- Email: [email protected]
Visual motion: Dendritic integration makes sense of the world
Dispatch
R15
Visual motion: Dendritic integration makes sense of the world Simon B. Laughlin
Flies use a system of specialised neurons to read the patterns of visual motion – optic flow — induced by the their movements. Recent experiments illustrate how the dendrites of these neurons reach out to assemble patterns of optic flow and encode them reliably. Address: Department of Zoology, University of Cambridge, Downing Street, Cambridge CB2 3EJ, UK. Current Biology 1999, 9:R15–R17 http://biomednet.com/elecref/09609822009R0015
scientists, Werner Reichardt and Bernhard Hassenstein, struggled to resume research amid the devastation in Germany. They used a very cheap preparation, a beetle walking on an ingenious Y-maze globe, made of straw. When Hassenstein and Reichardt rotated stripes around the beetle, it turned the Y-maze globe to the left or the right, to follow the movement. By varying the stripes and measuring the effect on the beetle’s turning tendency, Hassenstein and Reichardt [3] were able to deduce the simple algorithm used to detect motion.
© Elsevier Science Ltd ISSN 0960-9822
How does a visual system make sense of the world? One strategy is to determine local cues, such as edges and movements, and then piece these cues together to establish a particular pattern, such as a shape or an object’s trajectory. The local cues can be detected early on in visual processing, by arrays of elementary detectors, spread across the visual field. Each elementary detector signals the presence of its cue within its small patch of the field, and takes little notice of what is going on around it. The bigger picture is assembled by neurons at higher levels which, by collecting and integrating inputs from specific sets of elementary detectors, determine the presence of a pattern. Two recent studies of large, motion-sensitive neurons in the blowfly brain [1,2] show how a neuron reaches out with its dendrites to collect local cues and sense pattern. The motion-sensitive pathways of the blowfly’s visual system provide an exceptional opportunity to study the neural basis of pattern recognition. This opportunity was created by a research programme that started almost 50 years ago. Shortly after the Second World War, two young
In the model developed by Hassenstein and Reichardt, ‘elementary motion detectors’ (EMDs) inside the beetle’s brain compare inputs from neighbouring sampling stations — adjacent ommatidia on the insect’s compound eye. An EMD delays the input from one ommatidium and multiplies it with the input from the neighbour (Figure 1). When something moves from one ommatidium to the neighbour in a time equal to the delay, two identical signals will be multiplied by the EMD. Multiplication produces a positive output that reflects the correlation between the signals at the two stations. This ‘Reichardt correlator’, arguably the first, simplest and most robust of the EMDs known, is thought to operate in our own visual system. Their success launched Hassenstein and Reichardt on distinguished careers. Reichardt established the MaxPlanck-Institute for Biological Cybernetics in Tübingen. With Valentino Braitenberg, Karl Götz and Kuno Kirschfeld, Reichardt initiated a powerful four-pronged attack on the problem of visual pattern recognition [4]. Using flies, the Tübingen group combined studies of behaviour, anatomy, physiology and computation in a
Figure 1 The basic unit of the Reichardt correlator receives inputs from two adjacent ommatidia in the compound eye. The correlator delays the signal from ommatidium 1 and multiplies it by the signal from ommatidium 2 to produce a positive output in response to image movement from 1 to 2. This is illustrated with the signals produced by the image of a small dark target, moving from ommatidium 1 to ommatidium 2 in time ∆t. For simplicity, ∆t happens to equal the delay introduced by the correlator. Thus, the delay brings the two signals into register and, when multiplied together, a large positive output is produced. The pre-filters (PFs) set the signal to a baseline of zero. Note that the correlator responds only to local motion from 1 to 2. (See [15] for further details.)
Ommatidia Pre-filters
1
PF
2
PF
EMD
Delay ∆t
∆t
Current Biology
R16
Current Biology, Vol 9 No 1
Figure 2
EMD array
Lobula plate Top
Receptive field
Top VS7
Front
Back
Bottom
150 µm
Front
Back
Bottom Current Biology
The dendritic tree and receptive field of the large motion-sensitive neuron VS7 in the blowfly lobula plate. Inputs from the array of elementary motion detectors (EMDs) project onto the lobula plate to form an orderly spatial map of the visual field (the topography of the projection has been simplified for this diagram). The dendrites of VS7 collect signals from EMDs in a vertical strip of the lobula plate. They select EMDs that respond to downwards movement to produce a receptive field that is dominated by high sensitivity to downwards motion in a vertical strip of visual space, to the side and slightly to the
back of the fly. Moving away from this strip, sensitivity falls (the arrows get shorter) and directional tuning curls to follow the pattern of optic flow induced by roll about a horizontal axis. The change in directional tuning is apparently achieved by weighting and summing inputs from EMDs that respond to movements along different axes of the facet mosaic of the compound eye [1]. Calcium imaging (not illustrated) shows that independent signals are injected into dendrites by EMDs, and that these sum to generate a common output [2]. (Diagram modified from [1].)
