Novel Approaches to Visual Information Processing in Insects: Case Studies on Neuronal Computations in the Blowfly

Full text

179
0-8493-2024-0/05/$0.00+$1.50
© 2005 by CRC Press LLC
7
Novel Approaches to
Visual Information
Processing in Insects:
Case Studies on
Neuronal Computations
in the Blowfly
Martin Egelhaaf, Jan Grewe, Katja Karmeier
Roland Kern, Rafael Kurtz, and
Anne-Kathrin Warzecha
CONTENTS
7.1 Introduction ................................................................................................. 179
7.2 Methods for Analyzing Visual Information Processing in Small Neural
Networks ..................................................................................................... 181
7.2.1 Identifying Neural Networks .......................................................... 181
7.2.2 Computational Properties of Synaptic Interactions........................ 185
7.3 Approaches to Study the Reliability of Encoding Visual Motion
Information.................................................................................................. 188
7.3.1 Reliability of Neural Coding by Individual Nerve Cells............... 188
7.3.2 Coding of Motion Information by Neural Populations.................. 194
7.4 Approaches to Investigate the Encoding of Natural Visual Stimuli.......... 195
7.5 Conclusions and Outlook............................................................................ 200
Acknowledgments................................................................................................. 200
References............................................................................................................. 201
7.1 INTRODUCTION
Vision guides behavior in virtually all animals, especially in numerous insects.
The array of photoreceptors in the eye typically receives a wildly fluctuating
pattern of image flow when the animal moves through its environment. It is the
2024_C007.fm Page 179 Monday, August 16, 2004 8:54 AM

180
New Frontiers in Insect Neuroscience
task of the brain to interpret this complex spatio–temporal input and to make use
of it in guiding behavior. Biological nervous systems outperform existing artificial
vision systems with regard to their capabilities to process retinal image flow.
Insects make efficient use of visual information, for instance, during the aerobatic
chasing maneuvers of male flies in the context of mating behavior (Boeddeker and
Egelhaaf, 2003a, 2003b; Boeddeker et al., 2003; Land and Collett, 1974; Wagner,
1986) and the ability of bees to assess, on the basis of visual motion cues, the
distance traveled in an unknown environment (Esch and Burns, 1996; Esch et al.,
2001; Srinivasan et al., 2000). These extraordinary capabilities are most remark-
able given the small number of neurons in insect brains and the astonishing speed
with which retinal images are processed.
Because the nervous systems of insects are well amenable to electrophysiological
and imaging techniques under
in vivo
conditions, insects have served for many years
as model systems for analyzing the processing of retinal image flow (review: Egel-
haaf and Kern, 2002). Novel experimental approaches at both the behavioral and
the neuronal levels, as well as new techniques for data analysis, are beginning to
unveil the mechanisms underlying the amazing visual capabilities of insects (see
also Chapter 1, Chapter 2, Chapter 6, and Chapter 8). In particular, blowflies have
been used as a model system for understanding how visual motion information is
processed (Borst and Haag, 2002; Egelhaaf et al., 2002; Kurtz and Egelhaaf, 2003).
Results on blowflies will, therefore, form the basis of this review, but wherever
possible we also refer to the work done on other insect groups.
Visual motion has been shown to play an important role in behavioral control
in blowflies. Examples are visual course control, landing behavior, and object–back-
ground discrimination. In mediating these behavioral maneuvers, the nervous system
of blowflies relies on extracting behaviorally relevant information from the contin-
ually changing brightness patterns that are generated on the eyes during locomotion.
This so-called optic flow contains information both about the direction and speed
of the animal’s self-motion and about the environmental layout (Koenderink, 1986;
Lappe, 2000). The behavioral significance of motion vision in blowflies and other
insects is reflected in an abundance of motion-sensitive neurons in their visual system
(reviews: Wehner, 1981; Hausen, 1981; Hausen and Egelhaaf, 1989; Rind and
Simmons, 1999; see also Chapter 8).
An ensemble of large visual interneurons, the so-called tangential cells (TCs),
in the blowfly’s third visual neuropil (the lobula complex) has been characterized
in particular detail and is assumed to play a key role in processing visual motion
information in the context of visually guided behavior. Most TCs receive their main
input from two sets of retinotopically organized, local, motion-sensitive interneurons
with opposite preferred directions. As a consequence, they respond in a directionally
selective manner to motion in large parts of the visual field (reviews: Borst and
Haag, 2002; Egelhaaf and Warzecha, 1999; Egelhaaf et al., 2002; Hausen and
Egelhaaf, 1989). Neither the preferred directions of motion of TCs nor their motion
sensitivities are homogeneous, but these change in a systematic way across the visual
field. These intricate receptive field structures, which represent a phylogenetic adap-
tation rather than the outcome of sensory experience (Karmeier et al., 2001), suggest
that TCs are tuned to certain patterns of optic flow, as seen in the blowfly during
2024_C007.fm Page 180 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects
181
specific types of self-motion (Hengstenberg, 1982; Krapp and Hengstenberg, 1996;
Krapp et al., 2001).
Some TCs receive, either exclusively or in addition to their retinotopic input,
excitatory or inhibitory input from other TCs (Egelhaaf et al., 1993; Haag and
Borst, 2001, 2002; Hausen and Egelhaaf, 1989; Horstmann et al., 2000; Warzecha
et al., 1993). This input is assumed to enhance the selectivity of TCs for optic
flow. As a consequence, some TCs respond best during coherent wide-field motion,
as may occur while an animal turns around a particular body axis (reviews:
Egelhaaf and Warzecha, 1999; Egelhaaf et al., 2002; Hausen, 1981; Hausen and
Egelhaaf, 1989; Krapp, 2000). Others respond best to object motion, as may occur
while the animal pursues a moving target or passes a stationary object in its
environment (Collett, 1971; Egelhaaf, 1985a, 1985b; Gauck and Borst, 1999;
Gilbert and Strausfeld, 1991; Kimmerle and Egelhaaf, 2000a, 2000b; Olberg, 1981,
1986; Olberg and Pinter, 1990).
By employing novel approaches, three different but related aspects of motion
computation have been addressed in recent years:
1. Visual information processing in small neuronal networks
2. The reliability of encoding of visual motion by neuronal populations
3. The neuronal representation of natural optic flow
The merits and limitations of these novel approaches will be summarized in the
following pages, and we discuss how they have contributed to our understanding of
how behaviorally relevant visual information is processed in the insect brain.
7.2 METHODS FOR ANALYZING VISUAL INFORMATION
PROCESSING IN SMALL NEURAL NETWORKS
From a methodological point of view, the intricate network of TCs is a particularly
advantageous model system to analyze the cellular basis of neural computation under
in vivo
conditions. Dendritic information processing and synaptic transmission can
be analyzed by electrophysiological and optical recording techniques while the
animal is stimulated by its natural visual input. Imaging of the intracellular activity
distribution with calcium-sensitive dyes (see also Chapter 6 and Chapter 13) is
feasible in the virtually intact brain, because TCs arborize in a plane almost parallel
and close to the brain surface. In this way, it has not only been possible to unravel
major parts of the neuronal circuitry at this level of the visual pathway, but also to
elucidate how these neuronal circuits process information under their normal oper-
ating conditions.
7.2.1 I
DENTIFYING
N
EURAL
N
ETWORKS
Dual electrophysiological recordings (see also Chapter 12 and Chapter 14) have
been successfully applied to analyze synaptic connections within a neuronal circuit
and are still the technique of choice, because they provide a means of establishing
2024_C007.fm Page 181 Monday, August 16, 2004 8:54 AM

