Nerve Cells and Animal Behaviour (2003)

Nerve cells and animal
behaviour,
SECOND EDITION
PETER J SIMMONS
and
DAVID YOUNG
CAMBRIDGE UNIVERSITY PRESS

Nerve Cells and Animal Behaviour
Second Edition
This new edition of Nerve Cells and Animal Behaviour has been
updated and expanded by Peter Simmons and David Young in
order to offer a comprehensive introduction to the field of
neuroethology while still maintaining the accessibility of the
book to university students. Two new chapters have been added,
broadening the scope of the book by describing changes in
behaviour and how networks of nerve cells control behaviour.
The book explains the way in which the nervous systems of
animals control behaviour without assuming that the reader has
any prior knowledge of neurophysiology. Using a carefully
selected series of behaviour patterns, students are taken from an
elementary-level introduction to a point at which sufficient detail
has been assimilated to allow a satisfying insight into current
research on how nervous systems control and generate
behaviour. Only examples for which it has been possible to
establish a clear link between the activity of particular nerve cells
and a pattern of behaviour have been used.
Important and possibly unfamiliar terminology is defined
directly or by context when it first appears and is printed in bold
type. At the end of each chapter, the authors have added a list of
suggestions for further reading, and specific topics are
highlighted in boxes within the text.
Nerve Cells and Animal Behaviour is essential reading for
undergraduate and graduate students of zoology, psychology and
physiology and serves as a clear introduction to the field of
neuroethology.
           is a Lecturer in the Department of

Neurobiology, University of Newcastle upon Tyne, UK, and
        is a Reader in the Department of Zoology,
University of Melbourne, Australia. Both authors regularly
publish their research in insect neuroethology.
This page intentionally left blank
Nerve cells and
animal behaviour
SECON D EDITION
P E T E R J S I M M O N S and D AV I D Y O U N G
PUBLISHED BY CAMBRIDGE UNIVERSITY PRESS (VIRTUAL PUBLISHING)
FOR AND ON BEHALF OF THE PRESS SYNDICATE OF THE UNIVERSITY OF
CAMBRIDGE
The Pitt Building, Trumpington Street, Cambridge CB2 IRP
40 West 20th Street, New York, NY 10011-4211, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
http://www.cambridge.org
© Cambridge University Press 1999

This edition © Cambridge University Press (Virtual Publishing) 2003
First published in printed format 1989

Second edition 1999
A catalogue record for the original printed book is available

from the British Library and from the Library of Congress
Original ISBN 0 521 62216 6 hardback
Original ISBN 0 521 62726 5 paperback
ISBN 0 511 01961 0 virtual (netLibrary Edition)

CONTENTS
Preface ix
1 Introduction 1
1.1 Nervous systems and the study of behaviour 1
1.2 Scope and limitations of neuroethology 2
1.3 Neural implications of ethological results 4
1.4 Sign stimuli in amphibians 9
1.5 Neuroethology of a releasing mechanism 11
1.6 Control theory and nervous systems 15
1.7 Conclusions 17
Further reading 18
2 Nerve cells 20
2.1 Basic organisation of nerve cells 20
2.2 Neuron physiology and action potentials 26
2.3 Synapses 30
2.4 Integration of postsynaptic potentials 33
2.5 Comparison of spikes and graded potentials 39
2.6 Additional mechanisms in integration 39
2.7 Conclusions 40
Further reading 41
3 Giant neurons and escape behaviour 42

3.1 Introduction 42
3.2 Giant neurons and the crayfish tail flip 44
3.3 The lateral giant interneuron: input and output 49
3.4 The decision to initiate startle behaviour 52
3.5 Executive functions of the lateral giant neuron 55
v
vi Contents
3.6 Summary of pathways in crayfish startle behaviour 59

3.7 Mauthner neurons and the teleost fast start 61
3.8 Excitation and inhibition in the Mauthner neuron 65
3.9 Outputs and executive functions of the Mauthner neuron 67
3.10 The startle reaction of a cockroach 69
3.11 Conclusions 74
Further reading 75
4 Capturing sensory information 76

4.1 Introduction 76
4.2 Basic receptor mechanisms: the campaniform organ 77
4.3 Summary of basic mechanisms 83
4.4 Essential properties of eyes 85
4.5 Design features of eyes 86
4.6 Photoreceptors and the receptor potential 91
4.7 Conclusions 96
Further reading 97
5 Stimulus filtering: vision and motion detection 99

5.1 Introduction 99
5.2 The insect visual system 100
5.3 Neuronal coding in the insect lamina 102
5.4 Optomotor neurons in flies 109
5.5 Figure-ground neurons of the lobula plate 114
5.6 A mechanism for directional selectivity 114
5.7 Summary of fly optomotor neurons 121
5.8 Collision warning neurons in the locust 122
5.9 Conclusions 127
Further reading 128
6 Hearing and hunting: sensory maps 129

6.1 Introduction 129
6.2 Prey localisation by hearing in owls 131
6.3 Auditory interneurons and sound localisation 135
6.4 Synthesising a neuronal map of auditory space 138
6.5 The echolocation sounds of bats 142
Contents vii
6.6 Interception of flying prey by bats 147

6.7 The auditory system and echolocation 148
6.8 Auditory specialisations for echo ranging 152
6.9 Auditory specialisations for Doppler shift analysis 158
6.10 Conclusions 162
Further reading 163
7 Programs for movement 165

7.2 Locusts and their flight 166
7.3 The flight engine 167
7.4 The flight program 169
7.5 Generation of the flight rhythm 171
7.6 Interneurons of the flight generator 172
7.7 Proprioceptors and the flight motor pattern 178
7.8 Steering and initiating flight 182
7.9 Overall view of locust flight 184
7.10 Triggering and maintaining escape swimming in Tritonia 186
7.11 Swimming by young Xenopus tadpoles 191
7.12 Circuit reconfiguration in the stomatogastric ganglion of the
lobster 194
Further reading 199
8 Circuits of nerve cells and behaviour 201

8.2 Neuronal activity during different behaviours in Aplysia 202
8.3 Optical monitoring of neuronal activity 204
8.4 Local bending reflexes in the leech 206
8.5 Modelling a network of neurons 209
8.6 Local reflex movements of a locust’s leg 212
8.7 Local spiking interneurons 213
8.8 Non-spiking interneurons 215
8.9 Organisation of neurons that control reflex movements 217
Further reading 220
viii Contents
9 Nerve cells and changes in behaviour 221

9.2 Growth and metamorphosis in insects 222
9.3 Ecdysis in a hawk moth 222
9.4 Remodelling neurons during metamorphosis 225
9.5 Associative learning and the proboscis extension reflex in 228
honey bees
9.6 Neuronal pathways and conditioning 230
9.7 The role of an identified neuron in conditioning 232
9.8 Bird song and its production 235
9.9 The development of song 239
9.10 Neural centres for hearing and singing 240
9.11 Development of song nuclei 242
Further reading 246
References 248
Index 262
P R E FA C E
Our aim in this book is to introduce university students to research on

nervous systems that is directly relevant to animal behaviour, and to do so
at a level that assumes no detailed knowledge of neurophysiology. Many
topics that fall within the scope of neurobiology are omitted or passed over
lightly, and attention is concentrated on particular examples that illustrate
clearly how the activity of nerve cells is linked with animal behaviour. Since
the first edition was published, many new books on neurobiology have
appeared, but most concentrate on the cellular and physiological aspects of
the nervous system. By reviewing some of the modern stories in neu-
roethology, we hope that this book will also be useful to postgraduate stu-
dents and others who wish to learn something of the way in which
behaviour is controlled.
Each major topic in Chapters 3–9 is dealt with as far as possible by
introducing a particular type of behaviour and then working towards a
description of how nerve cells control it. We have selected subjects from
studies in which the links between nerve cells and animal behaviour are
particularly clear. In doing this, we hope to illustrate the principles that
have been revealed in modern research in neuroethology. Inevitably,
there are many interesting stories that we have not been able to touch
upon.
Readers who are familiar with the first edition of the book will notice
several changes in content and arrangement. The final two chapters, on cir-
cuits of nerve cells and on plasticity in behaviour, are completely new. In
order to provide an early illustration of how activity in nerve cells can be
related to animal behaviour, we now describe work on prey detection by
toads in the first chapter, and the chapter on startle behaviour is placed
earlier in the book than it was in the previous edition. New material has
been added in several places, particularly in Chapters 3, 5 and 7. In order to
ix
x Preface
make room for this new material, we have had to omit the chapter on
intraspecific communication.
To help readers to come to grips with unfamiliar terms and concepts,
many of these are set in bold type the first time they appear in the book, as
well as in the index. Anyone who studies neurobiology will soon discover
that it has many side branches, linking one story to another or to other
branches of biology. We have included brief introductions to a few of these
by means of boxes in some chapters. These boxes do not have to be read as
part of the main text, but are meant to complement it by providing useful
and interesting, relevant information.
Suggestions for further reading are given at the end of each chapter, and
major references to points of detail are scattered through the text and listed
at the end of the book. The references in the figure legends also draw atten-
tion to relevant papers as well as indicating our grateful acknowledgement
of material from other authors that we have incorporated into the figures.
We would like to thank many colleagues who have given useful com-
ments on various aspects of the book, particularly Claire Rind and a
number of undergraduate and postgraduate students. We are also very
grateful to members of our families for their support during the preparation
of the book.
1 Introduction
1.1 Nervous systems and the study of behaviour
People in antiquity seem to have had no idea that the brain was in any way
connected with behaviour. Even that great practical biologist Aristotle was
mistaken in his ideas. He observed the rich vascular supply of the brain and
concluded that it was an organ for cooling the blood. The ancient Egyptians
were positively cavalier in their attitude: when the body of a monarch was
being prepared for mummification, the brain was extracted with a spoon
and thrown away. The brain was considered unnecessary for the future life,
but the entrails were carefully preserved in a jar and kept beside the
mummified body.
Modern opinion emphasises the paramount importance of the brain as
the source of an individual’s behaviour and personality. This trend has gone
so far that many a successful work of science fiction has been based on the
idea that the brain might be kept alive or transplanted, and that by this
means the essential personality of the original individual might be pre-
served after the rest of the body has been disposed of. This vast change in
prevailing opinion about the brain is, of course, due to the anatomical and
physiological research of the last 200 years, which has revealed the nature
and importance of the central nervous system.
Our present understanding of the way in which nervous systems control
animal behaviour owes much to a group of biologists working in the middle
of the twentieth century, who pioneered an experimental approach to ana-
lysing behaviour. The approach they adopted came to be known as ethol-
ogy, and one of early ethology’s most thoughtful exponents was Niko
Tinbergen. In an important paper, ‘On aims and methods of ethology’
(Tinbergen, 1963), he defined ethology simply as ‘the biological study of
behaviour’.
1
2 Introduction
Tinbergen himself made an impact on ethology by concentrating on field

observations or on elegantly simple experiments carried out on intact
animals. But he expected that the results of this work would be integrated
with a neural analysis as this became available. This is seen clearly in his
book synthesising ethology, entitled The Study of Instinct (1951), in which
he referred to contemporary research in neurophysiology and formulated
his concepts in terms of the nervous system as far as possible. He expected
that the biological methods of ethology would yield ‘concrete problems that
can be tackled both by the ethologist and the physiologist’, and he wrote of
‘the fundamental identity of the neurophysiological and the ethological
approach’.
The long-term goal of such an approach is to analyse patterns of behavi-
our in terms of the activity of the underlying neural components. Hence,
this field of research is sometimes given the title of neuroethology, a term
that first came into use in the 1960s. Neuroethology tries to combine the
approaches of both ethology and neurobiology so as to understand the
neural basis of behaviour. Often, this involves examining groups of recep-
tors or networks of nerve cells in order to elucidate the interactions relevant
to behaviour. In some cases it is possible to bring both neurobiological
and ethological analysis to bear on a single phenomenon, as Tinbergen
expected.
In the chapters that follow, selected examples are considered in which
neural analysis has been carried out in a way that is helpful to an under-
standing of animals’ natural behaviour. As far as possible, attention is con-
centrated on specific case histories in which a connection has been
established between a particular group of nerve cells (also termed neurons)
and a particular pattern of behaviour. This field of study is developing
rapidly and enough has been accomplished to enable initial conclusions to
be drawn about the operation of many basic areas. These studies and con-
clusions form an essential and fascinating part of ethology, the biological
study of behaviour.
1.2 Scope and limitations of neuroethology
As might be expected, neuroethology has been most successful in tackling

those elementary components of behaviour with which ethology itself
began. The simple kinds of behaviour that first caught the attention of the
Scope and limitations of neuroethology 3
founders of ethology are often also the kinds of behaviour most readily ana-
lysed in terms of the underlying neural events. One good example is
intraspecific communication, which requires both that the sender delivers
a clear signal and that the receiver has the appropriate sensory apparatus to
analyse it. The interactions between predator and prey have also been the
subject of many neuroethological studies because the neural mechanisms
involved must be simple in order to be swift. When life-or-death decisions
have to be made in a small fraction of a second, there is just not time for
elaborate neural circuits to operate.
A good many of the cases that have been analysed successfully involve
dedicated systems. A dedicated neural system is one that is largely devoted
to a single, important function such as escape (see Chapter 3). Dedicated
systems are easier to analyse than multipurpose systems, not merely
because they tend to be simpler, but more importantly because their
behavioural function is clearly known. As neural systems become more
flexible in the tasks that they can perform, it becomes more difficult for
experimenters to determine what is behaviourally important in their
neurophysiological recordings. In a multipurpose system, it is difficult to
discern which of several possible functions is pertinent to neural activity
recorded in a dissected animal. In a dedicated system, any recorded activ-
ity is likely to relate to the one and only behavioural function, provided the
system is in a healthy state.
For example, the large amount of neurophysiological work that has been
done on hearing in cats has been of little interest to ethologists because it is
so difficult to correlate particular properties of the auditory system with
particular episodes in the animal’s normal behaviour. It is almost impos-
sible to know what a cat is listening to at any given moment, simply because
its hearing is used for so many purposes. By contrast, in the study of hearing
in bats, we know precisely what the animals are listening to: they are listen-
ing to themselves. The auditory system of bats is largely dedicated to ana-
lysing the echoes of their own cries as part of the sonar system by which
they find their way around (see Chapter 6). Knowing this central fact, the
physiological properties of nerve cells in the auditory system are readily
correlated with their behavioural function in the intact animal.
Whether or not the system under study is a dedicated one, it obviously
makes the neuroethologist’s task easier if the absolute number of nerve
cells involved is small. Unfortunately, most of the higher vertebrates have
4 Introduction
very large numbers of nerve cells in even the smallest subsection of their
central nervous systems. By and large, therefore, neuroethologists have
looked to the lower vertebrates and the invertebrates for suitable study
material. Among the invertebrates, the arthropods show behaviour that is
complex enough to be interesting yet they also show a remarkable economy
in the number of nerve cells involved. Whereas mammals may employ hun-
dreds of nerve cells to excite a single muscle, for example, arthropods
usually make do with no more than half a dozen (see Chapter 7). It does not
necessarily follow that, because the number of nerve cells involved is
smaller, the neural principles of operation will be simpler. But working with
a smaller number of nerve cells does increase the chances of discovering
the principles in the first place.
1.3 Neural implications of ethological results
The behaviour of an animal is to a large extent the product of activity in its

nervous system. The patterns of behaviour that are recognised in ethologi-
cal studies must therefore reflect the underlying organisation of the
nervous system. In the case of the elementary components of behaviour
studied by the early ethologists, this correspondence may be fairly close.
Consequently, a careful study of behaviour patterns at the level of the intact
organism will often produce results that provide valuable clues about the
underlying neural organisation.
Consider the classic case of the egg-retrieval behaviour found in many
ground-nesting birds, which was first studied in the greylag goose (genus
Anser) by Lorenz and Tinbergen in the 1930s. A nesting goose employs a
stereotyped sequence of movements to retrieve an egg that has become dis-
placed from the nest. The bird leans out of the nest, places its beak beyond
the egg, and then draws the beak back towards its chest so that the egg is
rolled back into the nest. Superimposed on this movement towards the chest
are little side-to-side movements of the beak, which serve to keep the egg in
place. This sequence of movements is used by all members of the species for
egg retrieval; none uses an alternative method. Indeed, a very similar
pattern of movement is found in other birds, such as the herring gull (Larus),
on which many tests have been carried out (Fig. 1.1). Stereotyped move-
ments of this kind were originally called fixed action patterns; nowadays,
more general terms like motor pattern are used instead by most ethologists.
Neural implications of ethological results 5
Figure 1.1 Egg retrieval in the herring gull (Larus): an incubating gull will
retrieve an egg that has become displaced from the nest, using a stereo-
typed pattern of movement. Here, the retrieval response is being used to
test what the gull perceives to be an egg. Two different models, both of
which differ considerably from the real egg in the nest, are placed on the
rim of the nest to compare their effectiveness in eliciting the retrieval
response. (Redrawn after Baerends & Drent, 1982.)
It was noticed that many such motor patterns seem to occur in response
to specific stimulus situations in the natural environment. During the
1930s, ethologists developed the technique of using models, in which one
feature at a time could easily be varied, to find out what features of a situa-
tion are important in triggering an animal’s response. Lorenz and
Tinbergen found that the greylag geese would retrieve wooden models
painted to resemble natural eggs. The goose would still retrieve the models
when they were made the wrong shape, such as cubes or cylinders, or when
they were made the right shape but the wrong size, including models that
were much larger than a normal egg. It was evident from these results, and
many others, that only certain features of the natural stimulus are needed
to produce a response. These essential features were called sign stimuli or,
where they were found in the context of social behaviour, social releasers.
Ethologists rightly sought to account for the fact that animals often
respond to only a small selection of the available stimuli by postulating
neural mechanisms in the responding animal. Response selectivity might
6 Introduction
Figure 1.2 A flow diagram showing early ethological concepts of the mecha-
nisms involved in a simple behaviour pattern such as egg retrieval.
(Redrawn after Shepherd, 1983.)
be due partly to the capacities of the sense organs, but it was already known
that an animal may respond to a specific sensory cue in one behavioural
context and not in another. Hence, the occurrence of sign stimuli must also
be due to stimulus selection by more centrally located mechanisms pro-
cessing the information received from the sense organs. The term releasing
mechanism was coined for this central processing and, because it was
assumed to develop independently of experience with the sign stimuli, the
adjective innate was attached to it, giving innate releasing mechanism
(IRM). The adjective innate is not much used by modern ethologists, but
the term releasing mechanism continues to call attention to an important
phenomenon of behaviour.
The way in which the various components might interact to produce a
behaviour pattern is illustrated in Fig. 1.2, which represents the results of
the early ethological period. In egg retrieval, the visual stimuli from around
the nest are passed from the sense organs along a neural pathway to the
central nervous system, where the releasing mechanism responds to the
sign stimuli that indicate ‘egg’. This central mechanism then releases or trig-
gers activity in the motor regions of the nervous system that generate the
fixed action pattern for retrieval. This sequence is not invariable in its oper-
ation but is enhanced or prevented by other factors. Thus, the releasing
mechanism is inhibited in the short term (arrows from above in Fig. 1.2)
when the bird is away from the nest foraging or escaping from a predator,
and in the long term (arrows from below) retrieval cannot be elicited
Neural implications of ethological results 7
outside the breeding season, which is controlled by reproductive hor-

mones.
Further insight into this phenomenon has been made possible by the
detailed studies of egg retrieval in the herring gull carried out by Baerends
and his colleagues (Baerends & Drent, 1982; Baerends, 1985), who placed
two egg models side by side on the rim of the nest and then watched from
a hide to see which of the models the gull retrieved first. Thousands of these
tests were made, carefully varying only one feature at a time, in order to
determine what the gulls’ preferences were. It was found that the gulls pre-
ferred larger eggs to smaller ones, green eggs to any other colour, speckled
eggs to uniformly coloured ones, strongly contrasting speckles to weakly
contrasting ones, and natural egg shapes to abnormal ones. This last pref-
erence was not nearly as strong as might have been expected, and a cylin-
drical model was almost as effective as an egg-shaped model of the same
size and colour.
These results show that the gulls do, indeed, respond selectively to a
limited number of stimuli, which match a real gull’s egg only in a rough
way. It is not even necessary for all the stimuli to be present for a response
to occur. The stimuli that are present add together independently to deter-
mine the overall effectiveness of an egg model in producing a response. For
instance, a smaller green egg will be as effective as a larger brown egg; if
speckling is then added to the green egg, it will become more effective than
the larger brown egg. One consequence of this property is that models can
be made more effective than the real object they represent. A gull will
retrieve a model 50 per cent larger than normal, green and with black
speckling in preference to one of its own eggs; such a model is what ethol-
ogists call a supernormal stimulus.
The experiments with models show that this releasing mechanism
involves perception of a number of simple visual cues, which add together
quantitatively to determine the degree of ‘egginess’ as far as the gull is con-
cerned. Clearly, these properties reflect the way in which visual perception
occurs in the gull’s nervous system, and the flow diagram shown in Fig. 1.3
tries to incorporate this. The response to a limited number of simple cues
may well reflect the occurrence in the early stages of the visual system of
units that respond selectively to visual cues such as colour, contrast, edges
and shapes (represented as selectors, S1to S9 , in Fig. 1.3). The way in which
the separate cues add together suggests the presence of a more central unit
8 Introduction
Figure 1.3 Releasing mechanism for egg retrieval in the herring gull: a flow
diagram based on experiments with egg models. The boxes represent major
systems or operations and the circles indicate sites where summation of
inputs occurs. Visual perception (top) is represented as a series of selectors
(S1 to S9) that respond to particular features of the stimulus. Some of these
feed on to a specific detector for egg recognition, which in turn feeds on to
the motor control for egg retrieval. This response is maintained during the
period of incubation but may be overridden by other factors such as the
need to escape (left) or the bird’s memory based on experience with real
eggs (right). (Redrawn after Baerends, 1985.)
that combines information from a specific set of selectors so as to act as a

detector for specific objects in the environment, in this case an ‘egg detec-
tor’. Units that correspond closely with this description are found widely in
the visual systems of both vertebrates and invertebrates, as shown in the
following example of prey detection in toads (see also Chapter 5). It is easy
to see how these units could be excited more strongly by a supernormal
combination of stimuli than by the natural combination.
Sign stimuli in amphibians 9
1.4 Sign stimuli in amphibians
The way in which frogs and toads recognise their prey provides another
example of a releasing mechanism. In this case, the ethological results are
even more compelling because they have been combined with a neuro-
physiological study of the same system. This combined approach clearly
shows how the selective properties of nerve cells (neurons) are involved in
releasing particular patterns of behaviour (Ewert, 1985, 1987).
In the visual world of a frog or toad, just a few, simple criteria serve to
categorise moving objects as prey, enemy or lover. Once the visual system
has placed a given object in one of these categories, the animal reacts
accordingly. These reactions can be used to analyse the criteria involved in
prey recognition because the animals are readily deceived by small card-
board models moving in front of them. A special study of prey detection has
been made in the common European toad (genus Bufo), using such models
to analyse the behavioural responses of the intact animal and the responses
of specific classes of neuron in the visual system. The natural prey of Bufo
consists of small animals such as beetles, earthworms and millipedes. If
one of these animals appears in its peripheral visual field, the toad
responds by turning its head and/or body so as to bring the animal into the
frontal visual field. The toad then walks towards the prey in order to capture
it.
The sign stimuli, by which the prey is recognised, can be analysed quan-
titatively in the laboratory. A hungry toad is confined in a glass vessel, from
which it can see a cardboard model circling around (Fig. 1.4 a). If the toad
interprets the model as a prey animal, it tries to bring it into the frontal
visual field, and in doing so turns around jerkily after the moving model.
The number of orientating turns per minute elicited by a given model,
compared to the number elicited by others, can therefore be taken as a
measure of the resemblance between that model and prey, from the toad’s
point of view.
In this experimental situation, the toad is not much impressed by a small
2.5 ⫻ 2.5 mm model, which elicits only a few orientating movements.
However, the stepwise elongation of this shape in the horizontal dimension
(Fig. 1.4b, shape x) greatly increases its releasing value. That is to say, elon-
gation of the model in the direction of movement increases its resemblance
to prey, up to a certain limit, and this long, small stripe has been called the
10 Introduction
Figure 1.4 Analysis of prey recognition in the toad (Bufo). (a) The experi-
mental set-up, with the toad confined in a glass vessel and a prey model (P)
circling around it. The toad turns to follow the model when it has moved
through a sufficient angle, the effective displacement (D). (b) The response
of the toad to moving models of three shapes (x, y, z) as these are enlarged
in one dimension (shapes x, y) or two dimensions (shape z). The toad’s
response is measured by the number of times it turns to follow the model in
1 min. (Redrawn after Ewert, 1980, 1983.)
worm configuration. If the small, square shape is elongated in the vertical

dimension (Fig. 1.4b, shape y), its releasing value decreases to zero. In fact,
the toad often interprets it as a threat and freezes in a defensive posture.
This shape has been called the antiworm configuration. If both dimensions
of the model are lengthened equally, so that the toad is presented with
squares of increasing size (Fig. 1.4b, shape z), the prey-catching activity ini-
tially increases but then declines rapidly to zero. This is probably the result
of non-linear summation of the horizontal (worm) and vertical (antiworm)
edges.
The toad’s ability to distinguish between worm and antiworm does not
vary with other stimulus parameters, such as the colour of the model or its
velocity of movement. It is also independent of the direction in which the
Neuroethology of a releasing mechanism 11
stimulus traverses the toad’s visual field. If the models are moved past the
toad in a vertical direction, then the vertical stripe elicits prey catching and
the horizontal stripe elicits no response or a defensive posture. Thus, the
worm/antiworm distinction is based on the combination of just two stim-
ulus parameters: the elongation of the object in relation to its direction of
movement. These parameters, then, are the sign stimuli that release prey-
catching behaviour in a hungry toad, and it is obvious that they correspond
only very approximately to a real worm. Nevertheless, they will normally
enable a toad to distinguish correctly between potential prey and inedible
objects in its natural environment.
1.5 Neuroethology of a releasing mechanism
As with other vertebrates, early visual processing in amphibians takes place

in the neuronal circuits of the retina. The neurons of the vertebrate retina
(see Fig. 2.4, p. 25, and Box 5.1, p. 107) are arranged in a way that provides
for both lateral interaction and through transmission. The through route of
the visual pathway is made up of receptors, bipolar cells and ganglion cells.
Recording with microelectrodes shows that the receptors respond in a
simple way to changes in the intensity of light that falls on them as part of
the image formed by the eye. These responses are mirrored by the bipolar
cells, each of which receives input from several receptors. In turn, the gan-
glion cells each pool input received from a large number of bipolar cells.
The ganglion cells are able to respond to more complex features of the
image because of the way information from the receptors is processed and
combined on its way to the ganglion cells. Ganglion cells can be divided
into several different classes on the basis of the type of feature to which
each is most sensitive. Some are most sensitive to objects of a given angular
size, others to objects moving in a particular direction or to the difference
in brightness between adjacent areas of the image. This information is
passed along the optic nerve to the brain (Fig. 1.5), where these basic
parameters are used to distinguish between mate, prey and predator.
The majority of retinal ganglion cells are connected to the optic tectum,
a specialised region of the midbrain visible as a large bulge on either side
(Fig. 1.5). A smaller number of ganglion cells are connected to the thala-
mus, which is the most prominent part of the posterior forebrain, and to
the pretectal areas of the midbrain. These connections are spread out in an
12 Introduction
Figure 1.5 The layout of the main visual pathways concerned in prey detec-
tion in the brain of an anuran amphibian. The axons of most ganglion cells
travel from the eye to the optic tectum on the opposite (contralateral) side
of the brain, via the optic nerve (other cranial nerves are not shown).
Feature-detecting neurons of the optic tectum send their axons to the
motor regions of the contralateral hindbrain.
orderly manner in the superficial layer of the optic tectum, with each gan-
glion cell keeping the same relative position with respect to its neighbours
that it has in the retina.
The responses of the tectal neurons can be recorded by probing the
deeper layers of the optic tectum with a microelectrode (Fig. 1.6). Like the
ganglion cells, the neurons of the thalamus and tectum can be divided into
different classes according to their patterns of response. Of the thalamic
and tectal neurons that have been investigated, at least three classes
show differing responses to moving stimuli of worm and antiworm
configurations. The thalamic Class TH3 neurons respond best to squares;
stripes with the antiworm configuration elicit a lesser response, and the
worm configuration elicits the least response of all (Fig. 1.6b). In the optic
tectum, the Class T5(1) neurons also respond best to squares, but when
tested with stripes, they prefer the worm to the antiworm configuration
(Fig. 1.6c). Another class of tested cells, the Class T5(2) neurons, distinguish
much more clearly between the worm and antiworm configurations, with
the worm configuration eliciting the greatest response, the squares a lesser
response, and the antiworm by far the least response (Fig. 1.6d). Among all
the neurons tested so far, the response pattern of the T5(2) neurons shows
Neuroethology of a releasing mechanism 13
Figure 1.6 (a) Set-up for recording the responses of neurons in the brain to
moving visual stimuli. The toad is held in a fixed position and its brain is
probed with a microelectrode for recording the spikes in single neurons.
Each stimulus is moved in front of the toad by means of the perimeter
device. (b) The response of thalamic Class TH3 neurons to increasing
angular size of the same three shapes (x, y, z) used in the behavioural tests
(Fig. 1.4). (c) The response of tectal Class T5(1) neurons to the same three
shapes. (d) The response of tectal Class T5(2) neurons to the same shapes.
(a redrawn after Ewert, 1985; b–d redrawn after Ewert, 1980.)
the best correlation with the sign stimuli for prey-catching behaviour (cf.
Fig 1.6d and Fig. 1.4b).
It is evident from Fig. 1.6 that the responses of the Class T5(2) neurons
could be accounted for if they receive excitatory input from Class T5(1)
neurons and inhibitory input from Class TH3 neurons. The fairly strong
14 Introduction
response to the worm configuration in Class T5(1) neurons would be mini-

mally inhibited by the poor response to it in the Class TH3 neurons, result-
ing in a strong response in the Class T5(2) neurons. Similarly, the poor
response to the antiworm configuration in Class T5(1) would interact with
the moderate response in Class TH3 to give a very poor response in Class
T5(2).
This possibility has been tested by removing the input from the Class TH3
neurons, which can be accomplished by severing the pathway that is
known to run from the thalamus to the optic tectum. Whether this lesion is
done permanently by microsurgery or temporarily by local application of a
neurotoxin, the effect on Class T5(2) neurons is dramatic. The responsive-
ness of these neurons to all visual stimuli is increased and selectivity is lost,
with the neurons responding best to squares and failing to distinguish
clearly between stripes in worm and antiworm configurations. This shows
that the normal selective response of the Class T5(2) neurons is dependent
on inhibition from thalamic neurons, including the Class TH3 neurons.
When a toad with a pretectal lesion is allowed to recover from surgery
and is tested behaviourally, its responses closely parallel those of the T5(2)
neurons: the operated animal responds vigorously to all shapes, preferring
squares and failing to distinguish clearly between worm and antiworm
configurations of stripes. Such a close correspondence between the
responses of the Class T5(2) neurons and of the whole animal suggests that
these neurons are directly involved in prey detection and hence in releasing
prey-catching activity. This is confirmed by means of a small telemetry
system that enables the experimenter to record from and stimulate single
neurons in the optic tectum of a freely moving toad. Recordings made with
this system show that activity of Class T5(2) neurons precedes and contin-
ues during the orientation of the toad towards the prey. Having recorded in
detail from a particular T5(2) neuron, it can then be stimulated by passing
a tiny current through the microelectrode, and this consistently elicits
orientating movements that are directed to the appropriate part of the
visual field.
If they are involved in prey detection in this way, one would expect the
Class T5(2) neurons to be connected, directly or indirectly, with the motor
circuits in the hindbrain of the toad. Various histological methods demon-
strate that a number of the connections arriving in the motor regions on
one side of the hindbrain do come from the contralateral optic tectum (see
Control theory and nervous systems 15
Fig. 1.5). That these include the Class T5(2) neurons is confirmed by physio-
logical methods. Localised stimulation of the appropriate neural tract in
the hindbrain sends signals travelling back up to the optic tectum, where
they can be recorded in individual T5(2) neurons with a microelectrode.
Thus, the Class T5(2) neurons provide an excellent example of specific
brain cells that are involved in releasing a simple, important behaviour
pattern. On the basis of the stimulus parameters selected by the retinal gan-
glion cells, the neurons of the optic tectum are able to respond to specific
parameter combinations that carry information relevant to the toad’s way
of life. The specific combination of visual parameters to which the T5(2)
neurons respond carries information enabling the toad to distinguish
between its natural prey and inedible objects. These response properties of
the T5(2) neurons do not identify worms or beetles, but rather they under-
lie the sign stimuli that elicit prey-catching behaviour in a hungry toad.
1.6 Control theory and nervous systems
In the attempt to pursue analysis of the mechanisms of behaviour to the

neural level, most of the concepts used are drawn either from general
neurobiology or from ethology itself. An additional strand of thought that
has made a useful contribution is cybernetics or control systems theory,
developed to provide a formal analysis of human control systems. It is often
useful to treat the nervous system of a behaving animal, or some part of its
nervous system, as a control system through which there is an orderly flow
of information, with definite input and output elements.
This approach has provided a helpful terminology and way of thinking
about the mechanisms of behaviour. With the advent of powerful comput-
ers, it has also become possible to use this approach for constructing
models that mimic the interactions among groups of neurons. When used
carefully, this provides a means of testing whether circuits work in the way
that is predicted from physiological recordings. This is particularly useful
when large numbers of neurons are involved because it is rarely possible to
monitor activity in many neurons simultaneously.
An example of a useful concept from control theory is that of feedback;
this takes place when an event or process has consequences which affect
the occurrence of that event or process. In control systems, feedback
is usually negative, which means that action is taken in response to a
16 Introduction
Figure 1.7 Orientation to prey in a mantis (Tenodera), illustrating move-

ments with and without feedback. (a) A flow diagram of visual tracking of
prey with feedback. (b) A flow diagram of rapid visual location of prey
without feedback. In (a) and (b) the boxes represent major systems or oper-
ations and the circles indicate sites where summation of inputs occurs. (c)
Rapid location of prey, as in (b), analysed from videotape, showing orienta-
tion towards a stationary target (left) and towards a target that moves off at
a constant angular velocity after the mantis has started to turn (right). (c
redrawn after Rossel, 1980.)
disturbance so as to correct that disturbance. This usually involves a special

mechanism like a thermostat, by means of which the output of a system is
fed back to regulate the input. Many behaviour patterns also have this self-
regulatory character. In egg retrieval, the movement of the beak towards the
chest seems not to involve feedback because it is so stereotyped and
because it continues to completion even if the egg is removed during the
movement. However, the side-to-side movements that keep the egg centred
on the beak certainly appear to involve feedback of some kind; they tend to
disappear if the bird is retrieving a cylindrical model, which rolls smoothly
and so does not need centring.
The distinction between movements that do and do not involve feedback
Conclusions 17
is made clear by the visual orientation to prey in a praying mantis. The

insect follows potential prey with movements of its head or body so as to
keep the prey in the centre of its line of vision. In this behaviour pattern,
visual information triggers a movement of the praying mantis, and this
results in an altered visual input, which in turn influences the subsequent
movement. Hence, the flow of information forms a closed loop, with output
feeding back to the input (Fig. 1.7a). However, a different situation obtains
when the mantis first locates the prey. As soon as a suitable object appears
in the visual field, the mantis turns towards it with a rapid movement that
is not influenced by feedback from the visual system. Even if the object is
experimentally removed during the turn, the mantis still continues turning
until it faces the place where the object originally was. Hence, in this case,
the information flow forms an open loop, without feedback (Fig. 1.7b, c).
In common with ethological concepts considered above, such concepts
from control theory do not in themselves provide an explanation in terms
of underlying mechanisms. Rather, they are, in computer terminology,
‘software’ explanations that specify the job done and the relations between
the different components of a behaviour pattern. For a full understanding
of the mechanisms of behaviour, these concepts obviously need to be
coupled with a detailed analysis of the underlying neural ‘hardware’. Once
a neural analysis is accomplished, or at least is underway, then the ‘soft-
ware’ concepts come into their own as a vehicle for showing how the neural
‘hardware’ is organised so as to generate a given behaviour pattern. To take
a simple example, flow diagrams such as that in Fig. 1.3 have been bor-
rowed from control theory as a vehicle for summarising behavioural mech-
anisms. If such a diagram is based on known neural components, whose
physiological properties have actually been studied, rather than on hypo-
thetical ‘black boxes’, then the diagram becomes a truly effective way of
summarising the link between the nervous system and behaviour (see, for
example, Fig. 3.7, p. 60).
1.7 Conclusions
The biological study of animal behaviour was put on a sound footing by

the early ethologists in the middle of the twentieth century. These etholo-
gists pioneered techniques for investigating the natural behaviour of
animals, including the use of simple experiments in conjunction with field
18 Introduction
observations. In the course of this work, they developed a number of key

concepts that have helped to guide efforts to understand the mechanisms
of behaviour.
One such concept is that of the motor pattern, which is a relatively
stereotyped sequence of movements and is easily recognised as part of an
animal’s ongoing behaviour. The recognition of these motor patterns in the
animal’s natural behaviour clearly implies that there is a corresponding
pattern in the underlying organisation of the animal’s nervous system,
which generates these movements. Another concept is that of the releasing
mechanism, which may be envisaged as a kind of neural filter tuned to rec-
ognise specific sign stimuli in the environment. The nervous system must
be organised so as to sort out different stimuli and to make decisions about
which motor pattern to put into action at any one time. The egg retrieval
response in nesting birds provides a good example of the development of
these two concepts.
The way that particular nerve cells can be identified as playing specific
roles in the control of behaviour is well illustrated by the T5(2) neurons in
the brains of toads. The close correspondence between the responses of
these neurons and the sign stimuli for prey catching in the toad (Bufo) sug-
gests that these cells are part of the natural releasing mechanism for prey-
catching behaviour. This conclusion is backed up by lesion experiments,
which alter the responses of these nerve cells and the behaviour of the
intact animal in similar ways.
Much recent research in neuroethology is aimed at understanding how
nerve cells are organised into circuits that perform specific functions in
behaviour, such as filtering out sign stimuli or generating a particular
motor pattern. This work involves tracing the flow of signals from one nerve
cell to the next, and is most easily done where particular nerve cells can be
uniquely identified. An essential first step along this path to understanding
how nervous systems underlie behaviour is to examine the relevant proper-
ties of the fundamental units of the nervous system, the nerve cells.
Further reading
Alcock, J. (1998). Animal Behavior, an Evolutionary Approach, 6th edn.
Sunderland, MA: Sinauer Associates. A useful textbook, which includes some
neuroethology, setting it nicely in the context of the varied questions that can
be asked about the mechanisms and evolution of behaviour.
Further reading 19
Manning, A. & Dawkins, M.S. (1998). An Introduction to Animal Behaviour, 5th

edn. Cambridge: Cambridge University Press.
Another good textbook, which gives neuroethology its place among the diver-
sity of subject matter and levels of analysis included within the study of beha-
viour.
Ewert, J.-P. (1980). Neuroethology. Berlin: Springer-Verlag. One of the first text-
books treating neuroethology as a distinct area of study, it remains useful for its
expert treatment of prey recognition in toads.
Ewert, J.-P. (1985). Concepts in vertebrate neuroethology. Anim Behav 33, 1–29.
This brief essay clearly explains the ideas about neural filtering that have
emerged from research on prey recognition in toads, using both ethological
and neurophysiological methods.
The International Society for Neuroethology provides a Web site with many links
to interesting aspects of this branch of biology: http://www.neuro-
bio.arizona.edu/isn/
2 Nerve cells
2.1 Basic organisation of nerve cells
The very title of this chapter would have been contentious in the nine-
teenth century, when detailed scientific study of the nervous system got
underway. For it was then not generally agreed that the nervous system is
composed of many individual nerve cells. This was mainly due to the fact
that nerve cells are difficult to visualise with routine histological methods.
Many cells are packed tightly together in nervous tissue (there are 100 000
nerve cells in 1 mm3 of human brain) and they give off fine, branched pro-
cesses, so that it is almost impossible to determine the limits of a single cell.
Many scientists therefore believed that nerve cells were fused together in a
continuous network of branched processes, rather like the capillary beds
that link small arteries and veins.
The technique that was most powerful in challenging this view was silver
staining, first discovered by Camillo Golgi in 1873 and developed by others,
particularly Santiago Ramon y Cajal from 1888 onwards. Ramon y Cajal
examined many parts of the nervous system in a wide range of animal
species. He realised that the special feature of silver staining is that it only
stains a small percentage of cells in a piece of tissue but it stains them in
their entirety, so that the structure of an individual nerve cell can be
described (Fig. 2.1). Nowadays, single neurons are often stained by the
intracellular injection of dye through a microelectrode, or by the use of
methods that recognise chemicals characteristically found in particular
neurons. In 1891, an anatomist named Wilhelm Waldeyer reviewed all the
evidence from microscopical and developmental studies and concluded
that the nervous system, like other organ systems, is indeed composed of
discrete cells. He suggested the term neuron for the nerve cell.
Final confirmation that the nervous system is composed of discrete cells
20
Basic organisation of nerve cells 21
Figure 2.1 Drawing of a pyramidal cell from a mammalian brain, an

example of a complex nerve cell. This is a single cell, stained by the Golgi
silver method. At the bottom of the figure, the axon is shown joining a
bundle of axons from similar cells. The cell body is about 20 ␮m (0.02 mm)
across. (From Ramon y Cajal, 1911.)
22 Nerve cells
Figure 2.2 Diagrammatic comparison of nerve cell structure in (a) verte-

brates and (b) arthropods, illustrated by motor neurons.
was provided in the 1950s by studies with the electron microscope, which
revealed that a gap separates the plasma membrane of adjacent nerve cells.
These studies also showed that a nerve cell contains an assemblage of intra-
cellular organelles, such as endoplasmic reticulum and mitochondria, just
like any other animal cell. These organelles are required to carry out the
usual metabolic functions of an animal cell. At the same time, nerve cells
are specialised in their anatomy and physiology to carry out the particular
function of processing information.
The specialised anatomy of nerve cells can be seen in Fig. 2.2, which
diagrammatically compares a vertebrate neuron with that of an arthropod.
As in other cells, the main intracellular organelles are gathered around the
nucleus, and the region of the neuron that includes the nucleus is called the
Figure 2.3 General arrangement of the nervous system and three basic cate-
gories of neuron, illustrated by schematic diagrams based on an arthropod.
(a) Two segmental ganglia, showing lateral and connective nerves. (b) One
ganglion with one sensory neuron, one motor neuron and an interneuron.
cell body. Neurons involved in long-distance signalling have a long, cable-

like process, the axon, which is devoid of most intracellular organelles.
Other processes are shorter and often branch repeatedly; these are usually
called dendrites. In most vertebrate neurons, the dendrites and the axon
emerge from the cell body, but in many invertebrate neurons the cell body
is separated from the branching region and lies at the end of a thin process,
or neurite. In many neurons concerned with local signalling the axon is
short, and may even be absent altogether. Neurons with no axon are called
amacrine cells and all their processes are dendrites.
The precise arrangement of axon and dendrites varies greatly among
neurons, and different types of neurons can be identified from their partic-
ular branching pattern. Sometimes the pattern can be very complex, as in
the case of the pyramidal cell in Fig. 2.1. Here, the many dendrites arise not
only from the cell body but also from a stout dendritic trunk. The axon is
relatively thin, and it gives off axonal branches called collaterals fairly close
to the cell body.
In terms of function, neurons can be grouped into three basic categories
(Fig. 2.3b). Neurons with specialised endings that respond to energy from
the environment are called sensory neurons. Neurons that have axons ter-
minating on muscle fibres are called motor neurons. All other neurons are
24 Nerve cells
interneurons. Some interneurons connect relatively distant parts of the

nervous system and have long axons, but many are amacrine cells or have
short axons, and these are often referred to as local interneurons. As well as
neurons, the nervous system contains cells that surround and support the
neurons, and these are called glial cells (glia ⫽ ‘glue’).
Neurons tend to be clustered together in structures called ganglia (Fig.
2.3). In clearly segmented animals, such as arthropods or annelid worms,
there is one ganglion for each body segment, although in many species the
ganglia of adjacent segments are fused together. A segmental ganglion con-
tains the motor neurons that control the muscles of its segment (Fig. 2.3a),
and axons from the sense organs of the segment run into it along lateral
nerves. Ganglia of adjacent segments are linked by relatively stout nerves
called connectives, which consist of bundles of axons that almost all belong
to interneurons. Ganglia occur in many non-segmented animals, such as
molluscs, and some aggregations of neurons in the vertebrate central
nervous system are also called ganglia. In most invertebrate ganglia, the cell
bodies of motor neurons and interneurons occur in a thin rind on the
periphery of the ganglion, and the core that contains axons and dendrites
is called the neuropile.
Although neurons are separate entities, they are connected together in
specific ways to form circuits, in which signals travel from cell to cell along
particular pathways. In the simplest pathways, signals pass from sensory
neurons straight to motor neurons, but most pathways involve several
interneurons. During the operation of a nervous system, signals pass
sequentially from place to place in individual neurons, and from neuron to
neuron in a circuit. As a general rule, the dendrites of a neuron are where it
receives signals, and the axon terminals are where the neuron transmits
signals onwards to other neurons or to muscle cells. A neuron adds together
signals that it receives from other neurons and usually converts them into
a new signal that it may pass on. The process of combining inputs together
to produce a new output signal is called integration.
Ramon y Cajal deduced from his anatomical studies that neurons are
connected into functional circuits and clearly understood that signals flow
from cell to cell. One part of the nervous system he studied intensively was
the vertebrate retina (Fig. 2.4). He showed that the very short axons of
sensory neurons (rods and cones) make contact with the dendrites of
bipolar cells, which in turn have axons that contact the dendrites of retinal
Figure 2.4 The pattern of connections in the primate retina, as revealed by

Golgi silver staining. The diagram includes sensory neurons (rods and
cones, a, b, B, A), bipolar cells (c, d, C) and the retinal ganglion cells (e, E,
D). F indicates the fovea, where bipolar and retinal ganglion cells are dis-
placed from their receptors but the normal pattern of connections is main-
tained. Some amacrine cells (f) are included on the left, but horizontal cells,
at the sensory neuron–bipolar cell junctions, are omitted. (From Ramon y
Cajal, 1911.)
ganglion cells. The retinal ganglion cells have long axons with fine,
branched terminals in the optic region of the brain. Thus, information
about vision must flow through this sequence of connections from the
receptors through bipolar and retinal ganglion cells to visual centres in
the brain. In such a sequence, neurons are often referred to by their order
in the chain, so sensory neurons are first order, bipolar cells second order,
retinal ganglion cells are third order and cells in the brain are higher order.
Signals do not just flow sequentially from cells of one order to the next,
however, but also flow laterally. Ramon y Cajal indicated this on the left of
his diagram (Fig. 2.4) by showing retinal amacrine cells that contact the
dendrites of retinal ganglion cells and axon terminals of bipolar cells. The
amacrine cells serve to modify signals as they pass from bipolar cells to
retinal ganglion cells, and may allow activity in one bipolar cell to modify
the information that its neighbouring bipolar cells transmit to retinal gan-
glion cells. In addition, there is another layer of horizontally oriented cells
26 Nerve cells
in the retina, the horizontal cells, that lies at the level of the junctions
between sensory neurons and bipolar cells.
2.2 Neuron physiology and action potentials
Neurons are specialised in their physiology to receive, sort out and pass on
information. The signals that neurons deal with involve small changes in
the electrical voltage between the inside and outside of the cell, and inte-
gration is the process by which these voltage changes are combined
together to determine the neuron’s output signal. This is essentially how
neurons make decisions.
The most conspicuous of these voltage changes are action potentials,
which are the signals that neurons use to transmit information along long
axons. Each action potential lasts slightly less than a millisecond (1 ms) at a
particular location along the axon and travels along the axon at a speed that
varies from less than 1 m/s to nearly 100 m/s depending on the girth of the
axon. Action potentials may be recorded from outside an axon by using fine
wires as electrodes, as shown in Fig. 2.5a. Although the voltage change
between the inside and outside of the neuron is about a tenth of a volt, the
signal that the electrodes outside the cell pick up is much smaller, so con-
siderable amplification is needed to display and measure the action poten-
tial. Each action potential is conducted rapidly along the axon and passes
the two electrodes in succession. As it passes the first electrode, the latter
will become positive with respect to the second electrode, and as the action
potential passes the second electrode, the situation is rapidly reversed.
Consequently, the output signal that the amplifier delivers has an S-shaped
waveform when displayed on the oscilloscope (Fig. 2.5b). On a slower time
scale, action potentials appear as stick-like departures from the baseline,
and because of this appearance they are commonly called spikes.
A more precise method of recording a spike at one location along an axon
is to use an intracellular electrode. This consists of a glass capillary tube
that is drawn out to a very fine point and filled with a conducting salt solu-
tion such as potassium acetate. The electrode is connected to an amplifier
by way of a silver wire placed into the electrode (Fig. 2.6a). When the tip of
the electrode is inserted through the membrane of the cell, the salt solution
is in electrical contact with the inside of the cell, and the signal recorded
measures the difference in electrical potential, or voltage, between the
Neuron physiology and action potentials 27
Figure 2.5 Activity in the axon of a neuron, recorded with extracellular elec-
trodes. (a) The recording method shown schematically. (b) The typical
appearance of action potentials when recorded by this method and dis-
played on an oscilloscope. The oscilloscope is a sensitive voltmeter, and dis-
plays changes in voltage recorded with time.
inside and the outside of the axon. The voltage recorded by an intracellular
electrode is called the membrane potential. When the tip of the electrode
first enters the cytoplasm, it usually records a voltage of 60–80 mV across
the membrane, with the inside of the cell negative with respect to the
outside. This is called the resting potential.
The cell membrane of a resting cell is thus polarised: there is a standing
electrical voltage across the cell membrane. A reduction in electrical poten-
tial across the membrane, bringing the intracellular voltage closer to the
extracellular, is called a depolarisation. Most depolarising events are excit-
atory because they increase the likelihood that the neuron will generate an
action potential. An increase in the membrane potential from the resting
value is called a hyperpolarisation, and the effect of this is inhibitory as it
counteracts any depolarisation. Much of the process of integration involves
28 Nerve cells
Figure 2.6 Activity in a neuron recorded with an intracellular electrode. In

this method (a), one electrode is used to inject current pulses of different
magnitudes into the neuron and a second is used to record the voltage, or
potential, across the membrane. In (b), the typical appearance on an oscil-
loscope of responses by a neuron to pulses of current is shown.
an interplay between depolarising and hyperpolarising membrane poten-

tials.
A second intracellular electrode may be used to stimulate the neuron
with pulses of electrical current, as shown in Fig. 2.6. Negative pulses of
current hyperpolarise the neuron, producing downward deflections in the
voltage recording trace. Positive pulses of current depolarise the neuron,
causing upward deflections in the voltage trace. If the depolarising current
pulses are small, the neuron’s response is passive, and the size of the voltage
change is proportional to the size of the current stimulus. However, when
the depolarising potential reaches a critical threshold value, the neuron
responds actively by producing a spike. Membrane potential changes
rapidly, reaching a peak when the inside of the neuron is about 40 mV pos-
Neuron physiology and action potentials 29
itive in relation to the outside, and then returns to its resting level, often
after a transient hyperpolarisation. A second spike cannot be initiated until
a short time after the first one is complete, this time being called the refrac-
tory period.
When a spike occurs at one location, it depolarises the membrane for
some distance away. This depolarisation acts as the stimulus for new
spikes, but, because of the refractory period, only in the length of axon that
has not recently produced a spike. A spike is therefore a stereotyped event,
in which the membrane potential swings rapidly between resting potential
to about 40 mV positive and back. The amplitudes of spikes in extracellular
recordings appear to vary from axon to axon, but this is because the size of
the extracellular signals picked up by the electrodes depends on the diam-
eters of the axons and how far away the axons are from the electrodes.
A difference in the voltage inside and outside a cell is the consequence of
two features of cell physiology. The first is that charged ions are distributed
unevenly between the cytoplasm and extracellular fluid, and the second is
that the membrane contains pores that, when open, allow particular ions to
pass through. The resting potential arises because potassium ions are more
concentrated in the cell’s cytoplasm than in the extracellular fluid, and
pores that are open in the resting cell membrane allow these ions (but not
others such as sodium or chloride) to pass through readily. Driven by the
concentration gradient, potassium tends to flow out of the cell. As each
potassium ion flows out of the cell, however, it carries with it a positive
charge, making it successively more difficult for subsequent potassium ions
to leave. A balance is established between the concentration gradient that
tends to push potassium outwards and the electrical gradient that tends to
retain it inside the cell. The electrical gradient at the balance point is the
resting potential.
The pores that allow ions to pass through the membrane are usually
formed by specialised protein molecules that aggregate in particular forma-
tions. These proteins and their associated pore are called a channel. The
resting membrane potential is therefore determined by the action of one
kind of potassium channel. During a spike, two other types of channel, a
sodium channel and another type of potassium channel, come into play.
These two channel types have pores that are closed in the resting cell, but
tend to open when the neuron is depolarised to the threshold voltage and
beyond. Electrical excitation of the cell causes the proteins to alter their
30 Nerve cells
shape so that the central pore opens, and these channels are called voltage
sensitive. The voltage-sensitive sodium channels open more quickly than
the potassium channels so that at first sodium ions, which are more con-
centrated outside the neuron than inside, tend to pass into the cell. This
continues until the electrical gradient across the membrane balances the
concentration gradient for sodium ions, which occurs when the inside of
the cell is about 40 mV more positive than the extracellular fluid. This is why
the intracellular voltage reaches ⫹40 mV at the peak of a spike. The spike is
a brief event because the sodium channels do not remain open for very
long, and voltage-sensitive potassium channels open at about the same
time as the sodium channels close, so that potassium now leaves the
neuron and the membrane repolarises towards resting potential.
The number of ions that flow during a spike is only a small proportion
of the total number of ions in the cytoplasm and extracellular fluid.
Nevertheless, the neuron needs to keep its batteries topped up by main-
taining a difference in concentrations of sodium and potassium between its
inside and outside. It does this by means of pumps, proteins in the mem-
brane that consume metabolic energy to transport ions against concentra-
tion gradients.
2.3 Synapses
Neurons communicate with each other and other cells at specialised junc-
tions called synapses, a term introduced by Charles Sherrington in 1897.
Sherrington was convinced, from his physiological and anatomical studies
of reflex functions of the spinal cord, that the transfer of signals between
cells is a different type of process from the transfer of signals within a cell.
It was not until the electron microscope was developed, 50 years after this,
that the structure of synapses could be examined in detail. Synapses are
tiny, and also very diverse. Sometimes a functional connection between
two neurons is composed of just a single synaptic structure but more com-
monly there are several. Sometimes thousands of discrete anatomical con-
tacts make up a connection between two neurons. A complex neuron such
as a vertebrate motor neuron typically receives several tens of thousands of
individual synapses from a variety of sources.
The neuron that is passing a signal on is referred to as presynaptic, and a
neuron that is receiving a signal is referred to as postsynaptic. Transmission
Synapses 31
Figure 2.7 Operation of the three main types of synapse between nerve
cells. (a) Schematic arrangement for recording the transfer of signals across
a synapse by inserting electrodes close to the presynaptic and postsynaptic
sites. (b) Recordings of presynaptic and postsynaptic potentials at the three
types of synapse.
across a synapse can be studied by using two microelectrodes, one to

record presynaptic potential and the other to record postsynaptic potential
(Fig. 2.7). A significant majority of synapses are chemical synapses at which
signals are transferred by way of chemicals such as acetylcholine or some
amino acids that act as neurotransmitters.
The postsynaptic signal is called the postsynaptic potential, usually
abbreviated to PSP. There is a delay, usually of about a millisecond, between
the spike and the PSP. Generally, a spike in a presynaptic neuron, about 100
mV in amplitude, causes a PSP that is much smaller, a few millivolts in
amplitude, although a few synapses are specialised to act as relays where
the PSP is strong enough to trigger a spike in the postsynaptic neuron. At
32 Nerve cells
some synapses, the PSP excites the postsynaptic neuron by depolarising it

and these PSPs are called excitatory postsynaptic potentials (EPSPs). At
other synapses, the PSP hyperpolarises and inhibits the postsynaptic
neuron and these PSPs are inhibitory postsynaptic potentials (IPSPs).
When a neuron is excited, it is more likely to release neurotransmitter
and so pass signals to its postsynaptic targets. This is because the amount
of neurotransmitter that is released depends on the membrane potential at
the presynaptic sites. The way that membrane potential is linked to neuro-
transmitter release is by calcium ions which enter the presynaptic terminal
through voltage-sensitive channels. Like the voltage-sensitive sodium
channels that cause the upswing of a spike in an axon, these calcium chan-
nels open in response to depolarising signals. When a spike invades a pre-
synaptic terminal, it depolarises the membrane strongly for a short time, so
there is a strong pulse of calcium entry into the terminal and a brief squirt
of neurotransmitter is released. Most of the delay in transmission across
a chemical synapse is due to the time it takes for the voltage-activated
calcium channels to open.
Spikes are not necessary for neurotransmitter release, and the link
between membrane potential and the rate of neurotransmitter release is
used by many neurons to transmit information in a graded manner. Such
neurons have been studied in various visual systems (see section 5.3) and
the motor systems of arthropods (see section 8.8), and are called non-
spiking neurons because they normally operate without producing spikes.
Instead, small variations in membrane potential regulate the amount of
neurotransmitter release, probably because the rate of calcium entry into
the presynaptic terminals is directly controlled by voltage-sensitive calcium
channels. Rather than being released in squirts, neurotransmitter tends to
dribble from the presynaptic terminals of these neurons, and some of them
sustain a steady leakage.
Neurotransmitter diffuses extremely rapidly across the cleft that separ-
ates the membranes of the presynaptic and postsynaptic neurons. Some of
its molecules attach to receptor proteins on the postsynaptic membrane
and cause the shape of the receptor to alter. Some receptors are parts of ion
channels, and a frequent action of neurotransmitter is to cause this type of
ion channel to open. As with voltage-sensitive channels, these chemical-
sensitive channels allow particular types of ions to pass through when they
are open, and the direction and strength of flow are governed by the con-
Integration of postsynaptic potentials 33
centration and electrical gradients across the membrane. Many of these

transmitter-activated channels allow sodium ions to pass through, and
sodium will tend to enter the neuron, exciting it by causing an EPSP. Other
transmitter-activated channels allow chloride ions to pass, and these ions
tend to enter the neuron causing an IPSP. Potassium-conducting transmit-
ter-activated channels also mediate IPSPs. Integration involves weighing
up the balance of EPSPs and IPSPs within a cell and this determines how
excited it is.
At a few synapses, electrical current can flow directly from one neuron
to another. These are called electrical synapses (see Heitler, 1990, for a
review). Under an electron microscope, an electrical synapse can be recog-
nised as a region where the cell membranes of two neurons come close and
touch. Protein channels called connexons link the two cells so that, when
the synapse is active, the cytoplasm in the two cells is in direct contact,
forming a conductive pathway for the electrical current. Sometimes electri-
cal synapses conduct equally well in both directions, so that each neuron is
both presynaptic and postsynaptic. A signal passes from one cell to another
with negligible delay at an electrical synapse (Fig. 2.7b), which means that
information passes across electrical synapses slightly more rapidly than
across chemical synapses. Possibly associated with this, some of the best-
known electrical synapses occur in pathways concerned with rapid escape
responses (see Chapter 3). Electrical synapses can also help ensure that the
neurons they connect are excited synchronously, which is useful for the co-
ordination of some motor activities. This is also useful in some sensory
systems such as the retina, in which electrical coupling helps to filter out
real signals from extraneous noise. However, compared with chemical syn-
apses, the scope for integration offered by electrical synapses is limited.
2.4 Integration of postsynaptic potentials
Integration involves weighing up the balance of EPSPs and IPSPs within a

