking5msfromthestimulusonsetduringanalysis.Also,thestimulatedelectrodewasexcludedfromanalysisduetohugestimulationartifacts.Whennecessarythestimulationprocesswasautomatedtostimulateselectedelectrodesinarandomorpredenedorderusingcustomsoftwareandhardware.Spikes(actionpotentials)aredetectedusingathresholdcrossingmethodwherethethresholdissetat5timesthestandarddeviationofnoisewhichisdenedasthetemporalsectionoftherecordedsignalwithoutanyactionpotentials.Thethresholdismonophasici.e.,spikesdetectedareeitherpositivegoingornegativegoing.Thisimpliesthatthemethodmightnotdetectspikesinchannelswhichhadbothpositivegoingandnegativegoingspikes.Inpracticehowever,thismethodwasappropriateasmostofthespikesrecordedweremonophasic.Incaseswherestimulationartifactappearedintherecordedsignal,theartifactissuppressedusingSALPAalgorithm( WagenaarandPotter , 2002 ).Thealgorithmremovesartifactbyttingacubicpolynomialateverypointandsubtractingitfromthesignal.Theblankingof5msisstillappliedassomeportionofthesignalisnotrecoverable. 2.3AnalyticalMeasures 2.3.1ElectrophysiologicalActivityofDissociatedNeuronalCulturesTheelectricalactivityobservedindissociatedculturesisgenerallycomposedofsinglespikesandbursts(groupsofspikesthatoccurtogether).Toquantitativelycharacterizetheactivityseeninthesecultures,anumberofmeasuresareused,includingmeanring 37

PAGE 38

rate,burstrate,burstdurationandpeakringrate.Thefollowingparagraphsprovidethedescriptionofthemeasureswhichareusedinlaterchapters.Meanringrateisthenumberofspikesobservedinagiventimewindowdividedbythelengthofthetimewindow.ItismeasuredinHz.Sincetheactivityobservedisnotuniformthroughoutthedurationofobservation,thismeasuredoesnotrevealtheentirenatureofactivity.Burstsaredetectedusingatime-clusteringalgorithmdevelopedby Wagenaaretal. ( 2005a ).Eachelectrodeissearchedforasequenceofatleastfourspikeswithallinter-spikeintervalslessthanathresholdsettotheelectrode'sinverseaveragespikerate(orto100mswhenthespikerateislessthan10Hz).Whensuchasequencefoundonmorethanoneelectrodeandoverlappedintime,itistermedaburst.Whenthespiketrainisbinnedsuchthatmorethanonespikefallswithinabin,theratehistogramproducedisdenedasaburstprole.Thepeakofthisproleisdenedasthepeakrate.Thestartandendoftheburstiscalculatedbyttinganexponentialtotherisingandfallingphaseoftheburstproleanddeterminingthetimecorrespondingto10%ofthepeakrate.Thedierenceintimebetweenthestartandendoftheburstisburstduration.Themeanringratewithinthedurationoftheburstisdenedasintra-burstspikerate.Inter-BurstIntervalistheaveragetimedierencebetweentheendofaburstandbeginningofnextburstandgivesanindirectestimateoftheburstrate.Theproportionofspikesdetectedatanelectrodethatareobservedtobepartofburstsisanotherusefulmeasureforunderstandingthenatureofinformationpropagation. 2.3.2IdentifyingRepeatingSpatio-TemporalPatternsinBurstsDissociatednetworkssuchasthoseusedinthisstudyshowawidevarietyinburstpatternsdependingonanumberoffactorslikedensity,age,sizeofculture( Tatenoetal. , 2002 ; Wagenaaretal. , 2006 ).However,givenaculturethatismaturethespatio-temporalpatternsobservedwithinburstshavebeenobservedtoberepeatableandpersistforlongperiods( Madhavanetal. , 2007 ; Pasqualeetal. , 2008 ; VanPelt 38

PAGE 39

etal. , 2004 ).Inordertoidentifysuchrepeatedlyoccuringspatio-temporalpatternswithinbursts,thefollowingprocedureisperformed.Burstsaredetectedasexplainedintheprevioussectionandspikesoccuringwithineachburstineachelectrodearebinnedtoproducespiketrains.Thebinsizeischosentobe5msandwindowsizeissetat500ms.Followingthis,thespiketrainsaresmoothedwitha20mslongGaussianwaveformandnormalizedtogeneratespikedensityfunction.Nextthespikedensityfunctionsareconcatenatedseriallysothatavectorcontainingncxlelementswherencisthenumberofchannelsandlisthelengthofeachspikedensityfunctionvector.Theaboveprocessisrepeatedforallthedetectedburstsandasemi-supervisedclusteringprocedureisperformedtoproduceclustersthatrepresentthedierentrepeatingpatternsinbursts.Theprocedureinvolvesanunsupervisedclusteringalgorithmsegregatingtheactivityvectorsintoclustersbasedonthenumberofclustersprovidedbytheuser. 2.3.3FunctionalConnectivityConnectivityinbraincanbeanalyzedatdierentscales.Therstisatamicroscopicscalewhichisatthelevelofsynapses.Thesecondisatanintermediatemesoscopicscalecorrespondingtotheconnectionsbetweenneuronsandthethirdatmacroscopicscale,whichisatthelevelofconnectionsbetweendierentbrainstructures( Spornsetal. , 2000 ).Analysesattheallthesescalesinvolveeitheranatomicaldataorstatisticaltechniques.Accordinglytheconnectivityunderinvestigationcanbetermedrespectivelystructuralconnectivityandfunctionalconnectivity.Structuralconnectivityreferstotheactualphysicalconnectionspresentinthenetwork.Henceanalysisofstructuralconnectivityinvolvescompleteanatomicaldataandtechniquesthatmaybeinvasivesuchasstainingandtracingornon-invasivelikediusiontensorimaging.Functionalconnectivityontheotherhandisessentiallyastatisticalmeasureandreferstothetemporalcorrelationsbetweenactivitiesfromdierentregionsofthenetwork. 39

PAGE 40

Itdenoteshowthefunctionofoneregionofthenetworkisrelatedwiththatofanotherbutdoesnotprovideanyinformationontheunderlyingstructuralconnectionsofthenetwork.Someofthecommonlyusedmeasurestodeterminefunctionalconnectivityarecrosscorrelation( AertsenandGerstein , 1985 ; Aertsenetal. , 1989 ; GersteinandPerkel , 1969 ),directedtransferfunction( Eichler , 2006 ; KaminskiandBlinowska , 1991 ),partialdirectedcoherence( SameshimaandBaccala , 1999 ; Takahashietal. , 2010 ),GrangerCausality( Cadotteetal. , 2008 ; Dingetal. , 2006 ; Fanselowetal. , 2001 ; Kisperskyetal. , 2011 )andmeasuresfrominformationtheory( Borstetal. , 1999 ; DayanandAbbott , 2001 ; Rieke , 1999 ).ScaledCorrelationisanapproachdevelopedby Nikolicetal. ( 2012 )thatisolatesthecross-correlationhistogramoffastsignalcomponentsfromslowvariationsinrate.ThemethodessentiallycomputesaPearson'srcoecientbetweenthesignalinaveryshorttimesegment(25ms).Thecorrelationcoecientsareaveragedacrossallsegmentsinatrial.Incaseswithrepeatedtrials,theseresultsareaveragedovertrials.Mathematically,itcanbeexpressedasrs=1 kKXk=1rk (2{1)whererkdenotesthecorrelationcomputedinsegmentk,k2[1...K].Kisthenumberofsegmentsinthesignalunderstudy.Theadvantageofthismethodoverconventionalcross-correlationhistogramsisthattheeectofslowvariationsinrateisremoved.ThisishighlyadvantageousincasesofsignalsfromMEAsasthespikeratewithinburstsgenerallyvariesoverarangeoffewhundredmilliseconds.Also,sinceitisthePearson'srcoecient,itisboundedbetween0and1andhenceappropriateforcomparisonbetweendierentsubjects.GrangerCausality( Granger , 1969 )isanothermeasurethathasrecentlygainedprominenceintheanalysisoffunctionalconnectivityinneuroscience( Cadotteetal. , 40

PAGE 41

2008 ; Fanselowetal. , 2001 ; Fristonetal. , 2012 ; Hesseetal. , 2003 ; Kisperskyetal. , 2011 ; KruminandShoham , 2010 ; Roebroecketal. , 2005 ; Zhouetal. , 2009 ).Itisbasedontheideathatifincorporatingthepastknowledgeofonetimeseriespermitsmoreaccuratepredictionofasecondtimeseries,therstcouldbecalledcausaltothesecond.Themathematicsunderlyingthismeasureisdescribedindetailin( Dingetal. , 2006 ).ThePairwiseGrangerCausal(PGC)valuesarearatiooftheresiduesoftwoautoregressivemodels-onethatislinearlyregressedononlythepastofatimeseries(Y)andanotherthatisregressedonthepastofthetimeseries(Y)andanothertimeseries(X)whoseinuenceonthegiventimeseries(Y)istobeestimated.Itcanbeshownthatbyextendingtheideatoincludeathirdtimeseries(Z)inboththemodelsandobtainingtheratioofresiduesagainitispossibletoeliminateanymediatingeectsZmayhaveontheinteractionbetweenXandY.ThisistermedasConditionalGrangerCausality(CGC)andisusefulineliminatingtheeectofcommonsourceandmediatingnodesintheestimationoffunctionalconnectivity.PGCandCGCvaluesaregenerallycomputedfromcontinuouswaveforms.Hencespiketrainsfromspontaneousactivityorevokedactivity,constructedbybinningthespiketimesin1msbinsaresmoothedwithanexponentiallydecayingwaveformtogenerateacontinuouswaveformwithtimeconstantof4ms( Cadotteetal. , 2008 ; Kaminskietal. , 2001 ).CGCvaluesarethencomputedbetweeneverypairofelectrodes,eachofwhichhadameanringrategreaterthan0.5Hz,conditionedontherestoftheelectrodes,usingtheGCCAtoolboxdevelopedbySeth( Seth , 2010 ).Thecriterionof0.5Hzwassettoeliminatelessactiveelectrodestherebyreducingthetimeofcomputationwithoutsignicantlyalteringthemeasuresofthefunctionalconnectivityofthenetwork.CGCvaluesweredeterminedtobestatisticallysignicantifthecorrespondingcoecientsofthemultivariateautoregressiveprocessarejointlysignicantlydierentfromzero.Thetestisthencorrectedwithafalsediscoveryratetoaccountformultiplecomparisons(p<0.001). 41

PAGE 42

2.3.4SpikeTrainSimilarityAnintuitiveapproachtoassessifinformationhasbeenreliablytransmittedfromoneneuronalpopulationtoanotheristocomparethespikingpatternsobservedfromthepopulation.Thisrequiresameasurethatcapturesthesimilarity/dissimilarityofspiketrains.AnumberofmethodshavebeendevelopedandincludeVictor-Purpuradistance( VictorandPurpura , 1996 , 1997 ),vanRossumdistance( vanRossum , 2001 ),eventsynchronization( Quirogaetal. , 2002 )andISI-andSPIKE-distance( Kreuzetal. , 2011 , 2007 ).TheVictor-Purpuradistanceisbasedontheminimumcostoftransformingonespiketraintoanotherbyinserting,deletingormovingspikes.Thecostissetas1toaddordeleteaspikewhilethecostformovingaspikefromttoalignsynchronoustoanotherspikeatt+tissetasqjtjwhereqisthescalingparameterthatcontrolsthetemporalsensitivity.Whenthedierenceinspiketimingsislessthan2=q,thecostisproportionaltothedierence.Whenitismore,deletingandinsertingaspikeislesscostly.Thusbyvaryingqitispossibletoadjustthesensitivityofthemetric.Ahighvalueofqimpliesashorttimewindowinwhichthespikeismovedtherebymakingthemetricindicativeoftighttemporalcoding.Ontheotherhand,aqsetatlowvalueresultsinthecostofmovingaspikelessthanadditionordeletiontherebymakingthemetriclesssensitivetotemporallocationandmoreindicativeofratecoding.Thusbyvaryingqitispossibletocomparedelityofinformationpropagationatdierenttemporalscales.Themeasureisboundedbydierenceofthenumberofspikesinthespiketrains(q=0)andthesumofthenumberofspikesinthespiketrains(q=1).Inthisthseis,VPdistance(whichisameasureofdissimilarity)iscomputedbetweenpairsofelectrodesfordierentvaluesofq,normalizedtobounds[0,1]andconvertedtoameasureofsimilaritybysubtractingfrom1.AschematicoftheprocedureisshowninFigure 2-2 42

PAGE 43

Figure2-2. SchematicofcomputingVictor-Purpuradistancemetric.Spiketrain1consistingofspikeslabeled1,2,3,4,5,6istransformedtospiketrain2consistingofspikesa,b,c,d,e,f.Spikes1and6arecoincidentwithaanderequiringnoedits.Spikes2,4and5aremovedbyt1,t2,t3respectively.Spike3isdeletedandspikefisinserted. 2.3.5StatisticalAnalysisStatisticalanalysisiscarriedoutusingPythonScipylibraryandR.Theexacttestfordeterminingstatisticalsignicanceischosendependingonthedistributionofmeasureunderstudy.Incaseswherethedistributionisclearlynotnormal,nonparametricapproachesareused.Generally,Kruskal-WallisonewayANOVAisusedwithaMann-WhitneyUtestaspost-hocanalysis.Toaccountformultiplecomparisons,Bonferronicorrectionisapplied.Incaseswherenormaldistributioncanbeassumed,One-wayorrepeatedmeasuresANOVAisusedwithTukeyhonestsignicancedierenceaspost-hocanalysis. 43

PAGE 44

CHAPTER3INFORMATIONTRANSMISSIONINNETWORKSWITHCONTROLLEDCONVERGENCEThebrainisknowntobefunctionallysegregatedatvariouslevelsoforganization( FellemanandVanEssen , 1991 ; Mountcastle , 1998 ; Zeki , 1993 ).Howeverforanyoftherealworldcomputationsanumberofthesesegregatedregionsworkinunisontoproducebehaviororcognition.Thisfunctionalintegrationisachievedbyvirtueofthenatureofconvergent-divergentconnectionsbetweenthedierentregions( Negyessyetal. , 2008 ; Spornsetal. , 2004 ).Sincethisintegrationalsoinvolvestheowofinformationfromoneregiontoanother,itcanbeassumedthatknowledgeoffunctionalconnectivitywouldprovideinsightintothedelityofinformationtransmissionbetweendierentregions.Functionalintegrationisgenerallymeasuredwithfunctionalconnectivitywhichismeasuredasdeviationsfromstatisticalindependenceofactivityacrossdierentregionswithoutregardtoanyunderlyingstructuralconnections( Friston , 1994 ).Recentstudieshaveshownthatthefunctionalconnectivitycomputedduringrestingstatereectstheunderlyingstructuralconnectivity( Hagmannetal. , 2008 ; Honeyetal. , 2009 , 2007 ; Pontenetal. , 2010 )suggestingthatfunctioniscontingentonunderlyingstructures.Usingsubstratepatterningtechnologies,itispossibletostudysystematicallyhowconvergence-divergencepropertiesofstructuralconnectivityaectsthemeasuredfunctionalconnectivityandinformationtransfer. Abeles ( 1991 )showedtheoreticallythatforactivitytotravelreliablyalongneuronalnetwork,convergentanddivergentconnectionsbetweeneachgroupofneuronsareanessentialstructurally.Inthisstudy,networksoflivingratcorticalneuronswithdierentlevelsofconvergencewereconstructedusingpatterningandthespontaneouslyarisingactivitywasanalyzedtounderstandhowthefunctionalconnectivityandinturnthedelityofinformationtransmissionareaectedbytheimposedstructuralconnectivity.Convergenceiscontrolledbycontrollingthenumberofnodestowhicheachnodeinthenetworkwasconnected.Thepatterned 44

PAGE 45

networkswerecomparedagainstnonpatternedrandomnetworkswheretherewerenoimposedstructuralconstraints. Figure3-1. Schematicofnetworkswithdierentlevelsofconvergence.Nodesrefertoneuronswhilearrowscorrespondtophysicalconnectionbetweennodes.L1,L2,L3refertodierentlayers.Intwodegreenetworks,(a)eachchaintransmitsinformationfromoneendtoanotherwithoutanyconnectiontootherchains.Infourdegreenetworks(b)nodesineachchainhavephysicalconnectiontoonenodeinthenextlayerwhilealsoconnectedtonodeswithinthelayer.Ineightdegreenetworks(c)nodesineachchainareconnectedtonodeswithineachlayerwhilealsoconnectedtomorethanonenodeinthenextlayer. SubstrateswerepatternedasdetailedinSection 2.1.1 .Thepatternsconsistedofagridofcircularnodesconnectedbystraightlines.Thepatternswerenamedbasedonthenumberofnearestneighborseachnodewasconnectedto.Accordingly,patternsinwhichnodeswereconnectedtotwonearestneighborsweretermedastwo-degreewhilethosethatwereconnectedtofourweretermedasfourdegreeandthoseconnectedtoeightnearestweretermedeightdegreeasshowninFigure 3-1 andFigure 3-2 .Thenodeswere50mindiameterwhilethelineswere20minwidth.Amoatofrandomlyconnectedneuronssurroundedthepatternsonallsides.Thisservedtogeneraterobustactivitywhichthenpropagatedthroughthepatternedregionofthenetwork.Theentirenetworkcanbethoughtofastwophysicallydistantpopulationofneuronsinteractingwitheachotherthroughacommunicationlayer,alsomadeofnetworkofneurons,whoseconvergenceisvariedsystematically.Theexercisenowbecomestounderstandhowdierentregions 45

PAGE 46

withinthislayercoordinatetheiractivityandhowtheconvergenceanddivergenceofphysicalconnectionsaecttheirabilitytocoordinate.Thetwodegreenetworkscanbethoughtofasparallelchainsofneuronswithnointeractionbetweentheneuronsofeachlayerconveyinginformationbetweenthetwopopulationofneuronswhilefourdegreenetworkscanbethoughtasparallelchainswithneuronsineachlayerhavingconnectionsbetweenthemselvestoreinforcetheinformationtheytransmitwithinthelayer.Theeightdegreenetworkscanbethoughtofnetworkswithconverging-divergingconnectionstoreinforcetheinformationwithinthelayeraswellasbetweenthelayers.(Figure 3-1 )FunctionalconnectivitywasmeasuredusingConditionalGrangerCausalityanddelityofinformationtransmissionwasmeasuredusingtheVictor-Purpurametric.Measuresfromgraphtheorywereusedtoassessotheraspectsoftherelationbetweenfunctionalconnectivityanddelityofinformationtransfer. Figure3-2. Fluorescentmicrographsofpatternednetworksoncoverslips.Neuronsarestainedwithcalcein.Intwodegreeandfourdegreenetworks,cellbodiesareclusteredatnodesprovidedinthepatternwhileintheeightconnectnetworks,cellbodiesarealsoclusteredintheintersectionoflines.Althoughthelinesaremostlycomposedofaxonalbundles,cellbodiescanalsobefound. 46

PAGE 47

Figure3-3. Planarmultielectrodearrayswithpatternedneuronalcultures.Theinter-electrodedistanceis500mwhiletheelectrodediameteris30m.Amoatofneuronscanbeseenatthetop,rightandleftedgeofthepicture. 47

