doi: 10.1073/pnas.2109717119.
Epub 2022 Mar 17.
Binocular mirror-symmetric microsaccadic sampling enables Drosophila hyperacute 3D vision
Affiliations
- PMID: 35298337
- PMCID: PMC8944591
- DOI: 10.1073/pnas.2109717119
Item in Clipboard
Binocular mirror-symmetric microsaccadic sampling enables Drosophila hyperacute 3D vision
Proc Natl Acad Sci U S A.
.
Abstract
SignificanceTo move efficiently, animals must continuously work out their x,y,z positions with respect to real-world objects, and many animals have a pair of eyes to achieve this. How photoreceptors actively sample the eyes' optical image disparity is not understood because this fundamental information-limiting step has not been investigated in vivo over the eyes' whole sampling matrix. This integrative multiscale study will advance our current understanding of stereopsis from static image disparity comparison to a morphodynamic active sampling theory. It shows how photomechanical photoreceptor microsaccades enable Drosophila superresolution three-dimensional vision and proposes neural computations for accurately predicting these flies' depth-perception dynamics, limits, and visual behaviors.
Keywords: active sampling; adaptive optics; compound eyes; stereovision.
Conflict of interest statement
The authors declare no competing interest.
Figures
X-ray imaging Drosophila in vivo reveals global mirror-symmetric right and left eye photoreceptor contraction dynamics that tie in with local photomechanical photoreceptor responses. (A) Experiments were performed using synchrotrons (SI Appendix, Figs. S1–S3 and S7–S9 ). (B) ESRF beamline ID16b imaging configuration, using 100-nm resolution. (C) X-rays evoked fast synchronized mirror-symmetric photoreceptor (R1–R8) contractions inside the left and right eyes, causing the photoreceptors to sweep in global back-to-front vergence motion (arrows). (D) Photoreceptor movement began <10 ms from the X-ray onset, increasing with intensity until saturating. (E) The longest frontal forward-facing photoreceptors (15) moved the fastest, ∼15 to 20 µm/s. (F) High-speed light microscopy of R1–R8 rhabdomere photomechanics to blue-green flashes under deep-red antidromic illumination (740-nm LED + 720-nm edge filter), with a fly held in a pipette tip. (G) A 200-ms blue/green flash, delivered orthodromically (through the microscope optics) into the left fly eye (Above), excited local photoreceptors (orange highlight) to twitch photomechanically in a back-to-front direction (arrow). (H) Rhabdomeres moved only in the ommatidia facing the incident blue/green flash from above and remained still in the other ommatidia. Thus, R1–R8 motion did not involve intraocular muscles (each eye has a pair) (14), which otherwise would have moved the whole retina (15). (I) Local blue/green light–induced photoreceptor movements’ early fast phase depended upon the light intensity and closely resembled those evoked by X-rays (D). (G and H) R1–R8 of one ommatidium contracted together as a unit if any of their R1–R8 alone saw light changes, indicating intraommatidial mechanical photoreceptor coupling; see SI Appendix, Figs. S32 and S33 . (J) The experimental X-ray wavelength peak was ∼6,900 times shorter than R1–R6s’ peak sensitivity (∼480 nm).