cell and the outcome determines how excited the neuron is at any given
moment. In order to sort out all the PSPs that it receives, a neuron needs
to have a way of combining them together. In general, the dendrites of a
neuron function to combine PSPs as well as to receive them and this
function is made possible by their passive cell membrane. This contrasts
with the membrane of an axon, which is described as being active
34 Nerve cells
because it contains the voltage-sensitive channels responsible for gener-

ating spikes.
It is difficult to study the way in which PSPs travel in the dendrites of most
neurons because of their complex branching structure. However, an axon
or a muscle cell conducts potentials that are below the threshold for a spike
in the same way as a long, unbranching dendrite would, and can be used to
illustrate the properties of a signal as it is conducted along a length of
passive membrane (Fig. 2.8a). These properties can be examined by using a
pair of microelectrodes, one to inject pulses of current and the second to
record voltage changes at different distances away from the point of injec-
tion.
Two important changes occur as a postsynaptic potential is conducted
along a passive membrane. One is that its amplitude decreases as it travels
away from its point of origin. Signal size is not directly proportional to the
distance, but declines exponentially. This can be understood by consider-
ing the flow of electrical current along and across the cell membrane (Fig.
2.8b). When the cell membrane is passive, an axon or dendrite behaves
electrically like an insulated cable and this type of current flow was studied
by the engineers who first constructed long submarine cables to carry tele-
graph messages. The passive conducting properties of axons and dendrites
are therefore often described as their cable properties.
The electrical model of a cell membrane in Fig. 2.8b consists of a network
of electrical resistors, each of which offers a pathway for the flow of current.
The current spread along the axon will be greater if the membrane resis-
tance is high in relation to the internal resistance because less current will
then flow out through the membrane. The cell membrane is indeed a rela-
tively poor conductor of electricity, being composed chiefly of lipid, which
is a good electrical insulator. Ion channels are the major route for current
flow across the membrane and without them even less current would be
siphoned away from the pathway along the length of the axon. The electri-
cal resistance of the cytoplasm depends on the diameter of the axon, with
wider axons being better conductors than narrow ones. Signals are there-
fore conducted passively for greater distances along wide axons and den-
drites than along narrow ones.
The way in which a signal changes in size along the length of an axon or
dendrite is expressed as the space constant, which is defined as the dis-
tance over which a signal decays to 37 per cent of its original amplitude
Figure 2.8 Cable properties of a length of axon or dendrite. (a) Two elec-
trodes are inserted into the axon, one to inject a pulse of current and the
second, which is moved successively further from the first, to record mem-
brane potential. Drawings of voltage responses to a square current pulse are
drawn for three locations. Note how the signal changes in size and shape as
distance from its point of origin increases. (b) Electrical circuit for the mem-
brane. (c) Graph to show how membrane potential declines with distance,
including the definition of space constant for a cable.
36 Nerve cells
(Fig. 2.8c). The space constant may be as great as 1 cm in a really wide axon,
such the giant axon of a squid which has a diameter of 1 mm. It is difficult
to measure the space constant in small neurons, but values from 0.2 to 1
mm have been estimated for typical mammalian dendrites. These values
are large in relation to most neurons, which means that postsynaptic
potentials will readily carry along the length of the average dendrite.
The second change that happens to a signal as it travels along a passive
membrane is that its waveform becomes more rounded. This can be seen by
comparing a square-shaped pulse of electrical current injected into an axon
at one point with the responses at two locations further along (Fig. 2.8a).
This change in shape is due to electrical capacitance of the membrane, a
property that occurs because the relatively high-resistance membrane sep-
arates two low-resistance solutions having unlike charges (negative inside,
positive outside). A capacitor stores electrical charges and a store, whether
an electrical capacitor or a bucket of water, takes time to fill and empty. As
the signal decreases in size with distance, it also takes longer to alter the
charge on the membrane capacitor and this is why the waveform recorded
in an axon or dendrite changes more slowly as the distance from the origin
of the signal increases. A measure of the effect an axon or dendrite has on
the time course of a signal due to its capacitance is given by its time con-
stant, which is defined as the time taken for a constant current pulse to
change the membrane potential to 37 per cent short of its final value.
When the cell membrane is responding passively, the potential change is
graded in amplitude with the size of the electrical stimulus, in approximate
accordance with Ohm’s law. A number of these graded potentials can there-
fore sum, giving a resultant potential with an amplitude that is proportional
to the current flow. The way in which the passive membrane of a neuron
sums together different synaptic inputs is shown in Fig. 2.9.
In Fig. 2.9a a neuron is shown receiving one excitatory synapse, and the
EPSP it produces is recorded by a microelectrode inserted into the cell
body. A single EPSP depolarises the membrane, but not as far as the spike
threshold for the postsynaptic neuron. If the presynaptic neuron spikes
twice, the second EPSP sums with the first, and the postsynaptic neuron is
more excited than by a single EPSP. A third EPSP is then able to carry the
membrane potential past threshold. The ability to sum several postsynap-
tic potentials occurring in rapid succession depends on the time constant
of the membrane. The capacitative properties of the postsynaptic
Figure 2.9 Integration of postsynaptic

potentials. In each section, an experiment
to show a particular type of integration is
shown schematically, together with the
appearance of recordings on an oscilloscope
screen. (a) Temporal summation of EPSPs.
(b) Spatial summation of an EPSP with an
IPSP. (c) Comparison of EPSPs at two loca-
tions along a long dendrite.
38 Nerve cells
membrane slow the voltage changes so that an EPSP will decay more slowly
than the synaptic current that gave rise to it. Consequently, a second EPSP,
following soon after the first, will add its depolarisation to that remaining
from the first EPSP. Several EPSPs in rapid succession are thus able to depo-
larise the membrane to threshold even though a single one is insufficient to
do so. This is called temporal summation.
In Fig 2.9b a neuron is shown receiving an excitatory synapse on one den-
drite and an inhibitory synapse on another dendrite some distance away.
When both presynaptic neurons produce spikes simultaneously, an EPSP
will be generated in the one postsynaptic dendrite at the same time as an
IPSP is generated in the other dendrite. As these two potentials spread
towards the recording electrode, they will meet and sum, producing a much
reduced depolarisation in this case. Similarly, if two excitatory synapses are
active at the same time, the postsynaptic neuron will be more strongly
excited than if just one synapse were active. This integration of PSPs from
different sites is called spatial summation. The ability to sum synaptic
potentials occurring at different sites within the neuron depends on the
space constant of the membrane. This is particularly important because the
postsynaptic membrane itself is not usually capable of generating spikes; it
is not electrically excitable. So the summed potential change produced by
synaptic activity must spread passively to the nearest patch of membrane
that is capable of generating action potentials. This is called the spike-ini-
tiating zone, and it is usually at some little distance from the synapses. For
neurons with long axons, it is located close to the origin of the axon.
In this way, the sites of synaptic action and the sites of spike generation
are restricted to different parts of the neuron, and are all linked together by
the passive spread of graded potentials. This arrangement offers great
scope for varying the integrative properties of neurons by varying the space
constant, the dendritic geometry and the sites of synaptic input. For
instance, the closer a given synaptic input is to the spike-initiating zone, the
more influence it will generally have on the output, because its postsynap-
tic potentials will have decayed less than those of other inputs further away
by the time they reach the spike-initiating zone. This is illustrated in Fig.
2.9 c, where two excitatory synapses are shown at different distances from
the base of a long thin dendrite. An EPSP from the synapse towards the base
of the dendrite will have decayed far less than one from the more distant
synapse when recorded close to the spike-initiating zone. Similarly, an
Additional mechanisms in integration 39
inhibitory synapse close to the spike initiating zone can effectively veto
excitation that arises further out on the dendritic tree.
2.5 Comparison of spikes and graded potentials
Graded potentials and spikes are the universal language of neurons in all
animals that have been studied. Nerve cells carry out their varied functions
with just these two kinds of membrane potential. Spikes are clearly suitable
for conveying information over long distances within the nervous system.
Because spikes are relatively constant in amplitude, variation in signal
strength is expressed in the frequency of spikes, with stronger signals being
represented by higher numbers of spikes in a given time than weaker
signals. The refractory period restricts the frequency of spikes that can be
produced to a few hundred, or at most a thousand, per second. Graded
potentials are used for short-distance communication and play an essential
part at synapses and at sensory endings (see Chapter 4). Variation in signal
strength is coded directly in the amplitude of membrane potential: the
stronger a signal, the greater the change in membrane potential.
In electrical terms, graded potentials are analogue signals, whereas
spikes are digital signals. In a nervous system, information is constantly
transformed between these two types. Thus, chemical synapses transform
digital signals into analogue signals by converting a burst of spikes arriving
at the presynaptic terminal into a series of postsynaptic potentials. These
graded variations in membrane potential can simply add together and so
allow the neuron to perform its fundamental task of integration. In turn,
analogue signals are transformed into digital signals at the spike-initiating
zone of a neuron, where the frequency of spike generation depends on the
rate of arrival of EPSPs. As larger and more numerous EPSPs spread to the
spike-initiating zone, depolarisation of the membrane will occur more
rapidly. So the threshold will be reached sooner and the time interval
between successive spikes will be less, which means that the spike fre-
quency will be higher.
2.6 Additional mechanisms in integration
The processes of spatial and temporal summation of PSPs in passive den-

drites underlie much of the integrative activity of neurons, but additional
40 Nerve cells
features can also play significant roles. First, some dendrites have voltage-
sensitive channels that can boost the amplitude of their input signals. One
place where this has been studied is in output neurons of the thalamus of
the vertebrate brain, which have the job of relaying sensory inputs to par-
ticular processing areas in the cortex. Second, there is not necessarily
a physical separation between input and output regions of a neuron.
Sometimes dendrites make output synapses as well as receiving inputs, and
in some amacrine cells, input and output synapses are intermingled. This
produces local circuits in the nervous system, where information can flow
through processes of a number of different neurons without travelling very
far. In the operation of local circuits, we cannot necessarily consider a
whole neuron as an integrative unit because individual dendrites can act
independently, summing local inputs to regulate output from their own
synapses. The operation of local circuits has been particularly studied in
the vertebrate olfactory system and retina (see Box 5.1, p. 107), and in cir-
cuits that control insect legs (see Chapter 8). Finally, the way that inhibition
opposes excitation is not always by simple arithmetic summation of EPSPs
and IPSPs. Strategically placed inhibitory synapses can make it hard for
excitatory synapses to generate large EPSPs because, when postsynaptic
channels at the inhibitory synapse open, they provide pathways for current
flow through the cell membrane which effectively short circuits it. This
reduces both the amplitudes of PSPs and, by reducing the neuron’s space
constant, the distance that they can travel.
2.7 Conclusions
Neurons consist of the same components as other animal cells, but are spe-
cialised for processing information. This specialisation is evident morpho-
logically in the fine processes, the dendrites and axon, by which neurons
make contact with each other. Physiologically, the most important special-
isation is the maintenance of a membrane potential, which varies in
response to incoming signals.
The transmission of signals over long distances is accomplished by
spikes, which are active changes in membrane potential and are conducted
along an axon with a constant amplitude at a finite speed. Signal strength is
coded as spike frequency, a digital process. Over short distances, neurons
use graded potentials that vary in amplitude and duration. These graded
Further reading 41
potentials usually originate at postsynaptic sites. Their amplitude reduces

and their waveform becomes smoothed with increasing distance from the
point of origin. Signal strength is coded as the amplitude of change in
membrane potential, an analogue process. These graded potentials can be
summed together: like variations in potential reinforce each other, and
unlike variations tend to cancel each other out. This simple electrical prop-
erty is the basis of most of the integrative processing that occurs in neurons.
Thus, the nervous system uses analogue processes to combine informa-
tion and digital processes to transmit it over long distances. These are basi-
cally simple processes but a great variety of operation is achieved by
adjustment of the physical properties of each neuron and its connections
with other neurons. Physical properties such as membrane resistance,
which affect the space and time constants, determine how far signals that
originate at separate synapses are able to travel and combine together. The
input and output connections of each neuron, reflected in its specific
branching pattern, control the flow of information and create functional
circuits in the nervous system. By combining the basic components and
processes of neurons together in different ways, evolution has generated an
extremely efficient system for processing information, one that is well able
to control an animal’s behaviour.
Further reading
Delcomyn, F. (1998). Foundations of Neurobiology. New York: Freeman. A compre-
hensive survey of basic neurobiology, covering a wide range of examples.
Nicholls, J.G., Martin, A.R & Wallace, B.G. (1992). From Neuron to Brain: a Cellular
and Molecular Approach to the Function of the Nervous System, 3rd edn.
Sunderland, MA: Sinauer. An established text which is widely used for its clear
and thorough account of how nerve cells work.
Reichert, H. (1992). Introduction to Neurobiology. Stuttgart: Thieme Verlag; New
York: Oxford University Press. A clear and concise account that introduces the
reader to a range of topics.
3 Giant neurons and escape behaviour
3.1 Introduction
When an animal is suddenly attacked by a predator, it must respond with

great urgency if it is to escape. The neuronal circuits that initiate such an
escape response must be both straightforward and reliable in order to fulfil
their biological function. A staightforward circuit is essential to ensure
speed in initiating the escape, and a reliable circuit is needed not only to
make sure the response occurs when required but also to avoid false
alarms. These qualities of simplicity and reliability, which are of great survi-
val value to the animal, are also of service to the neuroethologist exploring
the role that nerve cells play in behaviour. Consequently, several of these
startle responses have been studied in detail and they provide valuable
insight into the flow of information through the nervous system from
sensory inputs to muscular output.
Furthermore, these neuronal circuits often involve neurons that are
exceptionally large and, because of this, are called giant neurons. The func-
tion of giant neurons is to conduct spikes rapidly along the body, but their
size also makes them readily accessible to study with microelectrodes. The
giant neurons therefore offer a major opportunity to investigate the role of
individual nerve cells in behaviour.
Two main functions must be carried out by the neuronal circuit that ini-
tiates any behaviour pattern, including escape. First of all, a decision to ini-
tiate an activity must be made at some point in the circuit. This is usually
done on the basis of incoming sensory information, which will often be
filtered by the sensory receptors and neurons closely associated with them
to extract particular stimulus features. This must occur rapidly because a
startled animal has only a few milliseconds left to it in which to initiate
escape action. The relevant processing must be fed to the decision point
42
Introduction 43
with minimal delay. Once the decision to initiate an escape has been made,
the second function is an executive one: the circuit must include connec-
tions that lead away from the decision point to excite those neurons which
are involved in the escape movement and to inhibit other neurons involved
in incompatible movements.
Early work in this area gave rise to the idea that there may be a special
class of high-order interneurons that carry out these two functions. The
term command neuron was applied to such interneurons by Wiersma &
Ikeda (1964), who found that electrical stimulation of single interneurons
elicited co-ordinated movements of the abdominal ventilatory append-
ages of crayfish (swimmerets), even when the system was deprived of
sensory feedback. Subsequently, a variety of instances was analysed in
which apparently normal behaviour could be elicited by stimulation of a
single neuron, and the term command neuron came into general use.
Following controversy as to exactly what a command neuron might be,
Kupferman & Weiss (1978) suggested that a command neuron should be
defined as a neuron that is both necessary and sufficient for the initiation
of a particular behaviour pattern. In this chapter, three cases of startle
behaviour in which identified interneurons have been studied in detail are
examined. Each case shows that this definition of a command neuron
cannot be strictly applied to any of the neurons examined. This gives rise
to a situation that will be entirely familiar to ethologists: one must either
be content to use the term ‘command neuron’ loosely, or abandon its use
altogether.
The three examples of startle behaviour considered here are the tail flip
of crayfish, the fast start of teleost fish, and escape running by cockroaches.
These have a number of features in common, including the fact that giant
interneurons play a key role in initiating the behaviour in each case. The
crayfish tail flip is probably understood, in terms of its neuronal circuitry,
more completely than any other behaviour pattern of comparable com-
plexity. Part of the reason for this relatively complete understanding is that
recording from identified neurons has gone hand in hand with an increas-
ingly exact study of the responses of intact animals, using film and video
techniques. This illustrates well how ethological and neurophysiological
methods of analysis can mutually reinforce one another during the inten-
sive study of a single system.
Figure 3.1 Giant interneurons involved in crayfish (Procambarus) startle

behaviour. (a) Crayfish showing the location of the central nervous system
(in solid black), a chain of ganglia. Also shown are electrodes implanted to
record neuronal activity in the freely moving animal. (b) Activity recorded
by these electrodes during a startle response: a tap on the abdomen (stimu-
lus) is followed by a spike in a lateral giant and a potential in the abdominal
flexor muscles. (c) Transverse section of the connectives between two
abdominal ganglia, showing the locations of the lateral and medial giants.
(a modified after Schramek, 1970; b redrawn after Krasne & Wine, 1975; c
redrawn after Krasne & Wine, 1977.)
3.2 Giant neurons and the crayfish tail flip
Crayfish escape from the strike of a predator by means of a rapidly executed

tail flip, produced by flexing and re-extending the whole abdomen. The
abdomen is able to act as an effective locomotory organ because the last
two (sixth and seventh) abdominal segments are modified to form the tail
fan (Fig. 3.1a). A single flip of the tail fan is capable of moving the animal
several centimetres through the water. The power for this movement is pro-
vided by the fast flexor muscles, which occupy much of the space within
each abdominal segment. These are called fast muscles because they
produce rapid twitch contractions, in contrast to a set of much smaller,
Giant neurons and the crayfish tail flip 45
slow muscles that produce graded postural movements of the abdomen. In

addition, each abdominal segment contains fast and slow extensor
muscles, and these are also much less substantial than the flexors.
The innervation of these muscles was first studied by Keis Wiersma
(1947), who showed that the giant neurons are involved in the control of
these muscles during a tail flip. About ten large motor neurons innervate
the flexor muscles on each side of each abdominal segment. One of these is
an exceptionally large motor neuron, called the motor giant, which sends
an axon branch to every fast flexor muscle fibre. Another is an inhibitory
motor neuron that also innervates every muscle fibre. The remaining motor
neurons are simply known as fast flexor neurons, and each of these inner-
vates only a localised group of fibres within the fast flexor muscles. During
a tail flip, the motor giant and the fast flexor motor neurons are excited via
two pairs of large interneurons, the lateral and medial giant interneurons
(Fig. 3.1b, c). The control exerted by these two giant interneurons is so
strong that a single spike in either a lateral or a medial giant interneuron is
sufficient to trigger a tail flip.
Both of these giant interneurons extend along much of the length of the
central nervous system but they differ considerably in structure. The
medial giants have their cell bodies and dendrites in the brain, where they
receive sensory input, and their axons extend down to the last abdominal
ganglion. In contrast, the lateral giants are segmentally repeated structures,
formed from separate cells linked end to end. Each segment contains a cell
body, dendrites and a length of axon that abuts against the corresponding
axon in the next segment. Where the axons abut, there is a segmental
synapse between two successive lateral giants (see Fig. 3.3a). The lateral
giants receive input only in the abdominal segments. At first, it was thought
that both kinds of giant interneurons initiate the same behaviour pattern
because a spike in either of them produces a rapid flexion of the abdomen.
However, a more detailed series of studies, using a combination of neuro-
physiological and high-speed filming techniques, showed that the lateral
and medial giants initiate different patterns of behaviour (Fig. 3.2).
Activation of the medial giants elicits contraction of the fast flexor
muscles in all abdominal segments, and this produces a uniform curling of
the abdomen that propels the animal straight backwards (Fig. 3.2a).
Activation of the lateral giants elicits flexor contraction in the anterior seg-
ments of the abdomen but not in the posterior segments; the latter remain
Figure 3.2 The different kinds of tail flip produced by the medial and lateral
giant interneurons. The precise pattern of movement is correlated with
activity in the giants by filming animals with electrodes implanted (as in
Figure 3.1a). Tracings from these high-speed films show (a) a medial giant
flip elicited by a tap on the head, and (b) a lateral giant flip elicited by a tap
on the abdomen. (c) Diagram showing the pattern of synaptic connections
between the two giants (horizontal lines) and the motor giant neurons (ver-
tical lines) in the abdominal segments. Direct synaptic connections are rep-
resented by filled circles and the absence of synaptic connections is
indicated by an asterisk. (From Wine, 1984.)
straight and so cause the thrust to be directed mainly downwards, thereby

pitching the animal forwards (Fig. 3.2b). These two kinds of movements are
well adapted to the different sorts of stimuli that excite the two types of
giant interneuron. The lateral giants are triggered only by sudden mechan-
ical stimuli that originate posteriorly, such as a sharp tap to the abdomen,
and a lateral giant flip appropriately carries the animal in an anterior direc-
tion. Similarly, the medial giants are triggered only by stimuli applied to the
head region and a medial giant flip carries the animal in a backward direc-
tion. In this way, each kind of tail flip removes the animal from the source
of the stimulus.
The differences between the two kinds of tail flip can be explained by
differences in the synaptic connections between the giant interneurons
and the motor giant neurons in the abdominal segments (Fig. 3.2c). At the
Giant neurons and the crayfish tail flip 47
Figure 3.3 Electrical synapses between neurons involved in crayfish startle

behaviour. (a) Drawing of an abdominal ganglion to show the relative posi-
tions of the lateral giant and giant motor neurons, the synapse between
them, and the arrangement for recording from each side of this synapse.
The segmental synapse between successive lateral giants is also shown. (b)
Simultaneous intracellular recordings from the lateral giant and motor giant
neurons close to the synapse, demonstrating the negligible delay due to
electrical transmission. (c) Intracellular recording from the lateral giant at a
point close to the synapse. Following electrical stimulation of the appropri-
ate sensory neurons, a compound EPSP and a spike are recorded in the
lateral giant with a very small delay. The two components of the EPSP (␣
and ␤) are produced by separate pathways from the sensory neurons, as
described in the text. (a and b modified after Furshpan & Potter, 1959; c
redrawn after Krasne, 1969.)
point of synaptic contact, the motor giant branches over the surface of the
interneuron’s axon in a characteristic manner, which is readily recognised
when the motor axons are filled with intracellular dye (Fig. 3.3a). In the
anterior segments of the abdomen, both lateral and medial giants receive
these synaptic branches from the motor giants, but in the posterior seg-
ments the synaptic branches to the lateral giant are clearly missing,
whereas those to the medial giant are present. This distribution of synapses
has been confirmed by testing for postsynaptic responses by recording with
a microelectrode: responses to medial giant activity can be obtained in all

abdominal segments, but responses to lateral giant activity are only
obtained in the anterior segments. In the thorax, the situation is reversed.
The medial giant makes no output connections to motor giants, but the
lateral giant does connect in the more posterior segments (Heitler & Fraser,
1993). Contraction of flexor muscles in the posterior part of the thorax helps
bend the body into the jackknife shape that propels it rapidly forwards.
Hence, the consistent difference in the pattern of abdominal flexion is
brought about by differences in the synaptic connections between the
respective controlling interneurons and a shared motor output system.
Another behavioural difference, noticed early in the study of crayfish
startle behaviour, is that sometimes only a single tail flip occurs, while at
other times an apparently similar stimulus evokes a whole series of tail flips
in rapid succession. The latter are produced by alternating flexions and
extensions of the abdomen, repeated at a frequency of 10 to 20 Hz, and this
behaviour is termed escape swimming. Studying animals with implanted
electrodes reveals that escape swimming does not involve the giants, which
are active only before the first tail flip. The fast flexor muscles of the
abdomen are used in swimming, and they are controlled by the fast flexor
motor neurons but not by the motor giants and the giants. Escape swim-
ming can be triggered on its own, without an initial giant-mediated flip, by
stimuli that are weaker or have a slower onset compared with those which
trigger the giant interneurons. Although a sharp stimulus can evoke a single
giant-mediated flip without subsequent swimming, it is more usual for the
giant-mediated flip to be followed by non-giant swimming.
The tail flips generated in escape swimming fall into three relatively
stereotyped classes: linear flips, which resemble medial giant flips; pitching
flips, which resemble lateral giant flips; and twisting flips, which tend to
rotate the animal about its longitudinal axis as well as having a backward
component. During swimming, these classes of tail flips are arranged in
sequences that result in the animal being propelled backwards away from
the original stimulus. Following a medial giant response, swimming
involves only linear flips, which simply continue the backward movement
away from the threat at the head end. A lateral giant response is followed by
one or two pitching flips, which turn the animal in a complete somersault
so that it lands on its back with its head facing the stimulus. Then two or
three twisting flips turn the animal dorsal side up again, and finally a series
The lateral giant interneuron: input and output 49
of linear flips carry it backwards away from the stimulus (Cooke &
Macmillan, 1985).
In addition, the swimming system is able to act upon directional infor-
mation in the stimulus that is ignored by the giant system. The lateral
giants, for example, always generate a bilaterally symmetrical response that
carries the animal straight forward, regardless of whether the stimulus
comes directly from behind or from one side. However, the path followed by
subsequent swimming has a lateral component that steers the animal away
from a stimulus delivered to one side of the abdomen (Reichert & Wine,
1983). All these results show that escape swimming is well adapted to
exploit the initial advantage gained by a giant-mediated tail flip. The follow-
ing account focuses on the lateral giant response and its relation to subse-
quent swimming because this system has been studied in most detail.
3.3 The lateral giant interneuron: input and output
Undoubtedly the most striking feature of the initial tail flip is its speed.
Within 50 ms, abdominal flexion is completed and the animal has usually
moved some distance through the water. The mean delay between the stim-
ulus and the onset of flexor muscle potential is 6 ms (see Fig. 3.1b). This
speed is required because the tail flip is probably a response to a predator
that is extremely near to or touching a crayfish. Physiologically, the speed is
partly achieved by extensive use of large neurons and of electrical synapses
(see section 2.3). The segmentally repeated synapse between successive
giant axons is electrical so that spikes are conducted across it with negli-
gible delay and the chain of neurons acts effectively as a single axon. Each
segmental lateral giant is also coupled via an electrical synapse with its
contralateral partner, and with the motor giant neuron, which provides a
pathway that conveys a spike from a lateral giant neuron to a fast flexor
muscle in about 2 ms.
In the anterior segments of the abdomen, besides exciting the motor
giants, the lateral giants also excite the other fast flexor motor neurons. A
lateral giant makes synapses that have a mixed chemical and electrical
nature with fast flexor motor neurons (Fraser & Heitler, 1991), but most
excitation is conveyed by an indirect route, through an interposed neuron
called the segmental giant because of the large size of its dendrite. There is
a segmental giant on each side of each abdominal ganglion. Because of the
Figure 3.4 Neuronal circuit for startle behaviour mediated by the lateral
giant interneuron. (a) Schematic representation of the excitatory pathway
from the mechanoreceptors to the flexor muscles, showing chemical (—䉳)
and electrical (—|) synapses. Labelled circles represent: the receptors (R);
sensory interneurons (SI); lateral giant (LG); segmental giant (SG); motor
giant (MoG) and fast flexor motor neurons (FF). The sensory pathways that
generate the two components of the compound EPSP (Fig. 3.3c) are labelled
␣ and ␤. (b) Diagrammatic representation of the arrangement of the above
components within the abdominal nervous system. The various compo-
nents are not drawn to scale and only one segment of the lateral giant is
shown; the medial giant and extensor motor neurons are not included.
(Modified after Wine & Krasne, 1982).
negligible delay in the transmission of spikes from the lateral giant to fast
flexor motor neurons across the two electrical synapses, the existence of
the segmental giants was not suspected for many years. An unusual feature
of the segmental giants is that they have a blind-ending axon in one of the
lateral nerve roots, which might indicate that they were originally motor
neurons. The output connections of a lateral giant are summarised in Fig.
3.4, which also shows the sensory input.
The lateral giant interneuron: input and output 51
The lateral giants receive input exclusively from sensory neurons that are
arranged in pairs, each pair being attached to a stout hair. These sensory
structures occur on the dorsal surface of the abdomen, including the tail
fan. The sensory neuron is a bipolar neuron, meaning that two processes
extend out from its cell body, which lies at the base of a hair. One process,
the dendrite, attaches to the inside of the hair, and is stretched when the
hair is deflected, initiating a chain of events that leads to the production of
an electrical signal. The second process is the axon, which carries signals in
the form of spikes to the central nervous system. These sensory neurons are
sensitive only to touch and to high-frequency water movements (about 80
Hz). They do not respond to low-frequency slosh of the water, which is
detected by other hair receptors on the cuticle. Consequently, the hair
receptors are well suited to detecting the shock wave in the water produced
by the acceleration of a predator towards the crayfish, or even actual
contact by the predator. This is a good example of sensory filtering:
because these mechanoreceptors respond only to particular types of dis-
turbance, the lateral giants, which they excite, will respond only to immi-
nent danger, and not to water movements caused by waves or movements
by the crayfish itself.
The main input pathway to the lateral giants runs from these receptors
via sensory interneurons located in the segmental ganglia of the abdomen
(Fig. 3.4). Most of these interneurons have relatively large dendrites and
axons, from which synaptic potentials and spikes can be recorded readily
with microelectrodes. The interneurons are an order of magnitude less
numerous than the hair receptors, so there is considerable convergence in
the input pathway. At the same time, there is some divergence, because
each receptor axon branches to make contact with several interneurons.
The receptors make chemical synapses onto the interneurons, which in
turn make electrical synapses onto the lateral giant. A small proportion of
the receptors make electrical synapses directly onto the lateral giant.
All the synapses on to the lateral giant are thus electrical but, because the
postsynaptic neuron is so much larger than the presynaptic one, single
spikes cannot generate enough current to depolarise the lateral giant above
threshold. Instead, single spikes generate EPSPs which summate in the
usual way (see section 2.4); hence, several spikes must arrive within a short
time of each other at these synapses in order to trigger a spike in the lateral
giant. Following stimulation of the hair receptors, a compound EPSP with
two components can be recorded in the lateral giant (see Fig. 3.3c). The first
component is due to receptors that synapse directly on the lateral giant, but
this input alone is insufficient to drive the giant to threshold. The second
component is due to input via the sensory interneurons. When the
mechanical stimulus is strong enough, the second component sums with
the first to carry the giant’s membrane potential beyond threshold.
Thus, the lateral giant has a high threshold for spike initiation. Also, its
membrane has a short time constant, which means that individual EPSPs
die away quickly, so that many EPSPs must arrive within a short time of
each other if they are to sum and trigger a spike. Together with the require-
ment of the mechanoreceptors for a high-frequency mechanical stimulus,
this ensures that the lateral giant responds only to a sudden, abrupt stimu-
lus. These characteristics are reinforced by the occurrence of habituation in
tail flips mediated by a lateral giant: when a crayfish is tapped on the
abdomen repeatedly, the probability of a tail flip in response diminishes
rapidly. Stimulation at the rate of one tap per minute can diminish respon-
siveness to zero within ten minutes, and then many hours rest are needed
for recovery. Physiological analysis has shown that one site at which habit-
uation occurs is in the input pathway. The strengths of the chemical synap-
ses between the receptor axons and sensory interneurons diminish (there
is no change in the properties of the electrical synapses). However, habitu-
ation of the reflex is not simply due to an automatic reduction in the
strengths of these synapses every time the sensory neuron is activated.
Experiments by Krasne and Teshiba (1995) indicate that interneurons that
originate in the brain exert a powerful influence over habituation, and
ensure that habituation only occurs if stimulation of the sensory neurons is
sufficiently intense to cause a tail flip.
3.4 The decision to initiate startle behaviour
It is clear that a lateral giant plays a crucial role in the circuit that generates
a tail flip (Fig. 3.4). In fact, it is the place in the circuit where the decision is
made to initiate this movement. Information from the hair receptors con-
verges on the lateral giant and is integrated in its dendrites. The decision to
initiate a tail flip is thus made when the compound EPSP reaches threshold
and triggers a spike that is conducted along the axons of the lateral giants.
Filtering in the sensory pathway ensures that a spike is only generated in
The decision to initiate startle behaviour 53
response to appropriate stimuli. Once a spike is initiated in the lateral giant,

the excitatory output to the motor giant produces contraction of all the fast
flexor muscles in the most direct manner possible, and this is backed up by
the slightly less direct route to the fast flexor motor neurons.
It can be shown that the lateral giant is essential to the production of a tail
flip by hyperpolarising the cell so that it cannot produce spikes. When a
lateral giant is prevented from firing in this way, stimuli that would nor-
mally cause a tail flip no longer do so (Olson & Krasne, 1981). For a given
stimulus intensity, the amount of hyperpolarisation needed to abolish the
tail flip is exactly that needed to prevent the lateral giant from firing. In the
hyperpolarised cells, the size of the compound EPSP increases gradually as
a function of the intensity of the stimulus delivered to the receptors, up to
and beyond normal threshold level. This confirms that the lateral giant is
the true point of decision in the circuit. If the decision were being made
earlier, so that the lateral giant acted merely as a relay, one would expect to
see an abrupt increase in EPSP size near threshold, due to firing of an
earlier decision-making neuron. This effect is never seen.
The rapid flexion of the abdomen in a tail flip is a direct consequence of
activity in the lateral giants, and it would be reasonable to assume that the
same would be true of the subsequent re-extension, which follows within
100 ms of the original stimulus. However, this is not the case. Re-extension
is a reflex initiated by particular mechanoreceptors stimulated by the
flexion (Reichert, Wine & Hagiwara, 1981). That actual movement of the
abdomen during flexion, and not just firing of the lateral giants, is neces-
sary for re-extension is shown by recordings made from freely moving
crayfish with electrodes implanted to record activity in the fast flexor and
extensor muscles. If the abdomen is then restrained so that it cannot flex,
or the motor nerves to the flexors are cut, no activity is recorded from the
extensors following a stimulus that is adequate to excite the lateral giants.
No synaptic connections are apparently present by which the lateral
giants could excite the extensor muscles after exciting the fast flexors
during a tail flip. However, there are two appropriate classes of mechano-
receptors that do have appropriate connection with the fast extensor motor
neurons. These are the hair receptors on the dorsal surface of the abdomen
and muscle receptor organs, which each consist of a single sensory neuron
with numerous branching dendrites that are attached to a slender receptor
muscle. The receptor muscle lies parallel to a powerful muscle that extends
the abdomen, and the muscle receptor organ, therefore, responds to move-
ments that flex the abdomen. There is one phasic (fast, adapting response)
and one tonic (slow, maintained response) muscle receptor organ on each
side of each abdominal segment.
Both the tonic and the phasic muscle receptor organs make direct synap-
tic connections on to fast motor neurons of the extensor muscles. These are
chemical, excitatory synapses, and each spike in the sensory neuron trig-
gers a discrete EPSP in the motor neuron, with a synaptic delay of about 1
ms. Ordinarily, this reflex is used for making postural extensions of the
abdomen in response to externally imposed loads, but it is equally suitable
for triggering rapid re-extension following the internally produced flexion
in a tail flip. The hair receptors also have a substantial excitatory input to
the fast extensor motor neurons and trigger EPSPs with a longer delay, indi-
cating a less direct pathway between sensory and motor neurons.
Recording from crayfish with electrodes implanted in muscles shows that a
waterborne stimulus to these hairs is able to excite brief activity in the
extensor muscles of a resting crayfish. During a normal tail flip, then, the
rapid water movement generated by abdominal flexion will excite the hair
receptors, which will add their excitatory input to that of the muscle recep-
tor organs and so activate the fast extensors of the abdomen.
The initial tail flip is normally followed by escape swimming, which con-
sists of a series of tail flips that are not mediated by the giant axons. The
lateral giants do not trigger escape swimming, any more than they do
abdominal extension (Reichert & Wine, 1983). Nor is swimming triggered by
sensory feedback from the first movements of flexion or extension. Instead,
stimuli to the smooth hairs on the abdomen activate another pathway, that
acts in parallel with, but more slowly than, the lateral giant pathway.
Evidence for this is obtained by electrical stimulation of the lateral giants
through electrodes implanted into resting, unrestrained crayfish. This stim-
ulation elicits a rapid flexion followed by a re-extension, but it bypasses the
pathways that are normally activated to trigger startle behaviour. In these
circumstances, fewer than 1 per cent of all tail flips are followed by swim-
ming. Such a result shows that neither activity in the lateral giants nor
sensory feedback from the actual tail flip is sufficient to elicit escape swim-
ming, so that this must be activated by an independent and parallel
pathway that leads from the mechanoreceptive hairs.
Taps on the abdomen that are just below threshold for firing the lateral
Executive functions of the lateral giant neuron 55
giants will often trigger escape swimming in well-rested crayfish. When this
happens, the delay between the stimulus and the onset of the first flexor
contraction is about 240 ms, slightly longer than the interval between the
stimulus and the first flexion in a swimming bout that follows lateral giant
spikes and a tail flip. This means that swimming is triggered through a much
slower pathway than that responsible for triggering a tail flip, and that
swimming is not simply inhibited during a tail flip. If inhibition was operat-
ing, one would expect swimming to begin with a shorter delay when it
occurs without a preceding lateral giant response, but this is not the case. In
fact the opposite is true: the average delay for swimming that is preceded by
a giant-mediated tail flip is somewhat less than the delay for swimming that
occurs without an initial tail flip (184 ms compared with 240 ms). This indi-
cates that the lateral giants may facilitate the onset of swimming, although
they do not trigger it. Hence, the smooth transition from a tail flip to swim-
ming in normal escape behaviour is largely due to the activation of two
different pathways that activate the two types of movement in sequence.
Escape swimming actually begins with a burst of activity in the abdomi-
nal extensor muscles, preceding the first flexor contraction by some 50 ms.
This extensor activity is an event that is distinct from the reflex re-extension
of the abdomen following a giant-mediated flip, but the two events often
overlap when the delay to the onset of swimming is shorter than average.
The extensors lead the flexors throughout swimming, with a nearly con-
stant delay between extension and flexion over a wide range of swimming
frequencies. As the frequency of abdomen movements declines during a
swim, the delay between flexion and subsequent extension lengthens and
the animal coasts with the abdomen in the flexed position; then it rapidly
extends and flexes it again at the start of the next cycle. Evidently, extension
of the abdomen during swimming is not produced by sensory reflexes,
which must be inhibited. This and other evidence combine to suggest that
the alternating extension and flexion during swimming is produced by a
pattern generator consisting of networks of interneurons that work in a
similar way to those described in Chapter 7 for other activities.
3.5 Executive functions of the lateral giant neuron
A lateral giant not only initiates a tail flip, but also extensively co-ordinates
the sequence of events involved. This executive function is achieved by a
massive and widely distributed array of inhibitory effects that follow a spike
in the giant axon. Pathways lead away from the lateral giant to exert inhibi-
tion at almost every point in the neuronal circuit generating a tail flip. The
IPSPs produced at these points in the circuit differ from one another, in the
delay to their onset and in duration, in such a way as to ensure that each
part of the response begins and ends at the right time.
The extensor motor system is the first place where inhibition is seen fol-
lowing a spike in a lateral giant axon. This inhibition is accomplished by
parallel actions at three points in the motor pathway to the extensor
muscles: the muscle receptor organs, the motor neurons, and the extensor
muscle fibres themselves. The muscle receptor organ is inhibited by a
special accessory cell associated with its sensory neuron; and the fast
extensor muscles are inhibited by a motor neuron that causes IPSPs in the
muscle fibres, called the extensor inhibitor motor neuron. The lateral giant
excites these two kinds of inhibitory neuron, and also inhibits the excitatory
motor neurons of the extensor muscles (Fig. 3.5a). These inhibitory actions
begin within a few milliseconds of a lateral giant spike; in fact, the delay to
the onset of the IPSPs in the extensor motor neurons is as short as that of
the EPSPs in the fast flexor motor neurons (Fig. 3.5b). The duration of the
IPSPs produced at all three points in the extensor pathway is relatively
short, having an average value of 30 ms.
These arrangements clearly function to ensure the accurate timing of the
reflex extension following giant-mediated flexion. The early onset of this
inhibition prevents premature activation of the extensor reflex from inter-
fering with the initial flexion. Inhibiting the extensor pathway at three sep-
arate locations makes quite certain that extension cannot occur while the
inhibition lasts. The average duration of the IPSPs at the three locations is
about the same as the average duration of the giant-mediated flexion, and
so the extensor system is released from inhibition just as flexion is com-
pleted. The inhibitory action of the lateral giant thus co-ordinates flexion
and re-extension effectively, even though the extensor system does not
receive any additional excitation from the lateral giant.
Co-ordination of the response is continued by inhibition of the flexor
motor system while re-extension takes place. Two inhibitory neurons have
been identified that are important in shutting down the flexor system: the
inhibitory motor neuron which innervates every fast flexor muscle fibre;
and the motor giant inhibitor, an interneuron which prevents the motor
Executive functions of the lateral giant neuron 57
Figure 3.5 Inhibition of the abdominal extensor muscles by the lateral

giant. (a) Neuronal circuit generating inhibition of the extensors, showing
representative neurons: the lateral giant (LG); fast extensor motor neuron
(FE); extensor inhibitor (EI); the muscle receptor organ (MRO); and its
accessory cell (AC). Inhibitory neurons are shown in solid black; excitatory
(—䉳) and inhibitory (—䊉) connections are shown, but electrical and chemi-
cal synapses are not distinguished. (b) Typical recording used to build up
the interpretation given in (a). The upper trace is an extracellular record
from the motor nerve to the flexors, showing the lateral giant spike and sub-
sequent compound spike from flexor motor neurons. The middle and lower
traces are intracellular records showing, respectively, an EPSP in the exten-
sor inhibitor and an IPSP in the fast extensor motor neuron. Note the short
delay of the postsynaptic potentials after the lateral giant spike. (b from
Wine, 1977; copyright Springer-Verlag.)
giant from firing. Both of these inhibitory neurons are excited indirectly by
the lateral giant (Fig. 3.6a). The motor giant inhibitor is strongly excited by
the fast flexor motor neurons and so brings about inhibition of the motor
giant very soon after the latter has fired. The flexor inhibitor is weakly
excited by a variety of sources, including the fast flexor motor neurons and
the corollary discharge interneurons, a class of interneurons which are
excited by the segmental giants and with axons that run between segments.
These sources of input bring about early depolarisation of the flexor inhibi-
tor, but the threshold for spiking is reached only after the corollary discharge
interneurons in other ganglia are recruited to provide additional input.
This roundabout pathway increases the delay, and consequently a spike
is not triggered until about 15 ms after the spike in the lateral giant (Fig.
3.6b), by which time the motor giant is already inhibited. The actual flexion
movement has only just begun at this time; but it must be remembered that
it takes a few milliseconds for the muscles to develop tension and for the
Figure 3.6 Delayed inhibition of the fast flexor muscles by the lateral giant.
(a) Neuronal circuit generating inhibition of the flexors, showing represen-
tative neurons as in Fig. 3.4 plus: the motor giant inhibitor (MoGI); corollary
discharge interneuron (CDI); and flexor inhibitor (FI). Symbols have the
same meaning as in Fig. 3.5, and chemical and electrical synapses are not
distinguished. (b) Recording demonstrating the delayed activation of the
flexor inhibitor. The upper trace is an extracellular record showing the delay
between the spike in the lateral giant and the spike in the flexor inhibitor;
the lower trace is a simultaneous intracellular record from the flexor
inhibitor showing the compound EPSP gradually rising to threshold. (b from
Wine & Mistick, 1977; reprinted by permission of Wiley-Liss, Inc., a sub-
sidiary of John Wiley & Sons Inc.)
tension to overcome inertia in the skeleton, so that by the time actual

movement begins the electrical events that triggered the movement have
already been completed. Hence, the delayed inhibition from the lateral
giant prevents any additional flexor activity and prepares the way for re-
extension. In addition, the flexor inhibitor receives an excitatory, monosyn-
aptic connection from the muscle receptor organ and this input becomes
active as soon as the muscle receptor organ is released from inhibition. This
input sums with that from the lateral giant, thereby prolonging activity of
the flexor inhibitor for the duration of re-extension.
A further important place where inhibition acts is on the input side of the
circuit, at the synapse between the hair receptors and the sensory interneu-
rons. This inhibition acts both postsynaptically, on the dendrites of the
interneurons, and also presynaptically, on the axon terminals of the recep-
tor neurons. Presynaptic inhibition of this type is common in mechano-
sensory systems, and its action is to reduce or prevent the release of
neurotransmitter. Both postsynaptic and presynaptic inhibition are
delayed to about 15 ms after the lateral giant spike due to the indirect
Summary of pathways in crayfish startle behaviour 59
pathway that mediates them, including the corollary discharge and other
interneurons. This inhibition has a long duration of about 50 ms. However,
not all the hair receptors are inhibited in this way; those that provide input
to the extensor motor neurons are not inhibited and they contribute to the
re-extension reflex.
The presynaptic inhibition of the first input synapse plays an important
part in co-ordinating the startle reflex because abdominal flexion is a dra-
matic movement that stimulates the same receptors that trigger the reflex.
If this input were not inhibited, it could cause a perpetual cycle of repeated
tail flips by positive feedback. As it is, the onset of inhibition coincides with
the onset of movement of the abdomen and so prevents this feedback
effect. The long duration of the inhibition makes sure that a second tail flip
is not triggered before the first one is completed. The occurrence of presyn-
aptic inhibition at the first synapse is well correlated with the fact that this
synapse is the site of habituation of the response. If postsynaptic inhibition
alone were present, then synaptic transmission could still habituate during
the repetitive stimulation caused by abdominal flexions, and this could
render the whole system unresponsive for several hours. This habituation
does not occur because presynaptic inhibition prevents the presynaptic
terminals from being fully activated.
3.6 Summary of pathways in crayfish startle behaviour
The startle response of crayfish is a simple behaviour, which is initiated

with the least possible delay. Nevertheless, the neuronal circuit underlying
this behaviour is quite sophisticated and involves complexities that could
hardly be predicted from the behaviour itself. The main pathways involved
in linking the response to an adequate stimulus are shown in Fig. 3.7. The
initial tail flip is triggered by sensory information, which is fed with
minimal processing on to the lateral giant, which acts as a command
neuron. The lateral giant produces excitation of the flexor muscles and, at
the same time, short-lasting inhibition of the extensors. This is followed by
delayed, long-lasting inhibition of the flexors and of sensory input to the
lateral giant itself. Finally, excitation of the extensor muscles is produced by
sensory feedback from the first flexion. Through the operation of these
pathways, the first tail flip is completed (to full re-extension) by about 110
ms from the initial stimulus.
Figure 3.7 A flow diagram summarising the functional relations between

the major components of the startle behaviour in crayfish. The usual
symbols are employed to represent excitatory (—䉳) and inhibitory (—䊉)
relations, but here the labelled boxes do not represent individual neurons,
except in the case of the lateral giant. The components of the initial, giant-
mediated tail flip are enclosed in the horizontal, dotted rectangle, and those
of non-giant swimming are enclosed in the vertical, dotted rectangle. The
flexor component has some vertical elements that are not common to both
systems, whereas the three separate sensory components may well have
some elements in common. (Modified after Wine, 1984.)
Sensory information that is adequate to trigger the giant-mediated tail

flip also triggers escape swimming by an independent pathway that
involves more elaborate sensory processing. This elaborate processing
enables the swimming system to take account of directional information in
the stimulus that is ignored by the giant-mediated first tail flip and also
introduces a considerable delay. The delayed excitation triggers a bout of
swimming, in which the extensors lead the flexors in each cycle. The delay
in triggering swimming is such that the first movement of escape swim-
ming, an extension, overlaps with or immediately follows the re-extension
of the first tail flip. When the crayfish swims, inhibition prevents both acti-
vation of the extensors, through sensory feedback from their receptor
organs, and activation of the lateral giant system.
As the startle behaviour of crayfish continues to be studied, many addi-
tional complexities are coming to light; these include additional circuit ele-
ments such as non-spiking interneurons that control abdominal muscles
Mauthner neurons and the teleost fast start 61
and circuits that control walking legs. Finally, it should be noted that the
startle response is not automatic, although it will override and inhibit any
conflicting, competing activity once it has been triggered. The readiness
with which the startle response can be triggered is modulated by a range of
factors, including strong control of elements in the circuit by neurons
descending from the brain that can make it almost impossible to elicit a
startle response in some circumstances (Krasne and Teshiba, 1995).
3.7 Mauthner neurons and the teleost fast start
When a sharp tap is delivered to the side of an aquarium, the fish inside
exhibit a characteristic startle response consisting of a brisk swivelling
movement that displaces the fish sideways by a small amount. In natural
circumstances, this is an effective escape movement that enables the
animal to dodge the strike of a predator. The key neuron in this startle
response is called the Mauthner neuron; there is one of these neurons on
each side of the brain of most species of fish and of amphibians. Most
studies of these neurons have been made in goldfish and zebra fish.
Because of the exceptionally large size of Mauthner neurons, it has been
possible to study them in both dissected preparations and intact animals;
these studies provide one of the few cases in vertebrates in which a clear
causal relationship has been established between activity in a particular
neuron and performance of a specific behaviour pattern. The way the
Mauthner neuron operates shows a number of instructive parallels with the
crayfish lateral giant neuron.
In teleost fish, the startle movement initiated by a Mauthner neuron is
known as a fast start, and consists of a stereotyped sequence of movements
that occur in three stages. In the initial stage, the trunk muscles contract all
along one side of the body so that the animal assumes a C-like shape with
the head and tail bent to the same side. A number of other actions are also
initiated, such as closing the mouth, drawing in the eyes and extending the
fins. During the second stage, muscle contractions proceed down the other
side of the body so that the tail straightens. These first two stages result in a
sudden acceleration that propels the animal in a direction that is deter-
mined by the extent of the body bend in the first stage and displaces the fish
by about one body length. The third stage consists of normal swimming
movements, or sometimes coasting, which carries the fish further away.
Figure 3.8 The Mauthner neuron in teleost fish. (a) The general location of
the bilateral pair of Mauthner neurons (M) shown schematically for a larval
zebra fish (Brachydanio). (b) The right Mauthner neuron is shown in a thick,
transverse section of the hindbrain at the level of the eighth cranial nerve,
together with the inputs from the lateral line and acoustico-vestibular
systems. The output of the Mauthner axon to the spinal motor neurons is
shown in a thick transverse section of the spinal cord. (c) The Mauthner
neuron of an adult goldfish (Carausius), reconstructed from transverse
sections of a cell injected with intracellular dye. (a modified after Prugh,
Kimmel & Metcalfe, 1982; b modified after Kimmel & Eaton, 1976; c redrawn
after Zottoli, 1978.)
The large cell body and two primary dendrites of a Mauthner neuron are
located in the hind brain (Fig. 3.8). The axon crosses to the opposite side of
the brain before descending the nerve cord to contact motor neurons of
trunk muscles. This basic anatomy was described by Bartelmez (1915), who
first suggested that it was involved in startle behaviour, although for several
Mauthner neurons and the teleost fast start 63
decades after that it was thought that the Mauthner neuron is involved in
normal swimming movements. The role of the neuron in startle behaviour
was firmly established when recordings from the neuron were correlated
with movements in freely moving animals – spikes in a Mauthner neuron
can be identified unambiguously in extracellular recordings because their
waveform provides a characteristic signature. When a spike is recorded
from a Mauthner neuron in response to a sudden sound, it is always imme-
diately followed by a large muscle potential in the trunk muscles on the side
of the body opposite to the Mauthner neuron cell body, and the fish per-
forms a fast start (Fig. 3.9a).
The precise timing of the Mauthner spike in relation to the stages of the
fast start is clarified by using an implanted electrode in conjunction with
high-speed filming of the response (Fig. 3.9b). This method shows that the
average delay from a stimulus to a Mauthner spike is about 6 ms. There is a
further delay of about 2 ms until the muscle potential starts and then a
delay of 6–10 ms until the muscle develops sufficient tension for actual
movement to begin. The speed of this startle response is comparable to that
of the crayfish tail flip, which is not surprising as both are natural responses
for evading the strike of a predator. Another parallel with the crayfish startle
system is that only a single spike occurs in the Mauthner neuron, and this
precedes the initial C-like bend. Mauthner spikes do not accompany the
second stage or subsequent swimming, which must therefore be due to the
activity of other interneurons acting in parallel with the Mauthner neuron.
The Mauthner neuron on the side closest to the stimulus is the only one
that produces a spike and, because this excites the contralateral muscula-
ture, the initial C-like turn is made to the side away from the stimulus. The
size of this initial turn is not constant but varies with the angle of the threat-
ening stimulus with respect to the fish. This is shown by dropping a metal
ball on to the water near the goldfish and analysing high-speed films of the
resulting startle response. In this experiment, the angle through which
the fish has turned by the end of stage one is inversely proportional to the
direction of the impact of the ball (Eaton & Emberley, 1991). When the
startle response is filmed in goldfish with electrodes implanted in the trunk
musculature (as in Fig. 3.9 a), the results show that progressively larger
turns in stage one are correlated with progressively longer and more
complex muscle potentials. The fish is evidently controlling how far it turns
by varying the activity of motor neurons supplying the trunk musculature.
Figure 3.9 Activity of the Mauthner neuron during startle behaviour. (a) A
goldfish in an aquarium with electrodes implanted for recording from one
Mauthner neuron and from both left and right trunk muscles in a freely
moving animal. The stimulus is delivered by a loudspeaker (at left) and
monitored by a hydrophone (at right). (b) Typical recording of a startle
response to a sound stimulus (bottom trace, On/Off). The brain electrode
(top trace) picks up the Mauthner spike (M) and also the prominent poten-
tial that spreads from the trunk musculature. The two myogram electrodes
(middle traces) show that this muscle potential originates from the con-
tralateral muscles (contra) and not the ipsilateral muscles (ipsi). (c)
Simultaneous recording with electrodes implanted in the brain and high-
speed filming in dorsal view (representative silhouettes at bottom) serves to
establish the precise timing of the Mauthner spike in startle behaviour.
(a and b modified after Zottoli, 1977; c modified after Eaton, Lavender &
Wieland, 1981.)
However, direct electrical stimulation of the Mauthner neuron consistently

results in a short and simple muscle potential. Hence, the size of the C-like
turn must be controlled by other interneurons that are active in parallel
with the Mauthner neuron (Eaton, Di Domenico & Nissanov, 1991).
Excitation and inhibition in the Mauthner neuron 65
Figure 3.10 Sensory input to the Mauthner neuron. (a) Schematic represen-
tation of input to the lateral dendrite from the eighth cranial nerve, showing
receptors with direct electrical (—|) synapses onto the Mauthner neuron
(M) and those with indirect chemical synapses (—䉳) by way of sensory
interneurons (SI). (b) Intracellular recording from the distal end of the
lateral dendrite, as indicated in (a), showing two superimposed records.
Following electrical stimulation of the receptor axons, a compound EPSP is
recorded, with an early component (␣) due to the electrical synapse and a
later component (␤) due to the chemical synapses. In one of the records,
the latter triggers a spike, which is small in size because it has spread pas-
sively from the spike-initiating zone to the recording site. (b modified after
Diamond, 1968.)
3.8 Excitation and inhibition in the Mauthner neuron
The Mauthner neuron is excited by a wide range of sensory modalities; as a

result, a large number of sensory neurons converge on this neuron, either
directly or indirectly, and this is reflected in the great size of the two main
dendrites (Fig. 3.8c and Fig. 3.10a). The cell body and the dendrites sum the
synaptic potentials which are generated by inputs from the sensory
neurons, and this summed signal is conducted passively to the initial part
of the axon. The membrane of the Mauthner neuron has a large space con-
stant, up to 5mm, which enables the dendrites to conduct passive signals
several hundred microns from the more distant synapses towards the axon.
The initial part of the axon plays a crucial part in the integrative mechanism
of the Mauthner neuron because it is where spikes are initiated when the
input delivered from the dendrites exceeds threshold – the dendrites and
cell body are incapable of supporting spikes.
The input that has been studied in most detail is that coming from the
auditory system. The underwater sounds that accompany the onrush of an
attacking predator can be detected by the sensory receptors in the inner ear
of the threatened fish. Using a loudspeaker (see Fig. 3.9a) or a high-pressure
jet of water as a stimulus in experiments shows that acoustic stimuli are
certainly capable of initiating fast starts. Visual stimuli and input to the
lateral line may augment the auditory stimulus but do not appear to be
capable of triggering a spike in the Mauthner neuron on their own. For
example, in the above experiment in which a ball was dropped on the
water, the goldfish could see the ball coming but a fast start was not initi-
ated until several milliseconds after the ball hit the water. This suggests that
the triggering stimulus for the Mauthner neuron was the acoustic one from
the impact of the ball on the water.
The sensory neurons from the inner ear provide the major input to the
lateral dendrite, where they make direct (monosynaptic) contact in the
form of large club-shaped endings. When postsynaptic potentials are
recorded by a microelectrode in the lateral dendrite, they clearly contain
two components: the ␣ component, which is an initial, sharply rising EPSP
after a short latency of about 0.1 ms; and the ␤ component, which is a more
gradually rising EPSP with a longer latency, of about 1 ms (Fig.3. 10b). The
latencies of these two events suggest that the first is mediated by electrical
synapses, and the second by chemical synapses, possibly from sensory
interneurons. The ␣ component can be elicited by relatively weak shocks to
the eighth cranial nerve, which excite the largest sensory axons but fail to
excite the smaller sensory axons that are responsible for the chemical EPSP.
Electron microscopy of the distal part of the lateral dendrites confirms that
these axons synapse with the Mauthner neuron by way of tight junctions
between the membranes of the two cells, which are characteristic of electri-
cal synapses.
As stimulus strength is increased further, both the ␣ and the ␤ compo-
nents grow in size, until their EPSPs are sufficiently large to trigger a spike.
In the recording shown in Fig. 3.10b, the spike appears small and rounded,
and the chemical EPSP that triggers it appears smaller than the preceding
electrical EPSP. This is because the site of the recording was towards the end
of the lateral dendrite, in the region where it receives the electrical synap-
ses; both the spike and the chemical EPSP, which is caused by synapses
closer to the cell body, were conducted passively along the cell back to the
recording site.
Thus, the electrical EPSP has a priming role; in natural conditions, it pro-
Outputs and executive functions of the Mauthner neuron 67
motes early depolarisation of the spike-initiation zone, so that a following,

large, chemically mediated EPSP rapidly triggers a spike. As in the input
circuit to the crayfish lateral giant, both electrical and chemical synapses
play roles in conveying sensory information to the Mauthner neuron. The
large diameter of the Mauthner neuron axon, together with extensive insu-
lation with myelin, ensure that, once initiated, a spike travels rapidly down
the spinal cord. In goldfish, the conduction velocity for this spike is 85 m/s,
which enables the signals to travel the whole length of the spinal cord
within 1 ms. Consequently, all the motor neurons to the trunk muscles on
that side of the body are excited virtually simultaneously.
3.9 Outputs and executive functions of the Mauthner neuron
The Mauthner neuron has monosynaptic connections with many of the

