1. Neuroscience
Download icon

Complementary mechanisms create direction selectivity in the fly

  1. Juergen Haag
  2. Alexander Arenz
  3. Etienne Serbe
  4. Fabrizio Gabbiani
  5. Alexander Borst Is a corresponding author
  1. Max Planck Institute of Neurobiology, Germany
  2. Baylor College of Medicine, United States
Research Article
Cited
9
Views
1,766
Comments
0
Cite as: eLife 2016;5:e17421 doi: 10.7554/eLife.17421

Abstract

How neurons become sensitive to the direction of visual motion represents a classic example of neural computation. Two alternative mechanisms have been discussed in the literature so far: preferred direction enhancement, by which responses are amplified when stimuli move along the preferred direction of the cell, and null direction suppression, where one signal inhibits the response to the subsequent one when stimuli move along the opposite, i.e. null direction. Along the processing chain in the Drosophila optic lobe, directional responses first appear in T4 and T5 cells. Visually stimulating sequences of individual columns in the optic lobe with a telescope while recording from single T4 neurons, we find both mechanisms at work implemented in different sub-regions of the receptive field. This finding explains the high degree of directional selectivity found already in the fly’s primary motion-sensing neurons and marks an important step in our understanding of elementary motion detection.

https://doi.org/10.7554/eLife.17421.001

eLife digest

The brain extracts information from signals delivered from the eyes and other sensory organs in order to direct behavior. Understanding how the interactions and wiring of a multitude of individual nerve cells process and transmit this critical information to the brain is a fundamental goal in the field of neuroscience.

One question many neuroscientists have tried to understand is how nerve cells in an animal’s brain detect direction when an animal sees movement of some kind – so-called motion vision. The raw signal from the light receptors in the eye does not discriminate whether the light moves in one direction or the other. So, the nerve cells in the brain must somehow compute the direction of movement based on the information relayed by the eye.

For more than half a century, major debates have revolved around two rival models that could explain how motion vision works. Both models could in principle lead to neurons that prefer images moving in one direction over images moving in the opposite direction – so-called direction selectivity. In both models, the information about the changing light levels hitting two light-sensitive cells at two points on the eye are compared across time. In one model, signals from images moving in a cell’s preferred direction become amplified. In the other model, signals moving in the unfavored direction become canceled out. However, neither model perfectly explains motion vision.

Now, Haag, Arenz et al. show that both models are partially correct and that the two mechanisms work together to detect motion across the field of vision more accurately. In the experiments, both models were tested in tiny fruit flies by measuring the activity of the first nerve cells that respond to the direction of visual motion. While each mechanism alone only produces a fairly weak and error-prone signal of direction, together the two mechanisms produce a stronger and more precise directional signal. Further research is now needed to determine which individual neurons amplify or cancel the signals to achieve such a high degree of direction selectivity.

https://doi.org/10.7554/eLife.17421.002

Introduction

Flies see the world through a hexagonal array of facets each equipped with its own small lens focusing the light onto 8 photoreceptors. Photoreceptors send their axons into the optic lobe, which consists of four consecutive layers of neuropil called lamina, medulla, lobula and lobula plate (Figure 1a). Each neuropil layer is made up of retinotopically organized columns together containing roughly 100 different neurons per column (Fischbach and Dittrich, 1989). Within the optic lobe, visual motion information is extracted in parallel pathways encoding light increments (ON) and decrements (OFF) (Joesch et al., 2010; Eichner et al., 2011; Joesch et al., 2013). Both pathways bifurcate in the lamina and lead, via a set of specific medulla interneurons, onto the dendrites of T4 and T5 cells, respectively (Takemura et al., 2013; Shinomiya et al., 2014). Both T4 and T5 cells exist in 4 subgroups tuned to one of the four cardinal directions of motion and project into four layers of the lobula plate (Maisak et al., 2013) (Figure 1b). There, they form monosynaptic cholinergic excitatory connections with the dendrites of the large-field lobula plate tangential cells (Mauss et al., 2014; Schnell et al., 2012) as well as with lobula plate intrinsic neurons. These intrinsic neurons, in turn, inhibit tangential cells in the adjacent layer (Mauss et al., 2015). Since none of the neurons upstream from T4 and T5 respond to visual motion in a direction-selective way (Behnia et al., 2014; Strother et al., 2014; Meier et al., 2014; Ammer et al., 2015; Serbe et al., 2016), T4 and T5 cells are the first neurons in the processing chain where directional information is represented explicitly (Maisak et al., 2013; Fisher et al., 2015).

Fly optic lobe and visual stimulation.

