Muscles that move the retina augment compound eye vision in Drosophila

Abstract

Most animals have compound eyes, with tens to thousands of lenses attached rigidly to the exoskeleton. A natural assumption is that all of these species must resort to moving either their head or their body to actively change their visual input. However, classic anatomy has revealed that flies have muscles poised to move their retinas under the stable lenses of each compound eye^1-3. Here we show that Drosophila use their retinal muscles to smoothly track visual motion, which helps to stabilize the retinal image, and also to perform small saccades when viewing a stationary scene. We show that when the retina moves, visual receptive fields shift accordingly, and that even the smallest retinal saccades activate visual neurons. Using a head-fixed behavioural paradigm, we find that Drosophila perform binocular, vergence movements of their retinas-which could enhance depth perception-when crossing gaps, and impairing the physiology of retinal motor neurons alters gap-crossing trajectories during free behaviour. That flies evolved an ability to actuate their retinas suggests that moving the eye independently of the head is broadly paramount for animals. The similarities of smooth and saccadic movements of the Drosophila retina and the vertebrate eye highlight a notable example of convergent evolution.

Extended Data Figure 1:. Retinal muscles that attach to the front of the orbital ridge can move the entire retinal sheet coherently.

(a) Schematic of the attachment of the retinal eye muscles to the orbital ridge in the Drosophila head (for an animated version see Extended Data Movie 1). (b) Staining of a part of the Drosophila head including the eye and the orbital ridge in a cleared specimen using Calcofluor White. Soft tissue was removed proteolytically. It reveals the vesica piscis-shaped opening of the orbital ridge and shows strongly sclerotized parts in the frontal region, where the two muscles attach (red stars), as well as two discontinuities on the dorsal and ventral poles (arrows). One possibility is that these discontinuities decouple the front of the orbital ridge from the back, mechanically. This may allow muscles that pull the front of the retina to also move the rear by the same amount through internal cohesion within a stiff set of ommatidia, rather than through a force vector that dissipates over space from front to back. Alternatively, the inhomogeneities at the top and bottom of the orbital ridge could act as a fulcra or pivot points, leading to the rear part of the orbital ridge to move outward when the frontal part moves inward (towards the midline), which could aid coherent motion of the retina. We will test these models in biomechanical studies in the future. For an animated version of the Calcofluor White image stack see Extended Data Movie 2. (c) We measured retinal movements using three cameras pointing at three different positions in one eye. To induce large movements, we optogenetically activated the retinal motoneurons. We expressed CsChrimson in a split-GAL4 line (w+;R44A07-AD;R13D09-DBD) and focused red light onto a spot between the fly eyes (Methods). The three plots to the right show the views of the three different cameras: we overlaid images before and after optogenetic activation and plotted the tracked centroid on top of the pseudopupil (white) to illustrate the retinal movements. (d) Optogenetic activation in this fly yielded movements roughly along the axis connecting photoreceptors 3–5, and we used this motion direction (dark arrow) for comparison across the different positions on the facet eye in (e). Schematic modified from Stavenga (1979) (e) Traces of the pseudopupil measured in front (blue), in the middle (orange) and in the back (green) of the fly eye. (f) Same traces as above but shifted in y (by hand) to illustrate the coherent motion of the pseudopupil across the eye.

Extended Data Figure 2:. Electrophysiological recordings from mDCNs reveal that they have small, front facing receptive fields and that they respond strongly, in a non-direction-selective manner, to moving spots and bars.

(a) GFP expression in the VT37804-GAL4 Line shows LC14 cells, which connect the lobula on one side to the contralateral lobula and medulla. Additional off-target expression in the anterior optic tubercle is also visible. (b) Example trace showing the mDCN membrane voltage (Vm) while it is modulated by a variety of visual stimuli (bars, spots and gratings). (c) Membrane potential responses from a single cell reveal strong, non-directional responses to moving spots and bars in a restricted portion of the visual field. Light gray: single trials. Black: mean response. (d) Single cell response to 100 ms spot flashes (gray region) on the LED screen. Light gray: single flashes. Black: mean response. (e) Heat map image representation of an LC14 cell’s receptive field (same cell as shown in d). (f) Population-averaged responses to moving bars and gratings (six cells) reveal more consistent responses to bars than gratings. (g) Heat-map representations of four more LC14 cells’ receptive fields. (Bottom left plot shows same data as Fig. 2b)

Extended Data Figure 3:. Optokinetic retinal tracking is interspersed with (nystagmus-like) counter-saccades and the largest counter-saccade magnitudes are observed in flight.

