Plasticity in auditory cortex on the grounds of learning discrimination

TABLE OF CONTENTS

1 Introduction
1.1 Overview
1.2 Delimitation of the topic

2 Theoretical framework
2.1 Cortical Plasticity
2.1.1 Basic principles of cortical plasticity
2.1.1.1 Synaptic plasticity - the weighting of synaptic strength
2.1.1.2 Long-term potentiation
2.1.1.3 Long-term depression
2.1.1.4 Axonal sprouting
2.1.1.5 Neurogenesis in enriched environments
2.1.2 From synapses to representational maps
2.1.3 Reorganization of cortical maps by deprivation
2.1.4 Reorganization after lesions in sensory areas
2.1.5 Experience-dependent plasticity
2.1.6 Reorganization of cortical maps after training
2.1.7 Reorganization of cortical maps in experienced learners
2.2 From cell assemblies to spatiotemporal patterns
2.3 Perceptual discrimination learning
2.3.1 Discrimination learning
2.3.2 Perceptual learning
2.4 Auditory sensory memory
2.5 Conclusions

3 Methodology
3.1 Basic Principles of MEG
3.2 Event Related Potentials and Fields
3.2.1 Dipole source model
3.2.2 The sensitivity of AERs for change
3.2.3 The Mismatch Negativity
3.2.3.1 The MMN change detection mechanism
3.2.3.2 Sources of MMN
3.2.3.3 Differential Impact of Native Language on the MMN
3.3 Conclusions

4 Experimental work
4.1 Experiment 1: Plastic changes as a result of frequency discrimination learning
4.1.1 Theoretical framework
4.1.2 Materials and Methods
4.1.2.1 Subjects
4.1.2.2 Discrimination Training
4.1.2.3 Discrimination Test
4.1.2.4 MEG Measurements
4.1.3 Results
4.1.3.1 Psychophysics
4.1.3.2 The Mismatch Field
4.1.3.3 The N1m responses
4.1.4 Discussion
4.1.5 Conclusions
4.2 Experiment 2: Plasticity due to discrimination learning of non- native mora-timing
4.2.1 Theoretical framework
4.2.2 Materials and methods
4.2.2.1 Subjects
4.2.2.2 Stimuli
4.2.2.3 Categorization test
4.2.2.4 Discrimination test and training
4.2.2.5 MEG recordings
4.2.3 Results
4.2.3.1 Behavioral results
4.2.3.2 Evoked Neuromagnetic Responses
4.2.4 Discussion
4.3 Experiment 3: Short-term plasticity in native speakers
4.3.1 Methods
4.3.2 Materials and methods
4.3.2.1 Subjects
4.3.2.2 Stimuli
4.3.2.3 Discrimination test and training
4.3.2.4 MEG measurements
4.3.3 Results
Behavioral results
4.3.3.2 Evoked Responses
4.3.4 Discussion

5 General Discussion
5.1 Considerations of methodological limitations
5.2 Final remarks on the ends of plasticity
5.3 Summary

6 Bibliography

LIST OF ABBREVIATIONS

illustration not visible in this excerpt

LIST OF FIGURES AND TABLES

Figure 1: Psychometric functions of frequency discrimination in an owl monkey

Figure 2: Expansion of representational areas in an owl monkey

Figure 3: Primary and secondary currents in a pyramidal cell and other types of neurons

Figure 4: Sensitivity of EEG and MEG for cortical activity

Figure 5: The most important AER to the physical features of a stimulus

Figure 6: Example of a MMF: subtraction of standard- from deviant waveforms

Figure 7: Experimental design and stimulation paradigm of Experiment 1

Figure 8: The magnetically shielded room with vacuum cast and dewar

Figure 9: Delta frequency at discrimination threshold in the first experiment

Figure 10: Global field power and latency of the MMF in Experiment 1

Figure 11: Dipole moment, global field power and peak amplitude of N1m

Figure 12: Stimuli with duration differences across 1 mora in Japanese words

Figure 13: Category boundary between long and short Japanese words

Figure 14: Training design

Figure 15: Reaction times of German subjects

Figure 16: MMF amplitudes in Experiment 2 for an exemplary German subject

Figure 17: PCA and grand-average localizations of the MMF

Figure 18: Follow-up case study results

Figure 19: Reaction times of the Japanese subjects

Figure 20: Hit rates of German and Japanese subjects

Figure 21: Psychometric functions of German and Japanese subjects

Figure 22: Global field power of the MMF of German and Japanese subjects

Figure 23: Results of MMF amplitudes and latency

Table 1: Scheme of possible synaptic intercorrelations

Table 2: Phonetic structure and meaning of the stimuli in Experiment 2

Table 3: Average Cartesian coordinates of German and Japanese subjects

ZUSAMMENFASSUNG NEUROMAGNETISCHE KORRELATE DER PLASTIZITÄT IM AUDITORISCHEN KORTEX AUFGRUND VON DISKRIMINATIONSLERNEN HANS MENNING

In einer Vielzahl von Tierexperimenten wurde lern-induzierte Plastizität nachgewie- sen. Obwohl Lernen und Erfahrung die Basis für menschliche Erkenntnis bilden, ist die Zahl der Studien über lern-induzierte Plastizität im menschlichen Gehirn gering. In dieser Dissertation wird der Frage nachgegangen, ob intensives Diskriminations- lernen plastische Veränderungen im auditorischen Kortex bewirkt. Im 1. Experiment wurde daher die Wirkung eines dreiwöchigen Frequenzdiskrimi- nationstrainings untersucht. Die auditorisch evozierten Gehirnantworten N1m und das „Mismatch Feld” (MMF) sind beide sensible „Änderungsdetektoren“. Zehn ge- sunde Probanden trainierten an 15 Tagen jeweils 2 Stunden täglich, immer kleiner werdende Frequenzunterschiede wahrzunehmen. Die Diskriminationsfähigkeit wurde anhand eines randomisierten Diskriminationstests ohne visuelle Rückmeldung er- fasst. Zwei magnetoenzephalographische (MEG) Messungen definierten die „nor- malen” MMN-Amplituden, eine Messung während, am Ende des Trainings und 3 Wochen danach wurden damit korreliert. Im Laufe des Trainings verbesserte sich die gerade noch unterscheidbare Frequenzdifferenz bis auf unter ein Tausendstel (1001 Hz wurde von 1000 Hz unterschieden). Parallel dazu erhöhten sich die Am- plituden der N1m und des MMF bis zum Ende des Trainings und fielen 3 Wochen danach leicht ab.

Im 2. Experiment wurden die neuronalen „Spuren” des Lernens von nicht-nativem Mora-timing untersucht. In der japanischen Sprache bildet ein „Mora” eine zeitliche, rhythmische Einheit ähnlich einer Silbe, die die Wörter in isochrone Segmente teilt (z.B. bestehen sowohl na-ka-mu-ra als auch to-o-kyo-o aus 4 Mora). Deutsche Probanden lernten in 10 Trainingseinheiten von 1 1/2 Stunden japanische Wortpaare zu diskriminieren, die sich in mehreren Stufen durch die Länge eines Mora unterschieden (in einer Konsonant- [anni-ani] und einer Vokalbedingung [kiyo- kyo]. Ein deutlicher Anstieg der Diskriminationsleistung korrelierte signifikant mit der MMF-Amplitude und mit der Abnahme der Reaktionszeiten im Training und der Latenzen der MMF.