spinal motor neurons that it excites; a short collateral of the Mauthner axon
makes contact with a special region of the motor axon several micrometres
from the cell body (see Fig. 3.8b). A spike in the Mauthner axon produces an
EPSP in the motor neuron with a synaptic delay of about 0.6 ms, so the
transmission is probably chemical. The excitation of a substantial number
of other motor neurons is relayed by way of a premotor interneuron and so
involves a slightly greater delay. The motor axons travel a relatively short
distance to the trunk muscles, where they have normal chemically trans-
mitting synaptic endings on the muscle surface. This short and direct
pathway is responsible for the short delay of about 2 ms between a
Mauthner neuron spike and the onset of a spike in the trunk muscles (see
Fig. 3.9b, c).
A spike in the Mauthner axon inhibits the spinal motor neurons contra-
lateral to the axon at the same time as it excites those ipsilateral to the axon.
The inhibition appears to be mediated by inhibitory interneurons that
cross the midline and are activated by electrical synapses from the
Mauthner axon (Fig. 3.11a). This crossed inhibitory circuit makes sure that,
if only a short interval in time separates spikes in the left and right
Mauthner neurons, only the earlier of the two Mauthner spikes is able to
fire its motor neurons. If the two Mauthner neurons spike simultaneously,
the crossed inhibition prevents any motor output at all.
Within the brain, branches from the axon of the Mauthner neuron
excite a number of cranial relay neurons, which carry out some important
Figure 3.11 Output circuitry of the Mauthner neuron. (a) Neuronal circuit
generating excitation and inhibition of spinal motor neurons, showing: the
Mauthner neuron (M); spinal relay neurons (SR); crossed inhibitory
interneuron (CI); and spinal motor neurons (SMo). Chemical excitatory
(—䉳), electrical excitatory (—|) and chemical inhibitory (—䊉) synapses are
shown. (b) Neuronal circuit generating excitation of the cranial motor
neurons and self-inhibition, showing the Mauthner neuron (M); cranial
relay neuron (CR); cranial motor neurons (CMo); and the Mauthner
inhibitor (MI). Synaptic symbols are as in (a), plus electrical inhibition of
the axon cap (—䊉|).
functions in the startle response (Fig. 3.11 b). The cranial relay neurons
excite motor neurons of muscles that close the jaw and draw in the eyes,
which helps to streamline the fish for its escape. In addition, the cranial
relay neurons monosynaptically excite a group of interneurons that inhibit
both the Mauthner neurons.
These Mauthner inhibitors have cell bodies clustered around the ventral
dendrite and they act to inhibit the Mauthner neuron in several ways. Some
of them exert an unusual form of electrical inhibition: they have axons
coiled tightly round the origin of the Mauthner axon, contributing to the
axon cap (see Fig. 3.8b). When spikes are generated in these axons, they
make the outside of the Mauthner neuron locally more positive, which has
the same effect as driving the intracellular potential more negative. Spikes
from these neurons thus exert a kind of brief stranglehold on the Mauthner
neuron, right at the zone where its spikes are initiated. This electrical inhi-
The startle reaction of a cockroach 69
bition follows a Mauthner spike with a delay of about 1 ms and so prevents

the Mauthner neuron from firing twice in response to a given stimulus.
Furthermore, the cranial relay neurons are excited by branches of both left
and right Mauthner axons, with the result that the contralateral Mauthner
neuron also receives electrical inhibition with the same short delay after a
spike in the ipsilateral axon.
The electrical inhibition of both Mauthner neurons is a brief event,
lasting no more than about 2 ms, but it is followed by inhibition through
chemical synapses that lasts much longer. The Mauthner inhibitors
produce this inhibition through their axon terminals on the cell body of the
Mauthner neuron, which make conventional chemical synapses. The
chemical inhibition begins about 1 ms after the electrical inhibition, and
lasts for about 45 ms, which takes it well into the second stage of the startle
response. There is evidence that this direct inhibition is backed up by pre-
synaptic inhibitory synapses made by the Mauthner inhibitors on to the
synaptic terminals of the sensory neurons on the Mauthner neuron den-
drites, and also by separate inhibitory circuits that are activated directly by
the sensory neurons. The result of these inhibitory mechanisms is not only
that the active Mauthner neuron and its contralateral partner are shut
down as soon as a spike has been generated, but also that they are kept shut
down long enough for the startle response to go to completion.
Although the involvement of Mauthner neurons in triggering startle
responses in teleost fish is clear, this does not mean that this is the only
function that these neurons serve. Sudden lunges by goldfish to capture
prey on the water surface involve movements similar to those of a startle
response, and follow a Mauthner neuron spike (Canfield & Rose, 1993).
Besides integrating information about a source of potential danger, a
Mauthner neuron must be able to collect information that triggers an
attack at the appropriate time. The input and output circuitry of a
Mauthner neuron is arranged so that it can initiate and co-ordinate a rapid,
directed movement.
3.10 The startle reaction of a cockroach
Cockroaches are extremely difficult to catch; this is partly because they are
exquisitely sensitive to very small air currents which trigger fast-running
responses that are accurately directed away from the source of wind. In
Figure 3.12 Startle response of the cockroach. (a) Video recording, made
from above the cockroach, of the turning response to a wind puff delivered
from the front left of the animal. The outlines of the body and head are
traced from every second frame. (b) Leg movements during an escape turn
caused by a puff of wind coming from the right and slightly to the front. The
cockroach was held so that its body could not move, but its legs slipped
over a lightly oiled glass plate. The initial positions of the legs are indicated
with dotted lines; the final positions of the middle and hind legs are drawn
as solid lines. The arrow above the cockroach indicates the direction in
which the animal would have faced if its body had been free to rotate.
(a redrawn after Comer & Dowd, 1993; b redrawn after Ritzmann, 1993.)
their natural habitat, which for the American cockroach Periplaneta amer-
icana is leaf litter on the forest floor, cockroaches have been shown to be
able to evade the strikes of one of their predators, the toad, by detecting the
movement of air caused by movement of the toad’s tongue (Camhi & Tom,
1978). Cockroaches are particularly sensitive to sudden increases in the
velocity of air currents, which is how they can distinguish an attack by a
toad from meteorological wind (Plummer & Camhi, 1981).
Unlike the startle response in crayfish, cockroach startle behaviour is
normally controlled by activity in several giants, and involves many spikes
in these neurons rather than the single spike that marks the decision-
making process in the crayfish lateral giant or Mauthner neuron.
Undoubtedly associated with this, the cockroach startle response is initi-
ated more slowly than that of the crayfish or fish, but is nevertheless a rapid
behaviour, with the first movement occurring within 50 ms of the start of a
wind stimulus. It is a rapid turn away from the direction of attack (Fig.
3.12a) that, unlike walking or running, involves the co-ordinated move-
ment of all six legs at the same time (Fig. 3.12b). Each leg pushes or pulls, so
that the cockroach swivels to face away from the direction from which the
air current is coming. The cockroach then runs forward, using the usual
tripod gait mode of locomotion in which the legs are moved in two sets of
three.
The sense organs that excite the giants are called filiform sensilla; each is
a slender, whip-like structure inserted in a socket on the ventral surface of
a cercus, the sensory structure that projects from each side of the last
abdominal segment. The sensillum articulates in its socket, and can move
in one particular plane. It is attached to a single bipolar sensory cell that is
excited when its sensillum is deflected in one direction. The axon of the
sensory cell enters the last abdominal ganglion, where it terminates in a
series of fine branches that are restricted to one side of that ganglion and
which make chemical synapses on to dendrites of the giant interneurons.
The filiform sensilla are arranged in 14 columns, most of which run the
length of the cercus. Sensilla of a particular column all share the same
direction for deflection, and so all respond best to air currents coming from
the same direction. A cercus of an adult cockroach has about 20 segments,
and each column is usually represented by one filiform sensillum on each
segment. The arrangement for two adjacent segments is shown in Fig.
3.13a.
Seven pairs of interneurons that originate in the last abdominal ganglion
and have axons that run in the ventral nerve cord have been labelled as
giants. The axons range from 20 to 60 ␮m across, so they are considerably
smaller than the giants of crayfish or Mauthner neurons, but are appre-
ciably larger than other axons in the cockroach nerve cord. Anatomically,
two separate groups of giants can be distinguished, one with axons located
dorsally and the other with axons located ventrally in each connective
nerve. The giants in the ventral group are larger than those in the dorsal
group and trigger escape running. The structure of one ventral giant,
number 1, is shown in Fig. 3.13b.
Carefully conducted experiments have revealed that each giant responds
most vigorously to air currents from a particular direction (Kolton & Camhi,
1995). To determine the directional selectivity of a giant, spikes were
recorded from its axon using a microelectrode that contained a stain so that
the neuron could later be identified from its anatomy. Carefully controlled
puffs of air, always with a peak velocity of 0.85 m/s, were directed at the
Figure 3.13 Giant interneurons and sensory analysis of wind-puff direction

in the cockroach. (a) Best directions of deflection for exciting sensory
neurons of the filiform sensilla of the seventh segment of the right cercus,
indicated by arrows. Each circle shows the location of one filiform sensillum
on the ventral surface of segment 7 or 8; the diameter of the circle depends
on the length of the hair. (b) Drawing of right giant 1 in the last abdominal
ganglion. (c) Polar plot of the directional sensitivities of the three ventral
giants (GI 1–3) that have their axons in the right connectives, constructed as
explained in the text. (a redrawn after Dagan & Camhi, 1979; b modified
after Harrow et al., 1980; c redrawn after Kolton & Camhi, 1995.)
right cercus from a nozzle that could be rotated to stimulate it from

different directions. Air currents were applied from an angle of 45° above
the cockroach, which is the type of angle from which an attack might be
made. Spikes were counted for the first 50 ms following the start of each
stimulus; this is roughly the time over which a giant neuron would respond
before a cockroach starts to move in response to an air current. The results
are expressed as a polar plot (Fig. 3.13c). The origin of the plot represents
the cercus, and each response is plotted as a point whose direction from the
origin represents the direction of the stimulus, and whose distance from
the origin represents the number of spikes.
In Fig. 3.13c, each point is the mean of several stimuli in experiments on

five different cockroaches. For each interneuron, the different points are
joined by a line, and the resulting shape gives an immediate impression of
the directional sensitivity of that giant. The shape and location of the plot
for each giant are called its receptive field, a term that is commonly used to
refer to the region around an animal from which a neuron receives sensory
input. All three interneurons respond preferentially, but not exclusively, to
stimuli coming from the same side as their axons (the right side in this
case), and interneuron 1 responds well to stimuli from the front or from the
rear. There is a little overlap between the receptive fields of interneurons 2
and 3, but if we look at the receptive field to determine which stimulus
direction elicits the best responses in these two neurons, we find that the
directions for the two giants are almost exactly at right angles to each other:
interneuron 2 prefers stimuli from behind and to the side of the animal,
whereas interneuron 3 responds best to stimuli from the front and the side.
In fact, the best direction for interneuron 3 corresponds to the direction in
which the cercus normally points, and the best direction for interneuron 2
is at right angles to this. Thus, a cockroach can determine whether an
air current is coming from the left or from the right by comparing the
responses of its left and right giants; and it can distinguish stimuli coming
from the front from those coming from the rear by the relative excitation of
interneurons 2 and 3. If a current of air is coming directly from the side,
these two giants will respond with almost equal vigour.
A cockroach giant interneuron can generate 25 spikes or more between
the times when it starts to respond to wind and when the cockroach starts
to move. By stimulating single giants to generate extra spikes during
responses to wind, Liebenthal, Uhlman & Camhi (1994) showed that the
most critical parameter in the response is the number of spikes in left com-
pared with right giants. It is, therefore, not possible to equate a single spike,
or any other single event, with a distinct decision to initiate an escape turn
in the same way as a spike in a Mauthner neuron or crayfish lateral giant
marks the decision to initiate the startle responses. No single giant inter-
neuron responds exclusively to air currents from the left or from the right,
yet the cockroach always makes quite accurately oriented turns that take it
away from a source of danger. Control of the activity in leg motor neurons
must, therefore, involve quite sophisticated mechanisms to combine
signals that originate in different giant interneurons (Levi & Camhi, 1996).
3.11 Conclusions
Behaviour is initiated as a result of a decision-making process in the central

nervous system. Startle behaviour provides simple and dramatic examples
of this, which illustrate how it is possible to trace the orderly flow of infor-
mation through pathways of interconnected neurons. These examples
enable us to understand how a specific behaviour pattern is triggered and
executed, and so provide some general lessons about the role of nerve cells
in behaviour.
Startle behaviour usually begins with the stimulation of specific groups of
sensory neurons, and these in turn excite one or more orders of interneu-
ron, which filter the sensory input. The filtered information is then inte-
grated by interneurons that act as a switch or trigger, determining whether
or not that particular behaviour is initiated. In the first two examples dis-
cussed here, this executive function is concentrated in a single giant
neuron, and this yields an extremely rapid response. But in cockroach
escape, the executive function is spread more diffusely over a group of
interneurons and this permits a more finely tuned response.
One important aspect of the way these neuronal circuits work is that both
inhibition and excitation play a vital role. The pattern of inhibitory connec-
tions is essential for ensuring that the behaviour is executed efficiently. This
is seen clearly in the way that motor systems that would generate move-
ments that are incompatible with a startle response are inhibited, such as
the inhibition of the extensor system during flexion of the crayfish
abdomen and the mutual inhibition between left and right Mauthner
neurons. A second role for inhibition is in switching off sensory systems
that are likely to be stimulated by movements originating from the startle
response itself. This prevents undesirable consequences such as triggering
fresh startle responses and habituation in sensory pathways. Finally, it is
common for an excitatory circuit that triggers a startle response to inhibit
itself, after a delay, which prevents fresh activity from being triggered until
the initial movement has gone to completion. The effectiveness of this
inhibitory control is often ensured by having inhibitory input at two or
three separate points in a given neural pathway.
Another important aspect of these neuronal circuits is that decisions are
made in the nervous system by the integration of synaptic potentials in
passive cell membranes. This is clearly illustrated by the way in which the
Further reading 75
crayfish lateral giant and the Mauthner neuron collect and sum synaptic
potentials that arise from sensory input on their dendrites. At each point in
the system, it is the properties of the synaptic potentials that determine the
outcome. The output of each neuron in a circuit depends on the way in
which inputs to it are combined.
Although they play a crucial role in the decision-making process, the
giant interneurons described above do not directly drive all the subsequent
events making up the escape behaviour. They excite only a small number of
neurons that are involved in co-ordinating motor activity; subsequently,
other neurons are activated by excitation from other, more slowly conduct-
ing interneurons and by feedback pathways from sense organs. The full
behaviour pattern is thus produced by co-ordinated activity in many parts
of the central nervous system, only some of which are directly driven by
triggering interneurons. This arrangement results in a behaviour pattern
that can greatly outlast the original sensory response that triggered it.
Startle behaviour is unusual in the simplicity of the decision-making
process and in the way that, once triggered, it completely overrides other
behaviour. But it is these features that have made it possible to bridge the
gap between sensory input and motor output. In much neuroethological
research, this gap seems to be unbridgeable, at least using currently avail-
able techniques.
Further reading
Eaton, R.C. ed, (1984). Neural Mechanisms in Startle Behavior. New York:Plenum
Press. This is a collection of several articles on startle behaviour in many
different types of animals; it is a good review of the state of research at the time
it was written.
Faber, D.S., Korn, H. & Lin, J-W. (1991). Role of medullary networks and postsyn-
aptic membrane properties in regulating Mauthner cell responsiveness to
sensory excitation. Brain Behav Evol 37, 286–97. This paper reviews the
physiology of the sensory input side of the Mauthner neuron. Other papers on
this neuron are in the same issue of the journal.
Hoy, R.R. (1993). Acoustic startle: an adaptive behavioural act in flying insects. In
Biological Neural Networks in Invertebrate Neuroethology and Robotics, pp.
139–58 ed. R.D. Beer, R.E. Ritzmann & T. McKenna, New York: Academic Press.
A favourite subject for neuroethologists for many years has been the way many
insects can react to the calls of hunting bats.
4 Capturing sensory information
4.1 Introduction
Most animals are active organisms and need up-to-date information about
their environment if they are to behave appropriately. Much information is
potentially available in the many forms of energy and chemicals that
impinge on the surface of the organism and act as stimuli. An animal must
be able to detect the various forms of energy and to sort them all out, a job
that is carried out by its sense organs, which act as instruments monitoring
stimuli coming in from the environment. Sense organs are thus an animal’s
mechanism for gathering up-to-date information, and as such it is hard to
exaggerate their importance in behaviour.
Clearly, a monitoring instrument is useful only if it measures one partic-
ular form of stimulus; having an instrument that responded indiscrimi-
nately to all forms of stimuli would be nearly as uninformative as having no
monitoring facility at all. Hence, one of the most fundamental properties of
sense organs is selectivity. Each sense organ contains specific receptor
cells that are tuned to be sensitive to one particular stimulus. In many sense
organs, the receptor cells are also nerve cells, having axons that convey
their information to the central nervous system, and receptor cells of this
kind are, therefore, often called sensory neurons. In other sense organs,
such as the eyes of vertebrates and insects, the receptor cells do not have
long axons themselves but instead make synaptic contact with separate
nerve cells that send trains of spikes to the central nervous system.
The particular form of stimulus to which a sense organ responds deter-
mines its sensory modality. The diversity of sensory modalities has been
recognised since the time of Aristotle, who distinguished the five primary
senses: sight, hearing, touch, smell and taste. These five are still the ones
recognised in everyday life, but many more modalities have been discov-
76
Basic receptor mechanisms: the campaniform organ 77
ered by physiologists, including unfamiliar ones such as the ability to detect

electric or magnetic fields. However, sensory modalities can be separated
broadly into three basic categories. First, there are chemoreceptors, sense
organs that are stimulated by chemical ions or molecules in various forms,
either as gases or in solution. These include not only the more familiar
senses of smell and taste but also less conspicuous receptors monitoring
such things as oxygen or carbon dioxide concentration. Second, there are
mechanoreceptors that are stimulated by some form of kinetic energy.
These include many sense organs that monitor internal functions, such as
muscle tension or joint position, as well as senses such as touch, balance
and hearing. Third, there are photoreceptors, which respond to electro-
magnetic energy in the form of photons.
Sense organs falling within these main categories are found throughout
the major animal groups. They exhibit a diversity of form as great as that of
the animals of which they form a part, even within a single modality. In
spite of this diversity, there are some basic receptor mechanisms that are
common to all the sensory modalities. Here, the operation of basic sensory
mechanisms is illustrated by a simple mechanoreceptor, the campaniform
organ of insects.
4.2 Basic receptor mechanisms: the campaniform organ
Campaniform organs detect strains in the cuticular exoskeleton of insects

and so provide information about loading on different parts of the body.
They occur in groups at those parts where stresses are most likely to be felt,
particularly near joints between segments (Fig. 4.1a, b). Mechanoreceptors
that detect the relative positions of, or stresses and strains in, different parts
of the body are known collectively as proprioceptors. The muscle receptor
organ in the crayfish abdomen (see section 3.4) is another example of a pro-
prioceptor, and the way that a proprioceptor influences the output from
motor neurons often fits well the description of feedback in a control system
(see section 1.6). The campaniform organs that occur near the joints of cock-
roach legs have been studied in detail, and play a significant role in normal
walking movements. They detect the strains that occur when the leg is being
applied to the ground and used to thrust the animal forwards. The response
of the campaniform organs initiates a reflex that prevents the leg being lifted
and swung forwards while it is still bearing a load (Zill & Moran, 1981).
Figure 4.1 A simple sense organ: the campaniform organ of insects. (a) A
leg of a cockroach (Periplaneta), showing the location of the groups of cam-
paniform sensilla (arrow) on the upper surface of the tibia. (b) The surface
of the tibia viewed from the direction of the arrow in (a), showing the distri-
bution of the caps of the campaniform organs. The arrow indicates the lon-
gitudinal axis of the leg, pointing towards the body. (c) Transverse section
through a campaniform organ, showing the relation of the sensory neuron
to the cuticular cap and to the accessory cells, as revealed by electron
microscopy. (a and b redrawn after Zill & Moran, 1981; c redrawn after
Gnatzy & Schmidt, 1971.)
The ability of a campaniform organ to detect strains arises from its spe-
cialised construction, which consists of a single sensory neuron coupled
with certain accessory cells and structures (Fig. 4.1c). Opposite the axon,
the sensory neuron bears a short process, the dendrite, with a specialised
ending consisting of a modified ciliary structure. This structure contains
numerous microtubules and is connected to a specialised region of cuticle
called the cap. Seen from the external surface, each cap is oval in shape,
having a central dome and a surrounding depressed region. In section, it
can be seen that the cuticle of the cap is thinner than elsewhere and that
the centre of the dome fits tightly around the tip of the sensory neuron. It
seems clear from this structural arrangement that the cap must act as a
mechanism by which the strains in the cuticle are transmitted to the ending
of the sensory neuron – it must act as a mechanism for coupling the stimu-
lus energy to the receptor cell.
The electrical activity of a campaniform organ can be studied using the
kind of arrangement shown in Fig. 4.2a. A cockroach leg is secured and
extracellular electrodes are arranged to record activity in the main leg
nerve. When bending forces are applied to the leg with a probe, a campan-
iform organ responds with a vigorous discharge of spikes (Fig. 4.2b). The
same response can be obtained by applying a much finer probe, with a tip
diameter of only 1 to 2 ␮m, directly to the cap of a single campaniform
organ. Furthermore, if that particular cap is ablated with a sharp needle, the
discharge is no longer recorded in response to a bending force applied to
the leg, which confirms the role of the cap in coupling the stimulus to the
receptor cell.
The sensory neuron acts as a biological transducer, converting mechani-
cal energy that bends the leg into electrical energy that generates the spikes.
This basic process of transduction is a defining feature of all receptor
cells. The precise mechanism of transduction of the sensory stimulus is
only partly understood in most cases, mainly because receptor sites are
small and the events occur rapidly. But it is not hard to see that in a mech-
anoreceptor like the campaniform organ, mechanical distortion of the
receptor cell membrane caused by the stimulus could open ion channels
in the membrane. This will allow particular ions to pass through the cell
membrane, resulting in a change in membrane potential. In campani-
form organs, transduction probably occurs in the ciliary region of the den-
drite. The dendrite lies in a lumen which almost certainly contains a special
Figure 4.2 Physiological recording from the tibial campaniform organs.

(a) Arrangement for recording spikes in response to controlled stimuli. The
stimuli are applied to the distal end of the tibia with a probe attached to a
piezoelectric crystal, which bends when a voltage is applied to it. Two fine
pins act as electrodes for recording from the leg nerve, and are attached
to an amplifier before the signal is displayed on an oscilloscope screen.
(b) Response to bending the tibia in a ventral direction. The upper trace
is a record of the voltage applied to the crystal to bend the tibia, and the
lower trace is the extracellular recording of the sensory spikes. (a redrawn
after Spinola & Chapman, 1975; b modified after Zill & Moran, 1981.)
high-potassium solution, a feature that is commonly found in mechano-

sensitive organs, including mammalian ears. The accessory cells are prob-
ably responsible for regulating the composition of this lumen.
The change in membrane potential that occurs at the site of transduction
is known as the receptor potential. In another mechanoreceptor of the
cockroach leg, a special spine near the knee, the receptor potential has
been measured directly with an intracellular microelectrode (Basarsky &
French, 1991), but less direct means have been used to record campaniform
organ receptor potentials. The result of such an experiment is shown in Fig.
4.3, where the record of membrane potential has been redrawn as if it were
recorded intracellularly. In this example, the fine stimulus probe applied to
the cap was driven sinusoidally, which is an effective way to stimulate a
receptor cell because the amplitude and frequency of the sine wave are
Figure 4.3 The response of a single campaniform sensillum to sinusoidally

modulated indentation of its cap with a fine probe. The top trace is the
monitor applied to the probe, with upward movement indicating increased
force. The bottom trace shows spikes recorded with pin electrodes, as in Fig.
4.2. The middle trace indicates the appearance of the receptor potential as if
it had been recorded with an intracellular electrode; upward movement of
the trace indicates depolarisation of the cell membrane. This trace has been
redrawn from a record of the receptor potential recorded extracellularly.
(Modified after Mann & Chapman, 1975.)
readily varied and the resulting responses lend themselves to quantitative

analysis.
The sinusoidal stimulus elicits a large receptor potential, which follows
the imposed force closely: as the force indenting the cap increases, the cell
becomes increasingly depolarised; and as the force decreases the cell
becomes repolarised (middle trace in Fig. 4.3). Spikes are generated by the
depolarisation and are superimposed on the basic shift in membrane
potential; they are propagated along the axon to the central nervous
system. The change from receptor potential to spikes is termed coding
because the analogue signal of the receptor potential is converted into the
digital code of spike frequency, which is used for long-distance communi-
cation in the nervous system. The conversion of receptor potential into
spikes occurs at the place where the axon originates from the cell body, the
spike-initiation zone. The mechanism by which it occurs is the same as that
by which spikes are triggered by summation of excitatory synaptic poten-

tials in other neurons (see Chapter 2). Transduction of an increasing stimu-
lus will cause a greater current flow and so depolarisation of the membrane
at the spike-initiating zone will occur more rapidly. This means that the
time interval between successive spikes will be less, and so the spike fre-
quency will be higher.
Inspection of Fig. 4.3 shows that, although the receptor potential mirrors
the changing stimulus rather faithfully, the train of spikes reflects the stim-
ulus pattern less satisfactorily. The axon must employ spikes for communi-
cation because the graded receptor potential could not travel far, being
limited by cable properties of the neuron. However, the change from an
analogue to a digital code involves a certain loss of information because the
use of a frequency code means that both encoding and decoding the signal
require time – a particular stimulus strength is immediately represented by
a particular receptor potential, but requires at least two separate spikes to
encode it. The closer together the spikes are in time, the stronger the stim-
ulus.
In addition to the intrinsic limitations of coding, there is another process
at work in shaping the response of the sensory neuron. Looking back to Fig.
4.2b, when a constant bending force is first applied, there is a vigorous
response to the onset of the stimulus but spike frequency then soon
declines. Similarly, in Fig. 4.3, spike frequency initially rises with increasing
stimulation, but begins to decline even before the stimulating force has
reached its maximum value. This pattern of response, whereby a receptor
cell responds vigorously to a changing stimulus but soon ceases to respond
to a steady stimulus, is known as adaptation and is a property of most
sensory cells. Generally, changes in stimulus strength are more significant
than constant stimuli, and adaptation is a means of enhancing the detec-
tion of changing stimuli.
A sensory neuron that is much more sensitive to a rapid change in the
stimulus is said to be quickly adapting or phasic; one that is sensitive to
slowly changing or maintained stimulation is said to be slowly adapting or
tonic. Naturally, phasic and tonic receptor cells provide an animal with
information about different types of situation. A phasic receptor provides
information about rapidly changing events, whereas a tonic receptor keeps
an animal informed about a steady background situation. Many sensory
neurons adapt to a stimulus in a way that represents some combination of
Summary of basic mechanisms 83
phasic and tonic properties. The campaniform organ is an example of this

because it is most sensitive to a rapid increase in applied force, but it also
exhibits a declining response to a maintained stimulus.
The tibial campaniform organs in the cockroach leg are arranged in two
subgroups with mutually perpendicular cap orientations (see Fig. 4.1b).
The longitudinal axes of the caps of the distal group lie parallel to the lon-
gitudinal axis of the tibia. In physiological recordings, it is found that cam-
paniform organs in the distal group respond only to downward bending of
the tibia and those in the proximal group respond only to upward bending.
This pattern of response can be understood by thinking of the tibia as a
cardboard tube and considering the effects of bending in one direction and
then in the opposite. The upper surface of the tibia, where the campani-
form organs are located, will undergo transverse compression when the
tibia is bent down and longitudinal compression when it is bent up. Hence,
the fact that the distal campaniform organs respond to downward bending
and the proximal ones to upward bending shows that each sensory neuron
is sensitive only to transverse compression of its cap (Zill & Moran, 1981).
The tibial campaniform organs, therefore, act as directionally sensitive
strain gauges and the two groups of campaniform organs below the knee
provide the animal with information about opposing forces to which the
tibia is subjected during walking. The organs are most sensitive to bending
forces in the plane of the femoro-tibial joint, whether these are produced by
external factors or by contraction of the muscles that move the tibia. Forces
that bend the leg at unnatural angles or other forces, such as those twisting
it about its longitudinal axis, produce little or no response in any of the
campaniform organs. The local forces of compression that best excite a
particular campaniform organ constitute its receptive field, a concept
previously introduced in the description of the wind-sensitive giant
interneurons of the cockroach (see Chapter 3).
4.3 Summary of basic mechanisms
The basic processes of sensory action, illustrated by the campaniform

organ, can now be reviewed by reference to Fig. 4.4. The incoming energy is
coupled to the sensory ending by means of highly specialised accessory
structures and is transduced in the sensory neuron terminal into a receptor
potential. This is a smoothly graded potential which spreads passively
Figure 4.4 Summary of the events that occur when a sensory receptor
responds to a stimulus. The normal sequence of processes is shown by the
arrows in the centre of the figure, and the structures in which these
processes occur are shown on the right. The arrows on the left indicate the
extent to which these events are specific to a particular modality (details in
text). Where the receptor cell works without trains of spikes, the receptor
potential spreads passively down the axon to the synapse with second-order
neurons.
along the neuron membrane. When it spreads to a special region at the start
of the axon called the spike-initiation zone, the receptor potential is con-
verted into a digital code of spikes. The spikes propagate along the axon
towards the central nervous system.
As indicated on the left of Fig. 4.4, the more central processes, such as
coding, are not specific to the particular type of receptor. Indeed, the
coding of passive potentials into spikes is not specific even to sensory
neurons, but is common to all neurons that employ trains of spikes. The
more peripheral processes are modality-specific: the transduction process
determines the sensory modality, and the coupling process, effected by the
accessory structures, narrows down the modality and sharpens the recep-
tive field.
Essential properties of eyes 85
4.4 Essential properties of eyes
The same basic processes that have been illustrated in a simple mechano-
receptor underlie the operation of more complicated sense organs. There is
no sense organ more appropriate to the study of animal behaviour than one
of the arrays of photoreceptors, commonly called eyes. The surface of the
Earth is continuously bathed in light, and the speed with which it travels
means that light can be a source of almost instantaneous information
about events that are occurring at a distance. Hence, it is not surprising that
eyes have evolved in many different animal groups, and that vision plays a
crucial role in guiding behaviour patterns in many species.
Two fundamental factors influence the design of any sense organ. First,
there are the physical constraints imposed by the particular modality
involved. In the case of eyes, the physical properties of light set the limits to
performance. Features such as the overall size of the eye, the design of
lenses, and the arrangement of light-sensing elements will partly be deter-
mined by the physical properties of light, just as in any other optical instru-
ment. The second factor that influences the design of a sense organ is the
need to gather information that is relevant to the animal’s particular way of
life. Thus, animals that are active at night time will have eyes of a construc-
tion different from those of animals that are usually only active in daylight.
The design of an animal’s sense organs must form an integrated part of its
overall adaptive strategy.
Eyes reach their highest level of complexity amongst the vertebrates and
arthropods, especially insects, as well as in cephalopod molluscs. The eyes
of insects are quite different in their design from those of vertebrates.
Nevertheless, both types of eye have to deal with the same problems in con-
verting patterns of change in light intensity into signals that are useful for
guiding behaviour. The same physical constraints influence the design of
the optics of both types of eye, and there are many similarities in the way
the two types of eyes operate. In photoreceptors, the same basic sensory
processes that have been described in a mechanoreceptor occur. Energy
from the environment is focused onto a photoreceptor by specialised
accessory structures (lenses) and it is then transduced into a receptor
potential, which typically adapts fairly rapidly. Most photoreceptors do not
have long axons, and the receptor potential itself regulates the transmission
of signals across synapses to the next layer of cells, the second-order
neurons. Here, the compound eyes of insects are used to illustrate the prin-
cipal features of eyes and the properties of the receptor potential.
4.5 Design features of eyes
The lens of an eye gathers light from the environment and usually focuses
it as a crisp image onto photoreceptor cells. The greater the amount of light
that one photoreceptor receives, the easier it is for it to provide an accurate
measure of light intensity. This is particularly important for animals that are
active in dim lighting conditions, when the amount of light available is
small. There are two main ways of maximising the amount of light that a
photoreceptor receives: one is for the eye to have a large lens, which
increases the amount of light caught from each part of the visual field; the
other is for the photoreceptors themselves to be large, so that each captures
light over a relatively large area of the image. But having large photorecep-
tors is not an advantage if the animal needs to extract detail from the image
on its retina, because what is required is for neighbouring photoreceptors
to sample small but distinct areas. The greater the number of sampling sta-
tions that cover one area of the image, the greater the amount of detail that
can be resolved in that area. The way in which an eye is constructed, there-
fore, involves compromise. The compromise between the need for large
photoreceptors, which have good ability to capture available light, and
small photoreceptors, to enable the image to be examined in detail, is the
most basic. The ability to measure light levels is called sensitivity, and the
ability to discriminate fine detail in an image is called resolution. The two
are quite distinct requirements, and it is difficult to improve the perfor-
mance of an optical instrument for one of them without deterioration in
the other.
Insects often depend heavily on vision, which they achieve by means of
compound eyes. A compound eye consists of many discrete optical units,
each of which has its own lens that focuses light from a small part of the
environment onto its own small group of photoreceptors (Fig. 4.5). Each
unit is called an ommatidium, and its lens is often referred to as a single
facet. Large dragonflies have almost 30 000 ommatidia in each eye, but
most insects have fewer than this – the housefly has 3000 and Drosophila
has 700. Compound eyes give the animal a very wide field of view; a fly can
see in almost every direction from its head. The size of eye that an insect
Design features of eyes 87
Figure 4.5 The arrangement of cells in the ommatidia of two diurnal

insects. (a) A locust (Locusta) ommatidium. Transverse sections of the
ommatidium at two levels (on the right) show how the rhabdomeres of indi-
vidual retinula cells form a single, central, light-collecting rod, the rhabdom.
(b) A blowfly (Calliphora) ommatidium. Here, the rhabdomeres of different
photoreceptors within an ommatidium remain separate. Seven of the
photoreceptors are numbered in the cross-section, on the left. Also shown
are pigment cells that surround each ommatidium. (a from Wilson, Garrard
& McGiness, 1978; copyright 1978 Springer-Verlag; b from Hardie, 1986;
reprinted with permission from Trends in Neuroscience; copyright © 1986
Elsevier Science.)
can carry around is obviously limited and, in order to achieve good resolu-
tion, the size of each ommatidium must be small so that as many as pos-
sible can be packed into the eye to provide a reasonable sampling density.
However, as the size of a lens is reduced, the image that it produces
becomes increasingly blurred, and there is a limit to how small individual
ommatidia can be.
The reason why small lenses produce blurred images is that when light
passes through an aperture or is bent by passing from one medium to
another, some waves are scattered – a property called diffraction. Because
of this scattering, light that passes through one part of a lens will spread out
and interfere with light that passes through another part of the lens. This
mutual interference causes blurring in the image that the lens projects onto
its focal plane. The image of a small spot of light is focused by a lens to a
diffuse spot called an Airy disk. The width of the Airy disk decreases as the
diameter of the lens increases, and so an eye that has a wide lens with a
large aperture will be able to work with a sharper image than an eye that has
a small lens.
If an eye is to make effective use of a sharply focused image, the width of
its photoreceptors must be matched to the width of the Airy disk. If the
photoreceptors were wider than the Airy disk, two closely spaced images
could not be distinguished from each other. However, interference effects
set a lower limit to the width of the rhabdom, the photoreceptor element
that catches the light which strikes its end, acting as a light guide. As the
diameter of a light guide is reduced towards the wavelength of light, inter-
ference effects frustrate total internal reflection and an increasing amount
of light travels outside the light guide. The parts of photoreceptors that
capture light energy and transduce it into an electrical signal are cylindrical
light guides. If a large amount of light is lost from these structures, the per-
formance of the eye is degraded because light travelling outside the recep-
tor element will not be captured by the receptor cell and may even cause
cross-talk by entering a neighbouring cell. In view of this, light guides made
of living tissue cannot usefully be less than 1 ␮m in diameter. Thus,
diffraction sets a lower limit both to blurring of the image and to the size of
the receptor elements.
In honey bees, which are typical of insects that fly on bright days, most
facets are about 25 ␮m across, and collect light from a cone with an angle
of a little over 1°. To resolve two small spots of light as separate rather than
one larger spot, a bee would have to use three ommatidia – one for each
spot, with an un-illuminated ommatidium between them. Therefore, bees
can distinguish between two objects that are about 3° apart. Compound
eyes of species that are active in dim light conditions often have ommatidia
that are larger than would be expected to give the best balance between
sampling frequency and a sharply focused image (Snyder, Stavenga &
Laughlin 1977). Usually, in dim light conditions, photoreceptors also collect
light not just from the lens of their own ommatidia, but from surrounding
lenses as well, an optical arrangement that is called superposition. Insects
and crustacea that are active in bright daylight usually operate with the best
resolution that a compound eye can deliver; this type of eye is called an
apposition compound eye, and ommatidia are shielded from their neigh-
bours so that each works quite independently.
Design features of eyes 89
Some diurnal insects depend on vision for the localisation of prey or

sexual partners. A good example is the mantis Tenodera, which depends on
vision to locate its prey. In order to achieve good resolution for a task such
as this, there is a need to dedicate as many ommatidia as possible to sample
the visual environment, and each individual ommatidium must have a
crisp image of a small part of the environment. This means that each
ommatidium must have a lens that is as large as possible. The size that an
insect’s eye can grow to is clearly limited, however, and there is a compro-
mise between the number of ommatidia and their size in a particular region
of the eye. The mantis eye has evolved so that only a small portion of each
eye has the high resolution required to examine potential prey. This region,
where resolution is enhanced, is called a fovea. The mantis fovea is func-
tionally equivalent to the more familiar fovea of vertebrate eyes.
The fovea of a mantis eye is a small region at the front of the eye (Fig.
4.6a). Individual ommatidia here have lenses that are 50 ␮m in diameter,
larger than elsewhere in the eye, and rhabdoms that are 1.5 to 2 ␮m across,
narrower than elsewhere in the eye, a specialisation to maximise resolution
by capturing light over a relatively narrow cone of acceptance. Also, the
angle between adjacent ommatidia (0.6°) is narrower than elsewhere in the
eye (Fig. 4.6b), and so the surface of the eye is less curved here than else-
where. These structural features are consistent with physiological proper-
ties of ommatidia, measured by recording the responses of photoreceptors
with intracellular electrodes. Intracellular recordings show that individual
ommatidia accept light from a cone of angle 7°, and this is smaller than the
acceptance angle for ommatidia measured elsewhere over the eye. In bright
light, acceptance angles, measured in electrophysiological experiments, are
very close to the angles between adjacent ommatidia, measured anatomi-
cally. In dim light, ommatidia in the fovea generate smaller responses than
those elsewhere in the eye, which means that they are less sensitive. This is
consistent with their narrow rhabdoms. The mantis eye is thus constructed
so that resolution is increased at the expense of sensitivity in one region of
the eye, the fovea, and is correspondingly reduced in the rest of the eye.
When a prey-like object appears in its peripheral field of view, a mantis
responds with a rapid movement of the head that brings the object’s image
into the fovea (see Fig. 1.7 p. 16). Subsequent movements of the prey are fol-
lowed by tracking movements of the head, which hold the image of the prey
in the fovea while the body is aligned and brought into range for a raptorial
Figure 4.6 The fovea of the mantis (Tenodera) eye. (a) Anterior view of the
head; the broken circle within each eye indicates the region of inter-omma-
tidial angles that constitutes the fovea. (b) Facet diameters (䊉) and inter-
ommatidial angles (䉱) plotted against position in the eye for the row of
facets indicated in (a). Note how facet diameters increase as inter-omma-
tidial angles decrease. The acceptance angles of individual ommatidia
follow the curve for inter-ommatidial angles very closely. (Redrawn after
Rossel, 1979.)
strike. There is a large area of overlap in the visual fields of the two eyes in
the mantis, so that this insect can use binocular cues to estimate prey dis-
tance. The foveae of the two eyes are included in the binocular field, and the
central axes of the left and right foveae intersect in the sagittal plane about
4 cm in front of the head. It is evident from watching the prey-catching
behaviour that the function of the foveae is close examination of spatial
detail of potential prey, while the peripheral eye is chiefly responsible for
the detection of novel objects.
In absolute resolving power, insect foveae are about an order of magni-
Photoreceptors and the receptor potential 91
tude poorer than human peripheral vision and two orders of magnitude
poorer than human foveal vision. However, a major function of high resolu-
tion is to enable animals to detect objects at a distance. Larger animals
need to detect objects at greater distances than small animals. A small
insect may need to react to objects that are only a few centimetres away,
whereas the appropriate reaction distance for a large primate may be
several metres. The force of this point can be made by multiplying the
angular separation between visual receptors by the height of the animal,
and then comparing the values obtained from a number of active species.
Expressed in this way, resolution turns out to be remarkably uniform over a
wide range of animals, from Drosophila to Homo. Thus, compound eyes
provide their small owners with resolving power that is as good as that of
vertebrates when considered in terms of biological needs rather than phys-
ical ideals.
4.6 Photoreceptors and the receptor potential
Within an ommatidium, the lens focuses light onto the tips of photorecep-
tor cells, which are called retinula (‘little retina’) cells (see Fig. 4.5). There
are usually eight or nine retinula cells per ommatidium. A retinula cell is
elongated, with its longitudinal axis parallel to that of the ommatidium.
Outermost is the cell body, and it includes a specialised structure, the rhab-
domere, where phototransduction occurs. The cell membrane is folded
into very regular finger-like projections, or microvilli, in a rhabdomere. The
microvilli contain the photopigment, a rhodopsin molecule, that absorbs
light and initiates the process of phototransduction. In most insects,
including bees, locusts and dragonflies, the rhabdomeres of all the retinula
cells of an ommatidium are adjacent to each other, creating a central rod-
like structure, the rhabdom (see Fig. 4.5a). In most insects, all the retinula
cells of one ommatidium share the same field of view, and all converge on
the second-order neurons that are beneath their ommatidium. However,
advanced dipteran flies, such as blowflies, have a different structure, where
individual rhabdomeres remain separate (see Fig. 4.5b), and each photore-
ceptor of a rhabdom looks in a slightly different direction.
Light has a dual physical nature. The way in which it is diffracted when it
passes through lenses and apertures is a characteristic of its wave-like
properties. It also consists of discrete packets of energy, called photons. The
amount of energy in a photon depends on the colour, or wavelength, of its

light. Light is actually radiation that occupies a very narrow part of the
electromagnetic spectrum, the part that animals can make use of with their
eyes. Infrared radiation has a longer wavelength than visible light, and has
insufficient energy to trigger phototransduction. At the other end of the
visible spectrum, the amount of energy in an ultraviolet ray is so high that
it is potentially damaging to biological molecules.
Photoreceptors are capable of responding to single photons of light, if the
eye has been left in darkness for some time to maximise its sensitivity. In an
insect, each photon gives rise to a discrete, depolarising potential, often
called a bump because of its shape. Bumps are usually 1–2 mV high in intra-
cellular recordings (Fig. 4.7a). In absolute darkness, bumps are very rare,
and the evidence that each bump represents the absorption of a photon
comes from a comparison between the statistical properties of the arrival of
photons from a very dim light source at a retinula cell and of the occurrence
of bumps in the cell. About 60 per cent of the photons that arrive at a facet
generate a bump (Lillywhite, 1977).
In very dim light conditions, photoreceptor potentials consist of a series
of discrete photon bumps. As light intensity increases, the bumps fuse
together, and the photoreceptor response to a change in light intensity
becomes much more smooth (Fig. 4.7b). In daylight, an insect retinula cell
receives millions of photons a second. If each bump generated a signal of 1
mV, this would lead to a signal of about 1000 V in the photoreceptor, which
it is clearly incapable of generating. When exposed to light, the sensitivity of
the photoreceptor response decreases quite dramatically, so that each
additional photon generates a smaller response. This is called light adapta-
tion. Conversely, adaptation to decreasing levels of ambient illumination,
which causes sensitivity to increase, is called dark adaptation. Light and
dark adaptation enable photoreceptor responses to be appropriate to the
average ambient light intensity. A number of different mechanisms are
involved, including changes in pigment-containing cells that regulate the
arrival of light at a rhabdom (analogous to the pupil of vertebrate eyes), and
changes in the sensitivity of transduction. It is important for photorecep-
tors to be able to adjust their sensitivity to ambient lighting conditions,
because these can vary enormously. In sunlight, each photoreceptor
receives about 40 million photons per second, and this number decreases
to 40 000/s inside a room and 40/s in moonlight. Because natural light
Figure 4.7 Coding of light intensity by an insect photoreceptor cell. (a–c)

Recordings from a dragonfly (Hemicordulia) retinula cell of the responses to
three different levels of light stimuli, delivered in darkness. Extremely dim
light (a) elicits two single photon bumps; moderately dim light (b) elicits a
steady receptor potential in which small fluctuations, caused by the random
nature of photon arrival, can still be discerned. Bright light (c) elicits a
receptor potential with a sharp, initial peak, followed by a more sustained
level. Notice how the voltage calibration changes in (a) to (c). (d)
Intensity–response curves for a dragonfly retinula cell. The amplitude of the
receptor potential is expressed as a percentage of the maximum response
recorded, and light intensity is expressed on a logarithmic scale. The curve
on the left (•-•-•) shows the peak size of the receptor potential to flashes of
increasing intensity in a dark-adapted cell. The three curves to the right
(䉱-䉱-䉱) are plots for cells adapted to particular levels of background illumi-
nation, indicated by the arrows. The row of dots connecting the bottoms of
the four plots shows the steady-state amplitude of the receptor potential
during different levels of sustained background illumination. (Recordings
from Laughlin, 1981; copyright © 1981 Springer-Verlag.)
intensity varies so much, it is convenient to use a logarithmic scale to

express it. On such a scale, each unit usually represents a tenfold change in
light intensity. A change in intensity from 10 to 100 occupies the same space
on such a scale as a change from 1000 to 10 000 (10 is 101, or log10 10 is 1; and
100 is 102, or log10 100 is 2, and so on).
One effect of light adaptation is that, at quite moderate light intensity, the
response to a step increase in light consists of a sharp, initial depolarising
peak, which quickly drops to a more sustained or plateau receptor poten-
tial (Fig. 4.7c). If the increase in light is maintained, the photoreceptor
potential declines very gradually, but stabilises over a timescale of a few

minutes. This means that a photoreceptor has both tonic and phasic char-
acteristics. The sustained receptor potential that is produced in response to
a long-lasting change in the overall level of illumination is a tonic charac-
teristic, and the initial peak response, which emphasises a change in light
intensity, is a phasic characteristic.
The second effect of adaptation is apparent in the way that information
about light intensity is encoded as a particular value of receptor potential.
This pattern of adaptation is usually examined quantitatively by plotting
intensity–response curves, based on intracellular recording of the receptor
potential (Fig. 4.7d). At the start of an experiment, the eye is kept in the fully
dark adapted state and the sizes of the peak responses to steadily brighter
flashes of light are recorded. The time intervals between successive flashes
are kept long enough to ensure that the eye remains fully dark adapted
throughout the experiment. The non-linear nature of bump summation is
reflected in the sigmoid shape of the curve of peak receptor potential
against a logarithmic scale of light intensity. For low intensities of stimulus,
there is a gradual rise in the size of the receptor potential as stimulus inten-
sity increases. The maximum intensity that the receptor is capable of sig-
nalling is about three log units (1000), on the horizontal axis of the graph.
At this intensity, the response has saturated. However, over a range of
intensities of three log units (a thousandfold change in absolute intensity)
the size of the response is proportional to the logarithm of light intensity.
The next stage in the experiment is to superimpose the stimulus light on
a steady, background level of illumination. The responses to superimposed
test flashes are then recorded in the same way as before. Fig. 4.7d shows
that background illumination shifts the intensity–response curve along the
intensity axis, and the size of the shift depends on the intensity of the back-
ground light. At low intensities at which adaptation in the response wave-
form is small, the shift is also small. However, at higher intensities, the shift
produced becomes larger. Hence, the range of intensities to which the cell
responds is shifted to match the level of background illumination. This shift
in the curves represents a loss in sensitivity: more light is now needed for
the photoreceptor to generate the same voltage response. The curves keep
approximately the same shape, indicating that the relation between test
flash intensity and peak response remains much the same, and response
amplitude changes in proportion to the logarithm of stimulus intensity. The
extent of adaptation varies between types of eyes, and a very good illustra-
tion of this comes from recent research on adaptation in the eyes of
different species of dipteran flies. This research shows that the physiology
of photoreceptors is well correlated with the ecology and lifestyle of a par-
ticular species (Box 4.1).
Box 4.1. Fast and slow photoreceptors
Tipula, a slow-flying, nocturnal cranefly, and Sarcophaga, a fast-

flying, diurnal fleshfly, are two of the 20 species of dipteran fly sur-
veyed by Laughlin & Weckström (1993). The intracellular recordings in
the figure are responses to 0.5s pulses of light delivered to dark-
adapted eyes. The receptor potential in Tipula rises relatively slowly,
and does not adapt during the light pulse, whereas that of Sarcophaga
rises rapidly and, for the brighter pulses, adapts by decaying quickly
from an initial peak to a more sustained level. Tipula is more sensitive
to light than Sarcophaga, as shown by the intensity versus response
curves and by the larger bump responses to individual photons in dim
light. Photoreceptors of Sarcophaga are fast, and able to respond
quickly to fluctuations in light as the insect flies around among twigs
and leaves. Those of Tipula are too slow to help a fast-flying animal
avoid colliding with objects, but are good at detecting light and dark at
night time. Sarcophaga’s photoreceptors are fast because they contain
a special potassium channel. The lifestyle of Tipula means it does not
need fast photoreceptors; in fact, to have them would be a significant
metabolic cost to the animal. (Figure modified from Laughlin &
Weckström (1993), copyright Springer-Verlag.)
It is a common pattern in sensory receptors for the amplitude of the

response to be proportional to the logarithm of stimulus intensity, and this
makes excellent functional sense. First of all, the large range of light inten-
sities to which an eye is exposed in the day-to-day life of an animal is com-
pressed into a manageable scale. Second, responding in this way has the
effect of making equal relative changes in light intensity generate equal
absolute changes in the size of the receptor potential. One log unit on the
light intensity scale is a tenfold change in absolute light intensity, which
causes a change in receptor potential of about 35 per cent of its maximum
amplitude. A tenfold change in light intensity is a very large change, and a
photoreceptor would normally experience much smaller changes in inten-
sity as it scans a natural image. Coding light intensity in this way enables
the eye to recognise different objects because they are distinguishable by
differences in the proportion of light that they reflect. The relative bright-
ness of two objects is called their contrast. Contrasts of objects do not vary
when ambient illumination changes. For instance, the contrast between
the print and the white paper on this page will be the same whether you are
reading the book in a dimly lit room or outside on a beach in bright sun-
light. The black print will actually reflect more light on the beach than the
white paper will when the book is read in the room. The signal that a photo-
receptor generates, therefore, consists of a small, steady depolarising
potential, which depends on the mean or background level of light, plus
fluctuations of a few millivolts caused by viewing objects of different con-
trasts.
The receptor potential is not coded into spikes in an insect retinula cell,
but is conducted by passive spread along the axon. Sometimes, the axon
can be as long as 2 mm, but the cable properties of these axons are such
that little of the signal is lost at the synaptic terminals.
4.7 Conclusions
Sense organs are essentially biological transducers, converting incoming

energy to the changes in membrane potential that are the currency of infor-
mation used throughout the nervous system. One of their key features is
selectivity. Selectivity starts at the level of the receptor cells, which are
essentially biological transducers, converting incoming energy to the mem-
brane potential changes that are used throughout the nervous system. Each
Further reading 97
sense organ is selective for a particular form of physical or chemical stimu-

lus and so acts as a tuned monitoring instrument. Just as in manufactured
instruments, there are physical constraints that set limits to their perfor-
mance. One of the outcomes of these constraints is that an improvement in
one direction cannot be made without detriment in another, which is well
illustrated for the compromise between resolution and sensitivity in the
design of eyes.
For a sense organ to function effectively as a monitoring instrument, the
stimulus energy must be appropriately coupled to the site of transduction
in its sensory receptors. Consideration of both campaniform organs and
eyes shows that coupling structures are highly specialised to enable a sense
organ to operate close to the limit of what is physically possible. The recep-
tive fields of individual sensory neurons are finely adjusted to enable a
group of receptors to pass on the most useful information to the central
nervous system. Thus, in gathering the sensory information that will regu-
late an animal’s behaviour, many events of crucial importance take place at
the coupling stage before any neural signal has been generated.
The result of sensory transduction is the generation of a receptor poten-
tial in each of the sensory neurons that have been stimulated. A common
feature of sensory receptors is that their sensitivity adapts to particular
background levels of stimulation. Adaptation ensures that signals about
changes in stimulus strength are enhanced, and also enables receptors to
operate efficiently over a wide range of stimulus intensities. In some sense
organs, the receptor potential is immediately coded into spikes and the
next stage of processing occurs within the central nervous system. In
others, notably eyes, the receptor potential itself is conducted passively
down a short axon for synaptic communication with the next layer of cells.
In either event, it is the array of receptor potentials in the sense organs that
constitutes the nervous system’s view of the world. All subsequent neuronal
processing depends on the receptor potentials; all later integration is built
on this initial basis.
Further reading
French, A.S. (1988). Transduction mechanisms of mechanosensilla. Ann Rev
Entomol 33, 39–58. An account of different types of mechanosensors in insects.
Nilson, D-E. (1990). From cornea to retinal image in invertebrate eyes. Trends
Neurosci 13, 55–63. A description of the large number of different kinds of com-
pound eyes that are found in crustaceans and insects.
Stavenga, D.G. & Hardie, R.C. eds, (1989). Facets of Vision. Berlin: Springer-Verlag.
A book containing well-written articles on various aspects of the structure and
function of compound eyes; especially relevant to this chapter are the chapters
by M. Land and S.B. Laughlin.
5 Stimulus filtering: vision and motion detection
5.1 Introduction
Sensory systems have evolved to provide information that is particularly

relevant to an animal’s way of life. Sensory neurons have modalities and
receptive fields that are strongly biased in favour of gathering information
that is behaviourally significant for that species. Whatever their bias, sense
organs can pick up large amounts of information about an animal’s envi-
ronment: for example, the photoreceptors of an insect’s eye provide a
point-by-point representation of light levels in the surrounding visual envi-
ronment. Higher-order neurons in a sensory system cope with all this infor-
mation by discarding much of it and keeping only the most significant
aspects. These neurons act essentially as filters, and transmit only certain
aspects of the signal they receive. A consequence of this is that much of the
information present at the level of the sensory receptors is thrown away.
Filtering is largely achieved by circuits, in which neurons interact with
each other through their synaptic connections. As a result of these interac-
tions, some features of the signal are enhanced and others are discarded at
each level in a sensory system. This progressive refinement of the sensory
signal begins at the very first synapse, between a sensory receptor and a
second-order neuron. Generally, lower-order neurons respond to fairly
simple characteristics of stimuli, such as changes in brightness. Higher-
order neurons, on the other hand, often respond to particular patterns of
stimuli in which information coming from particular groups of sensory
receptors is combined together. In fact, much of the complexity of the syn-
aptic interactions in a sensory system is concerned with ensuring that there
is a consistent response to a specific stimulus pattern, although the physi-
cal attributes of the stimulus may vary. An interneuron that responds to a
particular combination of stimuli is known as a feature detector, and
99
the specific pattern to which it responds may be termed its key feature.
Neuronal circuits that select different features commonly occur in parallel
at a given level in a sensory system, with each circuit forming a pathway for
processing a particular feature.
Feature detection has been studied in most detail in visual systems. The
clearest examples are those in which visual processing is involved in initiat-
ing a simple behaviour pattern of obvious importance, as in the case of prey
recognition in toads (discussed in sections 1.4 and 1.5). Early ethologists
invoked the concept of a releasing mechanism to explain how a sensory
system filters out a particular key feature to trigger an appropriate behavi-
our pattern. Study of insects has revealed a rich assortment of neurons that
respond to particular types of moving stimuli, and we probably know more
about the circuits involving movement-detecting neurons in insects than
about those in any other type of animal. For some of these neurons, their
role in behaviour is quite clear and we have a reasonable idea of how their
selective responses are produced by neuronal processing.
5.2 The insect visual system
The interneurons that process visual information in insects are arranged as

a series of neuropiles that lie beneath a compound eye and make up an
optic lobe of the brain (Fig. 5.1). Most interneurons connect two layers of
neuropile, with their cell bodies and input synapses in one layer and an
axon that conducts information towards the brain into the next layer of
neuropile. Information, therefore, tends to flow sequentially through
different layers of neuropile. But there are also circuits that make lateral
connections within one layer, and some neurons convey information out-
wards, from the brain towards the eye.
The photoreceptors convey a point-by-point representation of the envi-
ronment to the first neuropile, the lamina. Axons that connect the lamina
with the next neuropile, the medulla, cross over one another so that the rear
of the medulla is connected with the front of the lamina. The spatial repre-
sentation of the environment is preserved in an array of columnar elements
within the medulla, and the presence of a topographic map of the environ-
ment is a universal feature of complex visual systems. (Topographic means
‘description of place’ in Greek.) In each medulla column, there are several
tens of types of neuron. It is likely that different types of neuron in a
The insect visual system 101
Figure 5.1 The visual system of a diurnal insect, such as a dragonfly or a

locust, showing the main neuronal pathways involved in movement detec-
tion. The three optic neuropiles (lamina, medulla and lobula) are indicated
by light stippling. The small arrows show the pathway followed by informa-
tion originating in the uppermost ommatidium that is drawn. The general
arrangement is similar in flies, except that the photoreceptors of one
ommatidium are not fused, and the lobula contains a distinct neuropile
called the lobula plate, which would lie on top of the lobula in this drawing.
medulla column are dedicated to filtering out particular stimulus features

that occur within the visual field of one ommatidium, or of a group of
neighbouring ommatidia. Many of the neurons in the next layer, the lobula,
are also arranged in topographic columns. It is hard to make electrophysio-
logical recordings from these columnar neurons in the lobula and medulla
because they are tiny. However, some particularly large neurons also occur
in the lobula, and stable recordings can be made from them so that their
response properties can be studied in detail. These neurons have fields of
view that incorporate many ommatidia, and they abstract information
about particular types of movement which can occur anywhere within a
very large field of view. These movement-sensitive cells, therefore, provide

excellent examples of feature-detecting neurons.
The first neuropile, the lamina, is divided into an array of discrete neural
units called cartridges. Each cartridge receives its input from photorecep-
tors that share the same field of view. In most species, including locusts and
dragonflies, this means that a cartridge receives input from photoreceptors
of the ommatidium that lies above it. In advanced flies, however, each car-
tridge receives input from just one photoreceptor in the ommatidium
directly above, and from one photoreceptor from each of five surrounding
ommatidia (see Fig. 4.5b, p. 87). Each cartridge contains the same comple-
ment of neurons, and the set of connections made within a cartridge is so
precisely determined and repeated across the eye that the structure of the
lamina has been described as crystalline.
The principal cells of each lamina cartridge are called large monopolar
cells. As the name suggests, the axon of a large monopolar cell gives rise to
just one process. A tuft of short, brush-like dendrites arises from it and, in
blowflies, these dendrites make almost exactly 200 discrete anatomical syn-
apses with each connecting photoreceptor (Nicol & Meinertzhagen, 1982).
The large monopolar cell process then becomes a smooth axon which,
together with the axons of smaller neurons, conveys information from the
lamina to the medulla. From each ommatidium, two central photoreceptors,
that are tuned to detect ultraviolet and blue light, pass straight through the
lamina and terminate in the medulla. Within the lamina, amacrine cells
make lateral connections between adjacent cartridges, and there may also be
feedback neurons that carry signals back from the lamina to photoreceptors.
5.3 Neuronal coding in the insect lamina
The large monopolar cells are specialised to respond to changes, or con-

trasts, in the visual signal. Like the photoreceptor cells, they convey signals
as small, graded changes in membrane potential and do not generate trains
of spikes. They are, therefore, particularly sensitive to small fluctuations in
light intensity about an average value. Their responses also depend on the
contrast between light received by their own photoreceptors and those of
neighbouring cartridges.
Characteristic intracellular responses from a photoreceptor and a large
monopolar cell to a pulse of light are shown in Fig. 5.2. Photoreceptors
Neuronal coding in the insect lamina 103
Figure 5.2 The transfer of signals across the first synapse in an insect visual
pathway. The drawing on the left shows how the retinula cells of an omma-
tidium send short axons down to the layer of second-order neurons in the
lamina. Intracellular recordings of responses to a 0.5s flash of light were
recorded separately from the receptor cell body and axon, and from a large
monopolar cell (LMC). (Recordings from Laughlin, 1981; copyright © 1981
Springer-Verlag.)
make inhibitory synapses with large monopolar cells which, therefore,

respond to an increase in light with a hyperpolarising signal, a response of
the opposite polarity to that of photoreceptors. The response by a large
monopolar cell is not, however, a mirror image of the response in a photo-
receptor. These second-order neurons respond much more phasically than
photoreceptors to a change in light, which is well illustrated in the wave-
forms of the responses to a pulse of light shown in Fig. 5.2. The response of
the photoreceptor shows an initial peak depolarising potential, followed by
a sustained, smaller depolarisation that lasts until the end of the stimulus.
The large monopolar cell marks the start of the stimulus by a large, tran-
sient hyperpolarising signal, and the end of the stimulus by a transient
depolarising signal. During the light stimulus, its membrane potential
repolarises almost to its level in darkness. Thus, information about mean
Figure 5.3 Coding of light intensity by large monopolar cells (LMCs) of

dragonfly (Hemicordulia) or blowfly (Calliphora). (a) Peak response ampli-
tudes plotted as a function of the intensity of illumination delivered from a
small point source of light, centred on the receptive field of a LMC. For the
curve on the left, the LMC is dark adapted (● ), whereas for the others it is
light adapted to various different background intensities. For each curve,
background intensity can be read from the intersect with the axis that
shows zero response. The responses to both light-on and light-off were
measured relative to the largest hyperpolarising response to light-on. (b) A
schematic diagram to show coding of light-intensity in retinula cells and
LMCs. The light signal varies sinusoidally and gradually increases in mean
intensity, representing the envelope of light-intensity variations that might
be experienced by scanning a visual scene at different intensities of ambient
light. Because of adaptation, when the mean light level increases, the ret-
inula cell response does not grow as dramatically as the light-intensity
signal. The LMC response is proportional to the contrast in the light signal:
most of the signal about ambient light intensity is removed, and the
remaining signal is amplified. (Modified after Laughlin & Hardie, 1978.)
levels of illumination is lost in passing from photoreceptors to large mono-

polar cells, but information about changes in illumination is enhanced. The
loss of information about the mean level of illumination is a good example
of adaptation in a sensory system.
The way in which the large monopolar cells code visual stimuli has been
studied by Simon Laughlin and colleagues in the same way as has been
done for photoreceptors (Fig. 5.3). First, in the dark-adapted state,
responses are measured to brief light stimuli of different intensities deliv-

ered from a dark background. Then a series of intensity–response curves is
plotted, each one recording responses to increases and decreases in light
from a particular mean, or background, level. Because adaptation quickly
returns the membrane potential of a large monopolar cell to near its dark
resting potential following a change in light, there is no steady-state
response to background illumination. Consequently, the large monopolar
cell response to an increase or decrease in light always starts from the same
value of membrane potential, and the intensity–response curve moves hor-
izontally by an amount equal to the change in background light intensity.
Small increases and decreases in intensity produce, respectively, hyperpo-
larising and depolarising departures from resting potential.
The size of a response to a particular change in light intensity is always
greater in a large monopolar cell than in a photoreceptor. In other words, as
the signal is passed across the first synapse in the pathway it is amplified as
well as being inverted. In blowflies, the signal is amplified about six times.
The relative change in light during a particular stimulus is expressed as its
contrast, and so large monopolar cells generate larger responses than
photoreceptors to light signals of particular contrasts.
Quantitatively, the relationship between contrast and response can be
measured as the gradient of the intensity–response curve; the steeper the
gradient, the greater the sensitivity to small changes in light intensity. Large
monopolar cells have steeper intensity–response curves than photorecep-
tors, reflecting their greater sensitivity to small fluctuations in light from a
particular mean intensity. As a result, for a particular background level of
light, the response by a large monopolar cell to increases in light intensity
saturates at a lower intensity than the response by a photoreceptor.
Saturation in the response is shown by a levelling-off in the inten-
sity–response curve; it is the intensity at which a further increase in light
will no longer lead to an increase in response amplitude.
The range of decreases in intensity that a light-adapted, large monopolar
cell can cover is also reduced compared with a photoreceptor.
Consequently, there is a trade-off in response characteristics: the price of
an increase in sensitivity to changes in light is a reduction in the range of
changes in intensity that can be covered. The response by a large monopo-
lar cell to a signal with a particular contrast is the same, irrespective of the
background, adapting light level.
When an insect experiences a large change in background light intensity,

for example by flying out of dense shade into bright sunlight, its large
monopolar cells can be driven beyond the range of their intensity–response
curves. However, light adaptation is sufficiently rapid that large monopolar
cell membrane potential usually settles to near its original value within less
than a second, and the neuron’s ability to respond to small changes in light
intensity is restored. Following a change in light intensity of 100 times, the
initial, large response by a large monopolar cell has almost completely
decayed within 200 ms. Very large changes in light intensity such as this
are extremely rare in the day-to-day life of an insect, and the inten-
sity–response curve of a large monopolar cell spans the range of changes in
intensity that the insect will encounter most often as it moves around.
Thus, the synapses that connect a photoreceptor with a large monopolar
cell process the visual signal in two important ways. They filter it, so that
information about background intensity is subtracted; and they amplify it,
so that the signal about small changes in intensity is transmitted.
Amplification, to make the signal as large as possible at an early stage in the
visual pathway, is important because every time the signal passes across a
synapse from one neuron to another it can become contaminated with
noise (Laughlin, Howard & Blakeslee, 1987).
Another kind of transformation that occurs involves mutual inhibition
between nearby cartridges. This kind of inhibition between neighbouring
units in a sensory system is widespread, and is known as lateral inhibition.
It is a mechanism for sharpening the receptive field of a neuron so that it
responds well to small stimuli centred on its own receptive field, but not to
large stimuli that fall on the receptive fields of many neurons. The signal
about light levels detected by surrounding ommatidia is subtracted from
the signal in the cartridge of a central ommatidium.
Lateral inhibition was first investigated in a compound eye of the horse-
shoe crab (Limulus) and was subsequently demonstrated to occur in verte-
brate retinas (Box 5.1). Its action can be shown in an insect large monopolar
cell by stimulating photoreceptors with a spot of light that is small enough
to illuminate only a single photoreceptor. The spot can be directed at
different angles to the ommatidium, and the responses by a large monopo-
lar cell to the light change with the angle of the stimulus (Fig. 5.4). When the
light is centred on the central axis of the large monopolar cell’s receptive
field, the cell’s biggest responses are recorded, and the response amplitude
Box 5.1. Processing and filtering in the vertebrate retina

