Fly motion vision and resultant compensatory optomotor responses are a classic example for neural computation. Here we review our current understanding of processing of optic flow as generated by an animal's self-motion. Optic flow processing is accomplished in a series of steps: First, the time-varying photoreceptor signals are fed into a two-dimensional array of Reichardt-type elementary motion detectors (EMDs). EMDs compute, in parallel, local motion vectors at each sampling point in space. Second, the output signals of many EMDs are spatially integrated on the dendrites of large-field tangential cells in the lobula plate. In the third step, tangential cells form extensive interactions with each other, giving rise to their large and complex receptive fields. Thus, tangential cells can act as matched filters tuned to optic flow during particular flight maneuvers. They finally distribute their information onto postsynaptic descending neurons, which either instruct the motor centers of the thoracic ganglion for flight and locomotion control or act themselves as motor neurons that control neck muscles for head movements.

Fly Motion Vision

Flies have been buzzing around on earth for over 300 million years. During this time they have radiated into more than 125,000 different species (Yeates and Wiegmann 1999), so that, by now, roughly every tenth described species is a fly. They thus represent one of the most successful animal groups on our planet. This evolutionary success might, at least in part, be a result of their acrobatic maneuverability, which enables them, for example, to chase mates at turning velocities of more than 3000° s–1 with delay times of less than 30 ms (Land and Collett 1974; Wagner 1986). It is this fantastic behavior, which has initiated much research during the last decades, both on its sensory control and the biophysical and aerodynamic principles of the flight output (Dickinson et al. 1999, 2000). Here, we review the current state of knowledge about the neural processing of visual motion, which represents one sensory component intimately involved in flight control. Other reviews on this topic have been published with a similar (Hausen 1981, 1984; Hausen and Egelhaaf 1989; Borst 1996) or different emphasis (Frye and Dickinson 2001; Borst and Dickinson 2002). Because of space limitations, we do not review the extensive work that has been done on fly motion-sensitive neurons to advance our understanding of neural coding (Bialek et al. 1991; Rieke et al. 1997; de Ruyter et al. 1997, 2000; Haag and Borst 1997, 1998; Borst and Haag 2001). Unless stated otherwise, all data presented in the following were obtained on the blowfly Calliphora vicina which we will often casually refer to as 'the fly'.

Neural networks in the cockpit of the fly

Animals have several good reasons for detecting motion with their eyes. First, the motion of other animals — potential preys, mates, intruders or predators — provides essential information on which to base vital moves such as escape or chase. Secondly, information about self-motion is crucial, especially in the context of navigation, course stabilization, obstacle avoidance, and collision-free goal reaching. In fact, the wealth of information provided by passive, non-contact self-motion evaluation in visual systems has been likened to a kind of “visual kinaesthesis” (Gibson 1958). Even the 3D structure of the environment can be picked up by a moving observer (revs. Collett and Harkness 1982; Buchner 1984; Nakayama 1985; Hildreth and Koch 1987). Von Helmholtz (1867) was the first to clearly state the importance of this “motion parallax” in locomotion, and Exner (1891) proposed that arthropods make use of motion parallax as well as stereopsis to estimate distances (see also Horridge 1986).

Directionally Selective Motion Detection by Insect Neurons

The mechanisms underlying visual motion detection can be studied simultaneously in different cell compartments in vivo by using calcium as a reporter of the spatiotemporal activity distribution in single motion-sensitive cells of the fly. As predicted by the Reichardt model, local dendritic calcium signals are found to indicate the direction and velocity of pattern motion but are corrupted by spatial pattern properties. The latter are canceled out by spatial integration, thus leading to a purely directional selective output signal in the axon. These findings attribute a specific computational task to the dendrites of visual interneurons and imply a functional interpretation of dendritic morphology.

https://science.sciencemag.org/content/sci/281/5384/1848.full.pdf

Dendritic integration and its role in computing image velocity.