PAGE 48

3.1CharacterizationofNetworkActivityEachnetworkbecamespontaneouslyactivegeneratingisolatedactionpotentialsintherstfewdaysthatlatercoalescedintoshortnetworkwideburstsasisgenerallyseenindissociatedneuronalcultures.Patternednetworks(twodegree,fourdegreeandeightdegree)showedsignicantdierencesrelativetorandomnetworksintermsofactivitydynamics.Meanspikerate,burstrateandintra-burstspikeratewerehigherwhileburstdurationwaslower(ForexplanationoftermsreferSection 2.3.1 )asshowninFigure 3-4 .Alsothenumberofsingleunitsisolatedfollowingspikesortingwasmoreinfourdegreeandeightdegreenetworksrelativetorandomandtwodegreenetworks.Theseresultswereconsistentwithresultsfrompreviousstudieswheretheeectofpatterningonnetworkactivitylevelswerestudied( Boehleretal. , 2011 ; Changetal. , 2006 ; Jungblutetal. , 2009 ; Marconietal. , 2012 ; Nametal. , 2004 ).Inpatternednetworks,thecellbodiespreferentiallyattachtothenodesofthepatternwhicharedesignedtobeareasoverelectrodetherebyincreasingthenumberofdetectedsingleunits. 3.2FunctionalConnectivityandNetworkConvergenceItwashypothesizedthatconvergenceinanetworkshouldinuencethesynapticstrengthbetweenneuronsinneighboringnodes.Thestrengthshouldincreasewithconvergencetherationalebeinganincreaseinconvergenceleadstoincreaseincoincidentspikingfromtheprecedinglayerstherebyenhancingthesynapticstrengthbetweenthenodeswithplasticitymechanisms.Ideally,increaseinsynapticstrengthsistestedbycomparingtheamplitudeofpostsynapticpotentials.However,studieshaveshownthatchangesinsynapticstrengthsarereectedinotherfunctionalpropertiesofthenetworklikespikerates( Jimboetal. , 1999 ; TatenoandJimbo , 1999 )andinturnrevealedbythechangesinfunctionalconnectivitymeasureslikeCross-correlograms( Chiappaloneetal. , 2008 )andConditionalGrangerCausality( Cadotteetal. , 2008 ).TotestthehypothesisConditionalGrangerCausalitywascomputedfromthespontaneousactivityofpatternedandrandomnetworksasexplainedinSection 2.3.3 .Spiketrainswere 48

PAGE 49

Figure3-4. ActivityPropertiesofNetworkswithDierentLevelsofConvergence.PanelAshowsthespikerateaveragedoverallneuronsdetectedinpatternednetworks.PatternednetworksshowahigherspikeratecomparedtoRandomnetworks.Alsospikerateishigherin4degreeand8degreenetworkscomparedto2degreenetworks.AsimilartrendisobservedinBurstRate(PanelB)andPeakSpikeRatewithinBursts(PanelD).BurstDurationisshorterinPatternedNetworkscomparedtoRandomNetworks(PanelC) constructedfrom300secondlongspontaneousactivityrecordingwithbinsizesetat10msandsmoothedwithanexponentiallydecayingwaveform40mslong.AscanbeseeninFigure 3-5 themeanCGCvaluesin4degree,8degreeishigherthan2degreeandrandomnetworks.Howeverthedierencewasnotstatisticallysignicant(p<0.05onewayKruskal-Wallis;post-hoconewayMann-Whitneytestwithbonferronicorrectionformultiplecomparison).Thepreviousanalysiscomparedfunctionalconnectionstrengthswithoutanyreferencetospatiallocationofthenodes.However,distancebetweenthenodescanalsoalsoaectfunctionalconnectionstrength.Inpairsofnodesthatarecloser,thepostsynapticneuron 49

PAGE 50

Figure3-5. MeanCGCvaluesinnetworkswithdierentlevelsofconvergence.AllCGCvaluesthatpassedthestatisticalsignicancethresholdwithFDRcorrectionwereincludedinthisgure.Errorbarsindicatethe95%condenceintervalofthemean.Themeanvalueacrossallelectrodepairswashigherin4Degreeand8Degreenetworkscomparedto2DegreeandRandomNetworks.(p<0.05onewayKruskal-Wallis;post-hoconewayMann-Whitneytestwithbonferronicorrectionformultiplecomparison) maybeelicitedtorewithinthewindowinwhichsynapticstrengthsareincreasedduetoSTDP.Asthedistancebetweennodesincrease,theeectfallsoleadingtolessersynapticstrengths.Convergenceshouldalsoplayaroleinthisdecaywithdistanceashigherconvergenceleadstomultiplepathwaysforconnectionbetweenthenodesandhencegreaterprobabilityofthepostsynapticneuronringwithinthewindow.CGCvalueswerenormalizedandplottedagainstdistancebetweenthenodes.Figure 3-6 showsthedistributionofCGCvaluesfordierentdistances.ThedistributionwasagainskewedtowardslowerCGCvaluesandnon-normal.Darkerlinescorrespondtoelectrodepairswithshortdistancesbetweenthem.Whenthedistributionsfor2Degreenetworksareconsidered,itisevidentthatthereisadierenceinthe 50

PAGE 51

Figure3-6. DistributionofCGCvaluesversusdistance.Thedarkeralinetheshorterthedistancebetweentheelectrodes.In2Degreenetworks(TopLeft)thedistributionofCGCvaluesatshorterdistancewassignicantlydierentfromthoseatlongerdistance.In4degree(TopRight)and8degree(BottomLeft)networksthedierencesinCGCdistributionsaresubtlewhiletherewasnodierenceincaseofRandomnetworks(BottomRight) 51

PAGE 52

distributionsatdierentdistanceswhichsuggeststhatthereisafall-oofconnectionstrengthswithdistance.Toestimatethisrateoffall-oabootstrapmethodwasfollowed.OneCGCvalueateachdistancewasrandomlychosenfromthedistributionandalinewastwithdistanceastheindependentvariableandCGCvaluesasthedependentvariableandtheslopeofthelinewasmeasured.ThusforeverypossiblecombinationofCGCvaluesalinecanbetandaslopecanbemeasured.Theentireprocesswasrepeatedtogenerate1000suchslopesforeachgroup.Figure 3-7 showsthedistributionofslopesmeasuredforthegroupsunderstudy.Itcanbeseenthattheslopewasthemostnegativefor2Degreenetworksandgraduallyincreasedwithincreaseinconvergencesuggestingthatthefalloofconnectionstrengthsbecamelesspronouncedandmoregradualwithincreasingconvergence.Alsothevariationinslopesdecreasedwithincreasingconvergence.Thedierenceinslopeswasstatisticallysignicant(p<0.05onewayKruskal-Wallis;post-hoconewayMann-Whitneytestwithbonferronicorrectionformultiplecomparison).Figure 3-8 showsthemeanCGCvaluesnormalizedtothemaximumCGCvalueinadishplottedversusdistanceandthecorrespondinglineartcomputedfromthemeanofthedistributionofslopesandinterceptsforeachgroup.Itcanbeseenthatthefallowassteeperin2degreenetworkscomparedto4degreeand8degreenetworksandthefalloinrandomnetworkswaslesssteepcomparedtothepatternednetworks.Thissuggeststhattheconvergenceofconnectionsaectsthefunctionalconnectivityofthenetwork. 3.3FidelityofInformationTransmissionAnumberofstudieshaveshownthattemporalstructureofspiketrainsformsthebasisofinformationprocessinginthebrain.Howeveritisnotclearhowthetemporalpropertiesofthespiketrainsarealteredastheypassthroughmultipleanatomicalstructuresoriftheunderlyinganatomicalstructurehasanyinuenceonthepropagationofthisinformation.Patternednetworksserveasusefultoolstostudyhowreliablytheinformationistransferredbetweendierentregionsofnetworkswithdierentanatomical 52

PAGE 53

Figure3-7. DistributionofslopesoflinesttedtodecayofCGCvalues.Eachpointusedingeneratingthebox-plotcomesfromalinettedtoCGCvaluesateachdistance.Theprocesswasrepeatedfor1000iterations.Theslopesarelessnegativefornetworkswithhigherconvergencecomparedtothosewithlowerconvergence.(p<0.05onewayKruskal-Wallis;post-hoconewayMann-Whitneytestwithbonferronicorrectionformultiplecomparison) structures.Victor-PurpuradistanceisusedasameasureofthedelityofinformationpropagationandiscomputedasexplainedinSection 2.3.4 .Figure 3-9 showstheVPdistanceplottedasasimilarityfordierentvaluesofq.Lowervaluesof1/qcorrespondtonarrowtemporalwindowsformovingaspikeandhenceindicativeoftemporalcodingwhilehighervaluesof1/qcorrespondtowidetemporalwindowsandindicativeofratecoding.Itcanbeseenthatthesimilarityisloweratlowervaluesof1/qthanathighervaluesforalllevelsofconvergencesuggestingthatspikeratesarebeingpropagatedmoreecientlythanprecisetemporalpatterns.Alsoastheconvergenceincreases,similarityincreasessuggestingbetterdelityoftransmissionathigherconvergencelevels. 53

PAGE 54

Figure3-8. CGCvaluesvsdistanceinnetworkswithdierentlevelsofconvergence.SquareboxesdenotethemeanCGCvaluesnormalizedtomaximumCGCvalueinadishateachdistance(errorbarsshowstandarderrorofmeanoverallpairsofelectrodesineachgroup)whilethebluelinedenotesthet. Asinthecaseoffunctionalconnectivity,itwashypothesizedthatdelityofinformationpropagationshouldbeaectedbydistancebetweenthenodes.Howevernosuchtrendwasobserved.Itwasthenhypothesizedfunctionaldistancemightinuencemorethanphysicaldistanceasthereisnocontrolovertheexactphysicalconnectionsofneurons.Onecommonlyusedmeasureoffunctionaldistanceisshortestpathlength.Ingraphtheory,shortestpathlengthreferstotheminimumnumberofedgesbetweentwonodes( Newman , 2003 ).Nodesthatareconnecteddirectlyhaveapathlengthof1 54

PAGE 55

Figure3-9. Spiketrainsimilarityinnetworkswithdierentlevelsofconvergence.Thex-axisdenotestheinverseofscalingparameterqinms.Thelowervaluesinx-axisindicatehighervaluesforqimplyingsimilarityintemporalcodingdomainwhereashighervaluesinx-axisindicatelowervaluesforqimplyingsimilarityinratecodingdomain.Thesimilarityincreaseswithdecreasingqimplyingbetterdelitywithratecodes.Thesimilarityfor2degreenetworksistheleastwhilesimilarityforrandomnetworksisthehighest.Thesimilarityfor4and8degreenetworksfallbetween2degreeandrandomwiththeformerbeinghigherthanthelatter.Statistically,2degreenetworksweresignicantlydierentfromrandomnetworks.Allotherdierenceswerenotstatisticallysignicant(p<0.05,Kruskal-WallistestwithMann-WhitneyUposthocandBonferronicorrection.) whilenodesthatareconnectedthroughoneintermediatenodehaveapathlengthof2.UndirectedGraphswereconstructedusingscaledcorrelationsbetweensortedneuronsandpathlengthbetweenallpossiblepairsofelectrodeswascomputed.Thesearegraphswherenodesdenotetheneuronsandlinksdenotefunctionalconnections.Thegraphsdonotnecessarilyreectthestructuralconnectivityi.e.,geometryimposedbypatterning.VPDistance,averagedoverallqvalues,wascomparedbetweendierentpathlengths.ThedistributionofpathlengthsisshowninFigure 3-10 .Themaximumpathlengthbetween 55

PAGE 56

anytwonodeswas4althoughtheproportionofnodepairsthathadapathlengthof4wasminimal.Nodepairswithpathlength2occurredmoreoftenthanthosewithanyotherpathlength.Almost90%ofthenodepairshadapathlengthof2orlesssuggestingthatanypairofnodesinthenetworkmostlikelyhadtwolinksbetweenthem.Inotherwords,ifoneweretotransitthroughthenodesofthefuntionalnetwork,anynodecanmostlikelybereachedfromanothernodedirectlyorthroughoneothernode. Figure3-10. Distributionofpathlengthsinpatternednetworks.Thelegendcorrespondstopathlengths.About60%ofthepairsofneuronswereconnectedatapathlengthoftwoinallgroups.Afurther30%wereconnectedatapathlengthof1.Pathlengthsof3and4werecomparativelyrarer. Itwasseenthatasthepathlengthbetweennodesincreased,thesimilaritydecreasedforallgroupssuggestingasthenumberofintermediatenodesincreasesthereisalossinthedelityoftransmission(Figure 3-11 ).ContrarytowhatwasseenindecayofCGCvalueswithdistance,convergencedidnotplayanyroleinthemannerofdecrease.Howeveratshorterpathlengths(pathlengths1and2),thehighertheconvergenceinthenetworkthehigherwasthesimilaritybetweenspiketrainsrecordedfromthenodes.Atlongerpathlengthstherewasnoobserveddierenceinsimilarity.VPdistanceforqvaluesfrom2to20wereaveragedtogiveanestimateofsimilarityinthetemporalcodingregime 56

PAGE 57

whilevaluesfrom80to150wereaveragedtogiveanestimatedintheratecodingregimetostudyiftherewasanydierencearisingduetonatureofcoding.Nodierenceintrendwasobservedotherthansimilarityvaluesbeinghigherforratecodingregime. Figure3-11. SpiketrainsimilarityvsPathlength.Figureontheleftshowsthesimilarityvaluesaveragedoverall1/qvalues.Toprightshowssimilarityvaluesaveragedover1/qvalues80to150whilebottomrightshowssimilarityaveragedover1/qvaluesof2to20. 3.4DiscussionPatternednetworksprovideameanstostudystructurefunctionrelationshipinlivingneuronalnetworks.Inthisstudy,constraintswereimposedonthelevelsofconvergenceinthenetwork.Thisdierenceinthestructureresultedindierencesinspontaneousactivityproperties.Firingratesandburstratesweresignicantlyhigherinthepatternednetworkscomparedtotherandomnetworks.Also,thedurationofburstswasshorterandringratewithintheburstswashigherrelativetorandomnetworks.Theseresultswereconsistentwithresultsfromstudiesbyourcolleaguesandothers( Boehleretal. , 2011 ; Changetal. , 2001 ; Jungblutetal. , 2009 ; Marconietal. , 2012 ).Ithasbeenshownthatthe 57

PAGE 58

enhancementinspontaneousactivitypropertiescomparedtorandomnetworksisduetoenhancementinastroglialdevelopment( Changetal. , 2006 ).Inourstudywehaveshownthatconvergenceimposedbypatterningalsoplaysaroleasevidencedbythedierencesbetweentheactivitypropertiesof2Degree,4Degreeand8Degreenetworks.Byimposingvaryinglevelsofconvergence,thenumberofphysicalpathwaysconnectingthenodeshasbeencontrolled.Thisshouldaecttheinteractionsbetweentheneuronsmanifestinginthedynamicsofactivityseeninthenetworks.Heretheeectofconvergenceofstructuralconnectionsinaneuronalnetworkonthefunctionalconnectivityandinformationtransferwithinthenetworkwasstudied.Itwasseenthattheactualstrengthsoffunctionalconnectivitywithinnetworkswithdierentlevelsofconvergencewerenotsignicantlydierentsuggestingthattheconvergenceobtainedinthesepatternednetworksdoesnothaveaninuenceinthefunctionalconnections(atleastasmeasuredbyGrangerCausality).Itispossiblethatpatterningmightnotbesucienttoprovidethecontrolneededtoaecttheconvergenceforeachneuron,inwhichcasefunctionalconnectionswouldnothavereectedtheexpectedeect.However,patterningdoesindeedprovideasignicantchangeinthenetworktopologyascanbeseenbyhowthefunctionalconnectionstrengthfellowithdistance.Theelectrodearraysusedinthisstudyhad59electrodesarrangedinanarrayof6columnsand10rows(1referenceelectrodeisoutsidethisarray)with500minter-electrodespacingandspanupto2.5mmhorizontallyand4.5mmvertically.Thusthemaximalspatialextentpossibleforneuronstoformconnectionsinthe4degreeand8degreenetworksis2500mforhorizontalpathways(6columnsseparatedby500m),1000mvertically(threerowsseparatedby500m).Themaximumdistanceacrosstheentirenetworkis3500m(cityblockdistance).Howeverthepresenceofdiagonalconnectionsin8degreenetworksallowsconnectionsof2692maswell.Themaximumdistanceatwhichcorticalneuronsinthebrainarefoundtosynapticallyconnectisaround1mm.Verticallythe4degreeand8degreenetworksshouldbenearthislimitofstructural 58

PAGE 59

connectivity.However,horizontallyanddiagonallythedistancesarefarlarger.Also,theprobabilityofneuronsformingaconnectiondecreasesasthedistanceincreasesduetothefactthethefurtheraxonsfollowalongthepathwaysthemorelikelytheywillencountersomaanddendritefromanothercell,stop,andformsynapseswiththatcell.Also,thestrengthoftheconnectionsbetweennodesshouldbeaectedbytheconvergenceateachnode.Whenthenumberofpotentialpathwaystoanodeisgreater,coincidentspikingfromtheneighboringnodesshouldenhancethesynapticstrengthbetweenthenodesduetospiketimingdependentplasticity(STDP).Itmustalsobenotedthattherearemultiplecellbodiesinanodeofthepatternandmoreoftenthannottherearecellbodiesonthelinesconnectingthenodesaswell.Hencetheactivitymeasuredatnodesmaybeactivitythathaspropagatedthroughmultiplesynapsesand,consideringtheinuenceofSTDPandtheprobabilityofconnections,itisplausiblethatfunctionalconnectionstrengthsmeasuredusingstatisticaltechniquesshowadecreasewithincreasingdistancebetweenthenodes.Thefallowithdistanceisreectedinabilityofthenetworkstotransmitinformationwithinthem.Networkswherefunctionalconnectionstrengthfallssteeplywithdistancemightnotbeeectiveintransmittinginformationandthosewithlessergradientsshouldallowmoreeectivetransmissionofinformation.Thisisreectedinourobservationsofthespiketrainsimilaritymeasure.Ofallthetopologiestested,randomnetworkshadthegreatestdelityinneuraltransmissionandwerealsogenerallylessaectedbydistancetraveledcomparedtothepatternedarchitectures.Ithasbeenshowninotherstudiesthatinvitrorandomneuronalnetworksmaysometimesdisplaysmallworldnetworkproperties( deSantos-Sierraetal. , 2014 ; Downesetal. , 2012 ),whichmayenablethemtotransmitinformationwithgreaterdelity.Patterning,thougheectiveincontrollingthestructureofnetworks,didnotprovidecontrolovertheowofinformation.Networksinwhichdirectionalitycanbecontrolledwouldhelpusunderstandhowinformationistransformedasitpassedthroughmultiple 59

PAGE 60

regions.Patterningcanbeusedinconjunctionwithothertechniqueslikechemical,electricalorphysicalgradientstoallowprecisecontroloverdirectionofconnectionbetweenneurons( Dowell-Mesnetal. , 2004 ; Rajniceketal. , 1998 ; Srensenetal. , 2007 ; ThompsonandBuettner , 2006 ; WoodandWillits , 2009 ).Alsothenetworksinthestudywerenotresponsivetoelectricalstimulation.Thispreventedfromstudyingthenetworkseectivenessincodingformultiplestimuliwhichisanotherkeypropertyininformationtransmission.Althoughthetopologiesstudiedhereseemedtobeoflimitedvalueinstudyingeectivetransmissionofinformation,dierentarchitecturescanberealizedwithrelativeeaseusingsubstratepatterningtechniques.Suitablealternativearchitecturescanbeinspiredfrombiologicalstructuresortopologiesthathavebeenstudiedindetailinotherareaslikecomputernetworkswhichmayenabletounderstandcomputationalpropertiesofneuronalnetworksandtheroleindividualneuronsplayinsuchnetworks. 60