Im 3. Experiment wurden japanische Probanden als Muttersprachler mit dem gleichen Paradigma auf Kurzzeitplastizität untersucht. Nach 2 aufeinanderfolgenden Trainingseinheiten (3 h) waren nur geringe Verbesserungen der Diskriminations- leistung und keine signifikante Erhöhung der MMF-Amplitude zu beobachten, aber, im Vergleich zu den deutschen Probanden, eine erhöhte Sensitivität des MMF für kleinere Unterschiede auf dem „anni“- und „kiyo“-Kontinuum vor dem Training. Die früheren Antworten P1m und P2m waren signifikant größer als bei den deutschen Probanden. Ein verändertes Modell Hebb’scher Plastizität wurde auf die vorliegenden Untersuchungen übertragen. Es konnte erstmalig gezeigt werden, dass intensives Diskriminationslernen zu plastischen Veränderungen im menschlichen Gehirn führen, wie sie in neuromagnetischen Antworten gezeigt wurden. Muttersprachler weisen dafür eine erhöhte Sensitivität, jedoch keine Kurzzeitplastizität auf.

PREFACE

The motivation for this thesis came from the intriguing idea that we continuously restructure our brain through everyday learning. How can this highly complex, highly adaptive “learning device” change and reorganize itself all the time while keeping the illusion that we are constantly “ourselves”? The question is, whether learning has the power to trigger functional and structural changes in the brain. Several levels of thinking are involved in an interdisciplinary way. Thus, on a psy- chological level, 3 major topics enter this work: learning, memory and preconscious or pre-attentive perception and processing of information. Pre-attentive perception means that the subjects' attention and awareness is not mirrored in the neuronal response at a great deal. Learning is involved in this study as an improving discrimi- nation of fine frequency and word duration differences; the latter was examined in a group of native and non-native speakers. Memory is referred to as sensory memory, a short-time memory trace that is established through the repetition of the same “standard” stimulus. In the auditory modality this has been termed “echoic mem- ory”. A long, repetitive training engraves deep “traces” into the memory. The life- long training of one’s native language results in a very fast and highly automated long-term memory access. On a neurophysiological level the main topics are plasti- city and the reorganization of the underlying representational brain areas. Plastic changes on a molecular, synaptic and neuronal level and reorganization of cortical “maps” have been demonstrated abundantly in animal studies. On a physical level the measured magnetic fields and the calculation of the source parameters of their underlying neural generators are discussed in the light of the neurophysiological and psychological phenomena. Therefore, the aim of this dissertation thesis was, to transfer the insights of animal plasticity research onto the human brain and to draw a connection line between discrimination learning and the underlying neurophysio- logical changes. In a second step, these effects of discrimination learning are tested on speech perception.

1 Introduction

1.1 Overview

Chapter 1 gives a general overview and explains the purpose of the thesis. Why is the used method suited for human plasticity research? Which preliminary as- sumptions and major delimitations have to be made at the current state-of-the-art in plasticity research? Finally, what exactly do we mean when we talk about “cortical plasticity"? “Plasticity” is the nominal form of the adjective “plastic”, which is derived from the Latin plasticus, “of molding”, and this from the Greek plastikos, from plassein “to mold, form”. Webster's New Encyclopedic Dictionary (1995) gives sev- eral definitions of plastic: “capable of being molded or modeled”; “capable of adapting to varying conditions”; and “capable of being deformed continuously and permanently in any direction without rupture”. We experience every day a changing environment, to which we adapt more or less automatically. While we are concerned with problem solving in a general sense and integration of new information, our brain is performing a hard job: it changes in accordance to the new demands. It can be conceptualized as a continuously changing “learning machine” that adapts to the needs of the subject. Its ability to self-organize and reorganize itself in response to the perceptual input from the external world makes it a proficient self-adjusting “learning device". The purpose of the current thesis is, therefore, to document the changes in cortical evoked responses induced by discrimination learning with one of the foremost brain imaging techniques.

Chapter 2: Starting with basic mechanisms of plasticity on the synaptic level, a model of the structural and functional self-organization of the brain is presented on different levels: from synapses to Hebbian cell assemblies, connected through spe- cific spatiotemporal patterns, to large representational maps. Based on plasticity in the auditory cortex in animal studies (retuning and reorganization of tonotopic maps in owl monkeys, gerbils, rats and cats) attention is drawn to plasticity in human subjects. From existing assumptions of reorganization of cortical maps in clinical populations (amputees, brain-injured and post-operative patients) to congenital blinds, a line is drawn to experience-dependent reorganization of cortical maps in experts. A special - since natural - case of expert learners is in infants. The neurogenesis in infants is compared with plasticity in adulthood.

If there are such abundant changes in the human brain, are they also evoked by the learning of speech sounds, e.g. when learning a new language? What are the mechanisms involved, when new, long-term representations for non-native speech elements are created? How are these non-native speech sounds represented in normal hearing listeners? Human perceptual discrimination learning abilities in the acoustic and speech domain are described. The role of Gestalt and categorical perception of speech sounds is considered.

In chapter 3 the methodology is presented: The neuromagnetic approach is a modern technique which allows an accurate description of the time course of audi- tory evoked responses (AER) and, with a few constraints, the localization of their sources. This makes them suitable for the study of a large number of perceptual and cognitive functions. For instance, the components N1m and P2m of the auditory evoked response can reflect onsets and offsets of speech sounds; N2b and P3m in- dicate an attention switch, whereas the Mismatch Negativity (MMN) reflects the vio- lation of an auditory sensory memory trace and thus, pre-attentive discrimination abilities. When discrimination learning changes the acoustic and phonetic processing of a (speech) sound, this is reflected in the amplitude and latency of the MMN re- sponses. The differential impact of acoustic vs. speech sound features on auditory evoked responses is described and discussed in the next 2 chapters.

Chapter 4 comprises the experimental work. In experiment 1 the hypothesis is tested whether plastic changes in the auditory cortex are reflected by neuromag- netic responses as a result of frequency discrimination learning. After determining the exact procedure in a pilot study, the stimulation was arranged in a staircase pro- cedure and subjects were measured before, during, at the end and as follow-up of a discrimination training. The purpose of experiment 2 was to extend the insights of the first experiment to the processing of speech features. The hypothesis was that learning a non-native speech feature such as Japanese mora-timing is reflected by the sensitivity of the MMN to small changes in the auditory input. Classic non-native speech perception findings suggested that adults have difficulties in discriminating distinctions that are not employed contrastively in their own language. What changes occur in the neuronal substrate when these distinctions are trained and improved? Experiment 3 addresses hypotheses on short-time plasticity in native speakers. Is their well-trained brain still “malleable” by a short intensive training? If yes, is this training effect measurable by means of magnetoencephalography (MEG)? Are their evoked responses different from those of the German subjects? The same MMN paradigm as in Experiment 2 was used with the difference that subjects performed only one block training of 3 hours.

In Chapter 5 the experimental findings are discussed in the light of plasticity literature and conclusions are drawn in relation to the predictions of the models presented at the beginning.