(a) Circuit diagram of the ON and the OFF pathway of fly motion vision. Directionally selective signals are carried via T4 and T5 cells to four layers of the lobula plate, where T4 and T5 cells with the same preferred direction converge on the dendrites of the tangential cells (yellow). Inhibition is conveyed via local interneurons LPi (red). From Borst and Helmstaedter, 2015. (b) Confocal image of T4 and T5 cells and their directional tuning. The light green bands indicate the dendrites of T4 and T5 cells. The presynaptic terminals of both T4 and T5 cells form four distinct layers within the lobula plate. The inset shows the result of two-photon calcium imaging, revealing four subgroups of T4 and T5 cells tuned to the four cardinal directions. Scale bar, 20 μm. From Borst and Helmstaedter, 2015. (c) Continuous (top) versus apparent motion (bottom), shown as x-t-plots where the luminance distribution is shown along one spatial (x) and the time axis (t). During continuous motion at constant velocity, a luminance profile is smoothly drifting along one direction, giving rise to a slanted bar in the x-t-plot. During apparent motion, the luminance profile is stable for a while and then jumps to a new position. (d) Two mechanisms proposed to account for direction selectivity. In each model, the signal from one photoreceptor is delayed by a temporal filter (τ) and fed, together with the direct signal from the neighboring photoreceptor, into a nonlinearity. In case of preferred direction enhancement (left), the delayed signal (E) enhances the direct signal (D), e.g. by a multiplication (E x D), in case of the null direction suppression (right), the delayed signal (S) suppresses the direct signal (D), e.g. by a division (D/S). (e) Immunostaining of a single T4 dendrite in layer 10 of the medulla (green) covering multiple columns (counterstained against bruchpilot, purple). (f) Setup for telescopic stimulation of single lamina columns. Antidromic illumination of the eye (left) results in parallel beams from the 6+2 photoreceptors in neighboring facets with identical optical axes. These are focused in the back focal plane of the objective projected onto a CMOS camera. In addition, an AMOLED display is coupled into the beam path to precisely stimulate single lamina columns. Lower left inset: The fly eye and the principle of neural superposition. Light rays parallel to each other shown in the same color activate different photoreceptors in neighboring ommatidia that converge onto a single column in the lamina (‘neuro-ommatidium’). Lower right inset: Picture from the CMOS camera, showing the far field radiation pattern of the Drosophila eye. Dot stimuli can be precisely positioned such as to stimulate single lamina columns.

https://doi.org/10.7554/eLife.17421.003

To investigate the mechanism leading to direction selectivity in T4 cells, we applied apparent motion stimuli where, instead of continuously moving (Figure 1c, top), a bar is abruptly stepped from one location to a neighboring one (Figure 1c, bottom). These stimuli lend themselves well to discriminate between preferred direction (‘PD’) enhancement (Hassenstein and Reichardt, 1956) (Figure 1d, left) and null direction (‘ND’) suppression (Barlow and Levick, 1965) (Figure 1d, right) as the response to the sequence can be compared with the sum of the responses to the luminance pulses given in isolation (‘linear expectation’): in case of preferred direction enhancement, the response to the sequence along the preferred direction is larger than the linear sum of responses to the isolated pulses and identical if the sequence is along the null direction; In case of null direction suppression, the response to the sequence along the preferred direction is identical to the linear sum of responses to the isolated pulses and smaller if the sequence is along the null direction.

The columnar organization of the optic lobes stretches from the laminar cartridges through the medulla and into the lobula and lobula plate. Individual T4 and T5 neurons extend their dendrites across multiple columns (Figure 1e) and receive synaptic inputs from different medulla cell types located in different columns relative to the home columns of the T4 or T5 neuron and to each other (Takemura et al., 2013; Shinomiya et al.; 2014). To understand the particular contribution of those inputs it seemed necessary to precisely place the stimuli onto the columnar raster of the fly’s optic lobe.

The structure of the optical system and of the neuronal wiring in flies obeys the neural superposition principle (Kirschfeld, 1967; Braitenberg, 1967). Those photoreceptors R1-6 from 6 neighboring ommatidia that share the same optical axis converge on the same lamina cartridge, thus forming an optical column that represents the unit of spatial resolution (‘neuro-ommatidium’). To visually stimulate these neuro-ommatidia precisely and in isolation, stimulation must consist of parallel rays at angles along the optical axes of those photoreceptors and aligned to the columnar raster. For this, we adopted a telescopic stimulation device (Franceschini, 1975; Schuling et al., 1989) (Figure 1f). As the fly rhabdomeric photoreceptors work as light guides, the raster of optical columns can be visualized by shining light from within the head capsule (antidromic illumination) to align the raster of neuro-ommatidia to the stimulation locations on a micro-display with the help of a CMOS camera.