PAGE 61

CHAPTER4TRANSFEROFINFORMATIONINADIRECTIONALNETWORKNeuronalassembliesarethoughttobetheunderlyingunitofseveraldierenttypesofcomputationinbrainincludingsensoryprocessing( EngelandSinger , 2001 ; HummelandGerlo , 2005 ),cognitiveprocesses( Buzsaki , 2010 ; Pastalkovaetal. , 2008 )andmotoroutputs( Braitenberg , 1978 ; Riehleetal. , 1997 ; Wickensetal. , 1994 ).TherehasbeenconsiderableprogressinunderstandingdierentcodingmechanismsinvolvingtheseassembliesasexplainedinSection 1.2.1 .Howevernatureofthetransmissionofcodesfromoneassemblytoanotherhasbeenstudiedpredominantlyusingmathematicalmodelsoffeedforwardnetworks(referSection 1.2.2 )duetothedicultyinaccessingmultipleregionssimultaneouslyinvivo.InvitrodissociatedneuronalnetworksgrownonmicrouidicdevicescoupledwithMEAsprovideasuitablealternativetostudypropagationofneuralcodes.Feedforwardnetworkscanbeconstructedinabottom-upapproachandpropagationcanbestudiedsystematically.Inthisstudy,asysteminwhichtwoneuronalpopulationsinteractviatunnelssmallenoughtoallowonlyaxonstogrowthroughthemwasdevelopedandthenatureofactivitypropagationbetweenthetwopopulationswasstudied.Aswasseeninthepreviousstudy,informationtransmissioniseectivewhenthereisaconvergent-divergentconnectionpatternbetweenlayersinthefeedforwardnetwork.Henceanarchitecturewhichallowsforanumberofconvergent-divergentconnectionswasdevelopedandstudied. 4.1ConstructionofaTwoLayeredFeedforwardNetworkPDMSdevicesarefabricatedasexplainedinSection 2.1.2 .Thedeviceconsistedof2chambersapproximately30mm2inareaseparatedby51tunnelsthatwere400mlong,3mhighand10mwide.Theheightofthetunnelensuredthatonlyaxonspassedthroughthemandcellbodieswererestrictedtothechambers.Figure 4-1 showsanexampleofaPDMSdeviceattachedtoanMEA. 61

PAGE 62

Figure4-1. Twochamberedmicrouidicdeviceconnectedby51tunnels.InsetsshowmicrographofthetwoneuronalpopulationsgrowingonMEA.WellAreferstochamber1andwellBreferstochamber2. Whenneuronsareculturedinbothchambers,axonsgrowthroughthetunnelsandformsynapseswithneuronsintheotherchamber.Thiscreatesasystemwherebothnetworksinuenceeachother.Tocreateasystemequivalentofthemathematicalmodelofafeedforwardnetwork,axonsshouldprojectinasingledirectionwithoutanyconnectionsintheoppositedirection.Inordertoachievethis,neuronsareculturedwithatimeintervalofabout7daysbetweenthem.Whenneuronsareculturedintherstchambertheyformconnectionsbetweenthemselvesandprojectaxonsthroughthetunnelsinsearchofotherneuronstoformconnectionswith.Sincethedimensionsofthetunnelsaremuchgreaterthanthatofaxons(usually1mindiameter),morethanoneaxonpassthroughandphysicallyllupthetunnels.Whenthesecondchamberiscultured,thenewersetofneuronsformconnectionsbetweenthemselvesaswellastheaxonsprojectedfromtherstchamber.Figure 4-2 showsaschematicoftheprocedure.Inordertovalidatetheimposeddirectionality,extracellularsignalsrecordedfromtheelectrodesunderthetunnelsandchamberswereanalysed.TheMEAusedinthisstudyhadan8x8gridofelectrodes 62

PAGE 63

Figure4-2. Schematicofsequentialplatingprocedure.Thedeviceisshownastwosquarechambersconnectedthroughtunnels.Chamber1(topsquare)isplatedrstatDay0andchamber2isplatedatDay7.Duringtheintermittenttime,neuronsinchamber1formconnectionsbetweenthemselvesandextendaxonstochamber2.Cellsplatedinsecondchamberformconnectionsbetweenthemselvesaswellasneuronsfromchamber1. 30mindiameterandseparatedby200m.ByattachingthePDMSdevicetothecenterofthearray,eachchamberhad22electrodesunderitwhile8ofthe51tunnelshadapairofelectrodesunderthem.Thenetworksbecamespontaneouslyactive10daysafterplating.Thechamberplatedrstbecameactivebeforethechamberthatwasplatedlater.Spontaneousactivitywasrecordedfor10-30minutesafterbothchambersshowedrobustspontanousactivity.Byanalysingactivityrecordedfromtheelectrodesundertunnelsaxonalpropagationofactionpotentialswasstudied.Figure 4-3 showsactionpotentialsrecordedfrompairofelectrodesunderthesametunnel.Inthiscasethelowerchamberwasplatedrstandelectrode85isclosertochamber1whileelectrode84isclosertochamber2whichwasplatedatalaterdate.Itcanbeseenthattheactionpotentialontheleftoccursatelectrode85earlierthanatelectrode84.Thisdenotesthepropagationofactionpotentialalongaxongrowingfromchamber1tochamber2whichistheintendeddirection.Howeveritcanalsobeseen 63

PAGE 64

thattheactionpotentialontherightoccursatelectrode84earlierthanatelectrode85suggestingthattheseareactionpotentialspropagatingalongadierentaxongrowingfromchamber2intochamber1. Figure4-3. Delayinactionpotentialrecordedfromelectrodesundertunnel.Theactivityisrecordedfromasingletunnelplacedoverelectrodes84and85ofthearray. Toquantifythefrequencyofoccurenceofaxonsintheintendeddirectionagainstthatintheoppositedirection,spikes(actionpotentials)weredetectedfromtherawwaveformsandsortedusingOineSorter(PlexonInc).Whenawaveformofaparticularshaperepeateditselfovertherecording,itwasassumedtobegeneratedfromthesameaxonandwasidentiedasoneunit.MultipleunitswereidentiedinasingleelectrodeaswasevidentfromFigure 4-3 .Delayhistogramswereconstructedfromspikesoccuringintwoelectrodesunderthesametunnel.Spikepairs,denedaspairsofspikesoccuringwithinatimeintervalof1msatthetwoelectrodes,weregeneratedandthedierenceintimingbetweenthemwascalculatedasthedelay.Thedistributionofdelaysthuscalculatedwasplottedforallpossiblepairsofunits.Itwasseenthatonlyafewofthepossiblepairs 64

PAGE 65

ofunitsshowedsignicantpeakssuggestingthespikeswereindeedactionpotentialspropagatingalongtheaxondetectedatthetwoelectrodesunderthetunnel.Figure 4-4 showsanexample.Topgureshowsthedistributionofdelaysbetweenallspikepairsdetectedbetweenelectrode84and85.Bottomguresshowthedelayhistogramsforpairsofunitsthatshowedsignicantpeaks. Figure4-4. Delayhistogramofactionpotentialsdetectedatelectrodesunderatunnel.Thehistogramofdelaysbetweenallspikepairsdetectedinelectrodes84and85isshownontop.Sortingthespikesinthesechannelsyielded3and4unitsrespectively.Constructingthehistogramofdelaysbetweenspikesofallpossiblepairsofunitsshowedaonetoonecorrespondencevalidatingtheideaofpropagatingactionpotential.Dependingonthepeakofthehistogram,itcanbeseentherstunitinbottomrowdenotesanaxongrowinginthereversedirectionwhilethesecondandthirdunitsdenoteaxonsgrowingintheintendeddirection. Theabovementionedanalysiswascarriedoutin6subjects.Ineachsubjecttherewere14electrodesundertunnels.Atotalof261unitswereidentiedofwhich100unitswere 65

PAGE 66

identiedinbothelectrodesunderthesametunnelscorrepondingto100uniqueaxons.Ofthese100axons,83propagatedactionpotentialsintheintendeddirection(fromchamber1tochamber2)whiletherestshowedpropagationintheoppositedirection.Fromthisitisinferredthatthedirectionofaxonalgrowthisstronglyindesireddirection.Thepreviousanalysisprovidedinsightintohowactivityfromindividualneuronspropagated.Tostudyhowtheentirepopulationofneuronsrespondedtotheimposedconstraint,networkburstswereanalysed.Burstsweredetectedusingatimeclusteringalgorithm(referSection 2.3.1 )inthechambersindependently.Burststhatoccurredinonechamberwithinatemporalwindowof500msofaburstintheotherchamberwereassumedtobeburststhatpropagatedfromthechamberinwhichtheburstwasdetectedrsttotheotherchamber.Iftheconnectionswereunidirectionalasintended,burstsshouldpropagatefromchamber1tochamber2moreoftenthanintheoppositedirection.Todeterminethis,thefrequencyofburstinitiationineachchamberwascomputed.Figure 4-5 showsthepercentageofburststhatoriginatedinChamber1(blackbar)andChamber2(whitebar).Inallcases,percentageofburstsinitiatinginchamber1washigherthanthepercentageofburstsinitiatinginchamber2.Overall,75%ofburstsinitiatedinchamber1andpropagatedtochamber2implyingthatthedirectionalitywaspredominantlyintheintendeddirection. 4.2PropagationofBurstsinTwoLayeredNetworkIntuitively,reliabletransmissionofinformationcanbethoughttoinvolvetwomaincomponents-highpercentageofsuccessfultransmissionandaccuraterepresentationofdierencesinthereceivedsignal.Inthetwolayerednetworksdescribedabove,transmissionofinformationcanbestudiedintermsofburstpropagation.Tostudysuccessfultransmissionofbursts,thepercentageofburstsobservedinbothchambersconcurrentlywascomparedwiththepercentageofburstsobservedonlyinoneofthechambers.Itwasseenthatthenumberofburstsoccuringonlyinchamber1oronlyinchamber2wassignicantlylesserthanthoseobservedinbothchambers.Thisimplies 66

PAGE 67

Figure4-5. Burstinitiationintwochamberdevice.Blackbarsshowproportionofburststhatoriginatedinchamber1andpropagatedtochamber2andwhitebarsshowproportionofburststhatoriginatedinchamber2andpropagatedtochamber1.Inallcasestheproportionofburststhatpropagatedfromchamber1tochamber2washigher. thatahighpercentageofburststhatoriginateinchamber1propagatetochamber2orviceversasuggestingatightcouplingbetweenbothchambers.Althoughthepercentageofburstsoccuringonlyinchamber2washigherthanthatoccuringonlyinchamber1,thedierencewasnotstatisticallysignicant(One-wayANOVA,p<0.01,TukeyHSDpost-hoc).Tostudythesecondaspectofinformationtransmissionnamelytheaccuraterepresentationofdierences,spatio-temporalpatternsofburstswereanalyzed.Networkwideburstsweredetectedinchambers1and2separatelyandtheclusteringproceduredescribedinSection 2.3.2 wasperformed.AnexampleofclustersidentiedinasubjectisshowninFigure 4-7 .Nextthecoincidenceoftheserepeatedactivitypatternsinthetwochamberswasstudied.Inotherwords,ifactivitypattern1wasobservedinchamber1whatisthedistributionofoccurenceofactivitypatternsinchamber2.Apredominantlyone-to-onerelationshipwasobservedbetweentheclustersofbothchambersi.e.,therewas 67

PAGE 68

Figure4-6. PercentageofburstsobservedthatoccuronlyinChamber1orChamber2orinbothchambers.Mostoftheburstswereobservedinbothwellsimplyingreliablepropagationofbursts. onespecicpatterninchamber2thatoccuredmoreoftenthanotherswhenaparticularpatternwasobservedinchamber1.IntheexampleshowninFigure 4-7 ,activitypatternsinclusters(A),(B)and(C)inchamber2occuredmoreoftenthanotheractivitypatternswhenactivitypatterninclusters(1),(2)and(3)inchamber1occuredrespectively(Figure 4-8 ).Thissuggeststhatchamber2respondstochamber1inamannerthatisdictatedbytheactivityinchamber1whichisanimportantaspectofinformationtransmission.Theaboveanalysiswascarriedouton6cultures.Anaverageof3clusterswereidentiedineachchamberindicating3patternsofactivitywithinburstsoccurredrepeatedly.In5outofthe6cultures,aone-to-onecorrespondencesimilartotheoneshowninFigure 4-8 wasobserved. 68

PAGE 69

Figure4-7. Exampleofidentiedclusterprolesinonesubject.3clusterswereidentiedineachchamber.Thenumberontherightofeachproledenotesthepercentageofburstsdetectedthatwereclusteredtogetherinagivencluster.Eachprolesisgeneratedbyaveragingthespikedensityfunctionsoftheconstituentsofthecorrespondingcluster.X-Axis-Timeinms;Y-Axis-Spikedensity.ThelayoutreectstheMEAelectrodelayout. 69

PAGE 70

Figure4-8. Distributionofconcurrenceofactivitypatterns.Thebarsrepresentthenumberoftimesanactivitypatterninchamber2wasobservedwhenagivenactivitypatternwasobservedinchamber1. 4.3EectofNumberofConnectionsonElectrophysiologicalPropertiesHavingstudiedtheinteractionbetweentwoneuronalpopulations,propertiesthataecttheinformationtransmissionwerestudiednext.Astraightforwardassumptionisthatthenumberofconnectionsbetweenthetwopopulationssignicantlyaectstheinformationtransmission.Thiscanbeachievedbyalteringthenumberoftunnelsconnectingthetwochambers.Deviceswerefabricatedwith2tunnels(4Subjects),5tunnels(5Subjects),10tunnels(5Subjects)and15tunnels(5Subjects)andthedeviceswereplatedwithneuronsinthesequentialmannerdescribedbefore.Aftertwoweeks,spontaneousnetworkactivitywasrecordedandanalysedtostudytheeectofconnectionsondynamicsoftheactivity.Therewasnosignicantdierenceobservedbetweenthegroupsintermsofmeanringrate,burstrateorburstduration.Thedierencesbetweenthechamberswerealsonotsignicant.However,inburstsdetectedinchamber2,spikeratewithinburstandpeakspikerateshowedanincreasewithincreasingnumberof 70

PAGE 71

tunnels(Figure 4-9 ).Thepeakspikerateinburstsdetectedinchamber1showedanincreasingtrendthoughnotasmarkedasthoseinchamber2.Therewasnosuchtrendobservedinintra-burstspikerate. Figure4-9. Dynamicsofspikingwithinbursts.Ashowsthespikeratewithinburstsdetectedinbothchambers.Thereisanincreasewithincreasingnumberoftunnelsinchamber2whereasthereisnosuchtrendinchamber1.Bshowsthepeakspikeratewithinbursts.Againthereisanincreasewithincreasingnumberoftunnelsinchamber2.Inthismeasure,chamber1alsoshowsanincreasingtrend. Anothersignicantobservationfromthespontaneousactivitywasthattheburstpropagationwasaectedbythenumberoftunnels.Thepercentageofburststhatpropagatedsuccessfullyfromchamber1tochamber2decreasedwithdecreasingnumberoftunnels.Alsothedelayinthepropagationofburstsincreasedwithincreasingnumberoftunnels.Thistrendwasobservedwhenthenetworkinchamber1waselectricallystimulatedaswell.Shortbiphasicelectricpulseswereappliedtoanelectrodeinchamber1toevokeburstsinthatchamberandafterashortdelay,oftenproducedenoughactivitytransmittedthroughthetunnelstoinitiateaburstofactivityinchamber2.Toquantifythetemporaldelaybetweenanevokedburstinchamber1andappearanceofaburstinchamber2thedelaywasestimatedbasedonthetimebetweenpeakringinchamber1versuschamber2(showninFigure 4-10 ).Thepeakringratewaschosenas 71

PAGE 72

thiswasfeatureoftheburstthatisreliableandrobusttodeterminecomparedtostarttimeswhichmaybeconfoundedbythedirectresponsesofstimulation.Decreasingthenumberoftunnelsthatconnectedeachchamberledtosignicantincreasesinthetimedelaybetweenbursts.Withtwotunnelsthedurationofdelaysbetweenpeakringwasapproximately300msanddeclinedmonotonicallywiththeincreasingnumberoftunnelstolessthan100msforfty-onetunnels(Figure 4-11 left). Figure4-10. Delayinpropagationofevokedburstsintwolayerednetworkswith51,15and5tunnels.TheplotsshowspikedensityfunctiongeneratedbybinningthespikesinaburstevokedbyelectricstimulationandsmoothingitwithaGaussianfunction.Itcanbeseenthatthereisdelayinthetimetoreachpeakringrateinallcases.Thepeakofthisfunctionisdetectedforeachchamberandthedierenceiscalculatedasthedelayinpropagation. Theprobabilityofactivityevokedinchamber1topropagatetochamber2wasalsoaectedbythenumberoftunnelsthatconnectedeachchamber.Wecalculatedthepercentageoftrialsinwhichactivityevokedinchamber1alsoproducedaburstofactivityinchamber2andplottedinFigure 4-11 (right).Thepercentageofevokedglobalburstsrepresentsasimplemeasureofhoweectivetheconnectionsmaybeatconductingbursts 72

PAGE 73

acrosschambers.Thispercentageincreasedrapidlyandmonotonicallyasthenumberoftunnelsincreasedfrom2(20%)to5(50%)to10(80%).Onlysmallchangeswereseenastunnelnumberchangedfrom10to15to51.Interestingly,eventwotunnels,carryingapproximately20axons,wereabletopropagatebursts,evenifimperfectly(approximately20%ofthetime),acrossnetworks. Figure4-11. Analysisofburstpropagationbetweentwochambers.(Left)Delaysofburstsbetweenchamber1andchamberasafunctionofthenumberoftunnels.Thedotsrepresentthemeandelaysforeachcase.Thebarsindicate1.96standarderrorofthemeanfor95%condencelimits.(Right)Percentagesofglobalburstsasafunctionofthenumberoftunnels.Thedotsrepresentthemeanpercentagesforeachcase.Thebarsindicate1.96standarderrorofthemeanfor95%condencelimits. 4.4FunctionalConnectionStrengthTounderstandhowthemanipulationinnumberoftunnelsaectedthefunctionalconnectionsbetweenthetwopopulations,weusedConditionalGrangerCausalityasameasureoffunctionalconnectivity,computedbetweentheneuralspiketrainsfrompairsofelectrodesinchamber1andchamber2.Figure 4-12 representsnetworkgraphsforthe5,10,and51tunnelgroupswhosenodes(electrodes)areclusteredaccordingtotheGrangercausalweightsconnectingeachnode(Force-DirectedAtlas).Inthislayout, 73

PAGE 74

thenodesincreaseinproximity(i.e.drawtogether)astheGranger-causalestimateoffunctionalstrengthbetweenthosenodesincreases.ThewidthoftheedgesdenotestheGranger-causalstrengthbetweenthosenodes.Electrodesunderchamber1aredepictedbyblacknodesandthoseunderchamber2ingray.Electrodesthatwerelocatedalongthetunnelswereremovedtoimproveclarity.IneachgrouptheGranger-causalstrengthsamongnodeswithinachamberdrawthosenodestogetherformingclustersforeachchamber.Howeverasthenumberoftunnelsincreaseeachclusterappearstodrawtogetherfromthe5tunnelgroupinwhichclustersareclearlyseparatedtothe15andnally51tunnelgroupsinwhichtheclustersappeartomerge. Figure4-12. FunctionalconnectivityreectedbyCGCvaluesinthreerepresentativesubjectsofcorrespondinggroups.Thenodesdenotetheelectrodeandarecoloredaccordingtothelocation(Reddenoteselectrodesinchamber1andGreeninchamber2).Thenetworkisdisplayedusingaforce-directedmethodtoshowthedierenceinthefunctionalconnectivitybetweendierenttypesofnetworks( Bastianetal. , 2009 ) Figure 4-13 (Left)portraysthestrengthoftherelationshipbetweenchambersasthepercentageoftotalconnectionsamongelectrodesinchamber1thatshareconnectionswithnodesinchamber2.Asthenumberoftunnelsincreases,thefractionofnodesinchamber1withsignicantGranger-causallinkstonodesinchamber2approaches50%,implyingthat,with51ormoretunnels,theneuronsinchamber1areequallywellconnectedtoneuronsinchamber1astochamber2(p=0.0029).CGCwasusedwiththesignalsfrom 74