Although the insect eye is quite different in structure from the verte-
brate retina, both have the function of converting a physical image
into a neuronal representation that the brain can act upon, and there
are striking parallels in the way the two operate, particularly in the
early stages. The figure summarises some of the principal types of cells
and their signals in response to a short pulse of light directed to the
photoreceptor on the left, based mainly on work on the amphibian
Necturus (see Dowling, 1987). Rod and cone photoreceptors transduce
light into receptor potentials, and signals pass first to bipolar cells and
then to retinal ganglion cells, which have axons that travel into the
brain (about 1.2 million in humans; Sterling, 1998). Signals are
modified by two layers of horizontally oriented cells, horizontal cells
and amacrine cells. Adaptation to changes in background light levels
occurs at several stages, including in the machinery responsible for
transduction in the photoreceptors, in the strengths of electrical
synapses that link photoreceptors, and in the chemical output
Box 5.1. (cont.)

synapses from photoreceptors. Bipolar cells have steeper
intensity–response curves than photoreceptors, and these curves
remain centred on the mean background light level as it alters so, like
the large monopolar cells in the fly eye, bipolar cells signal contrasts in
the visual stimulus (Laughlin, 1994). Bipolar cells also have a centre-
surround organisation to their receptive fields, in which the action of
the photoreceptors directly above a bipolar cell is opposed by the sur-
rounding photoreceptors, an action mediated via the horizontal cells.
(Figure redrawn after Dowling, 1970)
decreases as the light is moved off axis. When the spot is shone from about
1.5° off axis, it begins to illuminate neighbouring ommatidia, and the large
monopolar cell starts to generate a depolarising rather than a hyperpolaris-
ing response. The receptive field of the large monopolar cell, therefore, has
two different regions: a central region where stimuli elicit the largest
responses (⫹ sign in Fig 5.4), and a surrounding region that elicits
responses with the opposite polarity (⫺ sign in Fig. 5.4). These two regions
of the receptive field work in opposition to each other, and the organisation
of the receptive field is described as centre-surround. This type of organisa-
tion is extremely common for neurons in sensory systems. It acts as a type
of sensory filter because it enables neurons to react strongly to small
stimuli that are centred on their own receptive field, but less strongly or not
at all to stimuli that are large enough to enter the receptive fields of several
neurons.
A centre-surround receptive field organisation is an efficient way of pro-
cessing visual information because it enables the size of neuronal signals to
be related to the likelihood of a particular stimulus occurring. Two different
points in space are more likely to be equally illuminated if they are close
together than if they are far apart. This means that if we know how bright
one point is, we could predict how bright the second is with a certainty that
is related to their separation. In terms of recognising objects, we do not
need to know the exact brightness of the two points, but whether they are
different from each other – information about their exact brightness is
redundant. Careful study of the responses by large monopolar cells in the
fly has shown that the strengths of feedback and lateral inhibition are tuned
to reduce redundant signals and enhance the detection of contrasts
between neighbouring cartridges (Srinivasan, Laughlin & Dubs, 1982).
Optomotor neurons in flies 109
Figure 5.4 The sensitivity of a large monopolar cell to light incident upon
the retina from different directions, expressed as a percentage of the
maximum response. Zero response level is the membrane potential in the
dark. Through lateral inhibition, off-axis light elicits a small response of
opposite polarity to on-axis light, so that the receptive field has an excita-
tory centre and an inhibitory surround, as shown in the inset. (Redrawn
after Srinivasan et al., 1982.)
5.4 Optomotor neurons in flies
The detection of different kinds of movements plays a very significant part

in controlling the behaviour of most fast-moving animals, such as flies.
These insects are able to make high-speed turns during flight, and are able
to land precisely on small targets, like a blade of grass or the edge of a cup.
They mate on the wing, and males pursue females under visual control
(Land & Collett, 1974). Many of these behaviours are almost certainly under
the control of a group of large, fan-shaped neurons that occupy a distinct
region of neuropile towards the posterior of the lobula called the lobula
plate. There are about 60 of these neurons (Hausen & Egelhaaf, 1989), and
many of them are involved in optomotor responses. Unlike the large mono-
polar cells, they are not concerned with stimuli that affect single omma-
tidia, but they respond to particular movements that occur anywhere
within a large receptive field.
Optomotor responses result in movements that tend to keep images of
the environment in constant positions on the retina. These simple behavi-
ours play vital roles in enabling more complex behaviour patterns to be
executed smoothly. They cause a stationary animal to maintain a constant
position and orientation, or help a moving animal to maintain a particular

course of locomotion.
An easy way to observe an optomotor response is to place an animal in
the centre of an upright cylinder that has vertical dark and light stripes
painted on its inner wall. If the cylinder is rotated, the animal will swivel
around its vertical axis, following the movement of the stripes.
Measurements of optomotor responses in a variety of types of animal have
shown that the animal attempts to minimise slippage of the images of the
stripes over the retina. Because most of the visual field of the animal is
occupied by the stripes, the animal interprets movement of the drum as
movement of itself relative to the environment, and optomotor responses
naturally tend to stabilise the animal’s eyes.
Flies reliably exhibit optomotor responses during flight. The strengths
and directions of these responses can be measured in the laboratory by
gluing a holder to the fly’s back and suspending it in the air. When the fly’s
feet lose contact with the ground, it beats its wings up and down as if flying,
and it will attempt to twist around its holder when it sees vertically oriented
stripes drifting past its eyes. The strength of the twisting force that the fly
applies to its holder can be measured, and the direction of the force
changes whenever the direction of travel of the stripes is reversed (Fig. 5.5a,
b). This kind of movement occurs in the yaw plane, and will tend to correct
the animal’s flight path when it twists around its vertical axis. Similar opto-
motor responses also operate in the roll and pitch planes (Fig. 5.5c). A twist-
ing movement by the fly, such as yawing, is characterised by a difference in
the direction in which images move over the two eyes, whereas a translat-
ing movement, such as flying straight forwards, is characterised by images
that move over the two eyes in the same direction and at the same speed.
When an animal moves relative to its environment, a particular pattern of
direction of movements is detected by the eyes and that pattern depends
on the type of movement. During forward movement, for example, images
move from the front of the animal towards the rear. The slowest movements
will be detected by the parts of the eyes directed towards the front of the
animal, and the fastest movements will be detected by the parts of the eyes
directed towards the sides of the animal. On the other hand, when a flying
animal pitches downwards, a rotating pattern of movements is set up, with
the front part of each eye receiving upward motion and the back part
receiving downward motion. In Fig. 5.5d speed and direction of movement
Figure 5.5 Optomotor responses and flow fields. (a) A fly, such as Calliphora
or Musca, is suspended in the centre of a striped drum that can be rotated.
The wire tethering the fly is attached to a meter that measures the torque,
or turning force, produced by the fly. (b) Recordings of drum movements
and torque produced by a fly. (c) The three axes for rotatory and translatory
movements that an airborne fly can make. (d ) Flow fields that stimulate the
right eye when the insect flies forwards or pitches downwards. (b recordings
from Egelhaaf, 1985; copyright © 1985 Springer-Verlag.).
of images over local regions of the eye are indicated by the lengths and
directions of the arrows. The pattern of coherent motion associated with a
particular kind of movement is called a flow field, and an animal can gather
information about its movement through the environment by detecting the
kind of flow field that its eyes are receiving.
Many of the large, fan-shaped neurons of the fly lobula plate are unique
individuals, and generate their largest responses to particular types of stim-
ulus movement. They filter out particular stimulus features and ignore
others. The exact location of a moving stimulus is not significant to these
neurons, but direction of movement is, and all these neurons show partic-
ular direction selectivity. For example, the three HS neurons respond most
strongly to images that rotate around the animal in the horizontal plane,
and the 11 VS neurons respond to movements upwards or downwards (Fig.
5.6a). Each HS neuron in the right lobula plate is most strongly excited by
movements occurring in the clockwise direction around the animal (back-
wards over its own eye or forwards over the other eye), and those in the left
lobula plate are most excited by anticlockwise movements. The most
effective direction for exciting a neuron is called the preferred direction.
Often, neurons are inhibited by movement in the opposite direction, which
is called the null direction. Each VS neuron is tuned to respond most
strongly to a particular type of flow field (Krapp & Hengstenberg, 1996). One
VS neuron, for example, responds most strongly to one direction of rolling
about the longitudinal axis, but other VS neurons are tuned to detect twist-
ing movements that pitch the head upwards or downwards. The VS
neurons, therefore, provide an array of filters, in which a particular combi-
nation of pitch and roll during flight corresponds to the strongest excitation
of one of the VS neurons. Abstraction of particular stimulus features is often
achieved by an array of neurons like this, each of which is tuned slightly
differently from its neighbours.
Although the HS neurons have no branches that extend across the brain
into the opposite optic lobe, they respond to movements over both eyes.
They are most sensitive to stimuli that rotate about the animal in one direc-
tion, being excited by stimuli that move backwards over their own eye or
forwards over the opposite eye. Neurons such as H1 (Fig. 5.6b) are respon-
sible for carrying information from one lobula plate to the other, and
provide pathways that enable the fly to compare stimuli that affect the two
eyes. These pathways enable the fly to distinguish whether it is rotating or
Figure 5.6 Optomotor neurons in the lobula plate of a blowfly. (a) On the
left is one of the three HS neurons that would respond to anticlockwise
movements around the yaw axis; and on the right is one of the VS neurons
that responds to downward motion over its eye. (b) Neuron H1, which is
excited by movements in the back-to-front direction over its own eye, the
right eye for the H1 drawn here. Its axon carries information from one
lobula plate to the other. (a modified from Dvorak, Bishop & Eckert, 1975;
copyright © 1975 Springer-Verlag; b modified from Franceschini et al., 1989;
copyright © 1989 Springer-Verlag.)
moving forwards. During rotation, movement flows forwards over one eye
and backwards over the other, whereas when the fly moves forwards, both
eyes see backwards movement.
The ways that the HS and VS neurons respond to moving stimuli suggest
that they could be important in controlling optomotor responses and there
is strong, indirect evidence that they play such a role (Hausen & Egelhaaf,
1989). The output processes of the HS and VS neurons intermingle with

dendrites of a small group of neurons that carry information from the brain
to motor neurons that control flight and neck muscles. Optomotor
responses are impaired in flies after minute cuts are made into the brain to
sever the axons of these neurons, or if the HS neurons are killed by irradiat-
ing them with a laser. Also, in Drosophila, some mutants that lack the HS
neurons do not produce normal optomotor responses. Therefore, it is likely
that these neurons are an important part of the cockpit of the fly, helping it
to maintain a particular flight course.
5.5 Figure-ground neurons of the lobula plate
While an insect is flying around, it needs to detect and react to nearby objects
as well as to monitor the way the background moves. A fly, for example, tends
to chase other flies, and to select suitable targets such as twigs on which to
land. When a fly approaches a twig, the image of the twig will move more
rapidly over the fly’s eyes than will the images of more distant objects.
Tethered, suspended flies turn strongly towards a small object that is moving
backwards relative to them, and this response is enhanced if the object is
moving over a background pattern that is also moving, rather than remain-
ing stationary (Kimmerle, Warzecha & Egelhaaf, 1997). Likely candidate
neurons for recognising a small object and initiating a turn by the fly towards
it are a group of four figure-ground neurons in each lobula plate (Egelhaaf,
1985). These neurons select a figure, or object, that is distinct from the
general background. Like the HS neurons, they are excited by movements
backwards over the eye. They also have large receptive fields but, unlike the
HS neurons, they respond much more briskly to movements of small targets
than to large parts of the visual field. This is because they receive a balance
of excitatory and inhibitory synaptic inputs in response to moving stimuli
(Warzecha, Egelhaaf & Borst, 1993). A figure-ground neuron responds
strongly to a stimulus that extends over about 5° of the visual field, but pro-
gressively less strongly to stimuli that are seen by larger regions of the eye.
5.6 A mechanism for directional selectivity
In sensory systems, there are many examples of neurons that act as filters,
responding most vigorously to one particular stimulus configuration. The
A mechanism for directional selectivity 115
selectivity of these neurons for their preferred stimulus must arise from the
way in which their presynaptic neurons are arranged, and study of fly
lobula plate neurons has enabled insights into the way in which one partic-
ular kind of selectivity – directional selectivity – arises. The neurons are
considered to be driven by an array of local motion-detecting units, each of
which detects the movement of images in a particular direction over a small
part of the eye. These units are commonly referred to as elementary motion
detectors. Their properties are known in some detail, although the neurons
of which they are composed have not been identified. Many of the proper-
ties of elementary motion detectors were first discovered in the 1950s from
experiments on the optomotor behaviour of beetles and flies, well before
methods had been developed for recording signals from single neurons in
the fly brain. More recently, their properties have been inferred by record-
ing the responses by lobula plate neurons to moving stimuli that are seen
by small regions of the eye, when the number of local motion detectors
stimulated would be very small.
When the image of a moving object travels across an eye, different photo-
receptors are stimulated in a particular order. Information about the direc-
tion of movement could, therefore, be obtained from the sequence in which
the photoreceptors are stimulated. The open rhabdom design of the fly eye
has enabled Nicholas Franceschini and colleagues (Franceschini, Riehle &
Le Nestour, 1989) to perform a remarkable experiment in which direction-
ally selective responses from a lobula plate neuron were elicited by stimu-
lating just two photoreceptors in sequence (Fig. 5.7a). The neuron they
chose to study was H1 (see Fig. 5.6b), which is the easiest lobula plate
neuron to record from. A special optical instrument enabled an experi-
menter to view a single ommatidium and to direct a tiny spot of light onto
each of two individual photoreceptors which view adjacent points in space.
H1 is excited when the posterior receptor is stimulated just before the ante-
rior receptor, and it is inhibited when the two receptors are stimulated in
the reverse order (Fig. 5.7b). This corresponds with the directional selectiv-
ity by H1 for moving objects, which is forwards over its own eye.
Illumination of either receptor alone, or both at the same time, does not
cause any response in H1; there must be a delay between stimulation of one
and stimulation of the other. Illumination of the posterior receptor facilitates
a response by H1 to stimulation of the anterior receptor. This facilitation does
not start immediately when the posterior receptor is stimulated, but after a
Figure 5.7 Directional selectivity in the housefly (Musca) H1 neuron to

stimulation of two individual photoreceptors. (a) A fly ommatidium viewed
through its lens, showing the seven individual rhabdomeres that are visible.
The anterior (a) and posterior (p) photoreceptors that were stimulated in
the experiment are indicated. (b) Responses by H1 to different sequences of
stimulating the two photoreceptors. The times when a pinpoint of light was
directed at each photoreceptor are indicated. There was a vigorous response
when a stimulus to the posterior receptor preceded that to the anterior, but
almost no response when the two receptors were stimulated simultane-
ously. Note that the light-on and light-off stimuli generated separate peaks
in the response. In another experiment (shown in the bottom trace), stimu-
lation of the anterior receptor before the posterior one (the null direction)
caused inhibition of H1. Each recording was an average from 100 stimulus
repetitions, and shows the response of H1 as instantaneous spike frequency.
(c) Diagram of an elementary motion detector. Open triangles represent the
two photoreceptors. (b modified from Franceschini et al., 1989; copyright ©
1989 Springer-Verlag.)
delay of at least 10 ms. It reaches maximum strength after 50 ms, and lasts for
up to 250 ms. This means that H1 is most sensitive to movements that stim-
ulate the anterior receptor 50 ms after the posterior receptor (these move-
ments are produced by images travelling over the surface of the eye at a speed
of 72 degrees per second). The delay in facilitation ensures that H1 is not
excited when overall illumination of the eyes alters or if light intensity
flickers, which would occur when the insect flies into or out of shade.
The wiring diagram in Fig. 5.7c summarises how a directionally selective
motion detector, with a preferred direction from posterior (p) to anterior
(a), might be organised. The diagram summarises how information is pro-
cessed. The inputs are elements that respond to light-on or light-off stimuli,
and each probably includes a photoreceptor and various neurons in
the lamina and medulla. The signal from the posterior photoreceptor is
delayed before it is combined with that from the anterior photoreceptor.
The combination is indicated as a switch that enables one photoreceptor to
regulate the output of the other, although in reality the way the outputs
from the two photoreceptors is combined is more sophisticated than a
simple on–off device. The two signals are thought to be multiplied together,
which ensures that the output is very small when only one of the two recep-
tors is stimulated (multiplying by zero gives a result of zero). The result is
that a strong output from the motion detector is only produced when stim-
ulation of the second photoreceptor follows stimulation of the first photo-
receptor with a certain delay.
It would be reasonable to assume the medulla contains many such cir-
cuits, perhaps one per column, so that H1 is excited by movement in its pre-
ferred direction over any part of its eye. In fact, at each location there must
be four copies of similar circuits. The first duplication is necessary to
account for inhibition of H1 by stimuli moving backwards over the eye, the
null direction. Inhibition is apparent in experiments in which H1 shows
quite a high rate of spontaneous spike discharge in the absence of any
movement stimulation, and sequential stimulation of the two photorecep-
tors in the null direction briefly inhibits H1, reducing its spike rate. This
inhibition shows the same time-dependent properties of facilitation as
excitation by movement in the preferred direction. To account for this,
there must be two mirror-image, motion-detecting circuits working
together, one exciting H1 and the other inhibiting it. The second duplica-
tion, in which there are two of each excitatory and inhibitory circuit, is
Figure 5.8 Responses by elementary motion detectors. (a) When a fly

viewed a moving pattern through a vertical slit cut in a mask, only a narrow
band of ommatidia and elementary motion detectors was stimulated. (b) A
wide-field HS neuron responded to each lightening or darkening of this
band, as shown in the intracellular recording. When the stimulus moved in
the null direction, each response by the strip of motion detectors caused
hyperpolarising potentials; and when the stimulus moved in the preferred
direction, another set of motion detectors responded, causing depolarising
potentials. When the mask was removed so the eye saw the whole moving
pattern, responses from different strips of motion detectors summed so that
sustained excitatory or inhibitory responses were recorded from the HS
neuron. (b modified after Egelhaaf & Borst, 1993.)
necessary because light-on stimuli are processed separately from light-off

stimuli. Although photoreceptors are excited by light-on stimuli and inhib-
ited by light-off stimuli, neurons such as H1 respond to moving light or dark
objects, so signals from photoreceptors must be channelled through
different, parallel pathways before they reach the lobula plate.
Microstimulation of single pairs of receptors shows that light onto the first
receptor facilitates responses to light on, but not to light off, delivered to the
second. This duplication ensures that a local motion detector will respond
only when there is correspondence in the stimulus that the two photore-
ceptors see.
In another type of experiment, Martin Egelhaaf and Alexander Borst have
made intracellular responses from an HS neuron (Egelhaaf & Borst, 1993).
They stimulated the eye with a pattern of vertically oriented stripes that
drifted backwards or forwards over it (Fig. 5.8a). The darkness of each stripe
varied sinusoidally, so there were no sharp borders between light and dark
in the pattern. A mask with a narrow slit cut into it was placed between the
moving pattern and the eye so that only a narrow, vertically oriented band
of ommatidia was stimulated. When the stimulus moved, the photorecep-
tors in this band experienced sinusoidal changes in light intensity, rather
than the abrupt switches in light that occurred during the microstimulation
experiments. When the stimulus moved in the preferred direction, the
response by the HS neuron consisted of a series of discrete depolarising
potentials in which a large potential alternated with a smaller one (Fig.
5.8b). Each large potential corresponded with a darkening of the photore-
ceptors in the band of stimulated ommatidia, and each small potential cor-
responded with a lightening of this band. When the stimulus moved in the
null direction, the neuron responded with a series of alternating large and
small hyperpolarising potentials. When the mask was removed so that a
large part of the eye saw the moving pattern, the intracellular recording
showed a sustained depolarisation of the neuron, indicating excitation, or
a sustained hyperpolarising potential, indicating inhibition, depending on
which direction the stripes moved. These results are consistent with the
hypothesis that motion is detected by local circuits, similar to Fig. 5.7c.
When the eye sees a large, extended pattern rather than a narrow slit, an HS
neuron will receive signals from many vertical bands of ommatidia.
Because the outputs from the different bands are out of phase with each
other, the summed signal in the HS neuron will be a steady depolarising or
hyperpolarising potential in which the fluctuations in output from individ-
ual motion detectors are ironed out.
Neurons like H1 and the HS neurons are excited most strongly by move-
ment of large-field stimuli in a particular direction. The amount of excita-
tion also depends on the speed and repeat pattern of the stimulus. These
two stimulus features, speed and repeat pattern, cannot be distinguished
from each other by an elementary motion detector because the response to
a narrow series of stripes moving quite slowly over the eye is the same as the
response to stripes that are twice as broad but moving twice as rapidly (Fig.
5.9). The strength of excitation, therefore, depends on the frequency with
which the detector is stimulated with changes in the contrast of light.
Slowly moving, closely spaced stripes will stimulate a detector with the
same contrast frequency as more rapidly moving, broader stripes. The
fact that the responses of H1 and HS neurons also depend on contrast
Figure 5.9 An elementary motion-detector circuit cannot distinguish

between narrow stripes moving slowly and wider stripes moving more
quickly. In this circuit, each receptor is excited whenever a light–dark edge
moves over it. The amount by which the detector is excited depends on the
delay between excitation of the two receptors, which is short if either the
speed of movement is fast or if the stripes are spaced close together.
frequency indicates strongly that they are driven by this type of elementary
motion detector. The strengths of optomotor turning responses by a teth-
ered fly also depend on contrast frequency of stimuli, rather than speed or
spatial pattern, and this lends support to the hypothesis that the HS
neurons are responsible for controlling these behaviours. Some move-
ments by insects, however, are controlled in a way that suggests the insects
can measure stimulus speed independently of image structure (Box 5.2).
Visual systems channel information about movements through a number
of independent pathways, each focusing on particular aspects of the stim-
ulus. Ambiguities, such as that between stimulus velocity and pattern, can
be removed at later stages by combining the outputs of different channels.
Summary of fly optomotor neurons 121
Box 5.2. Bees can measure image speed to fly a straight course
Bees can be trained to fly along a tunnel if it is part of the route

between their hive and a good food source. They tend to maintain a
course straight along the centre of the tunnel, and they could do this
by balancing the relative speed of motion that is detected by the left
and right eyes. Kirchner & Srinivasan (1989; see also Srinivasan,1992)
showed that bees detect the speed of motion over each eye by observ-
ing flight paths along tunnels with side walls decorated with vertical
stripes. No matter how broad the stripes on each side are, the bee flew
straight down the middle (a). If the stripes on one wall moved (short
arrow) in the same direction as the bee, the bee flew closer to that wall
(b), but if the stripes moved in the opposite direction, the bee flew
closer to the other wall (c). These results suggest quite strongly that
bees have neurons that can compare the speeds of the images of the
two walls, something that would be difficult to achieve with the opto-
kinetic neurons of the fly lobula plate.
5.7 Summary of fly optomotor neurons
Motion over the eye is first detected by small circuits, called elementary
motion detectors, that respond to the sequence in which a pair of photore-
ceptors is stimulated. Each retinotopic column of neurons in the medulla
probably contains several of these detectors, each tuned to a different
direction of movement or of light-intensity change. In an elementary
motion detector, the signal that originates in one of the photoreceptors is
delayed, and is then combined with the signal from the second photorecep-
tor. The detector, therefore, correlates the signal in one receptor with the
signal that occurs slightly later in the second. The detectors drive an array
of fan-shaped neurons in the lobula plate, each of which is tuned to a
particular direction of motion occurring anywhere in a very large visual
field. Most lobula plate neurons respond best to movement of a large pano-
ramic image, and are well suited to controlling optomotor responses that
stabilise the image of the environment on the animal’s eyes. A few respond
selectively to movements of small objects, and these probably elicit turning

responses by the fly towards small objects such as other flies or perches.
5.8 Collision warning neurons in the locust
Optomotor responses help to stabilise an animal when it is stationary or

proceeding along a straight course, and they do this by referring to move-
ments that occur in the background. The animal also needs to be able to
respond to individual objects, such as potential predators, mates or perch-
ing sites. Like many animals, locusts and grasshoppers will escape from
rapidly approaching objects. This is apparent to anyone who has tried to
catch one of these insects; they respond to approach by a powerful jump,
caused by rapid extension of the hind legs. Often, the wings are opened
during the jump, and the animal extends the range of its jump by flying or
gliding.
A particular neuron in the hindbrain of the locust has been found to
respond vigorously to the images of approaching objects. The axon of the
neuron travels to the thorax, where it excites motor neurons and interneu-
rons that are concerned with the control of the jumping and flying.
Probably, therefore, it plays a vital role in channelling sensory information
about an approaching object to the motor control circuits responsible for
executing escape movements.
It is relatively easy to record responses from this neuron by using extra-
cellular electrodes placed around a nerve cord because it produces spikes
that are usually much larger than those from the axons of other neurons.
Because the axon of the neuron crosses the brain, and responds to move-
ments detected by the opposite (contralateral) eye, the neuron is called the
descending contralateral movement detector (DCMD). The DCMD receives
its input from a single, large, fan-shaped neuron in the lobula, called the
lobula giant movement detector (LGMD; Fig. 5.10). Spikes in the DCMD
follow those in the LGMD one for one. The two neurons are extremely sen-
sitive to small movements anywhere within the visual field of their eye, and
respond with a brisk, but brief, burst of spikes. As is to be expected from
neurons that receive their input through large monopolar cells in the
lamina, the response to a sustained stimulus rapidly disappears and they
respond to changes in contrast over a wide range of background light inten-
sities.
Collision warning neurons in the locust 123
Figure 5.10 Two feature-detecting neurons in the visual system of a locust

(Locusta), the lobula giant motion detector (LGMD) and the descending
contralateral motion detector (DCMD). The drawings were made from
neurons that had been stained by injecting them with cobalt ions. The
LGMD receives excitatory input from lower-order interneurons on the fan-
like array of dendrites on the left. (Modified from Rind, 1984.)
The LGMD generates a vigorous and prolonged train of spikes in

response to an approaching object. The frequency of spikes increases
throughout the approach movement, as if the neuron has locked on to the
approach movement (Fig. 5.11a). Images of receding objects, or of objects
that are moving around the locust, generate only brief responses.
Deviations as small as 2–3° from a direct collision course result in a reduc-
tion in response in the LGMD by a half, so it is remarkably tightly tuned to
respond to objects that will collide with the animal (Judge & Rind, 1997).
The intensity of response depends on the speed with which the object is
approaching, and the neuron would respond very well to a predatory bird
swooping towards a locust. Wide field movements, like those that cause
optomotor responses, inhibit responses by the LGMD.
To determine the types of movement that are most effective at exciting
the LGMD, Claire Rind made recordings while a locust viewed a video of a
space movie (Rind & Simmons, 1992). This approach was an effective way
of providing a wide variety of visual stimuli and rapidly indicated that the
Figure 5.11 Detection of approaching objects by the locust LGMD neuron.

(a) Response by the LGMD to the image of an object approaching and then
receding at 3.5 m/s. The top trace is an intracellular recording from the
LGMD, and the lower trace is a monitor of the size of the object’s image,
which was displayed on a screen. (b) Stimulation of one part of the visual
field inhibits the response to stimulation of another part. The drawing
shows the stimulus screen, with two vertically oriented bars that could
move from left to right across the screen. Movement of either bar alone
across the screen caused EPSPs and spikes in the LGMD, but when the left
bar moved just before the right one, the response to movement of the right
bar was greatly reduced. (c) Drawing of an electron micrograph of a den-
drite of the LGMD with input synapses onto it from two adjacent presynap-
tic elements, pre-1 and pre-2. Each presynaptic element contains a densely
staining bar-like structure that is believed to direct vesicles of neurotrans-
mitter to their site of release at an insect synapse. The LGMD receives
synapses from each presynaptic element, and the two elements also
synapse with each other as indicated by the arrows. (a recordings from
Rind, 1996; reprinted with permission of the American Physiological
Society; b from Rind & Simmons, 1998.)
Collision warning neurons in the locust 125
LGMD responded more vigorously to images of approaching objects than

to other types of movement. Subsequently, responses were analysed in
more detail by generating carefully controlled moving images with a com-
puter, and then determining exactly which features in the image of an
approaching object are the important cues (Simmons & Rind, 1992).
Although many invertebrates show avoidance reactions when a shadow is
cast over them, the LGMD only responds weakly to overall decreases in light
intensity. To excite it, an image of the edges of an approaching object must
move over the surface of the eye. Two different features of the image are
important as cues that an object is moving nearer to the eye: first, the edges
grow in length; and, second, the edges accelerate as they move across the
retina.
As with the lobula plate neurons in the fly, selectivity by the LGMD for a
particular kind of moving stimulus is established by circuits among
neurons in the medulla. The way that these circuits work has been estab-
lished by using a variety of experimental approaches. Electrophysiological
recordings demonstrated that lateral inhibition operates among the
columns of neurons in the medulla (O’Shea & Rowell, 1976). When one area
of the retina is stimulated by a moving image, the response to stimulation
of other parts of the retina is depressed (Fig. 5.11b). Although the first
movement clearly decreases the response by the LGMD to the second
movement, no inhibitory postsynaptic potentials are recorded from the
LGMD. This means that the inhibition must occur at an earlier stage, pre-
synaptically to the LGMD.
Using electron microscopy, it has been shown that the dendrites of the
LGMD are covered with input synapses, probably from neurons that origi-
nate in the medulla. These synapses are arranged in a remarkable manner
(Fig. 5.11c). Each neuron that synapses with the LGMD also synapses with
its neighbours; in other words, the neurons that drive the LGMD are recip-
rocally coupled to each other. These microanatomical circuits provide a
route for local lateral inhibition among the elements that excite the LGMD,
which is a key feature of the input circuitry to the LGMD, allowing it to filter
out approaching objects. In the lamina, the lateral inhibition between car-
tridges serves a different function – to sharpen the detection of edges in the
image.
The way in which the LGMD filters out the image of an approaching
object can be envisaged as a kind of race between excitation and inhibition
Figure 5.12 A diagram of the circuits thought to drive the LGMD. An array of
units drives the LGMD through excitatory inputs (—䉳) and inhibits their
neighbours. Each unit is excited by a small number of photoreceptors (䉯—)
and inhibits its neighbours (—䊉). When an object approaches the eye, its
edges spread outwards (arrows), and this creates a race between excitation
of units by edge movement, and inhibition between neighbours. The LGMD
is excited strongly when excitation is winning the race. A second input to
the LGMD causes inhibition in response to wide-field movements.
of the units in the medulla that synapse with the LGMD (Fig. 5.12). The exci-
tation comes from photoreceptors that are stimulated sequentially by the
edges of the image as it travels over the eye; and the inhibition travels later-
ally between the medulla units. As the number of excited units increases,
the strength of the lateral inhibition also increases. Consequently, in order
for the excitation to outstrip the inhibition and be passed on to the LGMD,
a large number of new medulla units must next be stimulated. For this to
occur, the extent of the edges in the image or the speed with which they
move must be increasing. A network like this has been modelled in a com-
puter, and it responds as predicted, being excited most strongly by the
images of approaching objects (Rind & Bramwell, 1996). The network incor-
porates a second kind of inhibition that also occurs in the LGMD system,
and this acts directly on the LGMD itself, causing IPSPs in it. This inhibition
acts to reduce responses by the LGMD when the whole background moves,
or when overall light intensity changes.
Conclusions 127
5.9 Conclusions
In a sensory system, the receptors transmit signals to a series of interneu-

rons and, together, these constitute a neuronal pathway in which signals
are progressively modified as they are transmitted from stage to stage.
Through their synaptic interactions, the interneurons act as selective filters,
enhancing some aspects of the received signal and discarding others. A
complex sense organ with many receptors is associated with a series of
neuropiles made up of many parallel pathways, each processing informa-
tion from a small group of receptors. Separate classes of interneuron within
each pathway extract different types of information from the same receptor
input. The neuronal pathways are arranged in an orderly manner, with each
one maintaining a particular position relative to its neighbours, so that the
central nervous system contains a map of the receptor array and hence, in
a visual system, of the external world.
Early in the pathway, interneurons act as signal conditioners, providing a
clear and unambiguous signal for the filtering networks that lie down-
stream. This is illustrated by the large monopolar cells of the insect lamina,
in which the signal provided by photoreceptors is processed so that it is
amplified and information about background light intensity is discarded.
The signals received by interneurons in the medulla are, as a result, largely
about contrasts in the light signal – changes both in time and in space. The
kinds of operation that medulla interneurons perform are illustrated by the
elementary motion detector in flies, or the array of laterally connected
neurons thought to drive the LGMD in locusts. We have some good indica-
tions of the types of computation that these circuits perform, although
direct experimentation, using microelectrodes to record from and stimu-
late different elements in these circuits, is extremely challenging.
Large neurons in the insect lobula are good examples of feature detec-
tors. They abstract information about a particular aspect of the stimulus,
usually the direction of movement, but discard others, such as the location
of a target. Neurons of the lobula plate in flies, and the LGMD in the locust,
are tuned quite tightly to respond most briskly to particular types of move-
ment. Information about other aspects of a stimulus is processed by other
neurons in the visual system, so that a particular scene in the environment
is analysed by parcelling different aspects into an array of different inter-
neurons, each concentrating on a different aspect. Understanding the
manner in which a nervous system resynthesises a visual scene, combining

together information from different feature-detecting pathways, provides a
major challenge to neuroscientists.
Further reading
Franceschini, N., Pichon, J.M. & Blanes, C. (1992). From insect vision to robot
vision. Phil Trans R Soc Lond B 337, 283–94. This describes how principles con-
cerning vision, derived from the study of motion-detecting neurons in flies,
have inspired the design of artificial eyes used to guide the movements of
robots.
Laughlin, S.B. (1989). The role of sensory adaptation in the retina. J Exp Biol 146,
39–62. A useful review of sensory adaptation, especially in the insect retina.
Strausfeld, N.J. (1976). Atlas of an Insect Brain. New York: Springer-Verlag. A classi-
cal description of the different types of neuron found in the brains of flies, par-
ticularly in their visual systems.
6 Hearing and hunting: sensory maps
6.1 Introduction
The interaction between a predator and its prey represents a dramatic

example of animal behaviour in which the capabilities of nervous systems
are stretched to the limit. A hunting animal faces the fundamental prob-
lems of detecting and localising the prey, and it must solve them on the
basis of purely passive information given out inadvertently by the prey. This
is a formidable task and it has led to the evolution of some remarkably
sophisticated neuronal systems in species that are adapted for hunting.
If one is asked to name a hunting species, the natural choice is a suitably
complex animal such as a large cat or a hawk. These animals do, indeed,
possess central nervous systems with the necessary sophistication to
handle the complex task of tracking prey, but this sophistication makes
most birds and mammals unsuitable as subjects for neuroethological
research. However, the difficulty can be overcome by looking at species with
a highly specialised method of hunting, based on a sensory system that is
dedicated to the specialised method of prey detection and localisation. It
then becomes easier to correlate the properties of particular neurons in
that system with the particular behavioural task (see section 1.2).
Such dedicated systems are found in two groups of animals that employ
hearing as a means of tracking prey, namely owls and bats, which use spe-
cialised auditory systems to hunt at night when visually guided predators
are at a disadvantage. Owls are able to locate small animals on the ground
by listening for the tiny rustling noises made by an animal moving among
fallen leaves and twigs. On most nights, an owl’s hearing is used in conjunc-
tion with its excellent eyesight, but on very dark nights some species can
hunt by sound alone.
Insectivorous bats both hunt and find their way around exclusively by
129
sound, using a method that is akin to human sonar and is technically

known as echolocation. This method involves the bat emitting loud pulses
of sound and then analysing the returning echoes in order to find out what
lies ahead. Echolocation is a characteristic of bats in the suborder
Microchiroptera, a diverse group that is quite distinct from the non-echolo-
cating fruit bats, the Megachiroptera (cf. Macdonald, 1984).
The basic properties of sound and their variation over time determine
what is possible by way of detection and localisation for a hunting owl or
bat. Sound is created when air molecules are set in motion by a vibrating
structure such as a loudspeaker. The vibrations of the speaker generate
alternating waves of compression and rarefaction of the air, which propa-
gate out from the speaker at the speed of sound. The molecules involved in
propagating the sound move back and forth from regions of high pressure
into regions of low pressure, which thereby become regions of high pres-
sure, and so on.
The pressure generated by a sound wave is technically expressed as
sound pressure level and is measured on a logarithmic scale. The unit most
often employed in studying animal sounds is the decibel (dB), which is
equivalent to an increase or decrease of about 12.2 per cent in relative
sound pressure. Absolute pressure levels are described relative to a refer-
ence sound pressure, usually 20 ␮Pa (⫽ 2 ⫻ 10⫺5 N m⫺2), which is roughly
the threshold of human hearing to 1 kHz sound. The interval from a given
point on one sound wave to the equivalent point on the next sound wave is
the wavelength, which is usually expressed as a frequency (the reciprocal of
the wavelength) and measured in cycles per second (Hertz). Most sounds
that animals produce or listen to have frequencies of thousands of cycles
per second, or kiloHertz (kHz).
Neuroethological research on owls has concentrated on the owl’s ability
to localise a sound source in space by measuring these basic properties of
sound. Most research on bat echolocation has paid more attention to the
bat’s ability to determine the distance from which an echo has returned
and to the suitability of different bat sounds for different sonar tech-
niques. Taken together, these studies on bats and owls are providing valu-
able insight into the neuronal basis of some sophisticated behaviour
patterns.
Prey localisation by hearing in owls 131
Figure 6.1 The hunting technique of an owl, drawn from photographs of

Tengmalm’s owl (Aegolius funereus) in its natural habitat. (a) The owl about
to strike prey with its talons, after flying down from an observation perch.
(b) The owl on its perch immediately before striking, with a diagram
showing the errors involved in localising prey by hearing. The prey (O) is
observed at a shallow angle (␣), with the result that a given angle of error
converts into a greater distance along the ground for a vertical (elevation)
error than for a horizontal (azimuth) error. (Modified after Norberg, 1970,
1977.)
6.2 Prey localisation by hearing in owls
Adult owls hunt within a well-defined territory, which they know well and
patrol regularly at night. During its patrol, an owl visits a number of obser-
vation perches, from which it can survey the ground round about. If it hears
potential prey, the owl swiftly turns its head so that it directly faces the
object of interest. Then, after adequate scrutiny, it flies down to capture the
prey in its outspread talons (Fig. 6.1). An owl listening from a perch or in
low-level flight must be able to pinpoint the sound source both in the hor-
izontal plane (azimuth) and in the vertical plane (elevation). In fact, unless
an owl is looking down from directly above its prey, its orientation is more
critical in the vertical than in the horizontal plane (Fig. 6.1b).
The use of hearing in prey capture has been studied mostly in the ubiq-
uitous barn owl, Tyto alba. Early behavioural studies were carried out using
tame individuals, which proved able to locate prey even in total darkness,
and a number of simple experiments demonstrated that hearing is used to
accomplish this (Payne, 1971). For example, it was found that the birds
could strike accurately at a concealed loudspeaker quietly broadcasting a
recording of leaf-rustling noises. This result not only showed the barn owl’s
ability to locate prey by hearing alone, but also opened the way for testing
which features of the sound are important in localisation.
Using this technique, the accuracy of localisation was found to vary with
sound frequency. The greatest accuracy was achieved with a sound con-
taining frequencies from 6 to 9 kHz. Typically, birds are not sensitive to fre-
quencies above 5 kHz, but the barn owl can hear up to 10 kHz, and 6–9 kHz
is the range to which its ear is most sensitive. In addition, more than half the
auditory neurons in a barn owl’s ear are devoted to this extended frequency
range of 5–10 kHz. The usual pattern is for approximately equal numbers of
auditory neurons to be devoted to each doubling of frequency (octave), but
the barn owl devotes a disproportionate number to the higher frequencies
(Koppl, Gleich & Manley, 1993). Hence, these important frequencies can be
analysed in greater detail and this arrangement may be thought of as an
acoustic fovea (cf. section 6.9 on bats).
The accuracy of sound localisation has been studied in more detail by
exploiting the natural response in which an owl turns to face a novel sound.
The owl is trained to remain on its perch and the angle through which its
head turns in response to a sound is measured using an electromagnetic
angle detector (Fig. 6.2 a). In each test, the head is first aligned by attracting
the owl’s attention with a sound from the zeroing speaker, and then the owl
is stimulated with a sound from the target speaker.
When the target speaker is placed in front of its face, the barn owl’s local-
isation is exceptionally accurate, with an error of less than 2° in both
azimuth and elevation. But the owl’s accuracy deteriorates in both planes as
the angle between the source and the axis of the head increases (Fig. 6.2 b).
The rapid flick of the head, with which the owl responds, is initiated about
Prey localisation by hearing in owls 133
Figure 6.2 Orientation of the head to sounds by the barn owl (Tyto alba). (a)
The method used to measure the accuracy with which the owl locates sounds
coming from different positions in space. Sound stimuli originate from either
a fixed source (the zeroing speaker) or a movable source (the target speaker).
The search coil on top of the owl’s head lies at the intersection of horizontal
and vertical magnetic fields generated by the induction coils. Movement of
the head in response to sound from a speaker induces a current in the search
coil, which is analysed by computer to give horizontal and vertical angles of
movement. (b) Localisation accuracy as a function of the position of the
target speaker, showing the mean degree of error in judging target position in
the horizontal plane (left) and in the vertical plane (right) for an individual
owl. (Modified after Knudsen, Blasdel & Konishi, 1979.)
100 ms after the onset of the sound. However, maximum accuracy can be
achieved even with brief sounds (75 ms duration) that end before move-
ment of the head begins. This shows that the owl does not locate the sound
by successive approximation but can determine the sound’s precise loca-
tion in space without feedback. That is to say, the owl is operating under
open-loop conditions (see section 1.6).
Figure 6.3 The barn owl’s ability to locate sounds in elevation. (a) A plot of
auditory space in front of the owl in degrees of azimuth (L and R) and of
elevation (⫹ and ⫺). The symbols show individual errors in open-loop
localisation of a sound source in the centre of the plot (Target) produced by
partly blocking one ear. A tighter ear plug (closed circles and triangles) pro-
duces a greater localisation error than a looser ear plug (open circles and
triangles). (b) The facial structures of the barn owl that contribute to locali-
sation of sound in elevation: the facial ruff is formed from tightly packed
feathers projecting from the relatively narrow skull, and the ear openings
are located behind the preaural flaps. These structures are revealed by
removing the sound-transparent feathers of the facial disc, which give the
owl’s face a flat appearance. (Modified after Knudsen & Konishi, 1979.)
An indication of just what cues are involved in locating a sound is pro-

vided by partially blocking one ear, which effectively reduces the sound
pressure level at that ear without significantly altering the sound’s time of
arrival. A plug in one ear leads to significant errors in elevation but only
slight errors in azimuth (Fig. 6.3a). A plug in the left ear causes the owl to
direct its head above and a little to the right of the target, and a plug in the
right ear results in the owl facing below and a little to the left. The tighter
the ear plug, the greater is the degree of error. This result indicates that the
intensity difference between the ears is the principal cue for locating a
sound in elevation.
The owl is able to use a comparison of intensity between the left and right
ears to locate a sound accurately in elevation due to the arrangement of
Auditory interneurons and sound localisation 135
feathers on its face. The ear openings and the protective, preaural flaps are
vertically displaced, the left flap being above the midpoint of the eye and
the right one below it (Fig. 6.3b). There is also a slight asymmetry in the
facial ruff, which is composed of dense, tightly packed feathers and forms a
vertical trough behind each ear opening: the left trough is orientated down-
wards and the right one upwards. Because of its dense feathers, the facial
ruff acts as an effective sound collector at frequencies above 4 kHz.
Consequently, the asymmetries in the ruff and in the ear openings give rise
to a vertical asymmetry in the directionality of the two ears: the left ear is
more sensitive to high-frequency sounds from below and the right ear from
above the horizontal plane. If the ruff feathers are removed, the owl is quite
unable to locate sounds in elevation and always faces horizontally regard-
less of the true elevation of the source, but it can still locate in azimuth.
When a source of sound is not directly in front of or behind the head,
sound will reach the two ears at slightly different times. In order to test the
importance of these time differences, sound stimuli are delivered using
miniature earphones installed in the owl’s ear canals, rather than sound
delivered from a distant speaker, which would inevitably produce
differences in both time and intensity. With the earphones, stimuli can be
made equal in intensity but different in time of arrival at the two ears. An
owl responds to such a stimulus by turning its head horizontally in a direc-
tion that the time difference would represent if it were an external source of
sound. The owl turns its head to the side that receives the stimulus earlier,
and the angle of turning is positively correlated with the magnitude of the
time difference between the ears (Moiseff & Konishi, 1981). These experi-
ments show that the owl depends mainly on differences in time of arrival at
the two ears to locate sounds in azimuth.
6.3 Auditory interneurons and sound localisation
The fact that a barn owl can localise sounds accurately under open-loop
conditions implies that its auditory system can recognise each point in
space from a unique combination of the cues that specify azimuth and ele-
vation. Interneurons that respond to such complex combinations of stimu-
lus features are found among higher-order neurons in the midbrain. The
main auditory region of the midbrain in owls is situated on the inner edge
of the optic tectum (Fig. 6.4a) and is divided functionally into an inner and
Figure 6.4 The neuronal map of auditory space in the midbrain of the barn
owl. (a) The left side of the owl’s brain, showing the location of the auditory
midbrain (in bold outline) on the inner side of the optic tectum. (b) The left
auditory midbrain enlarged from (a), with the co-ordinates of the neuronal
map indicated in degrees of azimuth (L and R) and of elevation (⫹ and ⫺)
of auditory space. (c) A plot of auditory space in front of the owl in degrees
of azimuth and of elevation, showing the receptive fields (bold rectangles)
of ten neurons recorded in three separate electrode penetrations. The pene-
trations were made with the electrode parallel to the transverse plane at the
positions indicated by the arrows linking (c) to (b). (Modified after Knudsen,
1981.)
Auditory interneurons and sound localisation 137
outer portion. In the inner portion, the interneurons are tuned to particu-
lar frequencies and are arranged according to their best frequency, which is
termed a tonotopic arrangement. The great majority of these neurons
respond to their best frequency regardless of where the sound source is
located in space.
However, in the outer portion of the auditory midbrain, there are inter-
neurons that respond quite differently; all have similar best frequencies
near the upper end of the owl’s range (6–8 kHz) and respond only when the
sound originates from a specific region of space. These are therefore called
space-specific neurons. When recording from such a neuron with a micro-
electrode, the size and shape of the specific region to which it responds,
termed the receptive field, can be determined with the movable speaker
(Fig. 6.2a). Typically, the neuron is excited by sounds coming from a region
of space shaped like a vertically elongated ellipse. Unlike most auditory
neurons, the space-specific neurons are insensitive to changes in intensity,
and even a 20 dB increase in intensity has little effect on the size of the
receptive field.
As the recording electrode is advanced through the outer portion of the
auditory midbrain, it samples neighbouring interneurons, which are found
to have receptive fields representing neighbouring regions of space. In fact,
the space-specific neurons are arranged systematically according to the
azimuth and elevation of their receptive fields, so that they form a neuronal
map of auditory space (Fig. 6.4b, c). Two-dimensional space is mapped on
to the auditory midbrain, with azimuth being arrayed longitudinally and
elevation being arrayed dorsoventrally. On each side of the brain, the map
extends from 60° contralateral to 15° ipsilateral, but a disproportionately
large number of neurons is devoted to the region between 15° contralateral
and 15° ipsilateral. This arrangement means that the 30° of space directly in
front of the owl is analysed by an especially large population of neurons on
both sides of the brain. In elevation, the map extends from 40° upward to
80° downward but the majority of neurons are devoted to the region below
the horizontal (Fig. 6.4c).
Partially blocking one ear, which changes the normal intensity
differences between the ears, causes a significant vertical displacement of
the receptive field in the space-specific neurons but makes little difference
in azimuth. Changes in time differences between the ears can be delivered
with the miniature earphones, and it is found that individual neurons
respond only to a narrow range of time differences, which correspond with

the horizontal location of the neuron’s receptive field. Neurons that
respond to small time differences have receptive fields in front of the face
and those that respond to larger time differences have receptive fields at
greater angles to the face. This shows that the azimuthal position of the
receptive field is determined largely by time differences between the ears
(Moiseff & Konishi, 1981).
Thus, the acoustic cues that are used to create the receptive fields of these
auditory interneurons are exactly the same cues as the intact owl depends
on for sound localisation. This fact makes it almost certain that the neuro-
nal map of auditory space in the midbrain underlies the barn owl’s skill in
open-loop localisation. Certainly, the greater density of space-specific
neurons devoted to the 30° arc in front of the face would account for the
owl’s greater accuracy in locating sounds in this region. The large propor-
tion of the neuronal map that is devoted to the region of space somewhat
below and in front of the face also makes good functional sense because
this is likely to be the prey-containing region for an owl scanning the
ground from its perch.
Barn owls normally use both sight and hearing to track their prey, and in
keeping with this a combined auditory and visual map of space is found in
the optic tectum. The auditory map in the tectum is derived by topograph-
ical projection from that in the auditory midbrain and is precisely aligned
with the visual map of space derived from the retina. The mutual alignment
is adjusted through sensory experience early in the life of the owl (Knudsen,
1983, 1998). In turn, this sensory map in the optic tectum is linked via
neural circuits in the hindbrain to the motor circuits responsible for gener-
ating head movements (Masino & Knudsen, 1990).
6.4 Synthesising a neuronal map of auditory space
Each space-specific neuron in the owl’s midbrain responds to stimuli deliv-

ered via the earphones only when both the time difference and the inten-
sity difference fall within the range to which it is tuned. It is not excited by
either the correct time difference alone or the correct intensity difference
alone. Evidently, the receptive fields are formed by tuning of the neurons to
specific combinations of time differences and intensity differences, which
are coded separately by lower-order neurons. The initial separation takes
Synthesising a neuronal map of auditory space 139
place at the level of the earliest staging post of the auditory pathway
in the brain, consisting of the two cochlear nuclei, the angular nucleus and
the magnocellular nucleus.
The information available to the brain is coded in spikes generated by the
sensory neurons of the owl’s inner ear. Each of these auditory neurons
responds to a particular frequency of sound, and the neurons produce their
spikes at or near a particular point on the arriving sound wave. This latter
property is called phase locking and it is important for measuring a sound’s
time of arrival accurately. The number of spikes generated on each occa-
sion is proportional to the sound pressure level at the ear. Thus, the train of
spikes travelling along each axon of the auditory nerve carries information
about both the time of arrival and the intensity of a particular sound fre-
quency. When it reaches the brain, each of these sensory axons divides into
two branches; one enters the angular nucleus and the other enters the mag-
nocellular nucleus.
The role played by these two nuclei is revealed by an ingenious experi-
ment in which a tiny amount of local anaesthetic is injected into one of
them so as to inactivate most of its neurons (Takahashi, Moiseff & Konishi,
1984). The responses of the space-specific neurons are then re-examined to
see what changes have occurred. The results are clear cut: inactivation of
the angular nucleus alters the responses of the space-specific neurons to
interaural intensity differences without affecting their responses to time
differences, and inactivation of the magnocellular nucleus alters the
neurons responses to time differences without affecting their responses to
intensity differences. When recordings are made from the interneurons in
these two nuclei, their properties are found to be consistent with this result.
Neurons of the magnocellular nucleus preserve the phase locking shown by
the sensory neurons but are insensitive to changes in intensity, whereas the
neurons in the angular nucleus are sensitive to intensity changes but do not
show phase locking. Evidently, these two cochlear nuclei serve as neural
filters, which pass along information about either time of arrival or inten-
sity, but not both.
The first place in the auditory pathway at which information from both
ears is compared is the lamina nucleus, which receives excitatory input
from both the left and right magnocellular nuclei. It seems clear from the
anatomy and physiology of the lamina nucleus that this staging post serves
to measure interaural time differences. The laminar neurons are arranged
Figure 6.5 A neuronal circuit for measuring interaural time differences in

the barn owl, shown as a highly diagrammatic section through the brain.
The left laminar nucleus receives excitatory input from both the left and
right magnocellular nuclei, represented here by a single axon from each.
When spikes conducted along the left and right magnocellular axons reach
a given laminar neuron simultaneously, that neuron will be strongly excited.
This will happen whenever the difference in a sound’s arrival time at the
two ears compensates for the difference in time taken for spikes to travel to
that laminar neuron along the left and right magnocellular axons. (Modified
after Konishi, 1992.)
in an elongated array, and the axons of the ipsilateral magnocellular

neurons pass along this array from one end whereas the contralateral mag-
nocellular axons pass along from the other end (Fig. 6.5). Both ipsilateral
and contralateral axons give off terminal branches that make synaptic
contact with the laminar neurons.
This circuit is able to compute interaural time differences on the princi-
ple of delay lines and coincidence detection. The laminar neurons fire max-
imally when they receive excitatory input simultaneously from both
ipsilateral and contralateral magnocellular neurons and so function as
coincidence detectors. The magnocellular axons function as delay lines
because the time it takes each spike to travel along the axon from one end
Synthesising a neuronal map of auditory space 141
Figure 6.6 A simplified flow diagram showing how the neuronal map of
auditory space is synthesised in the brain of the barn owl. The boxes on the
left represent successive regions of the brain, and the process that takes
place in each region is shown on the right of the corresponding box. The
arrows indicate the flow of information along pathways (left) and along the
sequence of computational steps (right). Note the separation of the time
and intensity pathways. (Modified after Konishi, 1992, 1993.)
of the lamina nucleus towards the other end causes a delay in the spike’s
time of arrival at a given laminar neuron. The point in the array of laminar
neurons where the left and right magnocellular spikes arrive simultane-
ously will, therefore, vary systematically as a function of the difference in a
sound’s time of arrival at the left and right ears plus the conduction time
along the magnocellular axons from the two ears. Consequently, for each
frequency band, the array of laminar neurons effectively constitutes a
neuronal map of interaural time differences (Carr & Konishi, 1990). The
axons of the laminar neurons convey this information forward to the audi-
tory midbrain, via the anterior part of the lateral lemniscal nucleus (Fig.
6.6).
The posterior part of the lateral lemniscus on each side of the brain
receives axons directly from the contralateral angular nucleus. This input is
excitatory and is in proportion to the sound pressure level at the contralat-

eral ear. At the same time, each lateral lemniscus receives inhibitory input
from the lemniscal nucleus on the opposite side of the brain and this
reflects the sound pressure level at the ipsilateral ear. So the response of the
lemniscal neurons depends on the balance between the degree of excita-
tion and the strength of inhibition, which in turn depends on the intensity
difference between the ears. Furthermore, the neurons of the posterior
lateral lemniscus vary systematically in the intensity difference that causes
them to respond maximally, forming a topographical array along the dorso-
ventral axis (Manley, Koppl & Konishi, 1988). In effect, this array constitutes
a neuronal map of interaural intensity differences for a given frequency
band.
This information is passed along by the axons of the lemniscal neurons to
the auditory midbrain, where it is eventually combined with the informa-
tion on interaural time differences to generate the receptive fields of the
space-specific neurons (Fig. 6.6). By this means, the neuronal map of audi-
tory space is synthesised centrally from sensory cues that are not them-
selves spatially organised. The inner ear is organised tonotopically, with the
topographical array of receptors coding frequency rather than space. For
each frequency band, the ear monitors intensity and time of arrival, cues
that are separated in the brain and processed in parallel pathways. In both
pathways, the information from the left and right ears is brought together
to form a topographical array of interaural differences. Finally, the intensity
and time pathways are brought together, thereby enabling neurons in the
auditory midbrain to encode specific combinations of time and intensity
differences.
6.5 The echolocation sounds of bats
Whereas owls locate prey by listening passively to the noises produced by

the prey, insectivorous bats actively interrogate the environment using the
technique of echolocation (Fig. 6.7). For this purpose, a flying bat produces
a succession of loud calls, each of which consists of a brief pulse of sound.
These sounds are often described as ultrasonic because they contain fre-
quencies (20–200 kHz) beyond the range of human hearing. The sound
pulses travel out in front of the bat and, when they encounter a target, are
reflected back and picked up by the ears. The essence of echolocation lies
The echolocation sounds of bats 143
Figure 6.7 A greater horseshoe bat (Rhinolophus ferrumequinum) about to

capture a moth in flight. The complex structure between the ears and the
mouth is the nose-leaf, through which echolocation sounds are emitted in
horseshoe bats. Insects are usually captured by being scooped up in the
wing membrane rather than being seized in the mouth. (After a photograph
by S. Dalton.)
in the ability of the brain to reconstruct features of the target, most impor-
tantly its position in space, by comparing the neural representation of the
echo with that of the original signal.
Analysis of the sound pulses used for echolocation reveals what is at first
sight a bewildering variety of form as one compares one species with
another, but it has become clear that there are basically only two kinds of
sound signal used by bats. The first kind are broadband signals, which
consist of short pulses, less than 5 ms in duration, that are frequency mod-
ulated (FM). An example of this kind of signal is found in the mouse-eared
bat (Myotis) in the widespread family Vespertilionidae: each pulse starts at
a high frequency and sweeps downwards in frequency during the course of
the pulse (Fig. 6.8a, b). At any given instant, the sound within the pulse is a
fairly pure tone, corresponding to the fundamental frequency generated by
the larynx, with traces of a second harmonic towards the end of the pulse.
Figure 6.8 Echolocation sounds of the mouse-eared bat (Myotis myotis).