Results

As a proof of principle, we first expressed the genetically encoded calcium indicator GCaMP6m (Chen et al., 2013) in lamina cells L2, recorded from their terminals in the medulla by 2-photon microscopy (Denk et al., 1990) and stimulated them with light spots of 1176 ms duration covering a single optical column at various positions. L2 cells responded maximally to stimuli at a certain column, with less than 30% response amplitude to stimuli positioned on surrounding columns, and negligible responses (<12%) to stimuli positioned on columns in the next outer ring (Figure 2a). The slight activation of neighboring columns results directly from the optics of the fly’s eye, where the visual fields of single ommatidia have Gaussian sensitivity profiles with an acceptance angle roughly matching the inter-ommatidial separation (Götz, 1965). These experiments illustrate the specificity of the telescopic stimulation to single lamina cartridges and thus optic lobe column with minimal cross-stimulation of neighboring columns at the physical limit of the fly optical system.

Receptive field of L2 (a) and T4 (b) cells.

Three example traces from a single terminal (top, stimulated ommatidium indicated in black) and mean responses (bottom) of L2 cells (a, n = 23 cells from six flies) and T4 cells (b, n = 10 cells from 10 flies) to flicker stimuli presented at 19 different columnar positions. The responses of individual cells were averaged after alignment to the maximum and normalization and are shown in false color code (left) as well as in 3D bar plots (right). In addition, responses are presented as bar plots along the three axes (dashed lines) of the hexagonal array (mean ± SEM). D = dorsal, V = ventral, L = lateral, F = frontal, DL = dorso-lateral, VF = ventro-frontal, VL = ventro-lateral, DF = dorso-frontal.

https://doi.org/10.7554/eLife.17421.004

We next used a driver line specific for those T4 and T5 cells sending their axons into layer 3 of the lobula plate that are hence sensitive to upward motion. We recorded from single T4 cells by selecting individual processes in layer three of the lobula plate and confirmed their preference for luminance increments. Repeating the above experiment with T4 cells, we again found maximal responses to the stimulus placed in a single column. Compared to L2, however, the receptive field was found to be larger, with about 50% amplitude to the stimuli onto the six surrounding columns, and roughly 25% to the next outer ring (Figure 2b). This indicates that T4 neurons pool excitatory synaptic input from more than one column, consistent with their morphology (Figure 1e) and expected from a motion detector that is required to integrate information from spatially offset input.

In order to discriminate between preferred direction enhancement and null direction suppression, we tested T4 cells with three light pulses of 472 ms duration positioned along the dorso-ventral axis of the eye (Figure 3a, left). T4 cells responded to the individual pulses with different amplitudes, depending on the position of the stimulus (Figure 3a, ‘Flicker’; see also Figure 2b). When stimulated sequentially from ventral to dorsal (Figure 3a, ‘Sequence’, top middle), the cell responded more strongly (thick blue line) than expected from the sum of the responses to the individual stimuli (thin blue line). The opposite was observed when we stimulated the cell sequentially from dorsal to ventral (Figure 3a, ‘Sequence’, bottom middle): now the cell responded more weakly (thick red line) than expected from the sum of the responses to the individual stimuli (thin red line). We then calculated the nonlinear response component by subtracting the linear expectation from the actual response and found that both preferred direction enhancement (Figure 3a, ‘Sequence’, top right) and null direction suppression (Figure 3a, ‘Sequence’, bottom right) contributes to the directionally selective responses of T4 cells.

Responses of T4 neurons to apparent motion stimulation.

