Defining the Computational Structure of the Motion Detector in Drosophila
Citation: Damon A. Clark, Limor Bursztyn, Mark A. Horowitz, Mark J. Schnitzer, Thomas R. Clandinin (2011) Defining the Computational Structure of the Motion Detector in Drosophila. Neuron (RSS)
DOI (original publisher): 10.1016/j.neuron.2011.05.023
Semantic Scholar (metadata): 10.1016/j.neuron.2011.05.023
Sci-Hub (fulltext): 10.1016/j.neuron.2011.05.023
Internet Archive Scholar (search for fulltext): Defining the Computational Structure of the Motion Detector in Drosophila
Tagged: Neuroscience (RSS) Drosophila (RSS), Reichardt detector (RSS), vision (RSS), motion (RSS)
This paper tests the hypothesis that information about motion is extracted from shifting intensity patterns on the back of the retina. It shows that a Reichardt-detector model can closely match the neuronal response to visual inputs, and identifies two specific pathways that may carry out pieces of the related computation.
These two pathways (via laminar cells L1 and L2) are shown to respond differently to light vs. dark moving edges, and it is shown they both carry out part of the idealized computation described by a Reichardt detector. A theoretical model is proposed that would be tuneable to match observed responses closely (possibly changing over time).
Goals and Methods
These experiments probe the structure and possible component parts of the part of motion detection in Drosophila that responds to moving edges like a Reichardt detector. While no specific physical components have been identified as carrying out the individual steps of the Reichardt detector algorithm, this work tests the hypothesis that two pathways map to specific subsets of the overall Reichardt computation.
Directional selectivity of flies is observed in response to light and dark edges in intensity patterns, the traditional input used to characterize Reichardt detection. Responses to "phi" and "reverse phi" stimuli are also observed: changes in contrast in the same direction or in opposite direction. Reverse phi is related to an illusion: it causes the observer to turn in the opposite direction that the spatial sequence of contrast change would -- into rather than away from an edge. As it is also observed in animals other than flies, this suggests some overlap in analysis for future study.
In this experiment the flies studied are conscious and held in place suspended just above a ball, which floats at low friction in an airstream. The fly can move its legs to 'walk' along the ball, and the rotation of the ball captures its motion. While in this setup, the flies are shown three screens, ahead and to the left and right; shown the image of a rotationg virtual cylinder with periodic patterns.
First the response of wild-type Drosophila's Reichart correlation is characterized, by showing them two bars side by side, with a randomized range of contrast changes and time delays between successive images. They are also shown rotating cylinders with a fixed periodic square wave. Genetically modified Drosophila missing either the L1 or L2 pathway are then bred and tested with the same system.
Various tests of the L1 and L2 terminals are carried out to see whether each of them responded to both light and dark stimuli in isolation, and to both increases and decreases in intensity. The response of flies to both phi and reverse phi stimuli are noted and compared to motion sensing in other species.
To understand the role and function of edge selectivity in the L1 and L2 pathways, flies with only one or the other are shown a variety of inputs to find differential responses: including light-light, dark-dark, and phi and revers-phi stimuli.
It is a point of contemporary debate whether all four possible unit computations of the Reichardt detector are carried out: dark-dark, dark-light, light-dark, light-light. Drosophila are shown not only the four combinations of dark and light bars, but also phi and reverse-phi stimuli. Reverse phi stimuli are predicted to produce the opposite response of phi stimuli if all four computations are possible.
A numerical model is developed to match observed edge preferences and responses, with parameters weighting the individual computations of the Reichardt detector to fit observations.
Finally, the implications of two distinct pathways within each Reichardt detector process, for fly motion detection and for understanding other observations of fly neuron responses, are analyzed.
Results and Analysis
The fly response to randomly generated bars roughly matches that of an ideal Reichardt correlator.