182
New Frontiers in Insect Neuroscience
synaptic connections between two cells and of characterizing their functional prop-
erties. For instance, one of the cells of the so-called horizontal system, the HSE cell
(horizontal system equatorial cell), was shown using this method to receive input
from two TCs that reside in the contralateral half of the visual system. Both are
excited by back-to-front motion. Owing to its ipsilateral retinotopic input, HSE is
excited by front-to-back motion, so it can be expected to respond best to wide-field
motion as is generated on the eyes when the animal turns about its vertical axis
(Figure 7.1) (Horstmann et al., 2000; Krapp et al., 2001).
Recently, two other experimental approaches were employed for network anal-
ysis. These approaches allow us to establish synaptic connectivity without requiring
simultaneous recording from two cells. The presumptive presynaptic neuron is
recorded intracellularly after the presumed postsynaptic cell was intracellularly
injected with an activity-dependent fluorescent dye (e.g., a calcium-sensitive dye
such as fura 2, calcium-green, or Oregon-green).
(14)
Synaptic connectivity is estab-
lished by depolarizing the presynaptic cell by current injection via the recording
electrode and monitoring the resulting fluorescence changes in the postsynaptic cell
(Haag and Borst, 2001, 2002, 2003). Synaptic network interactions have been further
established by the so-called fill-and-kill technique (Farrow et al., 2003; Miller and
Selverston, 1979; Warzecha et al., 1993). The presynaptic cell is injected with a
fluorescent dye, 6-carboxy-fluorescein,
(26)
which becomes phototoxic after laser illu-
mination. After photoablation of the presynaptic neuron, the functional consequences
for the postsynaptic cell can be characterized. Using this method, it was established
that a particular TC, a so-called
figure detection neuron
(FD1), is inhibited by one
of the GABAergic
centrifugal horizontal cells
(CH cells) (Figure 7.2) (Warzecha et
al., 1993). As a consequence of this inhibition, the FD1 neuron responds best to
object motion but shows little activity during wide-field motion (Figure 7.2b). This
circuit was further analyzed by photoablation of neurons that are presynaptic to the
CH cells: CH cells receive their main ipsilateral motion input not from retinotopic
FIGURE 7.1
a
: Schematic of synaptic input organization of the HSE cell in the right half
of the brain. It receives input from many retinotopically organized local motion-sensitive
input elements (indicated by thin lines) in the ipsilateral visual field. Information about back-
to-front motion in the contralateral visual field is mediated by the H1 and the H2 cell. The
H2 cell contacts the HSE cell close to its output terminal; the H1 cell is likely to make a
multitude of synaptic connections with its extended terminal region on the dendritic tree of
the HSE cell. Gray arrows indicate the direction of signal flow in the cells. The light gray
insets illustrate, seen from above, the fly looking at various motion stimuli and indicate the
preferred directions of motion of the different cells. The graphs (below) display the EPSPs
recorded in the HSE cell as evoked by the H1 and the H2 cell, respectively. Vertical arrows
mark the occurrence of the H1 and H2 spike, respectively.
b
: Time courses of responses of
the HSE cell in the right half of the brain to rotational optic flow, as well as the corresponding
monocular components. The arrows indicate front-to-back and back-to-front motion in the
right and left visual field, respectively. Although the cell mainly shows graded depolarizations
during ipsilateral front-to-back motion (first trace), there are many response transients during
contralateral back-to-front motion (second trace) and clockwise rotational optic flow (third
trace). The duration of motion is indicated by the black horizontal bar. (From Horstmann, W.
et al.,
Eur. J. Neurosci.
12
:2157–2165, 2000. With permission.)
2024_C007.fm Page 182 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects
183
small-field elements, as is the case for most other TCs, but via dendro–dendritic
interactions from the HS cells. After laser ablation of HS cells, the CH cells could
no longer respond to motion in front of the ipsilateral eye, and only responses to
motion in the contralateral visual field remained (Farrow et al., 2003).
FIGURE 7.1
...
...
...
...
H2-cell
HSE-cell
right half of the brain
left half of the brain
H1-cell
H1-induced EPSP
0
4
8
depolarisation [mV]
02468
msec
H1 spike H2 spike
H2-induced EPSP
0
4
8
depolarisation [mV]
024 86 msec
(a)
10 mV 50 msec
10 mV
50 msec
10 mV
50 msec
motion stimulation
HS responses
(b)
2024_C007.fm Page 183 Monday, August 16, 2004 8:54 AM

184
New Frontiers in Insect Neuroscience
By applying these approaches to neuronal network analysis, it has been possible
to unravel major parts of the wiring diagram of the intricate network of TCs in the
blowfly’s third visual neuropil. The interactions that take place within the network
are thought to play an important role in tuning TCs to particular types of optic flow,
as are generated on the eyes when the animal turns around a particular body axis
or passes a nearby object in front of a more distant background. The network
FIGURE 7.2
a
: Left: Schematic of input organization of one of the figure detection cells,
the FD1 cell. It receives input from many retinotopically organized local motion sensitive
input elements (indicated by thin lines) in the ipsilateral visual field. It is inhibited via
GABAergic synapses by one of the CH cells, the VCH cell. The VCH cell receives excitatory
input from the contralateral eye via the H1 and the H2 cell. Black arrows indicate the direction
of signal flow within the cells. Insets illustrate, seen from above, the fly looking at various
motion stimuli and indicate the preferred directions of motion (black arrows) of the different
cells or of the inhibitory input of the FD1 cell (open arrows). Bottom plot: Response of the
FD1 cell as a function of pattern size, illustrating that it is most sensitive to the motion of a
small pattern. (From Egelhaaf,
Biol. Cybern.
52
:195–209,1985a. With permission.)
b
: Spike
frequency histogram of responses of FD1 cell to wide-field and small-field motion (indicated
in the insets below the diagrams) before and after photoablation of the ipsilateral VCH cell.
Horizontal bars below the responses indicate the duration of motion. Arrows in insets sym-
bolize the size and direction of the moving pattern. Dotted horizontal lines indicate mean
response amplitudes during small-field motion. (From Warzecha, A-K. et al.,
J. Neurophysiol.
69
:329–339, 1993. With permission.)
right half of the brain
left half of the brain
...
object motion
right half of the brain
left half of the brain
...
...
GABAGABA
FD1-cell
-
-
VCH-cell
H1-cell
H2-cell
0
0.5
1
response [rel. Units]
612 24 36 48 72 120
pattern size [deg.]
200
100
0
1 sec
response sp ikes /sec][
normal response
response [rel. Units]
200
100
0
1 sec
after ablation of VCH-cell
(a) (b)
+
+
++
2024_C007.fm Page 184 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects
185
interactions comprise heterolateral interactions, recurrent inhibitory interactions, and
feed-forward inhibition. For instance, the HSE cells are major output elements of
the visual system. In parallel they act, via dendro–dendritic synapses, as excitatory
input elements of the CH cells, the inhibitory inputs of the object-motion sensitive
FD1 cell (see previous), as well as of an excitatory heterolateral element, the H1
cell (see Chapter 8). The latter neuron in turn excites, among other cells, the
contralateral HSE cell and the contralateral CH cells (Figure 7.3). Modeling is
required to infer the functional significance of this intricate connection pattern.
7.2.2 C
OMPUTATIONAL
P
ROPERTIES
OF
S
YNAPTIC
I
NTERACTIONS
Understanding the computational properties of a neuronal network requires knowl-
edge about details of the neuronal wiring scheme, but also about the functional
properties of the synaptic connections. In general, synapses are particularly important
sites of cellular information processing, because they may have peculiar nonlinear
properties and may even change their transmission properties depending on their
activation history (Fortune and Rose, 2001; Juusola et al., 1996; Paulsen and
Sejnowski, 2000; Sabatini and Regehr, 1999; Simmons, 2002; Thomson, 2000).
Meaningful representations of optic flow are often only achieved by specific
synaptic interactions between TCs (see previous section). To be beneficial, these
synaptic interactions must be carefully adjusted to the natural operating range of the
system. Otherwise, synaptic transmission may severely distort the information being
transmitted. This hazard is particularly daunting because synaptic transmission is
inherently noisy and the underlying biophysical processes have been found in many
FIGURE 7.3
Relationship of the two neural circuits sketched in Figure 7.1a and Figure 7.2,
tuning blowfly tangential cells either to coherent wide-field motion or object motion, respec-
tively. The cells are indicated by boxes. Excitatory and inhibitory synapses are indicated by
triangles and circles, respectively. Note the reciprocal recurrent inhibitory connections
between neurons in both halves of the visual system.
H2
H1
H2
H1
HSE
FD1
VCH
HSE FD1
VCH
right half of the brain
left half of the brain
2024_C007.fm Page 185 Monday, August 16, 2004 8:54 AM

186
New Frontiers in Insect Neuroscience
systems to be intrinsically nonlinear. Moreover, the transformation of the postsyn-
aptic potential into spike activity may also be nonlinear. Combined electrophysio-
logical and
in vivo
optical imaging experiments were performed in the blowfly to
analyze the relationship between the activity of a given presynaptic TC and the spike
rate of its postsynaptic target in the contralateral visual system. It was found that
the entire range of presynaptic depolarization levels that can be elicited by motion
in the “preferred direction” is transformed approximately linearly into the postsyn-
aptic spike rate (Figure 7.4) (Kurtz et al., 2001). This is surprising, especially given
the potential nonlinearities mentioned earlier. Linearity characterizes the transmis-
sion of membrane potential fluctuations up to frequencies of 10 Hz (see below;
Figure 7.6B) (Warzecha et al., 2003). Thus, the linear synaptic regime covers most
of the dynamic range within which visual motion information is transmitted with
high gain (Haag and Borst, 1997; Warzecha et al., 1998). In addition to slow graded
membrane potential changes, rapid presynaptic depolarizations, such as spikes, are
also transmitted reliably at this synapse (Warzecha et al., 2003). As a consequence
of the computational properties of the analyzed synapse, visual motion information
is transmitted largely undistorted to the contralateral visual system. This ensures
that the characteristic dependence of neural responses on stimulus parameters such
as velocity or contrast is not affected by the intervening synapse.
Presynaptic transmitter release is thought to be controlled by changes in the
presynaptic calcium concentration. Therefore, presynaptic calcium concentration
changes were monitored after the presynaptic cell was injected with a calcium-
sensitive fluorescent dye. This allows us to monitor the time-dependent changes in
calcium concentration in different parts of the axon terminal (Figure 7.4a). Interest-
ingly, a linear relationship was found between presynaptic depolarization as is
induced during preferred direction motion and the presynaptic calcium concentra-
tion. Moreover, the postsynaptic spike rate was also linearly related to presynaptic
calcium concentration increases (Figure 7.4b) (Kurtz et al., 2001). Although these
findings are in accordance with the overall characteristics of synaptic transmission,
one should be careful in interpreting the measured presynaptic calcium signals to
reflect precisely the calcium concentration that is relevant in the control of transmitter
release: with conventional
in vivo
imaging techniques, only the bulk calcium con-
centration changes within presynaptic arborizations can be monitored (see also
Chapter 13). In contrast, the calcium concentration changes that are relevant for
transmitter release are likely to be regulated very close to the presynaptic membrane
(review: Neher, 1998).
Because conclusions on the role of calcium in synaptic transmission are
critically dependent on the exact determination of presynaptic calcium changes,
two new technical approaches are currently tested. First, to overcome spatial
resolution limits of conventional intracellular imaging, two-photon laser scanning
fluorescence microscopy has recently been adjusted for
in vivo
analysis of pre-
synaptic calcium concentration changes at high spatial and temporal resolution
(Kalb et al., 2004). Second, to relate presynaptic calcium concentration changes
and postsynaptic activity in a more systematic way than is possible by visual
stimulation, the flash-photolysis technique is used to release calcium in the pre-
synaptic neuron. Here, the presynaptic neuron is injected with a photolabile cal-
AU: Pls
specify
where (which
section).
2024_C007.fm Page 186 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects
187
cium cage (BAPTA-1 hexapotassium salt or NP-EGTA tetrapotassium salt).
(14)
This compound is loaded with calcium that is rapidly released upon illumination
of the preparation with a brief, high-intensity flash of the appropriate wavelength
(Kurtz, 2004). These techniques are currently adjusted for their use in the intact
FIGURE 7.4
Transmission of optic flow information between a pair of TCs.
a
: Upper
diagram: Presynaptic calcium accumulation in a VS cell filled with a calcium-sensitive
fluorescent dye (raw fluorescence images of the entire cell and of the presynaptic region, left
diagrams) during presentation of preferred direction motion (black horizontal bar). White and
light gray values in the images correspond to increases in calcium concentration (measured
as relative change in fluorescence:
D
F/F). Bottom diagrams: Time courses of presynaptic
calcium concentration changes in three regions of the axon terminal (indicated in the insets)
for variable stimulus strengths (different line types). The insets correspond to the terminal
region as seen on the left raw fluorescence image.
b
: Linearity of the transfer of preferred
direction motion. Left: Postsynaptic spike rate (relative to resting activity) is plotted vs. the
presynaptic membrane potential change (
D
E
pre
) for visual stimuli of variable strengths, moving
either in the preferred direction (positive
D
E
pre
values) or in the null direction (negative
D
E
pre
values). The gain of signal transfer is about constant for the entire range of visually induced
excitations, resulting in a linear relationship between presynaptic potential and postsynaptic
spike rate upon motion in the preferred direction. A rectification is prominent for null direction
motion. Linear dependencies for preferred direction motion are also present in the relationship
between changes in presynaptic calcium and in presynaptic membrane potential (middle) and
in that between postsynaptic spike rate and changes in presynaptic calcium (right). (From
Kurtz, R. et al.,
J. Neurosci.
21
:6957–6966, 2001. With permission.)
(a)
∆F/F 15%
0%
–10%
25 µm200 µm
5%
1 sec
PD 0
PD1
PD2
PD3
PD4
–
–
––
–
––
–
–
–––
(b)
15
15
2024_C007.fm Page 187 Monday, August 16, 2004 8:54 AM