(a) A single frequency-modulated pulse, recorded from a bat in flight (left)
and displayed on an oscilloscope (right). (b) Computer-generated sonagram
of a single pulse, showing the downward sweep in frequency in more detail.
The computer generates the sonagram by plotting curves of the relative
intensity of different frequencies at successive intervals of time during the
sound pulse. This pulse was emitted by the bat at a distance of 4 m from the
target; (c) and (d) are sonagrams of pulses emitted respectively at 36 cm and
at 7 cm from the target. (a modified after Sales & Pye, 1974; b–d from
Habersetzer & Vogler, 1983.)
But the downward sweep results in the pulse having a total bandwidth of
some 60–70 kHz.
Frequency-modulated pulses appear to have evolved in bat echolocation
because the wide range of frequencies makes them suitable for target
The echolocation sounds of bats 145
description and accurate ranging. Behavioural tests show that vespertili-

onid bats can discriminate between targets that differ only in distance
(range) with an accuracy of 10–15 mm. When these bats are tested with a
pair of loudspeakers, each producing an electronically synthesised ‘echo’
after each echolocation pulse, they can discriminate differences between
the speakers down to about 60 µs, which corresponds to a difference in
target range of about 10 mm. Hence, it seems probable that bats estimate
target range from the time it takes sound pulses to travel out to the target
and return as echoes, just as radar sets do. This is confirmed by studies of
the brain, which show that bats take advantage of FM signals for target
ranging by making multiple estimates of pulse–echo delay with interneu-
rons tuned to different frequencies within the FM sweep (see section 6.8).
The second basic kind of echolocation sound used by bats consists of
narrow band signals, in which the sound has a constant frequency (CF).
These signals are generally longer, between 10 and 100 ms in duration, and
form part of an alternative strategy of echolocation employed by many
species. This alternative is particularly well developed in the horseshoe
bats, which are members of the specialised family Rhinolophidae (see Fig.
6.7). The echolocation sound of the greater horseshoe bat (Rhinolophus fer-
rumequinum) consists mainly of a long component (about 60 ms) with a
constant frequency of just over 80 kHz, which is followed by a brief down-
ward frequency-modulated sweep and is often preceded by an even briefer
upward sweep (Fig. 6.9).
A long CF signal is unsuitable for target description but is well suited to
measuring the Doppler shift, which is the shift in sound frequency experi-
enced by an observer listening to a moving sound source such as a passing
train. That horseshoe bats actually perceive the Doppler shifts generated
during echolocation is shown clearly by the way they modify their sounds
in flight. If a horseshoe bat is trained to fly down a long room to a landing
platform, it is observed to alter the sound frequency of its echolocation
pulses so as to keep the Doppler-shifted echoes from the landing platform
at a constant, species-specific frequency, which is 83 kHz for the greater
horseshoe bat.
In human affairs, measurement of the Doppler shift in a radar signal
provides an accurate estimate of the relative velocity of a moving target;
radar measurement of motorists’ speed is perhaps the most familiar
example. Similarly, bats using a long CF signal are able to determine the
100
Sound frequency (kHz)
75
50
25
1 2 43 45 57
Time (ms)
Figure 6.9 Echolocation sound of the greater horseshoe bat (Rhinolophus
ferrumequinum). The computer-generated sonagram shows the three com-
ponents of the long pulse: the initial, upward sweep in frequency; part of
the long constant-frequency component (note breaks in time axis); and the
final, downward sweep in frequency. The faint traces of the fundamental
frequency at around 40 kHz indicate that the broadcast frequency is actu-
ally the second harmonic. (From Neuweiler, Bruns & Schuller, 1980.)
relative velocity of their prey by comparing the frequency of the outgoing

pulse with that of the returning echo. In addition, it has been found that
horseshoe bats are able to perceive the relatively small Doppler shifts in
echo frequency produced by the beating wings of a flying insect. They use
this as a means of detecting insect prey in the face of the extensive echo
clutter produced by dense foliage or other background objects (see
section 6.9).
Constant frequency and frequency-modulated sound pulses thus repre-
sent two different strategies for extracting information from the environ-
ment by means of echolocation. It is evident that Rhinolophus depends
largely on the former, whereas Myotis employs the latter exclusively.
However, there are a number of other genera that employ a mixture of the
two strategies and emit pulses in which both CF and FM components are
well developed. One example that has been the subject of a detailed neuro-
ethological study is the moustached bat (Pteronotus) in the family
Mormoopidae.
Interception of flying prey by bats 147
6.6 Interception of flying prey by bats
The way in which bats use their echolocation signals to track prey has been
studied by combining high-speed photography with tape recordings of the
normal sequence of airborne interception. These observations are neces-
sarily made under controlled conditions but they have been supplemented
by numerous, and increasingly exact, observations made in the field. All the
species studied so far follow a fairly standard routine, which can be broadly
divided into three stages, known as the search, approach and terminal
stages. Much the same sequence is followed whether the bat is intercepting
prey, avoiding an obstacle or landing on a perch.
During the search stage, a bat emits sound pulses of a constant, species-
specific form at a low repetition rate of about 10 Hz or less, as described in
the previous section. As its name implies, the main function of the search
stage is to detect potential prey or obstacles. Behavioural tests with bats
using FM signals show that they can detect a small sphere (0.5 cm diame-
ter) at almost 3 m and a larger sphere (2 cm diameter) at over 5 m.
Calculation of the intensity of echoes returning from targets at these dis-
tances suggests that the maximum range of echolocation corresponds with
the threshold of hearing in bats.
However, in natural interception, bats do not visibly react to targets at
these distances nor do they appear to react to progressively larger targets at
progressively greater distances. Instead, the onset of the approach stage,
which represents the first visible reaction of the bat to the target, occurs
when the bat is between 1 and 2 m away in nearly all cases. Therefore, it is
probable that the approach stage begins at a critical distance between the
bat and the target and does not necessarily begin when the target is first
detected. The transition to the approach stage is marked by the bat turning
its head, especially its ears, directly towards the target and by an increase in
the repetition rate of the echolocation sounds to a value of about 40 Hz. In
bats such as Myotis, which use only FM pulses, the pulses become shorter
but the slope of the FM sweep becomes steeper so that the bandwidth of
the signal is maintained (see Fig. 6.8c). Species that use long CF pulses for
Doppler shift echolocation do not drop the CF component during the
search stage but it becomes shorter and the small FM component increases
in bandwidth.
There is thus a shift towards brief, FM pulses at the approach stage in all
species of echolocating bats. It is probable that bats are taking advantage of

the greater information content of broadband signals as they approach the
target, especially because the decision about whether or not to catch an
item of potential prey is evidently made during the approach stage. When a
mixture of living insects and similar-sized plastic discs or spheres is thrown
into the air for bats to catch, the bats break off pursuit of the plastic objects
at the end of the approach stage. Again, bats trained to discriminate
between two targets in flight make their choice, as judged by the orienta-
tion of their ears, towards the end of the approach stage. The increase in
repetition rate of the echolocation sounds is also understandable as a
response to the need for increasingly frequent estimates of range and direc-
tion as the bat closes upon the target. When the approach stage goes to
completion, it normally takes the bat to within 50 cm of the target.
The transition to the terminal stage is marked by an abrupt increase in
pulse repetition rate, which rises to about 100 Hz or even 200 Hz in some
cases. This increased rate clearly provides a rapid updating of information
about the target’s position as the bat makes its final manoeuvres to capture
the target. In most species, the pulses emitted during the terminal stage are
FM sweeps, often with several harmonics, that are 0.5 ms or less in duration
(see Fig. 6.8d). Only in horseshoe bats, and others that exploit the Doppler
shift, is a CF component retained during the terminal stage; even then, the
CF component is reduced to a length of about 10 ms or less.
Bats do not usually capture flying insects in their mouths but rather use
their outspread wing or tail membranes as a scoop for collecting the prey.
The accuracy with which this is accomplished is illustrated by experiments
in which horseshoe bats are fed on flour-covered mealworms thrown into
the air. Flour marks are then found on the wing membrane at the base of
the third to fifth digits, in an area with a diameter of 2 to 3 cm (cf. Fig. 6.7).
This suggests that the bats are consistently able to localise prey in space to
within about 1 cm3.
6.7 The auditory system and echolocation
When echolocation sounds return as echoes to a bat, they are received by

an auditory system that conforms to the general mammalian pattern. The
sounds are collected by the external ear and enter the ear canal, where they
impinge on the tympanum (Fig. 6.10). The vibrations of the tympanum are
The auditory system and echolocation 149
Figure 6.10 The middle and inner ear of a bat; diagram based on a horizon-
tal section. The external ear opening is off picture to the right. Structural
elements of the middle and inner ear are shown in solid black; the sur-
rounding bone of the skull is shown in stipple. Note that the bones of the
middle ear are acted upon by various muscles, known collectively as the
middle ear muscles. (Redrawn after Sales & Pye, 1974.)
transmitted by the bones of the middle ear to the oval window of the inner
ear. From here, the vibrations travel through the cochlea along the basilar
membrane, which forms a helical ribbon, wide at the apex of the cochlea
and narrow at its base. Transduction takes place in receptor cells, the hair
cells, which are distributed along the length of the basilar membrane. The
receptor potentials that are produced in the hair cells by the vibration of the
basilar membrane are transmitted across synapses to first-order neurons
with axons that carry spikes to the brain along the auditory nerve.
The information that is provided by the traffic of spikes in the auditory
nerve is processed through a sequence of levels in the mammalian brain, as
it is in other vertebrates. The auditory pathway consists of a rather complex
network of brain regions but may be illustrated by three main staging posts.
Firstly, there is the cochlear nucleus in the hindbrain, which receives input
from the auditory nerve and contains mainly second-order neurons.
Secondly, there is the inferior colliculus, which is one of a number of impor-
tant nuclei in the midbrain containing higher-order auditory neurons.
Thirdly, there is the auditory cortex in the forebrain, which is the final
staging post of the auditory pathway. Echolocation is possible because a
bat’s auditory system has several striking specialisations that enable it to
receive and analyse faint echoes. These specialisations start at the periph-
eral level, as described below.
The first of these specialisations is that a bat’s hearing is particularly sen-
sitive to sounds that have similar frequencies to its own echolocation
pulses. This is shown by testing bats with sounds of different frequencies
and measuring the auditory threshold, which is the lowest sound intensity
that elicits a detectable response. The results are expressed as threshold
curves, in which the sound pressure level at threshold is plotted against fre-
quency (Fig. 6.11). In bat species that use broadband (FM) signals in the
search stage, it is found that the frequencies with the lowest threshold coin-
cide with the dominant frequencies in the echolocation signal (Fig. 6.11b).
Apart from this feature, the threshold curve is not very different from that
of a non-echolocating fruit bat (Fig. 6.11a). Nor is the absolute threshold of
hearing exceptional by mammalian standards. In bat species that use
narrow band (CF) signals in the search stage, the threshold curve is much
more sharply tuned to a narrow frequency band. In Rhinolophus, for
example, hearing is very sharply tuned (Fig. 6.11c) and the echoes are kept
close to this best frequency by the Doppler-shift compensation.
A second specialisation of the peripheral auditory system is that echolo-
cating bats have highly directional hearing, in contrast to most mammals,
which have good all-round hearing. This is shown clearly by behavioural
tests carried out on restrained horseshoe bats. A loudspeaker directly in front
of the bat’s head produces ‘echoes’ with an electronically shifted frequency
following each of the bat’s CF pulses, and the bat then responds by compen-
sating for this apparent Doppler shift. The directionality of hearing is tested
by presenting sounds at the normal echo frequency from another speaker at
different angles around the bat’s head. When this second sound is perceived
by the bat, it effectively masks the first sound and the bat does not show the
compensation response. Hearing proves to be most sensitive directly in front
of the head, and sensitivity falls off by about 45 dB from the midline to the
side. A similar fall in sensitivity occurs at angles below the horizontal but the
drop is less severe above the horizontal. Consequently, returning echoes are
useful only if they fall within a narrow cone in front of the head, not more
than 30° off the direction of flight (Grinnell & Schnitzler, 1977).
The auditory system and echolocation 151
Figure 6.11 Hearing threshold curves for three species of bat, derived from
recordings of summed potentials in the inferior colliculus; thresholds are
expressed as sound pressure level in decibels (dB SPL), as a function of
sound frequency. (a) Curve for a non-echolocating fruit bat from southern
India. (b) Curve for an echolocating bat from the same locality, with a son-
agram of the echolocation signal shown above for comparison (rotated
through 90°). Note the similarity of the curves in (a) and (b), apart from the
tuning of the curve in (b) to the strongest frequencies in the echolocation
signal (24 to 26 kHz). (c) Curve for the greater horseshoe bat (Rhinolophus
ferrumequinum) showing the sharp tuning to 83 kHz and the notch of
insensitivity to frequencies just below this. (a and b modified after
Neuweiler, Singh & Sripathi, 1984; c modified after Neuweiler, 1983.)
A third specialisation for echolocation in bats is that the peripheral audi-

tory system shows a reduced sensitivity to the emitted pulse of sound. The
echolocation sounds emitted by bats are very intense, with a sound pres-
sure level of around 110 to 120 dB when measured 5 to 10 cm in front of the
head. If the bat’s auditory system were directly exposed to such intense
sounds, it would not be able to recover fully by the time the echo arrived. In
bats using brief FM signals, such as Myotis, this problem is overcome by
contraction of the middle-ear muscles, which partially uncouples the inner
ear from the vibrations of the tympanum (see Fig. 6.10). During echoloca-
tion, these muscles begin to contract before each sound pulse and develop
maximal tension at the onset of sound emission; they then relax very
rapidly, within about 8 ms, so that the response to the echo is not attenu-
ated except at very close range. In Myotis, the attenuation produced by
contraction of the middle-ear muscles is some 20 to 25 dB.
In addition, neural attenuation takes place in the midbrain prior to the
inferior colliculus. Recording with simple wire electrodes shows that the
summed potential elicited by the emitted sounds is smaller in the midbrain
than in the cochlea nucleus, but this is not the case for external stimuli.
Hence, there must be a central, neural mechanism that attenuates self-
stimulation by the emitted sounds. This lasts only for a short time and
echoes returning after 4 ms are not affected. The magnitude of the neural
attenuation averages about 15 dB in Myotis, and so the total attenuation
available from both mechanical and neural mechanisms amounts to some
35 to 40 dB.
Bats that use long CF signals, such as Rhinolophus, cannot exploit these
mechanisms because the long outgoing pulse overlaps the returning echo.
However, when the bat is on the wing, the emitted sound normally has a
lower frequency than the echo, which is kept at a constant frequency by
Doppler-shift compensation. Consequently, Rhinolophus is able to solve
the problem of self-stimulation simply by being rather deaf at the normal
emission frequency, which is a few kiloHertz below the echo frequency (Fig.
6.11c).
6.8 Auditory specialisations for echo ranging
The mechanisms of attenuation outlined above are particularly important

in facilitating accurate measurement of the echo delay, and hence of dis-
tance, because they prevent the auditory system from being overloaded by
the outgoing sound pulses. In order to measure echo delay, the bat’s audi-
tory system must be capable of resolving very small time intervals. Most
mammals are unsuited to this task because their auditory systems take too
Auditory specialisations for echo ranging 153
long to recover between successive sound stimuli – usually some tens of

milliseconds need to elapse after a pulse of sound before full sensitivity is
recovered. But in echolocation, most echoes will return within 30 ms of the
outgoing pulse, given that bats track prey at distances of no more than 3 or
4 m and that sound travels at 334 ms⫺1.
It is therefore not surprising that the auditory neurons of bats are found
to recover rapidly when stimulated with paired pulses of sound, roughly
resembling the natural pairing of pulse and echo. For instance, summed
potentials recorded from the midbrain of Myotis show some response to a
second pulse that follows the first after only 0.5 ms and full recovery is
achieved in 2 ms. Similar tests of hearing in fruit bats have shown that most
species are slow to recover from the first pulse, but rapid recovery is found
in one genus, Rousettus, which has evolved an echolocation capability
independently of the insectivorous bats. Because this is the only respect in
which the hearing of Rousettus differs conspicuously from that of its non-
echolocating relatives, this rapid recovery may represent the most funda-
mental adaptation of the auditory system for echolocation (Grinnell &
Hagiwara, 1972).
There is more to it than this, however, if a bat is to measure the arrival
time of an echo rather than merely be aware of its existence. For echo delay
to be coded by the auditory system as a time interval, some classes of inter-
neurons must be able to act as accurate time markers. Interneurons that
appear to be specialised as time markers for echolocation have been found
in the inferior colliculus of bats belonging to several different genera. In
fact, the majority of neurons in the inferior colliculus are found to have
suitable properties, the most important of which is that their response is
highly phasic. When one of these neurons is stimulated with a pair of FM
pulses, it responds to each sound pulse by generating a single spike, or at
the most two (Fig. 6.12a).
Repeated presentation of the stimulus shows that the delay between
stimulus and response, the response latency, remains very consistent from
one presentation to the next. The neuron’s recovery is also sufficiently rapid
that its response to the second pulse is essentially independent of that to
the first pulse. As a result, the time interval between the two pulses is coded
with remarkable precision in the pattern of spikes in the interneuron (Fig.
6.12a). Furthermore, the response latency remains almost constant regard-
less of the intensity of the sound stimulus, which is a crucial specialisation
Figure 6.12 Properties of time-marking interneurons in the inferior collicu-

lus of bats. (a) An oscilloscope display showing a sound stimulus, consisting
of a pair of identical, frequency-modulated pulses 6 ms apart (above), and
the interneuron’s response (below). Each dot represents the occurrence of a
single spike in the responding neuron, and the pair of dot columns is gener-
ated by 16 consecutive presentations of the stimulus. (b) A similar display
showing the response of a collicular neuron to a single frequency-modu-
lated pulse, consisting of a downward sweep from 40 kHz to 20 kHz,
stretched over three different durations (shown diagrammatically, above).
The interneuron’s response (dot columns, below) shifts in register with the
occurrence of 25 kHz in the FM sweep (indicated by the arrowheads,
above). This suggests that the neuron is responding to this particular fre-
quency in the stimulus. (a from Pollak, 1980; b from Bodenhamer, Pollak &
Marsh, 1979.)
for time marking in echolocation. The rate of recovery remains fast enough
for the neurons to respond consistently to the second pulse, at short pulse
intervals, even when the first pulse is as much as 30 dB louder than the
second (Pollak et al., 1977; Pollak, 1980).
Each of these interneurons shows an exceptionally sharp tuning to a par-
ticular frequency within the FM sweep used for echolocation. This can be
seen in their threshold curves, which have nearly vertical slopes on either
side of the best frequency rather than the more usual V-shaped curve. As
soon as the intensity of the appropriate frequency rises a few decibels
above threshold, the neuron produces its single spike. Consequently, the
neuron is fired promptly by the same frequency component in each sound
pulse and so locks on to each event with consistent precision. This is shown
nicely by stretching the sound pulse used as a stimulus over a longer time:
for each stimulus duration, the neuron responds at the same relative posi-
tion in the FM sweep (Fig. 6.12b).
Information about time of arrival provided by these neurons is passed on
up the auditory pathway, particularly to the auditory region of the cerebral
cortex. Neurons that are sensitive to time delays between the outgoing
pulse and the returning echo have been found in the auditory cortex of all
bat species tested. Such neurons are well suited to measuring the range of
a target, based on the information provided by the time-marking neurons
of the inferior colliculus. It is often the case that these cortical neurons are
selective for different frequencies in the pulse and in the echo. This may
reflect the fact that, during active flight, the echo will inevitably be Doppler
shifted to a higher frequency than the outgoing pulse. The equivalent point
in time for pulse and echo will therefore be indicated by separate time-
marking neurons in the inferior colliculus.
The delay-sensitive neurons of the auditory cortex respond vigorously to
paired FM pulses simulating natural pulse/echo pairs but hardly respond at
all to FM pulses presented singly or to CF pulses. The response consists of
a short train of spikes and occurs, in the majority of neurons, only if the
time interval between pulse and echo is appropriate. If one of these
neurons is presented with a sequence of pulse/echo pairs, in which the
echo delay is progressively reduced to simulate the bat’s approach to a
target, it responds strongly only to a narrow range of delays (Fig. 6.13a). By
systematically varying the echo delay and recording the neuron’s responses,
it can be seen that each of these neurons is sharply tuned to a particular
echo delay, termed the best delay (Fig. 6.13b). Some of these delay-tuned
neurons even respond preferentially to a narrow range of pulse-repetition
rates, which correspond to the approach stage in intercepting prey (Wong,
Maekawa & Tanaka, 1992).
Neurons having similar best delays are grouped together in inwardly
directed columns within a specific area of the auditory cortex (the location
of the auditory cortex is shown in Fig. 6.13c). In the moustached bat,
Pteronotus parnellii, these columns are arranged systematically with best
delay increasing along the cortical surface from anterior to posterior (Fig.
6.13d). The delay-tuned neurons are thus arranged topographically accord-
ing to the distance they encode and so provide a neural representation of
target range. This delay-tuned area of the auditory cortex is quite distinct
Figure 6.13 Echo-ranging neurons in the auditory cortex of bats. (a) A his-
togram (above) summarising the response of an echo-ranging neuron from
Myotis to a sequence of pulse/echo pairs (below) simulating the natural
approach to a target. (b) The response of another neuron from Myotis,
expressed as a percentage of the maximum number of spikes, as a function
of the delay between a simulated pulse and echo of constant amplitude.
(c) A diagram of the brain of Myotis, viewed from the left and above, showing
the auditory region of the cerebral cortex. (d) The FM area of the auditory
cortex in Pteronotus, showing the systematic distribution of echo-ranging
neurons according to their best delay. The solid lines labelled with a
number are contours of best delay in milliseconds. There are three horizon-
tal clusters of neurons, indicated by broken lines, each tuned to only one of
the three harmonics (H2, H3, H4) in the echo. (a from Wong et al., 1992;
b and c modified after Sullivan, 1982; d modified after O’Neill & Suga, 1982.)
from the tonotopic area. For reasons that are not yet understood, the audi-
tory cortex of Myotis is less highly differentiated: there is a tendency for
neurons with larger best delays to be located more posteriorly but this
hardly constitutes a neural map of best delay. Nor is the delay-tuned area
clearly separated from the neighbouring tonotopic area; in fact, there is a
considerable overlap (Wong & Shannon, 1988).
The echolocation pulses of Pteronotus contain three higher harmonics in
addition to the fundamental frequency, and each delay-tuned neuron
responds to only one of these three harmonics in the echo (H2, H3 and H4 in
Fig. 6.13d). By varying the intensity of the simulated echo when testing
these neurons, it has proved possible to measure the threshold of the
response at each delay tested and so to construct a threshold curve. A con-
spicuous result is that each of these curves has an upper threshold as well
as a lower one, which means that the neuron fails to respond if the echo is
too loud as well as if it is too quiet. Also, neurons tuned to shorter delays
tend to have a higher threshold coupled with a narrower delay range. Each
of the delay-tuned neurons is thus tuned to a particular combination of
time delay and intensity appropriate to echoes returning from a certain dis-
tance (O’Neill & Suga, 1982).
The best delays in the FM area of the auditory cortex of Pteronotus cover
a range from 0.4 ms at the anterior edge to 18 ms at the posterior edge (Fig.
6.13d), which corresponds to target ranges from 7 to 310 cm. This agrees
closely with the range over which bats are observed to detect and react to
targets (see section 6.6). An especially large number of neurons, reflected in
cortical surface area, is devoted to delays from 3 to 8 ms (50 to 140 cm), cor-
responding roughly with the approach stage of target interception.
Especially with the larger values of echo delay, it is obvious that the
neural response to the emitted pulse must be considerably delayed if it is to
reach the cortex at the same time as the response to the echo. In fact, a large
range of response latencies is found among the time-marking neurons of
the inferior colliculus. It is therefore possible that the neural pathways
through the colliculus are acting as delay lines and the cortical neurons are
effectively coincidence detectors for particular combinations of pulse and
echo latencies. The neural map of target range could thus be assembled in
a similar manner to the barn owl’s map of interaural time differences.
Because the FM area of the auditory cortex is sharply separated from the
tonotopic area in Pteronotus, it is possible to inactivate them separately
with drugs. As with the corresponding experiments in the barn owl’s brain
(see section 6.4), this provides a test of the behavioural function of these
regions. When the FM area is inactivated with the drug, fine discrimination
of target range is impaired, though coarse discrimination is still possible,
and frequency discrimination remains unaffected. This confirms that the
FM area is involved in the perception of distance, as expected from the
responses of its interneurons and their topographical arrangement, but it
appears to have little to do with frequency discrimination (Riquimaroux,
Gaioni & Suga, 1991). The opposite result is obtained when the tonotopic
area is inactivated with the drug, and this area appears to be specialised for
analysis of the Doppler-shifted CF signal.
6.9 Auditory specialisations for Doppler shift analysis
All echolocating bats need to be able to analyse sound frequencies accu-

rately. With FM signals, fine frequency analysis yields accurate range meas-
urement and detailed description of the target, and with CF signals it yields
an accurate measure of the Doppler shift. Frequency analysis is certainly
excellent in bats that use only FM signals, such as Myotis, but their abilities
are not so very different from those of other mammals. However, in bats
that use long CF signals, such as Rhinolophus, frequency resolution is
quite exceptional and greatly exceeds the abilities of non-echolocating
mammals.
Specialisations for fine-frequency analysis begin at the inner ear, with the
basilar membrane, which is the accessory structure that couples the sound
stimulus to the receptor cells (hair cells) in the cochlea (see Fig. 6.10).
Mammalian hair cells are tuned to particular frequencies according to their
position on the basilar membrane, typically with equal lengths of the mem-
brane being devoted to each octave. But in the greater horseshoe bat, the
representation of frequencies from 80 to 86 kHz is greatly expanded, and
the greatest expansion is found at the reference frequency of 83 kHz. This
expanded representation on the basilar membrane is reflected in the
number of first-order neurons that innervate the hair cells. Compared to
frequencies below 70 kHz, which are not involved in echolocation, frequen-
cies of the expanded region are over-represented approximately 10 times,
with the result that 21 per cent of all first-order auditory neurons represent
the frequency range from 80 to 86 kHz (Fig. 6.14a). By analogy with the
visual system, this expanded representation is termed an acoustic fovea.
Auditory specialisations for Doppler shift analysis 159
Figure 6.14 The acoustic fovea in the greater horseshoe bat (Rhinolophus
ferrumequinum). (a) A histogram of frequency representation among first-
order auditory neurons, showing the great over-representation of frequen-
cies around 83 kHz. (b) The sharpness of tuning, expressed as the Q10 dB
value, for neurons with different best frequencies in the cochlear nucleus,
showing the exceptional sharpness of tuning in neurons with best frequen-
cies around 83 kHz. See also Fig. 6.11c. (a redrawn after Bruns &
Schmieszek, 1980; b redrawn after Suga, Neuweiler & Moller, 1976.)
The fovea is almost certainly related to structural specialisations in the

basal part of the basilar membrane, where the highest frequencies are rep-
resented. These structural peculiarities abruptly disappear at a distance of
4.5 mm from the oval window, and the expanded frequency region is
located immediately beyond this critical point. As a result, the basilar mem-
brane acts as a mechanical filter, which tunes a disproportionate length of
the membrane to a narrow frequency band (Vater, Feng & Betz, 1985). A
consequence of this arrangement is that the first-order neurons innervat-
ing the hair cells within the foveal region are each extremely sharply tuned.
A measure of the sharpness of tuning is provided by the Q10 dB value, which
is the neuron’s best frequency divided by the bandwidth of its threshold

curve 10 dB above the minimum threshold. Very high Q10 dB
values of
between 50 and 200 are found in the foveal region and, in the region of
greatest expansion around 83 kHz, even values of over 400 are found (Fig.
6.14b). For frequencies below 70 kHz, the Q10 dB values fall below 20, which
is within the range found in other mammals.
This over-representation and sharp tuning are conserved at all higher
levels of the auditory pathway. During echolocation, Doppler-shift com-
pensation serves to clamp the echo of the CF component within this
expanded frequency range. In effect, this behavioural response creates a
constant carrier frequency, on which the small frequency modulations pro-
duced by the wing beats of flying insects are superimposed, so enabling
them to be analysed by the sharply tuned neurons. A similar combination
of peripheral tuning and Doppler-shift compensation has evolved indepen-
dently of the horseshoe bats in the moustached bat and this combination
therefore probably represents a general strategy for Doppler shift analysis.
The kinds of echo modulation produced by flying insects have been
examined by the simple expedient of echolocating them with a simulated
horseshoe bat, consisting of a loudspeaker broadcasting a pure tone of 80
kHz and a microphone. These tests show that the fluttering wings of an
insect produce a strong echo or acoustic glint only when they are approxi-
mately perpendicular to the impinging sound waves, which happens for a
short moment in each wing-beat cycle, but there are no acoustic glints from
non-flying insects. A glint consists of a momentary increase in echo ampli-
tude and a concomitant broadening of echo frequency, which represents
Doppler shifts caused by the movement of the wings with respect to the
sound source.
In echoes returning from a flying insect, glints modulate the echo at a
rate corresponding to the wing-beat frequency of the insect, and this is
termed the modulation frequency. The extent of frequency modulation
involved is termed the modulation depth and is generally between 1 and 2
kHz above or below the carrier frequency. By perceiving these glints, a
horseshoe bat should be able to distinguish with certainty between echoes
from a fluttering insect and echoes from inanimate objects. Neurons spe-
cialised to encode the acoustic glints produced by flying insects are found
among those processing the CF echo (foveal) frequencies in all the major
staging posts of the auditory pathway.
Auditory specialisations for Doppler shift analysis 161
At the level of the inferior colliculus, the over-representation of the CF

echo frequencies is actually enhanced, being approximately 24 times that
for frequencies below 70 kHz. Consequently, the normal tonotopic arrange-
ment is distorted by the substantial block of interneurons devoted to the CF
echo frequency range. Within this block, the neurons are very sharply tuned
and the great majority are extremely sensitive to small frequency modula-
tions. When stimulated with sinusoidal frequency modulations that sweep
as little as ⫾ 10 Hz around the 83 kHz carrier frequency, these neurons
respond with discharges that are phase-locked to the frequency modula-
tion. The response remains phase-locked at all modulation frequencies up
to about 500 Hz, but some neurons show a preference for rates between 20
and 100 Hz. The ability of these collicular neurons to follow the complex
modulations produced by real wing beats is confirmed by using the
recorded echoes from a flying moth as stimuli in the experiments. With this
natural stimulus, the neurons reliably encode the wing-beat frequency of
the moth, as well as more subtle features of the echo (Pollak & Schuller,
1981; Schuller, 1984).
The encoded information is passed on to the tonotopic area of the audi-
tory cortex, within which neurons devoted to the CF echo frequencies form
a disproportionately large block, often termed the CF area. Neurons in the
CF area respond to sinusoidal frequency modulations in much the same
way as the collicular neurons but are more selective in the range of modula-
tion to which their response is phase-locked. Most of them prefer modula-
tion depths around ⫾ 1 kHz and modulation frequencies of 100 Hz or less,
with a strong preference for frequencies between 40 and 70 Hz. Among the
nocturnal moths that are potential prey for horseshoe bats, many species
have wing-beat frequencies around 40 to 60 Hz, and such moths produce
acoustical glints with a modulation depth of about 1 kHz. The phase-
locking of the cortical neurons is thus tuned to a behaviourally relevant
range, in contrast to the wide range of responses found in the collicular
neurons.
Behavioural observations confirm that horseshoe bats do in fact detect
their natural prey by means of these modulations of the CF echo. Newly
caged horseshoe bats only pursue insects that are beating their wings and
ignore stationary insects or those walking on the sides of the cage. The
caged bats will take dead, tethered insects when these are associated with
an artificial wing-beat simulator placed nearby. The crucial role of the CF
signal is also shown by observations on horseshoe bats foraging in

flycatcher style in natural forests: while hanging on twigs, they scan the sur-
rounding area for flying insects and take off on a catching flight only after
the prey has been detected. During this stationary scanning for prey, the
bats emit long CF pulses without any FM component, indicating that prey
detection is based on the CF signal alone (Link, Marimuthu & Neuweiler,
1986; Neuweiler et al., 1987).
6.10 Conclusions
A hunting animal depends on adequate sensory systems in order to detect

its prey and to pinpoint the prey’s location in space. Bats and owls have
auditory systems that are dedicated to the task of detecting and localising
prey by means of sound. In both groups, specialisations are found at many
levels of the auditory pathway that enable spatial information about the
prey to be extracted from measurement of the intensity, frequency and time
of arrival of sound at the ears. Hearing in owls is adapted to process the
noises made by their prey because owls depend on listening only, but
hearing in bats is adapted to process the echoes of their own cries because
bats hunt and navigate by echolocation.
Bats employ two basic kinds of echolocation signal, which reflect
different strategies for extracting information about the prey from the
returning echoes. Frequency-modulated sounds are used for prey descrip-
tion and for estimating the distance to the prey, both of which are based on
accurate analysis of a wide range of frequencies. Distance is estimated from
the delay between sound emission and echo arrival: this time interval is
encoded by sharply tuned neurons that respond promptly to a given fre-
quency in the outgoing pulse or in the returning echo. This enables higher-
order neurons to be tuned to particular echo delays and so to transform the
peripheral responses into a neural map of target range.
Constant-frequency sounds are used for prey detection, whether in open
spaces or among dense clutter by picking out fluttering prey from the back-
ground. Analysis of CF echoes is based on an acoustic fovea, consisting of a
large number of auditory neurons that are very sharply tuned to a small
range of frequencies, combined with a behaviour pattern, Doppler-shift
compensation, which clamps the CF echo within this range. This combina-
tion makes the auditory system extremely sensitive to small Doppler shifts,
Further reading 163
the acoustic glints, imposed on the echo frequency by the wing beats of a
flying insect.
Barn owls are able to detect the angular direction of prey from sounds
made by the prey. The angle in azimuth is determined from the time of
arrival of the sound at the two ears, and the sound’s angle of elevation is
determined from intensity differences between the ears. This cue for eleva-
tion is made possible by the specialised arrangement of non-neural acces-
sory structures, the external ear openings and associated feathers. On the
basis of input from the two ears, auditory interneurons are tuned to partic-
ular combinations of time and intensity differences that correspond to par-
ticular locations in space. These neurons are arranged to form a neural map
of acoustic space, which underlies the owl’s skill in open-loop localisation
of the prey’s sounds.
In both bats and owls, the higher levels of the auditory pathway tend to
be organised functionally into discrete populations of interneurons, each
specialised for encoding a small set of acoustic parameters. Each set of
parameters represents a particular category of information that is directly
relevant to the animal’s hunting or navigational behaviour, such as echo
delay and intensity representing target range. Within each of these discrete
populations, the neurons are arranged topographically so that a particular
value of an acoustic parameter is encoded at a particular place in the brain.
Brain areas that encode sound frequency are arranged tonotopically, an
arrangement that is produced by simple topographical projection from the
array of receptors in the cochlea. But the representation of acoustic space
or target range is synthesised by neuronal interactions and does not arise
by topographical projection of the receptor array. Discrete parallel path-
ways extract different types of information from the simple frequency and
intensity coding carried out by the receptors. Individual neurons at the
higher levels then respond only to a specific combination of parameters
synthesised by neuronal interactions. Thus, they act as feature detectors for
specific values of behaviourally relevant information, such as angular loca-
tion or target range.
Further reading
Fenton, M.B., Racey, P. & Rayner, J.M.V., eds. (1987). Recent Advances in the Study
of Bats. Cambridge: Cambridge University Press. This book contains some
useful reviews of bat echolocation; in particular, Suthers and Wenstrup and

also Roverud show how much insight has been gained from behavioural
studies; O’Neill reviews target ranging in the auditory system; and Vater com-
pares the highly specialised mechanisms of frequency tuning in Pteronotus and
Rhinolophus.
Konishi, M. (1993). Listening with two ears. Sci Amer 268(4), 34–41. This article
explains how the neuronal map of auditory space in the barn owl is synthe-
sised from separate pathways processing time and intensity data from the ears.
Neuweiler, G. (1990). Auditory adaptations for prey capture in echolocating bats.
Physiol Rev 70, 615–41. This looks at the echolocation signals of bats and their
corresponding auditory specialisations, in the context of the bats’ natural
habitat.
Popper, A.N. & Fay, R.R. eds. (1995). Hearing by Bats. New York: Springer-Verlag.
Contained in this book are detailed reviews of major topics in bat echolocation.
The overview by Grinnell and the review of the auditory cortex by O’Neill are
the most relevant to the material covered in this chapter.
Suga, N. (1990). Biosonar and neural computation in bats. Sci Amer 262(6), 34–41.
The article deals almost exclusively with Pteronotus, describing its echoloca-
tion sounds and showing how both the FM and CF components are repre-
sented in the auditory cortex.
7 Programs for movement
7.1 Introduction
Understanding the mechanisms which generate and control locomotory

movements is fundamental to a complete knowledge of the neuronal
control of behaviour.We can regard locomotion, such as jumping, walking or
flying, as basic building blocks for much of an animal’s behavioural reper-
toire; and we can pose three basic questions about the control of such move-
ments. First, what mechanisms ensure that muscles contract in the
appropriate sequence? In walking, for example, the basic pattern is repeated
flexion and then extension of each leg, with flexion of the left leg coinciding
with extension of the right. Second, how does a nervous system select, initi-
ate and terminate a particular type of movement? For example, what initi-
ates the pattern of walking; and how is walking rather than running or
swimming selected? Third, how is the basic pattern for movement mod-
ulated appropriately? Stride pattern changes, for example, when a person
walks up a flight of steps or turns a corner.
Experimental approaches to these questions have often involved work on
invertebrates and lower vertebrates, animals in which the parts of the
nervous system that generate programs for movement contain a limited
number of neurons. This offers the opportunity to identify and characterise
all the components involved in generating a particular movement. A
specific question that has occupied many investigators is how to determine
the source of rhythmical activity that underlies many regularly repeated
movements, such as walking or flying. One possible source is propriocep-
tive reflexes, and muscles could be activated in a particular sequence by
joining reflexes in a chain. This view originated from Sherrington’s elucida-
tion of spinal reflexes in mammals and predominated for the first half of
the twentieth century. However, a number of experiments, particularly on
165
insects from 1960 onwards, showed that proprioceptive reflexes are not
required for the co-ordination of quite long, complex sequences of move-
ments. From these experiments emerged the concept that the central
nervous system contains central pattern generators, responsible for gener-
ating programs for movement.
Nowadays, it is widely acknowledged that timing cues for movements are
generated both within the central nervous system and by feedback from
proprioceptors. Because more than one mechanism operates, the overall
control of a rhythmical movement is rugged. Even within the central
nervous system, it now appears to be usual that central pattern generators
employ several mechanisms that operate in parallel to generate rhythms of
activity. One mechanism is by circuits, such as pairs of neurons that inhibit
each other so that excitation alternates between the two neurons. Another
way is by pacemaker neurons, which have the intrinsic property of gener-
ating regular, clock-like waves of depolarisation and repolarisation of their
membrane potentials.
A recent development is the realisation that circuits of neurons respon-
sible for generating movement patterns are often quite plastic. In the locust
flight system, for example, such plasticity allows the animal to adapt its
movements in the best way for maintaining its course. Another aspect of
plasticity is found when neurons participate in more than one type of
movement. This has been most extensively investigated in a small group
of neurons responsible for controlling movements of the teeth and foregut
of spiny lobsters and crabs, which have shown how neuronal circuitry can
be dramatically and rapidly reconfigured.
7.2 Locusts and their flight
A number of basic concepts of the way in which rhythmical movements are

generated have arisen from experiments on locust flight. Locusts are infa-
mous for their swarms, which can destroy enormous areas of crops in trop-
ical and subtropical regions. Study of locusts in the laboratory was first
promoted by the establishment of the Anti-Locust Research Centre in
London, and locusts have become a favoured animal in research for many
groups of neurobiologists (Burrows, 1996). Locusts are large insects, and
easy to maintain and breed in the laboratory. Their nervous system is
amenable to analysis at the single cell level. Locusts are capable of long
The flight engine 167
migratory flights, covering several hundred kilometres a day at heights

greater than 1000 m. In the laboratory, they can generate sustained flight-
like movements even when securely tethered to a bar, or dissected to allow
access to the nervous system.
7.3 The flight engine
As in most insects, flying is achieved by cyclical movements of four wings,

borne on the posterior two segments of the thorax. As the wings move up
and down, they twist so that they constantly generate lift to keep the insect
airborne (Fig. 7.1a). Each wing is moved by ten muscles, which can be
divided into three main groups (Fig. 7.1b). The first group contains just one
large muscle for each wing, oriented along the animal’s long axis. When
these dorsal longitudinal muscles contract, they distort the stiff cuticular
box structure of the thorax in a manner that causes the wing tips to move
downwards, and they are called ‘indirect depressor’ muscles. The other two
groups of muscles lie upright in the thorax and pull directly on the wing
base. One group pulls outside the fulcrum of the wing hinge, and so these
muscles are direct depressors, and the other group pulls on the inside of
the fulcrum and so these muscles are direct elevators (Fig. 7.1c). The three
largest direct depressor muscles of each wing control the way the wing
twists around its long axis, and are important in altering the pattern of wing
beat during steering manoeuvres.
The pattern of innervation of the wing muscles is straightforward
because each flight muscle is usually controlled by just one or two motor
neurons. A spike in a flight motor neuron mediates a rapid, strong twitch of
its muscle. The motor neurons make synapses along the length of each
muscle fibre, and these operate in the same way as synapses in the central
nervous system. When the presynaptic terminals are depolarised by the
arrival of a spike, each releases a tiny squirt of neurotransmitter, in this case
the amino acid glutamic acid. The ion channels in the muscle cell mem-
branes that bind glutamic acid cause large EPSPs. Through a series of
events, the electrical signal is transduced into the development of tension
by individual muscle fibres.
The simple pattern of innervation of locust flight muscles contrasts
with that found in most vertebrate skeletal muscles, where each muscle
is controlled by several tens or hundreds of motor neurons. It also
Figure 7.1 Movements and muscles involved in locust flight. (a) Movements
during one wing-beat cycle. The drawings show three stages during a down-
stroke; the hindwings move slightly before the forewings. Below, the path of
movement of the tip of the left forewing during one wing beat and the angle
of the wing during one wing-beat cycle are shown. The wing is twisted
during the upstroke to help maintain lift throughout the cycle. (b) and (c)
The main flight muscles of the third thoracic segment (bold outline in the
left drawing in a). The right half of the segment is shown in medial view (b)
and anterior view (c). Muscles 112, 128 and 129 are wing depressors;
muscles 113, 118 and 119 are wing elevators. The wing hinge is indicated by
an arrow in (c). (a redrawn after Pringle, 1975; b and c redrawn after
Snodgrass, 1935.)
The flight program 169
contrasts with the innervation of other muscles in locusts, such as those

which move the legs. Each leg muscle is innervated by only a few motor
neurons, but not all of them mediate fast, twitch-like contractions.
Some, the slow motor neurons, mediate small, individual twitches that
sum together gradually to develop graded, strong, slow contractions
when the motor neuron generates a train of spikes. Other motor neurons
are inhibitory and employ gamma aminobutyric acid as their neuro-
transmitter. The inhibitory motor neurons oppose the action of the slow
excitors by causing hyperpolarising postsynaptic potentials in their
muscle fibres. For further information about muscle innervation, see
Aidley (1998).
7.4 The flight program
Recordings of the electrical signals that cause contractions of a muscle are

called electromyograms (EMGs). The technique employed for recording
EMGs during locust flight is similar to that used for recording electrocardi-
ograms in humans, except that the electrodes are fine wires, inserted
through small holes in the cuticle and secured in place with glue or wax
(Fig. 7.2a). It is possible to fit a locust with a tiny radio transmitter to record
EMGs from pairs of muscles while it is flying unrestrained (Kutsch et al.,
1993). More usually, a locust is tethered, often to a solid bar but sometimes
to a harness attached to counterweights so that some of the forces gener-
ated by the wing movements can be measured. It is quite easy to induce a
suspended, tethered locust to flap its wings in a flight-like manner by
blowing a current of air over its head. The air current excites wind-sensitive
sensory hairs. During straight, level flight the wing-beat frequency is
between 15 and 20/s. The frequency tends to fall gradually during a long
flight, perhaps because the locust needs less power after it has burned
some of its fuel.
Each flight motor neuron usually produces one or two spikes per wing
beat, and excitation of elevator motor neurons alternates with excitation of
depressors (Fig. 7.2b). Over a range of different wing-beat frequencies, the
delay between excitation of the elevator and depressor motor neurons is
fairly constant (Fig. 7.2c), as is the duration of the depressor activity. The
most variable event during a wing-beat cycle is the duration of the burst of
spikes in elevator motor neurons (Fig. 7.2c). Hindwing depressors spike
Figure 7.2 The motor program for locust flight. (a) The locust is tethered to
a rod, and induced to fly by wind directed at the head. Fine wire electrodes
are inserted into flight muscles, and are attached to amplifiers (triangle
symbols). The positions of the leg attachments are shown as circles.
(b) Electromyograms from a depressor and an elevator muscle of a
forewing. Usually, two motor neuron spikes are registered in each muscle
per wingbeat cycle. In midcycle, small spikes are picked up from other
muscles as cross talk. (c) During a long flight, the gradual increase in the
duration of wing-beat cycles was due to an increase in the delay between
spikes in depressor and elevator motor neurons. The delay between spikes
in elevator and depressor motor neurons remained constant. (a modified
after Horsman, Heinzel & Wendler, 1983; b from Pearson & Wolf, 1987;
copyright © 1987 Springer-Verlag; c modified after Hedwig & Pearson, 1984.)
5ms before forewing depressors, although elevators of all four wings are
active synchronously. The result is that the hindwings are depressed before
the forewings in each wing-beat cycle, so that they are not moving into the
turbulent air caused by forewing movements.
Generation of the flight rhythm 171
7.5 Generation of the flight rhythm
An influential paper in the study of the neuronal control of movement was

by Donald Wilson, published in 1960. This challenged the prevailing view
that sequential activation of reflex loops generated rhythmical movements,
such as locust flight and walking in vertebrates. Locust wings bear a large
number of sense organs that report details of their movements. However,
Wilson showed that when he removed the wings and destroyed the sense
organs of the wing bases, a locust could still generate a pattern similar to
flight. This consisted of regularly repeating, alternating spikes in wing ele-
vator and wing depressor motor neurons. Either an air current directed at
the head or a series of randomly timed electrical shocks to the connective
nerves could elicit this flight-like activity. Flights are short in duration, and
the wing-beat frequency is about half that of an intact locust. Wilson con-
cluded that the basic program for generating flight movements is situated
in the central nervous system. Generation of the rhythm, or the co-ordi-
nated activity of different motor neurons in the correct sequence, does not
require proprioceptors to report details of the movements caused by
muscle contraction. Even if all the flight muscles and the head are removed,
the thoracic ganglia are capable of generating a slow, repeating pattern of
alternating excitation of elevator and depressor motor neurons.
Some of the neurons involved in generating the flight rhythm have been
identified and characterised by intracellular recording. The thorax is opened
dorsally to allow access to the thoracic ganglia, which are stabilised against
movements by supporting them with a metal platform (Fig. 7.3a). Further
stability is achieved by removing the legs and cutting most of the nerves that
supply the flight muscles. The nerve to one muscle, usually one of the dorsal
longitudinal muscles, is left intact so that EMGs recorded from this muscle
provide a monitor of the flight rhythm. This nerve also carries the axons of
most of the proprioceptors of the wing base. A locust prepared in this way will
produce sequences of flight-like activity when wind flows over its head, or if
the neurohormone octopamine is applied to the thoracic ganglia. There are
many differences in the detail of the pattern from that generated by intact
animals, and flight sequences are rather short in duration. However, the basic
pattern of repeated, alternating activation of depressors and elevator motor
neurons is present. As in an intact animal, there is a constant delay between
activity in elevator and depressor motor neurons, and a delay of 5 ms between
Figure 7.3 Intracellular recordings during fictive flight. (a) A method for
preparing the locust, which is opened mid-dorsally and the body walls
pinned, exposing the ganglia. Two glass capillary microelectrodes penetrate
neurons in the ganglia while wire electrodes record EMGs from muscle 112.
(b) Intracellular recordings from an elevator (el) and a depressor (dep)
motor neuron, together with an EMG recording from muscle 112 (lower
trace). (c) A phase-resetting experiment with interneuron 501. The upper
trace is an intracellular recording from the neuron and the lower trace is the
EMG from muscle 112 (when the neuron was stimulated with depolarising
current, the intracellular recording could not be registered). The arrowheads
indicate the times when bursts of spikes in the depressor muscle would
have been expected to occur without any stimulus to the interneuron.
(a and b modified after Robertson & Pearson, 1982; c modified after
Robertson & Pearson, 1983.)
spikes in motor neurons of the hindwing and forewing depressor muscles.