single tractable system. Quantitative behavioural analysis defined what flies saw. Definitive neuroanatomy and the detailed physiological analysis of photoreceptors and neural circuits revealed the means by which information was processed. And computational models established that the anatomical and physiological means achieved the behavioural end, a superb visual ability. The group’s description of neural circuitry, optics, physiology and behaviour made flies a productive model system for studying the neural mechanisms of vision. Among many notable discoveries, was the identification of a set of neurons that process a visual pattern called optic flow [5].
lobula plate, an area that is four synapses removed from photoreceptors and is the fly’s equivalent of area MST in the primate visual cortex [7].
Optic flow is a pattern of movement, generated by movement of the observer, that extends across the entire visual field. You can see for yourself the relationships between optic flow and the observer’s movements. Turn right and objects flow to the left; dip one’s head, and the scenery flows upwards; move forwards and things flow radially outward from a focus at one’s apparent destination. So informative are these patterns, that flight simulators create the vivid impression that you are steering a flying aircraft by generating flow fields on your computer screen. Flies, like pilots and cyclists, read patterns of optic flow quickly and accurately to deduce their trajectory and avoid crashing [6]. They use a system of neurons whose large size (Figure 2) promotes a rapid and reliable response, and favours experimental analysis. These neurons sit in the
The large, motion-sensitive neurons of the lobula plate detect patterns of optic flow by collecting signals from EMDs. Centred on every ommatidium of the compound eye is a group of Reichardt correlators [8]. Each correlator senses movement from the central ommatidium to a particular neighbour [9]. By covering all directions, the group of correlators is able to capture the position, direction and magnitude of local motion at their ommatidium. The group of correlators is replicated at all ommatidia, generating a field of detectors that can capture the essential cues — the directions of local motion — at all points on the image. Hausen [10] found that the large neurons of the lobula plate reach out with their dendrites to collect inputs from specific subsets of correlators. These dendrites define the position of the neuron’s receptive field and, by selecting inputs from correlators that respond to motion in a particular direction, the dendrites determine the neuron’s directional tuning (Figure 2). Two recent papers [1,2] from groups in Tübingen have examined the dendritic integration of local cues in remarkable detail. Krapp et al. [1] used rotating spots to make a detailed map of the receptive fields (Figure 2) of the large motion-sensitive neurons. Every neuron that
Dispatch
they mapped was injected with dye to visualise its morphology. This allowed them to correlate the fine structure of the receptive field with the shape of the dendritic tree. As expected, the projection of the dendritic tree into the field of EMDs defined the region of the visual field within which the neuron was most sensitive to motion. There are, however, weaker inputs from areas beyond that defined by the EMDs that the dendrites directly synapse with. These signals must be fed through from correlators indirectly, by other neurons. The most remarkable finding is that the directional sensitivity of a large motion-sensitive neuron changes within its receptive field, to follow the whorls and eddies of optic flow (Figure 2). Every one of the ten vertical neurons in the blowfly’s lobula plate is a matched filter, tuned to the pattern of optic flow induced by rotation about a particular horizontal axis. This fine tuning of receptive fields exemplifies a dendrite’s ability to collect and weight specific combinations of inputs from EMDs. Krapp et al. [1] promise a theoretical analysis that demonstrates how the patterns read by this array of matched filters signal deviations in flight. We await the outcome with interest.