X-rays activate phototransduction. (A) Remote-controlled stimulation and recording system for head-fixed Drosophila, including two piezo micromanipulators, an ERG amplifier and a white LED, fitted in the DESY P10 beamline. (B) Microelectrodes recorded the fly eyes' combined response, ERG, to white-light and X-ray pulses. (C) Wild-type ERGs to a white light (i) and X-rays (ii) show on and off transients, indicating normal histaminergic synaptic transfer. Hyperpolarizing photoreceptor ERG component and (iii) R1–R8 photomechanical contraction increased with X-ray intensity. (iv) White-light ERG control recorded 20 s after the X-rays. (D) hdcJK910 (i) white and (ii) X-ray ERGs lacked on and off transients, indicating missing synaptic transfer. (ii) ERG photoreceptor component and (iii) R1–R8 photomechanics increased with X-ray intensity. (iv) hdcJK910 white ERG control recorded 20 s later. (E) Blind norpAP24 do not generate ERG responses or photomechanical photoreceptor contractions to white-light or X-ray pulses. (F) Blind trp;trpl do not generate ERG responses while their photoreceptors contract photomechanically to white or X-ray pulses but in a less coordinated way. (C–F) In the R1–R6/LMC cartoons, green indicates the normal function, gray R1–R6 no contraction, and the black LMC no synaptic output. (G) Wild-type and hdcJK910 ERG photoreceptor components increased sigmoidally with X-ray intensity (following a logistic dose–response function), while those of norpAP24 and trp;trpl did not respond. (H) Wild-type and hdcJK910 photomechanical responses grew sigmoidally with X-ray intensity (here approximated with Michaelis–Menten kinetics), while those of norpAP24 did not respond. The maximal X-ray–induced photoreceptor contraction in trp;trpl (orange) was comparable to the wild type and hdcJK910. (G and H) The normalized maximum intensity corresponds to 2.2 × 106 photons/s/μm2.
Left and right eye photoreceptor RFs overlap frontally and move mirror symmetrically, tracing forward translation-induced optic flow. (A) A goniometric high-speed imaging system for mapping photoreceptors’ RFs. Inset: Infra-red (IR) back-lit R1–R7/8 photoreceptor rhabdomeres, forming the left and right eye DPPs (23) (circled), ∼10× magnified by the ommatidial lenses. Each eye’s DPP shows rhabdomeres from neighboring ommatidia that collect light from overlapping RFs (superposition). (B) Rotating the fly head through its central x,y axes revealed its DPPs’ stereoscopic visual field (vine color); see Movie S3 . (C) Because the frontal photoreceptors’ RFs overlap binocularly (∼23.5° azimuth, ∼180° elevation), these mirror-symmetric pairs could enable depth perception. (D) Ommatidial lenses invert the eyes’ fast up medially recoiling microsaccades (DPP fast phase; big arrows), evoked by a 10-ms light flash within their overlapping RFs, to sweep their respective RFs down laterally (small arrows). (E) Microsaccade fast-phase directions mapped across the left (red) and right (blue) eyes; slower-phase return in the opposite direction (compare Movie S4 ; mean of 5 ♂ flies). (F) Brightening (10-ms light flash) contracts R1–R8 front to back (fast phase), and darkening returns them back to front (slower phase); their RFs move in the opposite directions. The mean (black) and 14 consecutive R1–R7/8 contractions (light gray), recorded through cornea-neutralized optics (compare Fig. 1G); see SI Appendix, Figs. S23 and S24 for fully light-adapted dynamics (Movie S5 ). (G) The corresponding slower-phase RF vector map (Left) compared to the forward flying fly’s optic flow field (Center), as experienced with the fly head upright. Their difference (error) is shown for the slower and fast phases. The fast phase matches the ground flow, the slower phase the sky flow. (H) By adjusting microsaccadic sampling to optic flow through head tilt, a fly can actively keep the passing world longer within its photoreceptors’ RFs, which theoretically should improve acuity to resolve the world in motion (SI Appendix, Figs. S56–S61 and
Movie S6 ). d, dorsal; a, anterior; v, ventral viewpoints. (I) Upright (0°) head, and normal tilting around this position (yellow), keep RFs’ fast and slower phases in a balanced push–pull sampling state. Optimizing vision for specific behaviors, like object tracking, requires further self-adjustments in locomotion speed and head and body movements (Movie S7 and
SI Appendix, Figs. S25–S27 ).