Flight-like activity in a dissected, immobile locust is called fictive flight.
At the start of a flight sequence, elevator motor neurons depolarise and
generate spikes first. In an intact animal, this would open the wings. During
flight, repeated smooth oscillations in membrane potential, up to 25 mV
in amplitude, are recorded from flight motor neurons (Fig. 7.3b).
Depolarisation of the elevator and depressor motor neurons alternates and
sometimes a motor neuron might hyperpolarise from its resting potential
between cycles of depolarisation.
7.6 Interneurons of the flight generator
None of the motor neurons contributes to the generation of the flight

rhythm, and one good functional reason is that motor neurons, and there-
Interneurons of the flight generator 173
fore muscles, can be used independently of each other. This is important

during steering, and during some movements of the legs in which muscles
that also move the wings are involved. The regular waves of depolarisation
that occur in motor neurons during flight must, therefore, originate in
interneurons and be communicated to the motor neurons by synaptic
transmission. Mel Robertson, Keir Pearson and others have characterised
many different interneurons which show cycles of rhythmical activity,
similar to those in motor neurons, during fictive flight. All of these inter-
neurons generate spikes, and the interneurons generally produce a greater
number of spikes per wing-beat cycle than the motor neurons. At the end
of an experiment, stain is injected into a neuron so that its structure can be
examined. In the thorax, about 85 different types of interneuron that are
active during flight have been distinguished. All of the interneurons exist as
bilateral pairs. Some probably belong to groups in which individual
members have not been distinguished from each other, so that the total
number of interneurons involved in the flight rhythm probably exceeds
100. The interneurons that form the central pattern generator for flight are
distributed among several ganglia, and most of them branch extensively in
both the second and third thoracic ganglia. In male crickets, which have
very similar flight machinery to locusts, the forewing muscles are used for
singing as well as for flying, and different interneurons are responsible for
the two motor programs (Box 7.1).
There are two criteria for establishing whether an interneuron is part of
the central pattern generator. First, it must be rhythmically active at the
flight frequency; and, second, injection of a brief pulse of current into it
should reset the flight rhythm. If an interneuron plays a role in generating
the flight rhythm, a brief pulse of depolarising current injected into it will
reset the rhythm by delaying or advancing the time of subsequent bursts
of spikes (Fig. 7.3c), not only in the interneuron but also in flight motor
neurons. Whether a pulse of depolarising current delivered to an interneu-
ron advances or delays the time of the next wing-beat cycle depends on
the phase of the cycle in which the stimulus current is delivered.
Experiments of this type are called phase-resetting experiments. They
reveal whether or not a particular neuron is involved in the clock mecha-
nism that determines the timing of cycles in a rhythmically repeating
activity. However, they do not reveal the exact mechanism for generating
the rhythm.
One interneuron, number 301, is shown in Fig.7.4a. This neuron has its
Box 7.1. The control of singing in crickets
Male crickets sing by slightly elevating their forewings (a), and repeat-
edly rubbing a comb-like file on one against a hardened scraper on the
other, shown by the cross-section through a cricket’s thorax in (b). The
pattern of activity in some wing muscles is very similar to their activ-
ity during flying. In Teleogryllus, during singing the wings move ini-
tially at 20 Hz, and later speed up to 35 Hz. Muscles that elevate the
wings in flight cause wing closure and a pulse of sound during singing;
and muscles that depress the wings in flight cause opening during
singing. Despite this, different interneurons are involved in generating
the two behaviour patterns. The pattern of intracellular activity
recorded from motor neurons is different in the two behaviours and
some interneurons are active during one activity but not the other
(Hennig, 1990). This is illustrated by the recording in (c), which shows
an interneuron producing clear rhythmical activity during singing,
but not after flight was initiated by a puff of wind. This finding was
surprising because it had been assumed that, because the two behav-
iours were similar, the same interneurons would be involved in gener-
ating the rhythms for flying and for singing. However, the evolution of
singing behaviour must have involved the development of new neu-
ronal circuitry. (b after Kutsch, 1969; c from Hennig (1990), copyright
Springer-Verlag.)
Figure 7.4 Interneurons of the central pattern generator for flight. (a) The
anatomy of interneuron 301 in the second and third thoracic ganglia. The
neuron was stained by injection of dye from an intracellular electrode. The
cell body is indicated by the arrow. (b) Excitatory connection from interneu-
ron 301 to interneuron 501, demonstrated by simultaneous intracellular
recordings from the two neurons. Multiple sweeps of the oscilloscope are
overlain, each triggered by a spike in 301. (c) Inhibitory connection from 501
to 301. (d) Schematic circuit representing the excitatory (⫹) and inhibitory
(⫺) connections between 301 and 501. (Modified after Robertson & Pearson,
1985.)
cell body in the second thoracic ganglion, and many branches both in this
ganglion and the third thoracic ganglion. Another interneuron, number
501, is arranged the other way round, with its cell body in the third thoracic
ganglion. Intracellular recording from 501 shows that this interneuron is
excited at the same time as depressor motor neurons in each wing-beat
cycle. Interneuron 301 is excited slightly before the activation of depressor
motor neurons.
When intracellular recordings from interneurons 301 and 501 were
examined in detail, interactions between the two interneurons were found.
Consistency in these interactions is shown by overlaying several sweeps on
the oscilloscope. Each sweep is triggered from a spike in one of the inter-
neurons. A spike in 301 is always followed, after a delay of 6 ms, by a small
depolarising potential in 501; and a spike in 501 is always followed, after 3
ms, by a brief, hyperpolarising IPSP in 301 (Fig. 7.4b, c). Allowing for time
for neuronal signals to be conducted to the recording sites, the delay of only
3 ms in transmission from 501 to 301 suggests that this connection is direct,
or monosynaptic. The greater delay in the connection from 301 to 501
leaves room for at least one additional neuron to be involved.
These two neurons form part of a circuit that could generate bursts of
activity (Fig. 7.4d; Robertson & Pearson, 1985). Excitation of 301 leads to
excitation of 501, after a short and fixed delay; and 501, in turn, inhibits 301
and terminates its burst of spikes. If 301 were excited from another source,
it would become active again once 501’s excitation had died away, and the
circuit would reverberate on its own. There is some evidence that the circuit
can work in this way because, in some experiments, steady excitation of 301
by current injection causes steady, rhythmical activity in 501 and in flight
motor neurons. However, it is not feasible to perform the critical test of iso-
lating this circuit from others in the thoracic ganglia. The circuit is just one
of many that have been found in the flight pattern generator. It is unlikely
that any single interneuron is indispensable for generating the flight
pattern.
The mechanism by which 301 excites 501 has not been fully elucidated.
An interneuron called number 511 is known to be interposed between 301
and 501 (Fig. 7.5). Neuron 301 inhibits 511, which in turn inhibits 501. This
implies that the delayed excitation which 301 causes in 501 is by way of two
successive inhibitions, a kind of interaction called disinhibition. However,
for disinhibition to work in this circuit, it is necessary for 511 to cause con-
tinuous, tonic inhibition of 501. This would mean that its membrane poten-
tial was continually depolarised above the threshold for its synapses to
release transmitter. Only in this way can a discrete inhibitory potential in
511, caused by a spike in 301, be converted into a discrete depolarising
potential in 501. Physiological characteristics of the depolarising potential
in 501 suggest that it is caused by a reduction in inhibition rather than by a
conventional excitatory synapse. However, we do not know whether output
synapses from 511 release transmitter tonically, without requiring spikes.
The circuit in Fig. 7.5 represents a small part of the central pattern gener-
ator for flight, and allows glimpses into its mode of operation. Wind on the
head activates pathways that excite 206. Interneuron 206 excites elevator
motor neurons through 504, and causes delayed excitation of depressor
motor neurons through 301. As explained above, a feature of the flight
program is that the duration of bursts of spikes in depressor motor neurons
Figure 7.5 Synaptic relationships of some of the elements of the central

pattern generator for flight in the locust. Excitatory connections are indi-
cated as triangles and inhibitory connections as circles. Interneurons are
numbered; and E and D indicate elevator and depressor wing motor
neurons. (Modified after Robertson & Pearson, 1985.)
is relatively constant over a range of different wing-beat frequencies. This

constancy can be explained by the actions of 301 in indirectly exciting these
motor neurons while at the same time removing inhibition of them from
511. This creates a discrete time slot, following a burst of spikes in 301,
during which it is possible to excite the depressor motor neurons. Another
feature of the flight program, the constant delay between bursts in elevator
and depressor motor neurons, can be explained by the dual action of 504
which excites the elevator motor neurons directly and the depressor motor
neurons indirectly through interneuron 301.
Not all of the thoracic interneurons that are involved in flight participate
in generating the rhythm. For example, some interneurons seem to play a
role in initiating and maintaining flight, but not in the timing of wing beats
(Pearson et al., 1985). The 204 interneurons are a small group of about four
on each side of the second thoracic ganglion. They all have axons that
describe a tight loop in that ganglion and then travel anteriorly, so they
cannot make direct contact with most of the interneurons of the flight gen-
erator, and they are excited by air currents blown at the head. Injection of
depolarising current into a single 204 neuron to excite it strongly can trigger
fictive flight, and injection of hyperpolarising current makes it less likely
that air currents will trigger a flight. These interneurons may, therefore, be
part of pathways that can start the flight motor pattern generator. During
fictive flight, they can spike tonically, so they neither contribute to nor
receive timing information for wing-beat cycles. Some stimuli which initi-
ate flight, such as wind to the tail end of a locust, do not excite the 204 inter-
neurons, which means that these neurons are not obligatory for starting the
flight motor program, so that different sensory pathways can trigger the
program independently.
7.7 Proprioceptors and the flight motor pattern
Locusts have a large array of sensors that report back to the central nervous
system on the mechanical effects which the commands it issued slightly
earlier caused. This is necessary because the effects of muscle contraction
are not totally predictable. A flying locust might, for example, experience a
gust of wind which causes its course to deviate. When it deviates, the wings
on one side of the body might move up and down more than those on the
other side for a few wing-beat cycles.
A large number of proprioceptors monitor movements of each wing and
two types have been most intensively studied. The first consists of just one
sensory receptor cell, called the wing hinge stretch receptor. Its cell body is
embedded in an elastic strand which spans the joint of the wing with the
side wall of the thorax. The large calibre of the axon of the stretch receptor
has enabled experimenters to record from it, even during tethered flight.
Wing elevation excites the stretch receptor and it fires a burst of spikes
during each wing-beat cycle, with the number of spikes reporting the extent
of elevation (Fig. 7.6a). The second sense organ is called the tegula, and this
Proprioceptors and the flight motor pattern 179
Figure 7.6 Proprioceptors and locust flight. (a) Spikes in a forewing stretch
receptor recorded simultaneously with up and down movement of the wing
and an EMG from a hindwing depressor muscle. (b) Arrangement of an
experiment in which a locust controls its angle of flight into the wind by
varying the interval between spikes in two flight motor neurons, here one
on the left and the other on the right. Wire electrodes record EMGs and,
after amplification, the time difference between excitation of the two motor
neurons is measured for each wing-beat cycle. The time difference is com-
pared with a predetermined reference value and if the two are equal, flight
course is unaltered. If there is a difference, the motor turns by a few degrees
to alter the direction of flight relative to the wind. (c) Recordings from an
experiment like that in (b), in which spikes from right and left motor
neurons 129 were recorded. The reference value for the right–left delay
switched every 25 s between 1 and 2 ms. Quite rapidly, the locust adjusted
the delay between right and left spikes to follow the reference value and,
after an initial swing, it maintained a flight direction more-or-less straight
into the wind. (a from Möhl, 1985; b and c modified from Möhl, 1988; copy-
right © 1985, 1988, Springer-Verlag.)
contains a number of stout hairs borne on a cuticular pad beneath the base
of the wing. Inside this pad is another proprioceptive sense organ called a
chordotonal organ. The hairs of the tegula are brushed and excited by wing
depression. The stretch receptor and the tegula, therefore, respectively
monitor wing elevation and depression. Other aspects of wing movement
are also monitored; even the tiny wing veins bear tension transducers on
their surfaces.
Both the stretch receptor and sensory cells of the tegula make direct,
monosynaptic connections onto flight motor neurons and interneurons
(Burrows, 1975; Pearson & Wolf, 1988). The stretch receptor excites almost
all of the depressor motor neurons on the same side of the body as its wing.
This means that when the wing is elevated, excitation of the stretch recep-
tor will enhance excitation of the depressor motor neurons whose action
opposes the elevation. If the wing is elevated more strongly than usual, the
action of this proprioceptive reflex is to ensure that the depressors are
excited more quickly than usual. The effect is to restore wing movement to
the usual or preferred pattern. Similarly, the tegula excites wing elevator
motor neurons when the wing is depressed.
In early experiments, attention was focused on the effects that stretch
receptors have on the flight rhythm. When stretch receptors were removed,
the flight rhythm dropped, and could be accelerated again by stimulating
the axon of a stretch receptor. Recent experiments have shown that these,
and other proprioceptors, play a vital role in ensuring that the flight motor
output is appropriate for achieving stable locomotion. Intracellular record-
ings from flight motor neurons show characteristic features that are due to
input from particular proprioceptors (Pearson & Wolf, 1988; Pearson &
Ramirez, 1990). Elevator motor neurons, for example, show two phases of
excitation during a wing beat, with synaptic inputs from the tegula sensory
neurons preceding excitatory input from interneurons. The wing proprio-
ceptors are, therefore, deeply embedded into the circuitry that generates
the flight pattern.
The central pattern generator and proprioceptive feedback loops work
together subtly to ensure that the animal’s motor output is continually
adjusted to maintain its desired course. This can be illustrated by some
experiments conducted by Bernhard Möhl (1993). If a locust is tethered to
a holder which allows it to swivel to the left or right, the locust will contin-
ually adjust its angle of yaw so that it flies straight into an air stream. In his
Proprioceptors and the flight motor pattern 181
experiments, Möhl did not allow the locust to provide its own power for
changing its angle of yaw but used a small motor attached to the holder to
twist it (Fig. 7.6b). Rotation of the motor was controlled by a computer,
which measured the time interval between spikes in the EMGs recorded
from two different flight muscles and compared this with a reference value
selected by the experimenter. The locust, therefore, could control its own
yaw angle by slight variations in the interval between spikes in the two
motor neurons, which were often the left and right hindwing muscles 129.
In the experiment shown in Fig. 7.6c, the first reference value chosen was a
time difference of 2 ms between spikes in the left and right 129 motor
neurons, so that the locust remained on a flight course straight into the
wind when it maintained this interval. The reference value was then
switched to 1 ms, and the locust quite rapidly reset the interval between
spikes in the left and right motor neurons. Whenever the reference value
was altered, the locust reset the interval between spikes in the two selected
motor neurons. Different pairs of muscles were selected in a series of
experiments.
These experiments have revealed a degree of plasticity in the locust
nervous system which was unexpected. The locust cannot know which pair
of motor neurons the experimenter has selected for a particular experiment.
This means that the locust must be able to compare the results of varying
the timing among different pairs of motor neurons before discovering which
ones are effective in maintaining course. Cutting the pair of muscles
selected destroys the ability of the locust to maintain course in this way,
which shows that proprioceptive feedback about the mechanical effects of
muscle contraction is required. We do not yet know which proprioceptors
are involved: the stretch receptor and tegula hairs both report movements of
skeletal elements rather than individual muscle contractions; and flight
muscles are not known to have sense organs similar to the muscle spindles
of humans embedded in them. Möhl’s experiments show that small, contin-
ual variations in the motor pattern are important because they allow
flexibility in the most effective pattern at maintaining course. The central
nervous system does not generate an unchanging, ‘perfect’ motor pattern;
the pattern is tuned through the action of proprioceptors. Instead, we can
view one important role for proprioceptors as participating in selecting the
most effective program for maintaining a desired course. The ability to make
continued, small adjustments to the motor output means that the pattern
can be continually updated, and rapidly adjusted to particular conditions.

For a locust, these conditions would include wind direction, and turbulence
caused both by the wind and neighbouring locusts in a swarm. A similar
strategy is becoming adopted by engineers in designing control systems for
robots, because it provides a way to compensate for inevitable inaccuracies
in manufacturing. It is difficult, for example, to ensure that the friction in
wheel axles on the left and right sides is exactly balanced.
7.8 Steering and initiating flight
Flying animals require sensory mechanisms both for maintaining a certain

flight course and for altering it so that they avoid collisions or predators.
The DCMD neuron (see section 5.8) might be responsible for the ability of
locusts to avoid colliding with each other in dense swarms; and locusts, like
some other insects, possess neurons which respond to the high-frequency
calls of insect-eating bats and which can probably cause a locust to steer its
flight path away from a hunting bat (Robert, 1989; Baader, 1991).
More is known about neurons that play a role in maintaining a particular
flight course, called DN neurons. These neurons are multimodal – they
respond to a number of different sensory modalities in a way that could
report details of deviations from a straight flight course. Many of them are
excited by wind flowing over the head, which is detected by groups of short,
curved hairs mostly situated dorsally, between the compound eyes. Each
hair is attached to a bipolar sensory neuron which projects into the brain
and probably makes synapses directly with some of the large interneurons.
Other sense organs which affect the interneurons include the simple eyes
or ocelli, the compound eyes, and proprioceptors that monitor movements
of the neck joints.
One wind-sensitive interneuron is called the tritocerebral commissure
giant (TCG), after the small nerve which contains its axon. The small nerve
runs between the two connective nerves near to where they leave the brain,
and this unusual anatomical feature has enabled experimenters to record
spikes from the TCG in loosely tethered, flying locusts (Bacon and Möhl,
1983). When a locust is flying into a steady air stream, the TCG does not
spike at a steady rate. Instead, it generates a brief burst of spikes to coincide
with elevation during each wing-beat cycle. The reason is that movement of
the wings and nodding motions of the head create their own air currents,
Steering and initiating flight 183
which interact with wind caused by the locust’s movement forward through
the air. This interneuron is, therefore, rhythmically active during flight.
Stimulating it artificially, to induce extra spikes, can reset the flight rhythm
by lengthening or shortening the cycle in which the stimulus is delivered.
Therefore, the TCG meets the same criteria for being part of the central
pattern generator as some of the local, thoracic interneurons.
Two additional roles have been assigned to the TCG. First, it could initi-
ate flight during a jump, when the TCG is excited by the rush of wind over
the head. Stimulating the cut axon of a TCG is an effective way of initiating
fictive flight (Bicker & Pearson, 1983). Second, the left and right TCGs could
stabilise flight direction with respect to wind direction. If the locust yaws so
that it is no longer facing directly into the wind, excitation increases to one
TCG and decreases to the other (Möhl & Bacon, 1983). Stimulating one TCG
electrically can induce yaw movements during tethered flight.
Another example of a DN neuron is DNC (Fig. 7.7a). Besides being
excited by wind currents, DNC responds to rotation of the visual horizon.
When the horizon rolls about the long axis of the locust, DNC is excited by
movements which roll the animal towards the side contralateral to its axon
(Fig. 7.7b), and inhibited by movements which roll it in the opposite direc-
tion. In experiments, the thorax of the locust is fixed, and a panorama
painted on the inside of a sphere can be moved around the head to simu-
late movements of the horizon. Inputs from the wind-sensitive and visual
pathways sum, so that when a locust is flying along a straight course, DNC
is tonically excited by air currents. Rolling towards one side causes an
increase in spike frequency, and rolling to the other side causes a reduction
in spike frequency. Probably, two separate visual pathways converge on
DNC: one from the compound eyes, and the other from the simple eyes, or
ocelli, which have enormous fields of view and monitor changes in horizon
position by measuring changes in the total amount of light they receive
(Wilson, 1978). If a locust is carefully prepared to allow access to a short
length of connective nerves in the neck, but is capable of moving its wings,
abdomen and head freely, steering movements can be produced by stimu-
lating DNC with trains of electrical pulses delivered through a micro-
electrode in its axon (Hensler & Rowell, 1990). A train of spikes lasting a few
hundred milliseconds causes a change in the relative latency of action
potentials in motor neurons of steering muscles within one wing-beat cycle
(Fig. 7.7c), and causes the head to roll after a slightly longer delay. Besides
Figure 7.7 Brain neuron DNC and its role in flight control. (a) Anatomy of
the DNC neuron with its axon on the right of the nervous system. The cell
body, on the left of the brain, is indicated with an arrow. In different experi-
ments, intracellularly injected dye revealed the structure of DNC in the
brain, and in the second and third thoracic ganglia. (b) Spikes recorded
from the axon of the right DNC in response to a wind current and to rolling
motion of an artificial visual horizon. Clockwise motion of the horizon – as
if the locust was rolling away from the side of the DNC axon – excited this
DNC. Rolling in the opposite direction inhibited it. (c) Electrical stimulation
of a DNC axon induced a change in the relative latency of spikes in the left
and right hindwing motor neurons, 127. Similar shifts in relative latency
were also recorded in response to rolling movements of the visual horizon.
(Modified after Hensler & Rowell, 1990.)
the TCG and DNC neurons, between 20 and 30 other DN neurons on either
side of the brain send their axons to the thoracic ganglia (Hensler, 1992).
The ways in which these neurons generate steering movements are not
clear, but involve some direct output connections with motor neurons as
well as connections with local, thoracic interneurons.
7.9 Overall view of locust flight
The overall picture we now have of the flight control system of the locust is
of a series of intermeshed, overlapping neuronal loops (Fig. 7.8). It contains
Overall view of locust flight 185
Figure 7.8 A flow diagram to show the relationships between some of the
elements involved in generating the flight motor program in a locust. Some,
but not all, of the thoracic interneurons included here are involved in gen-
erating the rhythm. Rhythmic input to motor neurons and interneurons is
also derived from wing proprioceptors and brain neurons. As a result of the
forward and rather irregular movement through the air, both the wind cur-
rents and the visual stimuli that the locust experiences are modified, so that
the locust receives stimuli that are a combination of those caused by the
external environment and those resulting from the locust’s own movements.
a surprisingly diverse array of different nerve cells. Many of them, including

the proprioceptors and at least some of the wind-sensitive interneurons,
are rhythmically active at the wing-beat rhythm. We must consider these
neurons to be part of the flight generator itself because they are able to reset
the flight rhythm in the same way as some of the thoracic interneurons. The
flight generator is highly redundant – it contains many elements with
similar or overlapping functions, none of which is indispensable to its
normal operation. This redundancy brings the advantages, first, that the
flight system is rugged and not easily perturbed, and, second, that the
pattern produced is flexible and able to adapt rapidly to changing demands
from the environment. Flight involves additional elements, not shown in
Fig. 7.8. One of these is that the neurohormone octopamine is released into
the haemolymph. Octopamine affects the contractile properties of muscle

cells (Box 7.2) and increases the supply of fuel to muscles. In addition, it can
affect the properties of some of the interneurons of the flight motor
program generator, making them liable to fire spikes in cyclical bursts
(Ramirez & Pearson, 1991), although it has not been demonstrated that
octopamine is released within the central nervous system during flight.
A strong stimulus for flying is wind on the locust’s head and, in the labor-
atory, flight ceases rapidly when wind stops. However, many rhythmical
movements of animals are initiated by a brief stimulus and continue for
some time after the stimulus finishes. Two quite different behaviours of this
kind are examined below, both of which serve to take an animal away from
a noxious stimulus. The first behaviour is escape swimming by the large
nudibranch mollusc Tritonia to escape from starfish predators. The smell of
starfish elicits fairly stereotyped movement patterns in many marine inver-
tebrates and in Tritonia they consist of about four to seven cycles of alter-
nate flexion of the body upwards and then downwards. The second
example is swimming by young tadpoles of the frog Xenopus, which is elic-
ited by touching the body surface.
7.10 Triggering and maintaining escape swimming in Tritonia
Some of the first intracellular recordings from nerve cells in an animal

behaving in a more-or-less normal pattern were made from Tritonia
(Willows, Dorsett & Hoyle, 1973). By exposing the brain through a small
incision and then fastening it to a support platform, intracellular record-
ings can be made from single neurons. The animal is suspended in a tank
of seawater and can perform normal muscular movements. This experi-
mental technique allowed identification of motor neurons and interneu-
rons that are involved in local withdrawal reflexes as well as in the dorsal
and ventral movements that occur during escape swimming (Fig. 7.9a).
Ideas about the neuronal mechanisms that control escape swimming have
been revised quite radically a number of times, and both the behaviour and
the neuronal circuits that mediate it are more simple than those involved in
locust flight.
A swimming episode lasts up to a minute, and consists of four to seven
cycles of alternating bursts in dorsal and ventral flexion neurons. As in
locust flight, interneurons generate the rhythm and communicate it to
Triggering and maintaining escape swimming in Tritonia 187
Box 7.2. The neuromodulator octopamine
Octopamine, like some other amines, has a wide variety of effects on

behaviour. In lobsters, injection of octopamine into the blood causes
an animal to adopt submissive behavioural postures, whereas another
amine, serotonin, causes a lobster to behave aggressively (Kravitz,
1988). There is considerable interest in elucidating the roles of these
substances in behaviour: do they help prepare the body of an animal
for certain kinds of behaviours, or do they also act as the triggers for
particular behavioural acts? In locusts, octopamine has widespread
effects. For example, it activates pathways that metabolise fat to
provide energy during flying; it increases the responsiveness of pro-
prioceptors such as the wing hinge stretch receptor; and it potentiates
transmission at connections made by the stretch receptor in the
central nervous system (Orchard, Ramirez & Lange, 1993). It is
released from some of the dorsal unpaired midline (DUM) neurons
which are thought to release octopamine from diffusely placed
swellings of their axons rather than from conventional synapses. In
(a), a DUM cell in the third thoracic ganglion of a locust was stained
by intracellular injection. Sites of octopamine release, if any exist,
are unknown in the central nervous system. One effect, which
octopamine mediates directly on the contractile proteins, is to
increase the rate of relaxation following a twitch. This is shown in
recordings of twitches by the muscle that extends the hind tibia of a
locust (b). Each twitch is caused by a spike in this muscle’s slow motor
neuron. Support for a role for DUM neurons in preparation for partic-
ular movements comes from the observations that some DUM
neurons are active before and during flight (Ramirez & Orchard, 1990),
and others are active just before a kick by the hind legs (Burrows &
Pflüger, 1995). (a modified after Watson, 1984; b modified after Evans
& Siegler, 1982.)
Figure 7.9 Escape swimming in the sea slug (Tritonia). (a) Swimming con-
sists of several alternating dorsal and ventral flexion movements. These
movements do little to propel the animal through the water; they lift it from
the substrate, and it is carried by water currents. (b) The pattern of connec-
tivity between interneurons involved in generating the program for swim-
ming. In each side of the brain, receptors, including chemoreceptors, excite
DRI, the dorsal ramp interneuron. The swim rhythm is generated by a
network involving three dorsal swim interneurons, (DSIs), two ventral swim
interneurons (VSIs), and one C2 interneuron. (c) Intracellular activity
recorded simultaneously from a DSI, C2 and DRI during a swim episode,
caused by electrical stimulation of a sensory nerve. At the start of the swim,
the DSI is excited strongly, and its excitation declines slowly in a ramp-like
manner during the swim. The source of this excitation is the DRI. (d)
Modulation of the strength of a synapse from C2 to a motor neuron by
activity in a DSI. A short burst of spikes in C2, elicited by intracellular stim-
ulation, caused an EPSP in the motor neuron. When stimulation of C2 was
preceded by stimulation of a DSI for several seconds, the size of the PSP
caused by C2 increased dramatically. A similar increase was also produced
by bathing the brain in seawater containing serotonin. (b and c modified
after Frost & Katz, 1996; copyright National Academy of Sciences, USA; d
from Katz et al., 1994; reprinted with permission from Nature; copyright
© 1990 Macmillan Magazines Ltd.)
Triggering and maintaining escape swimming in Tritonia 189
motor neurons. Three groups of interneurons on each side of the brain are
involved in rhythm generation: three dorsal swim interneurons, two ventral
swim interneurons, and one other interneuron called C2 (Fig. 7.9b). None
of these interneurons has an intrinsic capacity to generate rhythmical oscil-
lations in membrane potential, and the rhythm is generated by a circuit of
linked interneurons. In an isolated brain, electrical stimulation of a nerve
elicits a very similar pattern of intracellular activity to that which occurs in
an intact animal that is swimming in response to starfish odour (Fig. 7.9c).
The swim neurons are strongly excited at the start of a swim. As the swim
progresses, the excitation gradually diminishes and the cycle period
becomes longer. A single interneuron on each side of the brain collects
input from sensory neurons and conveys the excitation to the swim inter-
neurons (Frost & Katz, 1996). The interneuron is called the dorsal ramp
interneuron (DRI) because its excitation decays in a slow, ramp-like
manner during a swim. Excitation of the DRI always precedes that of other
interneurons before a swim starts. During a swim, a DRI continues to spike,
with a brief interruption during each ventral flexion caused by an inhibitory
synapse from the ventral swim interneurons. Stimulating a DRI with
enough current to make it spike at similar frequencies to those that occur
during a swim will trigger swim activity in the network. There are similar-
ities between a DRI and some of the wind-sensitive interneurons of the
locust flight system. Both respond to the same sensory stimuli that trigger
a particular motor pattern, and electrical stimulation of the neurons trig-
gers the pattern. However, the DRI is unlike any single wind-sensitive
interneuron in that swimming will not start unless it is excited. If
hyperpolarising current is injected into a DRI so that the DRI does not spike
when a peripheral nerve that normally elicits swimming is stimulated,
swimming does not begin. Also, ramp excitation of the three swim inter-
neurons does not occur. If a DRI is hyperpolarised after a swim has started,
the swim episode stops. The DRI, therefore, acts as a channel through
which all sensory excitation must pass in order to activate the central
pattern generator for swimming.
Each DRI makes excitatory connections with the three dorsal swim inter-
neurons, which play a dual role in generating the program for swimming
(Katz, Getting & Frost, 1994). First, the dorsal swim interneurons contribute
to the pattern of each cycle of the rhythm through their synaptic connec-
tions with the ventral swim interneurons and with C2. Second, they sustain
the rhythm because the neurotransmitter that they release, serotonin, has
multiple effects within the network. Not only does serotonin act as the neu-
rotransmitter at the output synapses made by a dorsal swim interneuron, it
also increases the excitability of C2 and enhances the release of neurotrans-
mitter from C2. For example, the EPSPs which C2 mediates in a motor
neuron increase in amplitude when a dorsal swim interneuron is excited
(Fig. 7.9d). The facilitatory action which the dorsal swim interneurons exert
on the output synapses from C2 is called intrinsic neuromodulation,
because the neuromodulator is released by neurons that are intrinsic to the
pattern-generating network and is only released when the pattern genera-
tor is active. The neurons of the swim generator are also involved in less
dramatic, local withdrawal responses. When swimming starts, the dorsal
swim interneurons generate an intense burst of spikes, and it is likely that
the serotonin which this releases ensures that the network is configured to
generate co-ordinated, rhythmical swimming movements.
7.11 Swimming by young Xenopus tadpoles
The most complex patterns of behaviour occur in vertebrates, and under-

standing the neuronal basis of these behaviours must involve unravelling
the circuits of the spinal cord. One approach, which has been particularly
fruitful, is to examine very simple behaviours in lower vertebrates, particu-
larly swimming in lampreys (Grillner et al., 1995) and in tadpoles. The
pattern of swimming in newly hatched tadpoles of the clawed frog Xenopus,
turns out to be especially simple (Roberts, 1990). For the first few days of
life, these tadpoles spend much of their time attached to leaves and stems
by a cement gland on the head. If the skin is touched, a tadpole will swim
by side-to-side undulating movements of its trunk and tail until it finds a
new attachment site (Fig. 7.10a). At the start of a swim, the body flexes to
the left and then to the right up to 20 times per second. The rate of flexion
declines during a swim. Waves of movement travel towards the tail end and
propel the tadpole forwards at up to 5 cm/s.
The spinal cord of a young tadpole contains about 1000 neurons of eight
types, including five types of interneuron. Alan Roberts and his colleagues
have pioneered a method of making intracellular recordings from these
neurons during fictive swimming. A tadpole is immobilised with the drug
curare, which blocks neuromuscular transmission. When the skin is
touched, the pattern of spikes in motor neuron axons is the same as in free-
Swimming by young Xenopus tadpoles 191
Figure 7.10 The generation of the swim pattern in Xenopus tadpoles. (a)
Tracings from a video recording of a short swim by a young tadpole, initiated
by touching its skin. (b) Intracellular recordings from motor neurons of left and
right trunk muscles during fictive swimming. During the swim, both neurons
were depolarised from resting potential (dotted lines). The neurons spiked out
of phase with each other, and received IPSPs in midcycle. (c) Rebound spike in
a motor neuron. A small amount of steady, depolarising current was injected
into the neuron, and a rebound spike occurred at the end of a brief pulse of
hyperpolarising current. (d) Dual component EPSP in a motor neuron, caused
by stimulating the axon of an interneuron. When the spinal cord was bathed in
a solution containing an antagonist of the NMDA glutamate receptor, the slow
and prolonged component of the EPSP disappeared. (e) The network that is
thought to generate the rhythm for swimming. The descending interneurons
(d) are involved in circuits of mutual excitation, and the commissural
interneurons (c) convey inhibitory signals across the nerve cord so that excita-
tion of motor neurons (mn) alternates in the two sides. (After Roberts, 1990;
reprinted with permission from Science and Technology Letters.)
swimming tadpoles, demonstrating that the central nervous system con-

tains a central pattern generator for swimming. Touching the skin excites
sensory neurons called Rohon Beard cells and a single spike in one of these
cells can be sufficient to elicit swimming. During a swim, a motor neuron
remains depolarised from resting potential. It produces just one spike per
swim cycle, and receives an IPSP when neurons on the opposite side fire
(Fig. 7.10b). When tadpoles mature sufficiently to feed, the pattern of activ-
ity during swimming becomes considerably more complex, and motor
neurons generate bursts of spikes (Sillar, Wedderburn & Simmers, 1991).
The pattern generator for swimming provides an excellent lesson on how
both synaptic circuitry and intrinsic membrane properties shape the
output of a central pattern generator. Like the other rhythmical behaviours
examined here, interneurons generate the pattern and, as in the case of
locust flight, different mechanisms for generating the rhythm operate in
parallel. An essential element of the swim pattern is that motor neurons on
the left and right sides are excited alternately. This is achieved by commis-
sural interneurons that are excited in time with motor neurons on the same
side as their cell bodies and dendrites. The axons of the commissural inter-
neurons cross the spinal cord and inhibit motor neurons and interneurons
of the opposite side.
The midcycle IPSP that these interneurons mediate has a dual role.
Besides ensuring that motor neurons are inhibited when their contralateral
partners are excited, an IPSP triggers the next spike in the motor neuron.
The spike is called a rebound spike because it is triggered by the end of an
event that hyperpolarised the motor neuron. A pulse of hyperpolarising
current injected from a microelectrode can also be a trigger for a rebound
spike. However, rebound spikes are not initiated if the membrane potential
simply repolarises to resting at the end of a hyperpolarising pulse. A small,
continual depolarising current must also be injected, to deliver a small
amount of excitation at the end of the hyperpolarising pulse (Fig. 7.10c).
During a swim, steady excitation is provided through excitatory synapses
from interneurons. The steady depolarising potential sets up a critical
balance between currents carried through two types of voltage-sensitive
channels. The first type is sodium channels, and entry of sodium through
these channels tends to excite the neuron. The second type is potassium
channels, through which potassium will leave the neuron. The current
carried by potassium ions will increase as the neuron depolarises, opposing
Swimming by young Xenopus tadpoles 193
the excitatory action of the sodium channels. The effect of a brief hyperpo-
larisation is to close both types of channel. When the hyperpolarisation
ends, the sodium channels open more rapidly than the potassium chan-
nels, allowing the generation of a single spike.
The steady depolarisation is necessary for the generation of a rebound
spike because it enables the sodium channels to open following an IPSP.
During swimming, it is provided by a population of interneurons that excite
motor neurons on their own side of the nerve cord. There are probably cir-
cuits of mutual excitation among these interneurons, which helps to
sustain swimming. Another factor that sustains excitation throughout a
swim is that the EPSPs which these interneurons cause in motor neurons
have relatively long durations, up to 200 ms, which is much longer than one
swim cycle. This is because the EPSPs have two different phases, each
mediated by different types of receptor for the neurotransmitter glutamate.
The two components of an EPSP can be dissected apart by pharmacologi-
cal agents (Fig. 7.10d). The first, fast component can be mimicked by apply-
ing the drugs kainate or quisqualate to the surface of a motor neuron, while
the second can be mimicked by applying another drug, N-methyl-D-aspar-
tate (NMDA). Both of these drugs are agonists of glutamate – they bind to
receptors and activate their ionic channels in a similar way to glutamate.
The motor neurons, therefore, have two different types of receptor for glu-
tamate at their input synapses: one type causes the initial, fast excitation,
whereas the second type, called the NMDA receptor, mediates the longer-
lasting excitation. The fast EPSPs help excite the motor neuron at the
appropriate time in a swim cycle, reinforcing the rebound from an IPSP.
A characteristic of the NMDA receptor is that its channel can only open if
the postsynaptic neuron is already strongly depolarised, through the action
of other synapses. This feature makes the NMDA receptor suitable for
responding to combinations of signals, and it is heavily implicated in
mechanisms of learning. In the tadpole spinal cord, strong excitation of
motor neurons by the faster-acting, glutaminergic synapses is necessary to
ensure that the NMDA-mediated EPSPs can switch on.
The mechanism for generating the swim pattern, therefore, depends on
circuitry that ensures that left and right sides alternate in their activity and
that there is mutual excitation between interneurons (Fig. 7.10e). Cellular
properties underlie the sustained excitation of motor neurons by NMDA
receptors, and the generation of rebound spikes. Simulation by computer
has shown that circuits that have these characteristics can generate an
output pattern like the tadpole swim pattern (Roberts & Tunstall, 1990).
However, the interneurons that convey inhibition across the cord are not
essential for the generation of a rhythm because when the spinal cord is
split longitudinally, each half can generate rhythmically repeating activity
on its own.
Some of the features of the swim generator of young tadpoles have also
been discovered in neurons of the mammalian spinal cord, including
rebound excitation and long-lasting EPSPs mediated by NMDA receptors.
The relative complexity of movements that the mammalian spinal cord is
capable of producing might arise partly from a greater diversity of types of
neuron compared with the young tadpole. However, it is equally possible
that the two spinal cords contain similar circuits of neurons but, in the
mammal, there are many variations in the basic wiring of these circuits.
Recent work on a relatively simple ganglion, the stomatogastric ganglion of
decapod crustacea, has demonstrated that circuitry can, indeed, be
reconfigured extensively, allowing groups of neurons to participate in quite
different activity patterns.
7.12 Circuit reconfiguration in the stomatogastric ganglion of

the lobster
In crustacea, movements of the foregut are achieved by contraction of dis-

crete striated muscles, very similar to the muscles that move limbs during
movements of the body. Quite complex, regularly repeating patterns of
activity occur in these muscles. A major reason for investigating their
control is that the neurons responsible for generating these movements are
contained in four discrete, small ganglia and the nerves that connect them
(Fig. 7.11a). The most intensively investigated is the stomatogastric gan-
glion which, in lobsters, contains just 30 neurons. A small nerve connects it
with three other ganglia of the stomatogastric nervous system. Almost all of
the neurons in the stomatogastric ganglion are motor neurons and, unlike
the situation for the three examples described earlier in this chapter, the
motor neurons are essential components of the central pattern generator.
The stomatogastric ganglion controls two different parts of the foregut,
called the gastric mill and the pyloric region. The gastric mill contains three
teeth, which cut and grind food after it has been churned in a region called
Circuit reconfiguration in the stomatogastric ganglion of the lobster 195
Figure 7.11 The foregut and stomatogastric nervous system of a spiny

lobster (Panulirus). (a) The foregut with its ganglia and nerves. The stomato-
gastric ganglion lies in a blood vessel (not shown) on top of the stomach.
The stomatogastric and oesophageal ganglia are unpaired, and there is a
circumoesophageal ganglion attached to each of the large connective
nerves which link the brain and suboesophageal ganglion. (b) A diagram
showing some of the synaptic connections between neurons of the pyloric
section of the stomatogastric ganglion. All neurons except the anterior
burster (AB) are motor neurons; the two PD neurons control dilatation of
the pylorus and the other motor neurons (VD, LP, IC and PY) control
different phases of constriction. The stomatogastric nerve contains axons of
neurons which have a variety of different effects on neurons of the stomato-
gastric ganglion. (b modified from Selverston & Miller, 1980.)
the cardiac sac. Muscles of the gastric mill are activated in a rhythm with a
cycle period of 5–10 s. The pyloric region, into which the gastric mill
empties, has a shorter cycle time, of 0.5–2 s; it squeezes and mixes food par-
ticles.
The pattern of synaptic connections within the ganglion and the array of
active, membrane properties which single neurons possess are startlingly
complex. The 14 neurons of the pyloric network are connected in a network
that includes over 20 electrical synapses and over 60 chemical, inhibitory
synapses (Fig. 7.11b). At one time it was thought that the gastric mill
rhythm is primarily generated by reciprocal, inhibitory circuit interactions,
whereas the pyloric rhythm is generated by intrinsic bursting in one inter-
neuron. However, we now know that the generation of each rhythm
involves both mechanisms. One experimental approach that was particu-
larly useful in revealing the different mechanisms involved isolating single
cell circuits of small numbers of cells from others in the ganglion by killing
the cells that were presynaptic to the neurons of interest. To kill a particular
neuron, a fluorescent dye was injected into it through a microelectrode and

then an intense spot of blue light was shone onto the cell body (Selverston
& Miller, 1980). Using this technique, it was possible to isolate a single pair
of neurons that are connected by reciprocal, inhibitory synapses (Miller
and Selverston, 1982). A simple two-neuron network is able to generate
rhythmical activity, although the pattern differs from that produced by an
intact stomatogastric ganglion.
Neurons of the stomatogastric nerve play a vital role in pattern genera-
tion by unmasking particular properties that are intrinsic to individual
cells. The nerve contains 60–120 axons and rhythmic activity is not
recorded from the stomatogastric ganglion if the stomatogastric nerve is
cut or if its axons are silenced. However, rhythmic co-ordinated output
from the ganglion can be restored either by stimulating axons within the
nerve or by adding certain transmitters, particularly some amines or pep-
tides, to the seawater bathing the ganglion. Because the pyloric neurons
only burst in the presence of particular transmitters, they are called condi-
tional oscillators. All of the neurons in the pyloric region are conditional
oscillators but one, the anterior burster (AB in Fig. 7.11b), has the fastest
rhythm and acts as the master clock, setting the pace for the whole pyloric
region.
One particular bilateral pair of neurons that project into the stomatogas-
tric ganglion, called the pyloric suppressors, have widespread effects on
motor patterns (Meyrand et al., 1991; 1994). When the pyloric suppressor
neurons are active, the separate pyloric and gastric mill rhythms stop; and
the pyloric and gastric mill neurons become active in a new, co-ordinated
pattern of rhythmic activity (Fig. 7.12a). An electric synapse links the two
pyloric suppressor neurons, so they operate as a single functional unit.
When the stomatogastric ganglion is expressing the pyloric and gastric mill
patterns, the pyloric suppressor neurons are silent. However, they can be
excited by applying food to chemoreceptors that are situated on the valve
that separates the oesophagus from the stomach. When a pyloric suppres-
sor neuron is excited, either through these sense organs or by injection of
depolarising current, its membrane potential oscillates and it generates
bursts of spikes. These bursts of spikes excite motor neurons which cause
the valve to open, and intracellular recordings from pyloric and gastric mill
neurons show profound changes in the activity patterns they express. Some
neurons are tonically inhibited from producing spikes, but others, such as
Circuit reconfiguration in the stomatogastric ganglion of the lobster 197
Figure 7.12 Action of the pyloric suppressor neuron (PS) in reconfiguring

the stomatogastric ganglion. (a) Intracellular recordings from a neuron of
the pyloric region and a neuron of the gastric mill region of the stomatogas-
tric ganglion. On the left, PS was silent and the two neurons expressed
different rhythms of activity. When PS was stimulated, the two neurons
became active in the same new rhythm. (b) A diagram to illustrate the effects
of PS on the circuitry of the stomatogastric ganglion. When PS is silent, net-
works for the oesophageal, pyloric and gastric mill rhythms operate inde-
pendently of each other. Activity in PS causes some neurons of these three
networks to be incorporated into a new functional network, and silences
others (dotted). (c) Behavioural correlates of PS activity. On the left, different
foregut regions are acting independently and valves between them are
closed. On the right, the lobster is swallowing. (a from Meyrand, Simmers
& Moulins,1991; reprinted from Nature; copyright © 1991 Macmillan
Magazines Ltd; b and c redrawn after Meyrand, Simmers & Moulins1994.)
the pyloric and gastric mill neurons, express a new pattern in which
neurons from both regions are co-ordinated.
The new pattern is probably that for swallowing, and the pyloric suppres-
sor neurons also cause the valve between the oesophagus and stomach to
open. During swallowing, movements of the whole foregut must be co-
ordinated to move food onwards (Fig. 7.12c) and the swallowing pattern has
a frequency between that of the pyloric and gastric mill regions. After the
pyloric suppressor neurons have stopped firing, the new pattern persists
for several tens of seconds and the original two pyloric and gastric mill
rhythms are slowly re-established. This observation is important because it
shows that circuits within the ganglion itself are reconfigured by activity in
the pyloric suppressor neurons (Fig. 7.12b). If the stomatogastric ganglion
neurons were all driven by bursts of spikes in the pyloric suppressor
neurons during swallowing, we would expect the pyloric and gastric mill
rhythms to be re-established as soon as the pyloric suppressor neurons
stopped firing. The pyloric suppressor neurons reconfigure the circuits of
the stomatogastric ganglion, so that neurons which previously participated
in different and unco-ordinated activities now express a new, single
pattern. Many of the details of how this reconfiguration is achieved remain
to be worked out.
7.13 Conclusions
Usually, motor neurons are not involved in the generation of rhythms for
movements, but are driven through synapses from interneurons and
sensory neurons. This arrangement allows motor neurons to be excited
independently of each other, so that their muscles can be used in different
movement patterns. Networks of neurons that can generate programs for
movement are often called pattern generators. In general, rhythmical
movements are generated by a number of different mechanisms operating
together, which makes the pattern generator robust.
Phase-resetting experiments demonstrate whether a particular neuron is
involved in generating rhythmical activity. In locust flight, the rhythm is
generated by networks of interneurons, reinforced by cyclical feedback
from proprioceptors such as the wing hinge stretch receptors, and from
wind-sensitive neurons. There is good evidence for the existence of rever-
berating circuits of interneurons in the flight generator, but because so
Further reading 199
many interneurons are involved it is difficult to assign a precise role to a

single neuron and to disentangle different circuits from each other.
There are a number of different ways in which the excitation of a rhythm-
generating network can be sustained. In locust flight, sensory activity, par-
ticularly from wind-sensitive neurons, is important. Neuromodulators can
play important roles by strengthening synaptic connections, as in Tritonia
swimming, or by switching on bursting properties, such as in the pyloric
rhythm of lobsters or, perhaps, locust flight. Finally, long-lasting postsyn-
aptic potentials such as those mediated by NMDA receptors in tadpole
swim motor neurons can also be important in sustaining a rhythm.
Circuits that generate programs for movement are not ‘hard wired’ and
inflexible. One way in which this is evident is the manner in which sensory
feedback works in controlling locust flight. At its simplest, sensory feedback
allows a motor program to compensate for changing demands on muscles
as the nature of the terrain alters. Another more subtle role is to select
between patterns that differ in their effectiveness, for example in moving an
animal along a straight path. Continual small variations in motor output
allow the motor program to be updated continually, allowing compensa-
tion for changes in the mechanical properties of an animal’s skeleton and
muscles as it grows or is injured. Interneurons can also cause radical
changes in the way that networks are configured. This is well illustrated by
the way in which a single interneuron can reconfigure the lobster stomato-
gastric ganglion so that the neurons participate in new groupings and pat-
terns of activity.
Further reading
Arshavsky, Y.I., Orlovsky, G.N., Panchin, Y.V., Roberts, A. & Soffe, S.R. (1993).
Neuronal control of swimming locomotion – analysis of the pteropod mollusk
Clione and embryos of the amphibian Xenopus. Trends Neurosci 16, 227–33. A
review that compares and contrasts the mechanisms that generate simple pro-
grams for swimming in frog tadpoles and in a planktonic snail. The review
deals with general problems in the production of rhythmical motor activity.
Harris-Warrick, R.M., Marder, E., Selverston, A.I. & Moulins, M. eds. (1992).
Dynamic Biological Networks: the Stomatogastric Nervous System. Cambridge,
MA: MIT Press. Although this book is primarily about the crustacean stomato-
gastric nervous system, its reviews synthesise material in a way that throws
light on the general properties of motor control systems.
Stein, P.S.G., Grillner, S., Selverston, A.I. & Stuart, D.G. eds. (1996). Neurons,
Networks and Motor Behaviour. Cambridge, MA: MIT Press. This collection of
reviews was written to accompany a conference that is held every ten years to
discuss and present work on the generation of motor patterns, particularly to
highlight the most recent advances. The meeting is reviewed briefly in Katz, P.S.
(1996). Neurons, networks and motor behaviour. Neuron 16, 245–53.
8 Circuits of nerve cells and behaviour
8.1 Introduction
There are very few instances in which complete neuronal pathways can be
traced from the level of sense organs all the way to that of motor neurons.
Notable exceptions are some startle behaviours like those described in
Chapter 3, in which the size of the giant neurons involved makes experi-
mental study relatively easy and in which the links between sensory pro-
cessing and motor control are short. However, most of an animal’s
behavioural repertoire is not performed with the same urgency as escape
movements. Much sensory analysis, particularly in visual and auditory
pathways, involves several different stages, distributed over different
regions of a brain. How are different sensory messages identified, and how
are appropriate motor programs selected? The type of problem can be illus-
trated with a simple example. If a fly lands on your cheek, or if the skin of
your knee itches, you can move your hand without thinking to those loca-
tions to remove the source of annoyance. This might seem like a trivial
example of behaviour, but the neuronal mechanisms that allow us to
perform such an act are far from being understood. It is relatively straight-
forward to map the locations of sensory receptors in the skin in an orderly
manner within the brain, which generates a somatotopic map. However, it
is not simple to generate the correct commands that will generate the
correct balance of activation in different muscles of a jointed limb so that
its end arrives at a specific location on the body surface. How is a map of
position transformed into a series of finely graded muscle contractions? For
an electrical engineer, it would be a major challenge to build a robot that
could locate an object on its surface and then move the tip of a jointed
appendage to the correct location to scratch itself.
This chapter, examines some extremely simple behaviours that involve
201
reflex withdrawal movements of restricted regions of animals’ bodies.

Different experimental approaches have been adopted to study these beha-
viours. Microelectrodes are the most widely used tool for studying the
activity of neurons, and allow connections between neurons to be charac-
terised in detail. However, only a few neurons can be monitored at any one
time by using microelectrodes. There is great interest, therefore, in tech-
niques that allow an investigator to record the activity of many neurons
simultaneously. One approach is to use optical methods, and another is to
study the way that model networks of neurons work. A common theme that
is emerging is that the central nervous system is rather diffusely organised,
with relatively few interneurons dedicated to single behaviour patterns.
8.2 Neuronal activity during di≈erent behaviours in Aplysia
Aplysia is a shell-less gastropod mollusc that spends most of the time

grazing on seaweed near or just below the low-tide mark. It breathes by
drawing currents of water over a delicate gill, which is normally partially
extended from under a fleshy shelf of the body wall, but is retracted for
defence if nearby skin is touched lightly (Fig. 8.1a). The gill is also strongly
retracted in time with respiratory pumping movements. Some species of
Aplysia grow in excess of 30 cm long and have attracted the attention of
neurophysiologists since the 1930s because the nervous system contains a
number of neatly arranged ganglia with large, often distinctly coloured,
neuron cell bodies that are ideal targets for microelectrodes. The abdomi-
nal ganglion has been particularly intensively investigated, and is the gan-
glion that contains cell bodies of gill-withdrawal motor neurons and of
some of the sensory neurons that innervate the skin near to the gill.
Some sensory neurons make direct, monosynaptic, excitatory synapses
onto gill-withdrawal motor neurons, forming direct, simple reflex path-
ways. These pathways are suitable for mediating the gill-withdrawal beha-
viour. As the intensity of a stimulus applied to the skin increases, the
sensory neurons generate more spikes, and more sensory neurons are
recruited. This increases the frequency with which motor neurons generate
EPSPs and spikes. The classical processes of temporal and spatial summa-
tion of postsynaptic potentials in the motor neurons could readily explain
the smoothly graded relationship between stimulus strength and response
intensity that characterises this simple behaviour (Kandel, 1976). However,
Neuronal activity during different behaviours in Aplysia 203
Figure 8.1 The gill-withdrawal reflex of Aplysia. (a) A view of Aplysia from
above, showing the location of the delicate gill. (b) and (c) Two alternative
ways in which circuits controlling behaviours such as gill withdrawal could
be organised. (d) A diagram of the dorsal surface of the abdominal ganglion
showing some of the larger neuron cell bodies. Superimposed is a grid of 25
squares, each of which is the size of a photocell used to monitor spikes, as
explained in the text. (b and c modified after Wu et al., 1994; d modified
after Falk et al., 1993.)
these direct monosynaptic pathways are not the only ones responsible for
mediating the gill-withdrawal reflex (estimates of their contribution
vary from 5 per cent to 60 per cent). Sensory neurons make extensive
connections with motor neurons through indirect pathways that involve
interneurons – this parallel arrangement of direct and indirect pathways
linking sensory and motor neurons should be familiar from the way that
proprioceptors influence locust flight (see section 7.7). Aplysia also has a
peripheral nervous system. The involvement of interneurons not only
increases the complexity of the integrative processes underlying the rela-
tion between stimulus and response, but also introduces the possibility
that responses by motor neurons to skin stimulation will be influenced by
other behaviours in which these interneurons are involved.
Besides responding to mechanical stimulation of the skin, large gill-with-
drawal movements occur as part of ordinary ventilatory pumping move-
ments, and smaller withdrawals often occur apparently spontaneously. An
engineer would probably design separate circuits responsible for reflex and
ventilatory withdrawal and feed the outputs of these circuits through a
switch so that the gill could be controlled either by one or the other. In this
kind of configuration, particular elements would be dedicated to specific
behaviours, as shown in Fig. 8.1b. Circuit 1 might be responsible for gener-
ating respiratory pumping movements, while circuit 2 might organise
retraction in response to touching the skin and circuit 3 might control
another kind of gill retraction. All three circuits are subject to input from
sensory neurons and they each drive some of the gill-withdrawal motor
neurons (direct links between sensory neurons and motor neurons are
omitted in the diagram). In another arrangement, drawn in Fig. 8.1c, inter-
neurons are interconnected more widely and diffusely; different functions
are distributed within the whole circuit, so that single interneurons partic-
ipate in several different behaviours.
8.3 Optical monitoring of neuronal activity
The abdominal ganglion contains about 900 neurons, and it would be an

enormous task to catalogue all their interconnections by making recordings
with microelectrodes. Another approach, adopted by Lawrence Cohen and
colleagues, is to monitor the activity of as many different neurons as pos-
sible while the animal performs different behaviours. To do this, Cohen et
al. used a chemical (a pyrazolone oxonole) which has been specially devel-
oped as a tool in neurophysiological experiments. The dye attaches to cell
membranes, and the amount of light that the dye absorbs alters as the
voltage across the cell membrane changes. A ganglion is soaked in a solu-
tion of this dye and an image of the ganglion is projected by a microscope
objective lens onto an array of 144 or 464 tiny photodiodes, each of which
Optical monitoring of neuronal activity 205
measures the amount of light passing through a small part of the ganglion
(Fig. 8.1d). The electrical signal produced by each diode is amplified and
processed to produce pulses that indicate the occurrence of spikes. Some
neurons are large enough to cover more than one photodiode, and other
diodes pick up spikes from more than one neuron. Nevertheless, these
experimenters are able to process data in a way that records the occurrence
of spikes in about a third of the neurons of the abdominal ganglion at one
time, although they cannot assign spikes to identified neurons.
One of the first discoveries made using this technique was that hundreds
of neurons in the abdominal ganglion are active during any gill-withdrawal
reflex (Zečević et al., 1989). This observation does not allow us to distin-
guish between the two arrangements shown in Figs. 8.1b and 8.1c because
we need to know whether each neuron is active only during the reflex with-
drawals, or during other behaviours as well. Subsequently, three different
kinds of gill-withdrawal movements were studied: reflex withdrawal; weak,
spontaneously occurring withdrawals; and strong withdrawals that occur
during respiratory pumping (Wu, Cohen & Falk, 1994). During each of these
movements, between 62 and 72 neurons were usually active and almost all
of these neurons were active during all three behaviours. This means that
something like 200 neurons in the abdominal ganglion, most of which are
interneurons, would be excited during any activity of the gill-withdrawal
muscles. The neuronal networks responsible for controlling gill-withdrawal
movements seem, therefore, to be organised in a distributed pattern like
that in Fig. 8.1c, with few neurons active during one behaviour but not
during the other behaviours.
Control of gill movements is a major concern of the neurons in this gan-
glion and many of the other activities that it controls, such as heart beat,
need to be co-ordinated with gill movements. However, the abdominal gan-
glion is also responsible for controlling a more dramatic defensive response
to intense stimulation of the skin, which involves expulsion of a dense
cloud of dark purple ink from a special gland. The inking behaviour con-
trasts with gill-withdrawal in that it is an all-or-none behaviour (Carew &
Kandel, 1977; Byrne, 1981). The same sensory neurons that trigger gill with-
drawal responses also trigger inking, and they communicate through inter-
neurons to three, strongly coupled motor neurons. It would be interesting
to investigate whether these three interneurons are dedicated to inking
behaviour, or whether they are also active during other behaviours.
The abdominal ganglion of Aplysia is suited to optical monitoring of

neuronal activity because it is quite small and transparent. However, a
major drawback of this technique is that it cannot be used to stimulate or
inhibit an individual neuron. In some other ganglia, the technique is
capable of resolving small changes in membrane potential, so that there is
a chance that it can be used to trace circuits between neurons by correlat-
ing the occurrence of spikes in some neurons with the postsynaptic poten-
tials they cause.
8.4 Local bending reflexes in the leech
If you touch a small area of the skin of a leech, it will bend its body away
from the touch, a movement caused by contraction of longitudinal muscles
in a few segments close to the site of the touch. The body of the leech locally
assumes a U shape. Touching the top of the body causes a downward bend,
whereas touching the left side causes a bend towards the right, and so on
(Fig. 8.2a). The main sensory neurons involved in triggering these local
reflex movements are pressure-sensitive P neurons. Each segmental gan-
glion contains four P neurons, each of which has a receptive field that
covers one quadrant of the skin of its segment – upper or lower right, and
upper or lower left. The muscles that produce the bending movements run
longitudinally along the body wall, and in each segment there are six bilat-
eral pairs of dorsal, ventral and lateral longitudinal muscles. Each dorsal or
ventral longitudinal muscle is controlled by two motor neurons, an excitor
and an inhibitor, and each lateral muscle is controlled only by an excitatory
motor neuron.
Excitation of a particular set of P neurons reliably activates motor
neurons in a particular way (Fig. 8.2b). No direct connections between P
neurons and motor neurons have been found, so interneurons must be
involved in the local bending reflexes. A straightforward way of arranging
the circuitry of a ganglion to control these local avoidance reflexes would be
to connect sensory neurons with motor neurons through interneurons that
are responsible for particular reflex behaviours, in the manner indicated in
Fig. 8.2c. Shawn Lockery and Bill Kristan characterised these interneurons
by using intracellular recording and staining (Lockery & Kristan, 1990a,
1990b). They studied interneurons that are involved in dorsal bending, the
production of a downward U shape when the top surface of a leech is
touched. In a survey of 73 ganglia, they found nine types of interneuron
Local bending reflexes in the leech 207
Figure 8.2 Local bending reflex movements of the leech (Hirudo). (a)
Drawings of a leech viewed from its right side: resting (top), and responding
to touches to the top (middle) or side (bottom), as indicated by the arrows.
(b) Intracellular recordings from the excitor motor neuron of the left dorsal
muscle in response to stimulation of P neurons. In the upper and middle
recordings, the four P neurons were stimulated individually. In the lower
recordings, lateral touches were mimicked by stimulating the dorsal and
ventral P neurons of either side simultaneously. (c) A circuit that incorpo-
rates interneurons which are dedicated to particular bending movements.
The triangles indicate excitatory synapses and the circles indicate inhibitory
synapses. (a redrawn after Lockery et al., 1989; b and c redrawn from
Lockery & Kristan, 1990a.)
that were both activated by dorsal P neurons and made connections

onward to dorsal bend motor neurons. Each type had a distinct morphol-
ogy, allowing it to be distinguished from others and to have its cell body
position identified on a map of a segmental ganglion. One, neuron 125, is
illustrated in Fig. 8.3a.
Each interneuron responded to touch anywhere on the skin of its
Figure 8.3 Interneuron 125, which is involved in local reflex bending move-
ments in the leech. (a) A diagram of a ganglion in which the neuron was
stained by intracellular injection of the fluorescent stain lucifer yellow. (b)
Amplitudes of EPSPs in a left 125 caused by stimulating the four different P
neurons (dorsal and ventral left and right) of the same segment. Each bar
indicates, from three experiments, the mean amplitude and standard error
of the EPSP evoked by stimulating a particular P neuron. (c) Amplitudes of
postsynaptic potentials evoked by spikes in interneuron 125 in the excitor
(ex.) and inhibitor (inh.) motor neurons of dorsal and ventral longitudinal
muscles. Upward-directed bars indicate EPSPs and downward-directed bars
indicate IPSPs. Means and standard errors are indicated, as in (b). (a
redrawn after Lockery & Kristan, 1990b.)
segment (Fig. 8.3b), and made connections with most of the longitudinal
motor neurons in its ganglion (Fig. 8.3c). None of the interneurons
appeared to be dedicated to just the dorsal reflex bending behaviour
because each was also active during ventral or lateral bending. Therefore,
the networks responsible for analysing touch stimuli and triggering
bending movements seem to be arranged in a distributed manner in which
particular interneurons are not assigned to specific local bending
responses. However, is it possible that other interneurons, not found by
Lockery and Kristan, are the ones that are primarily responsible for specify-
ing particular bending responses?
One test for the involvement of a neuron in a behaviour is to remove
that neuron and observe any change in behaviour. When hyperpolarising
currents were injected into single interneurons to reduce their excitabil-
ity, responses by motor neurons to stimulation of P neurons were
reduced in amplitude. But these experiments were inconclusive because
the reductions in response by motor neurons were small, which is not
Modelling a network on neurons 209
Figure 8.4 A diagram indicating the kind of organisation of the circuit that
controls local bending movements in the leech. The diagram is simplified in
many ways so that it does not indicate connections between left and right
neurons or the relative strengths of different connections, and includes only
three of the 17 interneurons known to exist in each ganglion.
surprising because each motor neuron is driven by several different

interneurons.
8.5 Modelling a network of neurons
In order to find out whether specific behavioural responses could be pro-

duced using interneurons with such unspecific characteristics, computing
techniques were used to construct a model of the network involved in leech
local bending (Lockery et al., 1989). Software represented three levels of
units, with one level equivalent to sensory neurons, one equivalent to
motor neurons, and the third interposed between them like the interneu-
rons (Fig. 8.4). The 18 interneuron units connected only with sensory and
motor neurons and not with each other, which is the same pattern as in the
leech. The strengths of connections between interneurons and sensory or
motor neurons were variable.
In an experiment, a model network was gradually trained to reproduce
the behaviour of a leech segmental ganglion, so that activation of a partic-
ular sensory neuron would generate an appropriate local bending-avoid-
ance response. The way in which this was done was first to programme the
computer to make connections between sensory neurons and interneu-
rons and between interneurons and motor neurons. The pattern of the
connections was set randomly, except that interneurons were arranged as

left–right pairs, and all the connections were initially weak. The perfor-
mance of the circuit was then tested to find out how closely its behaviour
matched that of the leech. It would be very unlikely to perform well in this
first test. Next, the computer altered the strengths of some of the connec-
tions, and the behaviour of the circuit was again tested. If the performance
of the circuit had been improved, the same connections were again
strengthened, but if it had deteriorated, these connections were weakened
and others were strengthened. By repeatedly altering connection strengths
and then testing the circuit, the computer gradually improved the circuit in
a stepwise manner until the input–output relationship became the same as
that observed in a leech. This technique of training a model network of
neural elements is called back propagation. Successful training of the
network required between 10 000 and a 100 000 repetitions of the training
cycle.
Eighteen different experiments were performed. Each one produced a
network that reproduced the behaviour of a leech but was different from
the other 17 networks. All the networks contained model interneurons
that resembled the leech interneurons in the lack of specificity in their
connections, with more than 95 per cent receiving inputs from two or
more P cells and 88 per cent making output connections to at least seven
of the eight motor neurons. This means that it is possible to adopt a
strategy of using diffusely connected elements in a network to link inputs
and outputs in specific ways. Indeed, because the computer program
always constructed a diffuse network rather than one in which individual
units were dedicated to restricted functions, the strategy must be a good
one.
In a leech segmental ganglion, therefore, each of the interneurons
involved in dorsal bending makes many connections that are inappropriate
to that response. Stimulation of a particular area of skin always causes the
same bending response, which means that the particular balance of the
strengths of connections within the ganglion must offset the apparently
inappropriate nature of many of them. Although computer modelling gen-
erated a number of different circuits for controlling local bending, it is likely
that the network of neurons in one leech is very similar to that in another.
It would be interesting to know what factors in evolution have selected this
particular network. Careful use of computer modelling techniques, such as
Modelling a network on neurons 211
this study on the leech, is an invaluable tool for neuroethology because it