PAGE 75

pairsofelectrodeswithintunnelstodetectthedirectionofpropagation.AsshowninFigure 4-13 (right),thereisastrongbiasintheforwarddirection(fromchamber1towardchamber2;p<0.05).Thiswasconsistentwithouttheearlierresultsforstaggeredculturesshowing83%directionalpreferenceasindicatedbytimedelaysofindividualactionpotentialwaveforms. Figure4-13. Functionalconnectivitybetweendierentregions.(Left)Thepercentageoffunctionalconnectionsfromelectrodesinchamber1toelectrodesinchamber2amongallthefunctionalconnectionsinvolvingelectrodesinchamber1.(Right)ThemeannormalizedCGCpercentagesbetweenpairsofelectrodesundertunnels.Theblackbarrepresentsthemeanvaluesinthedirectionfromchamber1tochamber2whilethewhitebarrepresentsthemeanvaluesintheoppositedirection.Valuesfromchamber1tochamber2arehigherthanthoseintheoppositedirection. 4.5FidelityofInformationTransmissionSpiketrainsimilaritywascomputedusingVictor-Purpurametricbetweenallpairsofelectrodesinchamber1andchamber2.Onlythespikesoccuringwithinburststhatweredetectedtohavepropagatedfromchamber1tochamber2wasincludedtostudythedelityofinformationtransmissionandhowthenumberofconnectionsbetweenthelayersaectsthisdelitymeasure.Thecostparameterqwassetfrom1/2to1/150correspondingto2msto150mswindowrespectively.Asexplainedearlier,highervalues 75

PAGE 76

ofqpenalizemovingspikesandhencecorrespondtosimilarityvaluesintighttemporalcodingregimewhereaslowervaluesofqallowliberalwindowswithinwhichspikescanbemovedandhencecorrespondtosimilarityinratecodingregime.Figure 4-14 showsthesimilaritymetricfordierentvaluesofq. Figure4-14. Spiketrainsimilaritywithinbursts.Similarityincreaseswithincreasingnumberoftunnelsandthisrelationisconservedinboththetemporalcodingregimeandratecodingregime.Thegureonleftshowsthesimilarityvaluesforqrangingfrom1/2to1/50.Thegureontherightisapartofthegureontheleftexpandedintherange1/2to1/20.Theinsetsshowthesimilarityvaluesaveragedoverallqvalues. Itcanbeseenthatthesimilarityvaluesarelowerforhigherqvalues(lowervaluesof1/q)comparedtothoseforlowerqvalues(highervaluesof1/q).Thissuggeststhatthetimingbetweenspikeswithintheburstsaren'tconservedreliablywhereastheoverallnumberofspikesseemstobeconservedreliably.Also,thesimilarityvaluesincreasewithincreasingnumberoftunnelssuggestingthathigherthenumberofconnectionsbetweenthelayersbetteristhedelityoftransmissionofinformation.Thisrelationisconservedacrosstheentirerangeofthecostparameter. 76

PAGE 77

4.6DiscussionInthisstudy,asystemoftwointerconnectednetworkswithadominantdirectionofactivityowwasconstructedusingacombinationofmicrouidicdevicecontainingtwochambersandtunnelsandtemporallystaggeredcellplatingprocedure.Thissetupcanbeconsideredtobeatwo-layeredfeedforwardnetworkandinformationowbetweenthetwosub-populationscanbestudied.Thisisincontrasttothestudyofinformationowatsinglecell/fewcellslevelinChapter 3 .Thedirectionalitywasveriedbyanalyzingtheactivityinthetunnelsandinthechambers.Intheinitialexperiment,itwasfound51tunnelsbetweenthechambersprovidedenoughconnectivitybetweenthetwonetworkstoreliablyandrobustlytransmitbursts.Giventhecross-sectionalareaofthetunnel(3x10m)andaxondiameter(1m),itcanbesafelyassumedthereare25axonsinatunnel.Thisgivesanestimatethat1250axonscommunicatebetweenthetwochambers.Thenumberofcellsduringplatingwassetat60,000.Thisimpliesthat2%connectivitybetweenthetwonetworksissucienttoreliablytransmitburstactivity.Dissociatednetworkssuchasthoseusedinthisstudyshowawidevarietyinburstpatternsdependingonanumberoffactorslikedensity,age,sizeofculture( Tatenoetal. , 2002 ; Wagenaaretal. , 2006 ).However,givenaculturethatismaturethespatio-temporalpatternsobservedwithinburstshavebeenobservedtoberepeatableandpersistforlongperiods( Madhavanetal. , 2007 ; Pasqualeetal. , 2008 ; VanPeltetal. , 2004 ).Also,burstsareconsideredtobethemainmechanismofinformationtransmissioninsuchnetworks( FeinermanandMoses , 2006 ).Theone-to-onecorrespondencebetweenspikingactivitypatternsinthetwochamberssuggestthattheactivityfromtherstlayeraectsthedynamicsofthesecondlayertoelicituniqueactivitypatterns.Ifoneweretoassumethattheactivityinrstlayeractsasjustaninitiatorofactivityinthesecondlayer,theone-to-onecorrespondenceobservedinspikingactivitypatternswouldnotbeaspronounced.Alsothespikingwithinburstsshouldbeindependentofthenumberoftunnels.Howeverthiswasnotthecasestronglysuggestinga 77

PAGE 78

morecomplexinteractionbetweenthetwonetworks.Theincreasesinspikerateandpeakspikeratewithinburstswithincreasingnumberoftunnelsseemtosuggestthatincreasingthenumberofconnectionsdrivesthesecondnetworkwithincreasingstrengthcausingneuronstoremoreoften.Thenumberoftunnelsbetweenthetwochamberssignicantlyaectsthestrengthofcouplingbetweenthenetworksasevidentfromthepercentageofburststhatsuccessfullypropagatedbetweenthelayers.Thiseectisbothintuitiveandstraightforward.Asthenumberofconnnectionsincreasestheprobabilityofburstpropagationincreases.Howevertheinterestingaspectisthenatureofthiseect.Evenwith2tunnels(50neurons)burstspropagatedsuccessfullyaround20%ofthetimeimplyingaconnectivityof<0.1%beingsucientforcommunication.Thecaveatinsuchacalculationisthateachaxonmayformsynapseswithmultipleneuronsintheotherlayer.Alsothepercentageofsuccessfultransmissionincreasesquicklyreaching80%withjust10tunnels.Asimilartrendisobservedinthedelayofburstpropagation.Thedelaymaybeexplainedintermsofthetimetakenforrecruitmentofneuronsingeneratinganetworkburst.Withfewerconnections,thenumberofneuronsinthesecondlayerthataredirectlydrivenbyneuronsintherstlayerislessresultinginextratimetakentoexcitesucientnumberofneuronstoresultinarunawayexcitation.Thisnon-linear,almostlog-logrelationshipisconsistentwithalotofphenomenonobservedinneuronalactionpotentialactivity( BuzsakiandMizuseki , 2014 ).ByusingGrangerCausalityasameasureoffunctionalconnectivitybetweenthetwosub-networksitwaspossibletocapturetheeectnumberoftunnelshasonthefunctionalconnectivity.Also,beinganasymmetricmeasure,thedierenceinGrangerCausalvaluesbetweenthesamepairofelectrodesundertunnelspermittedthevalidationofthedirectionofaxonalactionpotentialpropagationinthetunnelsofthesenetworks( Panetal. , 2011 ).Thediversityinspatiotemporalpatternsthatariseinsuchdissociatedcultureshasbeenattributedtothevariedconnectivityarchitectureinthenetwork( Volmanetal. , 2005 ). 78

PAGE 79

Thusthedierenceinspatiotemporalpatternsobservedcanbeattributedtothedierenceinunderlyingfunctionalnetwork.Alsowhendierentsitesofdissociatedculturearestimulated,thepatternofactivitygeneratedisdierentduetothisinherentdierence( Panetal. , 2009a ; Segevetal. , 2004 ).ByusingGrangerCausalityontheresponsesevokedbythestimulusonasingleelectrodetodeterminefunctionalconnectivity,weareobtainingalimitedrepresentationofthefunctionalconnectivitiesinthenetwork.Itisremarkablethateveninthislimitedrepresentationthedierenceinnumberoffunctionalconnectionsbetweenthenetworksisstarkasshownintheincreasingpercentageofconnectionsfromchamber1tochamber2withincreasingnumberoftunnels.ThedierenceinfunctionalconnectionsbetweenthepopulationsisreectedinthedelityofinformationtransmissionasmeasuredbyVictor-Purpurametric.Spikesimilaritiestendedtobehigherwhenthecostparameterwaslowimplyingthatthenumberofspikeswithinburstsremainsrelativelyconstantacrossmultipleelectrodes.Theunderlyingmechanismmaybeoneofeectivecommunicationofratesacrosstheentirenetwork.Analternatehypothesiswouldbethattheratesobservedateachelectrodemaybetheeectoflocalconnectivityintheregionwithoutanyrelevantinformationbeingpassedon.Bymanipulatingthenumberoftunnels,itispossibletocomparetheabovehypotheses.Ifthesecondhypothesisweretrue,thereshouldnothavebeenanydierencesbetweengroupsinthesimilarityobservedbetweenelectrodesinlayer1andlayer2asthelocalconnectivitycanbeassumedtobefairlyuniformacrossgroups.Supportingevidenceforthisassumptionisthelackofdierenceinthespikerate,burstrateandburstdurationinbetweenthetwolayers.Thatthesimilarityincreasedwithincreasingnumberofconnectionssupportsthehypothesisthatratesareeectivelycommunicatedthroughoutthenetwork.TheincreaseinsimilaritywithincreasingnumberofconnectionswasconservedathighervaluesofthecostparameterinVictor-Purpurametric.Thisimpliesthateventhoughspiketimingisnotaseectivelytransmittedacrossthelayersasrate,increasing 79

PAGE 80

thenumberofconnectionsdoesinuencetheeectivenessofspiketimingtransmission.Thisresultsuggeststhattheunderlyingmechanismoftransmissionofinformationinsuchnetworksisnotpurelyaratebasedorprecisetimingbasedashasbeenshownindierentnumericalmodelingstudiesbutonethatisacombinationofboth.Recentstudiesbothinmodeling( Kumaretal. , 2010 )andinvivoexperiments( Ainsworthetal. , 2012 )supportthisidea.Thetwolayernetworkandanalysesdescribedaboveshowhowdierenthypothesescanbetestedinasystemwhereconnectionscanbecontrolledsystematicallybothintermsofdirectionandfunctionalcoupling.Howeveritislimitedbythenumberoflayersthroughwhichinformationtransmissioncanbestudied.Tocompareeectivelywithotherstudiesinfeedforwardnetworks,itisessentialtodevelopasystemwithmorelayerstounderstandthelimitswithinwhichinformationcanbetransmittedeectively.Thetwochambermicrouidicdevice,however,canbeusedforotherstudieswhereitmightbeinteresttostudytheinteractionbetweentwodierentpopulationofneurons( Breweretal. , 2013 ; Kanagasabapathietal. , 2012 )orstudytheeectofpharmacologicalagents. 80

PAGE 81

CHAPTER5INFORMATIONTRANSMISSIONTHROUGHMULTIPLELAYERSINANETWORKResearchinpropagationofneuralcodesinfeedforwardnetworkshasalmostalwaysinvolvednetworkswithmorethan2layers.Thisisduetothefactthatthemodelhasitsoriginsinthecorticalminicolumnidea.Theanatomyofcortexshowslaminarorganizationwithdistinctivelayersintheverticaldimensionandcanbethoughttobeorganizedascolumns.Ithasbeenhypothesizedthatthisorganizationhasfunctionalimplicationsandbehaviorarisesfromthepropagationofinformationbetweensuchcolumns( Bux-hoevedenandCasanova , 2002 ; Mountcastle , 1997 ).Sincethecortexiscomposedofmanysuchcolumns,itisimportanttostudypropagationofdierentcodesthroughmultiplelayers.Insomecases,foragivenmodeloffeedforwardnetwork,thenatureofcodepropagationwasinuencedbythenumberoflayersactivitypassedthrough( Litvaketal. , 2003 ).Given'Brain-on-a-chip'technology,itiseasytoextendthetwochambersystemdetailedinthepreviouschaptertoonecontainingmultiplechambers.Inthisstudy,asystemconsistingof4chamberswithtunnelsconnectingthemwasdevelopedandinformationpropagationwasstudied.Theschematicandanexampleofa4chamberdevicewithneuronsareshowninFigure 5-1 .Thedeviceconsistsof4chamberseach6mmx3mmandconnectedby3setsof50tunnelsthatare600mlongand3mx10mincross-section.ThelayoutischosentotthedeviceonanMEAwithelectrodesarrangedina6x10array.Theredlinesinthemicrographdenotesthedirectioninwhichconnectivityisdesired.Thesequentialplatingprocedurefollowedinthepreviousstudyisnotconduciveinthiscasetoprovideaunidirectionalnetwork.Itwouldrequirewaiting28daysafterplatingtherstchambertoplatethefourthchamberduringwhichthenetworkinchamber1wouldhavereachedmaturityanditmightnotbeappropriatecomparingtheactivityfromnetworkswithsuchwidedierenceinages.Therefore,mostofthe 81

PAGE 82

Figure5-1. Schematicandmicrographofinvitro4layernetwork.Theschematicofthedeviceisshownonleft.Thechambersare3mmx6mmwhilethetunnelswere600mlong,3mhighand10mwide.Thetunnelsconnectchambers1to2,2to3and3to4.AmicrographofthenetworkgrowingonanMEAshotat10xmagnicationisshownonright.Theedgesofthechambersarecurvedduetotheshapeofthepunchusedincuttingoutthepdmstocreatechambers.Theelectrodesare30mindiameterandspaced500mapart.TheredlinesindicatethedirectioninwhichconnectivityisdesiredtocreateafourlayeredfeedforwardnetworksimilartothetwolayerednetworkinChapter 4 82

PAGE 83

experimentsinthisstudywerecarriedoutinsubjectswhereallthefourchamberswereplatedatthesametime. 5.1PropagationofBurststhroughMultipleLayers3MEAswith4chamberPDMSdevicesattachedtothemwereplatedwith30,000dissociatedcorticalneurons.Spontaneousactivitywasrecordedatmultipledaysafterplatingwhenburstswereobservedtopropagatethroughall4layers.Althoughtherewerevariationsamongthe3MEAs,therewasnosignicantdierencebetweenthelayersintermsofburstdynamicsasshowninFigure 5-2 . Figure5-2. BurstDynamicsin4LayerNetwork.Nosignifantdierencewasobservedbetweenthelayersintermsof(A)BurstRate,(B)BurstDuration,(C)Inter-BurstIntervaland(D)Intra-BurstSpikeRate.ThenumbersintheX-AxistheMEAidenticationnumber.Thebarsin(B),(C)and(D)denotethemeanacrossallburstsdetectedinthelayer.Errorbarsdenotethestandarddeviation Notalltheburstsdetectedinthenetworkpropagatedthroughall4layers.Therewereburststhatwererestrictedtoonechamberorpropagatedthroughtwoorthreechambers.Theproportionofburstsdetectedineachwellthatpropagatedthroughall4 83

PAGE 84

layersisshowninFigure 5-3 (Left).Itcanbeseenthatsimilartootherburstdynamicsmeasures,therewasnosignicantdierencebetweenthelayersinthisaspect.Duetoabsenceofanydesigninconstrainingdirectionality,burstsoriginatedinallchambersandpropagatedinallpossibledirections.Howeverthelikelihoodofeachlayerbeingtheburstoriginationlayerwasnotequal.Figure 5-3 (Right)showstheproportionofburststhatoriginatedateachlayerofthenetworkinthesubsetofburststhatpropagatedthroughall4layers.Itappearsasthoughoneortwolayersdominatetheactivityofotherlayersandactastheburstoriginatormoreoftenthanothers.Additionalexperimentsarerequiredtoconrmthis. Figure5-3. Propagationofbursts.(Left)Proportionofburstsdetectedineachlayerthatpropagatedto/fromotherlayers.(Right)Burstoriginationlayerinthesubsetofburststhatpropagatedthroughall4layers. 5.2FidelityofInformationTransmissionVictor-Purpuradistancemetricwasagainusedasameasureofdelityofinformationtransmissionacrossthe4layerednetwork.OnlytheburststhatoriginatedinLayer1andpropagatedthroughalltheotherlayerswereusedinthisanalysistoremoveanyconfoundingeectofpropagationdirection.Thedissimilaritymetricwascomputedbetweenallpairsofelectrodesacrossthe4chamberswithqvaluesbetween2and500. 84

PAGE 85

Sincetherearemultiplelayers,thesimilaritycanbecomparedwithreferencetoeachlayerseparately.Thegeneraltrendwasadecreasingsimilaritywithincreasingdistancewhenlayers1,2and3werechosenasthereference.However,whenlayer4waschosenasthereference,notrendwasobserved.Furtherinvestigationintotheunderlyingcauseforthisisrequired. Figure5-4. Fidelityofinformationtransmissionin4layernetworks.Thesquaresrepresentthemeansimilarityacrossdierentqvaluesbetweenelectrodeswithinthelayeroracrossindicatedlayers.Errorbarsdenote1.96xS.E.M.Thesimilaritybetweenelectrodeswithinalayerishigherthanthesimilaritybetweentheelectrodesacrossdierentlayerswhenelectrodesinlayer1,2and3areconsidered.Alsothesimilaritiesseemtodecreasewithdistancefromtheselayers.Howeversincethesimilaritiesbetweenlayer4andanyotherlayerwerelower,thereisnoapparenttrend. 5.3EectofDisinhibitionInvivoandinvitrostudieshaveshownthatgammaoscillationsplayanimportantroleinthepropagationofinformationinthebrain( Colginetal. , 2009 ; Fisahnetal. , 1998 ; LismanandJensen , 2013 ; Sederbergetal. , 2007 ).Ithasalsobeenshownthatinhibitoryneuronsplayacentralroleingeneratinggammaoscillationsbothinintactbrainsandbrainslices( SohalandHuguenard , 2005 ; WangandBuzsaki , 85

