A computational model of mechanisms controlling experience-dependent reorganization of representational maps in auditory cortex

Cognitive Affective amp Behavioral Neuroscience2001 1 (1) 37-55

The dynamics of cortical processing are mediated byneuromodulatory mechanisms Neuromodulatory chem-icals are known to affect the sensitivities of cortical neu-rons to incoming signals the modifiability of synapticconnections between neurons and the oscillatory prop-erties of cortical networks (for a review see Hasselmo1995) How such effects are related to forming maintain-ing retrieving or adapting neural representations is poorlyunderstood Computational models of cortical process-ing provide a useful theoretical framework for examiningthe roles neuromodulators play in the representation andtransformation of sensory information (for reviews seedeCharms amp Zador 2000 Fellous amp Linster 1998 Gluckamp Myers 2001)

Past computational models of processing in sensorycortices have often focused on replicating network dy-namics and single-cell response properties Many mod-els have been proposed for describing how cortical sen-sitivities change during development and after injuries toreceptors (see eg Erwin Obermayer amp Schulten 1995Grajski amp Merzenich 1990 Sirosh amp Miikkulainen 1997

Sutton Reggia Armentrout amp DrsquoAutrechy 1994 Swin-dale 1996 Swindale amp Bauer 1998 Tanaka 1990) Lessattention has been given to modeling changes in sensoryrepresentations induced by experience (Armony Servan-Schreiber Cohen amp LeDoux 1995 Barkai BergmanHorowitz amp Hasselmo 1994 Benukskova Diamond ampEbner 1994 Elliott amp Shadbolt 1998 Joublin Spen-gler Wacquant amp Dinse 1996)

Although experience-dependent changes in the re-sponse properties of sensory cortical neurons have beenextensively investigated (for recent reviews see Buono-mano amp Merzenich 1998 Edeline 1999 Kaas 1997)the mechanisms underlying cortical plasticity in adultsremain unclear The most detailed theoretical model oflearning-induced plasticity developed to date describeshow associative conditioning can change spectral sensi-tivities in the auditory cortex (Weinberger et al 1990aWeinberger et al 1990b Weinberger amp Bakin 1998)This model specifies neuromodulatory actions that con-trol how and when cortical representations will becomereorganized

Qualitative models of neuromodulatory effects in com-bination with experimental data collected to test the pre-dictions of these models have led to significant progresstoward understanding auditory cortical plasticity Thesemodels are limited however in terms of how preciselythey can predict the effects that particular experimentaltreatments will have on cortical representations of soundMore quantitative theoretical frameworks are needed in

37 Copyright 2001 Psychonomic Society Inc

This work was supported by an APANIMH postdoctoral fellowshipand by the RutgersndashNewark MBRS program The authors thank twoanonymous reviewers of the manuscript for their insightful commentsand suggestions Correspondence concerning this article should be ad-dressed to E Mercado Center for Molecular and Behavioral Neuro-science Rutgers University 197 University Ave Newark NJ 07102(e-mail mercadopavlovrutgersedu)

A computational model of mechanismscontrolling experience-dependent reorganization

of representational maps in auditory cortex

EDUARDO MERCADO III CATHERINE E MYERS and MARK A GLUCKRutgers University Newark New Jersey

Cortical representations of sound can be modified by repeatedly pairing presentation of a pure tonewith electrical stimulation of neuromodulatory neurons located in the basal forebrain (Bakin amp Wein-berger 1996 Kilgard amp Merzenich 1998a) We developed a computational model to investigate the pos-sible effects of basal forebrain modulation on map reorganization in the auditory cortex The model isa self-organizing map with acoustic response characteristics mimicking those observed in the mam-malian auditory cortex We simulated the effects of basal forebrain modulation using parameters in-trinsic to the self-organizing map such as the learning rate (controlling the adaptability of map nodes)and the neighborhood function (controlling the excitability of map nodes) Previous research has sug-gested that both parameters can be useful for characterizing the effects of neuromodulation on plas-ticity (Kohonen 1993 Myers et al 1996 Myers Ermita Hasselmo amp Gluck 1998) The model suc-cessfully accounts for experimentally observed effects of pairing basal forebrain stimulation with thepresentation of a single tone but not of two tones suggesting that auditory cortical plasticity is con-strained in ways not accounted for by current theories Despite this limitation the model provides auseful framework for describing experience-induced changes in auditory representations and for re-lating such changes to variations in the excitability and adaptability of cortical neurons produced byneuromodulation

38 MERCADO MYERS AND GLUCK

order to rigorously test the validity of current theoreticalproposals concerning the role that neuromodulators playin cortical plasticity Previous simulations of experience-dependent changes in auditory cortical processing (egArmony et al 1995) have not explicitly considered howneuromodulatory systems and preexisting topographicalcortical sensitivities might affect the reorganization ofauditory representations Such variables can critically af-fect cortical reorganization (M E Diamond Petersen ampHarris 1999 Dykes 1997)

In this paper we describe a connectionist model of au-ditory cortical processing that can be used to quantitativelydescribe the effects of neuromodulation on experience-dependent cortical plasticity We constrained the organi-zation and sensitivities of a neural network so that theyparalleled properties seen in the mammalian auditorycortex and modeled the effects of modulatory processeson cortical reorganization using parameters intrinsic tothe learning algorithm used by the neural network Weinvestigated how variations in parameter values affectedexperience-induced changes in the responses of the neuralnetwork and compared these changes with recent electro-

physiological descriptions of reorganization in the audi-tory cortex induced by pairing electrical stimulation ofthe basal forebrain with the presentation of tones

Our goal was to develop a simple computational modelof auditory cortical processing that could provide a pre-cise theoretical framework for (1) investigating the rolesof neuromodulation in experience-dependent reorgani-zation of auditory representations and (2) testing the ac-curacy and sufficiency of more qualitative models of experience-induced plasticity in the auditory cortex Pre-liminary reports of some of these data appeared in abstractform (Mercado Myers amp Gluck 1999a 1999b)

A COMPUTATIONAL MODEL OFAUDITORY CORTICAL PROCESSING

Numerous cortical models of varying computationalcomplexity and physiological realism have been devel-oped (see eg Abbott amp Sejnowski 1999 Arbib 1998Bower amp Beeman 1998) Two general classes of auditorycortex models are prevalent signal-processing models andneural network models (see Figure 1) Signal-processing

Figure 1 Models of audition CN cochlear nucleus IC inferior colliculus MGB medialgeniculate body The biological model involves a multimodule loop composed of converging anddiverging neural connections Signal-processing models typically involve sequential transfor-mations of a one-dimensional waveform x(t ) measuring variations in pressure over time Firstx(t ) is transformed by a set of bandpass filters simulating cochlear processing Then these sig-nals are normalized simulating processing by hair cells and auditory ganglia to generate an au-ditory spectrum Finally the auditory spectrum is transformed into an n-dimensional spacey(t fn) that describes features of the sound such as pitch and signal bandwidth Neural networkmodels are typically much more abstract Inputs are often represented as binary vectors that codemeasured properties of sounds and outputs often reflect activity levels of nodes Transitions be-tween stages (or layers) involve matrix transformations Signal-processing models of audition usu-ally involve static feedforward processing whereas neural network models are usually adaptive

MODEL OF AUDITORY CORTICAL REORGANIZATION 39

models of auditory cortical processing characterizetransformations performed by cortical circuits ratherthan detailed response properties of cortical neurons(see eg K Wang amp Shamma 1995a) Most of thesemodels involve nonadaptive processing of acoustic eventsFor example Suga (1995) has modeled the auditory sys-tem of bats as a sequence of hierarchically organized fil-ters In contrast neural network models of auditory cor-tical processing are generally adaptive (ie their responsesensitivities are affected by training) making them use-ful for simulating experience-induced adaptation in theauditory cortex These models either can be biologicallybased focusing on the dynamics of cellular interactions(de Pinho amp Roque-da-Silva 1999 Elliot amp Shadbolt1998 Palakal amp Wong 1999 Sanchez-Montanes Ver-schure amp Konig 2000) or may characterize corticalprocessing more abstractly (Armony et al 1995 MyersGluck amp Granger 1995) Neural network models typi-cally do not incorporate the types of transformations de-scribed by signal-processing models of audition Inputsto neural networks usually consist of either binary pat-terns (Armony et al 1995) or numerical values describ-ing properties of sounds or neural responses (Bauer Deramp Herrmann 1996 Guenther amp Gjaja 1996 PalakalMurthy Chittajallu amp Wong 1995 Ritter Martinez ampSchulten 1992)

Most computational models of auditory cortical pro-cessing in the past have involved nonadaptive filteringof inputs (Palakal et al 1995 Suga 1990 K Wang amp

Shamma 1995a) and thus have not accounted for theplasticity seen in adults The few modelers that have at-tempted to characterize auditory plasticity have focusedon changes in the sensitivities of individual neurons (Ar-mony et al 1995 Armony Servan-Schreiber Cohen ampLeDoux 1997 Armony Servan-Schreiber RomanskiCohen amp LeDoux 1997 de Pinho Mazza amp Roque2000 Sanchez-Montanes et al 2000) These models(with the exception of de Pinho Mazza amp Roque 2000)do not address how the initial spatial organization ofcortical response properties might affect cortical map re-organization

Our approach was to take a neural network model thathas served as the basis for many past cortical models (theself-organizing map) and integrate it with advanced sig-nal processing models of auditory representation (Fig-ure 2) The computational structure of the self-organizingmap is straightforward and has been studied extensively(for reviews see Kohonen 1993 1997 Ritter et al1992) This connectionist model has a highly flexible ar-chitecture that can be easily customized to model pro-cessing in sensory cortices (see eg Erwin et al 1995Palakal et al 1995 Ritter et al 1992 Sirosh amp Miik-kulainen 1997 Swindale amp Bauer 1998) Signal-processing models of auditory representations have ad-vanced considerably over the past decade (see egMeyer-Base amp Scheich 1995 Pitton Wang amp Juang1996 Robert amp Eriksson 1999 Tchorz amp Kollmeier1999 K Wang amp Shamma 1995a) Such models pro-

Figure 2 Models of basal forebrain modulation of cortical plasticity In the self-organizing map modelmodulation by neurons in the nucleus basalis (NB) is simulated as changes in the area of the map acti-vated by the most responsive node andor by changes in the rate at which response properties adapt Inthe Weinberger et al (1990a 1990b) model NB modulation increases the excitability and adaptability ofcortical neurons when behaviorally relevant events are occurring


vide a way to precisely characterize how auditory corti-cal response sensitivities are likely to change in responseto learning (Mercado Myers amp Gluck 2000)