allows an experimenter to test whether a neuronal network operates in the
way expected. An alternative approach to reconstructing circuits is to
isolate individual neurons and allow them to make synaptic connections in
culture (Box 8.1).
Box 8.1. Reconstructing circuits of live neurons
A very direct test of whether a circuit of neurons is sufficient to gener-

ate an activity is to isolate the circuit concerned from the nervous
system. In the lobster stomatogastric ganglion, which contains a very
small number of neurons this has been done by killing some neurons,
so that only those involved in a particular circuit are left. Another
technique was adopted by Syed, Bulloch & Lukowiak (1990) in a study
of the pond snail, Lymnaea. This is a diving, air-breathing animal; it
has a lung, and comes to the surface to exchange the air it contains by
rhythmically opening and closing the aperture, called the pneu-
mostome (a). Experiments using intracellular recording revealed a
simple circuit of three interconnected interneurons that are active
during the ventilatory rhythm (b). Is this circuit sufficient to generate
the ventilatory rhythm, or are additional neurons required? In order to
test this, cell bodies of the three neurons were carefully dissected from
the central nervous system and placed in a dish containing tissue
culture medium. Quite quickly, the isolated cell bodies began to
sprout new processes and, within a day, the three neurons established
synaptic connections with each other, in the same pattern as in the
intact brain. Brief electrical excitation of one of the neurons caused
several cycles of rhythmical activity very similar to that found in an
intact brain. None of the neurons was capable of generating rhythmi-
cal activity on its own and, although I.P3.1 and V.D4 (which connect
with motor neurons) made a simple circuit in which each inhibited
the other, rhythmical activity was not produced unless the giant,
dopamine-containing cell (Rpe.D1) was also present.
8.6 Local reflex movements of a locust’s leg
Jointed limbs, such those of a mammal or an insect, are used in a wide

variety of ways. Besides being involved in locomotion, they are capable of
finely controlled movements such as grooming, in which the end of the
limb is brought precisely to a particular spot on the animal’s surface. Since
the 1970s, Malcolm Burrows and colleagues have investigated in detail the
neurons responsible for controlling local movements of a locust’s hind leg,
providing a very complete description of the links between sensory analy-
sis and motor control (see Burrows, 1996). Compared with the local
bending movements of the leech, local reflexes of a locust’s leg involve
many more neurons, arranged in two layers between the sensory and
motor neurons. Each of these layers consists of local interneurons, which
have their branches restricted to the third thoracic ganglion, often to one
part of it. The first layer mostly receives inputs from sensory neurons and
consists of spiking interneurons. The second layer makes connections with
motor neurons, and consists of non-spiking interneurons. The major route
for the flow of information is from sensory neurons to spiking interneurons,
then to non-spiking interneurons and finally to motor neurons. Within this
route there are no feedback pathways, so motor neurons do not synapse
back onto either type of interneuron, and non-spiking interneurons do not
synapse back onto local spiking interneurons. Feedback is provided by
sense organs that monitor movements of the limbs, and these mainly
connect with interneurons rather than directly with motor neurons.
If a locust’s leg is touched anywhere on its surface, it will be moved away
from the touch (Fig. 8.5a). Usually, the movement involves several joints so
that, for example, when the ventral surface of the tibia is touched, the tibia is
raised and during this response the angle between the tibia and the foot
changes to keep the foot parallel with the ground. These movements are con-
trolled by neurons contained within the third thoracic ganglion, including
about 70 motor neurons and a few hundred local interneurons. About 10 000
sensory neurons project into each half of the third thoracic ganglion, and
many of these originate on the surface of the leg, so there is a great deal of
convergence as information flows from sensory neurons to motor neurons.
Many of the sensory neurons belong to basiconic sensilla, each of which
consists of a short tactile hair with one mechanosensory neuron and four
chemosensory neurons. Trichoid sensilla are another type of sense organ
Local spiking interneurons 213
Figure 8.5 Local reflex movements of a hind leg of a locust, Schistocerca. (a)
Movements evoked by touching the tibia on its anterior or posterior (large
arrows). The leg is shown as an outline in its initial position and in black in its
final position. (b) The projection of a trichoid sensillum on the right foot into
the third thoracic ganglion, viewed from below. The dashed line indicates the
extent of the neuropile into which sensilla on different parts of the right leg
project. Only the anterior section of the right half of the ganglion is drawn. (a
redrawn after Siegler & Burrows, 1986; b from Newland, 1991; reprinted by
permission of Wiley-Liss, Inc., a subsidiary of John Wiley & Sons Inc.)
which consist of a sensory cell that responds phasically when its hair
(which can be up to 0.8 mm long) is deflected towards the centre of the
body. Other types of sense organs, such as campaniform sensilla (see
Chapter 4), monitor strains in the cuticle or span joints, responding when-
ever the leg is moved. Sensory neurons project in an orderly manner into
the ganglion, so that their terminal branches form a somatotopic map of
the surface of the leg (Newland, 1991). All the hair-like sensilla project to a
particular region of neuropile in the ventral part of the third thoracic gan-
glion. A sensillum near the foot projects to a relatively posterior region of
this neuropile (Fig. 8.5b), whereas a sensillum close to the joint of the femur
with the base of the leg projects more anteriorly.
8.7 Local spiking interneurons
Each local spiking interneuron has two layers of branches (Fig. 8.6a).
Studies with the electron microscope have shown that the ventral layer
Figure 8.6 Local spiking interneurons of the locust third thoracic ganglion.
(a) Diagrams of an interneuron showing its dorsal and ventral fields of
branches. This interneuron was excited by hairs on the ventral surface of
the tibia near to the foot, and by proprioceptors that responded to exten-
sion of the tibia. The two oval shapes indicate locations of cell bodies of two
different groups of spiking local interneurons. (b) Intracellular recordings of
the responses by a local spiking interneuron to stimulation of trichoid sen-
silla at three different locations on the femur, indicated in the diagram. Two
or three recordings for each hair are superimposed. The lower trace shows a
spike from the sensory neuron, and the upper trace is the intracellular
recording from the interneuron. The middle recording shows the strongest
connection: each sensory neuron spike evoked a spike in the interneuron,
whereas spikes in the other two sensory neurons evoked subthreshold
EPSPs. (a from Burrows, 1985; reprinted with permission of the American
Physiological Society; b modified after Burrows, 1992b; reprinted with per-
mission from Trends in Neuroscience; copyright © 1992 Elsevier Science.)
mostly receives input synapses, whereas the dorsal layer mostly makes
output synapses (Watson & Burrows, 1985). The two layers are connected
by a long, thin neurite that may be an axon, carrying spikes from the ventral
to the dorsal layer, but its small size makes it hard to confirm this experi-
mentally. One group of local spiking interneurons have their cell bodies
clustered near the middle of the ventral surface of the ganglion, and these
interneurons all make inhibitory output synapses (Siegler & Burrows, 1984).
Another group have their cell bodies close to the anterior connective, and
these interneurons make excitatory output synapses (Nagayama, 1989).
Each local spiking interneuron responds to stimulation of basiconic and
trichoid sensilla in a well-defined receptive field on particular regions of the
leg (Burrows & Siegler, 1985; Burrows & Newland, 1993). Some interneurons
have fields restricted to particular parts of one segment of the leg, but
Non-spiking interneurons 215
others have broader receptive fields that extend in both directions from a
joint. The interneuron from which the recordings in Fig. 8.6b were made
responded to stimulation of trichoid sensilla on the ventral surface of the
femur and there was a gradient in the strength of connections that declined
in a direction away from the body. In the most sensitive part of the recep-
tive field, a single spike in a sensillum will often be sufficient to elicit a spike
in a local spiking interneuron. However, when two sensory neuron spikes
occur in quick succession, the second may only elicit a small PSP, and this
change in the effectiveness of the synapses is a mechanism that enhances
responsiveness by the locust to new, rather than maintained, tactile stimuli
(Burrows, 1992a).
Each spiking interneuron excites or inhibits a number of different targets,
which are mainly non-spiking interneurons and motor neurons. The overall
pattern of connectivity is diffuse, because each non-spiking interneuron or
motor neuron receives inputs from a number of different local spiking
interneurons. However, the pattern of connections preserves information
about the location of tactile stimuli on the surface of the leg, so that non-
spiking interneurons and motor neurons, as well as the local spiking inter-
neurons, have well-defined receptive fields. When part of the leg is touched,
therefore, a particular set of local spiking interneurons and another set of
non-spiking interneurons are activated.
8.8 Non-spiking interneurons
The non-spiking interneurons (Fig. 8.7a) provide a major source of input to

leg motor neurons. As their name suggests, these interneurons do not spike
when they are excited, but changes in membrane potential directly affect
the rate at which neurotransmitter is released from their output synapses.
There is a smoothly graded relationship between the membrane potential
of a non-spiking interneuron and of a motor neuron that it drives (Burrows
& Siegler, 1978), which is seen when one electrode is used to inject current
into a non-spiking interneuron while another one is used to record the
membrane potential of a postsynaptic motor neuron. The same relation-
ship between presynaptic and postsynaptic potentials is found at all chem-
ical synapses, but the all-or-nothing nature of a spike often obscures it. In
the recordings shown in Fig. 8.7b, recordings were made from two different
motor neurons at the same time. The interneuron inhibited the flexor
Figure 8.7 Non-spiking interneurons in the third thoracic ganglion of a

locust. (a) The morphology of a non-spiking interneuron that excites the
slow extensor tibiae motor neuron. (b) Intracellular recordings that show
the graded nature of transmission from a non-spiking interneuron. One
electrode was used to inject pulses of depolarising current into an interneu-
ron, while a second electrode recorded the intracellular potential of a motor
neuron of a leg extensor muscle, and a third electrode recorded from a
motor neuron of a leg flexor muscle. When the strength of current
increased, the potential changes in the two motor neurons also increased.
As the circuit on the right indicates, the non-spiking interneuron probably
made a direct, inhibitory connection with the flexor motor neuron, and
excited the extensor motor neuron by reducing tonic inhibition from
another non-spiking interneuron. (a from Watkins, Burrows & Siegler, 1985;
reprinted by permission of Wiley-Liss, Inc., a subsidiary of John Wiley &
Sons Inc.; b modified after Burrows, 1989.)
motor neuron and excited the extensor, probably by disinhibition of a

second, inhibitory non-spiking interneuron (cf. section 7.6).
Many of the non-spiking interneurons release neurotransmitter toni-
cally, exerting a steady synaptic drive on their postsynaptic targets. One
important result is that they regulate the passage of sensory information to
motor neurons, illustrated as follows for a pathway in which a sensory
neuron excites a non-spiking interneuron which, in turn, excites a motor
neuron. If the non-spiking interneuron is relatively hyperpolarised as a
result of a particular combination of sensory inputs, its membrane poten-
tial will be below the threshold for transmitter release and its output synap-
ses will effectively be switched off. As a result, exciting the sensory neuron
to produce a spike will cause an EPSP in the interneuron, but this will be too
Organisation of neurons that control reflex movements 217
small to cause the interneuron to release neurotransmitter and excite the

motor neuron. If the angle of the joint is altered, the new combination of
sensory inputs onto the interneuron might excite it sufficiently for it to
release transmitter tonically, which will excite the motor neuron with a
steady, depolarising potential. Now if the sensory receptor is excited, the
EPSP it causes in the interneuron will be passed on to the motor neuron
(Burrows, 1979). The non-spiking interneurons, therefore, can act as
summing points, where signals including those from proprioceptors, tactile
sense organs, and interneurons that co-ordinate activity between different
segments are integrated before being passed on to the motor neurons. The
result is that local reflex movements are modulated according to behaviou-
ral context. Interestingly, when the angle of a joint changes, the size of the
change in membrane potential that occurs in non-spiking interneurons
depends on whether the joint angle has increased or decreased (Siegler,
1981). This means that the properties of the interneuron and its output syn-
apses depend upon the recent history of movements the locust has made.
8.9 Organisation of neurons that control reflex movements
A summary of the main patterns of interconnections found among neurons

controlling local reflexes of a locust’s leg is drawn in Fig. 8.8. Some non-
spiking interneurons make excitatory output connections and others make
inhibitory connections, but none has been found to excite some of its
targets and inhibit others. As with the local spiking interneurons, the
pattern of output connections from non-spiking interneurons is diffuse in
that a non-spiking interneuron can control several different motor neurons
and each motor neuron is controlled by several non-spiking interneurons
(Burrows, 1980). Some non-spiking interneurons connect laterally with
others, always making inhibitory connection, which might help to ensure
that incompatible movements involving antagonistic muscles do not occur.
The most intriguing aspect of the organisation of the neurons that
control local reflexes of a locust’s leg is that one type of interneuron oper-
ates with spikes and the other without them. The situation contrasts with
the way information is processed in visual systems, where the first layers
of neurons often operate without spikes but later neurons transmit trains
of spikes along axons. It is interesting that all the interneurons that have
been found to drive flight motor neurons in locust thoracic ganglia
Figure 8.8 A circuit diagram indicating the kinds of connections that have
been discovered between neurons involved in local reflexes of a locust’s
hind leg. Excitatory connections are indicated by triangles, and inhibitory
connections by circles.
produce spikes (see section 7.6). It might be that the graded way in which
non-spiking interneurons control motor neurons is useful for the kinds of
small postural adjustments that legs can make. Spiking rather than non-
spiking communication is not simply related to neuronal size, because
both the extent and the complexity of the branching pattern are similar in
non-spiking and spiking local interneurons (compare Fig. 8.6a with Fig.
8.7a). It is possible that signals other than spikes cannot travel along the
narrow, constricted neurite that connects the dorsal and ventral fields of a
spiking local interneuron. Spikes may also provide a mechanism for syn-
chrony by depolarising all of a neuron’s presynaptic terminals together and
by a similar amount. In contrast, the use of smoothly graded membrane
potentials in a non-spiking interneuron probably allows different regions
of the neuron to operate independently to some extent. In support of this
idea, electron microscopy has shown that input and output synapses are
intermingled along all the processes of non-spiking interneurons and no
separate regions of input or output can be distinguished (Watson &
Burrows, 1988), and signals do seem to reduce in amplitude as they travel
from one part of a non-spiking interneuron to another (Burrows & Siegler,
1978).
Conclusions 219
8.10 Conclusions
All three studies in this chapter have described interneurons that are not
dedicated to particular behaviours, but which operate as parts of ensem-
bles in the interface between sensory analysis and motor control. Whenever
an Aplysia retracts its gill, many interneurons are active no matter whether
the movement is a reflex in response to a sensory stimulus or a movement
associated with respiratory pumping. A similar conclusion applies to the
interneurons involved in local bending movements of the leech because
most of them respond to touch anywhere on the surface of their segment.
In the locust, both local spiking and non-spiking interneurons have distinct
receptive fields in their sensory responses, so are probably more restricted
in their patterns of activity. Experiments in the leech and locust involving
paired recordings with microelectrodes in order to construct circuit dia-
grams of the neurons involved have shown that information mostly flows
through one or two layers of interneurons rather than directly from sensory
to motor neurons. Each interneuron sums inputs from a number of presyn-
aptic sources and distributes outputs onwards to many different targets.
Information flows in one direction, with no feedback from the motor
neurons to the interneurons. Although the pattern of connections made by
interneurons is complex, the movements controlled by each motor neuron
are oriented appropriately in response to stimulation of an area of skin.
It is difficult to appreciate how networks of neurons organise well-co-
ordinated behaviour simply be examining a diagram of the connection
pattern. Computer modelling can be very useful because factors such as the
strengths of different connections can be incorporated, and because an
experimenter can trace the way that information flows sequentially
through the network. Also, as the study with the leech showed, modelling
with computers can test whether a particular network works in a specific
way.
There is undoubtedly a continuum in the degree to which interneurons
are dedicated to particular behaviours, with some giant interneurons (see
Chapter 3) being apparently concerned with just one behaviour. Probably
the major advantage in using neurons that are not strongly dedicated to
particular behaviours is that the circuits are flexible, so that different activ-
ities in which an animal is engaged are properly integrated together.
Although we can neatly categorise different activities performed by an
animal as different patterns of behaviour, the diffuse nature of the organ-

isation of nervous systems makes it difficult to recognise discrete circuits
that are concerned with particular behaviours.
Further reading
Posner, M.I. & Raichle, M.E. (1997). Images of Mind, 2nd edn. New York: Scientific
American Library. A book that describes a number of ways that have been used
to monitor changes in the activity of populations of neurons in the human
brain, often associated with the performance of particular tasks. A good
example is the technique of positron emission tomography, which has pro-
vided much valuable clinical information. This and other techniques monitor
changes in the flow of blood to different brain regions, providing an indication
of how active they are.
Lewis, J.E. & Kristan, W.B. Jr (1998). A neuronal network for computing population
vectors in the leech. Nature 391, 76–9. This describes work that extends the
study of the interneurons responsible for controlling local bending movements
in the leech by examining how information about the exact location of a touch
to the animal’s surface is coded by a population of neurons.
Jellema, T. & Heitler, W.J. (1997). Adaptive reconfiguration of a reflex circuit during
different motor programmes in the locust. J Comp Physiol A 180, 659–69.
Before a jump by a locust, both the extensor and the flexor muscles of the hind
leg tibiae contract together, building up tension for sudden release causing the
jump. These muscles are activated in a different pattern during walking or
scratching, and this paper reports one aspect of how the circuitry is modified,
by altering the strengths of synapses from proprioceptors.
9 Nerve cells and changes in behaviour
9.1 Introduction
One of the most important, intriguing aspects of animal behaviour is that

it continually changes. Some of the changes are parts of the processes of
development and maturation, whereas others allow the animal to learn
about alterations in its environment. Learning enables an animal to make
and modify predictions based on experience, for example that a particular
action will be followed by a rewarding or an aversive event. The ability to
change is a fundamental property of many nerve cells and their intercon-
nections, and much recent research has focused on events at the molecu-
lar and cellular level that could underlie learning or events during the
development of a nervous system. A major challenge in neuroethology is to
relate these changes in cellular properties to alterations in animal behavi-
our.
One type of trigger for changes in behaviour is provided by hormones,
which can ensure that events are initiated at particular times. Steroid hor-
mones act by regulating gene expression and can produce modifications in
the morphology of nervous systems that are correlated with changes in
behaviour (Breedlove, 1992; Weeks & Levine, 1995). Polypeptide hormones,
on the other hand, often act to trigger particular behaviour patterns by
exciting particular target neurons. Another type of mechanism triggers
learning, in which an animal forms a new association between a sensory
stimulus and a motor program. This commonly occurs when a particular
sequence of stimuli is received by sensory pathways, leading to changes in
neuronal excitability and the strengths of synapses. In the longer term,
learning can also involve alterations in the pattern of gene expression in a
neuron.
221
9.2 Growth and metamorphosis in insects
During the life history of many animals, the body form alters as the animal
grows and matures. Each stage has a different pattern of behaviours from
the other stages, which means that quite profound changes can occur in
nerve cells and the circuits in which they participate. The mechanisms that
underlie such changes have been extensively studied in a type of hawk
moth, the tobacco hornworm Manduca sexta. The first stage in the life cycle
after hatching from the egg is the larva, which spends most of its time
eating, walking and growing. The second stage is the pupa, which is almost
immobile and undergoes a major reorganisation of tissues, or metamor-
phosis, to produce the final, adult stage. The lifestyles of all three stages are
quite different and, together with changes in body form, the sensory, mus-
cular and nervous systems are extensively remodelled. The life history of
Manduca is punctuated by ecdyses, in which the animal sheds its old skin
and its whole behaviour is dedicated to ensuring an orderly transition from
one developmental stage to the next.
9.3 Ecdysis in a hawk moth
Towards the end of each developmental stage, the larva grows a new, soft
exoskeleton underneath the old one, and the behaviour of ecdysis loosens
the new skin from the old and then allows the old skin to be shrugged off.
Ecdysis lasts for about six hours, and it provides an excellent example of
how particular neurohormones can orchestrate a co-ordinated behaviour
at a particular time during an animal’s life history. The activity of a small
number of neurons commits the whole animal to ecdysis.
Manduca has five larval stages, each one almost identical in form to the
last but slightly larger, until it becomes a pupa. This final larval ecdysis is
the best understood. The first event is inflation of tracheal air sacs, partic-
ularly in the head, with air. Next, during pre-ecdysis, muscles contract and
then relax synchronously in all segments of the abdomen once every 5–10 s
(Fig. 9.1a; Copenhaver & Truman, 1982; Miles & Weeks, 1991). At the same
time, abdominal appendages called pro-legs are retracted and re-
extended. Pre-ecdysis loosens the old cuticle from the new and takes about
half an hour. The rhythm for pre-ecdysis is generated by a network of inter-
neurons in the last abdominal ganglion, and a single left–right pair of long
Ecdysis in a hawk moth 223
Figure 9.1 Ecdysis and its control in the hawk moth (Manduca sexta).
(a) Representation of the patterns of activity that occur in pre-ecdysis and
in ecdysis. Times when the same muscle in the third (A3) and fourth (A4)
abdominal segments contract are shown. (b) A flow diagram to illustrate the
actions of ecdysis-triggering hormone, released from Inka cells, and eclo-
sion hormone, released from four brain cells. The pre-ecdysis central
pattern generator (cpg) is activated by low concentrations of eclosion-
triggering hormone, and the pre-ecdysis rhythm is communicated to
segmental motor neurons (mn) by a pair of co-ordinating interneurons
(int). Eclosion is switched on by crustacean cardioactive peptide, released
from CCAP cells throughout the central nervous system, and this peptide
also switches off pre-ecdysis.
interneurons communicates the rhythm to other abdominal ganglia,

ensuring that contractions occur synchronously in all segments (Novicki
& Weeks, 1995). Ecdysis uses the same muscle as pre-ecdysis, but in a
different motor pattern in which a wave of contraction progresses anteri-
orly from segment to segment (Fig. 9.1a; Weeks & Truman, 1984). The
neurons responsible for generating the ecdysis motor program have not
yet been located.
Two polypeptide hormones are responsible for triggering and orchestrat-
ing pre-ecdysis and ecdysis (Fig. 9.1b; Zitnan et al., 1996; Gammie &
Truman, 1997). First, ecdysis-triggering hormone is released into the blood
from cells called Inka cells which are situated near to the openings of the
respiratory tracheal system to the outside. The pattern generator for pre-
ecdysis is very sensitive to ecdysis-triggering hormone and is initiated as
soon as the concentration of this hormone starts to rise. Another effect of
ecdysis-triggering hormone is to excite four cells in the brain that release
the second polypeptide, eclosion hormone. These cells were identified with
a stain for the RNA for eclosion hormone. Their axons extend right down
the nerve cord to the hind end of the animal and then run anteriorly in
nerves that lie alongside the gut. Activation of these cells depends on levels
in the blood of the steroid hormone ecdysone, the hormone that acts as a
master switch for the genes that instruct the production of new body struc-
tures during metamorphosis. Eclosion hormone is released both within the
central nervous system and into the blood, and the Inka cells are excited by
it, so there is a feed-forward system between the Inka cells and the eclosion
hormone-releasing brain cells. Once activated, this mutual excitation is
unlikely to stop, ensuring the animal is fully committed to ecdysis. The Inka
cells and the brain cells discharge all of their hormone within about half an
hour.
In the central nervous system, the targets for eclosion hormone are a set
of 50 nerve cells, scattered fairly uniformly among the segmental ganglia.
The effect of eclosion hormone is to excite these cells, by way of an intracel-
lular messenger pathway involving cyclic GMP, and to cause them to release
another peptide hormone (called crustacean cardioactive peptide). This
peptide ensures the correct change in motor program by activating the
pattern generator for ecdysis and at the same time switching off the pattern
generator for pre-ecdysis. Ecdysis is sustained after the initial intense
release of ecdysis-triggering and eclosion hormones because the increase
Remodelling neurons during metamorphosis 225
in cyclic GMP is quite slow in the cells that release crustacean cardioactive
peptide and they maintain a steady level of discharge. These cells are also
excited by sensory cells that signal the continued presence of the old
cuticle.
9.4 Remodelling neurons during metamorphosis
The lifestyle and behaviour of the adult moth are quite different from those
of the larva, and the two stages have completely different locomotory and
sensory systems. The larva moves by crawling movements involving stubby
pro-legs on the abdomen and thorax, and these legs and their muscles dis-
appear during the pupal stage. Almost all of the larval muscles die during
metamorphosis, and are replaced by new muscles that move adult struc-
tures such as wings and thoracic legs. The adult has more sophisticated
sense organs than the larva, such as compound eyes and antennae. Rather
than being dismantled and rebuilt, the central nervous system is exten-
sively remodelled. A few new neurons are born and some die, but many are
remodelled and incorporated into new circuits.
One neuron that is remodelled is MN1, a motor neuron found in each
abdominal ganglion (Levine & Truman, 1982). In the larva, the muscle it
controls bends the abdomen sideways. This muscle dies during metamor-
phosis, and MN1 comes to innervate a new muscle that bends the
abdomen upwards. This means that the left and right MN1 neurons are
often used in opposition to each other in the larva, but are used together
in the adult. Associated with the change in target muscle are changes in the
morphology of MN1 and its connections with other neurons (Fig. 9.2). In
the larva, dendrites of the neuron are nearly all on the same side of the
ganglion as its axon, but it sprouts new dendrites during the pupal stage,
and has a separate dendritic region on each side of the ganglion in the
adult.
In the larva and in the adult, MN1 is excited by a stretch receptor which
responds whenever the muscle controlled by MN1 is stretched. This stretch
receptor probably makes a direct, excitatory connection with MN1 (the
unbroken arrow in Fig. 9.2) because each sensory spike evokes an EPSP in
MN1 after a short delay. Therefore, there is a short feedback pathway that
excites MN1 whenever its muscle is stretched. The stretch receptor on the
other side of the segment also participates in a circuit with MN1, but this
Figure 9.2 Changes in the morphology and central connections of an

identified motor neuron, MN1, in Manduca. The MN1 drawn innervated a
muscle on the left side of an abdominal segment but its cell body is on the
right side of the ganglion, which is a relatively unusual arrangement in an
insect. Vertical dotted lines indicate the midline of the ganglion. Arrows
beneath each diagram indicate connections made by identified stretch
receptor cells on the left and right sides of the animal. (Drawings of larva,
pupa and adult modified after Levine & Weeks, 1990; MN1 modified after
Truman & Reiss, 1988; reproduced with permission; copyright Society for
Neuroscience.)
Remodelling neurons during metamorphosis 227
circuit alters during metamorphosis. In the larva, a spike in the right stretch
receptor causes an IPSP in the left MN1. The delay between the spike and
the IPSP is longer than the delay for the EPSP, so the circuit probably
includes an interneuron, and is indicated by a broken arrow in Fig. 9.2. The
different inputs from the two receptors help to ensure that the larva tends
to straighten its abdomen if it becomes bent, and that the left and right
MN1 neurons act in opposing ways.
During metamorphosis, the stretch receptor continues to make connec-
tions with the left and right MN1 motor neurons. It continues to excite the
MN1 on the same side of the segment, but its inhibitory connection with
the opposite MN1 becomes weaker and is then replaced with a direct excit-
atory connection. What is almost certainly occurring is that the new den-
drites of the motor neuron come into contact with the right stretch
receptor, and excitatory synapses form between the two neurons. In the
adult, therefore, the left and right stretch receptors are excited whenever
the abdomen is bent downwards, and act in a new proprioceptive reflex
that excites the left and right MN1s. Up and down movements of the
abdomen occur during flight-steering manoeuvres, during mating and
during egg laying.
The growth of new dendrites of MN1 occurs at the same time as a rapid
rise in the level of ecdysone within the body during the pupal stage
(Truman & Reiss, 1988). Experiments on diapausing pupa, in which the
emergence of the adult can be delayed for months in response to environ-
mental stress, have shown that it is ecdysone that causes the motor
neuron to grow new dendrites. In diapausing pupae, the concentration of
ecdysone is unusually low, and MN1 does not grow new dendrites.
Injecting ecdysone into a diapausing pupa causes MN1 to sprout its new
dendrites.
The effect that ecdysone has on MN1 is regulated by another hormone, a
turpenoid called juvenile hormone. If juvenile hormone is present when
ecdysone levels rise, MN1 retains its larval characteristics. The same com-
bination of rising levels of ecdysone and low levels of juvenile hormone has
widespread effects on the central nervous system and elsewhere in the
body (Streichert & Weeks, 1995; Levine, Morton & Restifo, 1995). It causes
many of the larval muscles and some of the motor neurons to die, as well as
triggering the development of new dendrites in motor neurons that survive,
such as MN1. However, not all of the events involved in remodelling the
nervous system are triggered by rising ecdysone levels. For example, the
inhibitory connection from the right abdominal stretch receptor to the left
MN1 disappears before the rise in ecdysone, and its cause is not yet known.
9.5 Associative learning and the proboscis extension reflex in

honey bees
Some of the most interesting behaviour patterns amongst insects are found
in the social hymenoptera, bees, wasps and ants. These animals are capable
of memorising landmarks, which they use for navigation (Dyer, 1996; Judd
& Collett, 1998), and they have elaborate systems for communication
amongst individuals, such as the waggle dance of honey bees (Lindauer,
1967). The disposition of bees to learn to associate colours and shapes of
flowers with a good source of food was recognised by one of the earliest
experimental ethologists, Karl von Frisch, and he exploited this to investi-
gate several features of the sensory capabilities of bees. Bees remain a
favoured subject for behavioural experiments because they are quite easy
to obtain and to train in laboratory conditions. It is also possible to make
electrophysiological recordings from single neurons in a bee’s brain during
the performance of some simple behaviour patterns in order to reveal cir-
cuits that are important in learned behaviours.
One of the simplest behaviours that a honey bee (Apis mellifera) pro-
duces is to extend its proboscis, or tongue, in response to a drop of sucrose
solution applied to chemosensory hairs on the proboscis or antenna (Fig.
9.3a). A bee will occasionally extend its proboscis without any obvious
stimulation, or if a tiny puff of a particular odour such as the smell of car-
nation or orange is blown at an antenna. If the bee has recently tasted
sucrose, the likelihood that it will extend its proboscis in response to a sub-
sequent odour puff increases – stimulation with sucrose is said to sensitise
the proboscis-extension response. A much greater enhancement of the
response to a puff of odour occurs, however, after the odour puff has been
paired with delivery of a drop of sucrose. For maximum effect, the odour
puff must be delivered between 1 and 3 s before the drop of sucrose
(Bitterman et al., 1983). The next time that the odour is directed at an
antenna, there is a very high chance that the bee will extend its proboscis
(Fig. 9.3b). In the bee’s brain, an association has been made between the
odour and the likely presence of sucrose. The order in which the two
Associative learning and the proboscis extension reflex in honey bees 229
Figure 9.3 Conditioning of the proboscis extension reflex in honey bees

(Apis) to odours. (a) A diagram to illustrate the conditioning procedure. A
puff of air laden with a chosen odour is blown across an antenna, and
sucrose delivered to the proboscis when it is extended acts as the reward.
(b) A graph to show the acquisition of the conditioned response. Groups of
bees were trained by pairing a puff of odour with a sucrose reward 2 s later.
After this pairing procedure, about 80 per cent of bees tested responded to
the odour with proboscis extension. In comparison, few untrained bees
responded to the odour. (b redrawn after Menzel, 1990.)
different stimuli are delivered is vital for the formation of memory. If the
sucrose is applied to the proboscis just before the odour is blown over an
antenna, a second puff of the odour delivered a minute later is unlikely to
cause proboscis extension.
The proboscis-extension reflex is a good example of conditioning, a type
of behavioural change often studied in vertebrates. The odour stimulus
becomes conditioned so that after it has been paired with a sucrose
reward, it reliably evokes a stimulus that was previously only rarely linked
with this stimulus. A single pairing of a puff of carnation with a taste of
sucrose is sufficient to enhance the new coupling between carnation and
proboscis extension for many hours. However, the coupling will weaken
and become extinguished if several odour puffs are delivered with no
sucrose as a reward, and can be replaced by a new association between
another odour, such as orange, and proboscis extension. The memory of
the association between an odour and sucrose goes through different
phases in time, so that the initial short-term memory is transferred into
longer-lasting medium-term and long-term memories (Hammer &
Menzel, 1995).
9.6 Neuronal pathways and conditioning
Taste-sensitive sensory neurons that detect sucrose project from the pro-
boscis into the suboesophageal ganglion, which is the first ganglion in the
nerve cord after the brain and is also the ganglion that contains the motor
neurons of the proboscis muscles. Odours are detected by sensilla on the
antennae, which project into antennal lobes of the brain. After processing
in the brain, information about odours is carried to the suboesophageal
ganglion by neurons that originate in the forebrain, or protocerebrum.
During conditioning, information about the sucrose reward must be
carried to the brain from the suboesophageal ganglion and cause
modification of the pathways that link olfactory sensory receptors with the
protocerebral neurons. The pathways involved are shown in Fig. 9.4.
About 60 000 olfactory receptors run from each antenna to a discrete area
on either side of the brain called the olfactory lobe. Each sensory neuron ter-
minates in a spherical structure called a glomerulus, and olfactory informa-
tion is processed both within and between the 160 glomeruli in an olfactory
lobe before being distributed by projection neurons to other brain regions.
Some neurons project directly from glomeruli to the protocerebrum, and
Neuronal pathways and conditioning 231
Figure 9.4 The morphology of the brain and suboesophageal ganglion of

the bee to show the structures and pathways involved in conditioning the
proboscis-extension reflex to odours. (a) Information about odours is
processed in the olfactory lobes, which send outputs both to the proto-
cerebrum and to the calyces of the mushroom bodies, where Kenyon cells
have their dendrites. Kenyon cell axons have branches to two separate
output lobes of a mushroom body, from where interneurons travel to
various brain regions, including the protocerebrum. Some interneurons
travel from the protocerebrum to the suboesophageal and the segmental
ganglia. Motor neurons that control proboscis muscles are situated in the
suboesophageal ganglion, which is also where sucrose-detecting chemo-
sensory neurons terminate. (b) A diagram to show the extent of the innerva-
tion pattern of neuron VUMmx1, whose cell body and dendrites are in the
suboesophageal ganglion. (a and b after Hammer, 1993; reprinted with
permission from Nature; copyright © 1993 Macmillan Magazines Ltd.)
others project to another brain region, called the mushroom body. The
mushroom bodies were named after their resemblance to some kinds of
horn-shaped mushrooms, and have been implicated in the more complex
types of insect behaviour, including learning and social behaviour. They are
particularly well developed in bees, and also in some other types of insects,
including cockroaches, where they may play roles in spatial memory
(Mizunami, Weibrecht & Strausfeld, 1993). Each mushroom body contains a
parallel array of neurons called Kenyon cells (one is drawn in Fig. 9.4a).
About a third of all the neurons in the brain of a bee are Kenyon cells, with
170 000 in each mushroom body. It is extremely difficult to make recordings
from them because they are small and tightly packed together. Their den-
drites are arranged in structures called calyces, and olfactory information
arrives at the outermost rim region of each calyx. Kenyon cell axons branch
into two, one branch going into the ␣ lobe of the mushroom body and the
other branch into the ␤ lobe. Various output neurons combine information
from many Kenyon cells and deliver information from the mushroom
bodies to other parts of the brain. Response properties of these output cells
can alter during conditioning (Mauelshagen, 1993).
Two types of experiment suggest that both the antennal lobe and the
mushroom body play roles in associative conditioning. In one, a small
probe was used to cool local regions of the brain to 1–5 °C for 10 s, transi-
ently blocking or reducing activity (Erber, Masuhr & Menzel, 1980). Cooling
either an olfactory lobe or a mushroom body reduced conditioning, and the
effect was not simply to reduce sensory activity because conditioning was
blocked if several seconds or even minutes elapsed between delivery of the
odour stimulus and the start of cooling. In the second type of experiment,
a small amount of the neuromodulator octopamine was injected into a
brain region just after delivery of an odour pulse. The octopamine substi-
tuted for the effects of stimulation of taste receptors with sucrose, and
when the odour was presented later on its own, it evoked proboscis exten-
sion. Octopamine had this effect when injected into the olfactory lobe or
calyx of a mushroom body, but not elsewhere.
9.7 The role of an identified neuron in conditioning
A bee’s brain must be able to form associations between a sucrose reward

and many different other stimuli, including a variety of odours. We could,
The role of an identified neuron in conditioning 233
therefore, expect that a neuron that plays a key role in conditioning would
have branches distributed in many different regions of the brain. One such
neuron is VUMmx1 (Fig. 9.4b), an unpaired neuron which has branches
that are arranged symmetrically on both the left and right sides of the sub-
oesophageal ganglion and brain, and Martin Hammer (1993) demonstrated
experimentally that this neuron can induce conditioning. Its branches
overlap the pathways that link olfactory stimuli to the extension reflex at
several places. Like many other unpaired, median neurons in insects, such
as DUM neurons in the thoracic ganglia (see Box 7.2, p. 187), VUMmx1
appears to contain and release octopamine.
Sucrose excites VUMmx1, and this neuron continues to spike for at least
half a minute after the sucrose has been removed. VUMmx1 is also excited,
but only for a brief time, by various odours directed at the antennae.
However, Hammer found that when sucrose was delivered just after a par-
ticular odour, the next time the odour was delivered alone VUMmx1
responded with a brief, intense burst of spikes followed by excitation that
continued for 20 s, a similar response to that given by VUMmx1 to sucrose.
Like proboscis extension, this response enhancement depended strictly on
the two stimuli being delivered within a short time, with the start of the
odour preceding the start of sucrose delivery. The neuronal mechanism
that requires a particular stimulus order is unknown.
Exciting VUMmx1 by injecting current into it through a microelectrode
can substitute for stimulation with sucrose in conditioning (Fig. 9.5). In
forward pairing, the odour stimulus started 2 s before the electrical excita-
tion of VUMmx1, and during backward pairing, VUMmx1 excitation pre-
ceded the odour by 5 s. Activation of the proboscis-extension muscle was
monitored in an electromyogram recording. After forward pairing, a test
pulse of odour 10 min later produced a large electromyogram response
(Fig. 9.5b). In contrast, backward pairing of VUMmx1 excitation with odour
did not lead to any enhancement of the response to the test odour. This
experiment clearly shows that VUMmx1 can condition the proboscis-
extension response to a particular odour. However, it does not show that
VUMmx1 is solely responsible for conditioning, and it probably normally
works in parallel with other similar ventral, unpaired median neurons in
the suboesophageal ganglion.
VUMmx1 can deliver information about a sucrose reward to the olfactory
lobes and the mushroom body, the sites that the experiments with local
Figure 9.5 Electrical excitation can substitute for stimulation with sucrose in
conditioning the proboscis-extension response to odours. (a) An outline of
the protocol used in training experiments. First, any response by proboscis-
extension muscles to a puff of odour was measured in a pre-test, Next, a puff
of the odour was paired with electrical excitation of VUMmx1 by a pulse of
depolarising current. In forward pairing, the odour preceded the start of the
current pulse, and in backward pairing the current pulse started before the
odour. Finally, responses to a test pulse of odour were recorded. (b)
Responses to a test puff of odour after training with either electrical excita-
tion of VUMmx1 or a sucrose reward. The response was measured as the
number of spikes in a proboscis extensor motor neuron during 10 s after a
test or pre-test puff of odour, and the histograms plot the median response
from groups of 11–15 bees. The error bars indicate the maximum responses
from each group. (a and b redrawn after Hammer, 1993.)
Bird song and its production 235
cooling and octopamine injection showed are important in conditioning. A

key event in conditioning is that excitation of VUMmx1 is closely associated
in time with stimulation of particular odour-activated pathways. A great
deal of information about the molecular mechanisms for coincidence
detection in neurons has been gained in studies on the gill-withdrawal
reflex of the mollusc Aplysia (Box 9.1). The mechanism by which VUMmx1
establishes conditioning is likely to involve intracellular messenger mole-
cules such as cyclic AMP and cyclic GMP and, in Drosophila, mutations that
affect production of these molecules also cause impairment of memory
(Belvin & Yin, 1997). Conditioning of proboscis extension in bees is quite
complex because a large range of possible stimuli can become associated
with the response, and it is probable that the complexity of the mushroom
bodies is in some way related with the need to be able to deal with a large
number of possible stimulus configurations.
9.8 Bird song and its production
Almost all birds produce calls of various kinds but one group of passerines,
4000 species called song birds, produce extended and complex songs. A
defining characteristic of these songs is that their full expression requires
learning by a juvenile bird from mature, singing adults that act as tutors.
Song birds are believed to have evolved from one common ancestor, and
include finches, warblers and thrushes. The most complex songs are
usually produced by males during the breeding season, and the functions
of song include establishing territories, attracting potential mates, and
maintaining pair bonds (Catchpole & Slater, 1995). Because birds incorpo-
rate parts of the songs of other birds into their own songs, local dialects
develop in some species (Marler & Tamura, 1964; Baker & Cunningham,
1985).
For a neuroethologist, bird song provides some general lessons about the
way a complex type of behaviour is controlled by a central nervous system.
Within the brain, a number of discrete areas, or nuclei, have been shown to
be involved in song. Each nucleus contains cell bodies and dendrites of a
number of types of neuron. Some of the neurons have axons that run in
tracts to other nuclei, and other neurons participate in processing informa-
tion within their nucleus. One group of nuclei is responsible for generating
the song in the adult, and another is involved in laying down motor
Box 9.1. Plasticity in the gill-withdrawal reflex of Aplysia
Aplysia withdraws its delicate gill if skin near to the siphon is touched
(see section 8.2). Properties of the synapse that links a sensory and
motor neuron in the abdominal ganglion (*) change as a result of
different kinds of sensory experience, and contribute to simple plastic
changes in behaviour. If the synapse is activated repeatedly, the
strength of transmission declines and this is one mechanism underly-
ing habituation of the gill-withdrawal reflex. A noxious stimulus
applied elsewhere to the body, such as to the tail, sensitises the reflex
so that the next siphon stimulus will evoke a larger gill-withdrawal
response than previously. Sensitisation is due to the action of facilitat-
ing interneurons which release the transmitter serotonin onto the
presynaptic terminals of the siphon sensory neurons. This causes
presynaptic facilitation, an increase in the amount of neurotransmit-
ter released (Byrne & Kandel, 1996). The circuit can also be condi-
tioned by following a touch to the siphon immediately by a noxious
stimulus to the tail. After this pairing, the next touch to the siphon
causes a greatly enhanced postsynaptic potential in the gill-with-
drawal motor neuron. At least two coincidence-detecting mecha-
nisms operate to enhance synaptic activity. The first is activated when
serotonin stimulates the sensory terminal of the sensory neuron while
the sensory neuron is electrically excited, and this increases the man-
ufacture of cyclic AMP in the sensory terminal (Hawkins, Kandel &
Siegelbaum, 1993). The second is similar to a phenomenon known as
post-tetanic potentiation, familiar in some parts of the vertebrate
central nervous system. It occurs if the sensory neuron is stimulated
at a time when the motor neuron is already excited, and involves a
postsynaptic receptor protein called an NMDA receptor (Glanzman,
1995).
Bird song and its production 237
programs for singing. The subject of many studies is the zebra finch
(Taeniopygia guttata), an Australian species that breeds readily in captivity
and develops to maturity in only 100 days. An individual male produces
song that is more easy to characterise than the songs of many other birds
because it is relatively stereotyped.
A lot of information about a song can be expressed in a sonograph, in
which records of sounds are broken down to show the relative contribu-
tions of different frequencies. Sonographs are particularly useful for com-
paring the songs of different individuals, or of one bird at different stages in
its development. The sonograph in Fig. 9.6a illustrates different levels of
organisation within the song of a zebra finch. Series of notes are linked
together into discrete syllables, and a series of syllables is linked together in
a unit called a motif. When birds are interrupted during singing, they
always finish a syllable (Cynx, 1990), which means that a syllable is a basic
unit in the organisation of song. In zebra finches, the motif that a particu-
lar male sings is fixed in form, although the number of motifs in a bout of
singing varies from song to song, as do the brief introductory notes that
precede the song and separate successive motifs. Each note of the zebra
finch song is composed of sounds of many different frequencies. Notes of
most other species contain more restricted tones and, as a result, have a
more musical quality to a human listener.
The organ responsible for producing sounds during song is the syrinx,
located where the trachea joins the bronchi of the two lungs (Fig. 9.6b). Four
to six muscles on either side are attached to the syrinx, and sound is pro-
duced when air is expelled through it. The exact way in which sound is pro-
duced has not been conclusively demonstrated, but it may involve the
openings of the bronchi into the trachea forming whistles, or the two medial
tympaniform membranes vibrating like drum membranes. Some syrinx
muscles determine the tone of sound which is produced and others control
the timing of sounds by opening and closing the bronchi. Respiratory
muscles generate the force for expelling air through the syrinx, and so
control the volume of sound. The lungs of birds are rigid and air moves
through them in one direction by the action of large air sacs that act as
bellows. Each syllable of a song is produced by contraction of muscles that
expel air from the interclavicular air sac. Electromyograms and pressure
measurements have shown that, in canaries, each syllable is co-ordinated
with a cycle of inspiration and expiration, even at rates in excess of 20/s. It is
Figure 9.6 Bird song and its production. (a) A sonograph of the song of a
zebra finch (Taeniopygia guttata). Shown here are a few introductory notes,
followed by three repetitions of the same motif. This motif contained six syl-
lables, each one of which was a particular sequence of notes. (b) The main
structures associated with the syrinx of a song bird. (a sonograph kindly
supplied by Dr D. Margoliash; b from Suthers, 1990; reprinted with permis-
sion from Nature; copyright © 1990 Macmillan Magazines Ltd.)
The development of song 239
usual for the muscles on the left and right of the syrinx to act independently
of each other, and zebra finches sing mostly by using muscles on the right.
9.9 The development of song
The development of song in most species follows three distinct phases. In

the first, the sensory phase, the bird hears songs produced by tutor birds
around him. In temperate species, this usually occurs soon after hatching
in the spring or summer, almost a year before the bird begins to sing
himself. The requirement for tutor songs was first shown experimentally by
Thorpe (1958), who found that chaffinches (Fringilla coelebs) that had been
reared in the laboratory without hearing the songs of adults produced very
abnormal songs when they matured. The syllables of their songs were
simple and disorganised. However, chaffinches that had heard tape record-
ings of adult chaffinch song as fledglings produced normal songs when
they matured. Young chaffinches did not learn from other sounds, includ-
ing the songs of most other bird species, so a mechanism for recognising
the song of members of its own species must be built into the brain of
the young bird. The timing of exposure to tutor song is critical. A young
chaffinch must hear adult songs before the middle of its first summer if it is
to sing normally the next year.
The second period of song development, the sensorimotor phase, begins
early in the spring of a chaffinch’s second year. The bird starts to sing spon-
taneously, at first producing quiet and variable songs. Gradually, song
becomes louder and, because it is variable, it is called plastic. Plastic song
is often a mixture of two different types of song, and includes repeated syl-
lables. A bird must be able to hear its own plastic song for normal develop-
ment (Konishi, 1965a, 1965b). In the final period, in late spring, song
crystallises to its mature form, probably triggered by a rise in the level of the
steroid hormone testosterone in the blood. In most species, deafening at
this stage does not cause abnormalities in song.
This sequence of three phases is found in most species of song bird, but
there are many variations in detail. Some species, including starlings and
canaries, are called open-ended learners because they go through the pro-
cesses of song practising and crystallisation every season. Zebra finches
follow the general pattern, except that young males can still hear the songs
of tutors while they start to produce plastic song, which is associated with
their rapid development and social lifestyle. Their sensitive period for song
learning starts at 20 days old, when they fledge, and lasts until they are 40
days old. Young zebra finches begin to sing 25 days after hatching and song
crystallises 90–110 days after hatching, at sexual maturity.
9.10 Neural centres for hearing and singing
Three different groups of nuclei are known to play roles in the production
of song. One group of brain nuclei is involved in processing auditory infor-
mation, a second in producing the motor program for song (Fig. 9.7a), and
a third in learning and developing the song pattern (Fig. 9.7b). One nucleus,
HVc, is a member of all three groups. Sometimes, HVc is referred to as the
‘higher vocal centre’, but attaching functional names to brain regions can
be misleading, and this and other nuclei are referred to here by accepted
abbreviations. In an adult male zebra finch, HVc contains about 35 000
neurons. There is a good correlation between the size of HVc and the com-
plexity of song in different species.
HVc exerts direct control over the production of song through connec-
tions it makes with nucleus RA. From RA, axons project to groups of motor
neurons that innervate the syrinx and some respiratory muscles. Birds in
which HVc or RA has been destroyed cannot sing, although they still court
females by adopting the same postures as during singing (Nottebohm et al.,
1976), and they produce alarm cries and some other calls. Anthony Yu and
Daniel Margoliash (1996) implanted fine wire electrodes into the brains of
zebra finches and recorded spikes from single neurons while the birds were
moving around their cages and singing normally. Many of the neurons in
HVc started to spike before a bird began to sing, and remained excited until
just before the end of the song. During a song, spike rate increased and
decreased in a characteristic manner whenever the bird sang a particular
syllable. A particular neuron was most excited when a particular syllable
was sung, and each syllable would be associated with excitation of a unique
population of HVc neurons. In contrast, individual neurons in RA produced
discrete bursts of spikes coinciding with a specific series of notes that
occurred in a number of different syllables throughout the song. In other
experiments, Eric Vu and colleagues (Vu, Mazurek & Kuo, 1994) used similar
electrodes to stimulate neurons in singing birds with a short train of pulses
that lasted less than the duration of a syllable. Stimuli to small regions of
Neural centres for hearing and singing 241
Figure 9.7 Diagrams to show the principal brain nuclei involved in the
generation and development of bird song. (a) The control of singing in a
mature adult. Motor neurons of syringeal muscles are activated by neurons
from RA which, in turn, is activated by neurons from HVc. HVc receives
inputs from Field L, which is a major sensory area involved in auditory
processing, and from Uva and NIf. (b) Nuclei involved in the development
of song. Notice that nucleus lMAN receives feedback from HVc via Area X
HVc disrupted the syllable the bird was singing and interfered with the
order of subsequent syllables within a motif. For example, if a bird’s motif
consisted of syllables A B C D, a stimulus delivered just after syllable B
might alter the motif’s structure to A B G, combining syllables C and D to an
unusual, abbreviated form, G. The electrical stimuli neither stopped the
song nor interfered with the pattern of syllables during following motifs.
Stimuli delivered to RA had a more restricted effect, disrupting the current
syllable without affecting later syllables within the motif. A stimulus deliv-
ered to RA just after B in the sequence A B C D would cause the bird to sing
A B C’ D, where C’ is an altered form of syllable C.
These two kinds of experiment suggest that individual neurons in HVc
control particular sequences of syllables within a motif by recruiting groups
of neurons in RA in a specific order. The RA neurons each excite a pool of
motor neurons that control muscles of the syrinx to produce a particular
combination of notes, and this correlates well with the anatomy of RA,
which is arranged topographically so that a particular region in it corre-
sponds with a particular group of motor neurons and muscles (Vicario,
1991). HVc is not organised topographically.
HVc has sensory as well as motor properties. In an anaesthetised bird,
many HVc neurons respond well to recordings of the bird’s own song, but
not to the songs of other birds. The neurons recognise both the order of
notes within a syllable and the order of syllables within a motif. Specificity
for the order of syllables in a motif is particularly interesting because it
means the HVc neurons need to collect information over time, for example
to distinguish order A B C D from A C B D. It is not clear whether these audi-
tory responses play a role in the control of singing in the adult because they
are probably suppressed when the bird sings (McCasland & Konishi, 1981),
but they are important for the normal development of song.
9.11 Development of song nuclei
Nuclei lMAN, DLM and Area X are required for a bird to acquire the normal
adult song pattern (Bottjer, Meisner & Arnold, 1984), but not for an adult
bird to sing. If lMAN is removed from a fledgling bird, the bird will produce
an incomplete pattern of song, and if Area X is removed, song remains
plastic. In the adult, neurons in Area X and lMAN, like those in HVc,
respond specifically to tape recordings of the bird’s own song. This selectiv-
Development of song nuclei 243
ity arises gradually. Neurons in lMAN and Area X of young birds are excited
by many kinds of sounds, including songs played in reverse, and do not
prefer particular songs (Doupe, 1997; Solis & Doupe, 1997). In 60 day-old
zebra finches, some of the neurons prefer the bird’s own song, whereas
others respond best to tutor songs. When song crystallises, all the neurons
prefer the bird’s own song. The gradual nature with which these auditory
responses develop supports a mechanism of instruction during song devel-
opment. The less likely alternative is that during the sensorimotor phase,
particular song-producing circuits are selected from an array of alternative
circuits that are already fully formed in the young bird. If development
occurred by selection, it would probably be abrupt rather than gradual.
The anatomical organisation of the song nuclei changes considerably
during development. HVc is first recognisable as a distinct nucleus 10–15
days after hatching, and by day 50, half way through sensorimotor learning,
the number of neurons it contains increases by 50 per cent (Kirn &
DeVoogd, 1989). The neurons that are added to HVc are new cells that orig-
inate from a proliferative layer in the ventricle above HVc. In another bird,
the canary, new neurons are added to HVc every year (Box 9.2). During sen-
sorimotor learning, RA also increases in size, but lMAN loses about half its
neurons. To begin with, 37 per cent of the synapses in RA are from lMAN
neurons, but this drops to 4 per cent by the time song crystallises
(Herrmann & Arnold, 1991).
Box 9.2. Seasonal birth of new neurons in the canary brain
In all animals, most nerve cells are present at birth, and it is well
known that in adult humans brain cells that are lost through injury are
not replaced. It was surprising, therefore, when Goldman and
Nottebohm (Nottebohm, 1989) demonstrated that, in canaries, new
neurons are added to HVc. These neurons are born in the brain cavity,
or ventricle, and migrate along glial cells to reach HVc. Some of these
new neurons grow axons that extend several millimetres to RA. The
rate at which new neurons are added to HVc is highest in the autumn,
which is when adult male canaries learn new songs each year.
Although there seems to be an association between the development
of new songs and the annual birth of new neurons, female canaries
also add some new neurons to their HVc, as do adult zebra finches,
and it has now been shown that a few new neurons are born in the
brains of adult mammals.
The essence of the development of song is that a motor program is

modified by comparing the sound of the bird’s own song with the memory
of aspects of tutors’ songs. One potential site for comparisons is Area X,
where some signals arrive directly from HVc and others are fed through a
loop that takes them from Area X and returns them there after travelling
through DLM and lMAN (Fig. 9.7b). Another potential site for comparisons
is RA, which receives information from HVc both directly and through
lMAN (Fig. 9.7b). Clearly, lMAN plays a key role in the developing song-
control circuitry. Individual neurons from lMAN send axonal branches to
both Area X and RA (Vates & Nottebohm, 1995), and lMAN is the nucleus
that shows the greatest loss of neurons during song development. Perhaps
as the bird improves his song, particular neurons in lMAN may be selected
to survive and maintain their contacts, whereas other lMAN neurons die.
The feedback loops evident in Fig. 9.7b may function to delay copies of
the motor commands issued by HVc, so that they coincide in time with
sensory responses to the sound of the bird’s song (Margoliash, 1997). A
motor command issued by HVc precedes the sound of a syllable by at least
50 ms; and the response to the sound of the syllable does not arrive in HVc
until at least 20 ms after the sound. The brain could compare the motor and
sensory signals if the two signals were brought together at particular nuclei,
with a longer route for the copy of the motor command issued by HVc and
a shorter route for the auditory signal. In response to particular syllables of
its own song, many HVc neurons in an individual bird spike intensely and
synchronously, which would be a powerful trigger for change. When two
signals coincide in time, they might trigger changes in the excitability of
neurons, or in the strengths of particular synapses, as occurs during other
forms of learning. It has been found that the act of singing causes the acti-
vation of a particular set of genes in some brain nuclei (Jarvis & Nottebohm,
1997), so it is likely that morphological changes, perhaps involving synap-
ses, follow particular patterns of electrical activity.
9.12 Conclusions
Nervous systems are programmed to ensure that behaviour changes in

response to particular events during the normal life of an animal. Some
changes are part of the developmental program of an animal, and others
allow it to adapt its behaviour when its external environment alters. A good
Conclusions 245
example of how a series of events is orchestrated at a particular time during

development is provided by ecdysis in moths. The larva becomes fully com-
mitted to ecdysis by the positive feedback loop in which two different poly-
peptide hormones, ecdysis-triggering hormone and eclosion hormone,
reinforce the release of each other. Pre-ecdysis is the first motor activity to
be triggered because its pattern generator is very sensitive to ecdysis-trig-
gering hormone and is switched on as soon as this hormone starts to be
released. Eclosion hormone switches off the pre-ecdysis behaviour and, at
the same time, switches on ecdysis. It mediates its effects by exciting cells
that release crustacean cardioactive peptide, and these remain excited for
some time after their initial stimulation by eclosion hormone, until ecdysis
is complete.
Polypeptide hormones exert their effects on neurons by binding with
protein receptors on the cell surface and triggering intracellular messenger
pathways. Steroid hormones, on the other hand, operate directly in the cell
nucleus, regulating gene expression. Before each moult in a moth, the level
of the steroid hormone ecdysone rises. If this rise occurs when the titre of
juvenile hormone is low, it can trigger widespread effects that play an
important role in the reorganisation of the nervous system, sense organs
and muscles that occurs during metamorphosis. The rise in ecdysone
causes some neurons and muscles to die, and others, such as MN1, to grow
new dendrites which enable them to participate in new circuits with other
neurons.
Less persistent changes occur during associative learning, for example
when a honey bee is conditioned to associate a particular odour with a
sucrose reward. The trigger for the reinforcement of specific neuronal cir-
cuits is a particular pattern of activity, which happens when sensory signals
about a specific stimulus, such as an odour, just precede signals about the
sucrose reward. An identified neuron, VUMmx1, has been shown to carry
information about the reward and to be able to reinforce the association
between a specific odour and proboscis extension. This neuron branches to
many brain regions and could reinforce a large number of different sensory
stimuli, which is important so that the bee can make use of many types of
sensory cues to increase its foraging efficiency. During conditioning, two
areas where the action of VUMmx1 is vital are the antennal lobes and the
mushroom bodies. The large numbers of neurons in the mushroom bodies
probably enable the bee to make novel association between many different
sensory stimuli and particular actions. The wide range of possible associa-
tions that animals can make during learning provides a major challenge to
neurobiologists.
The process of development of a male song bird’s brain includes a
program which ensures that it stores memories of the songs of adult tutors
and uses the memories to mould its own, individual song pattern. There is
no obvious immediate pressure for a young bird to learn songs, but learn-
ing is vital because young males that are prevented from using the songs of
mature, singing adults as models for their own songs are unable to attract
mates. Birds do not copy the songs of tutors exactly, and an intriguing
problem is provided by the origin of the characters that distinguish the song
of one individual from another.
A specific set of brain nuclei is involved in the development of song,
although these nuclei play no known role in the control of mature song in
the adult. During development, new neurons are added to some of these
nuclei, and many neurons die in others. A number of loops and feedback
pathways involving the nuclei probably enable comparisons between the
sound of the bird’s own song and the remembered songs of tutors.
Increasing proficiency at singing is accompanied by an increase in the
auditory responses of neurons in song nuclei until, in the adult,
the neurons prefer recordings of the bird’s own song to any other sound.
The gradual change in the specificity of the auditory responses supports the
idea that the development of the program for song occurs by instruction
rather than by selection. Despite its complexity, bird song has become one
of the most active areas of research in neuroethology. This is partly because
it is intrinsically interesting, but also because the brains of animals that
specialise in particular behaviours, such as bird song, almost invariably
prove to be good sources of information about how the nerve cells dedi-
cated to that behaviour work.
Further reading
Dudai, Y. (1989). The Neurobiology of Memory. Concepts, Findings, Trends. Oxford:
Oxford University Press. The book outlines clearly, and often in a thought-pro-
voking way, many different types of learning in a variety of animals.
Ewer, J. & Truman, J.W. (1996). Increases in cyclic 3⬘5⬘-guanosine monophosphate
(cGMP) occur at ecdysis in an evolutionarily conserved crustacean cardioactive
peptide-immunoreactive insect neuronal network. J Comp Neurol 370, 330–41.
Further reading 247
This paper is about plasticity at an evolutionary level; it describes differences

among orders of insects in some of the neurons that control ecdysis.
Heisenberg, M. (1995). Pattern recognition in insects. Curr Opin Neurobiol 5,
475–81. A review of how insects learn visual patterns.
Jacobs, L.F. (1996). The economy of winter – phenotypic plasticity in behavior and
brain structure. Biol Bull 191, 91–100. This review outlines changes that occur
in the hippocampus of the brains of birds and mammals when food resources
become scarce in winter. The hippocampus shrinks every year in mammals
that hibernate, but grows in birds that store food.
Wehner, R. Lehrer, M. & Harvey, W.R. eds. (1996). Navigation. J Exp Biol 199. This
volume includes reviews about how insects and other animals use memorised
features, such as landmarks, to find routes.
REFERENCES
Aidley, D.J. (1998). The Physiology of Excitable Cells, 4th edn. Cambridge:
Cambridge University Press.
Baader, A. (1991). Auditory interneurons in locusts produce directional head and
abdomen movements. J Comp Physiol A 169, 87–100.
Bacon, J. & Möhl, B. (1983). The tritocerebral commissure giant (TCG) wind-sensi-
tive interneurone in the locust. 1. Its activity in straight flight. J Comp Physiol
150, 439–52.
Baerends, G.P. (1985). Do dummy experiments with sticklebacks support the IRM
concept? Behaviour 93, 258–77.
Baerends, G.P. & Drent, R.H. (1982). The herring gull and its egg. Part II. The
responsiveness to egg-features. Behaviour 82, 1–416.
Baker, M.C. & Cunningham, M.A. (1985). The biology of bird-song dialects. Behav
Brain Sci 8, 85–133.
Bartelmez, G.W. (1915). Mauthner’s cell and the nucleus motorius tegmenti.
J Comp Neurol 25, 87–128.
Basarsky, T.A. & French, A.S. (1991). Intracellular measurements from a rapidly
adapting sensory neuron. J Neurophysiol 65, 49–56.
Belvin, M.P. & Yin, J.C.P. (1997). Drosophila learning and memory: recent progress
and new approaches. BioEssays 19, 1083–9.
Bicker, G. & Pearson, K.G. (1983). Initiation of flight by stimulation of a single
identified wind sensitive neurone (TCG) in the locust. J Exp Biol 104, 289–94.
Bitterman, M.E., Menzel, R., Fietz, A. & Schäfer, S. (1983). Classical conditioning
of proboscis extension in honeybees (Apis mellifera). J Comp Psychol 97,
107–19.
Bodenhamer, R., Pollak, G.D. & Marsh, D.S. (1979). Coding of fine frequency infor-
mation by echoranging neurons in the inferior colliculus of the Mexican free-
tailed bat. Brain Res 171, 530–5.
Bottjer, S.W., Meisner, A. & Arnold, A. (1984). Fore-brain lesions disrupt develop-
ment but not maintenance of song in passerine birds. Science 224, 901–3.
Breedlove, S.M. (1992). Sexual dimorphism in the vertebrate nervous system.
J Neurosci 12, 4133–42.
Bruns, V. & Schmieszek, E. (1980). Cochlear innervation in the greater horseshoe
bat: demonstration of an acoustic fovea. Hearing Res 3, 27–43.
248
References 249
Burrows, M. (1975). Monosynaptic connexions between wing stretch receptors

and flight motorneurones of the locust. J Exp Biol 62, 189–219.
Burrows, M. (1979). Synaptic potentials affect the release of transmitter from
locust non-spiking interneurons. Science 204, 81–3.
Burrows, M. (1980). The control of sets of motoneurones by local interneurones in
the locust. J Physiol 298, 213–33.
Burrows, M. (1985). The processing of mechanosensory information by spiking
local interneurons in the locust. J Neurophysiol 54, 463–78.
Burrows, M. (1989). Processing of mechanosensory signals in local reflex pathways
of the locust. J Exp Biol 146, 209–27.
Burrows, M. (1992a). Reliability and effectiveness of transmission from exterocep-
tive sensory neurones to spiking local interneurones in the locust. J Neurosci
12, 1477–89.
Burrows, M. (1992b). Local circuits for the control of leg movements in an insect.
Trends Neurosci 15, 226–32.
Burrows, M. (1996). The Neurobiology of an Insect Brain. Oxford: Oxford University
Press.
Burrows, M. & Newland, P.L. (1993). Correlation between the receptive fields of
interneurons, their dendritic morphology, and the central projections of mech-
anosensory neurons. J Comp Neurol 329, 412–26.
Burrows, M. & Pflüger, H.-J. (1995). Activity of locust neuromodulatory neurons is
coupled to specific motor patterns. J Neurophysiol 74, 47–57.
Burrows, M. & Siegler, M.V.S. (1978). Graded synaptic transmission between local
interneurones and motor neurones in the metathoracic ganglion of the locust.
J Physiol 285, 231–55.
Burrows, M. & Siegler, M.V.S. (1985). The organization of receptive fields of spiking
local interneurons in the locust with inputs from hair afferents. J Neurophysiol
53, 1147–57.
Byrne, J.H. (1981). Comparative aspects of neural circuits for inking behavior and
gill withdrawal in Aplysia californica. J Neurophysiol 45, 98–106.
Byrne, J.H. & Kandel, E.R. (1996). Presynaptic facilitation revisited: state and time
dependence. J Neurosci 16, 425–35.
Camhi, J.M. & Tom, W. (1978). The escape system of the cockroach Periplaneta
americana: I. The turning response to wind puffs. J Comp Physiol 128, 193–201.
Canfield, J.G. & Rose, G.J. (1993). Activation of Mauthner neurons during prey
capture. J Comp Physiol A 172, 611–18.
Carew, T.J. & Kandel, E.R. (1977). Inking in Aplysia californica. I. Neural circuit of
an all-or-nothing behavioral response. J Neurophysiol 40, 692–707.
Carr, C.E. & Konishi, M. (1990). A circuit for detection of interaural time
differences in the brain stem of the barn owl. J Neurosci 10, 3227–46.
Catchpole, D.K. & Slater, P.J.B. (1995). Bird Song: Biological Themes and Variations.
Cambridge: Cambridge University Press.
Comer, C.M. & Dowd, J.P. (1993). Multisensory processing for movement: antennal
and cercal mediation of escape turning in the cockroach. In Biological Neural
250 References
Networks in Invertebrate Neuroethology and Robotics, ed. R.D. Beer, R.E.