1.2 Delimitation of the topic

What kind of plastic changes can be shown with MEG and what not? Are the results found in animal studies on brain plasticity - with some delimitations - transferable onto the human brain? Why was MEG chosen? Is this method better suited than others for this purpose? However, the basic question is whether plastic changes oc- cur as a result of perceptual discrimination learning in young adults. For this pur- pose, a non-invasive method like MEG has many advantages, but also some restric- tions. The advantages are well known and established. MEG is a non-invasive brain imaging method with a high temporal resolution (up to the milliseconds level) and a rather good spatial localization of the source (with an error of only a few millime- ters). Besides, event-related potentials or fields (ERP/ERF) like the Mismatch Nega- tivity / Mismatch Field (MMN/MMF) are meanwhile well studied and allow an objec- tive, mostly attention-independent measurement of the discrimination performance for acoustic and phonemic features of language. There is a vast literature on early, middle-latent and late evoked potentials and also about components that are spe- cific to a certain frequency band (e.g. gamma band). Several of these components are sensitive for plasticity (MMN, P1, N1, N1c, P2, N2b, P3b etc.) and thus provide a basis for the studies presented here.

For a better understanding of plasticity phenomena, correlates of the measured ac- tivity on the molecular and cellular level should be considered. For instance, Ca² + influx or neurotransmitter release plays a decisive role in plasticity. How is this re- flected in MEG? The neurotransmitters dopamine and serotonine modulate involun- tary and selective attention. In a recent study it has been shown that the MMN to duration and frequency changes was modulated by the serotonine level (Kähkönen et al., 2001).

For the visibility of a neuronal activity in MEG several thousands of parallel oriented neurons are supposed to be simultaneously active. Only the synchronized activity of larger cell assemblies reaches a strength level that is visible in MEG. When this neuronal activity increases post learning compared to before, large numbers of new neurons have been recruited or the synchronicity of the available neurons is enhanced. Although MEG is “blind” for smaller units than cell assemblies and its “resolution” ends at the level of several thousands of neurons, it is one of the foremost and most accurate methods in cognitive neuroscience, which, in comparison with other methods, delivers a high spatio-temporal resolution.

One of the major limitations in plasticity research is that we cannot delimitate with current MEG the exact cortical representation areas as in (histological) animal stud- ies, but rather determine a “center of gravity” of the activity, which is reflected by an equivalent current dipole (ECD). Nevertheless, the plasticity processes described on the molecular and cellular level reach visibility as soon as several thousands postsynaptic currents in parallel oriented dendrites or axons sum together. The pre- vailing assumption in current human plasticity studies is therefore, that experience or training induced plasticity is reflected by an enhanced strength of the source ac- tivity. However, the correlation between larger areas and an enhanced source activ- ity is not stringent. The larger generator amplitudes may also reflect a higher syn- chronization or could be an effect of attention switch and not of an enlargement of the underlying cortical representation area. More attention also results in higher amplitudes.

There are further limitations on the measurement of human brain plasticity. Al- though in animal studies there is a large body of evidence that plastic changes occur on the synaptic level (e.g. dendritic spine growth, sprouting, genesis and survival of new neurons, strengthening of the interconnections between neurons) as well as on the cortical level (e.g. reorganization of cortical representational maps), the studies which demonstrate these processes in humans are scarce. Especially changes in the brain's representation maps as a result of the learning of a new language (with its new phonemes) are still unknown. From hence the current thesis picks up the thread.

2 Theoretical framework

2.1 Cortical Plasticity

Παντα ρει¹ (Heraklitos)

The brain’s ability to change with experience is known as cortical plasticity, very pronounced in (early) childhood, but still possible during adulthood. Prenatally formed neurons merge together to neural networks, which are extended and strengthened during cerebral maturation processes in childhood. However, this strengthening and de-strengthening of neural connections is not finished with childhood, but pertains in some form throughout the entire life. During the past years, modifications of the neural connectivities through (sensory) experience have been studied extensively in animals and, less intensively, in humans.

In humans, the learning of language abilities or musical skills involves considerable flexibility of the brain and is therefore a candidate for auditory cortical plasticity re- search. These plastic changes in the auditory cortex highly depend upon the age at the onset of training, the length and intensity of practice, as early started musical training or learning of a second language in childhood is critical for the feasibility of later learning. In infants, auditory experience shapes the areas involved in the proc- essing of complex sounds and language processing. In adults, the deprivation of one sensory modality (e.g. visual), lifelong experience and intensive learning also can induce plastic changes in the brain.

First of all, a basic model of synaptic strength modifications in distributed neural networks is presented (2.1.1). A variety of phenomena related to this kind of plas- ticity are described (2.1.1.1 - 2.1.1.5). In a second step, the modified Hebbian plas- ticity model is extended to the reorganization of cortical representation maps (2.1.2 - 2.1.4). Then, the state of the art for cortical plasticity research is drawn from ele- mentary animal studies to clinical populations, congenitally blind humans and ex- perienced learners. Experience-dependent plasticity is presented (2.1.5 - 2.1.7). The model is extended from cell assemblies to spatiotemporal patterns and the inter- correlations between plasticity phenomena and perceptual learning are discussed (2.2 - 2.5).

2.1.1 Basic principles of cortical plasticity

Cortical plasticity can be defined as the potential of the neocortex its function by strengthening or loosening synaptic connections. At the end of the nineteenth century, the psychologist William James suggested that learning might alter synaptic connectivity (James, 1890). Half a century later, Donald Hebb developed a model, which assumed that learning and memory are based on changes of synaptic efficacy (Hebb, 1949). The tenet of Hebb's theory was that simultaneously active adjacent pre- and postsynaptic neurons mutually contribute to a strengthening of their synaptic connections. Non-simultaneously active adjacent neurons do not strengthen their connections. “Cells that fire together, wire together”.

Learning occurs throughout life. In terms of plasticity, learning can be conceived as the induction of neural change through experience and memory as the storage and application of these changes. The mechanisms underlying these changes are briefly explained in the next section. First, a short overview over the basic mechanisms and forms of plasticity that occur on the pre- and postsynaptic interface is given (2.1.1.1-2.1.1.4). The special contribution of enriched environment to plastic changes in the brain is described in the next section (2.1.1.5).

2.1.1.1 Synaptic plasticity - the weighting of synaptic strength

It is a fundamental assumption in cognitive neuroscience that changes in the synap- tic weights (efficacy) between two neurons or in a cell assembly are the substrate for learning and memory. Associative or Hebbian synaptic plasticity is thought to underlie the experience-dependent plasticity. On the molecular level, fast and slow changes in the postsynaptic neuron are induced on specific (ionotrope and me- tabotrope) receptors by presynaptic facilitation. The basic mechanism is this: A “second messenger” called cAMP (cyclic Adenosine Mono-Phosphate) activates an enzyme (Protein Kinase A or C) in the synapse, which regulates the biosynthesis of

proteins. The release of these enzymes closes the K+-channels, so that the duration of action potentials is prolonged (the repolarization of the depolarized neuron is hin- dered). This enables an enhanced influx of Ca² +-ions which causes an enhanced re- lease of neurotransmitters (Byrne & Kandel, 1996). On the cellular level, the re- peated activation of the presynaptic sensory synapse, which transmits the condi- tioning stimulus (CS, e.g. an electric shock), concomitant with an unconditioned stimulus (UCS) on the postsynaptic neuron, strengthens the connection of these two neurons. This association of the activity of two neurons is the simplest form of learning.