188
New Frontiers in Insect Neuroscience
fly brain to unravel the role of calcium in synaptic transmission — and thus in
neuronal processing of optic flow information.
7.3 APPROACHES TO STUDY THE RELIABILITY OF
ENCODING VISUAL MOTION INFORMATION
By simply looking at the architecture of neuronal circuits, it is hard to infer how
reliably they extract the relevant information under normal behavioral conditions.
One reason for this is the large variability of neuronal responses, even if these are
evoked by repeated presentation of the same stimulus. Although the across-trial
variance of the spike count in fly TCs is smaller than in motion-sensitive neurons
in the primate cortex (see e.g., Barberini et al., 2000; de Ruyter van Steveninck et
al., 1997; Warzecha and Egelhaaf, 1999; Warzecha et al., 2000), it is hard to deduce
from individual spike trains the extent to which variations in the interspike interval
are caused by the stimulus or by sources that are not time-locked to the stimulus
(
noise
). Neuronal response variability limits the timescale on which time-varying
optic flow can be conveyed.
The performance of a neuron in real time can only be understood by scrutinizing
individual response traces, not on the basis of ensemble averages. This requires
assessing the variability in a quantitative way by appropriate statistical measures.
The spike count variance across trials is the most straightforward measure of neu-
ronal variability. However, this measure does not take into account the time course
of neuronal activities that may be particularly important when analyzing responses
to time-varying stimuli. In the following section, we briefly summarize some meth-
ods to quantify the variability of time-varying neuronal data and to analyze the
timescale on which sensory information is encoded by the nervous system.
To understand the specificity of sensory coding, the analysis must be extended
from individual neurons to populations of neurons that all respond to somewhat
different aspects of the stimulus (see Chapter 14). The significance of population
coding can be analyzed particularly well on blowfly TCs, because the neurons
comprising the population are largely known and are accessible to experimental
analysis. Elaborate theoretical tools have been developed to assess (i) how well
different optic flow patterns, as induced by the animal’s self-motion, can be distin-
guished from each other and (ii) how specifically these optic flow patterns are
encoded on the basis of the population responses. Some of these theoretical tools
will be summarized with regard to optic flow encoding by fly TCs.
7.3.1 R
ELIABILITY
OF
N
EURAL
C
ODING
BY
I
NDIVIDUAL
N
ERVE
C
ELLS
There are various measures to quantify the reliability of neuronal responses to time-
varying stimuli. If repetitive stimulation with a given stimulus led always to identical
neuronal responses, the timing of spikes would be determined exclusively by the
stimulus. Obviously, sensory neurons do not respond with absolute accuracy, and
the timing of spikes varies considerably from trial to trial. Thus, it is most likely
that spike timing depends on noise sources, such as the stochastic absorption of
2024_C007.fm Page 188 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects
189
photons by the photoreceptors or the stochastic nature of transmitter release at
synapses and the subsequent opening and closing of ionic channels.
Several approaches to quantify neuronal variability are based on information
theory (reviews: Borst and Theunissen, 1999; Rieke et al., 1997). Applying these
measures allows us to analyze how much information about the stimulus is encoded
by a neuron. Information theory implies that a stimulus that assumes many states
contains more information than a stimulus assuming, for instance, only two states.
Moreover, the information content of a stimulus depends also on the probability of
the different stimulus states. Accordingly, the information conveyed by a neuron
depends on the probability distribution of the neuronal response levels. The mutual
information between stimulus and response specifies how much information about
the different states of the stimulus is represented by the neural response (Figure 7.5).
The mutual information between stimulus and response is constrained by the vari-
ability of neuronal responses upon repetitive presentation of a given stimulus. The
mutual information can be determined from the relationship of the probability
distribution of response levels when a particular stimulus is presented (conditional
probability) and the overall probability distribution of the different responses levels
obtained for all stimulus conditions (Figure 7.5).
Although, in principle, this procedure for determining the mutual information
is simple, it can be applied in practice only with some difficulty (Borst and Theu-
nissen, 1999). A major reason for this difficulty is that the different stimulus con-
ditions and the different response states need to be defined explicitly. This may be
feasible for simple stimuli (e.g., a pattern that moves only to the left or to the right,
or a moving pattern whose orientation is varied in discrete steps), but is hardly
possible for complex time-dependent stimuli. Therefore, more practical approaches
to quantify stimulus–response relationships have been developed that simplify mat-
ters but, as a trade-off, are based on certain assumptions about the properties of the
stimuli or of the neuronal responses. One type of approach,
linear reconstruction
techniques
, will be summarized briefly without going into formal details (for details,
see Bendat and Piersol, 2000; Borst and Theunissen, 1999; Marmarelis and Mar-
marelis, 1978; Rieke et al., 1997).
Here, the neuronal encoding of a signal is not quantified on the basis of prob-
ability distributions of stimulus conditions and response levels, but by relating the
measured responses to the output of an encoding model, that is, a model that accounts
for sensory information processing in formal terms. From a formal point of view,
the simplest encoding model is a linear one. The optimal linear model that allows
one to estimate a time-dependent signal from another signal can be determined by
the so-called
reverse reconstruction approach
(review: Rieke et al., 1997) (Figure
7.6a). This approach has frequently been applied to sensory systems, including
blowfly TCs, to investigate how well stimulus velocity can be estimated on the basis
of neuronal responses (Bialek et al., 1991; Haag and Borst, 1997, 1998). It was
shown that time-dependent stimulus velocity is linearly encoded by blowfly TCs as
long as pattern velocity and the velocity changes are relatively small. Otherwise,
acceleration and higher-order temporal derivatives play an increasingly large role in
shaping the time course of the motion response (Egelhaaf and Reichardt, 1987; Haag
and Borst, 1997). Recently, the reverse reconstruction approach has also been applied
2024_C007.fm Page 189 Monday, August 16, 2004 8:54 AM