To review the self-organizing map consists of ann-dimensional matrix of nodes each of which can po-tentially be activated by external inputs (represented asvectors) Inputs are sent to every node in the map viaweighted connections (which are also represented as vec-tors) Whether a node is activated by a particular inputdepends on how well the input matches the weights as-sociated with that node The basic self-organizing map-learning algorithm is as follows (detailed mathematicalformulations can be found in Kohonen 1997) (1) Choosean input vector at random from the set of all input vec-tors (2) measure the similarity between this input vectorand all the weight vectors (3) find the node with the mostsimilar weight vector (4) update the weights of this nodeand nearby nodes so that they are more similar to theinput vector and (5) continue this process until all the in-puts have been presented to the map

Input sets are typically presented to a self-organizingmap several times The number of surrounding nodes af-fected by activation of the best-matching node is definedin terms of a neighborhood function Two parameters inthe self-organizing map-learning algorithm can be usedto control the adjustment of node weights the learningrate function and the range of the neighborhood function(called the neighborhood size) Both parameters are usu-ally decreased as training proceeds so that early adjust-ments to weights are larger and have a more global effectthan do later adjustments The response properties ofself-organizing maps become topographically organizedwith training so that nearby nodes respond maximallyto similar inputs

Self-organizing maps with biologically based responsecharacteristics can emulate the spatial organization of re-sponse properties observed in the auditory cortex aswell as the competitive adaptation currently theorized tounderlie changes in auditory cortical organization

EFFECTS OF NEUROMODULATION ONAUDITORY CORTICAL REORGANIZATION

Kohonen (1987 1993) suggested that the adaptiveprocesses underlying plasticity in biological neural net-works may directly parallel the self-organizing map al-gorithm He postulated that active cells can control adap-tation in surrounding cells by extracellular transmissionof diffuse chemical agents The number of neighboringcells affected would depend on the extent and amount ofchemical agent transmitted providing a possible chemi-cal implementation of the neighborhood function usedin self-organizing maps

Although diffuse intracortical transmission of neuro-modulatory chemicals may provide a feasible mecha-nism for localized cortical plasticity (see eg Ahissaret al 1992 Cruikshank amp Weinberger 1996b Maldon-ado amp Gerstein 1996 Zoli et al 1998) neuromodula-

tory systems originating outside the cortex provide an al-ternative source of diffuse chemical agents that can reg-ulate reorganization in sensory cortices Several neuro-modulatory systems have been implicated in sensoryprocessing and cortical plasticity (for reviews see Ede-line 1999 Hasselmo 1995 Zaborszky Pang SomogyiNadasdy amp Kallo 1999) three of which have been im-plicated in auditory cortical reorganization the basal fore-brain cholinergic system (Bakin amp Weinberger 1996Edeline Hars Maho amp Hennevin 1994 Hars MahoEdeline amp Hennevin 1993 Metherate amp Ashe 19911993) the mesencephalic dopaminergic system (Bao ampMerzenich 2000 Stark amp Scheich 1997) and the nor-adrenergic system (Edeline 1995 Manunta amp Edeline1997) These neuromodulatory systems consist of con-centrations of neurons with axons that project throughoutthe cortex Many sites of neuromodulator release in theaxons of these neurons are not associated with clearly iden-tifiable synapses suggesting that extrinsic neuromodu-lation may affect cortical activity at least in part throughdiffuse chemical control Neuromodulators released fromneurons in the basal forebrain and brain stem may thusmediate local activity-dependent changes in corticaladaptability that are qualitatively similar to those con-trolled by neighborhood functions in self-organizing maps

Past attempts at explaining the role that neuromodu-lators play in auditory cortical plasticity have focusedprimarily on modulation by neurons projecting from thenucleus basalis a subpopulation of neurons in the basalforebrain For example Weinberger et al (1990a Wein-berger et al 1990b) proposed that acetylcholine re-leased into the auditory cortex by the nucleus basalisamplifies inputs from the thalamus by enhancing post-synaptic activation throughout the cortex (see Figure 2)They suggested that synaptic connections from activethalamic cells to cortical pyramidal cells are strength-ened and that nonactive synapses are weakened (ieadaptation occurs via Hebbian mechanisms) Dykes(1997) proposed a similar qualitative model to explainhow basal forebrain modulation affects reorganization inthe somatosensory cortex In this model acetylcholine-induced excitation combines with GABA-induced disin-hibition to create a permissive cortical state that facili-tates ldquoLTP-likerdquo synaptic plasticity

These qualitative models are supported by recentneurophysiological studies showing that the nucleusbasalis plays a prominent modulatory role in experience-dependent cortical plasticity (Bakin amp Weinberger 1996Baskerville Schweitzler amp Herron 1997 Bjordahl Dim-yan amp Weinberger 1998 Dimyan amp Weinberger 1999Kilgard amp Merzenich 1998a Sachdev Shao-Ming Wileyamp Edner 1998) For example when electrical stimula-tion of the nucleus basalis is repeatedly paired with pre-sentation of a sound auditory cortical neurons adapt sothat the cortical area responsive to that sound is greatlyexpanded (Kilgard amp Merzenich 1998a Mercado Sho-hamy Orduntildea Gluck amp Merzenich 2000) In contrastbasal forebrain lesions significantly reduce experience-


dependent plasticity (Baskerville et al 1997 JulianoMa amp Eslin 1991 Kilgard amp Merzenich 1998a Sach-dev et al 1998 Webster Hanisch Dykes amp Biesold1991)

Researchers studying the role of basal forebrain mod-ulation in cortical plasticity have typically focused oncholinergic effects Increasing the amount of acetyl-choline in the sensory cortex generally facilitates evokedresponses (Ashe amp Weinberger 1991 Edeline et al1994 Hars et al 1993 Metherate amp Ashe 1993 Meth-erate Ashe amp Weinberger 1990 Metherate Tremblayamp Dykes 1987 Tremblay Warren amp Dykes 1990 Web-ster Rasmusson et al 1991) whereas blocking or elim-inating cholinergic effects reduces responsiveness (Dela-cour Houcine amp Costa 1990 Edeline et al 1994 JacobsCode amp Juliano 1991 Maalouf Miasnikov amp Dykes1998 Metherate amp Ashe 1991 1993) In addition changesin cortical sensitivities can be induced by pairing the ap-plication of acetylcholine (or drugs that mimic cholinergicactions) with the presentation of tones (Ashe McKennaamp Weinberger 1989 McKenna Ashe amp Weinberger1989 Metherate amp Weinberger 1989 1990 Shulz Sos-nik Ego Haidarliu amp Ahissar 2000) Cholinergic effectson cortical plasticity are activity dependent For exam-ple basal forebrain stimulation without the presentationof sound does not lead to long-lasting facilitation of evokedresponses (Bakin amp Weinberger 1996 Edeline et al1994 Hars et al 1993)

Although cholinergic neurons in the nucleus basalisare a primary source of cortical acetylcholine recentanatomical studies suggest that cholinergic neurons rep-resent a relatively small fraction of the modulatory cellsprojecting from the nucleus basalis to sensory corticesThe majority of projecting neurons appear to be eitherGABAergic or peptidergic (Gritti Mainville Mancia ampJones 1997 Zaborszky et al 1999) Auditory corticalsensitivities can be substantially modified by blockingthe effects of GABA (Foeller Vater amp Kossl 2000Schulze amp Langner 1999 J Wang Caspary amp Salvi2000) providing support for Dykesrsquos (1997) proposalthat GABAergic modulation can significantly affect cor-tical reorganization (see also Giorgetti et al 2000Jiminez-Capdeville Dykes amp Myasnikov 1997) Theeffects of peptides on auditory cortical plasticity if anyare unknown

In summary neuromodulators released in the auditorycortex by basal forebrain and brain stem neurons appearto control experience-dependent changes in neuronalsensitivities Although the mechanisms underlying neuro-modulatory effects on cortical plasticity remain conjec-tural electrophysiological measurements provide clearevidence that basal forebrain neurons in particular modu-late both the activity and the adaptability of auditory cor-tical neurons Because the roles of other neuromodulatorysystems in cortical plasticity have been investigated in muchless detail the present simulations focus on modeling theeffects of basal forebrain modulation on experience-dependent changes in cortical representations

MODELING BASALFOREBRAIN MODULATION

How can the effects of basal forebrain modulation beincorporated within our model of auditory cortical pro-cessing Assume that in certain contexts basal forebrainneurons control when where and to what extent corticalneurons change their response properties In the self-organizing map where is determined by input featuresand the neighborhood size when is determined by inputfeatures and the learning rate and to what extent is de-termined by the neighborhood size and the learning rateThus adjusting the neighborhood size and the learningrate in our simulated auditory cortex affects reorganizationin ways that qualitatively parallel the activity-dependenteffects of basal forebrain modulation on reorganizationin the auditory cortex

In the qualitative models proposed by Weinberger et al(1990a 1990b) and Dykes (1997) (see also Gao amp Suga2000) neuromodulators affect auditory cortical plastic-ity by changing the likelihood that signals from cochlearreceptors andor intracortical connections will activatecortical neurons Increased or decreased activation withinparticular cortical regions is then hypothesized to control(via Hebbian processes) how synaptic connections inthose regions are modified (see also Ahissar Abeles Ahis-sar Haidarliu amp Vaadia 1998 Fregnac amp Shulz 1999)Our model dissociates this sequence of events into twosubprocesses excitation and adaptation When neuronalexcitability is increased in the auditory cortex largerpopulations of neurons respond to the presentation of apure tone (see eg J Wang et al 2000) This increasein responsive area is similar to that produced by increas-ing the intensity of a tone (Bakin Kwon Masino Wein-berger amp Frostig 1996) Paralleling these effects wemodel changes in excitability in terms of changes in neigh-borhood size Increases in neuronal adaptability havebeen qualitatively modeled as changes in synaptic plas-ticity based on modified Hebbian rules Similarly we usechanges in learning rate to model changes in synapticplasticity induced by basal forebrain modulation

MethodIn the present study we simulated plasticity in pri-

mary auditory cortex using the SOM Toolbox developedby students of Kohonen at the Laboratory of Informationand Computer Science in the Helsinki University of Tech-nology The SOM Toolbox is a software library for Mat-lab 5 that implements the self-organizing map algorithmWe used an 11 3 20 node two-dimensional map to grosslyapproximate the spatial organization of the primary audi-tory cortex in various mammals (as described in Aitkin1990 Recanzone Schreiner Sutter Beitel amp Merzen-ich 1999 Schreiner 1998 K Wang amp Shamma 1995a)and for computational tractability The local topographyof the map was hexagonalmdashthat is each node was sur-rounded by 6 other nodes A Gaussian-shaped neighbor-hood function was used Individual nodes in the self-


organizing map simulate cortical regions rather thanindividual neurons Recent research suggests that indi-vidual neurons within local regions respond similarly toacoustic stimuli and that it is likely that these clusters ofneurons serve as coherent functional processing units(see eg deCharms amp Merzenich 1996 M E Dia-mond amp Ebner 1990 X Wang Merzenich Beitel ampSchreiner 1995)