PAGE 86

1996 ; Whittingtonetal. , 2000 ; Wilson , 2007 ).Dissociatedcorticalnetworksreectthecompositionofcortexintermsofexcitatoryandinhibitoryneurons.80%ofneuronsareexcitatorywhile20%areinhibitory.Tostudyhowthedelityofinformationtransmissionisaectedbyabsenceofinhibition,networksweretreatedwith10MsolutionofBicucullineaGABA-Aantagonistthatisgenerallyusedinstudyofdisinhibition( Khalilovetal. , 2005 ; Panetal. , 2009b ).Thehypothesiswasthatdisinhibitionshouldleadtopoorertransmissionofinformationasinhibitoryneuronsareknowntoplayanimportantroleininformationpropagation( VogelsandAbbott , 2009 ).Amarkeddierenceinthespontaneousactivityofthenetworkwasobserved.HoweverthedierencewasnotconsistentacrossalllayersasshowninFigure 5-5 .Therewasasignicantincreaseinthedurationoftheburstwithaslightdecreaseintheinterburstinterval.Burstdurationandintraburstspikerateshowedincreaseanddecreaserespectivelyacrossalllayers.Nochangewasobservedintheburstinitiationsitesortheproportionofburststhatpropagatedthroughall4layers.Tostudytheeectofdisinhibitiononinformationpropagation,similaritywascomputedbetweentheelectrodesindierentlayersasbefore.Itwasseenthatthesimilaritiesbetweentheelectrodesofdierentlayersduringtreatmentdidnotshowanyconsistentdierencefromthosebeforetreatment.InFigure 5-6 ,bluelinesdenotethesimilaritybetweenlayersbeforetreatingthenetworkswithbicucullinewhileredlinesdenotethesimilaritybetweenlayersduringtreatmentwithbicuculline.Whenthesimilaritybetweenlayersareconsidered,therewasnosignicantdierenceinthesimilaritiesbeforeandduringtreatment.Nexttheeectofdisinhibitiononeachpairofelectrodeswasstudied.DierencebetweenVPsimilaritiesbeforeandduringtreatmentwascomputedacrossallpairsofconnections.Figure 5-7 showsthemeandierenceacrossallsuchcomparisonsandacrossallthreesubjects.Itcanbeseenthatthedierencesareconsistentlylessthanzeroindicatingadecreaseinsimilarityaftertreatmentwithbicuculline.Thesimilarities 86

PAGE 87

Figure5-5. Eectofdisinhibitiononburstdynamics.Theguresshowthedierenceinmeanofvariousmeasuresofburstdynamicscomparedbeforeandduringbicucullinetreatment(MeasureBIC-Measurepretreat).ThereisaslightincreaseinburstrateinLayers2,3,4(A)whereasLayer1showsnochange.Burstduration(B)showsanincreaseacrossalllayers.Thereisaslightdecreaseininter-burstinterval(C).Intra-burstspikerate(D)showsandecreaseacrossalllayers.Errorbarsdenote1.96xS.E.M withinlayer2showedanincrease.Butthiswastheonlyoutliertothegeneraltrend.Thedecreaseinsimilarityimpliesthatinhibitoryneuronsplayaroleinthetransmissionofinformation.Researchhasshownthatinhibitoryneuronshelpincoordinationofactivityamongexcitatoryneuronstherebyallowingecienttransmissionofinformation.Whenthismechanismisaected,run-awayexcitationmayoccurleadingtobettertransmissionofactivitybutpoorertransmissionofinformationwithinactivity.Alltheanalysisdescribedabovewereconductedinasinglebatchof3MEAs.AdditionalexperimentsinvolvingmoreMEAsandatleastonemorebatchisrequiredtoobtainstatisticallysignicantresults. 5.4ConstrainingDirectionalityusingPhysicalBarrierInordertoobtainthedesireddirectionalityinthe4chambersystem,amodiedsequentialplatingapproachwasused.AsmallsliverofPDMS200minwidth,andas 87

PAGE 88

Figure5-6. Eectofdisinhibitionondelityofinformationtransmission.Thebluelinesrepresentthemeansimilaritybetweenelectrodeswithinthelayeroracrossindicatedlayersbeforetreatmentandtheredlinesrepresentthosewhentreatedwithbicuculline.Errorbarsdenote1.96xS.E.M.Noconsistenttrendwasobservedinthecomparisonofsimilarities. Figure5-7. Changeindelityofinformationtransmission.ThelinesdenotethechangeinVPsimilarityvalueaftertreatmentwithbicuculline.Errorbarsdenote1.96xS.E.M.Itcanbeseenthattherewasaconsistentdecreaseinsimilarities 88

PAGE 89

highasthechamberswascutandplacedmanuallyinchamber3rightnexttothetunnelsconnectingchamber2andchamber3.ItwasensuredthattheblockformedatightsealwiththeMEAsurfaceatthebottomofthechamber.Cellswereplatedinchambers1and3onday0andchambers2and4wereplatedonday4.ThePDMSblockpreventedthegrowthofaxonsfromchamber3tochamber4inthemeantimeanditwasremovedcarefullyonday7afterinitialplatingsothataxonsfromchamber2canformsynapticconnectionswithneuronsinchamber3.Thisapproachreducedthetimerequiredforplatingthecellstherebyalleviatingthenetworkmaturationissueofa4stepsequentialplatingprocess.Abatchof3MEAswereplatedinthismannerandthespontaneousactivitywasobserved.ItwasseenthatLayers1and3werethemaininitiatorsofbursts.Burstsinitiatinginlayer3propagatedtolayer4anddidnotpropagateinthereversedirection.Burstsinitiatinginlayer1propagatedthroughall4layersorwererestrictedtojustlayers1and2dependingonifaburstwasinitiatedinlayer3inashortintervalbeforetheburstfromlayer1arrived.AnexampleisshowninFigure 5-8 .ThisapproachcanbefurthermodiedbyplacingaPDMSblockinchambers2,3and4andplatingallfourchambersatthesametime.Suchanapproachwouldeliminatethenecessityforsequentialplatingtoproducedirectionality. 5.5DiscussionInthisstudy,PDMSdevicesconsistingof4chamberswithtunnelsconnectingthemwerecreated.Whencellswereplatedinthematthesametime,networkswithnoapparentdirectionalitywereformed.Someoftheburstspropagatedthroughallthechambers,whileotherswerelocalizedtooneortwochambers.Failuretopropagateislikelyduetonon-uniformityintheintrinsicdynamicsofeachlayer.Thenon-uniformitymanifestsitselfinrefractorinessatdierenttimepointspreventingtheburstpropagatingthroughthelayer.However,thevariousmeasuresofburstdynamicsdidnotindicateanysignicantdierencesbetweenlayers.Additionalanalysesrelatingburstdynamicstoparticipationofalayerinpropagatedburstswillbehelpfultostudythisaspectofburst 89

PAGE 90

Figure5-8. Burstpropagationinasequentiallyplated4layernetworkshownasarasterplot.Theboxesdenotethelayersandthedottedlinedenotesthestartoftheburst.ThegureonleftshowsaburstthatinitiatedinLayer1andpropagatedthroughtheotherlayerswhiletheoneontherightshowstheburstthatinitiatedinlayer3andpropagatedjusttolayer4. propagation.Theresultswereconsistentwithotherstudiesthatlookedatcommunicationbetweendissociatednetworksconsistingofsubnetworkscreatedthroughothermethods( Baruchietal. , 2008 ; Maedaetal. , 1995 ; Shteingartetal. , 2010 ).FidelityofinformationtransmissionasmeasuredwithVictorPurpurametricshoweddierencesbetweenproximalanddistantlayers.Howeverthetrendwasnotuniversalwithlayer2beingtheoutlierinallcases.Sincetheresultsarefromjust3subjects,additionalexperimentsarerequiredtoseeifthiseectpersists.Dissociatednetworkshaveseldombeenthoughttogenerateoscillationsasitisbelievedthattheabsenceofadenedstructureprecludesthemfromsuchphenomenon.Howeverarecentstudyhasshownevidenceforthepresenceofrhythmssimilartothoseobservedinslicesandintactbrains( Leondopulosetal. , 2012 ).Dierentstudieshaveshownthattheoscillationsplayanimportantroleininformationtransmissioninthebrain( Canoltyetal. , 2006 ; Colginetal. , 2009 ; Fries , 2009 ).Tostudyifsuch 90

PAGE 91

arelationcanbefoundindissociatednetworks,anindirectapproachwasfollowed.BicucullineisknowntoactasaGABA-Aantagonistandabolishesgammaoscillationsinvitrocitepcunningham2004role,sahn1998cholinergic,bartos2007synaptic.Presumably,inhibitoryneuronsweresilencedwhentreatedwithbicucullinetherebyshuttingtheprocessbywhichoscillationsaregenerated.Changesinburstdynamicswereobserved.Also,adecreaseinsimilaritieswasobservedimplyingthatthedelitywasdecreasedandhenceinhibitionplaysaroleinthetransmissionofinformation.Additionalinvestigationisrequiredtostudythedirectrelationbetweeninformationpropagationanddierentfrequencybands.Thestudydescribedhereinvolvednetworkswithoutanyapparentdirectionalityintermsofconnections.Usingphysicalbarriers,itispossibletocreatenetworksthataredirectionalandusefultostudydierentcomputationalpropertiesinfeedforwardnetworks.ThedimensionsofthePDMSdeviceandthenumberofchamberswererestrictedbythelayoutofelectrodesintheMEAused.UsingMEAswithdierentlayoutswillallowtocreatedeviceswithmorethanfourchambersleadingtostudyoffeedforwardnetworkswithasmanylayersasdesired. 91

PAGE 92

CHAPTER6CONCLUSIONThesignicanceofthedissertationliesinthedevelopmentofaplatformtostudystructure-functionrelationshipsinnetworksofarelativelysmallerpopulationofneurons.MostofthestudiesinvolvingMEAsfocusonpropertiesofnetworkswithnoapparentstructure.Byusingvarioustechniquesexplainedinthepreceedingchapters,neuronalnetworkscanbeforcedtospecicstructuresandtheirpropertiesstudied.Themainresultsofthisdissertationcanbesummarizedintermsoftechnologydevelopmentandinformationprocessinginfeedforwardnetworks.Thetwotechnologiesusedvariedintheeectivenessofcontrollingstructure.Microcontactpatterning,thoughabletoprovideconstraintoverstructureataleveloffewneurons,wasnoteectiveinalteringthefunctionofthenetworksatthatlevel.Ontheotherhand,microfabricateddevicesprovidedcontrolatthelevelofasmallpopulationofneuronsandgiventheabilitytocontroldirectionalityapproximatedfeedforwardnetworksbetterthanpatternednetworks.Intermsofinformationtransmission,thefollowingwerethesalientresults Informationwasmorereliablytransmittedasaratethanintighttimingsofspikes. Thereliabilityoftransmissionwasaectedbythestructuralpropertiesofthenetwork. Thereliabilityoftransmissiondecreasedwithdistanceinboththepatternednetworksandnetworkswithmicrofabricateddevices. Informationtransmissionwaspartiallydisruptedwheninhibitoryneuronswereinterferedwith. 6.1GeneralDiscussionInteractionbetweennetworksofneuronshaslongbeenaninterestinneuroscience.Thisfollowsfromthefactthatarguablyallbehaviorarisesfrominteractionbetweendierentnetworksinthebrain.Feedforwardnetworkshavebeenawidelyusedmodelforstudyingthisinteractionintermsofpropagationofneuralcodes.Dissociatedcultureshavenotbeenconduciveforsuchstudiessincetheconnectionsthatformbetweenneurons 92

PAGE 93

appeartoberandominnature.However,withtheuseoftechnologieslikesubstratepatterningandmicrouidicdevices,itispossibletobringstructureatdierentlevelsofthenetwork.Patternednetworkscreatedusingmicrocontactprinting(Chapter 3 )providedcontrolatthelevelofsinglecell/smallgroupsofcells.Itwasseenthat,atsuchalevel,convergenceofconnectionsplayedanimportantroleineectivetransmissionofinformationconrmingthetheoreticworkbyAbeles( Abeles , 1991 ).Also,itprovidedevidenceforarelationbetweenthestructureofthenetworkanditsfunctionalconnectivity.Thisresultisconsistentwithsomeofthestudiesthathaveshownthestructuralconnectivityinthebrainaectsitsfunctionalconnectivityusingfunctionalimaginginintactbrains( Hagmannetal. , 2008 ; Honeyetal. , 2007 ; Pontenetal. , 2010 ).ThetwolayerednetworkstudiedinChapter 4 showedthatfeedforwardnetworkscanberealizedusingdissociatedneuronalculturesandmicrofabricateddevices.Incontrasttomicrocontactprinting,thedirectionofconnectionscanbecontrolled.Theconnectionswithinalayerintherealizedfeedforwardnetworkcannotbecontrolledatthelevelofindividualneurons,butcanbecontrolledatthelevelofsmallpopulationsofneurons.Sincemostoftheactivityobservedinbrainsuggestsapopulationcodingofbehavior,thisapproachoersnewpossibilitiesforinvestigatingpopulationconnectivity.Itwasshownthatbycontrollingthenumberoftunnelsbetweenthechambers,structuralconnectionsandinturnfunctionalconnectionscanbecontrolledrobustly.Thepropagationofburstactivityfromonechambertoanothersuggeststheinformationbeingpropagatedispredominantlyintheformofratecodes,althoughthereisenoughsimilarityathightemporalresolutionnottocompletelyruleoutratecoding.However,thattheburstpeaktimingsaretightlycorrelatedsupportsahypothesisthatthereisaburstbasedtemporalcode. 93

PAGE 94

Extendingthetechnologyusedincreatingtwolayerednetworks,fourlayerednetworkswerecreated.However,thedirectionalityimposedintheformercouldnotbeimposedinthelatterduetotheincompatibilityofsequentialplatingprocedure.Still,itservedasadecentmodelasenoughactivitywasobservedtopropagatethroughalllayersintheintendeddirection.Asthedistancebetweenlayersincreasedtherewasadecreaseinthedelityofinformationsuggestingaphysicallimitintheeectivetransmissionofinformationbyneurons.Thisphenomenonwasobservedinpatternednetworksaswell.Also,itwasseenthatdisinhibitionplayedaroleinthetransmissionofinformationacrossthenetwork. 6.2FutureWorkSubstratepatterningtechniqueusedinthisstudywasrobustenoughtoproducenetworkswithreproduciblestructuralproperties.However,thecontroloverspecicconnectionsofneuronswasnotpossibleandwouldprobablyrequireusingmethodslikechemicalgradientorelectricalgradients( Dowell-Mesnetal. , 2004 ; WoodandWillits , 2009 )inconjunctionwithmicrocontactprinting.Analternativewouldbetousemicrouidicdevicestocreatetherequiredpatternsasthereliabilityincontrollingtheconnectionsismorethanprintingproteinsonthesurface.Thepatternsusedinthisstudyweresimplisticindesignandsuchstructuresarediculttondinanactualbrain.Patternsthatarebiologicallyplausibleandthatareinspiredfromnetworktheorywouldenableunderstandingthecomputationalpropertiesofindividualneuronsaswellthecollectivecomputationalpropertyofthenetwork.Microuidicdeviceshavebeenusedinconjunctionwithco-culturesoforganotypicslicesaswellasdissociatedneurons( Breweretal. , 2013 ; Kanagasabapathietal. , 2012 ).Also,dierentcelltypescanbeplatedindierentchambersandtheinteractionbetweenthemstudied.Thisapproachhelpsinunderstandingtheunderlyingmechanismofcomputationinthesepartsinanintactbrain.Inaddition,utilisinghighdensityMEAs( Freyetal. , 2009 ; Gandolfoetal. , 2010 ; Patolskyetal. , 2006 )wouldenable 94

PAGE 95

samplingfrommanymoresinglecellsthanispossiblenowandwouldprovidedeeperinsightsintohowsinglecellactivitytranslatestonetworklevelactivities.Withrecentstudiesshowingthepresenceofoscillationspreviouslythoughtnotpossibleindissociatedcultures,onelineofinvestigationwouldbetostudytheroleofoscillationsinsuchmulti-layerednetworks.Thefrequencyofoscillationsinthebrainisthoughttobefunctionallyrelevant.Suchstudiescanbeextendedtoinvitronetworkswhereitispossibletounderstandthenetworkmechanismsthatunderliethegenerationofoscillations. 95

PAGE 96

REFERENCES Abeles,M.(1991),Corticonics:Neuralcircuitsofthecerebralcortex(CambridgeUniversityPress) Adrian,E.D.(1928),Thebasisofsensation.(WWNorton&Co) Aertsen,A.,Diesmann,M.,andGewaltig,M.(1996),Propagationofsynchronousspikingactivityinfeedforwardneuralnetworks,JournalofPhysiology-Paris,90,3-4,243{247 Aertsen,A.andGerstein,G.(1985),Evaluationofneuronalconnectivity:sensitivityofcross-correlation,Brainresearch,340,2,341{354 Aertsen,A.,Gerstein,G.,Habib,M.,andPalm,G.(1989),Dynamicsofneuronalringcorrelation:modulationof"eectiveconnectivity",JournalofNeurophysiology,61,5,900{917 Aggelopoulos,N.C.,Franco,L.,andRolls,E.T.(2005),Objectperceptioninnaturalscenes:encodingbyinferiortemporalcortexsimultaneouslyrecordedneurons,Journalofneurophysiology,93,3,1342{1357 Ahissar,E.,Sosnik,R.,andHaidarliu,S.(2000),Transformationfromtemporaltoratecodinginasomatosensorythalamocorticalpathway,Nature,406,6793,302{306 Ainsworth,M.,Lee,S.,Cunningham,M.O.,Traub,R.D.,Kopell,N.J.,andWhittington,M.A.(2012),Ratesandrhythms:asynergisticviewoffrequencyandtemporalcodinginneuronalnetworks,Neuron,75,4,572{583 Albright,T.D.(1984),Directionandorientationselectivityofneuronsinvisualareamtofthemacaque,JournalofNeurophysiology,52,6,1106{1130 Andrade,J.andHlady,V.(1986),Proteinadsorptionandmaterialsbiocompatibility:atutorialreviewandsuggestedhypotheses,inBiopolymers/Non-ExclusionHPLC(Springer),1{63 Arnoldi,H.-M.R.,Englmeier,K.-H.,andBrauer,W.(1999),Translation-invariantpatternrecognitionbasedonsynrechains,Biologicalcybernetics,80,6,433{447 Azouz,R.andGray,C.M.(2000),Dynamicspikethresholdrevealsamechanismforsynapticcoincidencedetectionincorticalneuronsinvivo,ProceedingsoftheNationalAcademyofSciences,97,14,8110{8115 Baddeley,R.,Abbott,L.F.,Booth,M.C.,Sengpiel,F.,Freeman,T.,Wakeman,E.A.,etal.(1997),Responsesofneuronsinprimaryandinferiortemporalvisualcorticestonaturalscenes,ProceedingsoftheRoyalSocietyofLondon.SeriesB:BiologicalSciences,264,1389,1775{1783 96

PAGE 97

Bani-Yaghoub,M.,Tremblay,R.,Voicu,R.,Mealing,G.,Monette,R.,Py,C.,etal.(2005),Neurogenesisandneuronalcommunicationonmicropatternedneurochips,Biotechnologyandbioengineering,92,3,336{345 Baruchi,I.,Volman,V.,Raichman,N.,Shein,M.,andBen-Jacob,E.(2008),Theemergenceandpropertiesofmutualsynchronizationininvitrocoupledcorticalnetworks,EuropeanJournalofNeuroscience,28,9,1825{1835 Bastian,M.,Heymann,S.,andJacomy,M.(2009),Gephi:anopensourcesoftwareforexploringandmanipulatingnetworks.,inICWSM Beggs,J.andPlenz,D.(2003),Neuronalavalanchesinneocorticalcircuits,TheJournalofneuroscience,23,35,11167{11177 Berdichevsky,Y.,Staley,K.J.,andYarmush,M.L.(2010),Buildingandmanipulatingneuralpathwayswithmicrouidics,LabonaChip,10,8,999{1004 Bhatia,S.K.,Shriver-Lake,L.C.,Prior,K.J.,Georger,J.H.,Calvert,J.M.,Bredehorst,R.,etal.(1989),Useofthiol-terminalsilanesandheterobifunctionalcrosslinkersforimmobilizationofantibodiesonsilicasurfaces,Analyticalbiochemistry,178,2,408{413 Biederlack,J.,Castelo-Branco,M.,Neuenschwander,S.,Wheeler,D.W.,Singer,W.,andNikolic,D.(2006),Brightnessinduction:rateenhancementandneuronalsynchronizationascomplementarycodes,Neuron,52,6,1073{1083 Blankenship,A.andFeller,M.(2009),Mechanismsunderlyingspontaneouspatternedactivityindevelopingneuralcircuits,NatureReviewsNeuroscience,11,1,18{29 Bliss,T.V.andLmo,T.(1973),Long-lastingpotentiationofsynaptictransmissioninthedentateareaoftheanaesthetizedrabbitfollowingstimulationoftheperforantpath,TheJournalofphysiology,232,2,331{356 Boehler,M.,Leondopulos,S.,Wheeler,B.,andBrewer,G.(2011),Hippocampalnetworksonreliablepatternedsubstrates,JournalofNeuroscienceMethods Bologna,L.,Nieus,T.,Tedesco,M.,Chiappalone,M.,Benfenati,F.,andMartinoia,S.(2010),Low-frequencystimulationenhancesburstactivityincorticalculturesduringdevelopment,Neuroscience,165,3,692{704 Borgers,C.andKopell,N.J.(2008),Gammaoscillationsandstimulusselection,NeuralComputation,20,2,383{414 Borst,A.,Theunissen,F.,etal.(1999),Informationtheoryandneuralcoding,Natureneuroscience,2,947{958 Botzolakis,E.,Maheshwari,A.,Feng,H.,Lagrange,A.,Shaver,J.,Kassebaum,N.,etal.(2009),Achievingsynapticallyrelevantpulsesofneurotransmitterusingpdmsmicrouidics,Journalofneurosciencemethods,177,2,294{302 97

