Neural correlates of illusory motion perception in Drosophila

Abstract

When the contrast of an image flickers as it moves, humans perceive an illusory reversal in the direction of motion. This classic illusion, called reverse-phi motion, has been well-characterized using psychophysics, and several models have been proposed to account for its effects. Here, we show that Drosophila melanogaster also respond behaviorally to the reverse-phi illusion and that the illusion is present in dendritic calcium signals of motion-sensitive neurons in the fly lobula plate. These results closely match the predictions of the predominant model of fly motion detection. However, high flicker rates cause an inversion of the reverse-phi behavioral response that is also present in calcium signals of lobula plate tangential cell dendrites but not predicted by the model. The fly's behavioral and neural responses to the reverse-phi illusion reveal unexpected interactions between motion and flicker signals in the fly visual system and suggest that a similar correlation-based mechanism underlies visual motion detection across the animal kingdom.

Fig. 1.

Flies faced with panoramic reverse-phi motion exhibit reverse-optomotor responses. (A) A schematic model for fly motion vision consists of two stages: (i) local motion is computed by columnar circuits within the lamina and medulla, which is followed by (ii) global integration of local motion signals in the lobula plate tangential cells. The output of the LPTCs is thought to control optomotor behavior. (B) The fly is suspended within a virtual flight arena where the amplitude of each wing-beat is tracked by an optical detector. The difference between the two wing-beats (left minus right wing-beat amplitude) is proportional to yaw torque (24). For example, when the amplitude of the left wing-beat is greater than the right, the fly is attempting to steer to the right with clockwise torque. (C) Space–time depictions of motion stimuli used in rotation experiments—all three are square-wave patterns moving from the top left to the bottom right (Movie S1). (D) Mean turning behavior of 10 flies (±SEM) in response to open-loop rotation of standard (Top), reverse-phi (Middle), and reverse-phi out-of-phase (Bottom) square-wave gratings (λ = 30°). The speed of reverse-phi out-of-phase stimuli moved at one-half of the speed of the standard and reverse-phi stimuli, because motion occurred only in every second frame (the space–time plot in C). Flies were presented with motion in both directions (CW and CCW), but responses are combined and plotted for CW rotation (SI Text has a complete description of data treatment).

Fig. 2.

The H-R EMD model of motion detection accounts for sensitivity to reverse-phi motion. (A) The H-R EMD requires two adjacent sampling units (e.g., photoreceptors) separated by a sampling base with distance Δϕ. Incoming luminance signals are high pass-filtered (τ_hp) before they are temporally delayed (τ_lp) and multiplied with the signal from the neighboring retinal sampling unit. The output signals of the two subunits are then subtracted (Σ). Standard motion in the EMD's preferred direction produces a positive output (38), whereas reverse-phi motion causes signals of opposite polarity to coincide at the multiplication stage, resulting in a negative output. (B) Behavioral tuning curves for standard and reverse-phi motion stimuli. Mean turning responses (±SEM; n = 10) to 3 s of standard (Left) or reverse-phi (Right) rotation. Stimulus velocity and spatial wavelength (λ) were varied across trials, and turning responses are shown as a function of both velocity and temporal frequency (Inset). Responses to standard motion exhibit temporal frequency tuning, whereas reverse-phi responses are velocity-tuned but feature an inversion at high speeds. (C) Modeled EMD responses to clockwise rotation are plotted as a function of both velocity and temporal frequency (Inset). For standard rotation, the model response is positive for a CW stimulus and exhibits a constant temporal frequency optimum across spatial wavelengths. For the reverse-phi stimulus, the model output is negative and tuned to the velocity of the stimulus.

Fig. 3.

The LPTC HSN responds to reverse-phi motion with inverted direction selectivity. (A) The experimental setup used to record calcium transients from LPTC dendrites consists of a tethered fly walking on an air-supported ball. (B Left) Low magnification of the left hemisphere of the fly brain (dorsal to the left and midline to the top) with horizontal system north (HSN) and equatorial (HSE) neurons labeled with gCaMP3.0. (Scale bar: 25 μm.) (Right) Higher magnification view of the dendritic arbors of the HS neurons showing the region of interest selected to analyze HSN responses to moving stimuli. (Scale bar: 7 μm.) (C) Example from a single fly of the HSN responses to visual motion stimuli updated at the same frame rate (corresponding to a temporal frequency of 1 Hz for standard motion; four to five repetitions each; individual trials are gray and the mean is depicted in the corresponding color: blue, standard rotation; red, reverse-phi; green, reverse-phi out of phase). The shaded region denotes the onset and duration of the visual stimulation. (D) Normalized HSN dendrite responses (n = 9 flies; mean ΔF/F ±SEM during the last 0.5 s of the stimulation) to motion (Left) and flicker-containing stimuli (Right). Flicker rates within the response range of the HSN neuron elicited mean responses that were significantly lower than those induced by either CW reverse-phi motion (Right; P < 0.001, u test with the sole exception of the lowest flicker rate tested, P = 0.09) or CCW reverse-phi out of phase (P < 0.0005, u test).

Fig. 4.

Independently varying contrast and flicker rates reveals local motion computations not predicted by the EMD model. (A) Mean turning responses (±SEM; n = 10 flies) to reverse-phi rotating stimuli in which the velocity and rate of contrast reversal were controlled independently (Movie S2). Rotation occurred at one of five velocities, and the contrast of the stimulus was inverted at one of four flicker rates. Blue traces indicate standard motion (no flicker), and red traces indicate reverse-phi (flicker rate is equal to motion frame rate). Blue asterisks denote conditions with low flicker rates and high motion frame rates, black asterisks indicate when the stimulus flickers at an integer multiple of the motion frame rate, and red asterisks indicate when the motion frame rate is two times the flicker rate. Green lines represent the steady-state output of an elementary motion detector simulation (same model parameters as in Fig. 2C) to each visual stimulus. (B) Normalized HSN dendrite responses (±SEM; n = 6 flies) to reverse-phi motion stimuli in which the rate of contrast reversal was independently varied (identical stimuli as in columns 2 and 3 of A; traces color coded as in A). To facilitate comparison with the behavioral results, each time series is plotted as CCW responses minus CW responses (calculated from the individual traces in Fig. S5). Insets on the right of each trace show whether HSN calcium transients (black), fly behavior (magenta), and EMD model prediction (green) are significantly greater than (+), less than (−), or not different (0) from zero, measured as the normalized mean response to the same stimulus (P < 0.1, one-tailed t test).