The weights associated with each node in a self-organizing map are typically initialized either to randomvalues or to values chosen on the basis of the linear sub-space spanned by the eigenvectors of the input data setThis is not the normal initial state of the primary audi-tory cortex in adult mammals however Sensitivities inthe mammalian auditory cortex are moderately predict-able Researchers have mapped out how cortical neuronsin particular areas respond to different spectral and tem-poral components of sounds (for reviews see Ehret1997 Merzenich amp Schreiner 1992 Schreiner Read amp

Sutter 2000 K Wang amp Shamma 1995b) We initializedthe self-organizing map to emulate previously reportedspatial properties and spectral sensitivities of the pri-mary auditory cortex (see Figure 3) Specifically responsesensitivities were initialized to be cochleotopic (ie thetopography of cortical sensitivities parallels the spatialarrangement of cochlear receptors) and nodes in the in-ner regions of the map were initialized to respond to anarrower band of frequencies than do nodes in the outeredges (see eg Recanzone et al 1999 Schreiner 1998Schreiner et al 2000 K Wang amp Shamma 1995a) Notethat the organization of map response properties in ourmodel is grossly oversimplified in comparison with ac-tual cortical sensitivities which can vary greatly acrossindividuals species and recording contexts

Simulations were performed to examine how map re-organization was affected by variation in four parametersnumber of input exposures learning rate neighborhoodsize and number of different tones presented to a map

1 kHz 5 kHz

D

V

D

V

Frequency Response Properties of Initialized Map Nodes

1 kHz 5 kHz 1 kHz 5 kHzFrequency Frequency

(A)

(B) (C)

WIDEBAND

WIDEBAND

NARROWBAND

ISOFREQUENCY CONTOUR

How MapRespondsto 2 kHz


Figure 3 Frequency response properties of initialized map nodes (A) Each curve represents the fre-quency response of an initialized node in the self-organizing map (5 of the 11 rows of nodes are shown)Middle rows respond to a narrower band of frequencies than do nodes located either dorsal (D) or ven-tral (V) to the middle This organization makes the initial response properties of the map symmetricalreflecting the ldquoredundantrdquo spectral sensitivities measured electrophysiologically in the auditory cortex(B) Activity levels across the map in response to a 2-kHz tone (darker regions correspond to higher ac-tivity levels) The dark vertical band indicates a column of nodes activated by the 2-kHz tone similarbands (called isofrequency contours) have been observed in the auditory cortex (C) Activity levels inresponse to a 4-kHz tone


(see Table 1) Increases in the number of times a map wasexposed to a particular input simulate increases in thenumber of times a particular sound is paired with activa-tion of basal forebrain neurons The learning rate param-eter controls the extent to which connections to activenodes are changed in response to each input presentationIncreases in this parameter simulate enhanced neuraladaptability (see also Myers et al 1996 Myers et al1998) Neighborhood size determines the number of sur-rounding nodes activated by a ldquowinningrdquo node Increasesin this parameter simulate enhanced neural excitability(see also Piepenbrock amp Obermayer 2000)

Previous simulations of experience-dependent audi-tory cortical plasticity have typically focused on changesin the response properties of single cells induced by clas-sical conditioning (Armony et al 1995 de Pinho Mazzaamp Roque 2000) Such changes can be produced after asmall number (5ndash30) of training trials (Bakin amp Wein-berger 1990 Edeline amp Weinberger 1993a) Althoughobservations of rapid learning-induced changes in cor-tical sensitivities suggest that concomitant rapid changesin the topography of auditory cortical maps might also beoccurring (Weinberger et al 1990a 1990b) this possi-bility has yet to be verified experimentally Changes inthe topographical organization of representational mapshave been demonstrated however after larger numbers(600ndash40000+) of conditioning trials (Gonzalez-Lima ampScheich 1986 Recanzone Schreiner amp Merzenich1993) and after pairing basal forebrain stimulation withthe presentation of tones for thousands of trials (Kilgardamp Merzenich 1998a) Our simulations focus on reorga-nization in simulated auditory cortical maps that are in-duced by relatively large numbers of repeated exposuresto a particular sound It is important to note howeverthat the number of training iterations used in our simu-lations currently cannot be directly equated either withthe number of conditioning trials in a behavioral task orwith the number of pairings used in a stimulation exper-iment Thus 1000 training iterations could potentiallyproduce changes comparable with those observed afteronly 10 conditioning trials

To assess the utility of our model we compared changesobserved in our simulations with changes in cortical sen-sitivities induced by pairing presentations of tones withbasal forebrain stimulation in rats Specifically param-eters were modulated in an attempt to grossly replicatechanges in the organization of the auditory cortex reportedby Kilgard and Merzenich (1998a) Kilgard and Merzen-ich (1998a) stimulated rats 300ndash500 times a day for

7ndash25 days for a total of about 2500ndash12500 trials Forsome rats stimulation was paired with the presentationof a 9-kHz tone in all the trials whereas for other ratsstimulation was paired with a 9-kHz tone in half the tri-als and a 19-kHz tone in the remaining trials (trials of eachtype were randomly intermixed) Similarly self-organizingmaps (with varying learning rates and neighborhood sizes)were exposed to either a single 2-kHz pure tone or twodifferent tones (2 kHz and 4 kHz) 1250ndash 5000 times

Changes in the sensitivities of self-organizing mapswere measured in terms of differences in the responsecharacteristics of trained nodes relative to initializednodes These measurements allow qualitative compar-isons to be made between the results of our simulationsand the data reported by Kilgard amp Merzenich (1998a)and quantitative comparisons to be made across simula-tions Our simulations also allow us to make detailed pre-dictions about how the topography of neuronal sensitivi-ties should change depending on how basal forebrainmodulation affects neural excitability and adaptability

AUDITORY CORTICAL PLASTICITYEMPIRICAL AND MODEL DATA

Changes in Spectral Receptive Fieldsin Rats and Neural Networks

Plasticity in the auditory cortex is commonly describedin terms of changes in neural firing rates induced by pre-sentation of short-duration tonal sounds (tone pips) Neu-rons typically fire most strongly to tones within a par-ticular range of frequencies Thus each neuron can bedescribed in terms of its frequency response function (orspectral receptive field) The frequency of the lowest in-tensity tone pip that consistently produces a change infiring rate is called the characteristic frequency and thefrequency of the tone pip that produces the greatest changein firing rate is called the best frequency

Several learning- and stimulation-induced changes inspectral receptive f ields have been described Thesechanges are often reported relative to a particular sound(eg a tone pip of a particular frequency) that has ac-quired relevance or irrelevance as a predictor of an aver-sive event (eg an electric shock) or basal forebrain stim-ulation For example most studies of learning-inducedplasticity have compared changes in neuronal responsesto a conditioned stimulus (eg a tone pip that has beenrepeatedly paired with shock) with changes in responsesto other frequencies that have no predictive relevance(Bakin amp Weinberger 1990 Edeline Neuenschwander-El Massioui amp Dutrieux 1990 Edeline amp Weinberger1993b Ohl amp Scheich 1996 1997 Weinberger Javid ampLepan 1993) A smaller number of studies have examinedchanges in responses induced by pairing several differ-ent sounds with basal forebrain activation (Kilgard ampMerzenich 1998a 1998b)

Details of previous findings on experience-inducedchanges in receptive fields will not be described herebecause these data have been extensively reviewed (see

Table 1Parameter Values Used in Simulations of

Auditory Cortical Reorganization

Level Exposures Learning Rate Neighborhood Size Tones

Low 1250 0005 2 1Medium 2500 005 5 2High 5000 05 10 ndash


eg Ahissar amp Ahissar 1994 Edeline 1999 Kaas 1996Rauschecker 1999 Weinberger 1993 1995 1997 1998aWeinberger amp Bakin 1998 Weinberger amp Diamond1987) A wide variety of experience-dependent changesin spectral sensitivities can be induced by either basal fore-brain stimulation or behavioral training including (1) in-creased responses to a conditioned tone pip (Bakin ampWeinberger 1990 D M Diamond amp Weinberger 1986Edeline amp Weinberger 1993a Weinberger et al 1993)(2) decreased responses to a conditioned tone pip (D MDiamond amp Weinberger 1984 Kraus amp Disterhoft1982 Ohl amp Scheich 1996 1997 Weinberger Hopkinsamp Diamond 1984) (3) increased responses to frequen-cies surrounding a conditioned or stimulated tone pip(Bjordahl et al 1998 D M Diamond amp Weinberger1989 Metherate amp Weinberger 1989 Ohl amp Scheich1996 1997) (4) decreased responses to frequenciesother than that of a conditioned tone pip (Bakin Southamp Weinberger 1996 Bakin amp Weinberger 1990 Ede-line amp Weinberger 1993b Weinberger et al 1993)(5) changes in the size (eg bandwidth) of the receptivefield (Edeline amp Weinberger 1993b Kilgard amp Mer-zenich 1998a Weinberger et al 1993) and (6) generalincreasesdecreases in neural responsiveness (BakinLepan amp Weinberger 1992 Bakin amp Weinberger 1990

D M Diamond amp Weinberger 1989 Edeline amp Wein-berger 1993a) Cases in which neurons show no statisti-cally significant changes in responsiveness to a condi-tioned tone pip have also been observed (see eg Edelineamp Weinberger 1993b Weinberger et al 1984) Althoughthe wide range of reported changes in receptive fieldscan be attributed in part to differing methodologiesmanyvarieties of changes have been reported in individualstudies Experience-induced changes in neuronal responseproperties can also fluctuate over time (Bjordahl et al1998 Ohl amp Scheich 1996)

Not surprisingly changes in the spectral response prop-erties of nodes in simulated auditory cortices were muchless variable than those observed experimentally Themost common changes observed included increased sen-sitivity to a targeted frequency and decreased sensitivityto tone pips that were not presented to the map Fig-ure 4A shows how individual nodes in a simulated cortexrespond after being exposed to a 2-kHz pure tone for2500 trials using a high learning rate and a low neigh-borhood size (ie simulating enhanced neural adapt-ability without enhanced excitability) In this scenariotraining changed the response properties of 51 of thenodes For comparison Figure 5A shows how the map re-sponded after being exposed to a 2-kHz pure tone for

1 kHz 5 kHz

D

V

D

V1 kHz 5 kHzFrequency

(A)

(B)

How TrainedMap Respondsto 2 kHz

Adapted Nodes

Figure 4 (A) Response properties for a map trained with a 2-kHz pure tone for 2500 trials using alow neighborhood size and a high learning rate Note that the responses of nodes on the right side ofthe map have not been affected by training (B) A substantially increased number of nodes (10 timesmore than in the initial map) now respond strongly to 2 kHz