190
New Frontiers in Insect Neuroscience
FIGURE 7.5
Information transmitted by a neuron about an input signal (mutual informa-
tion).
a
: Simulated tuning curve for a range a stimuli. The mean responses and standard
deviations are shown as can be measured experimentally. The information conveyed by the
neuron depends on the probability distribution of the response amplitudes. The more the
probability distributions of responses to particular stimuli p(r|s
x
) deviate from the overall
response distribution that is obtained for all stimuli p(r), the better the different stimuli can
be distinguished on the basis of the neuronal responses and the higher mutual information.
b
: Definition of mutual information and explanation of symbols. The mutual information
between stimulus and response can be determined from the relationship of the probability
distribution of response levels when a particular stimulus is presented (conditional probability)
and the overall probability distribution of the different responses levels that is obtained for
all stimulus conditions.
p(r) probability distribution of the neural response to any stimulus
p(sx) probability that the stimulus takes the value sx
p(r|sx probability of neural responses when the stimulus sx is presented
(conditional probability)
Information about a particular stimulus sx
Isx = ∑ p(r|sx) log2 (p(r|sx) / p(r))
Average mutual information obtained from all stimulus conditions:
Is = ∑∑ p(sx) p(r|sx) log2 (p(r|sx) / p(r))
0 60 120 180 240 300 360
–25
–20
–15
–10
–5
0
5
10
15
20
25
stimulus s
neuronal response
0 0.05 0.1
probability
← P(r|sx)
← P(r|s)
(a)
(b)
)
2024_C007.fm Page 190 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects
191
FIGURE 7.6
Reverse reconstruction of the time-dependent presynaptic potential from
postsynaptic spike trains.
a
: Schematic outline of the procedure. The linear filter is determined
that, when convolved with the postsynaptic spike train, leads to the best estimate of the
presynaptic potential. Hyperpolarizations of the presynaptic neuron do not have much effect
on the activity of the postsynaptic cell (due to the low postsynaptic resting activity). Therefore,
the presynaptic potential was rectified at the resting potential of each response trace and was
then used for the reconstruction. The part of the response that was rectified is marked by the
shaded bar. The coherence function was determined as a measure of the similarity between
the recorded and the estimated presynaptic membrane potential traces.
b
: Coherence deter-
mined for a cell pair analyzed with random velocity fluctuations. Coherence values close to
1 for frequencies up to 10 Hz indicate that the system can be regarded as very reliable and
approximately linear in this frequency range. (Data from Warzecha, A-K. et al.,
Neurosci.
119
:1103–1112, 2003. With permission.)
(a)
(b) coherence
frequency [Hz]
0
0.2
0.4
0.6
0.8
1
10 100
presynaptic potential
20 mV
postsynaptic spikes
comparison
coherence
250 msec
best linear
filter
reverse
reconstruction
20 mV
12
8
4
0
–25 0 25 msec
mV
reconstructed presynaptic potential
2024_C007.fm Page 191 Monday, August 16, 2004 8:54 AM

192
New Frontiers in Insect Neuroscience
to relate the time-dependent postsynaptic signal to the corresponding presynaptic
input (Figure 7.6a). The results of these experiments have already been summarized
earlier (Warzecha et al., 2003).
The similarity between the responses of the encoding model and the experimen-
tally obtained responses can be determined on the basis of the so-called coherence
function, which relates two time-dependent signals as a function of their frequency
components (Bendat and Piersol, 2000; van Hateren and Snippe, 2001). In technical
terms, the coherence function is the normalized power spectrum of the cross-corre-
lation function between the two time-dependent signals. It varies between 0 (i.e.,
both signals are unrelated) and 1 (i.e., both signals can be linearly transformed into
each other) (Figure 7.6b). There is one ambiguity with the coherence function, which
is that a coherence value smaller than 1 can arise from two nonexclusive sources.
First, it can be due to inherent noise in the neural pathway, and thus arise from the
variability in the neural responses. Second, the neural responses and the responses
predicted on the basis of the encoding model are not linearly related. To account
for the noise in encoding the stimulus, the so-called expected coherence can be
assessed (Haag and Borst, 1998; van Hateren and Snippe, 2001). The expected
coherence is determined by calculating the coherence function between the stimulus-
induced response component (i.e., the ensemble average of a sufficiently large
number of individual responses to the same input) and the individual responses. The
expected coherence is related to the signal-to-noise ratio of the neuron (Haag and
Borst, 1998; van Hateren and Snippe, 2001). The expected coherence represents the
upper limit, given neuronal variability, of what can be maximally expected to be
encoded even if a perfect encoding model were available.
The reliability of neural encoding has recently been addressed on the basis of
another type of approach to assess the similarity of neural responses. In the context
of blowfly motion vision, this approach has been applied to pinpointing the domi-
nating noise source in the visual motion pathway. Here, the similarity of neural
responses to a given stimulus is related to the similarity of response traces evoked
by different stimuli. The similarity of spike trains can be determined in two ways.
The first approach compares spike trains by calculating the minimal costs of trans-
forming one spike train into another (Victor and Purpura, 1996). The transformation
is done by either deleting, inserting, or temporally shifting single spikes. Each of
these procedures is linked to defined costs (Figure 7.7a,b). By varying the costs for
shifting a spike relative to inserting or deleting one, the temporal resolution of the
procedure is changed. The second approach to determine similarity calculates the
distance between two spike trains that are smoothed by temporal filtering. The
distance is given by the square root of the squared differences between the smoothed
spike trains (Kretzberg et al., 2001b). Changing the filter width leads to a varying
temporal resolution of the procedure.
Both these approaches to determine the similarity of neural responses were used
in a blowfly TC to test whether the timing of spikes elicited by visual motion is
determined by photon noise, as was proposed in earlier studies (Borst and Haag,
2001; Lewen et al., 2001), or by noise inherent in the nervous system. Simulating
photon noise by random brightness fluctuations of moving dots allowed us to show
that the reliability of spike timing is dominated by noise intrinsic to the nervous
system (Figure 7.7c) (Grewe et al., 2003).
2024_C007.fm Page 192 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects
193
FIGURE 7.7
Similarity of spike trains as evoked by motion stimuli superimposed by variable
noise.
a
: The raster plot shows sections out of the spike responses of a blowfly TC (the H1 cell)
to a reference and a test stimulus. Both stimuli consisted of the same pattern of ten randomly
positioned dots that were moved at a constant velocity in the cell’s preferred direction. The dots
were superimposed by random luminance fluctuations (noise). These were the same for each
presentation of the reference stimulus, but differed, although they had the same statistical
properties, for each presentation of the test stimulus (for details, see Grewe et al., 2003). A
common spike pattern can be seen in all responses, despite differences in timing of individual
spikes. The way of analyzing whether the temporal structure of the reference responses are more
similar to each other than to the test responses is sketched by the lines connecting the highlighted
reference response and all other reference and test responses.
b
: The similarity between two
spike trains is defined as the inverse of the minimal costs of transforming one spike train into
another one (Victor and Purpura, 1996). The transformation is done by either deleting, inserting,
or temporally shifting single spikes. Deleting or inserting single spikes has the cost of 1. The
cost of a temporal shift (
q
per second) is variable and determines the temporal resolution of the
measure. For example, a
q
value of 200 equals a temporal resolution of 10 msec: for this time-
scale spikes in two response traces are considered nearly coincident if a given spike in one of
the spike trains is shifted by less than
±
10 msec with respect to the corresponding spike in the
other spike train. As long as this condition is met, it is “cheaper” to adjust the spikes temporally
in the two spike trains by shifting than it is by deleting and inserting one of them.
c
: The pairs
of mean similarities, of each reference response to all test responses, and of each reference
response to all other reference responses are attributed either to the reference or the test stimulus.
Assuming that the reference responses are more similar to each other than to the test responses,
the larger similarity value was assigned to the reference response. The percentage of correct
decisions is plotted as a function of added noise level. The shaded area represents the domain
of uncertainty
(see Grewe et al., 2003, for details). Discrimination performances falling into
this range are likely to be a consequence of chance. A significant effect of the added noise on
the responses can be assumed if the actual percentage-correct value is outside the domain of
uncertainty. (From Grewe, J. et al.,
J. Neurosci.
23
:10776–10783, 2003. With permission.)
(a)
(b)
48163264
20
30
40
50
60
70
80
90
100
noise level
percentage correct
reference responses
spike train 1:
spike train 2:
t
transformation costs:
temporal shifting
deleting/inserting
1 + 1 = 2
deleting inserting q[costs/s] defines the
temporal resolution
(c)
test responses
2024_C007.fm Page 193 Monday, August 16, 2004 8:54 AM