Hyperacute L2-terminal sensitivity follows microsaccade directions. (A) A UV fly saw ultrafine (∼0.5°-pixel resolution) UV stimuli on a 150° × 50° screen 38 mm away, while its L2 neurons’ GCaMP6f-fluorescence changes (Ca2+ responses) were recorded by high-speed two-photon imaging. In UV flies (25), UV-sensitive Rh3-opsin is expressed in R1–R6s, containing nonfunctional Rh1-opsin (ninaE8). (B) Each L2 RF samples information from six optically superimposed R1–R6 RFs. L2 retinotopy through axonal crossing: Distal lamina L2s project terminals to the frontal medulla. Inset: Single L2-terminal Ca2+-fluorescence responses to UV stimulation were analyzed as regions of interest (ROI) (red). (C) L2-terminal responses resolve in time hyperacute moving bars (here, showing a larger second-bar response) and black-and-white gratings (interbar distance <4.5°, gray), crossing their RFs, over a broad range of orientations and velocities. (D) L2 resolvability for a dynamically narrowing grating, moving 20.9°/s. Red arrow indicates the finest resolvable angle (interbar distance, as a rounded-up conservative estimate). (E) Interbar distance resolvability depends on stimulus velocity. L2s’ GCaMP6f readout resolved hyperacute patterns moving 60°/s. Note, the finest L2 resolvability, ∼1.09°, approaches the visual display’s two-pixel limit (∼0.5° pixels) and that L2 voltage can encode even faster/finer inputs (34, 35). (F) An L2 terminal’s motion-direction sensitivity map is broadly hyperacute, here primarily along the vertical axis (the black line shows its fitted orientation tuning). The map shows the finest resolvable interbar distances to a dynamically narrowing moving grating stimulus (C–E), covering 360° directions at different speeds. (G) Neighboring L2 terminals show a gradual shift in their dominant motion-direction sensitivity (black arrows; see SI Appendix, Section IV for analytical details). (H) Drosophila's combined L2-terminal motion-direction sensitivity map for the tested left eye region shows retinotopic organization (Left, n = 4 flies), mainly colinear to the corresponding left eye microsaccade directions (Right, compare Fig. 3E). (I) Eye-location–specific L2-terminal direction sensitivities map R1–R8 microsaccade directions. Thus, L2 terminals collectively generate a high-resolution neural representation of the moving world, enhancing visual information transfer during forward locomotion. The dotted rectangle specifies the visual area covered by the display screen.
Forward-facing binocular photoreceptors' biophysically realistic multiscale modeling predicts phasic motion disparity for hyperacute stereopsis. (A) With the corresponding left- and right-eye photoreceptors being a fixed distance, k, apart, their RF disparities inform about the object depth, z. (B) R1–R8’s beam-traced (40) RFs (half-width circular cuts of broadly bell-shaped functions; right eye, blue; left eye, red) tile the fly's visual fields over completely; shown at virtual planes 200- and 5-mm depths from the eyes. (C) R1–R7/8 rhabdomeres of each paired left and right ommatidia lay mirror symmetrically (compare Movies S3 and S9 ). Because rhabdomeres are of different sizes (11) and distances away (22) from the ommatidium center, so too are their projected RFs (B). Therefore, in the neural superposition pooling, the resultant R1–R7/8 RFs do not overlay perfectly into one 4.5° pixel (classic view), but instead tile over completely each small area in the eyes’ visual fields. (D) Phasic voltage response differences of binocularly paired photoreceptors enhance object resolvability in time and carry information about the object depth, z, to the fly brain (SI Appendix, Figs. S55–S61 ). Two dots, 3.5° apart moving left to right at 50°/s, cross binocular RFs of the corresponding left and right R6s 25 mm away. The resulting mirror-symmetric microsaccades make the RFs move along (right R6) and against (left R6) the passing dots, shaping their light inputs and voltage outputs (Movie S10 ). (E) The proposed binocular mirror-symmetric microsaccadic sampling model (SI Appendix, Fig. S59 ) translates the depth of a moving object into the distance in neural time. The closer the object to the fly’s eyes, the shorter the time difference between the responses. Error bars indicate stochastic jitter. (F) The model predicts that Drosophila cannot estimate the depth of more distant objects accurately. The error is >10% when an object is >100 mm from the fly eyes, comparable to their distance discrimination estimate (2 to 200 mm) (41).