PAGE 98

Braitenberg,V.(1978),Cellassembliesinthecerebralcortex,inTheoreticalapproachestocomplexsystems(Springer),171{188 Branch,D.,Corey,J.,Weyhenmeyer,J.,Brewer,G.,andWheeler,B.(1998),Microstamppatternsofbiomoleculesforhigh-resolutionneuronalnetworks,MedicalandBiologicalEngineeringandComputing,36,1,135{141 Brewer,G.J.,Boehler,M.D.,Leondopulos,S.,Pan,L.,Alagapan,S.,DeMarse,T.,etal.(2013),Towardaself-wiredactivereconstructionofthehippocampaltrisynapticloop:Dg-ca3,FrontiersinNeuralCircuits,7,165 Buxhoeveden,D.P.andCasanova,M.F.(2002),Theminicolumnhypothesisinneuroscience,Brain,125,5,935{951 Buzsaki,G.(2004),Large-scalerecordingofneuronalensembles,Natureneuroscience,7,5,446{451 Buzsaki,G.(2010),Neuralsyntax:cellassemblies,synapsembles,andreaders,Neuron,68,3,362{385 Buzsaki,G.andMizuseki,K.(2014),Thelog-dynamicbrain:howskeweddistributionsaectnetworkoperations,NatureReviewsNeuroscience Cadotte,A.,DeMarse,T.,He,P.,andDing,M.(2008),Causalmeasuresofstructureandplasticityinsimulatedandlivingneuralnetworks,PLoSOne,3,10,e3355 Callaway,E.M.(1998),Localcircuitsinprimaryvisualcortexofthemacaquemonkey,Annualreviewofneuroscience,21,1,47{74 Callaway,E.M.etal.(2004),Feedforward,feedbackandinhibitoryconnectionsinprimatevisualcortex,NeuralNetworks,17,5,625{632 Canolty,R.T.,Edwards,E.,Dalal,S.S.,Soltani,M.,Nagarajan,S.S.,Kirsch,H.E.,etal.(2006),Highgammapowerisphase-lockedtothetaoscillationsinhumanneocortex,science,313,5793,1626{1628 Carelli,R.andDeadwyler,S.(1994),Acomparisonofnucleusaccumbensneuronalringpatternsduringcocaineself-administrationandwaterreinforcementinrats,TheJournalofneuroscience,14,12,7735{7746 Carr,C.andKonishi,M.(1990),Acircuitfordetectionofinterauraltimedierencesinthebrainstemofthebarnowl,TheJournalofNeuroscience,10,10,3227{3246 Carr,C.E.andKonishi,M.(1988),Axonaldelaylinesfortimemeasurementintheowl'sbrainstem,ProceedingsoftheNationalAcademyofSciences,85,21,8311{8315 Chang,J.,Brewer,G.,andWheeler,B.(2001),Modulationofneuralnetworkactivitybypatterning,BiosensorsandBioelectronics,16,7-8,527{533 98

PAGE 99

Chang,J.,Brewer,G.,andWheeler,B.(2006),Neuronalnetworkstructuringinducesgreaterneuronalactivitythroughenhancedastroglialdevelopment,Journalofneuralengineering,3,217 Chang,J.C.,Brewer,G.J.,andWheeler,B.C.(2003),Amodiedmicrostampingtechniqueenhancespolylysinetransferandneuronalcellpatterning,Biomaterials,24,17,2863{2870 Chapin,J.andNicolelis,M.(1999),Principalcomponentanalysisofneuronalensembleactivityrevealsmultidimensionalsomatosensoryrepresentations,Journalofneurosciencemethods,94,1,121{140 Chiappalone,M.,Bove,M.,Vato,A.,Tedesco,M.,andMartinoia,S.(2006),Dissociatedcorticalnetworksshowspontaneouslycorrelatedactivitypatternsduringinvitrodevelopment,Brainresearch,1093,1,41{53 Chiappalone,M.,Massobrio,P.,andMartinoia,S.(2008),Networkplasticityincorticalassemblies,EuropeanJournalofNeuroscience,28,1,221{237 Claverol-tintur,E.,Rosell,X.,andCabestany,J.(2007),Technicalstepstowardsone-to-oneelectrode-neuroninterfacingwithneuralcircuitsreconstructedinvitro,Neurocomputing,70,2716{2722,doi:10.1016/j.neucom.2006.06.018 Claverol-Tinture,E.,Ghirardi,M.,Fiumara,F.,Rosell,X.,andCabestany,J.(2005),Multielectrodearrayswithelastomericmicrostructuredoverlaysforextracellularrecordingsfrompatternedneurons,Journalofneuralengineering,2,2,L1 Colgin,L.L.,Denninger,T.,Fyhn,M.,Hafting,T.,Bonnevie,T.,Jensen,O.,etal.(2009),Frequencyofgammaoscillationsroutesowofinformationinthehippocampus,Nature,462,7271,353{357 Corey,J.,Wheeler,B.,andBrewer,G.(1996),Micrometerresolutionsilane-basedpatterningofhippocampalneurons:criticalvariablesinphotoresistandlaserablationprocessesforsubstratefabrication,BiomedicalEngineering,IEEETransactionson,43,9,944{955 Corner,M.,VanPelt,J.,Wolters,P.,Baker,R.,andNuytinck,R.(2002),Physiologicaleectsofsustainedblockadeofexcitatorysynaptictransmissiononspontaneouslyactivedevelopingneuronalnetworks{aninquiryintothereciprocallinkagebetweenintrinsicbiorhythmsandneuroplasticityinearlyontogeny,Neuroscience&BiobehavioralReviews,26,2,127{185 Corner,M.A.,Baker,R.E.,Pelt,J.v.,andWolters,P.S.(2005),Compensatoryphysiologicalresponsestochronicblockadeofaminoacidreceptorsduringearlydevelopmentinspontaneouslyactiveorganotypiccerebralcortexexplantsculturedinvitro,Progressinbrainresearch,147,231{248 99

PAGE 100

Corner,M.A.andRamakers,G.(1992),Spontaneousringasanepigeneticfactorinbraindevelopmentphysiologicalconsequencesofchronictetrodotoxinandpicrotoxinexposureonculturedratneocortexneurons,Developmentalbrainresearch,65,1,57{64 Cornish,T.,Branch,D.,Wheeler,B.,andCampanelli,J.(2002),Microcontactprinting:aversatiletechniqueforthestudyofsynaptogenicmolecules,Molecularandcellularneuroscience,20,1,140{153 Darbon,P.,Scicluna,L.,Tscherter,A.,andStreit,J.(2002),Mechanismscontrollingburstingactivityinducedbydisinhibitioninspinalcordnetworks,EuropeanJournalofNeuroscience,15,4,671{683 Dayan,P.andAbbott,L.F.(2001),Theoreticalneuroscience:Computationalandmathematicalmodelingofneuralsystems(Taylor&Francis) deSantos-Sierra,D.,Sendi~na-Nadal,I.,Leyva,I.,Almendral,J.A.,Anava,S.,Ayali,A.,etal.(2014),Emergenceofsmall-worldanatomicalnetworksinself-organizingclusteredneuronalcultures,PLOSONE,9,1,e85828 Diesmann,M.,Gewaltig,M.,andAertsen,A.(1999),Stablepropagationofsynchronousspikingincorticalneuralnetworks,Nature,402,6761,529{533 Ding,M.,Chen,Y.,andBressler,S.(2006),Grangercausality:basictheoryandapplicationtoneuroscience,Handbookoftimeseriesanalysis,437{460 Dinh,N.-D.,Chiang,Y.-Y.,Hardelauf,H.,Baumann,J.,Jackson,E.,Waide,S.,etal.(2013),Microuidicconstructionofminimalisticneuronalco-cultures,LabonaChip Dowell-Mesn,N.,Abdul-Karim,M.,Turner,A.,Schanz,S.,Craighead,H.,Roysam,B.,etal.(2004),Topographicallymodiedsurfacesaectorientationandgrowthofhippocampalneurons,Journalofneuralengineering,1,2,78 Downes,J.H.,Hammond,M.W.,Xydas,D.,Spencer,M.C.,Becerra,V.M.,Warwick,K.,etal.(2012),Emergenceofasmall-worldfunctionalnetworkinculturedneurons,PLoScomputationalbiology,8,5,e1002522 Edwards,D.,Stancescu,M.,Molnar,P.,andHickman,J.J.(2013),Twocellcircuitsoforientedadulthippocampalneuronsonself-assembledmonolayersforuseinthestudyofneuronalcommunicationinadenedsystem,ACSchemicalneuroscience Eichler,M.(2006),Ontheevaluationofinformationowinmultivariatesystemsbythedirectedtransferfunction,Biologicalcybernetics,94,6,469{482 Engel,A.K.andSinger,W.(2001),Temporalbindingandtheneuralcorrelatesofsensoryawareness,Trendsincognitivesciences,5,1,16{25 Erickson,J.,Tooker,A.,Tai,Y.,andPine,J.(2008),Cagedneuronmea:Asystemforlong-terminvestigationofculturedneuralnetworkconnectivity,Journalofneurosciencemethods,175,1,1{16 100

PAGE 101

Esposti,F.,Lamanna,J.,Gullo,F.,Wanke,E.,andSignorini,M.(2007),Howdottxandap5aectthepost-recoveryneuronalnetworkactivitysynchronization?,inEngineeringinMedicineandBiologySociety,2007.EMBS2007.29thAnnualInternationalConferenceoftheIEEE(IEEE),3012{3015 Eytan,D.andMarom,S.(2006),Dynamicsandeectivetopologyunderlyingsynchronizationinnetworksofcorticalneurons,TheJournalofneuroscience,26,33,8465{8476 Fanselow,E.E.,Sameshima,K.,Baccala,L.A.,andNicolelis,M.A.(2001),Thalamicburstinginratsduringdierentawakebehavioralstates,ProceedingsoftheNationalAcademyofSciences,98,26,15330{15335 Fee,M.S.,Kozhevnikov,A.A.,andHahnloser,R.H.(2004),Neuralmechanismsofvocalsequencegenerationinthesongbird,AnnalsoftheNewYorkAcademyofSciences,1016,1,153{170 Feinerman,O.andMoses,E.(2006),Transportofinformationalongunidimensionallayerednetworksofdissociatedhippocampalneuronsandimplicationsforratecoding,TheJournalofneuroscience,26,17,4526{4534 Felleman,D.J.andVanEssen,D.C.(1991),Distributedhierarchicalprocessingintheprimatecerebralcortex,Cerebralcortex,1,1,1{47 Fisahn,A.,Pike,F.,Buhl,E.,andPaulsen,O.(1998),Cholinergicinductionofnetworkoscillationsat40hzinthehippocampusinvitro,NATURE-LONDON-,186{188 Fodor,S.,Read,J.,Pirrung,M.,Stryer,L.,Lu,A.,Solas,D.,etal.(1991),Light-directed,spatiallyaddressableparallelchemicalsynthesis.,Science(NewYork,NY),251,4995,767{773 Frey,U.,Egert,U.,Heer,F.,Hazovic,S.,andHierlemann,A.(2009),Microelectronicsystemforhigh-resolutionmappingofextracellularelectriceldsappliedtobrainslices,BiosensorsandBioelectronics,24,7,2191{2198 Fries,P.(2009),Neuronalgamma-bandsynchronizationasafundamentalprocessincorticalcomputation,Annualreviewofneuroscience,32,209{224 Fries,P.,Reynolds,J.H.,Rorie,A.E.,andDesimone,R.(2001),Modulationofoscillatoryneuronalsynchronizationbyselectivevisualattention,Science,291,5508,1560{1563 Fries,P.,Womelsdorf,T.,Oostenveld,R.,andDesimone,R.(2008),Theeectsofvisualstimulationandselectivevisualattentiononrhythmicneuronalsynchronizationinmacaqueareav4,TheJournalofNeuroscience,28,18,4823{4835 Friston,K.,Moran,R.,andSeth,A.K.(2012),Analysingconnectivitywithgrangercausalityanddynamiccausalmodelling,Currentopinioninneurobiology 101

PAGE 102

Friston,K.J.(1994),Functionalandeectiveconnectivityinneuroimaging:asynthesis,Humanbrainmapping,2,1-2,56{78 Fujisawa,S.,Amarasingham,A.,Harrison,M.T.,andBuzsaki,G.(2008),Behavior-dependentshort-termassemblydynamicsinthemedialprefrontalcortex,Natureneuroscience,11,7,823{833 Gandolfo,M.,Maccione,A.,Tedesco,M.,Martinoia,S.,andBerdondini,L.(2010),Trackingburstpatternsinhippocampalcultureswithhigh-densitycmos-meas,Journalofneuralengineering,7,056001 Gautrais,J.andThorpe,S.(1998),Ratecodingversustemporalordercoding:atheoreticalapproach,Biosystems,48,1,57{65 Georgopoulos,A.,Schwartz,A.,andKettner,R.(1986),Neuronalpopulationcodingofmovementdirection,Science,233,4771,1416{1419 Georgopoulos,A.P.,Kettner,R.E.,andSchwartz,A.B.(1988),Primatemotorcortexandfreearmmovementstovisualtargetsinthree-dimensionalspace.ii.codingofthedirectionofmovementbyaneuronalpopulation,TheJournalofNeuroscience,8,8,2928{2937 German,P.andFields,H.(2007),Ratnucleusaccumbensneuronspersistentlyencodelocationsassociatedwithmorphinereward,Journalofneurophysiology,97,3,2094{2106 Gerstein,G.andPerkel,D.(1969),Simultaneouslyrecordedtrainsofactionpotentials:analysisandfunctionalinterpretation,Science,164,3881,828{830 Gerstein,G.L.,Bedenbaugh,P.,andAertsen,A.M.(1989),Neuronalassemblies,Biomed-icalEngineering,IEEETransactionson,36,1,4{14 Gewaltig,M.,Diesmann,M.,andAertsen,A.(2001),Propagationofcorticalsynreactivity:survivalprobabilityinsingletrialsandstabilityinthemean,Neuralnetworks,14,6-7,657{673 Granger,C.(1969),Investigatingcausalrelationsbyeconometricmodelsandcross-spectralmethods,Econometrica:JournaloftheEconometricSociety,424{438 Gray,C.M.(1999),Thetemporalcorrelationhypothesisofvisualfeatureintegration:stillaliveandwell,Neuron,24,1,31{47 Gray,C.M.,Konig,P.,Engel,A.K.,Singer,W.,etal.(1989),Oscillatoryresponsesincatvisualcortexexhibitinter-columnarsynchronizationwhichreectsglobalstimulusproperties,Nature,338,6213,334{337 Gray,C.M.andSinger,W.(1989),Stimulus-specicneuronaloscillationsinorientationcolumnsofcatvisualcortex,ProceedingsoftheNationalAcademyofSciences,86,5,1698{1702 102

PAGE 103

Gross,G.(1979),Simultaneoussingleunitrecordinginvitrowithaphotoetchedlaserdeinsulatedgoldmultimicroelectrodesurface,BiomedicalEngineering,IEEETransac-tionson,,5,273{279 Gross,G.,Rhoades,B.,Reust,D.,andSchwalm,F.(1993),Stimulationofmonolayernetworksinculturethroughthin-lmindium-tinoxiderecordingelectrodes,Journalofneurosciencemethods,50,2,131{143 Guillotin,B.andGuillemot,F.(2011),Cellpatterningtechnologiesfororganotypictissuefabrication,Trendsinbiotechnology,29,4,183{190 Gustafsson,B.andWigstrom,H.(1986),Hippocampallong-lastingpotentiationproducedbypairingsinglevolleysandbriefconditioningtetanievokedinseparateaerents,TheJournalofneuroscience,6,6,1575{1582 Haddon,R.andLamola,A.(1985),Themolecularelectronicdeviceandthebiochipcomputer:presentstatus,ProceedingsoftheNationalAcademyofSciences,82,7,1874{1878 Hagmann,P.,Cammoun,L.,Gigandet,X.,Meuli,R.,Honey,C.,Wedeen,V.,etal.(2008),Mappingthestructuralcoreofhumancerebralcortex,PLoSbiology,6,7,e159 Hahnloser,R.H.,Kozhevnikov,A.A.,andFee,M.S.(2002),Anultra-sparsecodeunderliesthegenerationofneuralsequencesinasongbird,Nature,419,6902,65{70 Harris,K.D.(2005),Neuralsignaturesofcellassemblyorganization,NatureReviewsNeuroscience,6,5,399{407 Hatsopoulos,N.G.,Ojakangas,C.L.,Paninski,L.,andDonoghue,J.P.(1998),Informationaboutmovementdirectionobtainedfromsynchronousactivityofmotorcorticalneurons,ProceedingsoftheNationalAcademyofSciences,95,26,15706{15711 Havenith,M.N.,Yu,S.,Biederlack,J.,Chen,N.-H.,Singer,W.,andNikolic,D.(2011),Synchronymakesneuronsreinsequence,andstimuluspropertiesdeterminewhoisahead,TheJournalofNeuroscience,31,23,8570{8584 Hebb,D.(1949),Theorganizationofbehavior;aneuropsychologicaltheory.(Wiley) Hengst,U.,Deglincerti,A.,Kim,H.J.,Jeon,N.L.,andJarey,S.R.(2009),Axonalelongationtriggeredbystimulus-inducedlocaltranslationofapolaritycomplexprotein,Naturecellbiology,11,8,1024{1030 Hesse,W.,Moller,E.,Arnold,M.,andSchack,B.(2003),Theuseoftime-varianteeggrangercausalityforinspectingdirectedinterdependenciesofneuralassemblies,JournalofNeuroscienceMethods,124,1,27{44 Hickman,J.J.,Bhatia,S.K.,Quong,J.N.,Shoen,P.,Stenger,D.A.,Pike,C.J.,etal.(1994),Rationalpatterndesignforinvitrocellularnetworksusingsurface 103