1250 trials when there was a lower learning rate and ahigh neighborhood size (ie simulating enhanced neuralexcitability with moderately enhanced adaptability) Eventhough this map was exposed to only half as many inputsas the map shown in Figure 4A most (82) of the nodesresponded strongly to 2 kHz Overall increases in excit-ability led to quicker and more extensive changes in re-sponse properties than did increases in learning rate (Fig-ure 6)

In all of our simulations we observed that the fre-quency response characteristics of individual nodes flat-tened (eg threshold and bandwidth increased) as theybecame ldquoretunedrdquo Consequently responses to charac-teristic frequencies decreased whereas responses to thefrequency paired with stimulation increased (Figure 7A)When neighborhood size was high individual nodesshowed a gradual decrease in flattening of the receptivefield as a function of distance from the target frequency(Figure 5A) In contrast a low neighborhood size led tomore discrete changes in the flatness of receptive fields(see Figure 4A) Overall such changes in spectral re-sponse properties are consistent with those observed inother neural network models of learning-induced audi-tory cortical plasticity (eg Armony et al 1995)

Figures 8 9 and 10 illustrate how maps respond afterbeing exposed to both a 2-kHz (625 trials) and a 4-kHz(625 trials) pure tone (exposures to each of the two tones

were randomly intermixed) When there was a lowneighborhood size nodes at the dorsal and ventral edgesof the map reorganized more slowly than inner nodes lo-cated an equivalent or greater distance from ldquowinningrdquonodes (ie nodes with characteristic frequencies match-ing the stimulated frequencies see Figures 8A and 10A)In contrast simulations in which there was a high neigh-borhood size showed no correlation between node posi-tion and speed of adaptation for an equal number of ex-posure trials (Figure 9A) All of the simulations with twotones consistently showed a tendency toward bimodallydistributed frequency response characteristics Whenneighborhood size was low some nodes became tunedto 2 kHz others became tuned to 4 kHz and a few nodesdeveloped multipeaked responses (see Figure 8A) Whenneighborhood size was high all the nodes became tunedto both 2 and 4 kHz (see Figure 9A) Simulations con-sistently showed reduced responsiveness to frequenciesbetween 2 and 4 kHz especially at 3 kHz the linear mid-point between these two frequencies

Map Reorganization in Rats and Neural NetworksRelatively few studies have examined global changes

in the spatial distribution of response properties across theauditory cortex induced by learning or basal forebrainstimulation This is somewhat surprising given the largeamount of research devoted to describing the spatial or-

1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Frequency

(A)

(B)

How Trained Map Responds to 2 kHz

Nodes Tuned to 2 kHz

Figure 5 (A) Response properties for a map trained with a 2-kHz pure tone for 1250 trials using a highneighborhood size and a medium learning rate (B) Over 80 of map nodes now respond best to 2 kHz


ganization of auditory cortical sensitivities (for reviewssee Aitkin 1990 Ehret 1997 Schreiner et al 2000)

Maps of cortical sensitivities have typically been mea-sured either anatomically or electrophysiologically Re-sponse sensitivities in the auditory cortex can be mappedanatomically using a variety of labeling techniques (egC-fos 2-DG) Labeling can reveal which neurons in thecortex have been activated in response to the presenta-tion and conditioning of particular acoustic stimuliChanges in cortical maps can be assessed by comparingpatterns of activity in naive subjects with those of trainedsubjects (for reviews see Edeline 1999 Scheich StarkZuschratter Ohl amp Simonis 1997) Changes in corticalmaps can also be measured by recording firing patternsfrom electrodes inserted throughout the auditory cortex(Kilgard amp Merzenich 1998a Recanzone et al 1993)or using optical imaging techniques (Bakin Kwan et al1996) Recently noninvasive techniques such as magneto-encephalography (MEG) have been used to measurechanges in auditory cortical sensitivities in humans(Cansino amp Williamson 1997 Molchan SunderlandMcIntosh Herscovitch amp Schreurs 1994 Morris Fris-

ton amp Dolan 1998 Pantev amp Lutkenhoner 2000 Pan-tev Wollbrink Roberts Engelien amp Lutkenhoner1999)

Studies of map reorganization based on anatomicaltechniques have shown that aversively conditioning ger-bils with a 1-kHz tone (eg by pairing the tone with ashock) leads to increased cortical sensitivity to frequen-cies below 1 kHz (Gonzalez-Lima amp Scheich 1986Scheich et al 1997) such shifts were not observed afteravoidance conditioning in which gerbils could escapethe aversive stimulus However both avoidance and aver-sion conditioning led to increased cortical responsive-ness relative to gerbils that were habituated to a 1-kHztone Electrophysiological studies with monkeys showedthat training on a frequency discrimination task led to in-creases in the cortical area responsive to behaviorally rel-evant frequencies (Recanzone et al 1993) Basal fore-brain stimulation paired with tone presentations in ratsled to massive reorganization of cortical topography sothat the majority of neurons responded to the stimulatedtone (Kilgard amp Merzenich 1998a) Changes in corticalresponsiveness have also been observed using MEG

Figure 6 Increases in excitability (neighborhood size) lead to larger increases in the number of nodes responding best to the tar-get frequency LR learning rate NS neighborhood size


during auditory discrimination tasks (Cansino amp Wil-liamson 1997) and after spectral deprivation (Pantevet al 1999) Such changes have not however been di-rectly related to changes in the organization of auditorycortical maps

Reorganization in self-organizing maps after trainingwith one tone was generally comparable with that seen inelectrophysiological experiments The map area respon-sive to the target tone increased with training the extentto which responsive regions expanded and the number ofexposures necessary to reach a particular level of expan-sion were primarily determined by the neighborhood sizeIncreasing the neighborhood size greatly reduced thenumber of exposures needed for maps to massively re-organize across a wide range of learning rates (Figure 6)Figure 4B shows how the map responds after being ex-posed to a 2-kHz pure tone for 2500 trials when there wasa high learning rate and a low neighborhood size Note that(1) the area of the map responding to 2 kHz has ex-panded considerably (2) regions of the map respondingto lower frequencies have compressed and (3) the dis-tribution of nodes sensitive to 2 kHz is roughly circularMuch less reorganization occurs when neither learningrate nor neighborhood size is high (Figure 10B) In com-

parison Figure 5B shows how the map responded afterbeing exposed to a 2-kHz pure tone for 1250 trials whenthere was a lower learning rate and a high neighborhoodsize In this map most of the nodes responded best to2 kHz and the nodes most sensitive to higher frequen-cies appear to have shifted outward

Simulations involving the presentation of two tonesproduced reorganization that was inconsistent with em-pirical observations For example in a map that was ex-posed to both a 2-kHz (625 trials) and a 4-kHz (625 trials)pure tone with a high learning rate and a low neighbor-hood size (Figure 8B) reorganization was focused at tar-get frequencies and a narrow region between 2 and 4 kHzwas not tuned to either tone Similarly a map exposed tothe same inputs when a high learning rate and a highneighborhood size were used (Figure 9B) showed a pre-dominance of nodes most sensitive to 2 and 4 kHz and alarge region between 2 and 4 kHz not tuned to eithertone In this map unlike the map shown in Figure 8B thenodes most sensitive to 2 and 4 kHz migrated to oppo-site corners In contrast when two tones are paired withbasal forebrain stimulation in rats cortical maps reorga-nize in such a way that most neurons respond best eitherto one of the stimulated frequencies or surprisingly to

1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)

How Trained Map Respondsto 2 kHz

Adapting Nodes Flatten

Figure 7 (A) Response properties for a map trained with a 2-kHz pure tone for 1250 trials using amedium neighborhood size and a medium learning rate Note how responses to characteristic fre-quencies decrease whereas responses to the target frequency increase (B) Comparison with Figures5A and 6A shows that intermediate parameter values give rise to intermediate levels of reorganization

Frequency


frequencies between the two stimulated frequencies(Kilgard amp Merzenich 1998a)

DISCUSSION

In this paper we directly tested Kohonenrsquos (1993) sug-gestion that the effects of neuromodulatory chemicals oncortical processing could be modeled as variations in theneighborhood functions of self-organizing maps Specif-ically we attempted to develop a self-organizing mapmodel of the auditory cortex that could be reorganized inways comparable with the cortical reorganization in-duced by pairing the presentation of sound with basal fore-brain activation

The results of our simulations demonstrate that a self-organizing map can be used to quantitatively characterizethe effects of enhanced synaptic adaptability and neural ex-citability on auditory cortical reorganization We found thatnode excitability (as controlled by neighborhood size) wasthe primary determinant of the number of exposures re-quired before massive reorganization was evident in mapsWhen excitability was high more nodes were adapted dur-ing each trial As nodes became sensitized to the target fre-quency the likelihood that a new best-responding nodewould emerge increased Even when adaptation was rela-

tively low numerous nodes were affected by trainingVarying either the adaptability or the excitability of nodesled to predictable changes in the response features of reor-ganized maps (summarized in Table 2)

Simulations of auditory cortical reorganization inducedby training or stimulation with a single tone produced re-sults generally comparable with those observed experi-mentally (eg by Recanzone et al 1993) and in priorcomputational studies Map nodes became tuned to thetarget frequency and the area of map representing thatfrequency expanded considerably Interestingly changesin response properties occurred first in the regions of themap where the most sharply tuned nodes were located(see eg Figure 10) We are not aware of any reports ofcorrelations between the initial bandwidth of a receptivefield and the probability (or rate) of retuning We were alsounable to find any reports describing differential potentialfor rapid plasticity as a function of spatial position withinthe auditory cortex Experiments examining whethersuch correlations do or do not exist could provide im-portant evidence for testing current models of auditorycortical plasticity

Simulations of changes in response sensitivities in-duced by pairing stimulation with two tones revealed alimitation in our model In empirical tests pairing basal

1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz 1 kHz 5 kHz

Frequency Frequency

(A)

(B) (C)

Nodes Tuned to 2 OR 4 kHz

Figure 8 (A) Response properties for a map trained with a 2-kHz tone (625 trials) and a 4-kHz tone(625 trials) using a low neighborhood size and a high learning rate (B) and (C) show that changes insensitivity are focused around the original nodes responding best to 2 kHz and 4 kHz respectively


forebrain stimulation with two tones led to unimodallydistributed neural response characteristics (Kilgard ampMerzenich 1998a) whereas in the present simulationsthis did not occur In addition pairing basal forebrainstimulation with two tones sharpens tuning in the audi-tory cortex (Kilgard amp Merzenich 1998a) we did notobserve this effect in our simulations with two tonesThese disparities suggest that stimulation-induced plas-ticity in the auditory cortex is constrained in ways notaccounted for by the competitive Hebbian learning algo-rithm used in self-organizing maps Whether such con-straints are specific to stimulation-induced plasticity isan important question for future research Weinberger(1998a) notes that the neural feasibility of Hebbian-likelearning processes is unclear and that the empirical sup-port for theories invoking Hebbian learning is not as strongas might be supposed (see also Cruikshank amp Weinberger1996a Edeline 1996 1999)