194
New Frontiers in Insect Neuroscience
These results are in accordance with findings of earlier studies based on dual
recordings of TCs with largely overlapping receptive fields and a correlation analysis
of their spike activity. Although the first spike after a rapid change in the direction
of motion may be precisely time-locked to the stimulus, the timing of most spikes,
even during dynamical motion stimulation, is determined by the refractory period
of the neuron and by random membrane potential fluctuations that are not directly
related to the stimulus (Kretzberg et al., 2001a; Warzecha et al., 1998).
7.3.2 C
ODING
OF
M
OTION
I
NFORMATION
BY
N
EURAL
P
OPULATIONS
The encoding of stimuli by neuronal populations is a basic operating principle of
nervous systems. The number of neurons that constitute such a neuronal population
may be very large. However, the population of TCs in the blowfly motion vision
system comprises only a relatively small number of 50 to 60 cells in each half of
the brain. Each of these cells is thought to encode different aspects of optic flow
(see above; reviews: Borst and Haag, 2002; Egelhaaf et al., 2002; Hausen and
Egelhaaf, 1989; Krapp, 2000). The representation of sensory information by neuronal
populations raises the question of the mechanism that eventually decodes this dis-
tributed information and uses it to guide behavior. In particular, it is important to
assess how well relevant stimulus parameters are preserved in the population
response and how specifically these parameters can be extracted by a readout mech-
anism. Whereas there are only few experimental studies addressing this problem,
the constraints that must be taken into account when interpreting neuronal population
responses are analyzed theoretically or by model simulations in a wide variety of
studies (reviews: Dayan and Abbott, 2001; Oram et al., 1998; Pouget et al., 1998,
2003).
Population coding is most frequently analyzed for a bank of sensory neurons
encoding a single stimulus parameter. The optimal stimulus of different neurons
varies along the stimulus axis. The tuning of an individual cell with respect to this
stimulus parameter is usually described by a Gaussian function; that is, the response
first increases with changing stimulus parameter, reaches an optimum, and then
decreases again. Alternatively, for periodic stimuli axes (e.g., for orientation tuning),
harmonic functions, such as cosine functions or truncated cosine functions, are
employed to describe tuning curves.
Neural population coding has recently been studied in a subpopulation of blowfly
TCs, the so-called VS cells. The ten VS cells in each half of the brain are thought
to signal rotations of the head around different axes lying in the equatorial plane of
the fly’s eye (Krapp et al., 1998). However, VS cell responses are highly ambiguous
because they are also activated during upward lift movements of the animal
(Karmeier et al., 2003). Even if we assume VS cells to be perfect detectors for self-
rotation, their responses are ambiguous due to their cosine-shaped tuning curves and
the variability of their responses (Figure 7.8a,d). Assuming a single stimulus dimen-
sion (i.e., the orientation of the axis of self-rotation), encoding accuracy of the
population of VS cells was determined (Figure 7.8b,f). For each stimulus condition
(i.e., for each rotation axis) the activity distribution of all ten VS cells is determined.
AU: Pls
specify
where.
2024_C007.fm Page 194 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects
195
Given these response probabilities, the likelihood of a particular rotation axis given
a certain response — and, thus, the most likely rotation axis — can be estimated
using Bayes’s theorem (Dayan and Abbott, 2001). The encoding error is determined
from the estimated and the real rotation axis (Figure 7.8c,e,f).
This approach has been applied to VS cells for different rotational velocities and
different time windows after the onset of motion. Despite the considerable neural
noise, the rotation axis can be estimated within 5 msec with an accuracy of less than
2
∞
. This result is surprising, given the ambiguous responses of individual sensory
neurons (Figure 7.8a,b), and it thus stresses the significance of population coding.
The decoding performance of VS cells does not improve for larger time windows
but deteriorates considerably if only three instead of the ten VS cells are taken into
account (Figure 7.8d,e). The finding that a very small time window is sufficient for
decoding the orientation axis from the population response has important functional
implications for flies, because they perform rapid acrobatic flight maneuvers, which
require fast and accurate sensory control signals for visual flight and gaze stabili-
zation (see Chapter 4).
Theoretical studies demonstrate that the accuracy of stimulus encoding depends
on a variety of other aspects, such as noise correlation between single elements of
the population, the width of the tuning curves, and the number of stimulus dimen-
sions to be encoded (e.g., Abbott and Dayan, 1999; Pouget et al., 1999; Wilke and
Eurich, 2001; Zhang and Sejnowski, 1999). These features must be taken into
account, although it is far from trivial to obtain the large amount of experimental
data required to understand fully the encoding of sensory information by neuronal
populations. All these analyses are only first steps toward understanding the encoding
of complex natural motion stimuli as experienced by the animals in normal behav-
ioral situations (see next section). This will require the development of novel theo-
retical approaches to analyze neuronal population data (see also Chapter 5).
7.4 APPROACHES TO INVESTIGATE THE ENCODING
OF NATURAL VISUAL STIMULI
Information processing within a neural circuit is traditionally analyzed with stimuli
that are much simpler with respect to their spatial and dynamical features than the
input an animal encounters in behavioral situations. Because visual systems evolved
in specific environments and behavioral contexts, the functional significance of the
information being processed can only be assessed by analyzing neuronal perfor-
mance under conditions that come close to natural situations.
One important aspect applies to the dynamical properties of the optic flow that
is encountered during behavior. These are determined by the dynamics of the ani-
mal’s self-motion and the three-dimensional layout of the environment. The char-
acteristics of the optic flow may differ greatly for different species and in different
behavioral situations. Some insects, such as hoverflies, dragonflies, and hawkmoths,
are able to hover in midair in front of a flower (Collett and Land, 1975; Farina et
al., 1994; Kern and Varjú, 1998). From their current position in space, these insects
can accelerate rapidly and dart off at high velocities. Blowflies usually change the
2024_C007.fm Page 195 Monday, August 16, 2004 8:54 AM

196
New Frontiers in Insect Neuroscience
FIGURE 7.8
Accuracy of population coding illustrated for a class of TCs, the ten VS cells.
a
: The response of each VS cell depends on the orientation of an horizontally aligned axis
about which a panoramic wide-field pattern is rotated. The tuning can be approximated by a
distorted cosine function.
b
: The rotation axis can be estimated from the measured response
distributions obtained for the different rotation axes p(r|s
i
) (compare Figure 7.5). By applying
Bayes’s theorem (see
f
), the probability distribution of rotation axes can be determined for a
given response p(s
i
|r).
c
: The coding error is given by the standard deviation between the
estimated orientation axis and the real orientation axis (for formal details see
f
). If the response
of only one VS cell is taken into account, the coding error varies strongly with the rotation
axis and may assume very large values. d: If the responses of more VS cells are taken into
account, the coding error decreases to very small values, even if only a 5 msec response
interval is used for decoding. Tuning curves of all ten VS cells (gray lines) and a subgroup
of three VS cells (black lines). e: Corresponding coding errors. f: Formal explanation of the
decoding procedure.
(f)
0 60 120 180 240 300 360
0
5
10
15
rotation axis [deg.]
activity [mV]
0 60120180240300360
20
10
0
10
20
activity [mV]
0
0.1
20 10 0 10 20
0
60
120
180
240
300
360
activity [mV]
rotation axis [deg.]
0
0.2
0.4
0.6
0 60 120 180 240 300 360
0
20
40
60
80
rotation axis [deg.]
error [deg.]
0 60 120 180 240 300 360
10
5
0
5
10
15
20
rotation axis [deg.]
activity [mV]
3 cells
10 cells
0 60 120 180 240 300 360
0
1
2
3
4
5
6
7
8
rotation axis [deg.]
error [deg.]
(a) (b)
P(r|s)P(s|r)
(c)
(d) (e)
Estimated stimulus for a
given population response r sest = ∑ p(sx|r) sx
sxError √(s – sest)2
r response vector describing the population response
p(r|sx) can be measured: probability distribution of the population response to the
stimulus sx
p(sx|r) to be determined: probability of the stimulus sx given the response vector r
Applying Bayes Theorem allows determination of p(sx|r) from measurable data:
p(r|sx) p(sx)
p(sx|r) = p(r)
2024_C007.fm Page 196 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects 197
direction of self-motion rapidly by saccadic turns during flight (van Hateren and
Schilstra, 1999) or, one order of magnitude more slowly, while walking (Horn and
Mittag, 1980; Kern et al., 2001b). Because the optic flow pattern and its dynamics
are species and context specific, it is reasonable to assume that the mechanisms
extracting motion information are adapted to behaviorally relevant conditions.
Because it is hard to record from visual interneurons in freely moving insects, more
indirect approaches have recently been used to determine the responses of motion-
sensitive neurons to a variety of approximations to behaviorally generated optic flow.
Recordings of the H1 cell have been made from the brains of blowflies that were
oscillated with a turntable with dynamics that mimic the rotational component of
flight trajectories of unrestrained small houseflies (Lewen et al., 2001). For technical
reasons, the most distinctive feature of optic flow of free-flying blowflies (i.e., the
succession of saccadic turns and stable gaze) (van Hateren and Schilstra, 1999) was
not taken into account and the angular velocity of the turntable was only half that
of the flight trajectory.
In an alternative approach, neural responses were recorded from the brain of
tethered moths flying in a flight simulator in which the animal can influence its
visual input as under free-flight conditions (Gray et al., 2002). So far, this elegant
approach is restricted to relatively large insect species, such as locusts or moths,
changing their direction of flight relatively slowly.
In another approach, the optic flow experienced by behaving blowflies was
reconstructed and replayed to the animal during nerve cell recordings. This approach
has been employed for various behavioral situations during tethered flight in a flight
simulator (Kimmerle and Egelhaaf, 2000b; Warzecha and Egelhaaf, 1996, 1997),
during unrestrained walking in a three-dimensional environment (Kern et al., 2000,
2001a), and recently during rapid free-flight maneuvers in a three-dimensional
environment (Figure 7.9a) (Kern et al., 2001b, 2004a, 2004b; Lindemann et al.,
2003).
The simulation of free flight has become possible thanks to the development of
sophisticated techniques. First, free-flight behavior can be monitored by means of
magnetic sensor coils mounted on the head and thorax of the animal (van Hateren
and Schilstra, 1999; Schilstra and van Hateren, 1999) or by high-speed digital
cameras (Oddos et al., 2003). Second, a panoramic visual stimulator (FliMax) for
presentation of optic flow has been designed that is sufficiently fast for visual stimuli
as experienced by free-flying insects (Figure 7.9a). FliMax is a special-purpose
panoramic VGA output device generating image frames at a frequency of 370 Hz
(Lindemann et al., 2003). FliMax is composed of printed circuit boards shaped like
equilateral triangles (side: 30 cm), assembled to form 14 of the 20 sides of an
icosahedron (radius of inscribed sphere: 22.4 cm). Each of these boards supports
512 regularly spaced, round light-emitting diodes, or LEDs (#WU-2-53GD, 5 mm,
emitting wavelength: 567 nm, effective viewing angle: 25º).(32) All 7168 LEDs are
controlled individually via a computer equipped with a standard VGA graphics card
and customized software. The luminance of each LED, adjustable to eight intensity
levels, is kept constant between updates by sample-and-hold circuits. FliMax is open
toward the back to mount an animal in its center and to make recordings from the
blowfly’s brain (Lindemann et al., 2003).
2024_C007.fm Page 197 Monday, August 16, 2004 8:54 AM