In conclusion, the combination of a CS with an UCS induces the cascade of proc- esses described above: an enhancement of cAMP in the synapse concomitant with a Ca² + influx and a transmitter release. These are the major factors of short-term fa- cilitation, which lasts for a few hours. The long-term storage implies that the second messenger cAMP activates the biosynthesis in the cell body of the neuron. When this process is blocked, no long-term storage can occur (Castellucci et al., 1989). The protein biosynthesis induces structural changes in the neuron. Long-term habituation in the Californian snail Aplysia results in less active transmitter release zones and less synaptic vesicles, whereas sensitization results in more and larger active transmitter releasing zones and more synaptic vesicles (Bailey & Chen, 1983, 1988). The transition from short-term facilitation to a long-term consolidation (over periods of hours, days or weeks) or, in psychological terms, from short-term memory (STM) to long-term memory (LTM) is known as long-term potentiation (LTP).

2.1.1.2 Long-term potentiation

LTP is one of the best-studied phenomena of neuroplasticity. The foundation of this theory goes back on Donald Hebb's assumption that the facilitation of synaptic transmission constitutes the basic mechanism for learning and memory (Hebb, 1949). He postulated that each experience evokes an unequivocal pattern of neu- ronal activity in so-called “reverberation circuits” which form the STM for that ex- perience. The repeated activation of these reverberation circuits induces a long-term potentiation of these synaptic connections and forms the basis for the LTM. The long lasting facilitation of synaptic transmission after activation of the postsynaptic neuron has been shown to follow an intensive stimulation of the presynaptic neuron (Bliss & Lømo, 1973). When afferent axons of the (CA1 region of the) hippocampus are repetitively stimulated with electric shocks in the natural hippocampal Theta- rhythm (4-8 Hz), the associative LTP is particularly strong. The simultaneous stimu- lation of the hippocampal neuron with a weak CS and a strong excitatory UCS en- hances the excitatory postsynaptic potential (EPSP) following the weak stimulus CS as well.

The main characteristics of LTP are that a single stimulation already evokes a long LTP of minutes and hours and several stimulations evoke a week-lasting potentia- tion. LTP is induced only when a postsynaptic firing follows the presynaptic firing. No LTP is induced if only the presynaptic cell is firing (but not the postsynaptic cell) or vice versa (Kelso et al., 1986). This co-occurrence of pre- and postsynaptic firing is known as “Hebb's rule” and forms the basis of neuroplasticity. This rule says that simultaneous pre- and postsynaptic activity results in a strengthening of that con- nection. A coincidence detector in the neurons records the co-occurrence of pre- and postsynaptic activity. The NMDA (N-Methyl-D-Aspartate)-receptor was found to fulfill these requirements.

LTP occurs due to the coincidence of partial depolarization (and partial Ca² + inflow, this time over a longer period) on the postsynaptic membrane and the simultaneous release of glutamate from the presynaptic neuron. Glutamate is the most important excitatory neurotransmitter in the brain. When released into the synaptic gap, glu- tamate is docking on the NMDA receptor of the postsynaptic membrane and elicits a maximal Ca² +-inflow, which then induces LTP through protein kinase in the cyto- plasm. This co-occurrence of pre- and postsynaptic activity strengthens the synaptic connection.

The glutamate-NMDA-coincidence-detection-mechanism was found in the CA1-re- gion of the hippocampus, in the cortex and in parts of the basal ganglia. LTP en- hances, after the presentation of learning material, the excitability of hippocampal and cortical neurons for hours. In this period occurs the consolidation of the learning content; it is transferred from STM to LTM. The supply with noradrenaline markedly enhances and prolongs LTP (for an overview see (Bliss & Frégnac, 1996).

2.1.1.3 Long-term depression

Additional to the induction of LTP for the establishment of changes in synaptic weights and neural excitability, the opposite phenomenon to LTP, called long-time depression (LTD), is observed as well. LTD is the long lasting, activity-dependent decrease in synaptic efficacy and was first described in the CA1 layer of the hippo- campus in vitro (Lynch et al., 1977). While most of the neurons involved in the modification of synaptic strength are LTP-specific, LTD is important for the capacity to modify synaptic efficacy bi-directional. Consistent with the synaptic plasticity and memory (SPM) hypothesis (Martin et al., 2000), any input-specific up- or down- regulation of synaptic strength of at least 1 hour is referred to as LTP and LTD. LTD seems to complement LTP and might play an important role for the ensurance of storage capacity. Although LTD is not dependent on NMDA receptors, a modifica- tion-threshold-model was proposed that accounts for both LTP and LTD (Artola et al., 1990). Both LTP and LTD rely on the influx of Ca² +. When the Ca² + influx is below a particular threshold, LTD occurs; when the threshold is exceeded, LTP occurs. The main role of LTD for the reorganization of cortical stimulus representations is seen in the depression of inputs from inactive afferents. Further, inhibitory plasticity might play an important role in cortical map reorganization (Jacobs & Donoghue, 1991, Jones, 1993). The balance of excitatory and inhibitory inputs on the presynaptic terminal determines whether an excitatory or inhibitory postsynaptic potential (EPSP or IPSP) occurs or not. Like the regulation of EPSP's in excitatory plasticity, both increases and decreases in IPSP's can be induced. Thus, it seems that plasticity applies not exclusively to synapses between excitatory cells, but also to inhibitory synapses.

2.1.1.4 Axonal sprouting

When an axon degenerates, the adjacent healthy axons sprout into the empty space left by the degenerated tissue and form new synaptic contacts. This collateral sprouting is activated by degenerating tissue (Piñel, 2001). This form of plasticity not only keeps the cortical tissue functioning throughout neural cell loss, but also allows the brain to recur on unused resources.

The earliest form of plasticity occurs in the maternal womb. In the course of the de- velopment of the neural plate in the embryo the cells in the dorsal ectoderm are totipotent. Each cell has the potential to develop into any cell of the body. During development, the axons sprout to their specific topographic place and form synaptic connections to the adjacent cells. During the following period, the active neurons are selected out and survive the inactive ones. Redundant neurons, which devel- oped “useless” connections, die through apoptosis. The gap on the postsynaptic membrane is replenished with axonal endings of the surviving “active” neurons. Thus, the cell death leads to a massive reorganization of synapses. In other words, the use or non-use of neurons is also apt to reorganize the brain. Inactive neurons and synapses do not survive (Hockfield & Kalb 1993, Kalil, 1989). Mental training prevents the loss of neurons by strengthening existing connections or building new ones.