Comparisons between specific changes in the responsecharacteristics of map nodes and response characteris-tics reported in electrophysiological studies are difficultto make because most reports only describe changes inthe responses of individual neurons and the types ofchanges reported vary considerably within and across

studies Despite this variability there do not appear to beany reports suggesting that response characteristics inauditory cortical neurons ldquoflattenrdquo before retuning to anew frequency as we observed in our simulations Moststudies report increasesdecreases in the sensitivity ofneurons to the conditioned frequency only minor modif-ications in previous spectral sensitivities have been re-ported (see eg Edeline et al 1990 Edeline amp Wein-berger 1993b Metherate amp Weinberger 1990 Ohl ampScheich 1996 1997)

Discrepancies such as those noted above suggest thatalthough self-organizing maps are able to generate globalchanges in the topographic structure of response proper-ties that are qualitatively similar to those induced bypairing tones with basal forebrain stimulation the pre-dictive value of such maps may be limited to describingend states of cortical reorganization involving individualtone pips How do self-organizing maps successfully pre-dict such end states Presumably they do so by capturingimportant processing characteristics of the auditory cortexincluding (1) incremental localized adaptability (2) pro-cessing units that are spatially organized on the basis ofsimilarities between inputs and (3) competition betweenprocessing units The limitations of self-organizing maps

1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)

Nodes Tuned to 2 AND 4 kHz

Figure 9 (A) Response properties for a map trained with a 2-kHz tone (625 trials) and a 4-kHz tone(625 trials) using a high neighborhood size and a high learning rate Sensitivities are clearly bimodal(B) and (C) show that map nodes most sensitive to 2 kHz (B) and 4 kHz (C) have migrated to oppositecorners (ie the symmetry of response properties has shifted)


probably stem from the properties of cortical processingthat they do not emulate (eg temporal dynamics species-specific predispositions effects of context and the orderof stimulus presentation) Additional simulations with moreadvanced self-organizing map-based architectures canclarify how such properties influence map reorganization

If the response characteristics of reorganized auditorycortices were measured in sufficient detail one could bet-ter assess the applicability of our simulation results Forexample if basal forebrain stimulation only increases theadaptability of synaptic connections our model predictsthat auditory cortices reorganized by such stimulationshould exhibit discrete shifts in response characteristicsat the borders of expanded regions of sensitivity Alter-natively if basal forebrain stimulation mainly increasesthe excitability of cortical neurons our model predictsthat cortical response characteristics should show a grad-ual change in sensitivity to the frequency paired withstimulation as distance from the area of greatest sensi-tivity increases Recently developed chronic invasive tech-niques (deCharms Blake amp Merzenich 1999) and non-invasive mapping techniques (Bakin Kwon et al 1996

Pantev et al 1999) that allow repeated cortical measure-ments to be made should provide much needed data re-garding the dynamics of auditory cortical reorganization

The motivation for these simulations was to develop asimple computational framework that would be usefulfor modeling how neuromodulation affects auditory cor-tical plasticity Our model makes specific predictionsabout how response topography will change as a func-tion of enhanced neural adaptability and excitability inauditory cortical networks An important line for futureresearch will be to dissociate the effects of different neuro-modulators (and neuromodulatory systems) on corticalreorganization We hypothesize that different neuro-modulators differentially affect neural excitability andadaptability depending on the behavioral context and thatthese differential effects can lead to measurable differ-ences in cortical reorganization For example if cholin-ergic modulation increases the excitability of auditorycortical neurons to a greater extent than does dopamin-ergic modulation this should lead to predictable differ-ences in the cortical reorganization induced by stimulat-ing neurons in the basal forebrain versus the ventral

1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)

Unalter ed Outer Nodes

Figure 10 (A) Response properties for a map trained with a 2-kHz tone (625 trials) and a 4-kHz tone(625 trials) using a low neighborhood size and a medium learning rate Changes in response sensitiv-ities of narrowband nodes are minimal and nodes in dorsal and ventral regions of the map are unal-tered (B) and (C) show that map nodes most sensitive to 2 kHz (B) and 4 kHz (C) have changed littlefrom their initial states


tegmental area (the primary source of dopaminergic neu-rons projecting to auditory cortex)

Preliminary data suggest that pairing sound with ven-tral tegmental stimulation does lead to characteristicchanges in cortical maps that differ from those associ-ated with basal forebrain stimulation (Bao amp Merzenich2000) Pairing tone pips with basal forebrain stimulationin rats causes the regions of the auditory cortex re-sponding to those tone pips to greatly expand (Kilgard ampMerzenich 1998a) consistent with the reorganizationwe observed in self-organizing maps with highly excit-able nodes In contrast pairing tone pips with ventraltegmental stimulation leads to changes that are spatiallyand spectrally selective (Bao amp Merzenich 2000) con-sistent with changes we observed in maps with highlyadaptable nodes Model-based characterizations of dif-ferences in the distribution and properties of adaptedreceptive fields provide a unified framework for quanti-tatively comparing and contrasting the effects of neuro-modulators on cortical plasticity

One of the major reasons for our interest in under-standing how neuromodulation affects auditory corticalreorganization is that most neuromodulatory systemsthat have been implicated in cortical plasticity have alsobeen implicated in modulating learning and memoryprocesses For example the basal forebrain system is be-lieved to be critical in mediating mnemonic and atten-tional processes (for reviews see Bartus Dean Pon-tecorvo amp Flicker 1985 Dekker Connor amp Thal 1991Everitt amp Robbins 1997 Kesner 1988 Sarter amp Bruno1997) Acetylcholine release is enhanced in the initialstages of learning and the neuromodulatory actions ofacetylcholine are often necessary for associative condi-tioning to occur (Butt Testylier amp Dykes 1997 Jacobsamp Juliano 1995 Maalouf et al 1998 Miranda ampBermudez-Rattoni 1999 Wellman amp Pelleymounter1999) These findings suggest that the role of the nu-cleus basalis in cortical plasticity is to raise acetylcho-line levels when behaviorally relevant events are occur-

ring (for reviews see Ahissar amp Ahissar 1994 Ashe ampWeinberger 1991 Metherate amp Weinberger 1990Richardson amp Delong 1991)

In this paper we focused on simulating the effects ofbasal forebrain modulation on cortical map reorganiza-tion rather than on simulating receptive field plasticityinduced by conditioning Learning-induced plasticitymay involve other modulatory systems in addition to thebasal forebrain and different systems may be involved atdifferent stages of training andor in different trainingtasks (Ahissar Haidarliu amp Shulz 1996 Dimyan ampWeinberger 1999 Sachdev et al 1998 Stark amp Scheich1997 Weinberger 1998b) Changes in receptive fieldscan be induced in as few as 5ndash30 training trials (Bakin ampWeinberger 1990 Bjordahl et al 1998 Edeline ampWeinberger 1993a) Whether such changes reflect topo-graphical reorganization in the auditory cortex is an im-portant question for future research Our model in itspresent form is not adequate for simulating several im-portant characteristics of learning-induced cortical plas-ticity such as consolidation (Bjordahl et al 1998 Ede-line amp Weinberger 1993b) and habituation (Condon ampWeinberger 1991) The roles neuromodulators play incortical changes induced either by neural processes lead-ing to consolidation or by learning processes such as ha-bituation and extinction remain poorly understood Ex-tension of the model to account for neuromodulatoryeffects on such processes remains a future line of research

Simulations with a single tone can potentially be usedto model a variety of learning tasks involving cortical plas-ticity such as classical fear conditioning sensory pre-conditioning and motor reflex conditioning Simulationsof plasticity-inducing learning tasks that involve eithermultiple stimuli (eg two-tone discrimination or rever-sal learning Bakin South amp Weinberger 1996 Bakinamp Weinberger 1990 Edeline et al 1990 Edeline amp Wein-berger 1993b Ohl amp Scheich 1996) or modulation byother neural regions such as the hippocampus would beparticularly interesting The self-organizing map model

Table 2Summary of Parameter Effects on Map Reorganization

Excitability

Adaptability Low High

Simulations With One ToneLow Little reorganization Moderate reorganization

Graded changes in sensitivitiesHigh Moderate reorganization Massive reorganization

Discrete changes in sensitivities Graded changes in sensitivities

Simulations With Two TonesLow Little reorganization Moderate reorganization

Graded changes in sensitivitiesBimodally distributed responsesSymmetry of map shifts

High Moderate reorganization Massive reorganizationDiscrete changes in sensitivities Graded changes in sensitivitiesBimodally distributed responses Bimodally distributed responsesSymmetry of map unaffected Symmetry of map shifts


can help identify behavioral tasks in which chemical al-teration of neural excitability or adaptability in the auditorycortex might be expected to differentially affect learningThus the model can be useful not only for characterizingneuromodulatory effects on auditory cortical plasticitybut also for predicting how such effects influence learn-ing and memory processes

REFERENCES

Abbott L amp Sejnowski T (Eds) (1999) Neural codes and distributedrepresentations Foundations of neural computation CambridgeMA MIT Press

Ahissar E Abeles M Ahissar M Haidarliu S amp Vaadia E(1998) Hebbian-like functional plasticity in the auditory cortex ofthe behaving monkey Neuropharmacology 37 633-655

Ahissar E amp Ahissar M (1994) Plasticity in auditory cortical cir-cuitry Current Opinion in Neurobiology 4 580-587

Ahissar E Haidarliu S amp Shulz D E (1996) Possible involve-ment of neuromodulatory systems in cortical Hebbian-like plasticityJournal of Physiology 90 353-360

Ahissar E Vaaida E Ahissar M Bergman H Arieli A ampAbeles M (1992) Dependence of cortical plasticity on correlatedactivity of single neurons and of behavioral context Science 2571412-1415

Aitkin L (1990) The auditory cortex London Chapman and HallArbib M A (Ed) (1998) The handbook of brain theory and neural

networks Cambridge MA MIT PressArmony J L Servan-Schreiber D Cohen J D amp LeDoux J E

(1995) An anatomically constrained neural network model of fearconditioning Behavioral Neuroscience 109 246-257

Armony J L Servan-Schreiber D Cohen amp LeDoux J E(1997) Computational modeling of emotion Explorations throughthe anatomy and physiology of fear conditioning Trends in CognitiveSciences 1 28-34

Armony J L Servan-Schreiber D Romanski L M CohenJ D amp LeDoux J E (1997) Stimulus generalization of fear re-sponses Effects of auditory cortex lesions in a computational modeland in rats Cerebral Cortex 7 157-165