198 New Frontiers in Insect Neuroscience
FIGURE 7.9 Responses of the HSE cell of a blowfly to behaviorally generated optic flow as
experienced during a free-flight maneuver. a: Upper diagram: Flight trajectory as seen from above
monitored in a cubic cage (0.4 ¥ 0.4 ¥ 0.4 m’) covered on its side walls with images of herbage
(for clarity shown only at low contrast). The position of the fly is displayed by small spheres
every 10 msec. The position and orientation of the head are shown every 130 msec. Bottom
diagram: Photograph of FliMax from behind. In the foreground, the micromanipulators can be
seen by which recording electrodes are inserted into the fly’s brain. b: Responses of an HSE cell
in the right half of the visual system to behaviorally generated stimuli. Top trace: Individual
response; cell responds to motion with graded de- and hyperpolarizations; spikelike depolariza-
tions superpose the graded potential changes. Second trace: Average response (n = 7). Third trace:
Angular velocity of the fly’s head. Sharp angular velocity peaks corresponding to saccadelike
turns of the fly dominate the time-dependent angular velocity profile. Positive (negative) values
denote counterclockwise (clockwise) turns of the head in a head-centered coordinate system. In
contrast to expectations based on the input organization of the HSE cell (Figure 7.1a), there are
no obvious response peaks during preferred direction motion evoked by counterclockwise sac-
cades. However, there are pronounced hyperpolarizations going along with clockwise saccades.
Dotted horizontal lines indicate resting potential. c: Saccade-triggered average of the HSE
responses. Counterclockwise saccades (solid line) go along with image motion in the HSE cell’s
preferred direction; clockwise saccades (dotted line) go along with image motion in the null
direction. Zero time corresponds to the maximum angular velocity. The resting potential was
subtracted before averaging. (Flight trajectory provided by JH van Hateren; Experimental data
from Lindemann, J.P. et al., Vision Res. 43:779–791, 2003. With permission.)
–50
–60
–40
–30
–50
–60
–40
-
0
1000
2000
3000
–1000
–2000
–3000 500 msec
individual response [mV]
average response [mV]
angular velocity [deg./sec]
(b)
(a)
15
10
5
0
–5
–10
response [mV]
–40 –20 0 20 40 60
time relative to saccade [msec]
(c)
Fly
Fly
2024_C007.fm Page 198 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects 199
Although the analysis of how natural visual stimuli are processed has started
only recently, it is already safe to conclude that the neuronal responses to complex
optic flow as experienced under behavioral conditions can be understood only partly
in terms of the concepts that were established on the basis of experiments done with
conventional motion stimuli. This is illustrated in the following example.
In moving animals, retinal image flow is distinguished from conventional (exper-
imenter-designed) visual stimuli by its characteristic dynamics, which are largely
determined by the animal’s own actions and reactions. During spontaneous flight,
blowflies execute a series of saccadic turns where the head shows angular velocity
peaks of up to several thousand degrees per second. Between saccades, the gaze is
kept basically stable (van Hateren and Schilstra, 1999; Schilstra and van Hateren,
1999). The resulting retinal image flow was reconstructed and replayed to blowflies.
The resulting complex time-dependent responses were related by reverse reconstruc-
tion approaches (see Section 7.3.1) to various self-motion parameters, such as yaw
rotation, forward translation, and sideslip. How well the original self-motion param-
eters can be estimated from the neuronal responses is quantified by the coherence
(see Section 7.3.1). Our results, obtained with natural optic flow on the so-called
HSE neuron, do not match conclusions based on systems analysis of the neuron’s
properties with experimenter-designed stimuli (Kern et al., 2004a):
1. It was previously concluded that the neuron should mainly act as a detector
of self-rotation of the animal around its vertical axis (see Figure 7.1). In
contrast, our results with behaviorally generated optic flow show that the
neuron fails to encode faithfully even the most prominent turns of the
animal as found during saccades.
2. Although the cell experiences the largest optic flow during saccades, it may
encode behaviorally relevant information, especially between saccades.
Between saccades blowflies keep their gaze stable apart from small, broad-
band yaw rotations, so they may gather useful information about the outside
world from the translational optic flow components that dominate at low
frequencies in intersaccadic intervals. Indeed, between saccades, neural
signals provide rich information about the spatial relation of the animal to
its surroundings. It should be noted that distance is signaled only relative
to the fly’s own velocity, because retinal velocities evoked during translation
are inversely proportional to distance and proportional to translation veloc-
ity. This implies that in walking flies, the visual surroundings should affect
the responses of the HSE cell only when the fly is very close to environ-
mental structures, just as has been found in electrophysiological experiments
(Kern et al., 2001a, 2001b). This implicit scaling of distance information
by the actual speed of the animal may be a parsimonious and advantageous
way to extract from optic flow behaviorally relevant information about the
outlay of the environment, because, for instance, evasive actions evoked by
obstacles in the path of locomotion need to be evoked only at a smaller
distance when the animal moves more slowly.
3. Based on experimenter-designed motion stimuli, motion-sensitive neurons
are frequently expected to encode stimulus velocity. Indeed, stimulus
2024_C007.fm Page 199 Monday, August 16, 2004 8:54 AM

200 New Frontiers in Insect Neuroscience
velocity can be reconstructed faithfully from the responses of blowfly
motion-sensitive neurons, as long as the velocities and velocity changes
are relatively small (Bialek et al., 1991; Egelhaaf and Reichardt, 1987;
Haag and Borst, 1997). However, we could show that under behaviorally
relevant stimulus conditions, the visual motion system of the blowfly
operates far beyond this linear range for a considerable portion of time
(Figure 7.9b,c). We concluded that, in contrast to previous views, the
nonlinearities of the visual motion system may be essential for the HSE
cell to encode information about the spatial relation of the animal to its
environment. If the neuron encoded linearly the entire velocity range the
system encounters in behavior, by far the largest responses would be
generated during body saccades. This would leave only a very small
response range for encoding optic flow in the intersaccadic interval. Given
the noisiness of neuronal signals (see above), it might be difficult to extract
meaningful information about the spatial organization of the surroundings
from the neuronal responses. Because angular velocities during saccades
are far beyond the linear range of the motion detection system, the HSE
cell appears to be able to encode useful information about translation and
thus about the spatial relation of the animal to the outside world. This
finding emphasizes that the significance of neuronal circuits can only be
assessed if they are probed in the natural operating range.
So far, only the specificity of individual neurons has been analyzed in encoding
of behaviorally relevant features from the natural optic flow. With the available
techniques, the next step will be to understand how this specificity is increased by
taking into account more of the relevant elements of the TC population. This analysis,
however, will require the development of novel conceptual and technical approaches
to handle the resulting complex time-dependent population data.
7.5 CONCLUSIONS AND OUTLOOK
Thanks to great methodological and conceptual developments, many aspects of the
neural computations underlying visually guided orientation behavior in the blowfly
have been elucidated in recent years. Analysis has ranged from the biophysical
properties of neurons and their synaptic interactions, to the performance and reli-
ability of neural populations in encoding behaviorally relevant motion sequences,
to orientation behavior in flight simulators and under free-flight conditions. Accord-
ingly, the employed methodological repertoire is very broad. On the one hand,
sophisticated techniques of cellular physiology had to be adapted to in vivo condi-
tions — to mention only the imaging of the spatially resolved intracellular activity
patterns and, in particular, the time-dependent distribution of ions involved in intra-
cellular information processing. Likewise, thanks to novel compounds such as pho-
tolabile calcium cages, targeted manipulation of intracellular ion concentrations
could be performed. On the other hand, developments, for instance in computer
graphics, have recently allowed for the first time the presentation of natural motion
stimuli in electrophysiological experiments, as they were encountered by freely
AU: Pls
specify
where.
2024_C007.fm Page 200 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects 201
behaving animals. The complex spatio–temporal properties of natural image
sequences and the resulting complex time-dependent neuronal responses have made
it necessary to establish and adapt sophisticated techniques for analysis of such
complex data structures. Moreover, taking into account the variability of responses
of individual neurons and the activity of whole populations of them has made another
set of novel approaches to data analysis necessary and neurobiology an increasingly
interdisciplinary effort.
Insect model systems, such as the visual motion pathway of the blowfly, are
particularly advantageous for this multifaceted interdisciplinary effort. Experimental
analysis down to the biophysical level can be done in vivo, where the system can
be probed with behaviorally generated input and, thus, under its natural operating
conditions. Because populations of neurons are relatively small compared to those
of mammals, it appears to be possible to boil down visually guided behavior to the
computational properties of neural networks of identified neurons. Despite the small
size of insects’ brains, these computations are far from trivial, given the fact that
many insects, such as blowflies, are able to perform extraordinary things — at least
when compared to autonomous manmade machines. Hence, disclosing computa-
tional design principles of insect brains down to the level of neurons and neural
networks is not only one of the most fascinating missions of basic science, but also
a strategic goal if robots with autonomous behavior are to be realized (see Chapter
8). To test the viability of biologically established computational principles, but also
to translate these principles into a language that can be implemented in artificial
systems, modeling is an indispensable tool. Indeed, in all laboratories investigating
blowfly vision, modeling was intensively employed in parallel with experimental
analysis at all levels, ranging from computations of single cells to overall behavioral
performance (reviews: Borst, 2003; Egelhaaf et al., 2003).
Finally, it should be mentioned that because of the efficiency of visually guided
orientation behavior in insects, there is great interest in applying principles of insect
motion information processing to autonomous artificial systems (see Chapter 8).
Although this has been successful for some behavioral components (Franz and
Mallot, 2000; Harrison and Koch, 2000; Huber et al., 1999; Rind, 2002; Srinivasan
et al., 1997, 2001), biomorphic autonomous robots still appear to be dull compared
with the original after which they are modeled. In contrast to manmade systems,
natural vision systems have been tested and improved on a much longer timescale
by many millions of years of evolution.
ACKNOWLEDGMENTS
The work in the authors’ laboratory is supported by the Deutsche Forschungsge-
meinschaft (DFG).
REFERENCES
Abbott LF, Dayan P (1999) The effect of correlated variability on the accuracy of a population
code. Neural Computation 11:91–101.
2024_C007.fm Page 201 Monday, August 16, 2004 8:54 AM