PAGE 104

photochemistry,JournalofVacuumScience&TechnologyA:Vacuum,Surfaces,andFilms,12,3,607{616 Higgins,C.M.,Douglass,J.K.,andStrausfeld,N.J.(2004),Thecomputationalbasisofanidentiedneuronalcircuitforelementarymotiondetectionindipterousinsects,VisualNeuroscience,21,4,567{586 Higgins,N.(1985),Biochips:factorction?,ElectronicsandPower,31,10,761{763 Honey,C.,Sporns,O.,Cammoun,L.,Gigandet,X.,Thiran,J.-P.,Meuli,R.,etal.(2009),Predictinghumanresting-statefunctionalconnectivityfromstructuralconnectivity,ProceedingsoftheNationalAcademyofSciences,106,6,2035{2040 Honey,C.J.,Kotter,R.,Breakspear,M.,andSporns,O.(2007),Networkstructureofcerebralcortexshapesfunctionalconnectivityonmultipletimescales,ProceedingsoftheNationalAcademyofSciences,104,24,10240{10245 Hosmane,S.,Yang,I.H.,Run,A.,Thakor,N.,andVenkatesan,A.(2010),Circularcompartmentalizedmicrouidicplatform:studyofaxon{gliainteractions,LabonaChip,10,6,741{747 Hubel,D.H.andWiesel,T.N.(1962),Receptiveelds,binocularinteractionandfunctionalarchitectureinthecat'svisualcortex,TheJournalofphysiology,160,1,106 Hubel,D.H.andWiesel,T.N.(1968),Receptiveeldsandfunctionalarchitectureofmonkeystriatecortex,TheJournalofphysiology,195,1,215{243 Hubel,D.H.,Wiesel,T.N.,andLeVay,S.(1977),Plasticityofoculardominancecolumnsinmonkeystriatecortex,PhilosophicalTransactionsoftheRoyalSocietyofLondon.SeriesB,BiologicalSciences,377{409 Hummel,F.andGerlo,C.(2005),Largerinterregionalsynchronyisassociatedwithgreaterbehavioralsuccessinacomplexsensoryintegrationtaskinhumans,CerebralCortex,15,5,670{678 Ichikawa,M.,Muramoto,K.,Kobayashi,K.,Kawahara,M.,andKuroda,Y.(1993),Formationandmaturationofsynapsesinprimaryculturesofratcerebralcorticalcells:anelectronmicroscopicstudy,Neuroscienceresearch,16,2,95{103 Ikegaya,Y.,Aaron,G.,Cossart,R.,Aronov,D.,Lampl,I.,Ferster,D.,etal.(2004),Synrechainsandcorticalsongs:temporalmodulesofcorticalactivity,Science,304,5670,559{564 Izhikevich,E.M.(2006),Polychronization:computationwithspikes,Neuralcomputation,18,2,245{282 Jacquemin,C.(1994),Atemporalconnectionistapproachtonaturallanguage,ACMSIGARTBulletin,5,3,12{22 104

PAGE 105

Jimbo,Y.andKawana,A.(1992),Electricalstimulationandrecordingfromculturedneuronsusingaplanarelectrodearray,BioelectrochemistryandBioenergetics,29,2,193{204 Jimbo,Y.,Kawana,A.,Parodi,P.,andTorre,V.(2000),Thedynamicsofaneuronalcultureofdissociatedcorticalneuronsofneonatalrats,Biologicalcybernetics,83,1,1{20 Jimbo,Y.,Robinson,H.,andKawana,A.(1993),Simultaneousmeasurementofintracellularcalciumandelectricalactivityfrompatternedneuralnetworksinculture,BiomedicalEngineering,IEEETransactionson,40,8,804{810 Jimbo,Y.,Tateno,T.,andRobinson,H.(1999),Simultaneousinductionofpathway-specicpotentiationanddepressioninnetworksofcorticalneurons,BiophysicalJournal,76,2,670{678 Jun,S.B.,Hynd,M.R.,Dowell-Mesn,N.,Smith,K.L.,Turner,J.N.,Shain,W.,etal.(2007),Low-densityneuronalnetworksculturedusingpatternedpoly-l-lysineonmicroelectrodearrays,Journalofneurosciencemethods,160,2,317{326 Jungblut,M.,Knoll,W.,Thielemann,C.,andPottek,M.(2009),Triangularneuronalnetworksonmicroelectrodearrays:anapproachtoimprovethepropertiesoflow-densitynetworksforextracellularrecording,Biomedicalmicrodevices,11,6,1269{1278 Jutras,M.J.,Fries,P.,andBualo,E.A.(2009),Gamma-bandsynchronizationinthemacaquehippocampusandmemoryformation,TheJournalofNeuroscience,29,40,12521{12531 Kaminski,M.andBlinowska,K.(1991),Anewmethodofthedescriptionoftheinformationowinthebrainstructures,Biologicalcybernetics,65,3,203{210 Kaminski,M.,Ding,M.,Truccolo,W.,andBressler,S.(2001),Evaluatingcausalrelationsinneuralsystems:Grangercausality,directedtransferfunctionandstatisticalassessmentofsignicance,Biologicalcybernetics,85,2,145{157 Kamioka,H.,Maeda,E.,Jimbo,Y.,Robinson,H.,andKawana,A.(1996),Spontaneousperiodicsynchronizedburstingduringformationofmaturepatternsofconnectionsincorticalcultures,Neuroscienceletters,206,2-3,109{112 Kanagasabapathi,T.T.,Ciliberti,D.,Martinoia,S.,Wadman,W.J.,andDecre,M.M.(2011),Dual-compartmentneurouidicsystemforelectrophysiologicalmeasurementsinphysicallysegregatedandfunctionallyconnectedneuronalcellculture,Frontiersinneuroengineering,4 Kanagasabapathi,T.T.,Massobrio,P.,Barone,R.A.,Tedesco,M.,Martinoia,S.,Wadman,W.J.,etal.(2012),Functionalconnectivityanddynamicsofcortical{thalamicnetworksco-culturedinadualcompartmentdevice,Journalofneuralengineering,9,3,036010 105

PAGE 106

Khalilov,I.,LeVanQuyen,M.,Gozlan,H.,andBen-Ari,Y.(2005),Epileptogenicactionsofgabaandfastoscillationsinthedevelopinghippocampus,Neuron,48,5,787{796 Kira,A.,Okano,K.,Hosokawa,Y.,Naito,A.,Fuwa,K.,Yuyama,J.,etal.(2009),Micropatterningofperuoroalkylself-assembledmonolayersforarrayingproteinsandcellsonchips,AppliedSurfaceScience,255,17,7647{7651 Kispersky,T.,Gutierrez,G.,andMarder,E.(2011),Functionalconnectivityinarhythmicinhibitorycircuitusinggrangercausality,NeuralSystemsCircuits,1,9 Kleinfeld,D.,Kahler,K.,andHockberger,P.(1988),Controlledoutgrowthofdissociatedneuronsonpatternedsubstrates,TheJournalofneuroscience,8,11,4098{4120 Konig,P.,Engel,A.K.,andSinger,W.(1996),Integratororcoincidencedetector?theroleofthecorticalneuronrevisited,Trendsinneurosciences,19,4,130{137 Kreiter,A.K.andSinger,W.(1996),Stimulus-dependentsynchronizationofneuronalresponsesinthevisualcortexoftheawakemacaquemonkey,TheJournalofNeuro-science,16,7,2381{2396 Kreuz,T.,Chicharro,D.,Greschner,M.,andAndrzejak,R.G.(2011),Time-resolvedandtime-scaleadaptivemeasuresofspiketrainsynchrony,Journalofneurosciencemethods,195,1,92{106 Kreuz,T.,Haas,J.S.,Morelli,A.,Abarbanel,H.D.,andPoliti,A.(2007),Measuringspiketrainsynchrony,arXivpreprintphysics/0701261 Krumin,M.andShoham,S.(2010),Multivariateautoregressivemodelingandgrangercausalityanalysisofmultiplespiketrains,Computationalintelligenceandneuroscience,2010,6{6 Kumar,A.,Abbott,N.L.,Biebuyck,H.A.,Kim,E.,andWhitesides,G.M.(1995),Patternedself-assembledmonolayersandmeso-scalephenomena,Accountsofchemicalresearch,28,5,219{226 Kumar,A.,Rotter,S.,andAertsen,A.(2008),Conditionsforpropagatingsynchronousspikingandasynchronousringratesinacorticalnetworkmodel,TheJournalofNeuroscience,28,20,5268{5280 Kumar,A.,Rotter,S.,andAertsen,A.(2010),Spikingactivitypropagationinneuronalnetworks:reconcilingdierentperspectivesonneuralcoding,NatureReviewsNeuro-science,11,9,615{627 Kumar,A.andWhitesides,G.M.(1993),Featuresofgoldhavingmicrometertocentimeterdimensionscanbeformedthroughacombinationofstampingwithanelastomericstampandanalkanethiolinkfollowedbychemicaletching,AppliedPhysicsLetters,63,2002 106

PAGE 107

Larkum,M.E.,Zhu,J.J.,andSakmann,B.(1999),Anewcellularmechanismforcouplinginputsarrivingatdierentcorticallayers,Nature,398,6725,338{341 Latham,P.,Richmond,B.,Nelson,P.,andNirenberg,S.(2000),Intrinsicdynamicsinneuronalnetworks.i.theory,JournalofNeurophysiology,83,2,808{827 Laurent,G.andDavidowitz,H.(1994),Encodingofolfactoryinformationwithoscillatingneuralassemblies,Science,265,5180,1872{1875 Leonardo,A.andFee,M.S.(2005),Ensemblecodingofvocalcontrolinbirdsong,TheJournalofneuroscience,25,3,652{661 Leondopulos,S.,Boehler,M.,Wheeler,B.,andBrewer,G.(2012),Chronicstimulationofculturedneuronalnetworksboostslow-frequencyoscillatoryactivityatthetaandgammawithspikesphase-lockedtogammafrequencies,JournalofNeuralEngineering,9,026015 Li,M.andGreenside,H.(2006),Stablepropagationofaburstthroughaone-dimensionalhomogeneousexcitatorychainmodelofsongbirdnucleushvc,PhysicalReviewE,74,1,011918 Lin,J.-N.,Herron,J.,Andrade,J.D.,andBrizgys,M.(1988),Characterizationofimmobilizedantibodiesonsilicasurfaces,BiomedicalEngineering,IEEETransactionson,35,6,466{471 Lisman,J.E.andJensen,O.(2013),Thetheta-gammaneuralcode,Neuron,77,6,1002{1016 Litvak,V.,Sompolinsky,H.,Segev,I.,andAbeles,M.(2003),Onthetransmissionofratecodeinlongfeedforwardnetworkswithexcitatory{inhibitorybalance,TheJournalofneuroscience,23,7,3006{3015 Lom,B.,Healy,K.E.,andHockberger,P.E.(1993),Aversatiletechniqueforpatterningbiomoleculesontoglasscoverslips,Journalofneurosciencemethods,50,3,385{397 Long,M.A.,Jin,D.Z.,andFee,M.S.(2010),Supportforasynapticchainmodelofneuronalsequencegeneration,Nature,468,7322,394{399 Lopez,G.P.,Albers,M.W.,Schreiber,S.L.,Carroll,R.,Peralta,E.,andWhitesides,G.M.(1993),Convenientmethodsforpatterningtheadhesionofmammaliancellstosurfacesusingself-assembledmonolayersofalkanethiolatesongold,JournaloftheAmericanChemicalSociety,115,13,5877{5878 Madhavan,R.,Chao,Z.,andPotter,S.(2007),Plasticityofrecurringspatiotemporalactivitypatternsincorticalnetworks,Physicalbiology,4,181 Maeda,E.,Robinson,H.,andKawana,A.(1995),Themechanismsofgenerationandpropagationofsynchronizedburstingindevelopingnetworksofcorticalneurons,TheJournalofneuroscience,15,10,6834{6845 107

PAGE 108

Maher,M.,Pine,J.,Wright,J.,andTai,Y.-C.(1999),Theneurochip:anewmultielectrodedeviceforstimulatingandrecordingfromculturedneurons,JournalofNeuroscienceMethods,87,1,45{56 Marconi,E.,Nieus,T.,Maccione,A.,Valente,P.,Simi,A.,Messa,M.,etal.(2012),Emergentfunctionalpropertiesofneuronalnetworkswithcontrolledtopology,PloSone,7,4,e34648 Marom,S.andShahaf,G.(2002),Development,learningandmemoryinlargerandomnetworksofcorticalneurons:lessonsbeyondanatomy,QuarterlyReviewsofBiophysics,35,01,63{87 Martinoia,S.,Sanguineti,V.,Cozzi,L.,Berdondini,L.,VanPelt,J.,Tomas,J.,etal.(2004),Towardsanembodiedinvitroelectrophysiology:theneurobitproject,Neuro-computing,58,1065{1072 Masuda,N.andAihara,K.(2002),Bridgingratecodingandtemporalspikecodingbyeectofnoise,Physicalreviewletters,88,24,248101 Matsuda,T.,Inoue,K.,andSugawara,T.(1990),Developmentofmicropatterningtechnologyforculturedcells,ASAIOJournal,36,3,M559{561 Matsuzawa,M.,Tabata,T.,Knoll,W.,andKano,M.(2000),Formationofhippocampalsynapsesonpatternedsubstratesofalaminin-derivedsyntheticpeptide,EuropeanJournalofNeuroscience,12,3,903{910 Mazurek,M.,Shadlen,M.,etal.(2002),Limitstothetemporaldelityofcorticalspikeratesignals,Natureneuroscience,5,5,463{471 Miller,R.(1996),Neuralassembliesandlaminarinteractionsinthecerebralcortex,Biologicalcybernetics,75,3,253{261 Milner,P.M.(1974),Amodelforvisualshaperecognition.,Psychologicalreview,81,6,521 Mooney,J.,Hunt,A.,McIntosh,J.,Liberko,C.,Walba,D.,andRogers,C.(1996),Patterningoffunctionalantibodiesandotherproteinsbyphotolithographyofsilanemonolayers,ProceedingsoftheNationalAcademyofSciences,93,22,12287{12291 Morin,F.,Nishimura,N.,Griscom,L.,LePioue,B.,Fujita,H.,Takamura,Y.,etal.(2006),Constrainingtheconnectivityofneuronalnetworksculturedonmicroelectrodearrayswithmicrouidictechniques:asteptowardsneuron-basedfunctionalchips,Biosensorsandbioelectronics,21,7,1093{1100 Mountcastle,V.B.(1997),Thecolumnarorganizationoftheneocortex.,Brain,120,4,701{722 Mountcastle,V.B.(1998),Perceptualneuroscience:Thecerebralcortex(HarvardUniversityPress) 108

PAGE 109

Muramoto,K.,Ichikawa,M.,Kawahara,M.,Kobayashi,K.,andKuroda,Y.(1993),Frequencyofsynchronousoscillationsofneuronalactivityincreasesduringdevelopmentandiscorrelatedtothenumberofsynapsesinculturedcorticalneuronnetworks,Neuroscienceletters,163,2,163{165 Nam,Y.,Branch,D.W.,andWheeler,B.C.(2006),Epoxy-silanelinkingofbiomoleculesissimpleandeectiveforpatterningneuronalcultures,BiosensorsandBioelectronics,22,5,589{597 Nam,Y.,Chang,J.,Khatami,D.,Brewer,G.,andWheeler,B.(2004),Patterningtoenhanceactivityofculturedneuronalnetworks,inNanobiotechnology,IEEProceedings-,volume151(IET),volume151,109{115 Negyessy,L.,Nepusz,T.,Zalanyi,L.,andBazso,F.(2008),Convergenceanddivergencearemostlyreciprocatedpropertiesoftheconnectionsinthenetworkofcorticalareas,ProceedingsoftheRoyalSocietyB:BiologicalSciences,275,1649,2403{2410 Newman,M.E.(2003),Thestructureandfunctionofcomplexnetworks,SIAMreview,45,2,167{256 Nicolelis,M.,Ghazanfar,A.,Stambaugh,C.,Oliveira,L.,Laubach,M.,Chapin,J.,etal.(1998),Simultaneousencodingoftactileinformationbythreeprimatecorticalareas,Natureneuroscience,1,621{630 Nikolic,D.,Muresan,R.C.,Feng,W.,andSinger,W.(2012),Scaledcorrelationanalysis:abetterwaytocomputeacross-correlogram,EuropeanJournalofNeuroscience,35,5,742{762 Oenhausser,A.,Bocker-Meert,S.,Decker,T.,Helpenstein,R.,Gasteier,P.,Groll,J.,etal.(2007),Microcontactprintingofproteinsforneuronalcellguidance,SoftMatter,3,3,290{298 O'Keefe,J.(1979),Areviewofthehippocampalplacecells,Progressinneurobiology,13,4,419{439 O'Keefe,J.andDostrovsky,J.(1971),Thehippocampusasaspatialmap.preliminaryevidencefromunitactivityinthefreely-movingrat,Brainresearch,34,1,171{175 O'Keefe,J.andRecce,M.L.(1993),Phaserelationshipbetweenhippocampalplaceunitsandtheeegthetarhythm,Hippocampus,3,3,317{330 Pan,L.,Alagapan,S.,Franca,E.,Brewer,G.,andWheeler,B.(2011),Propagationofactionpotentialactivityinapredenedmicrotunnelneuralnetwork,JournalofNeuralEngineering,8,046031 Pan,L.,Song,X.,Xiang,G.,Wong,A.,Xing,W.,andCheng,J.(2009a),First-spikerankorderasareliableindicatorofburstinitiationanditsrelationwithearly-to-reneurons,BiomedicalEngineering,IEEETransactionson,56,6,1673{1682 109

PAGE 110

Pan,L.,Song,X.,Xiang,G.,Zhu,J.,andCheng,J.(2009b),Eectsofdisinhibitiononspatiotemporalpatternofneuronalrstrecruitmentinneuronalnetworks,ProgressinNaturalScience,19,5,615{621 Park,J.W.,Kim,H.J.,Byun,J.H.,Ryu,H.R.,andJeon,N.L.(2009),Novelmicrouidicplatformforculturingneurons:culturingandbiochemicalanalysisofneuronalcomponents,Biotechnologyjournal,4,11,1573{1577 Pasquale,V.,Massobrio,P.,Bologna,L.,Chiappalone,M.,andMartinoia,S.(2008),Self-organizationandneuronalavalanchesinnetworksofdissociatedcorticalneurons,Neuroscience,153,4,1354{1369 Pastalkova,E.,Itskov,V.,Amarasingham,A.,andBuzsaki,G.(2008),Internallygeneratedcellassemblysequencesintherathippocampus,Science,321,5894,1322{1327 Patolsky,F.,Timko,B.,Yu,G.,Fang,Y.,Greytak,A.,Zheng,G.,etal.(2006),Detection,stimulation,andinhibitionofneuronalsignalswithhigh-densitynanowiretransistorarrays,Science,313,5790,1100{1104 Perez-Orive,J.,Mazor,O.,Turner,G.C.,Cassenaer,S.,Wilson,R.I.,andLaurent,G.(2002),Oscillationsandsparseningofodorrepresentationsinthemushroombody,Science,297,5580,359{365 Perkel,D.H.andBullock,T.H.(1968),Neuralcoding.,NeurosciencesResearchProgramBulletin Pesaran,B.,Pezaris,J.S.,Sahani,M.,Mitra,P.P.,andAndersen,R.A.(2002),Temporalstructureinneuronalactivityduringworkingmemoryinmacaqueparietalcortex,Natureneuroscience,5,8,805{811 Pine,J.(1980),Recordingactionpotentialsfromculturedneuronswithextracellularmicrocircuitelectrodes,JournalofNeuroscienceMethods,2,1,19{31 Ponten,S.,Daertshofer,A.,Hillebrand,A.,andStam,C.(2010),Therelationshipbetweenstructuralandfunctionalconnectivity:graphtheoreticalanalysisofaneegneuralmassmodel,Neuroimage,52,3,985{994 Quiroga,R.Q.,Kreuz,T.,andGrassberger,P.(2002),Eventsynchronization:asimpleandfastmethodtomeasuresynchronicityandtimedelaypatterns,PhysicalReviewE,66,4,041904 Raichman,N.,Volman,V.,andBen-Jacob,E.(2006),Collectiveplasticityandindividualstabilityinculturedneuronalnetworks,Neurocomputing,69,10,1150{1154 Rajnicek,A.M.,Robinson,K.R.,andMcCaig,C.D.(1998),Thedirectionofneuritegrowthinaweakdcelectricelddependsonthesubstratum:contributionsofadhesivityandnetsurfacecharge,Developmentalbiology,203,2,412{423 110