Ashe J H McKenna T M amp Weinberger N M (1989) Cholin-ergic modulation of frequency receptive fields in auditory cortexII Frequency-specific effects of anticholinesterases provide evidencefor a modulatory action of endogenous ACh Synapse 4 44-54

Ashe J H amp Weinberger N M (1991) Acetylcholine modulationof cellular excitability via muscarinic receptors Functional plastic-ity in auditory cortex In R T Richardson (Ed) Activation to acqui-sition Functional aspects of the basal forebrain cholinergic system(pp 189-246) Boston Birkhauser

Bakin J S Kwon M C Masino S A Weinberger N M ampFrostig R D (1996) Suprathreshold auditory cortex activationvisualized by intrinsic signal optical imaging Cerebral Cortex 6120-130

Bakin J S Lepan B amp Weinberger N M (1992) Sensitization in-duced receptive field plasticity in the auditory cortex is independentof CS-modality Brain Research 577 226-235

Bakin J S South D A amp Weinberger N M (1996) Induction ofreceptive field plasticity in the auditory cortex of the guinea pig dur-ing instrumental avoidance training Behavioral Neuroscience 110905-913

Bakin J S amp Weinberger N M (1990) Classical conditioning in-duces CS-specific receptive field plasticity in the auditory cortex ofthe guinea pig Brain Research 536 271-286

Bakin J S amp Weinberger N M (1996) Induction of a physiologicalmemory in the cerebral cortex by stimulation of the nucleus basalisProceedings of the National Academy of Sciences 93 11219-11224

Bao S amp Merzenich M M (2000) Auditory cortical reorganiza-tion induced by stimulation of ventral tegmental area Society forNeuroscience Abstracts 26 1477

Barkai E Bergman R E Horowitz G amp Hasselmo M E

(1994) Modulation of associative memory function in a biophysicalsimulation of rat piriform cortex Journal of Neurophysiology 72659-677

Bartus R T Dean R L Pontecorvo M J amp Flicker C (1985)The cholinergic hypothesis A historical overview current perspec-tive and future directions In D S Olton E Gamzu amp S Corkin(Eds) Memory dysfunctions An integration of animal and humanresearch from preclinical and clinical perspectives (Annals of theNew York Academy of Sciences Vol 444 pp 332-358) New YorkNew York Academy of Sciences

Baskerville K A Schweitzler J B amp Herron P (1997) Effectsof cholinergic depletion on experience-dependent plasticity in thecortex of the rat Neuroscience 80 1159-1169

Bauer H-U Der R amp Herrmann M (1996) Controlling the mag-nification factor of self-organizing feature maps Neural Computa-tion 8 757-771

Benukskova L Diamond M E amp Ebner F F (1994) Dynamicsynaptic modification threshold Computational model of experience-dependent plasticity in adult rat barrel cortex Proceedings of theNational Academy of Sciences 91 4791-4795

Bjordahl T S Dimyan M A amp Weinberger N M (1998) In-duction of long-term receptive field plasticity in the auditory cortexof the waking guinea pig by stimulation of the nucleus basalis Be-havioral Neuroscience 112 467-479

Bower J M amp Beeman D (1998) The book of GENESIS Explor-ing realistic neural models with the GEneral NEural SImulation Sys-tem Berlin Springer-Verlag

Buonomano D V amp Merzenich M M (1998) Cortical plasticityFrom synapses to maps Annual Review of Neuroscience 21 149-186

Butt A E Testylier G amp Dykes R W (1997) Acetylcholine re-lease in rat frontal and somatosensory cortex is enhanced during tac-tile discrimination learning Psychobiology 25 18-33

Cansino S amp Williamson S J (1997) Neuromagnetic fields revealcortical plasticity when learning an auditory discrimination task BrainResearch 764 53-66

Condon C D amp Weinberger N M (1991) Habituation producesfrequency-specific plasticity of receptive fields in the auditory cortexBehavioral Neuroscience 105 416-430

Cruikshank S J amp Weinberger N M (1996a) Evidence for theHebbian hypothesis in experience-dependent physiological plasticityof neocortex A critical review Brain Research Reviews 22 191-228

Cruikshank S J amp Weinberger N M (1996b) Receptive-fieldplasticity in the adult auditory cortex induced by Hebbian covarianceJournal of Neuroscience 16 861-875

deCharms R C Blake D T amp Merzenich M M (1999) A multi-electrode implant device for the cerebral cortex Journal of Neuro-science Methods 93 27-35

deCharms R C amp Merzenich M M (1996) Primary cortical rep-resentation of sounds by the coordination of action-potential timingNature 381 610-613

deCharms R C amp Zador A (2000) Neural representation and thecortical code Annual Review of Neuroscience 23 613-647

Dekker A J Connor D J amp Thal L J (1991) The role of cholin-ergic projections from the nucleus basalis in memory Neuroscienceamp Biobehavioral Reviews 15 299-317

Delacour J O Houcine O amp Costa J C (1990) Evidence for acholinergic mechanism of ldquolearnedrdquo changes in the responses of barrelfield neurons of the awake and undrugged rat Neuroscience 34 1-8

de Pinho M Mazza M amp Roque A C (2000) A biologically-plau-sible computational model of classical conditioning induced reorga-nization of tonotopic maps in the auditory cortex Neurocomputing 32685-691

de Pinho M amp Roque-da-Silva A C (1999) A realistic computa-tional model of formation and variability of tonotopic maps in theauditory cortex Neurocomputing 26-7 355-359

Diamond D M amp Weinberger N M (1984) Physiological plastic-ity of single neurons in auditory cortex of the cat during acquisitionof pupillary responses II Secondary field (AII) Behavioral Neuro-science 98 189-210

Diamond D M amp Weinberger N M (1986) Classical conditioningrapidly induces specific changes in frequency receptive fields of sin-


gle neurons in secondary and ventral ectosylvian auditory corticalfields Brain Research 372 357-360

Diamond D M amp Weinberger N M (1989) Role of context in theexpression of learning-induced plasticity of single neurons in audi-tory cortex Behavioral Neuroscience 103 471-494

Diamond M E amp Ebner F F (1990) Emergence of radial and mod-ular units in neocortexIn B L Finlay G Innocenti amp H Scheich(Eds) The neocortex Ontogeny and phylogeny (pp 159-171) NewYork Plenum

Diamond M E Petersen R S amp Harris J A (1999) Learningthrough maps Functional significance of topographic organization inprimary sensory cortex Journal of Neurobiology 41 64-68

Dimyan M A amp Weinberger N M (1999) Basal forebrain stimu-lation induces discriminative receptive field plasticity in the auditorycortex Behavioral Neuroscience 113 691-702

Dykes R W (1997) Mechanisms controlling neuronal plasticity insomatosensory cortex Canadian Journal of Physiological Pharma-cology 75 535-545

Edeline J-M (1995) The a2-adrenergic antagonist Idazoxan en-hances the frequency selectivity and increases the threshold of auditorycortex neurons Experimental Brain Research 107 221-240

Edeline J-M (1996) Does Hebbian synaptic plasticity explainlearning-induced sensory plasticity in adult mammals Journal ofPhysiology 90 271-276

Edeline J-M (1999) Learning-induced physiological plasticity inthalamo-cortical sensory systems A critical evaluation of receptivefield plasticity map changes and their potential mechanisms Prog-ress in Neurobiology 57 165-224

Edeline J-M Hars B Maho C amp Hennevin E (1994) Tran-sient and prolonged facilitation of tone-evoked responses induced bybasal forebrain stimulation in the rat auditory cortex ExperimentalBrain Research 97 373-386

Edeline J-M Neuenschwander-El Massioui N amp Dutrieux G(1990) Frequency-specific cellular changes in the auditory systemduring acquisition and reversal of discriminative conditioning Psy-chobiology 18 382-393

Edeline J-M amp Weinberger N M (1993a) Rapid development oflearning-induced receptive field plasticity in the auditory cortex Be-havioral Neuroscience 107 539-557

Edeline J-M amp Weinberger N M (1993b) Receptive field plastic-ity in the auditory cortex during frequency discrimination trainingSelective retuning independent of task difficulty Behavioral Neuro-science 107 82-103

Ehret G (1997) The auditory cortex Journal of Comparative Physi-ology A 181 547-557

Elliot T amp Shadbolt N R (1998) A model of activity-dependentanatomical inhibitory plasticity applied to the mammalian auditorysystem Biological Cybernetics 78 455-464

Erwin E Obermayer K amp Schulten K (1995) Models of orien-tation and ocular dominance columns in the visual cortex A criticalcomparison Neural Computation 7 425-468

Everitt B J amp Robbins T W (1997) Central cholinergic systemsand cognition Annual Review of Psychology 48 649-684

Fellous J-M amp Linster C (1998) Computational models of neu-romodulation Neural Computation 10 771-805

Foeller E Vater M amp Kossl M (2000) Influence of GABAergicinhibition on temporal response properties and frequency tuning ofauditory cortex neurons in the gerbil Abstracts of the Association forResearch in Otolaryngology 23 12

Fregnac Y amp Shulz D E (1999) Activity-dependent regulation ofreceptive field properties of cat area 17 by supervised Hebbian learn-ing Journal of Neurobiology 41 69-82

Gao E amp Suga N (2000) Experience-dependent plasticity in the au-ditory cortex and the inferior colliculus of bats Role of the cortico-fugal system Proceedings of the National Academy of Sciences 978081-8086

Giorgetti M Bacciottini L Giovannini M G ColivicchiM A Goldfarb J O amp Blandina P (2000) Local GABAergicmodulation of acetylcholine release from the cortex of freely movingrats European Journal of Neuroscience 12 1941-1948

Gluck M A amp Myers C E (2001) Gateway to memory An intro-

duction to neural network modeling of the hippocampus and learn-ing Cambridge MA MIT Press

Gonzalez-Lima F amp Scheich H (1986) Neural substrates for tone-conditioned bradycardia demonstrated with 2 deoxyglucose II Au-ditory cortex plasticity Behavioural Brain Research 20 281-293

Grajski K A amp Merzenich M M (1990) Hebb-type dynamics issufficient to account for the inverse magnification rule in cortical so-matotopy Neural Computation 2 71-84

Gritti I Mainville L Mancia M amp Jones B F (1997)GABAergic and other noncholinergic basal forebrain neurons togetherwith cholinergic neurons project to the mesocortex and isocortex inthe rat Journal of Comparative Neurology 383 163-177

Guenther F H amp Gjaja M N (1996) The perceptual magnet effectas an emergent property of neural map formation Journal of theAcoustical Society of America 100 1111-1121

Hars B Maho C Edeline J-M amp Hennevin E (1993) Basalforebrain stimulation facilitates tone-evoked responses in the audi-tory cortex of the awake rat Neuroscience 56 61-74

Hasselmo M E (1995) Neuromodulation and cortical function Model-ing the physiological basis of behavior Behavioural Brain Research67 1-27