Losing L1 reduces response to only the light edges; losing L2 reduces response to only dark edges. To distinguish edge detection from overall changes in light level, an equiluminant stimulus is developed where light and dark edges move in opposite directions at equal speeds, at once. Control flies move only slightly; those without L1 turn towards the dark edge, those without L2 turned towards the light.
This strongly suggests that L1 and L2 pathways are tied to the ability to detect motion of light and dark edges, respectively. However both L1 and L2 are found to respond similarly to light and dark flashes, and to respond to the motion of both light and dark edges. They also respond highly linearly to changes in contrast - either increases or decreases. In contrast to contemporary work that suggested L2 terminals respond to decreases and not increases in brightness, this is shown to be linear quite uniformly across the spectrum of stimulus intensity.
As reverse phi stimuli produce the response predicted by a full Reichardt detector, the authors concluded that all four unit compoutations are carried out. This did not entirely settle the question for other researchers however. Reverse phi stimuli had another interesting property: they produced different complementary responses in flies missing L1 or L2 pathways.
To understand this better, a numerical model is designed for an array of Reichardt detectors, with responses to phi and reverse-phi inputs tuned by weighting the four different unit multiplications in the detector. Weighting phi stimuli equally while weighting reverse-stimuli differentially is enough to reproduce the edge-selction seen in L1 and L2 pathways.
As reverse phi responses are observed in many species other than Drosophila, the analysis and model presented at the end of this study are of potential interest to studies of edge and motion detection in other creatures.
- Notes for future and contemporary researchers
The idea of half-wave rectification of the signal at each input has often been proposed since it is not known how sign-correct multiplication of inputs could be implemented. Rectification may happen elsewhere in the neural circuit, however. These experiments demonstrate that the L1 and L2 cells do not themselves carry out such rectification.
The authors tackle the ongoing debate between "two-computation" and "four-computation" models of a Reichardt detector, which are related to ideas about whether the fly visual system is organized into ON and OFF pathways or simply detecting light v. dark edges. They conclude from their work that the fly visual system is not organized into ON / OFF pathways -- that is, that L1 transmits information about more than just increases in contrast, and L2 transmits information about more than just decreases. This is in contrast to Joesch and Eichner.
A few time dependencies are noted related to the laminar cells, as pointers to future study. The flies studied remember the last image they had seen for at least 1 second for the purposes of future reactions. L1 and L2 terminals show different kinetics in respond to prolonged stimuli. And previous work suggests that electrical changes in the cell membrane potential would last for 50-100ms, in contrast to the Ca responses in axonal terminals that last for seconds. This could indicate adaptation of the synapse, or processing within the axon.
Reverse phi is proposed as more than an illusion - but as a specific sort of motion captured by neurons, with functional use. In humans for instance, reverse phi shares properties with motion aftereffects, and may contribute to perception of moving edges of a specific polarity. Edge-detecting cells are a fundamental unit of visual computation, and this work suggests edge-polarity detection is as well.
Finally, as demonstrated most concisely in the numerical model presented, edge selectivity can be characterized by selective weighting of L1 and L2 pathways. Furthermore, this study may only have captured a subset of the actual computations carried out. Other cells may be involved as well - L4 cells may have a significant role to play.
- See Internal Structure of the Fly Elementary Motion Detector, Eichner, Joesch, Schnell, Reiff, and Borst. Neuron, 2011 (published in the same issue, back to back). An earlier related work by Joesh is referenced in this paper.
- Phi movement as a subtraction process. S.M. Anstis, Vision. (1970)
Theoretical and Practical Relevance
Changes in light and shadow over time and space are shown to be detected by a system that closely parallels a simple Reichardt detector (or Hassenstein-Reichardt correlator). A model for such a detector is proposed that can be tuned to detect moving edges of contrast.
This clearly distinguished two different neuronal pathways (L1 and L2) as responsible for different parts of vision, by selectively disabling each one separately and quantifying the different changes in edge detection.