202 New Frontiers in Insect Neuroscience
Barberini CL, Horwitz GD, Newsome WT (2000) A comparison of spiking statistics in motion
sensing neurons of flies and monkeys. In: Computational, Neural and Ecological
Constraints of Visual Motion Processing. Zanker JM, Zeil J, Eds., Springer, Heidel-
berg, pp. 307–320.
Bendat JS, Piersol AG (2000) Random Data: Analysis and Measurement Procedures. Wiley-
Interscience, New York.
Bialek W, Rieke F, de Ruyter van Steveninck R, Warland D (1991) Reading a neural code.
Science 252:1854–1857.
Boeddeker N, Egelhaaf M (2003a) Steering a model fly: simulations on visual pursuit in
blowflies. Proc. R. Soc. Lond. B 270:1971–1978.
Boeddeker N, Egelhaaf M (2003b) Chasing behaviour of blowflies: a smooth pursuit system
generates saccades. (submitted)
Boeddeker N, Kern R, Egelhaaf M (2003) Chasing a dummy target: smooth pursuit and
velocity control in male blowflies. Proc. R. Soc. Lond. B 270:393–399.
Borst A, Haag J (2001) Effects of mean firing on neural information rate. J. Comput. Neurosci.
10:213–221.
Borst A, Haag J (2002) Neural networks in the cockpit of the fly. J. Comp. Physiol. A
188:419–437.
Borst A, Theunissen FE (1999) Information theory and neural coding. Nature Neurosci.
2:947–957.
Collett TS (1971) Visual neurones for tracking moving targets. Nature 232:127–130.
Collett TS, Land MF (1975) Visual control of flight behaviour in the hoverfly Syritta pipiens
L. J. Comp. Physiol. 99:1–66.
Dayan P, Abbott LF (2001) Theoretical Neuroscience: Computational and Mathematical
Modeling of Neural Systems. MIT Press, Cambridge, MA.
Egelhaaf M (1985a) On the neuronal basis of figure–ground discrimination by relative motion
in the visual system of the fly. II. Figure-detection cells, a new class of visual
interneurones. Biol. Cybern. 52:195–209.
Egelhaaf M (1985b) On the neuronal basis of figure–ground discrimination by relative motion
in the visual system of the fly. III. Possible input circuitries and behavioural signif-
icance of the FD-cells. Biol. Cybern. 52:267–280.
Egelhaaf M, Borst A, Warzecha A-K, Flecks S, Wildemann A (1993) Neural circuit tuning
fly visual interneurons to motion of small objects. II. Input organization of inhibitory
circuit elements by electrophysiological and optical recording techniques. J. Neuro-
physiol. 69:340–351.
Egelhaaf M, Kern R (2002) Vision in flying insects. Curr. Opin. Neurobiol. 12:699–706.
Egelhaaf M, Kern R, Kurtz R, Krapp HG, Kretzberg J, Warzecha A-K (2002) Neural encoding
of behaviourally relevant motion information in the fly. Trends Neurosci. 25:96–102.
Egelhaaf M, Reichardt W (1987) Dynamic response properties of movement detectors: the-
oretical analysis and electrophysiological investigation in the visual system of the fly.
Biol. Cybern. 56:69–87.
Egelhaaf M, Warzecha A-K (1999) Encoding of motion in real time by the fly visual system.
Curr. Opin. Neurobiol. 9:454–460.
Esch HE, Burns JM (1996) Distance estimation by foraging honeybees. J. Exp. Biol.
199:155–162.
Esch HE, Zhang S, Srinivasan MV, Tautz J (2001) Honeybee dances communicate distances
measured by optic flow. Nature 411:581–583.
Farina WM, Varjú D, Zhou Y (1994) The regulation of distance to dummy flowers during
hovering flight in the hawk moth Macroglossum stellatarum. J. Comp. Physiol.
174:239–247.
AU: Pls
update.
2024_C007.fm Page 202 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects 203
Farrow K, Haag J, Borst A (2003) Input organization of multifunctional motion-sensitive
neurons in the blowfly. J. Neurosci. 29:9805–9811.
Fortune ES, Rose GJ (2001) Short-term synaptic plasticity as a temporal filter. Trends Neu-
rosci. 24:381–385.
Franz MO, Mallot HA (2000) Biomimetic robot navigation. Robots and Autonomous Systems
133–153. [volume #?]
Gauck V, Borst A (1999) Spatial response properties of contralateral inhibited lobula plate
tangential cells in the fly visual system. J. Comp. Neurol. 406:51–71.
Gilbert C, Strausfeld NJ (1991) The functional organization of male-specific visual neurons
in flies. J. Comp. Physiol. A 169:395–411.
Gray JR, Pawlowski VM, Willis MA (2002) A method for recording behavior and multineu-
ronal CNS activity from tethered insects flying in virtual space. J. Neurosci. Meth.
120:211–223.
Grewe J, Kretzberg J, Warzecha A-K, Egelhaaf M (2003) Impact of photon-noise on the
reliability of a motion-sensitive neuron in the fly’s visual system. J. Neurosci.
23:10776–10783.
Haag J, Borst A (1997) Encoding of visual motion information and reliability in spiking and
graded potential neurons. J. Neurosci. 17:4809–4819.
Haag J, Borst A (1998) Active membrane properties and signal encoding in graded potential
neurons. J. Neurosci. 18:7972–7986.
Haag J, Borst A (2001) Recurrent network interactions underlying flow-field selectivity of
visual interneurons. J. Neurosci. 21:5685–5692.
Haag J, Borst A (2002) Dendro-dendritic interactions between motion-sensitive large-field
neurons in the fly. J. Neurosci. 22:3227–3233.
Haag J, Borst A (2003) Orientation tuning of motion-sensitive neurons shaped by vertical-
horizontal network interactions. J. Comp. Physiol. A 189:363–370.
Harrison RR, Koch C (2000) A silicon implementation of the fly’s optomotor control system.
Neural Comput. 12:2291–2304.
Hateren JH van, Schilstra C (1999) Blowfly flight and optic flow. II. Head movements during
flight. J. Exp. Biol. 202:1491–1500.
Hateren JH van, Snippe HP (2001) Information theoretical evaluation of parametric models
of gain control in blowfly photoreceptor cells. Vision Res. 41:1851–1865.
Hausen K (1981) Monocular and binocular computation of motion in the lobula plate of the
fly. Verh. Dtsch. Zool. Ges. 74:49–70.
Hausen K, Egelhaaf M (1989) Neural mechanisms of visual course control in insects. In:
Facets of Vision. Stavenga D, Hardie RC, Eds., Springer, Berlin, pp. 391–424.
Hengstenberg R (1982) Common visual response properties of giant vertical cells in the lobula
plate of the blowfly Calliphora. J. Comp. Physiol. 149:179–193.
Horn E, Mittag J (1980) Body movements and retinal pattern displacements while approaching
a stationary object in the walking fly, Calliphora erythrocephala. Biol. Cybern.
39:67–77.
Horstmann W, Egelhaaf M, Warzecha A-K (2000) Synaptic interactions increase optic flow
specificity. Eur. J. Neurosci. 12:2157–2165.
Huber SA, Franz MO, Bülthoff HH (1999) On robots and flies: modeling the visual orientation
behavior of flies. Robotics and Autonomous Systems 29:227–242.
Juusola M, French AS, Uusitalo RO, Weckström M (1996) Information processing by graded-
potential transmission through tonically active synapses. Trends Neurosci.
19:292–297.
AU: Pls sup-
ply.
2024_C007.fm Page 203 Monday, August 16, 2004 8:54 AM