PAGE 111

Reyes,A.etal.(2003),Synchrony-dependentpropagationofringrateiniterativelyconstructednetworksinvitro,Natureneuroscience,6,6,593{599 Riehle,A.,Grun,S.,Diesmann,M.,andAertsen,A.(1997),Spikesynchronizationandratemodulationdierentiallyinvolvedinmotorcorticalfunction,Science,278,5345,1950{1953 Rieke,F.(1999),Spikes:exploringtheneuralcode(TheMITPress) Rivieccio,M.A.,Brochier,C.,Willis,D.E.,Walker,B.A.,D'Annibale,M.A.,McLaughlin,K.,etal.(2009),Hdac6isatargetforprotectionandregenerationfollowinginjuryinthenervoussystem,ProceedingsoftheNationalAcademyofSci-ences,106,46,19599{19604 Robinson,H.,Kawahara,M.,Jimbo,Y.,Torimitsu,K.,Kuroda,Y.,andKawana,A.(1993),Periodicsynchronizedburstingandintracellularcalciumtransientselicitedbylowmagnesiuminculturedcorticalneurons,Journalofneurophysiology,70,4,1606{1616 Roebroeck,A.,Formisano,E.,andGoebel,R.(2005),Mappingdirectedinuenceoverthebrainusinggrangercausalityandfmri,Neuroimage,25,1,230{242 Rolls,E.T.,Franco,L.,Aggelopoulos,N.C.,andJerez,J.M.(2006),Informationintherstspike,theorderofspikes,andthenumberofspikesprovidedbyneuronsintheinferiortemporalvisualcortex,VisionResearch,46,25,4193{4205 Rolston,J.,Wagenaar,D.,andPotter,S.(2007),Preciselytimedspatiotemporalpatternsofneuralactivityindissociatedcorticalcultures,Neuroscience,148,1,294{303 Ruiz,A.,Buzanska,L.,Gilliland,D.,Rauscher,H.,Sirghi,L.,Sobanski,T.,etal.(2008),Micro-stampedsurfacesforthepatternedgrowthofneuralstemcells,Biomaterials,29,36,4766{4774 Ruiz,A.,Ceriotti,L.,Buzanska,L.,Hasiwa,M.,Bretagnol,F.,Ceccone,G.,etal.(2007),Controlledmicropatterningofbiomoleculesforcellculturing,Microelectronicengineering,84,5,1733{1736 Sameshima,K.andBaccala,L.(1999),Usingpartialdirectedcoherencetodescribeneuronalensembleinteractions,JournalofNeuroscienceMethods,94,1,93{103 Schwartzkroin,P.A.andWester,K.(1975),Long-lastingfacilitationofasynapticpotentialfollowingtetanizationintheinvitrohippocampalslice,Brainresearch,89,1,107{119 Sederberg,P.B.,Schulze-Bonhage,A.,Madsen,J.R.,Bromeld,E.B.,McCarthy,D.C.,Brandt,A.,etal.(2007),Hippocampalandneocorticalgammaoscillationspredictmemoryformationinhumans,CerebralCortex,17,5,1190{1196 111

PAGE 112

Segev,R.,Baruchi,I.,Hulata,E.,Ben-Jacob,E.,etal.(2004),Hiddenneuronalcorrelationsinculturednetworks,PhysicalReview-SeriesA-,92,11,118102{118300 Serre,T.,Oliva,A.,andPoggio,T.(2007),Afeedforwardarchitectureaccountsforrapidcategorization,ProceedingsoftheNationalAcademyofSciences,104,15,6424{6429 Seth,A.(2010),Amatlabtoolboxforgrangercausalconnectivityanalysis,Journalofneurosciencemethods,186,2,262{273 Shadlen,M.andNewsome,W.(1994),Noise,neuralcodesandcorticalorganization,Currentopinioninneurobiology,4,4,569{579 Shadlen,M.andNewsome,W.(1998),Thevariabledischargeofcorticalneurons:implicationsforconnectivity,computation,andinformationcoding,TheJournalofNeuroscience,18,10,3870{3896 Shadlen,M.N.andMovshon,J.A.(1999),Synchronyunbound:acriticalevaluationofthetemporalbindinghypothesis,Neuron,24,1,67{77 Shahaf,G.andMarom,S.(2001),Learninginnetworksofcorticalneurons,TheJournalofNeuroscience,21,22,8782{8788 Shmiel,T.,Drori,R.,Shmiel,O.,Ben-Shaul,Y.,Nadasdy,Z.,Shemesh,M.,etal.(2006),Temporallyprecisecorticalringpatternsareassociatedwithdistinctactionsegments,JournalofNeurophysiology,96,5,2645{2652 Shriver-Lake,L.C.,Donner,B.,Edelstein,R.,Breslin,K.,Bhatia,S.K.,andLigler,F.S.(1997),Antibodyimmobilizationusingheterobifunctionalcrosslinkers,BiosensorsandBioelectronics,12,11,1101{1106 Shteingart,H.,Raichman,N.,Baruchi,I.,andBen-Jacob,E.(2010),Wrestlingmodeloftherepertoireofactivitypropagationmodesinquadrupleneuralnetworks,Frontiersincomputationalneuroscience,4 Sia,S.K.andWhitesides,G.M.(2003),Microuidicdevicesfabricatedinpoly(dimethylsiloxane)forbiologicalstudies,Electrophoresis,24,21,3563{3576 Sigrist,H.,Collioud,A.,Clemence,J.-F.,Gao,H.,Luginbuehl,R.,Saenger,M.,etal.(1995),Surfaceimmobilizationofbiomoleculesbylight,OpticalEngineering,34,8,2339{2348 Singer,W.(2001),Consciousnessandthebindingproblem,AnnalsoftheNewYorkAcademyofSciences,929,1,123{146 Singer,W.andGray,C.M.(1995),Visualfeatureintegrationandthetemporalcorrelationhypothesis,Annualreviewofneuroscience,18,1,555{586 Single,S.andBorst,A.(1998),Dendriticintegrationanditsroleincomputingimagevelocity,Science,281,5384,1848{1850 112

PAGE 113

Soderquist,M.andWalton,A.(1980),Structuralchangesinproteinsadsorbedonpolymersurfaces,JournalofColloidandInterfaceScience,75,2,386{397 Softky,W.R.andKoch,C.(1993),Thehighlyirregularringofcorticalcellsisinconsistentwithtemporalintegrationofrandomepsps,TheJournalofNeuroscience,13,1,334{350 Sohal,V.andHuguenard,J.(2005),Inhibitorycouplingspecicallygeneratesemergentgammaoscillationsindiversecelltypes,ProceedingsoftheNationalAcademyofSciencesoftheUnitedStatesofAmerica,102,51,18638 Srensen,A.,Alekseeva,T.,Katechia,K.,Robertson,M.,Riehle,M.O.,andBarnett,S.C.(2007),Long-termneuriteorientationonastrocytemonolayersalignedbymicrotopography,Biomaterials,28,36,5498{5508 Sorribas,H.,Padeste,C.,andTiefenauer,L.(2002),Photolithographicgenerationofproteinmicropatternsforneuroncultureapplications,Biomaterials,23,3,893{900 Sporns,O.,Chialvo,D.R.,Kaiser,M.,andHilgetag,C.C.(2004),Organization,developmentandfunctionofcomplexbrainnetworks,Trendsincognitivesciences,8,9,418{425 Sporns,O.,Tononi,G.,andEdelman,G.(2000),Connectivityandcomplexity:therelationshipbetweenneuroanatomyandbraindynamics,NeuralNetworks,13,8-9,909{922 Srinivas,K.,Jain,R.,Saurav,S.,andSikdar,S.(2007),Small-worldnetworktopologyofhippocampalneuronalnetworkislost,inaninvitroglutamateinjurymodelofepilepsy,EuropeanJournalofNeuroscience,25,11,3276{3286 Stiefel,K.M.,Tapson,J.,andvanSchaik,A.(2013),Temporalorderdetectionandcodinginnervoussystems,Neuralcomputation,25,2,510{531 Stopfer,M.,Bhagavan,S.,Smith,B.H.,andLaurent,G.(1997),Impairedodourdiscriminationondesynchronizationofodour-encodingneuralassemblies,Nature,390,6655,70{74 Streit,J.,Tscherter,A.,Heuschkel,M.O.,andRenaud,P.(2001),Thegenerationofrhythmicactivityindissociatedculturesofratspinalcord,Europeanjournalofneuroscience,14,2,191{202 Sun,J.-J.,Kilb,W.,andLuhmann,H.J.(2010),Self-organizationofrepetitivespikepatternsindevelopingneuronalnetworksinvitro,EuropeanJournalofNeuroscience,32,8,1289{1299 Suzuki,I.,Sugio,Y.,Moriguchi,H.,Jimbo,Y.,andYasuda,K.(2004),Modicationofaneuronalnetworkdirectionusingstepwisephoto-thermaletchingofanagarosearchitecture,Journalofnanobiotechnology,2,1,7 113

PAGE 114

Takahashi,N.,Sasaki,T.,Matsumoto,W.,Matsuki,N.,andIkegaya,Y.(2010),Circuittopologyforsynchronizingneuronsinspontaneouslyactivenetworks,ProceedingsoftheNationalAcademyofSciences,107,22,10244{10249 Tateno,T.andJimbo,Y.(1999),Activity-dependentenhancementinthereliabilityofcorrelatedspiketimingsinculturedcorticalneurons,BiologicalCybernetics,80,1,45{55 Tateno,T.,Kawana,A.,andJimbo,Y.(2002),Analyticalcharacterizationofspontaneousringinnetworksofdevelopingratculturedcorticalneurons,PhysicalReviewE,65,5,051924 Taylor,A.,Blurton-Jones,M.,Rhee,S.,Cribbs,D.,Cotman,C.,andJeon,N.(2005),Amicrouidiccultureplatformforcnsaxonalinjury,regenerationandtransport,NatureMethods,2,8,599{605 Taylor,A.,Dieterich,D.,Ito,H.,Kim,S.,andSchuman,E.(2010),Microuidiclocalperfusionchambersforthevisualizationandmanipulationofsynapses,Neuron,66,1,57{68 Taylor,A.,Rhee,S.,Tu,C.,Cribbs,D.,Cotman,C.,andJeon,N.(2003),Microuidicmulticompartmentdeviceforneuroscienceresearch,Langmuir,19,5,1551{1556 Thomas,C.,Springer,P.,Loeb,G.,Berwald-Netter,Y.,Okun,L.,etal.(1972),Aminiaturemicroelectrodearraytomonitorthebioelectricactivityofculturedcells,Experimentalcellresearch,74,1,61{66 Thompson,D.M.andBuettner,H.M.(2006),Neuriteoutgrowthisdirectedbyschwanncellalignmentintheabsenceofotherguidancecues,Annalsofbiomedicalengineering,34,1,161{168 Tourovskaia,A.,Li,N.,andFolch,A.(2008),Localizedacetylcholinereceptorclusteringdynamicsinresponsetomicrouidicfocalstimulationwithagrin,Biophysicaljournal,95,6,3009{3016 Uhlhaas,P.,Pipa,G.,Lima,B.,Melloni,L.,Neuenschwander,S.,Nikolic,D.,etal.(2009),Neuralsynchronyincorticalnetworks:history,conceptandcurrentstatus,Frontiersinintegrativeneuroscience,3,17 Vajda,I.,VanPelt,J.,Wolters,P.,Chiappalone,M.,Martinoia,S.,VanSomeren,E.,etal.(2008),Low-frequencystimulationinducesstabletransitionsinstereotypicalactivityincorticalnetworks,Biophysicaljournal,94,12,5028{5039 VanDenPol,A.,Obrietan,K.,andBelousov,A.(1996),Glutamatehyperexcitabilityandseizure-likeactivitythroughoutthebrainandspinalcorduponrelieffromchronicglutamatereceptorblockadeinculture,Neuroscience,74,3,653{674 VanHuizen,F.,Romijn,H.,andHabets,A.(1985),Synaptogenesisinratcerebralcortexculturesisaectedduringchronicblockadeofspontaneousbioelectricactivitybytetrodotoxin,DevelopmentalBrainResearch,19,1,67{80 114

PAGE 115

VanPelt,J.,Corner,M.,Wolters,P.,Rutten,W.,andRamakers,G.(2004),Longtermstabilityanddevelopmentalchangesinspontaneousnetworkburstringpatternsindissociatedratcerebralcortexcellculturesonmultielectrodearrays,Neuroscienceletters,361,1,86{89 vanRossum,M.,Turrigiano,G.,andNelson,S.(2002),Fastpropagationofringratesthroughlayerednetworksofnoisyneurons,TheJournalofneuroscience,22,5,1956{1966 vanRossum,M.C.(2001),Anovelspikedistance,NeuralComputation,13,4,751{763 Victor,J.D.andPurpura,K.P.(1996),Natureandprecisionoftemporalcodinginvisualcortex:ametric-spaceanalysis,JournalofNeurophysiology,76,2,1310{1326 Victor,J.D.andPurpura,K.P.(1997),Metric-spaceanalysisofspiketrains:theory,algorithmsandapplication,Network:computationinneuralsystems,8,2,127{164 Vinck,M.,Lima,B.,Womelsdorf,T.,Oostenveld,R.,Singer,W.,Neuenschwander,S.,etal.(2010),Gamma-phaseshiftinginawakemonkeyvisualcortex,TheJournalofNeuroscience,30,4,1250{1257 Vogels,T.andAbbott,L.(2009),Gatingmultiplesignalsthroughdetailedbalanceofexcitationandinhibitioninspikingnetworks,Natureneuroscience,12,4,483{491 Vogt,A.,Brewer,G.,andOenhausser,A.(2005a),Connectivitypatternsinneuronalnetworksofexperimentallydenedgeometry,Tissueengineering,11,11-12,1757{1767 Vogt,A.,Wrobel,G.,Meyer,W.,Knoll,W.,andOenhausser,A.(2005b),Synapticplasticityinmicropatternedneuronalnetworks,Biomaterials,26,15,2549{2557 Volman,V.,Baruchi,I.,andBen-Jacob,E.(2005),Manifestationoffunction-follow-forminculturedneuronalnetworks,Physicalbiology,2,2,98 VonDerMalsburg,C.(1985),Nervousstructureswithdynamicallinks,BerichtederBunsengesellschaftfurphysikalischeChemie,89,6,703{710 VonDerMalsburg,C.(1994),Thecorrelationtheoryofbrainfunction,Modelsofneuralnetworks,2,95{119 Wadhwa,G.(1990),Biochipsandbiocomputers-futureofcomputingandmedicine,JournalofScientic&IndustrialResearch,49,10,486{491 Wagenaar,D.,DeMarse,T.B.,andPotter,S.M.(2005a),Meabench:Atoolsetformulti-electrodedataacquisitionandon-lineanalysis,inNeuralEngineering,2005.ConferenceProceedings.2ndInternationalIEEEEMBSConferenceon(IEEE),518{521 Wagenaar,D.,Madhavan,R.,Pine,J.,andPotter,S.(2005b),Controllingburstingincorticalcultureswithclosed-loopmulti-electrodestimulation,TheJournalofneuro-science,25,3,680{688 115

PAGE 116

Wagenaar,D.,Pine,J.,andPotter,S.(2004),Eectiveparametersforstimulationofdissociatedculturesusingmulti-electrodearrays,Journalofneurosciencemethods,138,1,27{37 Wagenaar,D.,Pine,J.,andPotter,S.(2006),Anextremelyrichrepertoireofburstingpatternsduringthedevelopmentofcorticalcultures,BMCneuroscience,7,1,11 Wagenaar,D.andPotter,S.(2002),Real-timemulti-channelstimulusartifactsuppressionbylocalcurvetting,Journalofneurosciencemethods,120,2,113{120 Wang,X.andBuzsaki,G.(1996),Gammaoscillationbysynapticinhibitioninahippocampalinterneuronalnetworkmodel,ThejournalofNeuroscience,16,20,6402{6413 Wheeler,B.,Corey,J.,Brewer,G.,andBranch,D.(1999),Microcontactprintingforprecisecontrolofnervecellgrowthinculture.,Journalofbiomechanicalengineering,121,1,73{78 Wheeler,B.C.andBrewer,G.J.(2010),Designingneuralnetworksinculture,Proceed-ingsoftheIEEE,98,3,398{406 Whittington,M.,Traub,R.,Kopell,N.,Ermentrout,B.,andBuhl,E.(2000),Inhibition-basedrhythms:experimentalandmathematicalobservationsonnetworkdynamics,InternationalJournalofPsychophysiology,38,3,315{336 Wickens,J.,Hyland,B.,andAnson,G.(1994),Corticalcellassemblies:apossiblemechanismformotorprograms,JournalofMotorBehavior,26,2,66{82 Wilson,C.(2007),Gabaergicinhibitionintheneostriatum,Progressinbrainresearch,160,91{110 Wood,M.D.andWillits,R.K.(2009),Appliedelectriceldenhancesdrgneuritegrowth:inuenceofstimulationmedia,surfacecoatingandgrowthsupplements,Journalofneuralengineering,6,4,046003 Wyart,C.,Ybert,C.,Bourdieu,L.,Herr,C.,Prinz,C.,andChatenay,D.(2002),Constrainedsynapticconnectivityinfunctionalmammalianneuronalnetworksgrownonpatternedsurfaces,Journalofneurosciencemethods,117,2,123{131 Yan,M.,Cai,S.X.,Wybourne,M.,andKeana,J.F.(1993),Photochemicalfunctionalizationofpolymersurfacesandtheproductionofbiomolecule-carryingmicrometer-scalestructuresbydeep-uvlithographyusing4-substitutedperuorophenylazides,JournaloftheAmericanChemicalSociety,115,2,814{816 Yang,I.H.,Siddique,R.,Hosmane,S.,Thakor,N.,andHoke,A.(2009),Compartmentalizedmicrouidiccultureplatformtostudymechanismofpaclitaxel-inducedaxonaldegeneration,Experimentalneurology,218,1,124{128 Zeki,S.(1993),AVisionoftheBrain(OxfordUnivPress) 116

PAGE 117

Zeng,H.-C.,Ho,Y.-C.,Chen,S.-T.,Wu,H.-I.,Tung,H.-W.,Fang,W.-L.,etal.(2007),Studyingtheformationoflargecellaggregatesinpatternedneuronalcultures,Journalofneurosciencemethods,165,1,72{82 Zhou,Z.,Ding,M.,Chen,Y.,Wright,P.,Lu,Z.,andLiu,Y.(2009),Detectingdirectionalinuenceinfmriconnectivityanalysisusingpcabasedgrangercausality,Brainresearch,1289,22{29 117

PAGE 118

BIOGRAPHICALSKETCH SankaraleengamAlagapanwasborninMadurai,Indiain1985.HegraduatedwithaBachelorofEngineeringinelectricalandelectronicsengineeringfromAnnaUniversityinMay2007andaMasterofScienceinbiomedicalEngineeringfromUniversityofFloridainDecember2008.HeenrolledinthePhDprogramofJCraytonPruittFamilyDepartmentofBiomedicalEngineeringatUniversityofFloridaworkingwithDr.BruceWheelerandreceivedhisPh.DinAugust2014. 118