2.1.1.5 Neurogenesis in enriched environments

Until the second half of the twentieth century, the brain was considered by scientists to be immutable, subject only to genetic control. In the early sixties, however, in- vestigators started to study the capability of different environmental influences to alter the brain structure. Beginning in 1964, laboratory research proved that the morphology and physiology of the brain could be experientially altered (Hubel & Wiesel, 1965). Since then, the capacity of the brain to respond to environmental input, specifically “enrichment”, has become an accepted fact among neuroscientists. In fact, the demonstration that environmental enrichment can modify structural components of the rat brain at any age altered prevailing presumptions about the brain's plasticity. The cerebral cortex, the area associated with higher cognitive processing, is more receptive than other parts of the brain to environmental enrichment (Diamond, 2001).

There is plenty of evidence from animal studies that a rich, complex environment not only improves the learning performance of rats, but also induces thicker cortices and more synapses per neuron than in isolated rats raised in poor environments.

(Sirevaag & Greenough, 1988; van Praag et al., 1999b; 2000). Differential rearing conditions alter the spine density on neurons in the rat's corpus striatum when growing up in a complex environment vs. an individual cage (Comery et al., 1995, Varty et al., 2000). A complex, stimulating environment enhanced the levels of neurotrophin-3 (NT-3) mRNA in the rat's visual cortex and hippocampus, which triggered the growth of new dendritic spines (Torasdotter et al., 1996). Hungry rats, which observed the realization of a duration discrimination task, rapidly learned to perform that task by themselves. A “naïve” control group, which did not observe the task showed a slower and smaller amount of learning (Lejeune et al., 1997). En- riched environments, where mice received complex inanimate and social stimulation, fostered the genesis and survival of new cells in the dentate gyrus of the adult mouse hippocampus (Kempermann et al., 1998). Moreover, these enriched environ- mental conditions inhibit spontaneous apoptosis, prevent seizures, are neuroprotec- tive against exitotoxic injuries and induce expression of neurotrophic factors (Young et al., 1999). Several in-vivo high resolution imaging studies of living cells showed in the developing rat barrel cortex a rapid experience-dependent plasticity. Dendritic spines and filopodia were generated or elongated within only tens of minutes (Segal, 2001). Functional plasticity triggers both the formation of novel dendritic spines and pruning of others in cultured hippocampal networks (Geinisman et al., 2001, Goldin et al., 2001, Yuste & Bonhoeffer, 2001). Besides a thicker occipital cortex, more synaptic contacts per neuron and larger dendritic arbors, the cerebral blood capillaries of rats reared in complex environments were closer together than those of individual or pair wise reared rats. The closer capillary spacing suggests compensatory angiogenesis in response to increased metabolic demand due to previously occurred plastic changes (Black et al., 1987).

Furthermore, exploration of a novel environment leads to the expression of inducible transcription factors in rodents' barrel-related columns. This adaptive plasticity of cortical receptive field properties of the whiskers occurred after that adult rats were placed overnight in an enriched environment (Staiger et al., 2000). These studies show that an enriched sensory experience induces learning, which can alter the neuronal development profoundly. In humans, a similar impact of “enriched envi- ronment” on the mental development of children is claimed. For clinical groups, an association between higher educational attainment and reduced risk of Alzheimer and Parkinson-related dementia also indicates that a stimulating environment has positive effects on the cerebral health of these patients.

2.1.2 From synapses to representational maps

The molecular and neuromodulatory underpinnings of plasticity, including associa- tive LTP and LTD of excitatory postsynaptic potentials have been described on the molecular and cellular level in section 2.1.1. It was shown that learning, mainly in an enriched environment, can change the connectivity between neurons by assigning different weights to the synaptic junction and even can induce morphological changes through neuro- and spinogenesis. This cortical synaptic plasticity is strongly linked to cortical map plasticity. Recent experimental and theoretical work provides support for the hypothesis that synaptic and cellular mechanisms underlie cortical representational plasticity (For a review see (Buonomano & Merzenich, 1998). Corti- cal maps are, thus, dynamic constructs that are remodelled by experience, pre- sumably throughout life.

Both LTP of excitatory synapses and LTD of inhibitory synapses follow a modified Hebbian correlative learning rule. At the synaptic level, Hebbian plasticity is involved when synaptic strength between neurons that fire simultaneously is increased or decreased. At the cortical level, Hebbian learning refers to the detection of tempo- rally correlated inputs that lead to the formation of topographic maps and of (some- times widespread) cell assemblies that specifically represent newly learned material. Topographic representations arise from peripheral inputs that fire in close temporal proximity and which are most likely represented in adjacent areas in the sensory cortex (for reviews see (Merzenich 1987, Merzenich & deCharms, 1996, Merzenich et al., 1990a, Merzenich et al., 1990b, Merzenich & Sameshima, 1993).

In the following sections 2.1.3 - 2.1.6 different studies are presented which demon- strate that cortical maps of adult animals (and, in a few studies, of humans: section 2.1.7) are not statically fixed, but rather dynamically changeable. The cortex can allocate cortical space depending on the needs given by the peripheral input. When there is a lack of input (deprivation or deafferentation), the free space is allocated to new functions. When there is abundant input (intensive training or experience), new space is recruited. In chapter 4, plastic changes in the responses generated by lar- ger cell assemblies are shown for human subjects in three forms of perceptual learning: in an intensive frequency discrimination training, a short-term discrimina- tion training of a native and a long-term discrimination training of a non-native speech feature.

2.1.3 Reorganization of cortical maps by deprivation

Learning in enriched environment changes the neuronal development (2.1.1.5). However, what happens when the brain or parts of it are cut off from sensory input? Plasticity can also be induced by cortical deprivation or deafferentation. Based on this assumption, Hebbian mechanisms of neuroplasticity were found in the visual cortex (Rauschecker, 1991) as well as in the auditory cortex (Rauschecker, 1999). Animal studies with visually deprived cats (Rauschecker & Kniepert, 1994) and fer- rets (King & Parsons, 1999) demonstrated a reorganization of the auditory cortex towards improved sound localization. The anterior ectosylvian visual area is entirely re-dedicated to auditory and somatosensory inputs (Rauschecker & Kniepert, 1994, Rauschecker & Korte, 1993).

Early visual deprivation of only a few days already results in a remarkable reduction of axons in the corpus geniculatum laterale for the deprived eye and in a broadening of the ocular dominance columns for the undeprived eye (Hubel et al., 1977). Animals raised in the dark show less synapses and dendritic dorns than normally raised animals. In parallel, they have deficits in spatial vision and pattern recognition (Hata & Stryker, 1994).

Amputees report “phantom pain” in the somatosensory circuitry responsible for the amputated limb. As a result, the adjacent areas expand into the free areas. Auditory cortical plasticity can also be demonstrated after lesions of the cochlea. The tonotopic organization of primary auditory cortex is altered subsequently to re- stricted cochlear lesions and the topographic reorganization is correlated with changes in the perceptual acuity of the animal (Robertson & Irvine, 1989).

These exemplary studies show that deprivation or deafferentation from sensory pe- riphery results in a loss of synaptic and dendritic density, but frees these areas for new functions.

Cortical maps undergo plastic changes in response to both sensory loss of input as well as increased input through learning and experiences. Therefore, a lesion of the sensory pathway, an injury of the sensory cortex itself or a change in the sensory experience can induce reorganization.