R17
in the lobula plate, is a direct demonstration that a neuron uses its dendrites to collect local cues and averages them, according to Eccle’s principle of spatial summation on the neural membrane, to produce a reliable indication of pattern motion. These revelations of dendrites at work illustrate the advantages of working with an accessible set of identified neurons, of known structure and function. Neurobiologists will continue to build on the exceptional foundations laid by Reichardt’s Tübingen group, and address points of principle that are made apparent in the blowfly brain. Sophisticated mathematical techniques are being used to measure the quantities of information coded by these motion-sensitive neurons [11–13]. Here is the opportunity to probe deeply into neural design by relating dendritic mechanisms to the efficiency with which neurons do their job. Brave souls are exploring the tiny neurons that feed the lobula plate, to find the neural mechanisms responsible for elementary motion detection [14]. Such opportunities are a testament to Reichardt, and to the neurobiologists who have followed the lead of his Max-Planck-Institute. References
In the second paper, Single and Borst [2] have observed the signals collected by the different dendrites and described their summation in the neuron. They have been able to visualise the spatial distribution of dendritic signals, because the large motion-sensitive neurons are flat and lie close to the surface of the brain. On a good day, neurons can be impaled with microelectrodes, injected with calcium indicator dyes, anatomically identified, observed with a calcium imaging microscope, and have their electrical signals recorded: all in a living fly watching movies on a screen. Single and Borst [2] established that the neuron’s intracellular calcium concentration increases linearly with membrane potential. Consequently, by measuring calcium concentrations throughout the cell, calcium imaging allowed them to simultaneously monitor voltage signals in different regions. As stripes excite a neuron by moving across its receptive field, a flickering pattern of membrane potential passes across the dendritic tree. This flicker is produced by small groups of EMDs. Because a Reichardt correlator multiplies neighbouring inputs, its response rises and falls as stripes pass across its small patch of the visual field. These local fluctuations tend to obscure the steady output that reliably indicates the local velocity. The signal in each dendrite fluctuates out of phase with its neighbours, because its correlators are looking at a different part of the stripe pattern. A computer model of the neuron, based on its anatomical and electrical properties, predicts that the local fluctuations cancel when the electrical inputs from many dendrites are combined. Calcium imaging confirmed the modelling; the output signal observed in the neuron’s axon was found to be smooth and reliable. Here,
1. Krapp HG, Hengstenberg B, Hengstenberg R: Dendritic structure and receptive-field organization of optic flow processing interneurons in the fly. J Neurophysiol 1998, 79:1902-1917. 2. Single S, Borst A: Dendritic integration and its role in computing image velocity. Science 1998, 281:848-1850. 3. Reichardt W: Autocorrelation, a principle for the evaluation of sensory information in the central nervous system. In Sensory Communication. Edited by Rosenblith WA. Cambridge, Mass: MIT Press; 1961:303-317. 4. Reichardt W: The insect eye as a model system for the study of the uptake, transduction and processing of optical data in the nervous system. In The Neurosciences: Second Study Programme. Edited by Schmitt FO. New York: Rockefeller University Press; 1970. 5. Hausen K, Egelhaaf M: Neural mechanisms of visual course control in insects. In Facets of Vision. Edited by Stavenga DG, Hardie RC. Berlin: Springer; 1989:391-424. 6. Hengstenberg R, Sandeman DC, Hengstenberg B: Compensatory head roll in the blowfly Calliphora during flight. Proc R Soc Lond [Biol] 1986, 227:455-482. 7. Wurtz RH: Optic flow: A brain region devoted to optic flow analysis? Curr Biol 1998, 8:R554-R556. 8. Reichardt W: Evaluation of optical motion information by movement detectors. J Comp Physiol 1987, A161:533-547. 9. Schuling FH, Masterbroek HAK, Bult R, Lenting BPM: Properties of elementary movement detectors in the fly Calliphora erythrocephala. J Comp Physiol 1989, A165:179-192. 10. Hausen K: Motion sensitive interneurons in the optomotor system of the fly .2. the horizontal cells – receptive-field organization and response characteristics. Biol Cybernetics 1982, 46:67-79. 11. de Ruyter van Steveninck RR, Bialek W: Real-time performance of a movement-sensitive neuron in the blowfly visual-system – coding and information-transfer in short spike sequences. Proc R Soc Lond [Biol] 1988, 234:379-414. 12. Rieke F, Warland D, de Ruyter van Steveninck RD, Bialek W: Spikes — Exploring the Neural Code. Cambridge, Mass: MIT Press; 1997. 13. Warzecha AK, Kretzberg J, Egelhaaf M: Temporal precision of the encoding of motion information by visual interneurons. Curr Biol 1998, 8:359-368. 14. Douglass JK, Strausfeld NJ: Visual motion-detection circuits in flies — parallel direction-sensitive and non-direction-sensitive pathways between the medulla and lobula plate. J Neurosci 1996, 16:4551-4562. 15. Borst A, Egelhaaf M: Principles of visual-motion detection. Trends Neurosci 1989, 12:297-306.