Jacobs S E Code R A amp Juliano S L (1991) Basal forebrain le-sions alter stimulus-evoked metabolic activity in rat somatosensorycortex Brain Research 560 342-345

Jacobs S E amp Juliano S L (1995) The impact of basal forebrainlesions on the ability of rats to perform a sensory discrimination taskinvolving barrel cortex Journal of Neuroscience 14 697-711

Jiminez-Capdeville M E Dykes R W amp Myasnikov A A(1997) Differential control of cortical activity by the basal forebrainin rats A role for both cholinergic and inhibitory influences Journalof Comparative Neurology 381 53-67

Joublin F Spengler F Wacquant S amp Dinse H R (1996) Acolumnar model of somatosensory reorganizational plasticity basedon Hebbian and non-Hebbian learning rules Biological Cybernetics74 275-286

Juliano S L Ma W amp Eslin D (1991) Cholinergic depletion pre-vents expansion of topographic maps in somatosensory cortex Pro-ceedings of the National Academy of Sciences 88 780-784

Kaas J H (1996) Plasticity of sensory representations in the auditoryand other systems of adult mammalsIn R J Salvi amp D Henderson(Eds) Auditory system plasticity and regeneration (pp 213-223)New York Thieme Medical Publishers

Kaas J H (1997) Introduction Functional plasticity in adult cortexSeminars in Neuroscience 9 1-1

Kesner R P (1988) Reevaluation of the contribution of the basal fore-brain cholinergic system to memory Neurobiology of Aging 9 609-616

Kilgard M P amp Merzenich M M (1998a) Cortical map reorgani-zation enabled by nucleus basalis activity Science 279 1714-1718

Kilgard M P amp Merzenich M M (1998b) Plasticity of temporalinformation processing in primary auditory cortex Nature Neuro-science 1 727-731

Kohonen T (1987) Adaptive associative and self-organizing functionsin neural computing Applied Optics 26 4910-4918

Kohonen T (1993) Physiological interpretation of the self-organizingmapping algorithm Neural Networks 6 895-905

Kohonen T (1997) Self-organizing maps Berlin Springer-VerlagKraus N amp Disterhoft J F (1982) Response plasticity of single

neurons in the rabbit auditory association cortex during tone-signaledlearning Brain Research 246 205-215

Maalouf M Miasnikov A A amp Dykes R W (1998) Blockade ofcholinergic receptors in rat barrel cortex prevents long-term changesin the evoked potentials during sensory preconditioning Journal ofNeurophysiology 80 529-545

Maldonado P E amp Gerstein G L (1996) Reorganization in theauditory cortex of the rat induced by intracortical microstimulation Amultiple single-unit study Experimental Brain Research 112 420-430

Manunta Y amp Edeline J-M (1997) Effects of noradrenaline onfrequency tuning of rat auditory cortex neurons European Journal ofNeuroscience 9 833-847

McKenna T M Ashe J H amp Weinberger N M (1989) Cholin-


ergic modulation of frequency receptive fields in auditory cortexI Frequency-specific effects of muscarinic agonists Synapse 4 30-43

Mercado E III Myers C E amp Gluck M A (1999a) A computa-tional model of basal forebrain modulation of auditory cortical plastic-ity Cognitive Neuroscience Society Annual Meeting Program 102 81B

Mercado E III Myers C E amp Gluck M A (1999b) Plasticity ofspectral processing in simulated primary auditory cortex Society forNeuroscience Abstracts 29 391

Mercado E III Myers C E amp Gluck M A (2000) Modeling au-ditory cortical processing as an adaptive chirplet transform Neuro-computing 32-33 913-919

Mercado E III Shohamy D Orduntildea I Gluck M A amp Mer-zenich M M (2000) Plasticity of spectrotemporal sensitivities inauditory cortex Journal of the Acoustical Society of America A 1072835

Merzenich M M amp Schreiner C E (1992) Mammalian auditorycortex Some comparative observations In D B Webster R F Fayamp A N Popper (Eds) The evolutionary biology of hearing (pp 673-689) New York Springer-Verlag

Metherate R amp Ashe J H (1991) Basal forebrain stimulation mod-ifies auditory cortex responsiveness by an action at muscarinic re-ceptors Brain Research 559 163-167

Metherate R amp Ashe J H (1993) Nucleus basalis stimulation fa-cilitates thalamocortical synaptic transmission in the rat auditory cor-tex Synapse 14 132-143

Metherate R Ashe J H amp Weinberger N M (1990) Acetyl-choline modifies neuronal acoustic rate-level functions in guinea pigauditory cortex by an action at muscarinic receptors Synapse 6 364-368

Metherate R Tremblay N amp Dykes R W (1987) Acetylcholinepermits long-term enhancement of neuronal responsiveness in cat pri-mary somatosensory cortex Neuroscience 22 75-81

Metherate R amp Weinberger N M (1989) Acetylcholine producesstimulus-specific receptive field alterations in cat auditory cortexBrain Research 480 372-377

Metherate R amp Weinberger N M (1990) Cholinergic modula-tion of responses to single tones produces tone-specific receptive fieldalterations in cat auditory cortex Synapse 14 133-145

Meyer-Base U amp Scheich H (1995) Artificial implementation ofauditory neurons A comparison of biologically motivated models anda new transfer function oriented model Biological Cybernetics 77123-130

Miranda M I amp Bermudez-Rattoni F (1999) Reversible inacti-vation of the nucleus basalis magnocellularis induces disruption ofcortical acetylcholine release and acquisition but not retrieval of aver-sive memories Proceedings of the National Academy of Sciences 966478-6482

Molchan S E Sunderland T McIntosh A R Herscovitch Pamp Schreurs B G (1994) A functional anatomical study of asso-ciative learning in humans Proceedings of the National Academy ofSciences 91 8122-8126

Morris J S Friston K J amp Dolan R J (1998) Experience-dependent modulation of tonotopic neural responses in human auditorycortex Proceedings of the Royal Society of London Series B 265649-657

Myers C E Ermita B R Harris K Hasselmo M Solomon Pamp Gluck M A (1996) A computational model of cholinergic dis-ruption of septohippocampal activity in classical eyeblink condition-ing Neurobiology of Learning amp Memory 66 51-66

Myers C E Ermita B R Hasselmo M amp Gluck M A (1998)Further implications of a computational model of septohippocampalcholinergic modulation in eyeblink conditioning Psychobiology 261-20

Myers C E Gluck M A amp Granger R (1995) Dissociation ofhippocampal and entorhinal function in associative learning A com-putational approach Psychobiology 23 116-138

Ohl F W amp Scheich H (1996) Differential frequency conditioningenhances spectral contrast sensitivity of units in auditory cortex(field AI) of the alert Mongolian gerbil European Journal of Neuro-science 8 1001-1017

Ohl F W amp Scheich H (1997) Learning-induced dynamic recep-tive field changes in primary auditory cortex of the unanaesthetizedMongolian gerbil Journal of Comparative Physiology A 181 685-696

Palakal M J Murthy U Chittajallu S K amp Wong D (1995)Tonotopic representation of auditory responses using self-organizingmaps Mathematical amp Computational Modelling 22 7-21

Palakal M J amp Wong D (1999) Cortical representation of spatio-temporal pattern of firing evoked by echolocation signals Popula-tion encoding of target features in real time Journal of the AcousticalSociety of America 106 479-490

Pantev C amp Lutkenhoner B (2000) Magnetoencephalographicstudies of functional organization and plasticity of the human audi-tory cortex Journal of Clinical Neurophysiology 17 130-142

Pantev C Wollbrink A Roberts L E Engelien A amp Lut-kenhoner B (1999) Short-term plasticity of the human auditorycortex Brain Research 842 192-199

Piepenbrock C amp Obermayer K (2000) The effect of intracorticalcompetition on the formation of topographic maps in models of Heb-bian learning Biological Cybernetics 82 345-353

Pitton J W Wang K amp Juang B-H (1996) Time-frequencyanalysis and auditory modeling for automatic recognition of speechProceedings of the IEEE 84 1199-1214

Rauschecker J P (1999) Auditory cortical plasticity A comparisonwith other sensory systems Trends in Neurosciences 22 74-80

Recanzone G H Schreiner C E amp Merzenich M M (1993)Plasticity in the frequency representation of primary auditory cortexfollowing discrimination training in adult owl monkeys Journal ofNeurophysiology 13 87-103

Recanzone G H Schreiner C E Sutter M L Beitel R E ampMerzenich M M (1999) Functional organization of spectral re-ceptive fields in the primary auditory cortex of the owl monkey Jour-nal of Comparative Physiology A 185 493-508

Richardson R T amp Delong M R (1991) Functional implicationof tonic and phasic changes in nucleus basalis neurons In R T Rich-ardson (Ed) Activation to acquisition Functional aspects of the basalforebrain cholinergic system (pp 135-166) Boston Birkhauser

Ritter H Martinez T amp Schulten K (1992) Neural computa-tion and self-organizing maps An introduction Reading MAAddison-Wesley

Robert A amp Eriksson J L (1999) A composite model of the audi-tory periphery for simulating responses to complex sounds Journalof the Acoustical Society of America 106 1852-1864

Sachdev R N S Shao-Ming L Wiley R G amp Edner F F(1998) Role of the basal forebrain cholinergic projection in somato-sensory cortical plasticity Journal of Neurophysiology 79 3216-3228

Sanchez-Montanes M A Verschure P F amp Konig P (2000)Local and global gating of synaptic plasticity Neural Computation12 519-529

Sarter M amp Bruno J P (1997) Cognitive functions of cortical acetyl-choline Toward a unifying hypothesis Brain Research Reviews 2328-46

Scheich H Stark H Zuschratter W Ohl F W amp SimonisC E (1997) Some functions of primary auditory cortex in learningand memory formation In H-J Freund B A Sabel amp O W Witte(Eds) Brain plasticity Advances in neurology (Vol 73 pp 179-193) Philadelphia Lippincott-Raven

Schreiner C E (1998) Spatial distribution of responses to simpleand complex sounds in primary auditory cortex Audiology amp Neuro-otology 3 104-122

Schreiner C E Read H L amp Sutter M L (2000) Modular or-ganization of frequency integration in primary auditory cortex An-nual Review of Neuroscience 23 501-529

Shulz D E Sosnik R Ego V Haidarliu S amp Ahissar E(2000) A neuronal analogue of state-dependent learning Nature 403549-553

Schulze H amp Langner G (1999) Auditory cortical responses toamplitude modulations with spectra above frequency receptive fieldsEvidence for wide spread spectral integration Journal of Compara-tive Physiology A 185 493-508

Sirosh J amp Miikkulainen R (1997) Topographic receptive fields


and patterned lateral interaction in a self-organizing model of the pri-mary visual cortex Neural Computation 9 577-594