2.1.4 Reorganization after lesions in sensory areas

Studies of cortical plasticity are mostly conducted in adult primary sensory areas. Because the topographic organization of the primary sensory cortices is well known, their reorganization can be studied thoroughly. In the somatosensory cortex, maps of skin surface are organized somatotopically (in form of a “homunculus"). Neighboring skin sites activate neighboring cortical areas. The same principle is valid for motor areas. Similarly, auditory and visual cortical maps are organized according to tonotopic and retinotopic principles. Hereby, the cortex allocates cortical area in a use-dependent manner: for the more used functions more cortical space.

As an example of a lesion of the peripheral pathway, (Kaas et al., 1990) found after a small lesion in the cats' one retina and resection of the other retina a reorganization of the receptive field neurons in the primary visual cortex. Such changes started within minutes after the lesion (Gilbert & Wiesel, 1992). The complete amputation of an arm of a monkey showed ten years later a considerable reorganization of the underlying somatosensory neurons: the amputated arm area was systematically “invaded” by cortical face representations (Pons et al., 1991).

An injury of sensory brain areas also evokes a reorganization of adjacent areas. Jen- kins & Merzenich resected an area of the somatosensory cortex that reacted on the touch of a hand. Several weeks later, adjacent neurons reacted to the touch of the hand. They took over the function of the resected neurons (Jenkins & Merzenich, 1987).

Early blindness in humans results in a refinement of auditory perception and an ex- pansion of auditory areas into the parietal cortex. In blind people the ability to lo- calize sounds showed higher precision than in sighted subjects (Lessard et al., 1998, Röder et al., 1999). A positron emission tomography (PET) study of auditory localization shows that congenitally blind humans activate large parts of the occipital lobe additionally to the posterior parietal areas activated in sighted subjects. These findings suggest that a cross-modal expansion of auditory territory in the cerebral cortex enables a refinement of auditory perception as well. The cortical space freed by a lesion or deafferentation of the peripheral pathway or an injury of the cortical area is reassigned to other brain functions.

2.1.5 Experience-dependent plasticity

The ever-changing cerebral cortex, with its complex microarchitecture of unknown potential, is powerfully shaped by experiences before birth, during youth and, in fact, throughout life. Reorganization of topographic maps in the (primary) sensory cortices after changes in the sensory experience was demonstrated in a series of animal studies. For instance, the primary auditory cortex could entirely be converted with the “false” experience. In a study on ferrets, the evolution of retinal ganglion cells was directed towards the corpus geniculatum mediale of the auditory system instead of the (normal) corpus geniculatum laterale of the visual system. After growing up, the auditory neurons of these animals reacted to different visual stimuli and their auditory cortex showed a retinotopic organization (Roe et al., 1990). In another study, barn owls were raised with eye prisms in front of their eyes that switched the view by 23 degree to the right. This altered their auditory spatial map, which is aligned with the visual spatial map in the tectum. As a result, these neu- rons became tuned to sound source locations corresponding to their optically dis- placed, rather than their normal, visual receptive field locations. The auditory map was also switched by 23 degree to the right (Knudsen & Brainard, 1991). Learning and experience cannot only modify the spatial representation maps of sounds but even tonotopic maps. For instance, if during the development of a birds’ brain a partial hearing loss occurs, the tonotopic organization of the auditory cortex changes in accordance to the audible frequencies (Scheich, 1991).

These results show that both a damage of neuronal structure and experience can induce either a fast reorganization that occurs within hours and “tunes” the brain for new experiences, or slow, profound reorganizational changes, able to compensate damages of the neural system or to specialize the brain to new functions. Two mechanisms can account for these changes: the Hebbian rule predicts a rapid rein- forcement of simultaneously active synaptic connections; on the other hand, “collat- eral sprouting", the formation of new dendritic spines and synaptic connections leads over longer periods of time to a large-scale reorganization. Kilgard and Mer- zenich found that the repeated stimulation with a tone during a six weeks period markedly enlarged the corresponding area in the rat's auditory cortex (Kilgard & Merzenich, 1998a, Kilgard & Merzenich, 1998b). This reorganization occurred only when the tone stimulation was paired with electric stimulation of the basal prefron- tal cortex, an area involved in memory processes and attention. For a detailed dis- cussion of the importance of experience-dependent alterations for the generation of new cells or cell survival see (Gould et al., 1999, Kempermann et al., 1997, van Praag et al., 1999a).

2.1.6 Reorganization of cortical maps after training

Besides changes of cortical maps through altered sensory experience, training or conditioning can induce plasticity as well. Recanzone and his colleagues showed in several experiments that discrimination training enlarges the receptive fields (RFs) in primary somatosensory and auditory cortices (Recanzone et al., 1992a, 1992b, 1992c, 1992d, 1992e, 1993). For instance, adult owl monkeys were trained over a period of 3-20 weeks to detect with a digit small vibration differences from the standard frequency of 20 Hz. Psychophysical thresholds for the trained digit progressively decreased from a 6- to 8-Hz difference to a 2- to 3-Hz difference relative to a 20-Hz standard (Recanzone et al., 1992a). In a subsequent experiment with new world monkeys, an expansion of the receptive fields in area 3b in the somatosensory cortex was observed (spanning several hundred microns) as a result of intracortical microstimulation (ICMS) (Recanzone et al., 1992b). During a tactile frequency discrimination training, the representational area of the trained finger (area 3a) expanded concomitant with improvements of behavioral performance to a size up to 3 times greater than the representations of control digits (Recanzone et Discrimination Learning and Auditory Plasticity al., 1992c). In the auditory modality, changes of the frequency representation areas in the primary auditory cortex (A1) were observed concomitant with an increase of the animal's perceptual acuity at a frequency discrimination task. Adult owl monkeys were trained for several weeks to discriminate small differences in the frequency of sequentially presented tonal stimuli (Recanzone et al., 1993). In the course of the training, a progressive improvement of the discrimination performance was observed (cf. Fig. 1). At the end of the training period, the tonotopic organization of A1 was defined by recording multiple-unit responses at 70-258 cortical locations. These responses were compared to those derived from untrained monkeys. Fig. 2 shows the enlargement of the trained frequency of 2.5 kHz within the tonotopic

illustration not visible in this excerpt

Figure 1: Psychometric functions of the trained frequency of 2.5 kHz in different training sessions in the owl monkey “OM3”. (From Recanzone et al. 1993).

organization of A1 before and after the training. The cortical representation area and the sharpness of tuning were greater for the behaviorally relevant frequencies of the trained monkeys. These investigations demonstrate that attended stimulation can modify the cortical organization of primary sensory cortices in the adult primate, and that this alteration is correlated with changes in perceptual acuity.

The hypothesis, that such receptive field (RF) plasticity is induced by a Hebbian rule of mutual strengthening, when the cells are simultaneously active, was called “the covariance hypothesis”. Cruikshank and Weinberger (1996) tested whether a “covariance treatment” alone, without behavioral experience, is sufficient to induce RF plasticity.

They paired in guinea pigs a tone far from the best frequency with an excitatory juxtacellular current applied to a single postsynaptic cell in A1, thereby increasing covariance between activity of the postsynaptic cell and its afferents that were activated by the tone. In alternation, within the same cell a second tone close to the best frequency was paired with inhibitory juxtacellular current, decreasing covariance between the postsynaptic cell and afferents activated by the second tone. After treatment, responses to tones associated with increased covariance strengthened significantly relative to tones associated with decreased covariance, as predicted by the Hebbian presynaptic-postsynaptic covariance hypothesis. (Cruikshank & Weinberger, 1996).