Hyperacute stereopsis requires two eyes with mirror-symmetric microsaccades. (A and B) In a flight simulator, a torque-meter–tethered flying Drosophila controls how a white cup rotates around it, showing three black dots (3.9° Ø), 90° apart, one with a black 4-mm center pin (1° Ø). Axially 25 mm away, the 3D pin at −90° (Left), 0° (Middle), or 90° (Right) dot is monocularly indistinguishable from the 2D dots. (C) Yet, flies fixate more on the pins than on the competing dots, implying 3D-pin salience. A single fly’s (Above) and population (Below) frontal fixation probability to the Left, Middle, and Right pin/dot positions, and during a blank control. (D) Fixation probabilities for the three pin positions, blank control subtracted to minimize experimental bias. (E) Positional salience (e.g., Left pin vs. Left dots, Above) and competition (e.g., Left pin vs. Middle and Right dots, Below) statistics indicate that Drosophila see hyperacute 3D pins among 2D dots (superresolution stereopsis). (F) Testing hyperacute 2D-object detection. (G) Old theory simulation: A fly with static 2D vision and 4.5° ommatidial pixelation cannot detect a black 0.98° dot hidden among 1.2° stripes, as its optically corrected contrast difference over a photoreceptor’s RF (5.4° half-width) is only ∼1.6% of that of the stripes alone, evoking response differences < voltage variation (noise) for such a contrast pulse (green, 100 ms). (H) Nevertheless, flies fixate on the hidden dot, irrespective of its position. A single fly’s (Above) and population (Below) frontal fixation probability to the Left, Middle, and Right dot positions, and during a stripe control. (I) Fixation probabilities for the three dot positions, stripe control subtracted to minimize bias. (J) Positional detection (e.g., Left dot vs. Left stripes, Above) and salience (e.g., Left dot vs. Middle and Right stripes, Below) statistics/trends indicate that Drosophila find hyperacute dots visually interesting. (K) Drosophila learns to avoid hyperacute 3D pins or 2D lines/dots (Above and Middle), associated with infrared (IR)-heat punishment (training, triangles), equally well to the classic T vs. ┴ conditioning (Below). (L) One-eye–painted Drosophila fails to learn hyperacute 3D- and 2D-object avoidance (Above and Middle), demonstrating that superresolution stereovision requires two eyes. Yet monocular Drosophila shows normal T vs. ┴ conditioning (Below), indicating that one eye is enough to learn large 2D patterns, consistent with retinal-position–invariant pattern recognition (51). (M) Blind hdcJK910 (Above) and norpA36 (Middle and Below), with no synaptic photoreceptor outputs but normal audition and olfaction, failed to learn the test stimuli, validating that the wild-type learning (K and L) was visual. (N) Rh1-rescue norpA36 with functional R1–R6 photoreceptors and normal mirror-symmetric left and right eye microsaccades learned hyperacute 3D stimuli (Above) and large 2D patterns (Below), but less well than wild type. Thus, R7/R8s also contribute to stereopsis. (O) ∼10% of Rh1-rescue norpA36 flies showed only left or right eye lateral microsaccade components, leading to asymmetric and asynchronous binocular sampling. These flies neither learned hyperacute 3D stimuli (Above) nor large 2D patterns (Below), meaning, mirror-symmetric microsaccadic sampling is necessary for hyperacute stereopsis. (P) norpAP24 with rescued R7/R8 photoreceptors, showing normal microsaccades, learned to differentiate both coarse 2D and hyperacute 3D patterns. Thus, R7/R8s alone are sufficient for hyperacute stereopsis. Note, the exact microsaccadic movements during the experiments are unknown as it was too difficult to measure the DPP movement concomitantly.
References
-
- Livingstone M., Hubel D., Segregation of form, color, movement, and depth: Anatomy, physiology, and perception. Science 240, 740–749 (1988). - PubMed
-
- Nityananda V., Joubier C., Tan J., Tarawneh G., Read J. C. A., Motion-in-depth perception and prey capture in the praying mantis Sphodromantis lineola. J. Exp. Biol. 222, jeb198614 (2019). - PubMed
-
- Poggio G. F., Motter B. C., Squatrito S., Trotter Y., Responses of neurons in visual cortex (V1 and V2) of the alert macaque to dynamic random-dot stereograms. Vision Res. 25, 397–406 (1985). - PubMed
MeSH terms
LinkOut - more resources
Full Text Sources
Molecular Biology Databases