204 New Frontiers in Insect Neuroscience
Kalb J, Nielsen T, Fricke N, Egelhaaf M, Kurtz R (2004) In vivo two-photon laser-scanning
microscopy of Ca2+ dynamics in visual motion-sensitive neurons. Biochem. Biophys.
Res. Communic. 316:341–347.
Karmeier K, Egelhaaf M, Krapp HG (2001) Early visual experience and receptive field
organization of the optic flow processing interneurons in the fly motion pathway. Vis.
Neurosci. 18:1–8.
Karmeier K, Krapp HG, Egelhaaf M (2003) Robustness of the tuning of fly visual interneurons
to rotatory optic flow. J. Neurophysiol. 90:1626–1634.
Kern R, van Hateren JH, Michaelis C, Lindemann JP, Egelhaaf M (2004a) Eye movements
during natural flight shape the function of a blowfly motion sensitive neuron. (sub-
mitted).
Kern R, Lutterklas M, Egelhaaf M (2000) Neural representation of optic flow experienced
by unilaterally blinded flies on their mean walking trajectories. J. Comp. Physiol. A
186:467–479.
Kern R, Lutterklas M, Petereit C, Lindemann JP, Egelhaaf M (2001a) Neuronal processing
of behaviourally generated optic flow: experiments and model simulations. Network:
Computations in Neural Systems 12:351–369.
Kern R, Oddos F, Schwerdtfeger G, Egelhaaf M (2004b) Performance of fly H1 neuron during
free-flight. (submitted).
Kern R, Petereit C, Egelhaaf M (2001b) Neural processing of naturalistic optic flow. J.
Neurosci. 21:1–5.
Kern R, Varjú D (1998) Visual position stabilization in the hummingbird hawk moth, Mac-
roglossum stellatarum L.: I. Behavioural analysis. J. Comp. Physiol. A 182:225–237.
Kimmerle B, Egelhaaf M (2000a) Detection of object motion by a fly neuron during simulated
translatory flight. J. Comp. Physiol. A 186:21–31.
Kimmerle B, Egelhaaf M (2000b) Performance of fly visual interneurons during object
fixation. J. Neurosci. 20:6256–6266.
Koenderink JJ (1986) Optic flow. Vis. Res. 26:161–180.
Krapp HG (2000) Neuronal matched filters for optic flow processing in flying insects. In:
Neuronal Processing of Optic Flow. Lappe M, Ed., Academic Press, San Diego, CA,
pp. 93–120.
Krapp HG, Hengstenberg B, Hengstenberg R (1998) Dendritic structure and receptive-field
organization of optic flow processing interneurons in the fly. J. Neurophysiol.
79:1902–1917.
Krapp HG, Hengstenberg R (1996) Estimation of self-motion by optic flow processing in
single visual interneurons. Nature 384:463–466.
Krapp HG, Hengstenberg R, Egelhaaf M (2001) Binocular contribution to optic flow pro-
cessing in the fly visual system. J. Neurophysiol. 85:724–734.
Kretzberg J, Egelhaaf M, Warzecha A-K (2001a) Membrane potential fluctuations determine
the precision of spike timing and synchronous activity: a model study. J. Comput.
Neurosci. 10:79–97.
Kretzberg J, Warzecha A-K, Egelhaaf M (2001b) Neural coding with graded membrane
potential changes and spikes. J. Comput. Neurosci. 11:153–164.
Kurtz R (2004) Ca2+ clearance in visual motion-sensitive neurons of the fly studied in vivo
by sensory stimulation and UV photolysis of caged Ca2+. (submitted)
Kurtz R, Egelhaaf M (2003) Natural patterns of neural activity. Molecular Neurobiol. 27:1–19.
Kurtz R, Warzecha A-K, Egelhaaf M (2001) Transfer of visual information via graded
synapses operates linearly in the natural activity range. J. Neurosci. 21:6957–6966.
Land MF, Collett TS (1974) Chasing behaviour of houseflies (Fannia canicularis): a descrip-
tion and analysis. J. Comp. Physiol. 89:331–357.
AU: Pls
update.
AU: Pls
update.
AU: Pls
update.
2024_C007.fm Page 204 Monday, August 16, 2004 8:54 AM

Novel Approaches to Visual Information Processing in Insects 205
Lappe M (2000) Neuronal Processing of Optic Flow. Academic Press, San Diego, CA.
Lewen GD, Bialek W, de Ruyter van Steveninck R (2001) Neural coding of naturalistic
stimuli. Network: Comput. Neural Syst. 12:317–329.
Lindemann JP, Kern R, Michaelis C, Meyer P, van Hateren JH, Egelhaaf M (2003) FliMax,
a novel stimulus device for panoramic and high-speed presentation of behaviourally
generated optic flow. Vision Res. 43:779–791.
Marmarelis PZ, Marmarelis VZ (1978) Analysis of Physiological Systems. Plenum Press, New
York.
Miller JP, Selverston AI (1979) Rapid killing of single neurons by irradiation of intracellularly
injected dye. Science 206:702–704.
Neher E (1998) Vesicle pools and Ca2+ microdomains: new tools for understanding their roles
in neurotransmitter release. Neuron 20:389–399.
Oddos F, Kern R, Boeddeker N, Egelhaaf M (2003) Flight performance modified by envi-
ronmental changes in the blowfly Lucilia. In: Göttingen Neurobiology Report. Elsner
N, Ed., Georg Thieme, Stuttgart, p. 444.
Olberg RM (1981) Object- and self-movement detectors in the ventral nerve cord of the
dragonfly. J. Comp. Physiol. 141:327–334.
Olberg RM (1986) Identified target-selective visual interneurons descending from the drag-
onfly brain. J. Comp. Physiol. 159:827–840.
Olberg RM, Pinter RB (1990) The effect of mean luminance on the size selectivity of identified
target interneurons in the dragonfly. J. Comp. Physiol. A 166:851–856.
Oram MW, Földiák P, Perrett DI, Sengpiel F (1998) The “ideal homunculus”: decoding neural
population signals. Trends Neurosci. 21:259–265.
Paulsen O, Sejnowski TJ (2000) Natural patterns of activity and long-term synaptic plasticity.
Curr. Opin. Neurobiol. 10:172–179.
Pouget A, Dayan P, Zemel RS (2003) Inference and computation with population codes. Ann.
Rev. Neurosci. 26:381–410.
Pouget A, Deneve S, Ducom J-C, Latham PE (1999) Narrow versus wide tuning curves:
what’s best for a population code? Neural Comput. 11:85–90.
Pouget A, Zhang K, Deneve S, Latham PE (1998) Statistically efficient estimation using
population coding. Neural Comput. 10:373–401.
Rieke F, Warland D, de Ruyter van Steveninck R, Bialek W (1997) Spikes: Exploring the
Neural Code. MIT Press, Cambridge, MA.
Rind FC (2002) Motion detectors in the locust visual system: from biology to robot sensors.
Microsc. Res. Tech. 56:256–269.
Rind FC, Simmons PJ (1999) Seeing what is coming: building collision-sensitive neurones.
Trends Neurosci. 22:215–220.
Ruyter van Steveninck R de, Lewen GD, Strong SP, Koberle R, Bialek W (1997) Reproduc-
ibility and variability in neural spike trains. Science 275:1805–1808.
Sabatini BL, Regehr WG (1999) Timing of synaptic transmission. Annu. Rev. Physiol.
61:521–542.
Schilstra C, van Hateren JH (1999) Blowfly flight and optic flow. I. Thorax kinematics and
flight dynamics. J. Exp. Biol. 202:1481–1490.
Simmons, PJ (2002) Signal processing in a simple visual system: the locust ocellar system
and its synapses. Microsc. Res. Tech. 56:270–280.
Srinivasan MV, Chahl JS, Zhang SW (1997) Embodying natural vision into machines. In:
From Living Eyes to Seeing Machines. Srinivasan MV, Venkatesh S, Eds., Oxford
University Press, Oxford, pp. 249–265.
Srinivasan MV, Zhang S, Altwein M, Tautz J (2000) Honeybee navigation: nature and cali-
bration of the “odometer.” Science 287:851–853.
2024_C007.fm Page 205 Monday, August 16, 2004 8:54 AM

206 New Frontiers in Insect Neuroscience
Srinivasan MV, Zhang S, Chahl JS (2001) Landing strategies in honeybees, and possible
applications to autonomous airborne vehicles. Biol. Bull. 200:216–221.
Thomson AM (2000) Facilitation, augmentation and potentiation at central synapses. Trends
Neurosci. 23:305–312.
Victor JD, Purpura K (1996) Nature and precision of temporal coding in visual cortex: a
metric-space analysis. J. Neurophysiol. 76:1310–1326.
Wagner H (1986) Flight performance and visual control of the flight of the free-flying housefly
(Musca domestica). II. Pursuit of targets. Phil. Trans. R. Soc. Lond. B 312:553–579.
Warzecha A-K, Egelhaaf M (1996) Intrinsic properties of biological motion detectors prevent
the optomotor control system from getting unstable. Phil. Trans. R. Soc. Lond. B
351:1579–1591.
Warzecha A-K, Egelhaaf M (1997) How reliably does a neuron in the visual motion pathway
of the fly encode behaviourally relevant information? Europ. J. Neurosci.
9:1365–1374.
Warzecha A-K, Egelhaaf M (1999) Variability in spike trains during constant and dynamic
stimulation. Science 283:1927–1930.
Warzecha A-K, Egelhaaf M, Borst A (1993) Neural circuit tuning fly visual interneurons to
motion of small objects. I. Dissection of the circuit by pharmacological and photo-
inactivation techniques. J. Neurophysiol. 69:329–339.
Warzecha A-K, Kretzberg J, Egelhaaf M (1998) Temporal precision of the encoding of motion
information by visual interneurons. Curr. Biol. 8:359–368.
Warzecha A-K, Kretzberg J, Egelhaaf M (2000) Reliability of a fly motion-sensitive neuron
depends on stimulus parameters. J. Neurosci. 20:8886–8896.
Warzecha A-K, Kurtz R, Egelhaaf M (2003) Synaptic transfer of dynamical motion informa-
tion between identified neurons in the visual system of the blowfly. Neurosci.
119:1103–1112.
Wehner R (1981) Spatial vision in arthropods. In: Handbook of Sensory Physiology. Vol.
VII/6C: Comparative Physiology and Evolution of Vision in Invertebrates. Autrum
H, Ed., Springer, Berlin, pp. 287–616.
Wilke SD, Eurich CW (2001) Representational accuracy of stochastic neural populations.
Neural Comput. 14:155–189.
Zhang K, Sejnowski TJ (1999) Neuronal tuning: to sharpen or broaden? Neural Comput.
11:75–84.
2024_C007.fm Page 206 Monday, August 16, 2004 8:54 AM