Ahissar and colleagues assessed associative learning-induced changes in the correlation strength of distant neurons (Ahissar et al., 1992). One neuron of a pair received the conditioned stimulus (CS) and the other functioned as a conditioned response (CR) neuron. The unconditioned stimulus (UCS) was an auditory stimulus that was capable to drive the CR neuron and guided the monkey's performance on

illustration not visible in this excerpt

Figure 2: Expansion of representational areas for the frequency of 2.5 kHz before (A) and after discrimination training (B) in the owl monkey “OM3” (From Recanzone et al. 1993).

auditory task. The activity in the CS neuron triggered the UCS and such, the CR neuron. Cross correlations before and after training revealed an increase in the coupling strength between the CS and CR neuron. The results show that this associative conditioning induces a powerful Hebbian plasticity in excitatory intracortical connections. In another animal study the intracortical microstimulation (ICMS) of an area in the rat's primary motor cortex, which evoked an arm movement, was repeated for an hour. Following repetitive ICMS, significant changes in movement representations were observed in the motor cortex (Nudo et al., 1990). These results show that already a small amount of training is able to trigger a reorganization of cortical areas.

2.1.7 Reorganization of cortical maps in experienced learners

The neural mechanisms of plasticity are likely to be the same across all cortical regions and modalities (Rauschecker, 1999). Auditory perception largely varies between individuals and is modulated by previous learning.

The most impressive plastic changes occur in the brain during the early development. Most of the 10 - 100 billions of nerve cells develop in the first months of pregnancy with up to half a million per minute. However, after birth, these neurons are not yet entirely connected with each other. In the first year after birth a myriad of neuronal connections develop in the neocortex. In this period, babies are able to learn every language (Kuhl et al., 1992). In the following years, a genetically controlled reduction of superfluous connections is started in the brain. The baby develops the prototypes for its native language and for all other (cognitive) skills and abilities. At the age of 9 months, the phoneme prototypes for the native language are already formed (Jusczyk, 2002). All continuously used cell assemblies strengthen their connections. All unused connections are faded out. This way, the organization of the brain is optimized and representational maps are formed step by step. Decisive hard-wiring connections are triggered in this period. What is learned in this critical time window is critical for future learning. This is the period of greatest plasticity. The structures built in this period act like a filter for all further information.

Early musical training leads to an expansion of the representation areas of complex harmonic sounds in the auditory cortex (Pantev et al., 2001a). Similarly, the early phonetic environment has a strong influence on speech development (Cheour et al., 1998) and, presumably, on the cortical organization of speech. An MEG study of the auditory modality (Pantev et al., 1998) showed increased cortical representations of piano tones in musicians compared to non-musicians. In a similar study of the somatosensory modality (Elbert et al., 1995) the representation areas of the fingers of the left hand were larger in a group of string players than in a control group. In a recent study, conductors show greater differentiation between instrument positions in the peripheral auditory space than pianists or nonmusicians (Münte et al., 2001).

2.2 From cell assemblies to spatiotemporal patterns

Donald Hebb developed a model in which information is represented in spatiotemporal patterns of activity within cell assemblies in the higher nervous system. He assumed that these cell assemblies form the neurobiological representations of cognitive elements such as gestalt-like figures or words (Hebb, 1949). His model is based on three assumptions: The first, best known as: “cells that fire together, wire together” explains the basic principle of the genesis of a cell assembly. The second assumption is, that these interconnections are not restricted to adjacent neurons, but can cover large distances. It is well known from neuroanatomical studies, that most pyramidal cells have long axons that reach distant areas, including subcortical structures. Similarly, connections from one area project to several other areas. This implies, that the entire cortex acts as a large associative memory, consisting of highly specific neuronal assemblies. Such assemblies for speech features, words or concepts could be distributed, for example, over Broca's and Wernicke's areas and parts of the temporal, prefrontal and parietal lobe. The third assumption synthesizes the two previous ones: When cells that fire together, wire together over large areas, they become a functional unit, a cell assembly. Accordingly, the representation of a single word must not be restricted to a small area, but could be distributed over a large area connected through mutual functional activation. A variety of electrophysiological data support the first assumption that cortical neurons, that are active at the same time strengthen their connections. This effect is well established as LTP (Ahissar et al., 1992) and was discussed together with LTD in the previous sections 2.1.1.2 and 2.1.1.3. While LTP leads to a strengthening of the synaptic contact between two neurons after repeated activation, LTD leads to a weakening of the synaptic connection (Artola & Singer, 1987, 1993; Rauschecker & Singer, 1982). Both, the strengthening and the reduction of the influence of one neuron on the other are long-lasting phenomena, which may last for several hours, days or weeks. A slight modification of the synaptic learning model according to a Hebbian rule was proposed (Pulvermüller, 1999). Accordingly, connection strength between two neurons is not only modified by simultaneous activity; it also changes when only one of the two neurons is active while the other one is inactive.

This kind of learning is called correlation learning, because it delivers not only information about the frequency of coincident firing of the neurons, but also about the strength of the correlation between their activations. These correlations have a rather gradual character and suggest a fine-graded synaptic dynamics in the neocortex which is adjusted in conformity with new learned contents (Tsumoto, 1992). Table 1 depicts the four possible intercorrelations between two neurons.

illustration not visible in this excerpt

The correlated activity establishes an anatomical and functional unit called a “cell assembly” (Hebb, 1949). As we have seen for LTP, the simultaneous activation of pre-and postsynaptic neuronal cells strengthens their connection, whereas the activation of only one (pre- or postsynaptic) neuron concomitant with inactivity of the other neuron weakens their connection (LTD). Between inactive neurons no change of their connection happens.

The “ignition” of the cell assembly occurs explosion-like when a sufficiently large number of neurons are stimulated by external, sensory fibers or by internal cortico- cortical fibers. The activity will spread to additional members of the cell assembly and, finally, will activate the entire assembly (Braitenberg & Pulvermüller, 1992). After ignition of the whole assembly, the activity will not stop immediately. A wave of excitation will circulate and reverberate in the many loops of the assembly. These reverberation circles make the strong connections even stronger and give the assembly an inner structure. The excitation wave can be described as a spatiotemporal pattern of activation and sustained reverberation, in which many neurons participate. This principle is the same from small subgroups up to complex, highly organized packs (“columns”) of neurons.

There is abundant experimental evidence that exactly timed spatiotemporal firing patterns occur in neurons (Abeles, 1982, 1991; Abeles et al., 1993, Aertsen et al., 1989, Gerstein et al., 1989). These synchronously active neuronal connections were termed synfire chains, where one active subgroup of neurons activates the next subgroup and keeps the chain in an active state. These spatiotemporal activity patterns point to the reverberation circuits postulated by Hebb (1949) and are explained by loops in the synfire chain (Abeles et al., 1993).