(a) We isolated large saccades of the right retina with a simple, threshold-crossing algorithm and plotted these alongside the concomitant retinal movements of the left retina in 10 quiescent (i.e., non-flying) flies. Data are shown in the context of full-field rightward grating motion (top) (87 saccades), rightward motion in the right visual hemisphere (middle) (56 saccades), and bilateral front-back-motion (bottom) (66 saccades). (b) Same as panel a but during tethered flight. We also show the left-minus-right wing beat amplitude (L–R WBA) of the flies, with rightward deflections indicating a rightward steering response and vice versa (top, middle and bottom traces include 122, 136 and 152 saccades respectively).

Extended Data Figure 4:. Retinal saccade magnitudes and peak velocities are tightly correlated, akin to the main sequence in human saccades.

(a) Saccade peak velocity plotted against saccade amplitude of the saccades of the right retina from Extended Data Fig. 3a. Data are shown in the context of full-field rightward grating motion (top) (87 saccades), rightward motion in the right visual hemisphere (middle) (56 saccades), and bilateral front-back-motion (bottom) (66 saccades). (b) Same as panel a but for the saccades of the right retina from Extended Data Fig. 3b, which occurred during tethered flight (top, middle and bottom traces include 122, 136 and 152 saccades respectively). The mean saccade amplitudes for flight and quiescence were 2.6° and 1.1° for full-field rightward motion, 1.6° and 0.7° for rightward motion in the right hemisphere and 2.7° vs 0.6° deg for bilateral front-to-back motion. All saccade-magnitude differences between flight and quiescence were highly significant (p<10⁻¹⁰, two-sided t-test).

Extended Data Figure 5:. The gain of optokinetic tracking is below unity and was significantly modulated by flight for bilateral front-to-back motion.

(a) The initial pseudopupil velocity for full-field 15°/s leftward grating motion when analyzing the left eye (orange) and for full-field 15°/s rightward grating motion when analyzing the right eye (blue). Data for optokinetic responses made during quiescence and tethered flight are shown separately. Each point represents data from one fly. (b) Initial pseudopupil velocities for unilateral motion stimuli: 15°/s grating motion in the left side for the left eye and right side for the right eye. (c) Initial pseudopupil velocities for bilateral front-back-motion. The initial pseudopupil velocities during flight and quiescence were not significantly different for rotational stimuli (a, b) but they were significantly bigger in flight for bilateral front-to-back motion, which simulates forward translation (two-sided Wilcoxon signed-rank test, left eye: p=0.005, right eye: p=0.007).

Extended Data Figure 6:. Silencing neurons in the early fly visual system abolishes optokinetic responses to visual motion but preserves spontaneous retinal movements in flying flies.

(a) Immuno-stain of a split-GAL4 line labelling L1-L4 visual neurons (green) and neuropil (anti-brp, magenta) (b) Example traces showing retinal movement in a control fly, expressing inactive tetanus toxin in the L1-L4 cells, and an experimental fly, expressing active tetanus toxin in the same neurons, in flight and non-flight. All behavioral responses were made in the context of a full-field grating rotating at 15°/s (1 Hz temporal frequency). (c) Single fly averaged (thin lines) and population averaged (thick line) retinal responses for 8 control flies and 10 experimental flies during non-flight/quiescence. Right panel: Optomotor index (Supplementary Information) quantifying the response strength to visual motion in the expected optokinetic direction. (d) Same as panel c, but in flying flies. Black: Left-minus-right wingbeat amplitude (L–R WBA).

Extended Data Figure 7:. In both D. melanogaster and D. suzukii, the sign and magnitude of retinal optokinetic responses and walking optomotor responses are consistent with non-hyperacute sampling of the visual world by the fly retina.

