Direction Selectivity in Drosophila Emerges from Preferred-Direction Enhancement and Null-Direction Suppression

Jonathan C. S. Leong, Jennifer J. Esch, Brian D. Poole, Surya Ganguli, Thomas R. ClandininJournal of Neuroscience 2016 · DOI

Abstract

Feature selective neurons respond preferentially to astonishingly specific stimuli, providing the neurobiological basis for perception. Direction selectivity serves as a paradigmatic model of feature selectivity that has been examined in many species. While insect elementary motion detectors have served as premiere experimental models of direction selectivity for 60 years, the central question of their underlying algorithm remains unanswered. Using in vivo two-photon imaging of intracellular calcium signals, we measure the receptive fields of the first direction-selective cells in the Drosophila visual system, and define the algorithm used to compute the direction of motion. Computational modeling of these receptive fields predicts responses to motion and reveals how this circuit efficiently captures many useful correlations intrinsic to moving dark edges.

Introduction

How do neurons transform a sequence of static images into a perception of motion? This "elementary motion detection" problem has been a cornerstone of neurobiology for over 60 years. Two primary models have historically dominated the field: the Hassenstein–Reichardt Correlator (HRC), which suggests motion detection works by enhancing responses in the "Preferred Direction" (PD), and the Barlow–Levick (BL) Detector, which posits suppression in the "Null Direction" (ND). While T4 and T5 neurons in the Drosophila visual system were identified as the first cells to show direction selectivity, the underlying algorithm remained a "black box" until now.

Methods

Researchers employed a multi-pronged approach to decode the T5 neuron's function:

  • In Vivo Two-Photon Calcium Imaging: Neural activity was recorded in living flies using GCaMP6f.
  • Stochastic "Flp-out" Labeling: Genetic techniques were used to label random individual cells, providing single-unit resolution within dense neural tissue.
  • Spatiotemporal Receptive Field (STRF) Mapping: Flies were presented with "ternary noise" (randomly flickering bars), and reverse correlation was used to map the spatial and temporal points to which a neuron responds.
  • LNCN Modeling: A Linear-Nonlinear-Convolutional-Nonlinear model was developed to account for specific lags and nonlinearities introduced by the calcium indicator, allowing for an accurate fit to biological data.
Overview of the experimental setup, neural circuit, and receptive field mapping in the Drosophila visual system.

Key Insights and Results

The "Spacetime Tilt" and Hybrid Algorithm

The mapping revealed that T5 receptive fields are "tilted" in spacetime, meaning the neuron is physically tuned to the velocity of an object. If an object moves at the right speed and direction, it remains within the neuron's "sweet spot" as it travels across the eye.

Most significantly, the study found that T5 neurons do not choose between HRC and BL models; they utilize a hybrid strategy. The spatiotemporal organization creates both PD enhancement (signals add up perfectly) and ND suppression (signals from different subfields cancel each other out).

Spatiotemporal receptive field mapping (E-F) and the resulting subfield geometry (G) showing ON and OFF regions.

Model Performance and Specialization

  • Accuracy: The LNCN model achieved a correlation coefficient of 0.87 when predicting responses to moving gratings, despite being trained only on random noise.
  • Subfield Geometry: T5 receptive fields consist of two oppositely signed subfields (ON and OFF) offset by approximately 12° in space.
  • Dark Edge Specialization: T5 is specifically tuned for moving dark edges, capturing high-order correlations that explain why these neurons are highly efficient at detecting shadows.

Discussion and Future Work

The organization of the fly's T5 receptive field shows remarkable parallels to "simple cells" in the mammalian primary visual cortex, suggesting evolutionary convergence on a universal solution for motion detection.

While this study provides a definitive algorithmic description, the specific mechanistic hardware—the synapses and neurotransmitters implementing this math—remains to be fully identified. Future research will focus on mapping upstream neurons in the medulla to the specific ON and OFF subfields identified here, and expanding the depth of modeling for T4 (light-edge) detectors.

Summary updated

Abstract

Feature selective neurons respond preferentially to astonishingly specific stimuli, providing the neurobiological basis for perception. Direction selectivity serves as a paradigmatic model of feature selectivity that has been examined in many species. While insect elementary motion detectors have served as premiere experimental models of direction selectivity for 60 years, the central question of their underlying algorithm remains unanswered. Using in vivo two-photon imaging of intracellular calcium signals, we measure the receptive fields of the first direction-selective cells in the Drosophila visual system, and define the algorithm used to compute the direction of motion. Computational modeling of these receptive fields predicts responses to motion and reveals how this circuit efficiently captures many useful correlations intrinsic to moving dark edges.
BibTeX
@article{leong2016direction,
  author = {Jonathan C. S. Leong and Jennifer J. Esch and Brian D. Poole and Surya Ganguli and Thomas R. Clandinin},
  title = {{Direction Selectivity in Drosophila Emerges from Preferred-Direction Enhancement and Null-Direction Suppression}},
  journal = {Journal of Neuroscience},
  year = {2016},
  doi = {10.1523/jneurosci.1272-16.2016},
}
Figure