On the cognitive level, the ignition of the cell assembly may correspond to the perception of a stimulus and to the activation of its representation, whereas the reverberating activity may represent short-term memory (Fuster, 1995). The memory traces established in an oddball paradigm and measured by means of mismatch negativity may also correspond to this well-timed spatiotemporal activity found in animal studies (Fuster, 1995, Villa & Fuster, 1992). The reverberation in recurrent circuits may keep the network in an active state.

The question of where stimulus-specific cell assemblies are active received evidence from another line of research. Synchronized oscillations are a special case of well- timed activity (Abeles et al., 1993, Aertsen & Arndt, 1993). This coherent rhythmic activity was found over large distributed areas and indicates that these areas are strongly coupled and act as unite (Engel et al., 1990, 1991, Gray et al., 1989, Kreiter & Singer 1992). Because synchronized responses change with the stimulus (Eckhorn et al., 1988, Gray & Singer, 1989), the synchronous activity presumably represents a stimulus-specific neuronal assembly. Applied to the Hebbian cell assemblies, at least a subgroup of the neurons of an assembly would exhibit synchronous activity during reverberation.

Support for this hypothesis, that stimulus-specific distributed cell assemblies oscillate synchronously, comes from a research line in a higher frequency range (> 24 Hz), known as gamma-band (Bertand & Pantev 1994, Pantev, 1995, Tallon-Baudry & Bertrand, 1999, Tallon-Baudry et al., 1997). Synchrony of high frequency cortical activity indicates “Gestalt” perception, the combination of uncorrelated percepts to a perceptual unite. Gestalt-like figures such as Kanizsa's triangle evoked stronger gamma band responses around 30 Hz than physically similar figures that are not perceived as a coherent Gestalt (Tallon et al., 1995, Tallon-Baudry et al., 1996). This data support the idea that a coherent perception activates cell assemblies that generate the cortical representations of that triangle.

Furthermore, the activation of one cell assembly may spread to an adjacent assembly and form a more complex, higher order assembly (Braitenberg, 1976). This hierarchical organization of cell assemblies could be the substrate for the complex semantic representations of words. Cortical loci involved in the representation and processing of words may comprise not only the perisylvian language areas of Broca and Wernicke, but also areas in the temporal, prefrontal and occipital lobe. Words - but not pseudowords - are assumed to be cortically represented by such cell assemblies (Eulitz et al., 1996, Pulvermüller et al., 1994). Semantic networks may have their cortical representations in cell assembly networks. the modified Hebbian framework sketched in Table 1, a weakening of connections through LTD occurs between neurons when they are not simultaneously active. This would trace the limit of the neuronal assembly against neurons outside the assembly, which do not reach the threshold and are not active at the same time.

The sketched framework of distributed spatio-temporal networks shows, how the brain is organizing and reorganizing itself from the level of single neurons to highly complex, reciprocally connected assemblies and how it forms the representations of our percepts. Specific synfire chains are created by repeated activation of cortical areas and work together as functional units. Through repeated stimulation, a set of neurons will become active at the same time and build the representation of that percept. Changes in these representations are the basis for learning.

2.3 Perceptual discrimination learning

2.3.1 Discrimination learning

Discrimination learning is a wide spread phenomenon in everyday life. Infants learn to choose from the first day of life between alternatives. They learn to distinguish their parents from other adults by discriminating a number of cues that pertain only to their parents. They learn to discriminate good things to eat or to do from bad things; good places from bad places, interesting things from tedious things and so on. The adult discrimination is based on the preferences learned by discrimination in childhood, but is continuously refined throughout life. Discrimination learning sets the preferences for Beethoven or Rock music or determines, whether one decides to become an artist or a scientist. Preferred passages of music are often heard so that the connections within their cell assemblies are strengthened and their cortical representation areas are enlarged. Simultaneously, the repeated listening to the preferred passages continuously improves the discrimination performance. For example, “experienced” listeners of classic music already recognize the style of a conductor in the first minutes and connoisseurs of wine can “taste” not only the region of origin, but also the year in which the grapes grew. William James already noted in 1890:

That “practice makes perfect” is notorious in the field of motor accomplishments. Motor accomplishments depend in part of sensory discrimination. Billiard playing, rifle shooting, tightrope dancing demand the most delicate appreciation of minute disparities of sensation as well as the power to make accurately graduated muscular response thereto. In the purely sensation field, we have well-known virtuosity displayed by the professional buyers and testers of various kinds of goods. One man will distinguish by taste between the upper and the lower half of a bottle of Old Madeira. Another will recognize by feeling the flour in a barrel whether the wheat was grown in Iowa or Tennessee. The blind deaf-mute Laura Bridgman had so improved her touch as to recognize after a year’s interval the hand of a person who had once shaken hers (James, 1890, p. 509).

These are examples for discrimination learning and, as we assume today, for plasticity. The visually discriminable features of an object like size, color, brightness, shape, complexity etc. are discriminated step by step from another object. An auditory “object” is characterized by its frequency, intensity, duration, pitch, harmonic structure, envelope, complexity and so on. Speech sounds have further characteristics that can (and must) be discriminated from other speech sounds: formant frequencies, Voice-onset-time (VOT), places of articulation, timbre, intonation, accentuation etc. The discrimination learning of fine differences between these parameters are a necessary prerequisite for understanding. For the native language, the discrimination of its characteristic speech features is already learned in the first year of life (Jusczyk, 2002). An almost “classic” example for discrimination learning in the speech domain is the training of Japanese listeners to identify English /r/ and /l/ (Bradlow et al., 1997, Flege et al., 1996, Lively et al., 1993, Lively et al., 1994, Logan et al., 1991). A possible explanation for the distinction of slightly different phenomena is, at first, that a repetitive training produces a cumulative effect: the more training, the higher the probability that “statistical learning” occurs (cf. Breitenstein & Knecht, 2002); at second, each sensation calls up its own associations. If these associations are sufficient different, the sensations themselves are more readily discriminated, the associations enhance the distinction (for a comprehensive review of discrimination learning see: Riley, 1968). Both explanations fit well into the model described above: repeated activation strengthens the connections within a cell assembly and weakens the connections to neighboring cell assemblies that represent similar perceptions; on the other hand, the activation of a part of the cell assembly spreads through associative processes to the entire assembly and new connections with other cell assemblies are established.

2.3.2 Perceptual learning

Perceptual learning comprises all forms of learning, which change sensory perception. This learning is assumed to change both the synaptic connectivity within cell assemblies and the size and organization of cell assemblies at the level of representational maps. Studies of receptive field (RF) plasticity show changes in the absolute amplitude of neurophysiological responses as a consequence of frequency tuning. Highly specific RF plasticity was observed after a sensitization training, characterized by maximal increased responses to the conditioned stimulus (CS) frequency and decreased responses to the pre-training best frequency (BF) and other frequencies. These changes were often sufficient to produce a shift in frequency tuning such that the frequency of the CS became the new BF. The greater the behavioral relevance of a stimulus, the larger the number of cells tuned to that stimulus (Bakin et al., 1996, Bakin & Weinberger, 1990, Edeline et al., 1993, Edeline & Weinberger, 1993, Weinberger, 1993). These animal studies show that plastic changes in a particular neuron or set of neurons can play a critical role in the generation of a new “perceptual event".

[...]

---

¹ Everything is flowing