(a) Experimental setup: pin-tethered flies walked on floating ball. We recorded the ball’s rotations as a readout of the flies’ turning velocity (black) alongside the displacements of the deep pseudopupil (left: orange, right: blue). Visual stimuli were presented on a conical screen using a projector and consisted of full-field rotating gratings at varying spatial wavelengths (always 4 Hz temporal frequency) for 5 s. (b) The inter-ommatidial angle Δϕ of the compound eye limits the spatial wavelength λ of a grating that can be properly resolved (modified from Land 1997⁵²). Below the cut-off wavelength of λ_L = 2·Δϕ, direction-selective motion responses of the visual system are predicted to invert due to spatial aliasing^,. (c) Top and middle: displacements of the left and right pseudopupils during the stimulus period as a function of the grating wavelength, λ (1 to 40°) for left- and rightward motion. Bottom: concomitant average walking velocity during the stimulus period. Thin lines are the trial-averages from 11 single flies. Thick lines are population averages. Note the sign inversion in the range of λ=5–10° for both the retinal optokinetic reflex and the walking optomotor response. (d) Average responses for both retinas for front-to-back (thicker line) and back-to-front motion for Drosophila melanogaster (top, data as in b) in comparison to Drosophila suzukii (bottom). D. suzukii showed inverted responses at smaller wavelengths than D. melanogaster, i.e. had a higher spatial acuity, consistent with the fact that D. suzukii have approximately twice the number of ommatidia as D. melanogaster. The sign-inversion of behavioral responses around the critical wavelength inferred from the optics of the eye argues that any retinal movements flies are making in the context of moving gratings are not allowing them to perceive the motion direction of fine gratings better than would be expected from the first-order optics of the eye. The lack of evidence for hyperacuity in this context does not exclude it existing in other circumstances.

Extended Data Figure 8.. Visual neurons activate during spontaneous saccades with direction-selective responses.

(a) Left: sample Vm responses of an HS cell on the left side of the brain to rightward followed by leftward grating motion (1 Hz temporal frequency). (b) Vm of the same cell, alongside the x-movements of the left retina (orange), in the context a stationary vertical grating. Arrows indicate moments of spontaneous saccades. (c) Trial-averaged Vm of single flies (gray, N=12 HS or VS cells) and population averaged Vm (black) for left-eye retinal saccades (orange). Data from a dark arena (left), a uniformly lit arena (middle), and a stationary grating (right) are shown. Top plots shown downward retinal movements for VS cells and leftward retinal movements for HS cells, which should produce visual motion in the preferred direction. Bottom plots show the opposite, null-direction retinal movements. The direction-selective responses to gratings argue that HS and VS cells respond to the visual motion induced on the retina by < 1° eye movements. The weak response to eye movements in darkness, or with a uniformly lit screen, is opposite in sign to that observed with a grating, which may represent an efference copy of the predicted motion signal arriving to HS/VS cells with each eye movement. This efference copy is potentially superseded by the actual, grating motion input with a high contrast grating, in the rightmost column.

Extended Data Figure 9.. Evidence that flies are genuinely in the dark during the lights-off epoch of the gap crossing experiments.

(a) Flies walking on the gap-crossing wheel (Fig. 5) were presented with a grating printed on paper that was physically moved back and forth in front of the right eye with a motorized manipulator. A small slit in the printed grating allowed us to slide an InfiniStix lens through it, abutting the fly’s right eye, to track the deep pseudopupil. (b) Example traces showing the horizontal shift of the right retina in two flies (blue) together with the grating position (black). (c) We observed a clear optokinetic response with the lights on, but not during darkness, demonstrating that there was genuinely no light available for flies to see with the lights off, even after being dark adapted for 30 minutes. Light blue: single flies. Dark blue: population mean. Five repetitions of the grating’s movement were presented and averaged for each fly, in each 5-min. epoch shown. (d) We quantified the number of gap crossings from the data presented in Fig. 5 and observed a ~30%, statistically significant, drop in the rate of gap crossing during the dark period (t-test, p = 0.013 when comparing lights on #1 with darkness and p = 0.0056 when comparing lights on #2 with darkness).

Extended Data Figure 10.. Anatomical characterization of two the split GAL4 lines used for silencing retinal motor neurons.