A THEORETICAL OVERVIEW Introduction Deterministic Dynamical Systems Stochastic Dynamical Systems Information Theory Optimal Control ATOMISTIC SIMULATIONS OF ION CHANNELS Introduction Simulation Methods Selected Applications Outlook MODELING NEURONAL CALCIUM DYNAMICS Introduction Basic Principles Special Calcium Signaling for Neurons Conclusions STRUCTURE BASED MODELS OF NO DIFFUSION IN THE NERVOUS SYSTEM Introduction Methods Results Exploring Functional Roles with More Abstract Models Conclusions STOCHASTIC MODELING OF SINGLE ION CHANNELS Introduction Some Basic Probability Single Channel Models Transition Probabilities, Macroscopic Currents and Noise Macroscopic Currents and Noise Behaviour of Single Channels under Equilibrium Conditions Time Interval Omission Some Miscellaneous Topics THE BIOPHYSICAL BASIS OF FIRING VARIABILITY IN CORTICAL NEURONS Introduction Typical Input is Correlated and Irregular Synaptic Unreliability Postsynaptic Ion Channel Noise Integration of a Transient Input by Cortical Neurons Noisy Spike Generation Dynamics Dynamics of NMDA Receptors Class 1 and Class 2 Neurons Show Different Noise Sensitivities Cortical Cell Dynamical Classes Implications for Synchronous Firing Conclusions Generating Models of Single Neurons Introduction The Hypothalamo-Hypophysial System Statistical Methods to Investigate The Intrinsic Mechanisms Underlying Spike Patterning Summary and Conclusions GENERATING QUANTITATIVELY ACCURATE, BUT COMPUTATIONALLY CONCISE, MODELS OF SINGLE NEURONS Introduction The Hypothalamo-hypophysial System Statistical Methods to Investigate the Intrinsic Mechanisms Underlying Spike Patterning Summary and Conclusions BURSTING ACTIVITY IN WEAKLY ELECTRIC FISH Introduction Overview of the Electrosensory System Feature Extraction by Spike Bursts Factors Shaping Burst Firing In Vivo Conditional Action Potential Back Propagation Controls Burst Firing In Vitro Comparison with Other Bursting Neurons Conclusions LIKELIHOOD METHODS FOR NEURAL SPIKE TRAIN DATA ANALYSIS Introduction Theory Applications Conclusion Appendix BIOLOGICALLY-DETAILED NETWORK MODELING Introduction Cells Synapses Connections Inputs Implementation Validation Conclusions HEBBIAN LEARNING AND SPIKE-TIMING-DEPENDENT PLASTICITY Hebbian Models of Plasticity Spike-Timing Dependent Plasticity Role of Constraints in Hebbian Learning Competitive Hebbian Learning Through STDP Temporal Aspects of STDP STDP in a Network Conclusion CORRELATED NEURONAL ACTIVITY: HIGH-AND LOW-LEVEL VIEWS Introduction: the Timing Game Functional Roles for Spike Timing Correlations Arising from Common input Correlations Arising from Local Network Interactions When Are Neurons Sensitive to Correlated Input? A Simple, Quantitative Model Correlations and Neuronal Variability Conclusion Appendix A CASE STUDY OF POPULATION CODING: STIMULUS LOCALIZATION IN THE BARREL CORTEX Introduction Series Expansion Method The Whisker System Coding in the Whisker System Discussion Conclusions MODELING FLY MOTION VISION The Fly Motion Vision System: An Overview Mechanisms of Local Motion Detection: The Correlation Detector Spatial Processing of Local Motion Signals BY Lobula Plate Tangential Cells Conclusions MEAN-FIELD THEORY OF IRREGULARLY SPIKING NEURONAL POPULATIONS AND WORKING MEMORY IN RECURRENT CORTICAL NETWORKS Introduction Firing-Rate and Variability of a Spiking Neuron with Noisy input Self-Consistent Theory of Recurrent Cortical Circuits THE OPERATION OF MEMORY SYSTEMS IN THE BRAIN Introduction Functions of the Hippocampus in Long-Term Memory Short Term Memory Systems Invariant Visual Object Recognition Visual Stimulus-Reward Association, Emotion, and Motivation Effects of Mood on Memory and Visual Processing MODELING MOTOR CONTROL PARADIGMS Introduction: The Ecological Nature of Motor Control The Robotic Perspective The Biological Perspective The Role of Cerebellum in the Coordination of Multiple Joints Controlling Unstable Plants Motor Learning Paradigms COMPUTATIONAL MODELS FOR GENERIC CORTICAL MICROCIRCUITS Introduction A Conceptual Framework for Real-Time Neural Computation The Generic Neural Microcircuit Model Towards a Non-Turing theory for Real-Time Neural Computation A Generic Neural Microcircuit on the Computational Test Stand Temporal integration and Kernel Function of Neural Microcircuit Models Software for Evaluating the Computational Capabilities of Neural Microcircuit Models Discussion MODELING PRIMATE VISUAL ATTENTION Introduction Brain Areas Bottom-Up Control Top-Down Modulation of Early Vision Top-Down Deployment of Attention Attention and Scene Understanding Discussion

Computational neuroscience: a comprehensive approach

The centrifugal horizontal cells in the lobula plate of the blowfly, Phaenicia sericata

The anatomy and physiology of a motion-sensitive neurone, the vertical-horizontal (VH-) cell in the third visual neuropil (lobula plate) of the blowfly,Phaenicia was studied by intracellular recordings combined with dye injection. The cell possesses two dendritic fields in different layers of the lobula plate. The axon runs jointly with those of the vertical cells along the caudal surface of the lobula plate and terminates in the central protocerebrum lateral to the esophageal canal. The receptive field of the VH-cell is subdivided into two physiologically different parts which correspond to the two dendritic fields: if the input reaches the dendritic field residing in a more caudal layer (V-layer), the cell responds maximally to vertical pattern motion; whereas if the input reaches the dendritic field residing in a more rostral layer (H-layer), the cell responds maximally to horizontal pattern motion. The VH-neurone responds maximally to a contrast frequency of approximately ω/λ ≅1.8 Hz which coincides with the contrast frequency dependence of optomotor (following) responses. It is, therefore, considered to be a likely candidate mediating the pitch response (Blondeau and Heisenberg 1982) in flies.

The vertical-horizontal neurone (VH) in the lobula plate of the blowfly,Phaenicia

The dependence of the landing response on the direction of moving stimuli (periodic gratings, single or double stripes) was studied in blowflies, Calliphora erythrocephala, of both sexes. Directions of motion eliciting maximally strong responses (preference direction) vary with the eye region stimulated: they are distributed radially from a common origin forming a flow-field. This origin lies at the intersection of the eye equators with the median plane of the animal. By changing its body posture relative to the direction of flight, the fly may align the pole of this flow field-with its direction of flight thus maximizing signal flow for the landing approach. Sex-specific differences were found for dorsal eye regions in which the shift of preference directions from vertical to obliquely inclined directions of motion (against the median plane) could only be determined for male flies.

Hendrik Eckert

Papers

The centrifugal horizontal cells in the lobula plate of the blowfly, Phaenicia sericata

The vertical-horizontal neurone (VH) in the lobula plate of the blowfly,Phaenicia

On the landing response of the blowfly, Calliphora erythrocephala