R15
Visual motion: Dendritic integration makes sense of the world Simon B. Laughlin
Flies use a system of specialised neurons to read the patterns of visual motion – optic flow — induced by the their movements. Recent experiments illustrate how the dendrites of these neurons reach out to assemble patterns of optic flow and encode them reliably. Address: Department of Zoology, University of Cambridge, Downing Street, Cambridge CB2 3EJ, UK. Current Biology 1999, 9:R15–R17 http://biomednet.com/elecref/09609822009R0015
scientists, Werner Reichardt and Bernhard Hassenstein, struggled to resume research amid the devastation in Germany. They used a very cheap preparation, a beetle walking on an ingenious Y-maze globe, made of straw. When Hassenstein and Reichardt rotated stripes around the beetle, it turned the Y-maze globe to the left or the right, to follow the movement. By varying the stripes and measuring the effect on the beetle’s turning tendency, Hassenstein and Reichardt [3] were able to deduce the simple algorithm used to detect motion.
© Elsevier Science Ltd ISSN 0960-9822
How does a visual system make sense of the world? One strategy is to determine local cues, such as edges and movements, and then piece these cues together to establish a particular pattern, such as a shape or an object’s trajectory. The local cues can be detected early on in visual processing, by arrays of elementary detectors, spread across the visual field. Each elementary detector signals the presence of its cue within its small patch of the field, and takes little notice of what is going on around it. The bigger picture is assembled by neurons at higher levels which, by collecting and integrating inputs from specific sets of elementary detectors, determine the presence of a pattern. Two recent studies of large, motion-sensitive neurons in the blowfly brain [1,2] show how a neuron reaches out with its dendrites to collect local cues and sense pattern. The motion-sensitive pathways of the blowfly’s visual system provide an exceptional opportunity to study the neural basis of pattern recognition. This opportunity was created by a research programme that started almost 50 years ago. Shortly after the Second World War, two young
In the model developed by Hassenstein and Reichardt, ‘elementary motion detectors’ (EMDs) inside the beetle’s brain compare inputs from neighbouring sampling stations — adjacent ommatidia on the insect’s compound eye. An EMD delays the input from one ommatidium and multiplies it with the input from the neighbour (Figure 1). When something moves from one ommatidium to the neighbour in a time equal to the delay, two identical signals will be multiplied by the EMD. Multiplication produces a positive output that reflects the correlation between the signals at the two stations. This ‘Reichardt correlator’, arguably the first, simplest and most robust of the EMDs known, is thought to operate in our own visual system. Their success launched Hassenstein and Reichardt on distinguished careers. Reichardt established the MaxPlanck-Institute for Biological Cybernetics in Tübingen. With Valentino Braitenberg, Karl Götz and Kuno Kirschfeld, Reichardt initiated a powerful four-pronged attack on the problem of visual pattern recognition [4]. Using flies, the Tübingen group combined studies of behaviour, anatomy, physiology and computation in a
Figure 1 The basic unit of the Reichardt correlator receives inputs from two adjacent ommatidia in the compound eye. The correlator delays the signal from ommatidium 1 and multiplies it by the signal from ommatidium 2 to produce a positive output in response to image movement from 1 to 2. This is illustrated with the signals produced by the image of a small dark target, moving from ommatidium 1 to ommatidium 2 in time ∆t. For simplicity, ∆t happens to equal the delay introduced by the correlator. Thus, the delay brings the two signals into register and, when multiplied together, a large positive output is produced. The pre-filters (PFs) set the signal to a baseline of zero. Note that the correlator responds only to local motion from 1 to 2. (See [15] for further details.)
Ommatidia Pre-filters
1
PF
2
PF
EMD
Delay ∆t
∆t
Current Biology
R16
Current Biology, Vol 9 No 1
Figure 2
EMD array
Lobula plate Top
Receptive field
Top VS7
Front
Back
Bottom
150 µm
Front
Back
Bottom Current Biology
The dendritic tree and receptive field of the large motion-sensitive neuron VS7 in the blowfly lobula plate. Inputs from the array of elementary motion detectors (EMDs) project onto the lobula plate to form an orderly spatial map of the visual field (the topography of the projection has been simplified for this diagram). The dendrites of VS7 collect signals from EMDs in a vertical strip of the lobula plate. They select EMDs that respond to downwards movement to produce a receptive field that is dominated by high sensitivity to downwards motion in a vertical strip of visual space, to the side and slightly to the
back of the fly. Moving away from this strip, sensitivity falls (the arrows get shorter) and directional tuning curls to follow the pattern of optic flow induced by roll about a horizontal axis. The change in directional tuning is apparently achieved by weighting and summing inputs from EMDs that respond to movements along different axes of the facet mosaic of the compound eye [1]. Calcium imaging (not illustrated) shows that independent signals are injected into dendrites by EMDs, and that these sum to generate a common output [2]. (Diagram modified from [1].)