We visualized expression in these two split-Gal4 lines by driving CsChrimson-tdTomato in R44A07-AD;R13D09-DBD (top) and R414B04-AD;R13D09-DBD (bottom). Maximum z-projections of the brain are shown over roughly the posterior and anterior halves to better visualize the branching. VNC maximum projections are shown over the full stack. The green cells on the right and left side of the SEZ (arrows) are retinal motor neurons, based on their dendritic arborization, location of cell body and their axons leaving the brain just below the antennal lobe. We could optogenetically induce retinal movements via expression of CsChrimson expression in both split lines.

Figure 1.. Drosophila have two muscles per eye which act to move the retina.

(a) Frontal view through the Drosophila head showing immunostaining for muscles in red (phalloidin) and neurons, including retinal motor neurons, in green (mVenus). Musculus orbito-scapalis (MOS), which interconnects the antennal cup to the front of the orbital ridge (which surrounds the retina), is fully visible. Musculus orbito-tentoralis (MOT) is partially visible, at its insertion point to the orbital ridge. (b) A horizontal view allows one to fully visualize the MOT and its innervating motor neuron (arrow). (c) Corresponding schematics of the insertion of both muscles in the coronal (top) and horizontal (bottom) plane. Muscle tendons in blue, orbital ridge in yellow. Other muscles (proboscis, esophagus) in brown. (d) Retina seen through a water immersion objective. Two images are overlaid. The red image shows the retina during optogenetic activation of the retinal motor neurons, yielding a maximal shift of photoreceptor tips. The grey image was taken after turning off the optogenetic light, with the muscles fully relaxed. Note that whereas the photoreceptors moved with optogenetic activation (red and gray photoreceptor tips are offset), the lenses, which are in focus on the left of the image (arrow), were stable. (e) The schematic shows how the deep pseudopupil is an erect, virtual image of the photoreceptor tips at the center of the curvature of the compound eye (drawn after). Overlayed image of the deep pseudopupil in one fly during (red) and after (gray) optogenetic activation (as in panel d). (f) Simultaneously tracked center-of-mass traces of photoreceptor tips, via a water immersion objective visualizing ommatidia at the very top of the eye, and the deep pseudopupil, via an air lens aimed at a lower region of the same eye. See methods for how we converted pseudopupil movements from the units of pixels to degrees.

Figure 2.. Retinal movements yield the expected angular shifts in receptive fields of visual neurons.

(a) We recorded from LC14 neurons on the right side. (b) Example of an LC14 cell 2D receptive field, as estimated by responses to 100 ms (gray region) flashes on the screen. (c) We performed whole cell patch clamp recordings in rigidly tethered flies. We optogenetically activated the motoneurons (Methods) to induce large retinal shifts. We measured the concomitant retinal position with a camera as well as the membrane voltage (Vm) in visual neurons in response to a moving black bar (bar width: 9 deg width, bar velocity: 21 deg/s) in lieu of the 2-dimensional receptive field. (d) Example trace showing the retinal positions (top trace; orange: left retina: right retina) and the Vm (bottom). Grey and red rectangles indicate trials without and with optogenetic activation. (e) Left: Baseline subtracted membrane voltage for rightward bar motion in one fly. Red thin lines show trials with optogenetic activation (n = 9) and black lines trials without (n = 7), thick lines indicate the means. Right: As left, but for leftward bar motion (n = 9 for both conditions). (f) Averaged membrane voltage across trials and for both bar directions. Distance between red and black vertical lines indicates the shift in the pseudopupil induced by the optogenetic light. (g) As in (f) but for a population of six flies (light lines single flies, dark lines averages). Baseline Vm (mean during 0.5 s at trial onset) was in all flies slightly higher during optogenetic activation than in trials without (mean Vm: −71.1 vs −72.5 mV with standard deviations of 5.5 mV). (h) Left: As in (g) but for single fly data normalized to peak values. Right: Vm data is shifted by the measured angular retinal shift (indicated by vertical lines).

Figure 3.. Drosophila perform vertical and horizontal retinal optokinetic responses and these responses are independently controlled in the two eyes.