Stark H amp Scheich H (1997) Dopaminergic and serotonergic sys-tems are differentially involved in auditory cortex learning A long-term microdialysis study of metabolites Journal of Neurochemistry68 691-697

Suga N (1990) Cortical computational maps for auditory imagingNeural Networks 3 3-21

Suga N (1995) Processing of auditory information carried by species-specif ic complex sounds In M S Gazzaniga (Ed) The cognitiveneurosciences (pp 295-313) Cambridge MA MIT Press

Sutton G G III Reggia J A Armentrout S L amp DrsquoAutrechyC L (1994) Cortical map reorganization as a competitive processNeural Computation 6 1-13

Swindale N V (1996) The development of topography in the visualcortex A review of models Network Computation in Neural Sys-tems 7 161-247

Swindale N V amp Bauer H-U (1998) Application of Kohonenrsquosself-organizing feature map alg54orithm to cortical maps of orienta-tion and direction preference Proceedings of the Royal Society ofLondon Series B 265 827-838

Tanaka S (1990) Theory of self-organization of cortical maps Math-ematical framework Neural Networks 3 625-640

Tchorz J amp Kollmeier B (1999) A model of auditory perception asa front end for automatic speech recognition Journal of the Acousti-cal Society of America 106 2040-2050

Tremblay N Warren R A amp Dykes R W (1990) Electrophysio-logical studies of acetylcholine and the role of the basal forebrain inthe somatosensory cortex of the cat I Cortical neurons excited by glut-amate Journal of Neurophysiology 64 1199-1211

Wang J Caspary D amp Salvi R J (2000) GABA-A antagonistcauses dramatic expansion of tuning in primary auditory cortexNeuroReport 11 1137-1140

Wang K amp Shamma S A (1995a) Auditory analysis of spectro-temporal information in acoustic signals IEEE Engineering in Med-icine amp Biology 14 186-194

Wang K amp Shamma S A (1995b) Spectral shape analysis in thecentral auditory system IEEE Transactions on Speech amp Audio Pro-cessing 3 382-395

Wang X Merzenich M M Beitel R amp Schreiner C H (1995)Representation of a species-specific vocalization in the primary audi-tory cortex of the common marmoset Temporal and spectral charac-teristics Journal of Neurophysiology 74 2685-2706

Webster H H Hanisch U K Dykes R W amp Biesold D (1991)Basal forebrain lesions with or without reserpine injection inhibitcortical reorganization in a rat hindpaw primary somatosensory cor-tex following sciatic nerve section Somatosensory Motor Research8 327-346

Webster H H Rasmusson D D Dykes R W Schliebs RSchober W Bruckner G amp Biesold D (1991) Long-term en-hancement of evoked potentials in raccoon somatosensory cortex fol-lowing co-activation of the nucleus basalis of Meynert complex andcutaneous receptors Brain Research 545 292-296

Weinberger N M (1993) Learning-induced changes of auditory re-ceptive fields Current Opinion in Neurobiology 3 570-577

Weinberger N M (1995) Dynamic regulation of receptive fields andmaps in the adult sensory cortex Annual Review of Neuroscience 18129-158

Weinberger N M (1997) Learning-induced receptive field plasticityin the primary auditory cortex Seminars in Neuroscience 9 59-67

Weinberger N M (1998a) Physiological memory in primary auditorycortex Characteristics and mechanisms Neurobiology of Learningamp Memory 70 226-251

Weinberger N M (1998b) Tuning the brain by learning and by stimu-lation of the nucleus basalis Trends in Cognitive Sciences 2 271-273

Weinberger N M Ashe J H Metherate R McKenna T MDiamond D M amp Bakin J (1990a) Retuning auditory cortex bylearning A preliminary model of receptive field plasticity Conceptsin Neuroscience 1 91-132

Weinberger N M Ashe J H Metherate R McKenna T MDiamond D M Bakin J S Lennartz R C amp Cassady J M(1990b) Neural adaptive information processing A preliminarymodel of receptive field plasticity in auditory cortex during Pavlovianconditioning In M Gabriel amp J Moore (Eds) Neurocomputationand learning Foundations of adaptive networks (pp 91-138) Cam-bridge MA MIT Press

Weinberger N M amp Bakin J S (1998) Learning-induced physio-logical memory in adult primary auditory cortex Receptive field plas-ticity model and mechanisms Audiology amp Neuro-Otology 3 145-167

Weinberger N M amp Diamond D (1987) Physiological plasticity inauditory cortex Rapid induction by learning Progress in Neurobiol-ogy 29 1-55

Weinberger N M Hopkins W amp Diamond D M (1984) Physio-logical plasticity of single neurons in auditory cortex of the cat dur-ing acquisition of pupillary responses I Primary field (AI) Behav-ioral Neuroscience 98 171-188

Weinberger N M Javid R amp Lepan B (1993) Long-term retentionof learning-induced receptive-field plasticity in the auditory cortexProceedings of the National Academy of Sciences 90 2394-2398

Wellman C L amp Pelleymounter M A (1999) Differential ef-fects of nucleus basalis lesions in young adult and aging rats Neuro-biology of Aging 20 381-393

Zaborszky L Pang K Somogyi J Nadasdy Z amp Kallo I(1999) The basal forebrain corticopetal system revisited In J F Mc-Ginty (Ed) Advancing from the ventral striatum to the extended amyg-dala Implications for neuropsychiatry and drug abuse (Annals of theNew York Academy of Sciences Vol 877 pp 339-367) New YorkNew York Academy of Sciences

Zoli M Torri C Jansson A Zini I Fuze K amp Agnati L F(1998) The emergence of the volume transmission concept BrainResearch Reviews 26 136-147

(Manuscript received May 10 2000revision accepted for publication September 11 2000)


order to rigorously test the validity of current theoreticalproposals concerning the role that neuromodulators playin cortical plasticity Previous simulations of experience-dependent changes in auditory cortical processing (egArmony et al 1995) have not explicitly considered howneuromodulatory systems and preexisting topographicalcortical sensitivities might affect the reorganization ofauditory representations Such variables can critically af-fect cortical reorganization (M E Diamond Petersen ampHarris 1999 Dykes 1997)

In this paper we describe a connectionist model of au-ditory cortical processing that can be used to quantitativelydescribe the effects of neuromodulation on experience-dependent cortical plasticity We constrained the organi-zation and sensitivities of a neural network so that theyparalleled properties seen in the mammalian auditorycortex and modeled the effects of modulatory processeson cortical reorganization using parameters intrinsic tothe learning algorithm used by the neural network Weinvestigated how variations in parameter values affectedexperience-induced changes in the responses of the neuralnetwork and compared these changes with recent electro-

physiological descriptions of reorganization in the audi-tory cortex induced by pairing electrical stimulation ofthe basal forebrain with the presentation of tones

Our goal was to develop a simple computational modelof auditory cortical processing that could provide a pre-cise theoretical framework for (1) investigating the rolesof neuromodulation in experience-dependent reorgani-zation of auditory representations and (2) testing the ac-curacy and sufficiency of more qualitative models of experience-induced plasticity in the auditory cortex Pre-liminary reports of some of these data appeared in abstractform (Mercado Myers amp Gluck 1999a 1999b)

A COMPUTATIONAL MODEL OFAUDITORY CORTICAL PROCESSING

Numerous cortical models of varying computationalcomplexity and physiological realism have been devel-oped (see eg Abbott amp Sejnowski 1999 Arbib 1998Bower amp Beeman 1998) Two general classes of auditorycortex models are prevalent signal-processing models andneural network models (see Figure 1) Signal-processing

Figure 1 Models of audition CN cochlear nucleus IC inferior colliculus MGB medialgeniculate body The biological model involves a multimodule loop composed of converging anddiverging neural connections Signal-processing models typically involve sequential transfor-mations of a one-dimensional waveform x(t ) measuring variations in pressure over time Firstx(t ) is transformed by a set of bandpass filters simulating cochlear processing Then these sig-nals are normalized simulating processing by hair cells and auditory ganglia to generate an au-ditory spectrum Finally the auditory spectrum is transformed into an n-dimensional spacey(t fn) that describes features of the sound such as pitch and signal bandwidth Neural networkmodels are typically much more abstract Inputs are often represented as binary vectors that codemeasured properties of sounds and outputs often reflect activity levels of nodes Transitions be-tween stages (or layers) involve matrix transformations Signal-processing models of audition usu-ally involve static feedforward processing whereas neural network models are usually adaptive

































1 kHz 5 kHz

D

V

D

V



(A)

(B) (C)

WIDEBAND

WIDEBAND

NARROWBAND
























1 kHz 5 kHz

D

V

D


(A)

(B)


Adapted Nodes









1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Frequency

(A)

(B)















1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES






































































































































































































1 kHz 5 kHz

D

V

D

V



(A)

(B) (C)

WIDEBAND

WIDEBAND

NARROWBAND
























1 kHz 5 kHz

D

V

D


(A)

(B)


Adapted Nodes









1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Frequency

(A)

(B)















1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES
































































































































































































1 kHz 5 kHz

D

V

D

V



(A)

(B) (C)

WIDEBAND

WIDEBAND

NARROWBAND
























1 kHz 5 kHz

D

V

D


(A)

(B)


Adapted Nodes









1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Frequency

(A)

(B)















1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES





















































































































































































1 kHz 5 kHz

D

V

D

V



(A)

(B) (C)

WIDEBAND

WIDEBAND

NARROWBAND
























1 kHz 5 kHz

D

V

D


(A)

(B)


Adapted Nodes









1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Frequency

(A)

(B)















1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES











































































































































































1 kHz 5 kHz

D

V

D

V



(A)

(B) (C)

WIDEBAND

WIDEBAND

NARROWBAND
























1 kHz 5 kHz

D

V

D


(A)

(B)


Adapted Nodes









1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Frequency

(A)

(B)















1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES

























































































































































































1 kHz 5 kHz

D

V

D


(A)

(B)


Adapted Nodes









1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Frequency

(A)

(B)















1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES










































































































































































1 kHz 5 kHz

D

V

D


(A)

(B)


Adapted Nodes









1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Frequency

(A)

(B)















1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES













































































































































































1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Frequency

(A)

(B)















1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES

















































































































































































1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES











































































































































































1 kHz 5 kHz

D

V

D

V1 kHz 5 kHz

Fr equency

(A)

(B)




Frequency



DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES








































































































































































DISCUSSION






1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES











































































































































































1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B)

(C)








1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES











































































































































































1 kHz 5 kHz

D

V

D


Frequency Frequency

(A)

(B) (C)











Excitability










REFERENCES














































































































































































Excitability










REFERENCES








































































































































































REFERENCES













































































































































































































































































































































































































Documents

A computational model of mechanisms controlling experience-dependent reorganization of representational maps in auditory cortex