single tractable system. Quantitative behavioural analysis defined what flies saw. Definitive neuroanatomy and the detailed physiological analysis of photoreceptors and neural circuits revealed the means by which information was processed. And computational models established that the anatomical and physiological means achieved the behavioural end, a superb visual ability. The group’s description of neural circuitry, optics, physiology and behaviour made flies a productive model system for studying the neural mechanisms of vision. Among many notable discoveries, was the identification of a set of neurons that process a visual pattern called optic flow [5].
lobula plate, an area that is four synapses removed from photoreceptors and is the fly’s equivalent of area MST in the primate visual cortex [7].
Optic flow is a pattern of movement, generated by movement of the observer, that extends across the entire visual field. You can see for yourself the relationships between optic flow and the observer’s movements. Turn right and objects flow to the left; dip one’s head, and the scenery flows upwards; move forwards and things flow radially outward from a focus at one’s apparent destination. So informative are these patterns, that flight simulators create the vivid impression that you are steering a flying aircraft by generating flow fields on your computer screen. Flies, like pilots and cyclists, read patterns of optic flow quickly and accurately to deduce their trajectory and avoid crashing [6]. They use a system of neurons whose large size (Figure 2) promotes a rapid and reliable response, and favours experimental analysis. These neurons sit in the
The large, motion-sensitive neurons of the lobula plate detect patterns of optic flow by collecting signals from EMDs. Centred on every ommatidium of the compound eye is a group of Reichardt correlators [8]. Each correlator senses movement from the central ommatidium to a particular neighbour [9]. By covering all directions, the group of correlators is able to capture the position, direction and magnitude of local motion at their ommatidium. The group of correlators is replicated at all ommatidia, generating a field of detectors that can capture the essential cues — the directions of local motion — at all points on the image. Hausen [10] found that the large neurons of the lobula plate reach out with their dendrites to collect inputs from specific subsets of correlators. These dendrites define the position of the neuron’s receptive field and, by selecting inputs from correlators that respond to motion in a particular direction, the dendrites determine the neuron’s directional tuning (Figure 2). Two recent papers [1,2] from groups in Tübingen have examined the dendritic integration of local cues in remarkable detail. Krapp et al. [1] used rotating spots to make a detailed map of the receptive fields (Figure 2) of the large motion-sensitive neurons. Every neuron that
Dispatch
they mapped was injected with dye to visualise its morphology. This allowed them to correlate the fine structure of the receptive field with the shape of the dendritic tree. As expected, the projection of the dendritic tree into the field of EMDs defined the region of the visual field within which the neuron was most sensitive to motion. There are, however, weaker inputs from areas beyond that defined by the EMDs that the dendrites directly synapse with. These signals must be fed through from correlators indirectly, by other neurons. The most remarkable finding is that the directional sensitivity of a large motion-sensitive neuron changes within its receptive field, to follow the whorls and eddies of optic flow (Figure 2). Every one of the ten vertical neurons in the blowfly’s lobula plate is a matched filter, tuned to the pattern of optic flow induced by rotation about a particular horizontal axis. This fine tuning of receptive fields exemplifies a dendrite’s ability to collect and weight specific combinations of inputs from EMDs. Krapp et al. [1] promise a theoretical analysis that demonstrates how the patterns read by this array of matched filters signal deviations in flight. We await the outcome with interest.