(a) Apparatus to monitor the position of the deep pseudopupil in both eyes alongside the wing steering behavior of tethered, flying flies, while visual stimuli are presented on a panoramic LED display. Left eye: orange throughout. Right eye: blue throughout. (b) Sample traces of the positions of both pseudopupils as a fly viewed rightward and leftward moving gratings. Data from a quiescent fly is shown on left and a flying fly on right. Responses in flight are more variable, with varying frequency of nystagmus saccades on a trial to trial basis (compare left and right eye in flight). (c) Plots at top show the population-averaged 2-dimensional (x-y) movements of the pseudopupils in response to full-field rightward and leftward moving gratings. Traces in middle show the horizontal (x) component of the population-averaged pseudopupil movements over time (Gray region indicates when the stimulus is moving). Schematics on top illustrate the visual stimulus presented. (d) Data from the same flies as in c, but during flight. Wing-steering left-minus-right wingbeat amplitude (L–R WBA) responses are shown at bottom, with rightward deflections indicating a rightward steering response. (e-f) Same as panels c-d, but for unilateral rightward and leftward motion. Data from gratings presented to the fly’s right eye are shown; symmetrical results were obtained when the left eye was stimulated (not shown). (g-h) Same as panels c-d, but for bilateral back-to-front motion. (i-j) Same as panels c-d, but for bilateral front-to-back motion. (k-l) Same as panels c-d, but for up and down motion. Vertical (y) movements are shown in the pseudopupil time series for these panels. Thin lines: single fly averages (3–10 trials). Thick lines: population average (N=8–10 flies).

Figure 4. Drosophila spontaneously move their retina, often in a saccadic fashion.

(a) Spontaneous retinal movements in a flying fly viewing a dark screen (left) or a stationary vertical grating (right). Left eye: orange throughout. Right eye: blue throughout. Top: x-y plot of pseudopupil movements. Bottom: horizontal (x) component of the pseudopupil movement over time. (b) X-Y plots of both retinas in two more flies viewing a stationary grating (20 s). (c) We isolated moments in which the right eye generated large saccades in the x dimension and averaged the x movements of the right and left eye around those times for our population, in the context of either a dark screen or a stationary, vertical grating.

Figure 5.. Retinal movements are important for Drosophila when crossing gaps.

(a) Tethered flies walked on a custom, 3D-printed wheel with two, 2.5 mm gaps. One gap had horizontal stripes and the other vertical stripes painted on the walls. Using cameras, we tracked the position of the wheel and the positions of the two retinas. (b) Example time series of the right and left deep pseudopupil positions (top) and wheel position (bottom) in a fly that performed particularly large vergence movements at the time of gap crossing. The dotted horizontal lines indicate the threshold position used to detect a gap crossing; vertical grey lines indicate a gap crossing event in the forward direction (Methods). (c) Gap-crossing-triggered averages revealed consistent convergent retinal movements at the moment of gap crossing. We plotted the sign-inverted product of the left- and right-eye retinal shifts (Methods) (top) as a metric that goes positive during coincident vergence movements. (d) Same as (c) but for a population of 23 flies. (e) Quantification of the time traces in (d). For the wheel position and the left- and right-eye retinal shifts, we calculated the mean baseline signal in a 3 s window, starting 5 s before gap crossing and we subtracted this value from the mean signal in a 3 s window starting 2 s after gap crossing. For the vergence metric we subtract the mean signal in a 1 s baseline window starting 2.5 s before gap crossing from the mean signal in a 1 s window surrounding the gap crossing event. All distributions are significantly different from zero (t-test, P<0.05, with a Bonferroni correction for 9 tests), except the vergence-measure distribution in darkness, which has a P value of 0.0083 that is just above the 0.0056 needed after the Bonferroni correction. (f) We expressed Kir2.1 in two split GAL4 lines targeting retinal motor neurons. Schematic of the position of the retinal motoneurons, and immunostainings of two split GAL4 lines > Kir2.1-T2A-tdTomato (R44A07-AD;R13D09-DBD top, R14B04-AD;R13D09-DBD bottom). Scale bars: 100 μm). (g) Snapshot of a fly crossing a gap, to scale with the plots in h,i. (h) Left: x-y trajectories of gap crossings in a control example fly. Right: as on left for an example fly expressing Kir2.1. (i) Median of minimum positions for all crosses per fly (dots) together with mean +/− SEM (black lines) for R44A07-AD;R13D09-DBD control (black) and silenced (red) flies, and the equivalent plots for line R414B04-AD;R13D09-DBD. (j) Optokinetic responses of both eyes in in control (black) and silenced (red) flies for both split GAL4 lines.