Ritzmann, & T. McKenna, pp. 89–112. Boston: Academic Press.
Cooke, I.R.C. & Macmillan, D.L. (1985). Further studies of crayfish escape behavi-
our. I. The role of the appendages and the stereotyped nature of non-giant
escape swimming. J Exp Biol 118, 351–65.
Copenhaver, P.F. & Truman, J.W. (1982). The role of eclosion hormone in the larval
ecdyses of Manduca sexta. J Insect Physiol 28, 695–701.
Cynx, J. (1990). Experimental determination of a unit of song production in the
zebra finch (Taeniopygia guttata). J Comp Psychol 104, 3–10.
Dagan, D. & Camhi, J.M. (1979). Responses to wind recorded from the cercal nerve
of the cockroach Periplaneta americana: II. Directional selectivity of the
sensory nerves innervating single columns of filiform hairs. J Comp Physiol A
133, 103–10.
Diamond, J. (1968). The activation and distribution of gaba and -glutamate
receptors on goldfish Mauthner neurons: an analysis of dendritic remote inhi-
bition. J Physiol 194, 669–723.
Doupe, A.J. (1997). Song- and order-selective neurons in the songbird anterior
forebrain and their emergence during vocal development. J Neurosci 17,
1147–67.
Dowling, J.E. (1970). Organization of the vertebrate retina. Invest Ophthalmol 9,
655–80.
Dowling, J.E. (1987). The Retina: an Approachable Part of the Brain. Cambridge,
MA: Belknap Harvard University Press.
Dvorak, D.R., Bishop, L.G. & Eckert, H.E. (1975). On the identification of move-
ment detectors in the fly optic lobe. J Comp Physiol 100, 5–23.
Dyer, F.C. (1996). Spatial memory and navigation by honeybees on the scale of the
foraging range. J Exp Biol 199, 147–54.
Eaton, R.C., DiDomenico, R. & Nissanov, J. (1991). Role of the Mauthner cell in
sensorimotor integration by the brain stem escape network. Brain Behav Evol
37, 272–85.
Eaton, R.C. & Emberley, D.S. (1991). How stimulus direction determines the angle
of the Mauthner initiated response in teleost fish. J Exp Biol 161, 469–87.
Eaton, R.C., Lavender, W.A. & Wieland, C.M. (1981). Identification of Mauthner-ini-
tiated response patterns in goldfish: evidence from simultaneous cinematogra-
phy and electrophysiology. J Comp Physiol 144, 521–31.
Egelhaaf, M. (1985). On the neuronal basis of figure-ground discrimination by rel-
ative motion in the visual system of the fly. I. Behavioural constraints imposed
by the neuronal network and the role of the optomotor system. Biol Cybern 52,
123–40.
Egelhaaf, M. & Borst, A. (1993). Motion computation and visual orientation in flies.
Comp Biochem Physiol 104A, 659–73.
Erber, J., Masuhr, T. & Menzel, R. (1980). Localization of short-term memory in the
bee of the brain, Apis mellifera. Physiol Entomol 5, 343–58.
References 251
Evans, P.D. & Siegler, M.V.S. (1982). Octopamine mediated relaxation of main-
tained and catch tension in locust skeletal muscle. J Physiol 324, 93–112.
Ewert, J.-P. (1980). Neuroetholgy. Berlin: Springer-Verlag.
Ewert, J.-P. (1983). Neuroethological analysis of the innate releasing mechanism
for prey-catching behaviour in toads. In Advances in Vertebrate Neuroethology,
ed. J.-P. Ewert, R.R. Capranica & D.J. Ingle, pp. 413.75. New York: Plenum
Press.
Ewert, J.-P. (1985). Concepts in vertebrate neuroethology. Anim Behav 33, 1–29.
Ewert, J.-P. (1987). Neuroethology of releasing mechanisms: prey-catching in
toads. Behav Brain Sci 10, 337–403.
Falk, C.X., Wu, J.-Y., Cohen, L.B. & Tang, A.C. (1993). Nonuniform expression of
habituation in the activity of distinct classes of neurons in the Aplysia abdomi-
nal ganglion. J Neurosci 13, 4072–81.
Franceschini, N., Riehle, A. & Le Nestour, A. (1989). Directionally selective motion
detection by insect neurons. In Facets of Vision, ed. D.G. Stavenga & R.C.
Hardie, pp. 360–390. Berlin: Springer-Verlag.
Fraser, K. & Heitler, W.J. (1991). Photoinactivation of the crayfish segmental giant
neuron reveals a direct giant-fiber to fast-flexor connection with a chemical
component. J Neurosci 11, 59–71.
Frost, W.N. & Katz, P.S. (1996). Single neuron control over a complex motor
pattern. Proc Natl Acad Sci USA 93, 422–6.
Furshpan, E.J. & Potter, D.D. (1959). Transmission at the giant motor synapses of
the crayfish. J Physiol 145, 289–325.
Gammie, S.C.G. & Truman, J.W. (1997). Neuropeptide hierarchies and the activa-
tion of sequential motor behaviors in the hawkmoth, Manduca sexta.
J Neurosci 17, 4389–97.
Glanzman, D.L. (1995). The cellular basis of classical conditioning in Aplysia cali-
fornica – it’s less simple than you think. Trends Neurosci 18, 30–5.
Gnatzy, W. & Schmidt, K. (1971). Die Feinstruktur der Sinneshaare auf den Cerci
von Gryllus bimaculatus Deg. (Saltatoria, Gryllidae). I. Faden und Keulenhaare.
Z Zellforsch 122, 190–209.
Grillner, S., Deliagina, T., Ekeberg, O. et al. (1995). Neural networks that co-ordi-
nate locomotion and body orientation in lamprey. Trends Neurosci 18, 270–9.
Grinnell, A.D. & Hagiwara, S. (1972). Studies of auditory neurophysiology in non-
echolocating bats, and adaptations for echolocation in one genus, Rousettus.
Z vergl Physiol 76, 82–96.
Grinnell, A.D. & Schnitzler, H.-U. (1977). Directional sensitivity of echolocation in
the horseshoe bat , Rhinolophus ferrumequinum. II. Behavioural directionality
of hearing. J Comp Physiol 116, 63–76.
Habersetzer, J. & Vogler, B. (1983). Discrimination of surface-structured targets by
the echolocating bat Myotis myotis during flight. J Comp Physiol 152, 275–82.
Hammer, M. (1993). An identified neuron mediates the unconditioned stimulus in
associative olfactory learning in honeybees. Nature 366, 59–63.
252 References
Hammer, M. & Menzel, R. (1995). Learning and memory in the honeybee.

J Neurosci 15, 1617–30.
Hardie, R.C. (1986). The photoreceptor array of the dipteran retina. Trends
Neurosci 9, 419–23.
Harrow, I.D., Hue, B., Pelhate, M. & Sattelle, D.B. (1980). Cockroach giant interneu-
rones stained by cobalt-backfilling of dissected axons. J Exp Biol 84, 341–3.
Hausen, K. & Egelhaaf, M. (1989). Neural mechanisms of visual course control in
insects. In Facets of Vision, ed. D.G. Stavenga & R.C. Hardie, pp. 391–424. Berlin:
Springer-Verlag.
Hawkins, R.D., Kandel, E.R. & Siegelbaum, S.A. (1993). Learning to modulate neu-
rotransmitter release: themes and variations in synaptic plasticity. Ann Rev
Neurosci 16, 625–65.
Hedwig, B. & Pearson, K.G. (1984). Patterns of synaptic input to identified flight
motoneurons in the locust. J Comp Physiol A 154, 745–60.
Heitler, W.J. (1990). Electrical synapses. In Neuronal Communications, ed. W.
Winlow, pp. 28–52. Manchester: Manchester University Press.
Heitler, W.J. & Fraser, K. (1993). Thoracic connections between crayfish giant fibres
and motor giant neurones reverse abdominal patterns. J Exp Biol 181, 329–33.
Hennig, M. (1990). Neuronal control of the forewings in two different behaviours
in the cricket, Teleogryllus commodus. J Comp Physiol A. 167, 617–27.
Hensler, K. (1992). Neuronal co-processing of course deviation and head move-
ments in locusts. I. Descending deviation detectors. J Comp Physiol A 171,
257–71.
Hensler, K. & Rowell, C.H.F. (1990). Control of optomotor responses by descending
deviation detector neurones in intact flying locusts. J Exp Biol 149, 191–205.
Herrmann, K. & Arnold, A.P. (1991). The development of afferent projections to the
robust archistiatal nucleus in male zebra finches: a quantitative electron
microscope study. J Neurosci 11, 2063–74.
Horsman, U., Heinzel, H.-G. & Wendler, G. (1983). The phasic influence of self-
generated air current modulations on the locust flight motor. J Comp Physiol
150, 427–38.
Jarvis, E.D. & Nottebohm, F. (1997). Motor-driven gene expression. Proc Natl Acad
Sci USA 94, 4097–102.
Judd, S.P.D. & Collett, T.S. (1998). Multiple stored views and landmark guidance in
ants. Nature 392, 710–14.
Judge, S. J. & Rind, F.C. (1997). The locust DCMD, a movement-detecting neurone
tightly tuned to collision trajectories. J Exp Biol 200, 2209–16.
Kandel, E.R. (1976). Cellular Basis of Behavior. San Francisco: Freeman.
Katz, P.S., Getting, P.A. & Frost, W.N. (1994). Dynamic neuromodulation of synaptic
strength intrinsic to a central pattern generator circuit. Nature 367, 729–31.
Kimmel, C.B. & Eaton, R.C. (1976). Development of the Mauthner cell. In Simpler
Networks and Behavior, ed. J.C. Fentress, pp. 186–202. Sunderland, MA: Sinauer
Associates.
References 253
Kimmerle, B., Warzecha, A.-K. & Egelhaaf, M. (1997). Object detection in the fly
during simulated translatory flight. J Comp Physiol A 181, 247–55.
Kirchner, W.H. & Srinivasan, M.V. (1989). Freely flying honeybees use image
motion to estimate object distance. Naturwissenschaften 76, 281–2.
Kirn, J.R. & DeVoogd, T.J. (1989). Genesis and death of vocal control neurons
during sexual differentiation in the zebra finch. J Neurosci 9, 3176–87.
Knudsen, E.I. (1981). The hearing of the barn owl. Sci Am 245 (6), 83–91.
Knudsen, E.I. (1983). Early auditory experience aligns the auditory map of space in
the optic tectum of the barn owl. Science 222, 939–42.
Knudsen, E.I. (1998). Capacity for plasticity in the adult owl auditory system
expanded by juvenile experience. Science 279, 1531–3.
Knudsen, E.I., Blasdel, G.G. & Konishi, M. (1979). Sound localisation by the barn
owl (Tyto alba) measured with the search coil technique. J Comp Physiol 133,
1–11.
Knudsen, E.I. & Konishi, M. (1979). Mechanisms of sound localisation in the barn
owl (Tyto alba). J Comp Physiol 133, 13–21.
Kolton, L. & Camhi, J.M. (1995). Cartesian representation of stimulus direction:
parallel processing by two sets of giant interneurons in the cockroach. J Comp
Physiol 176, 691–702.
Konishi, M. (1965a). Effects of deafening on song development in American robins
and black-headed grosbeaks. Z Tierpsychol 22, 584–99.
Konishi, M. (1965b). The role of auditory feedback in the control of vocalization in
the white-crowned sparrow. Z Tierpsychol 22, 770–83.
Konishi, M. (1992). The neural algorithm for sound localisation in the owl. The
Harvey Lectures 86, 47–64.
Konishi, M. (1993). Listening with two ears. Sci Am 268 (4), 34–41.
Koppl, C., Gleich, O. & Manley, G.A. (1993). An auditory fovea in the barn owl
cochlea. J Comp Physiol A 171, 695–704.
Krapp, H.G. & Hengstenberg, R. (1996). Estimation of self-motion by optic flow
processing in single visual interneurons. Nature 384, 463–6.
Krasne, F.B. (1969). Excitation and habituation of the crayfish escape reflex: the
depolarising response in lateral giant fibres of the isolated abdomen. J Exp Biol
50, 29–46.
Krasne, F.B. & Teshiba, T.M. (1995). Habituation of an invertebrate escape reflex
due to modulation by higher centers rather than local events. Proc Natl Acad
Sci USA 92, 3362–6.
Krasne, F.B. & Wine, J.J. (1975). Extrinsic modulation of crayfish escape and beha-
viour. J Exp Biol 63, 433–50.
Krasne, F.B. & Wine, J.J. (1977). Control of crayfish escape behavior. In Identified
Neurons and Behavior of Arthropods, ed. G. Hoyle, pp. 275–92. New York:
Plenum Press.
Kravitz, E. (1988). Hormonal control of behavior: amines and the biasing of neuro-
nal output in the lobster. Science 241, 1775–81.
254 References
Kupferman, I. & Weiss, K.R. (1978). The command neuron concept. Brain Behav
Sci 1, 3–39.
Kutsch, W. (1969). Neuromuskulare Aktivität bei verschiedenen Verhaltensweisen
von drei Grillenarten. Z Vergl Physiol 63, 335–78.
Kutsch, W., Schwarz, G., Fischer, H. & Kautz, H. (1993). Wireless transmission of
muscle potentials during free flight of a locust. J Exp Biol 185, 367–73.
Land, M.F. & Collett, T.S. (1974). Chasing behaviour of houseflies (Fannia canicu-
laris). A description and analysis. J Comp Physiol 89, 331–57.
Laughlin, S.B. (1981). Neural principles in the peripheral visual systems of inverte-
brates. In Comparative Physiology and Evolution of Vision in Invertebrates:
Invertebrate Visual Centers and Behavior I, Handbook of Sensory Physiology, ed.
H. Autrum, pp. 133–280. Berlin: Springer-Verlag.
Laughlin, S.B. (1994). Matching coding, circuits, cells and molecules to signals:
general principles in retinal design of the fly’s eye. Prog Retin and Eye Res 13,
165–96.
Laughlin, S.B. & Hardie, R.C. (1978). Common strategies for light adaptation in the
peripheral visual system of fly and dragonfly. J Comp Physiol 128, 319–40.
Laughlin, S.B., Howard, J. & Blakeslee, B. (1987). Synaptic limitation to contrast
coding in the retina of the blowfly Calliphora. Proc R Soc B 231, 437–67.
Laughlin, S.B. & Weckström, M. (1993). Fast and slow photoreceptors – a compara-
tive study of the functional diversity of coding and conductances in the
Diptera. J Comp Physiol A 172, 593–609.
Levi, R. & Camhi, J.M. (1996). Producing directed behaviour: muscle activity pat-
terns of the cockroach escape responses. J Exp Biol 199, 563–8.
Levine, R. B., Morton, D. B. & Restifo, L.L. (1995). Remodelling of the insect
nervous system. Curr Opin Neurobiol 5, 28–35.
Levine, R.B. & Truman, J.W. (1982). Metamorphosis of the insect nervous system:
changes in morphology and synaptic interactions of identified neurones.
Nature 299, 250–2.
Levine, R.B. & Weeks, J.C. (1990). Hormonally mediated changes in simple reflex
circuits during metamorphosis in Manduca sexta. J Neurobiol 21, 1022–36.
Liebenthal, E., Uhlman, O. & Camhi, J.M. (1994). Critical parameters of the spike
trains in a cell assembly: coding of turn direction by the giant interneurons of
the cockroach. J Comp Physiol A 174, 281–96.
Lillywhite, P.G. (1977). Single photon signals and transduction in an insect eye.
J Comp Physiol 122, 189–200.
Lindauer, M. (1967). Recent advances in bee communication and orientation. Ann
Rev Entomol 12, 439–70.
Link, A., Marimuthu, G. & Neuweiler, G. (1986). Movement as a specific stimulus
for prey catching behaviour in rhinolophid and hipposiderid bats. J Comp
Physiol A 159, 403–13.
Lockery, S.R. & Kristan, W.B. (1990a). Distributed processing of sensory informa-
tion in the leech. I . Input–output relations of the local bending reflex.
J Neurosci 10, 1811–15.
References 255
Lockery, S.R. & Kristan, W.B. (1990b). Distributed processing of sensory informa-
tion in the leech. II. Identification of interneurons contributing to the local
bending reflex. J Neurosci 10, 1816–29.
Lockery, S.R., Wittenberg, G., Kristan, W.B. & Cottrell, G.W. (1989). Function of
identified interneurons in the leech elucidated using neural networks trained
by back-propagation. Nature 340, 468–71.
Macdonald, D., ed. (1984). The Encyclopedia of Mammals. London: Allen & Unwin.
Manley, G.A., Koppl, C. & Konishi, M. (1988). A neural map of interaural intensity
differences in the brain stem of the barn owl. J Neurosci 8, 2665–76.
Mann, D.W. & Chapman, K.M. (1975). Component mechanisms of sensitivity and
adaptation in an insect mechanoreceptor. Brain Res 97, 331–6.
Margoliash, D. (1997). Functional organization of forebrain pathways for song pro-
duction and perception. J Neurobiol 33, 671–93.
Marler, P. & Tamura, M. (1964). Song ‘dialects’ in three species of white-crowned
sparrow. Science 146, 1483–6.
Masino, T. & Knudsen, E.I. (1990). Horizontal and vertical components of head
movement are controlled by distinct neural circuits in the barn owl. Nature
345, 434–7.
Mauelshagen, J. (1993). Neural correlates of olfactory learning paradigms in an
identified neuron in the honeybee brain. J Neurophysiol 69, 609–25.
McCasland, J.S. & Konishi, M. (1981). Interaction between auditory and motor
activities in an avian song control nucleus. Proc Natl Acad Sci USA 78, 7815–9.
Menzel, R. (1990). Learning, memory and ‘cognition’ in honeybees. In
Neurobiology of Comparative Cognition, ed. R.P. Kesner & D.S. Olten, pp.
237–92. Hillsdale, N.J.: Erlbaum.
Meyrand, P., Simmers, A.J. & Moulins, M. (1991). Construction of a pattern gener-
ating circuit with neurons of different networks. Nature 351, 60–3.
Meyrand, P., Simmers, A.J. & Moulins, M. (1994). Dynamic construction of a neural
network from multiple pattern generators in the lobster stomatogastric
nervous system. J Neurosci 14, 630–44.
Miles, C.I. & Weeks, J.C. (1991). Developmental attenuation of the pre-ecdysis
motor pattern in the tobacco hornworm, Manduca sexta. J Comp Physiol A 140,
179–90.
Miller, J.P. & Selverston, A.I. (1982). Mechanisms underlying pattern generation in
lobster stomatogastric ganglion as determined by selective inactivation of
identified neurons. IV. Network properties of pyloric system. J Neurophysiol 48,
1416–32.
Mizunami, M., Weibrecht, J.M. & Strausfeld, N.J. (1993). A new role for the insect
mushroom bodies: place memory and motor control. In Biological Neural
Networks in Invertebrate Neuroethology and Robotics, ed. R.D. Beer, R.E.
Ritzmann & T. McKenna, pp. 199–226. Boston: Academic Press
Möhl, B. (1985). The role of proprioception in locust flight control. II. Information
relayed by forewing stretch receptors during flight. J Comp Physiol A. 156,
103–16.
256 References
Möhl, B. (1988). Short-term learning during flight control in Locusta migratoria.

J Comp Physiol A 163, 803–12.
Möhl, B. (1993). The role of proprioception for motor learning in locust flight.
J Comp Physiol A 172, 325–32.
Möhl, B. & Bacon, J. (1983). The tritocerebral commissure giant (TCG) wind-sensi-
tive interneurone in the locust. II. Directional sensitivity and role in flight sta-
bilisation. J Comp Physiol 150, 453–65.
Moiseff, A. & Konishi, M. (1981). Neuronal and behavioural sensitivity to binaural
time differences in the owl. J Neurosci 1, 40–8.
Nagayama, T. (1989). Morphology of a new population of local interneurones in
the locust metathoracic ganglion. J Comp Neurol 283, 189–211.
Neuweiler, G. (1983). Echolocation and adaptivity to ecological constraints. In
Neuroethology and Behavioural Physiology, ed. F. Huber, & H. Markl, pp.
280–302. Berlin: Springer-Verlag
Neuweiler, G., Bruns, V. & Schuller, G. (1980). Ears adapted for the detection of
motion, or how echolocating bats have exploited the capacities of the mam-
malian auditory system. J Acoust Soc Am 68, 741–53.
Neuweiler, G., Metzner, W., Heilmann, U., Rubsamen, R., Eckrich, M. & Costa, H. H.
(1987). Foraging behaviour and echolocation in the rufous horseshoe bat
(Rhinolophus rouxi) of Sri Lanka. Behav Ecol Sociobiol 20, 53–67.
Neuweiler, G., Singh, S. & Sripathi, K. (1984). Audiograms of a South Indian bat
community. J Comp Physiol 154, 133–42.
Newland, P.L. (1991). Morphology and somatotopic organisation of the central
projections of afferents from tactile hairs on the hind leg of the locust. J Comp
Neurol 311, 1–16.
Nicol, D. & Meinertzhagen, I.A. (1982). An analysis of the number and composition
of synaptic populations formed by photoreceptors of the fly. J Comp Neurol
207, 29–44.
Norberg, R.A. (1970). Hunting technique of Tengmalm’s owl, Aegolius funereus (L.).
Ornis Scand 1, 51–64.
Norberg, R.A. (1977). Occurrence and independent evolution of bilateral ear asym-
metry in owls and implications on owl taxonomy. Phil Trans R Soc Lond B 280,
375–408.
Nottebohm, F. (1989). From bird song to neurogenesis. Sci Am 214, 1368–70.
Nottebohm, F., Stokes, T. & Leonard, C. (1976). Central control of song in the
canary. J Comp Neurol 165, 457–86.
Novicki, A. & Weeks, J.C. (1995). A single pair of interneurons controls motor
neuron activity during pre-ecdysis compression behaviour in larval Manduca
sexta. J Comp Physiol A. 176, 45–54.
O’Neill, W.E. & Suga, N. (1982). Encoding of target range and its representation in
the auditory cortex of the moustached bat. J Neurosci 2, 17–31.
O’Shea, M. & Rowell, C.H.F. (1976). The neuronal basis of a sensory analyser, the
acridid movement detector system. II. Response decrement, convergence and
References 257
the nature of the excitatory afferents to the fan-like dendrites of the LGMD.
J Exp Biol 65, 289–308.
Olson, G.C. & Krasne, F.B. (1981). The crayfish lateral giants are command neurons
for escape behavior. Brain Res. 214, 89–100.
Orchard, I., Ramirez, J.-M. & Lange, A.B. (1993). A multifunctional role for octopa-
mine in locust flight. Ann Rev Entomol 38, 227–49.
Payne, R.S. (1971). Acoustic location of prey by barn owls (Tyto alba). J Exp Biol 54,
535–73.
Pearson, K.G. & Ramirez, J.-M. (1990). Influence of input from the forewing stretch
receptors on motoneurones in flying locusts. J Exp Biol 151, 317–40.
Pearson, K.G., Reye, D.N., Parsons, D.W. & Bicker, G. (1985). Flight-initiating inter-
neurons in the locust. J Neurophysiol 53, 910–23.
Pearson, K.G. & Wolf, H. (1987). Comparison of motor patterns in the intact
and deafferented flight motor system of the locust. J Comp Physiol A 160,
259–68.
Pearson, K.G. & Wolf, H. (1988). Connections of hindwing tegulae with flight neu-
rones in the locust, Locusta migratoria. J Exp Biol 135, 381–409.
Plummer, M. & Camhi, J.M. (1981). Discrimination of sensory signals from noise
in the escape system of the cockroach: the role of wind acceleration. J Comp
Physiol 142, 347–57.
Pollack, G.D. (1980). Organizational and encoding features of single neurons in the
inferior colliculus of bats. In Animal Sonar Systems, ed. R.G. Busnel & J.F. Fish,
pp. 549–587. New York: Plenum Press.
Pollak, G.D., Marsh, D., Bodenhamer, R. & Souther, A. (1977). Characteristics of
phasic-on neurons in the inferior colliculus of unanaesthetised bats with
observations relating to mechanisms of echo ranging. J Neurophysiol 40,
926–42.
Pollak, G.D. & Schuller, G. (1981). Tonotopic organization and encoding features of
single units in inferior colliculus of horseshoe bats: functional implications for
prey identification. J Neurophysiol 45, 208–26.
Pringle, J.W.S. (1975). Insect Flight. Oxford Biology Readers No. 52. Oxford: Oxford
University Press.
Prugh, J.I., Kimmel, C.B. & Metcalfe, W.K. (1982). Noninvasive recording of the
Mauthner neurone action potential in larval zebra fish. J Exp Biol 101, 83–92.
Ramirez, J.-M. & Orchard, I. (1990). Octopaminergic modulation of the fore-wing
stretch receptor in the locust Locusta migratoria. J Exp Biol 149, 255–79.
Ramirez, J.-M. & Pearson, K.G. (1991). Octopaminergic modulation of interneu-
rons in the flight system of the locust. J Neurophysiol 66, 1522–37.
Ramon y Cajal, S. (1911). Histologie du Système Nerveux de l’Homme et des
Vertébrés. Madrid: Instituto Ramon y Cajal.
Reichert, H. & Wine, J.J. (1983). Coordination of lateral giant and non-giant escape
systems in crayfish escape behaviour. J Comp Physiol 153, 3–15.
Reichert, H., Wine, J.J. & Hagiwara, G. (1981). Crayfish escape behavior: neuro-
258 References
behavioral analysis of phasic extension reveals dual systems for motor control.
Rind, F.C. (1984). A chemical synapse between two motion detecting neurones in
the locust brain. J Exp Biol 110, 143–67.
Rind, F.C. (1996). Intracellular characterization of neurons in the locust brain sig-
nalling impending collision. J Neurophysiol 75, 986–95.
Rind, F.C. & Bramwell, D.I. (1996). A neural network based on the input organisa-
tion of an identified neuron signalling impending collision. J Neurophysiol 75,
967–85.
Rind, F.C. & Simmons, P.J. (1992). Orthopteran DCMD neuron: a reevaluation of
responses to moving objects. I. Selective responses to approaching objects.
J Neurophysiol 68, 1654–66.
Rind, F.C. & Simmons, P.J. (1998). A local circuit for the computation of object
approach by an identified visual neuron in the locust. J Comp Neurol 395,
405–15.
Riquimaroux, H., Gaioni, S. J. & Suga, N. (1991). Cortical computational maps
control auditory perception. Science 251, 565–8.
Ritzmann, R.E. (1993). The neural organization of cockroach escape and its role in
context-dependent orientation. In Biological Neuronal Networks in Invertebrate
Neuroethology and Robotics., ed. R.D. Beer, R.E. Ritzmann & T. McKenna, pp.
113–37. New York: Academic Press.
Robert, D. (1989). The auditory behaviour of flying locusts. J Exp Biol 147, 279–301.
Roberts, A. (1990). How does a nervous system produce behaviour? A case study in
neurobiology. Sci Prog 74, 31–51.
Roberts, A. & Tunstall, M.J. (1990). Mutual re-excitation with post-inhibitory
rebound: a simulation study of the mechanisms of locomotor rhythm genera-
tion in the spinal cord of Xenopus embryos. Eur J Neurosci 2, 11–23.
Robertson, R.M. & Pearson, K.G. (1982). A preparation for the intracellular analysis
of neuronal activity during flight in the locust. J Comp Physiol 146, 311–20.
Robertson, R.M. & Pearson, K.G. (1983). Interneurons in the flight system of the
locust: distribution, properties and resetting properties. J Comp Neurol 215,
33–50.
Robertson, R.M. & Pearson, K.G. (1985). Neural circuits in the flight system of the
locust. J Neurophysiol 53, 110–28.
Rossel, S. (1979). Regional differences in photoreceptor performance in the eye of
the praying mantis. J Comp Physiol 131, 95–112.
Rossel, S. (1980). Foveal fixation and visual tracking in the praying mantis. J Comp
Physiol 139, 307–31.
Sales, G. & Pye, D. (1974). Ultrasonic Communication by Animals. London:
Chapman and Hall.
Schramek, J.E. (1970). Crayfish swimming: alternating motor output and giant
fiber activity. Science 169, 698–700.
Schuller, G. (1984). Natural ultrasonic echoes from wing beating insects are
References 259
encoded by collicular neurons in the CF-FM bat, Rhinolophus ferrumequinum.

Selverston, A.I. & Miller, J.P. (1980). Mechanisms underlying pattern generation in
lobster stomatogastric ganglion as determined by selective inactivation of
identified neurons. I. Pyloric system. J Neurophysiol 44, 1102–21.
Shepherd, G.M. (1983). Neurobiology. New York: Oxford University Press.
Siegler, M.V.S. (1981). Postural changes alter synaptic interactions between non-
spiking interneurons and motor neurons in the locust. J Neurophysiol 46,
310–23.
Siegler, M.V.S. & Burrows, M. (1984). The morphology of two groups of spiking
local interneurons in the metathoracic ganglion of the locust. J Neurosci 6,
507–13.
Siegler, M.V.S. & Burrows, M. (1986). Receptive fields of motor neurones underly-
ing local tactile reflexes in the locust. J Neurosci 6, 507–13.
Sillar, K., Wedderburn, J.F.S. & Simmers, A.J. (1991). The postembryonic develop-
ment of locomotor rhythmicity in Xenopus laevis tadpoles. Proc R Soc B 246,
147–53.
Simmons, P.J. & Rind, F.C. (1992). Orthopteran DCMD neuron: a reevaluation of
responses to moving objects. II. Critical cues for detecting approaching objects.
J Neurophysiol 68, 1667–82.
Snodgrass, R.E. (1935). Principles of Insect Morphology. New York: McGraw-Hill.
Snyder, A.W., Stavenga, D.G. & Laughlin, S.B. (1977). Spatial information capacity
of compound eyes. J Comp Physiol 116, 183–207.
Solis, M.M. & Doupe, A.J. (1997). Anterior forebrain neurons develop selectively by
an intermediate stage of birdsong learning. J Neurosci 17, 6447–62.
Spinola, S.M. & Chapman, K.M. (1975). Proprioceptive indentation of the campan-
iform sensilla of cockroach legs. J Comp Physiol 96, 257–72.
Srinivasan, M.V. (1992). How bees exploit optic flow: behavioural experiments and
neural models. Phil Trans R. Soc B 337, 253–9.
Srinivasan, M.V., Laughlin, S.B. & Dubs, A. (1982). Predictive coding: a fresh view of
inhibition in the retina. Proc R Soc B 216, 427–59.
Sterling, P. (1998). Retina. In The Synaptic Organization of the Brain, ed. G.M.
Shepherd, pp. 205–53. Oxford: Oxford University Press.
Streichert, L.C. & Weeks, J.C. (1995). Decreased monosynaptic input to an
identified motoneuron is associated with steroid-mediated dendritic regression
during metamorphosis in Manduca sexta. J Neurosci 15, 1484–95.
Suga, N., Neuweiler, G. & Moller, J. (1976). Peripheral auditory tuning for fine fre-
quency analysis by the CF-FM bat, Rhinolophus ferrumequinum. IV. Properties
of peripheral auditory neurons. J Comp Physiol 106, 111–25.
Sullivan, W.E. (1982). Neural representation of target distance in auditory cortex of
the echolocating bat, Myotis lucifugus. J Neurophysiol 48, 1011–32.
Suthers, R.A. (1990). Contributions to birdsong from the left and right sides of the
intact syrinx. Nature 347, 473–7.
260 References
Syed, N.I., Bulloch, A.G.M. & Lukowiak, K. (1990). In vitro reconstruction of the
respiratory central pattern generator of the mollusk Lymnaea. Science 250,
282–5.
Takahashi, T., Moiseff, A. & Konishi, M. (1984). Time and intensity cues are pro-
cessed independently in the auditory system of the owl. J Neurosci 4, 1781–6.
Thorpe, W.H. (1958). The learning of song patterns by birds, with especial refer-
ences to the song of the chaffinch, Fringilla coelebs. Ibis 100, 535–70.
Tinbergen, N. (1951). The Study of Instinct. Oxford: Oxford University Press.
Tinbergen, N. (1963). On aims and methods of ethology. Z Tierpsychol 20, 410–33.
Truman, J.W. & Reiss, S.E. (1988). Hormonal regulation of the shape of identified
motoneurons in the moth Manduca sexta. J Neurosci 8, 765–75.
Vater, M., Feng, A.S. & Betz, M. (1985). An HRP-study of the frequency–place map
of the horseshoe bat cochlea: morphological correlates of the sharp tuning to a
narrow frequency band. J Comp Physiol A 157, 671–86.
Vates, G.E. & Nottebohm, F. (1995). Feedback circuitry within a song-learning
pathway. Proc Natl Acad Sci USA 92, 5139–43.
Vicario, D.S. (1991). Organization of the zebra finch song control system: II.
Functional organization of the output from the nucleus robustus archistiriat-
alis. J Comp Neurol 309, 486–94.
Vu, E.T., Mazurek, M.E. & Kuo, Y.-C. (1994). Identification of a forebrain motor pro-
gramming network for the learned song of zebra finches. J Neurosci 14,
6924–34.
Warzecha, A.-K., Egelhaaf, M. & Borst, A. (1993). Neural circuit tuning fly visual
neurons to motion of small objects. I. Dissection of the circuit by pharmacolog-
ical and photoinactivation techniques. J Neurophysiol 69, 329–39.
Watkins, B.L., Burrows, M. & Siegler, M.V.S. (1985). The structure of locust nonspik-
ing interneurones in relation to the anatomy of their segmental ganglion.
J Comp Neurol 240, 233–55.
Watson, A.H.D. (1984). The dorsal unpaired median neurons of the locust meta-
thoracic ganglion: neuronal structure and diversity, and synapse distribution.
J Neurocytol 13, 303–27.
Watson, A.H.D. & Burrows, M. (1985). The distribution of synapses on the two
fields of neurites of spiking local interneurones in the locust. J Comp Neurol
240, 219–32.
Watson, A.H.D. & Burrows, M. (1988). The distribution and morphology of synap-
ses on nonspiking local interneurones in the thoracic nervous system of the
locust. J Comp Neurol 272, 605–16.
Weeks, J.C. & Levine, R.B. (1995). Steroid effects on neurons subserving behaviour.
Curr Opin Neurobiol 5, 809–15.
Weeks, J.C. & Truman, J.W. (1984). Neural organization of peptide activated ecdysis
behaviors during metamorphosis of Manduca sexta. J Comp Physiol A 155,
407–22.
Wiersma, C.A.G. (1947). Giant nerve fiber system of the crayfish: a contribution to
comparative physiology of the synapse. J Neurophysiol 10, 23–38.
References 261
Wiersma, C.A. G. & Ikeda, K. (1964). Interneurons commanding swimmeret move-

ments in the crayfish, Procambarus clarkii (Girard). Comp Biochem Physiol 12,
509–25.
Willows, A.O.D., Dorsett, D.A. & Hoyle, G. (1973). The neuronal basis of behavior in
Tritonia. III. Neuronal mechanism of a fixed action pattern. J Neurobiol 4,
255–85.
Wilson, D.M. (1960). The central nervous control of flight in a locust. J Exp Biol 38,
471–90.
Wilson, M. (1978). The functional organization of locust ocelli. J Comp Physiol 124,
297–316.
Wilson, M., Garrard, P. & McGiness, S. (1978). The unit structure of the locust com-
pound eye. Cell Tiss Res 195, 205–26.
Wine, J.J. (1977). Crayfish escape behaviour II. Command derived inhibition of
abdominal extension. J Comp Physiol 121, 173–86.
Wine, J.J. (1984). The structural basis of an innate behaviour pattern. J Exp Biol
112, 283–319.
Wine, J.J. & Krasne, J.B. (1982). The cellular organization of crayfish escape behav-
ior. In The Biology of Crustacea, ed. E.D. Bliss, pp. 241–292. New York: Academic
Press.
Wine, J.J. & Mistick, D.C. (1977). Temporal organization of crayfish escape behav-
ior: delayed recruitment of peripheral inhibition. J Neurophysiol 40, 904–25.
Wong, D., Maekawa, M. & Tanaka, H. (1992). The effect of pulse repetition rate on
the delay sensitivity of neurons in the auditory-cortex of the fm bat, Myotis-
lucifugus. J Comp Physiol 170, 393–402.
Wong, D. & Shannon, S. (1988). Functional zones in the auditory cortex of the
echo-locating bat, Myotis lucifugus. Brain Res 453, 349–52.
Wu, J.-Y., Cohen, L.B. & Falk, C.X. (1994). Neuronal activity during different behav-
iors in Aplysia: a distributed organization? Science 263, 820–3.
Yu, A.C. & Margoliash, D. (1996). Temporal hierarchical control of singing in birds.
Science 273, 1871–5.
Zečević, D., Wu, J.-Y., Cohen, L.B., London, J.A., Höpp, H.-P. & Falk, C.X. (1989).
Hundreds of neurons in the Aplysia abdominal ganglion are active during the
gill withdrawal reflex. J Neurosci 9, 3681–9.
Zill, S.N. & Moran, D.T. (1981). The exoskeleton and insect proprioception. I.
Responses of campaniform sensilla to external and muscle-generated forces in
the American cockroach, Periplaneta americana. J Exp Biol 91, 1–24.
Zitnan, D., Kingan, T.G., Hermesman, J.L. & Adams, M.E. (1996). Identification of
ecdysis-triggering hormone from an epitracheal endocrine system. Science 271,
88–91.
Zottoli, S.J. (1977). Correlation of the startle reflex and Mauthner cell auditory
responses in unrestrained goldfish. J Exp Biol 66, 243–54.
Zottoli, S.J. (1978). Comparative morphology of the Mauthner cell in fish and
amphibians. In Neurobiology of the Mauthner cell., ed. D.S. Faber & H. Korn,
pp. 13–45. New York: Raven Press.
INDEX
Entries in bold type refer to pages where an important term, set in bold in the text, is
defined or used for the first time.
acceptance angle 89 carrier frequency 161

action potential 26 cartridge 102
active membrane 33 cell body 23, 62
adaptation, sensory 82, 105 central pattern generator 166, 177, 192
dark 92, 104 centre surround receptive field 108
light 92 cercus 71
agonist 193 chaffinch 239
amacrine cell 23 channel 29
retinal 25, 107 chemical sensitive 32
amphibians 9 voltage sensitive 30, 40, 192
amplify 105, 106 circuit 24, 42, 99, 117, 125, 140, 176, 192,
analogue signals in neurons 39, 81 201, 204, 206, 211, 225
Anser 4 closed loop 17
Apis 228 cochlea 149, 158
Aplysia 202, 236 cochlea nucleus 139, 149
apposition 88 cockroach 69, 77
approach 123 coding, sensory 81, 102
arthropod 4, 22, 23, 85 coincidence detection 140, 157
association 228 collision 123
associative learning 228 command neuron 43
axon 23, 24, 26, 38, 45, 51, 62, 68, 76, 96, commissural interneurons 192
140 computer model 15, 126, 193, 209
conditional oscillator 196
back propagation 210 conditioning 230
barn owl 132 connective nerve 24
bat 129, 142 constant frequency signals 145
bee 88, 228 context 217
binocular 90 contrast 96, 102, 105, 108
bipolar neuron 51 frequency 119–20
blowfly 87, 102, 104 control systems theory 15
Brachydanio 62 corollary discharge interneuron 57
Bufo 9 cranial relay neuron 67
crayfish 44
cable properties 34 cricket 174
call, bat 142 cue
Calliphora 87, 104, 111 acoustic 138
campaniform organ 77 visual 7, 125
canary 243
capacitor 36 dark adaptation 92
Carausius 62 decibel 130
262
Index 263
decision 26, 42, 52, 75 filter, sensory 18, 74, 99, 106, 112, 114
dedicated fixed action pattern 4
circuit 204 flight 109, 166
system 3, 129 fictive 172
delay line 140, 157 initiation 178, 182
dendrite 23, 24, 33, 40, 51, 62, 65, 66, locust 167–82
125, 225 muscle 167
sensory 79 rhythm 171
depolarisation 27 steering 167, 182
descending contralateral movement flow diagram 7
detector (DCMD) 122 flow field 112
deviation detector (DN) 182 fly 109
diffraction 87 fovea
digital signals 39, 82 auditory 158–9
dipteran 95 eye 89
directional selectivity 71, 112 frequency analysis 158
mechanism of 114–20 frequency modulated signals 143
disinhibition 176, 216 Fringilla 239
Doppler shift 145, 158
dorsal ramp interneuron 189 ganglion 24
dragonfly 86, 93 giant neurons 42, 70, 71
Drosophila 86, 114, 235 glial cell 24
glomerulus 230
ear glutamate 193
bat 148–9 goldfish 61
owl 132–5 goose 4
ecdysis 222 graded potential 36, 39
ecdysone 224 as analogue signal 39
echolocation 130, 142 summation of 36
echo ranging 152 grasshopper 122
eclosion hormone 224 gull 4
ecology 95
egg retrieval 4 H1 neuron 112, 115
electrical inhibition 68 habituation 52, 59, 236
electrical synapse 33, 49, 66 hair cell 149
electromyogram (emg) 169 hair receptor 51, 53–4, 58
elementary motion detector 115 harmonics 146
encode 94 Hemicordulia 93, 104
escape running 43, 71 Homo 91
escape swimming honey bee 88, 228
crayfish 43, 48, 55 horseshoe bat 145
Tritonia 186 horseshoe crab 106
excitation 29, 40, 65, 74, 125, 189 housefly 86
excitatory 27 HS neuron 112, 118–19
excitatory postsynaptic potential HVc 240–4
(EPSP) 32, 33, 36, 51, 66, 193, 202, hymenoptera 228
216 hyperpolarisation 27, 53
eye
compound 86, 183 inhibition 40, 74
simple 183 lateral 106
presynaptic 58
facilitation 115, 236 inhibitory 27, 56, 68–9
feature detector 99, 127 inhibitory motor neuron 56, 206
feedback 15, 16, 108, 180, 212, 225, 244 inhibitory postsynaptic potential (IPSP)
feed forward 224 32, 33, 38, 56, 125
figure-ground neuron 114 insect 100
filiform sensillum 71 integration 24, 27, 33–40, 74
264 Index
intensity-response curve 94, 105 memory 235

interneuron metamorphosis 222, 225–8
definition 24 Microchiroptera 130
flight generator 172–8 mollusc 186, 202, 235
local 24, 212 monosynaptic connection 58, 66, 67,
non-spiking 212, 215–18 176, 203
spiking 212–15 Mormoopidae 146
intracellular electrode 26 moth 222
intrinsic neuromodulation 190 motif 237, 242
invertebrate 4, 23, 168 motor
command 244
juvenile 235 control 212
juvenile hormone 227 giant neuron 45
neuron 23, 45, 67, 167, 171–8, 192–4,
Kenyon cell 232 202, 206, 209, 215–17, 225–7
pattern 4, 18
lamina 100 mouse-eared bat 143
large monopolar cell 102–8 moustached bat 146
Larus 5 multimodal neurons 182
larva 222, 225–8 Musca 111, 161
lateral giant interneuron 45–60 muscle
lateral inhibition 106 flight 167
leech 206 receptor organ 53–4, 56, 77
lens 86–8 vertebrate 167
light adaptation 92 mushroom body 232
Limulus 106 Myotis 143, 146–7, 152, 158
lobster 187, 194
lobula 101 nerve cell 2, 20–2
giant movement detector (LGMD) neurite 23
122 neuroethology 2
plate 109 neurohormone 185
local circuit 40 neuromodulator 187, 190, 232
locust 122 neuron
flight 166–86 birth 243
leg 212–18 culture 211
Locusta 87 definition of 20
logarithmic scale 93 kinds of 23–4
Lymnaea 211 loss 242, 243
parts of 22–3
Manduca 222 neuronal loop 184
mantis 17, 89 neuropile 24
map neurotransmitter 31, 190, 193
neuronal 137, 138–42, 158 noise 33, 106
somatotopic 201, 213 nucleus 22, 235
topographic 100 auditory 138–42
Mauthner bird song 235, 240–4
inhibitor 69 nudibranch mollusc 186
neuron 61–9 null direction 112
mechanoreceptor 51, 53, 77, 79–83
medial giant interneuron 45–7 ocelli 183
medulla 100, 117, 125 octopamine 185–7, 232–3
Megachiroptera 130 ommatidium 86, 87–91, 101
membrane open loop 17, 133
cable properties 34 optic lobe, insect 100
potential 27, 80 optic tectum
space constant 34 amphibian 11–15
time constant 36 bird 138
Index 265
optical monitoring 204–6 proboscis extension 228

optomotor proprioceptive 165
neurons 109–14 refractory period 29
response 109 releasing mechanism 6, 11–15
owl 131 resistor 35
resolution, of eye 86
pacemaker neuron 166 resting potential 27
parallel pathway 100, 118, 142, 204 retina, vertebrate 11, 24–5, 107–8
passerine 235 retinula cell 91
passive membrane 33 rhabdom 91
pathway, sensory 100 rhabdomere 91
pattern generator 55, 166 Rhinolophidae 145
perception 7 Rhinolophus 145–6, 150–2, 158–62
Periplaneta 70, 78 rhythm generation 166, 171–99
phase locking 139 robot 201, 211
phase resetting 173 Rohon beard cell 192
phasic response 54, 82, 94, 103 Rousettus 153
photon bump 92
photoreceptor 86, 97, 100–4 Sarcophaga 95
potential 91–6 saturated response 94
plastic song 239 segmental giant neuron 49
plasticity 181 selectivity 76, 114–15
polar plot 72 sensitive period 239–40
polypeptide 221, 224 sensitisation 236
positive feedback 59 sensitivity 86, 92
postsynaptic neuron 30 sensory
postsynaptic potential (PSP) 31 analysis 212
post-tetanic potentiation 236 filtering 51
preferred direction 112 modality 76–7, 182
presynaptic neuron 23, 51, 65, 76, 79, 82–4, 202–3,
facilitation 236 205, 206–9, 212, 215, 218, 236
inhibition 58 sign stimuli 5, 9–11
neuron 30 silver staining 20
prey recognition 9–15, 161 simulation 193
Procambarus 44 snail 211
proprioceptor 77, 180, 217 social releaser 5
Pteronotus 146, 155–8 song
pupa 222, 226 bird 235, 237–44
pyramidal cell 21 cricket 173–4
sonograph 237
rebound spike 192 sound 130
receptive field 73, 83, 106, 137, 138, frequency 130
214–15 nature of 130
reconfiguration 194–8 pressure level 130
receptor pulse 142–6
cell 76, 149 space constant 34
NMDA 193, 199, 236 spatial summation 38
olfactory 230 spike 26, 51–2, 65–73, 82, 153–5, 192–3,
potential 80, 91–6 205, 219
protein 32 as digital signal 39, 82
redundant conduction velocity 26, 67
element 185 frequency 39, 74
signal 108 initiation 52, 65, 84, 192
reflex 54, 77 initiation zone 38, 81
gill withdrawal 202–6, 236 spinal cord 190
local bending 206–11 starfish 186
locust leg 212–18 steroid 221, 224, 245
266 Index
stomatogastric ganglion 194–8 tight junction 66

stretch receptor 178, 225–7 time
summation 38, 202 constant 36
supernormal stimulus 7 marker neuron 153
superposition eye 88 Tipula 95
swimming toad 9, 70
crayfish 48–9, 54–5 tobacco hornworm 222
tadpole 190–4 tonic response 54, 82, 94
Tritonia 186, 188–90 tonotopic arrangement 137
syllable 237, 240–1 topographic map 100
synapse 30, 36–40, 52, 103, 125, 167, 214, transduction 79, 85, 92
218, 227, 236, 243–4 transmitter 196
mixed 49 trichoid sensillum 212
strength 52, 221, 236, 244 tritocerebral commissure giant neuron
types of 30–3 (TCG) 182
synaptic connection 46, 102, 211 tuning 112, 150, 154, 157, 159–60
syrinx 337–8 Tyto 132
tadpole 186, 190 velocity 145

Taeniopygia 237 vertebrate 4, 22–3, 85, 165, 190
tail flip 44 Vespertilionidae 143
tectal cells 12–15, 135–8 VS neuron 112
tectum 11, 135, 138 VUMX1 neuron 233–5
tegula 178
Teleogryllus 174 wind-sensitive hairs 169
teleost 61 wing
temporal summation 38, 202 muscles 167
Tenodera 16, 89 proprioceptors 178–82
thalamus 11–14
threshold Xenopus 186, 190
curve 150–1
for spike 28, 38–9, 52, 57 zebra finch 237
sensory 147 zebra fish 61

Nerve Cells and Animal Behaviour (2003)

Uploaded by

Copyright:

Available Formats

Nerve Cells and Animal Behaviour (2003)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nerve Cells and Animal Behaviour (2003)

Uploaded by

Copyright:

Available Formats

Nerve cells and animal

CAMBRIDGE UNIVERSITY PRESS

           is a Lecturer in the Department of

© Cambridge University Press 1999

First published in printed format 1989

A catalogue record for the original printed book is available

ISBN 0 511 01961 0 virtual (netLibrary Edition)

3 Giant neurons and escape behaviour 42

3.6 Summary of pathways in crayfish startle behaviour 59

4 Capturing sensory information 76

5 Stimulus filtering: vision and motion detection 99

6 Hearing and hunting: sensory maps 129

6.6 Interception of flying prey by bats 147

7 Programs for movement 165

8 Circuits of nerve cells and behaviour 201

9 Nerve cells and changes in behaviour 221

Our aim in this book is to introduce university students to research on

1.1 Nervous systems and the study of behaviour

Tinbergen himself made an impact on ethology by concentrating on field

1.2 Scope and limitations of neuroethology

As might be expected, neuroethology has been most successful in tackling

1.3 Neural implications of ethological results

The behaviour of an animal is to a large extent the product of activity in its

outside the breeding season, which is controlled by reproductive hor-

that combines information from a specific set of selectors so as to act as a

1.4 Sign stimuli in amphibians

worm configuration. If the small, square shape is elongated in the vertical

1.5 Neuroethology of a releasing mechanism

As with other vertebrates, early visual processing in amphibians takes place

response to the worm configuration in Class T5(1) neurons would be mini-

1.6 Control theory and nervous systems

In the attempt to pursue analysis of the mechanisms of behaviour to the

Figure 1.7 Orientation to prey in a mantis (Tenodera), illustrating move-

disturbance so as to correct that disturbance. This usually involves a special

is made clear by the visual orientation to prey in a praying mantis. The

The biological study of animal behaviour was put on a sound footing by

observations. In the course of this work, they developed a number of key

Manning, A. & Dawkins, M.S. (1998). An Introduction to Animal Behaviour, 5th

2.1 Basic organisation of nerve cells

Figure 2.1 Drawing of a pyramidal cell from a mammalian brain, an

Figure 2.2 Diagrammatic comparison of nerve cell structure in (a) verte-

cell body. Neurons involved in long-distance signalling have a long, cable-

interneurons. Some interneurons connect relatively distant parts of the

Figure 2.4 The pattern of connections in the primate retina, as revealed by

2.2 Neuron physiology and action potentials

Figure 2.6 Activity in a neuron recorded with an intracellular electrode. In

an interplay between depolarising and hyperpolarising membrane poten-

across a synapse can be studied by using two microelectrodes, one to

some synapses, the PSP excites the postsynaptic neuron by depolarising it

centration and electrical gradients across the membrane. Many of these

2.4 Integration of postsynaptic potentials

Integration involves weighing up the balance of EPSPs and IPSPs within a

because it contains the voltage-sensitive channels responsible for gener-

Figure 2.9 Integration of postsynaptic

2.5 Comparison of spikes and graded potentials

2.6 Additional mechanisms in integration

The processes of spatial and temporal summation of PSPs in passive den-

potentials usually originate at postsynaptic sites. Their amplitude reduces

When an animal is suddenly attacked by a predator, it must respond with

Figure 3.1 Giant interneurons involved in crayfish (Procambarus) startle