R17
in the lobula plate, is a direct demonstration that a neuron uses its dendrites to collect local cues and averages them, according to Eccle’s principle of spatial summation on the neural membrane, to produce a reliable indication of pattern motion. These revelations of dendrites at work illustrate the advantages of working with an accessible set of identified neurons, of known structure and function. Neurobiologists will continue to build on the exceptional foundations laid by Reichardt’s Tübingen group, and address points of principle that are made apparent in the blowfly brain. Sophisticated mathematical techniques are being used to measure the quantities of information coded by these motion-sensitive neurons [11–13]. Here is the opportunity to probe deeply into neural design by relating dendritic mechanisms to the efficiency with which neurons do their job. Brave souls are exploring the tiny neurons that feed the lobula plate, to find the neural mechanisms responsible for elementary motion detection [14]. Such opportunities are a testament to Reichardt, and to the neurobiologists who have followed the lead of his Max-Planck-Institute. References
In the second paper, Single and Borst [2] have observed the signals collected by the different dendrites and described their summation in the neuron. They have been able to visualise the spatial distribution of dendritic signals, because the large motion-sensitive neurons are flat and lie close to the surface of the brain. On a good day, neurons can be impaled with microelectrodes, injected with calcium indicator dyes, anatomically identified, observed with a calcium imaging microscope, and have their electrical signals recorded: all in a living fly watching movies on a screen. Single and Borst [2] established that the neuron’s intracellular calcium concentration increases linearly with membrane potential. Consequently, by measuring calcium concentrations throughout the cell, calcium imaging allowed them to simultaneously monitor voltage signals in different regions. As stripes excite a neuron by moving across its receptive field, a flickering pattern of membrane potential passes across the dendritic tree. This flicker is produced by small groups of EMDs. Because a Reichardt correlator multiplies neighbouring inputs, its response rises and falls as stripes pass across its small patch of the visual field. These local fluctuations tend to obscure the steady output that reliably indicates the local velocity. The signal in each dendrite fluctuates out of phase with its neighbours, because its correlators are looking at a different part of the stripe pattern. A computer model of the neuron, based on its anatomical and electrical properties, predicts that the local fluctuations cancel when the electrical inputs from many dendrites are combined. Calcium imaging confirmed the modelling; the output signal observed in the neuron’s axon was found to be smooth and reliable. Here,
1. Krapp HG, Hengstenberg B, Hengstenberg R: Dendritic structure and receptive-field organization of optic flow processing interneurons in the fly. J Neurophysiol 1998, 79:1902-1917. 2. Single S, Borst A: Dendritic integration and its role in computing image velocity. Science 1998, 281:848-1850. 3. Reichardt W: Autocorrelation, a principle for the evaluation of sensory information in the central nervous system. In Sensory Communication. Edited by Rosenblith WA. Cambridge, Mass: MIT Press; 1961:303-317. 4. Reichardt W: The insect eye as a model system for the study of the uptake, transduction and processing of optical data in the nervous system. In The Neurosciences: Second Study Programme. Edited by Schmitt FO. New York: Rockefeller University Press; 1970. 5. Hausen K, Egelhaaf M: Neural mechanisms of visual course control in insects. In Facets of Vision. Edited by Stavenga DG, Hardie RC. Berlin: Springer; 1989:391-424. 6. Hengstenberg R, Sandeman DC, Hengstenberg B: Compensatory head roll in the blowfly Calliphora during flight. Proc R Soc Lond [Biol] 1986, 227:455-482. 7. Wurtz RH: Optic flow: A brain region devoted to optic flow analysis? Curr Biol 1998, 8:R554-R556. 8. Reichardt W: Evaluation of optical motion information by movement detectors. J Comp Physiol 1987, A161:533-547. 9. Schuling FH, Masterbroek HAK, Bult R, Lenting BPM: Properties of elementary movement detectors in the fly Calliphora erythrocephala. J Comp Physiol 1989, A165:179-192. 10. Hausen K: Motion sensitive interneurons in the optomotor system of the fly .2. the horizontal cells – receptive-field organization and response characteristics. Biol Cybernetics 1982, 46:67-79. 11. de Ruyter van Steveninck RR, Bialek W: Real-time performance of a movement-sensitive neuron in the blowfly visual-system – coding and information-transfer in short spike sequences. Proc R Soc Lond [Biol] 1988, 234:379-414. 12. Rieke F, Warland D, de Ruyter van Steveninck RD, Bialek W: Spikes — Exploring the Neural Code. Cambridge, Mass: MIT Press; 1997. 13. Warzecha AK, Kretzberg J, Egelhaaf M: Temporal precision of the encoding of motion information by visual interneurons. Curr Biol 1998, 8:359-368. 14. Douglass JK, Strausfeld NJ: Visual motion-detection circuits in flies — parallel direction-sensitive and non-direction-sensitive pathways between the medulla and lobula plate. J Neurosci 1996, 16:4551-4562. 15. Borst A, Egelhaaf M: Principles of visual-motion detection. Trends Neurosci 1989, 12:297-306.