(a) Response of a single T4 cell recorded in a single sweep to three-step apparent motion stimuli. The image shows the position of the three stimuli. Left: Responses to individual light pulses (‘Flicker’) delivered at the three different positions. The responses are shifted according to the stimulus protocol used for the subsequent apparent motion stimuli. Middle: Responses of T4 to apparent motion stimuli in preferred and null direction (thick line = measured response, thin line = linear expectation, i.e. sum of responses to the single light pulses). Right: Nonlinear response component defined as the difference between measured response and linear expectation. The responses are the mean obtained from n = 3 stimulus repetitions. Similar data were obtained in 12 experiments. (b) Dependence of the nonlinear two-step response component on the position within the receptive field of T4 cells. Left: Responses to individual light pulses at the four positions. Middle: Stimulus arrangement. Right: Nonlinear response component. Data represent the mean ± SEM (n = 15 cells in 11 flies). (c) Responses to 2-step apparent motion stimuli as a function of the onset time delay (top: preferred direction, location −1 -> 0; bottom: null direction, location 2 -> 0). Data represent the mean ± SEM (upper: n = 13 cells from 8 flies; lower: n = 7 cells from 7 flies). (d) Average responses to a flicker stimulus of the central neuro-ommatidium (black), to a two-step (red) and a three-step (green) null-direction apparent motion sequence (mean ± SEM, n = 11 T4 cells from eight flies). (e) Mean responses of T4 cells to pulses presented to the central column and simultaneously to one of the 18 surrounding columns. Responses were averaged after alignment to the receptive field center, normalized to the flicker response to the central column and are shown in false color code and as bar plots along the three axes of the hexagonal array (mean ± SEM, n = 10 T4 cells from 8 flies). Abbreviations as in Figure 2.

https://doi.org/10.7554/eLife.17421.005

We then asked whether these two mechanisms occupied separate or overlapping receptive fields. For that, we presented pulses in four neighboring columns along the dorso-ventral axis, individually (Figure 3b, left) as well as sequences of pairs in adjacent columns (Figure 3b, middle). Stepping the stimulus up and down between the uppermost columns, we observed only null direction suppression but no preferred direction enhancement (Figure 3b, right top). The opposite was true for stimuli at the two lower most locations: here, only preferred direction enhancement was observed, but no null direction suppression (Figure 3b, right bottom). For sequences between the two inner locations, both phenomena were visible (Figure 3b, right middle). We conclude that the different mechanisms are offset in their receptive fields: preferred direction enhancement is shifted towards the ‘null-side’, and null direction suppression towards the ‘preferred side’ with respect to each other.

To assess the time-course of both mechanisms, we presented two pulses of 472 ms duration and delayed one with respect to the other by a variable time-lag (Figure 3c). When we placed the pulses onto the ventral part of the receptive field, delaying the upper with respect to the lower one (PD), responses peaked at a delay of about 500 ms (Figure 3c, top). When the pulses were placed onto the dorsal part of the receptive field, delaying the lower with respect to the upper one (ND), responses were suppressed and returned to base line at a delay larger than 500 ms (Figure 3c, bottom).

We also noticed that this null direction suppression builds up over longer sequences of steps (Figure 3d). When the stimulation of the central column was preceded by the stimulation of one directly adjacent column on the ‘preferred side’, the response to the central pulse was strongly suppressed as before (red trace). This suppression led to a response to the sequence that on average not only falls below the sum of both individual stimulations, but even below the flicker response to the central column (black trace), indicating inhibition. When this null-direction sequence was extended from two to three columns, the resulting response was even smaller (green trace).

The experiments shown in Figure 3c also revealed that for zero onset time differences, i.e. simultaneous stimulation of two columns, the response was suppressed as compared to stimulation of just one column. We tested the receptive field of this suppression of the central column systematically by simultaneously stimulating the central column together with one of the columns in the 2 rings surrounding it (Figure 3e). We normalized the response to the two stimulated columns with respect to the response to the isolated stimulation of the central column. The resulting response in the T4 neuron was found to be suppressed in comparison to the isolated central column response for simultaneous stimulation of the central column together with another one in the dorsal part of the receptive field.

This suggests that employing apparent motion with flicker stimuli of larger bars should only lead to small flicker responses, possibly occluding null-direction suppression. We tested this prediction by presenting bright horizontal bars of different spatial extent at different locations within the receptive field of individual T4 cells on an LED arena (Figure 4a,c). Using a bar size of 180° × 4.5°, indeed, flicker responses of T4 cells were almost undetectable (Figure 4a). Apparent motion stimuli along the preferred direction led to pronounced responses, larger than the linear prediction, while null direction sequences did not suppress the negligible sum of flicker responses by a significant amount (Figure 4b). This was much different when repeating the same stimuli using bars of only 4.5° × 3° instead: now, as in the experiments employing telescopic stimulation of individual columns, strong flicker responses appeared (Figure 4c). Furthermore, in addition to the preferred direction enhancement, pronounced null direction suppression was observed with peak sensitivity within the dorsal part of the receptive field (Figure 4d).

Apparent motion stimulation on a LED arena.

(a) Flicker responses of T4 cells to the presentation of a large horizontal bar (180˚ × 4.5˚) at different elevations on a LED arena, aligned to the elevation evoking the maximum response (n = 17 cells from 5 flies). (b) Corresponding non-linear response components to two-step apparent motion sequences at different elevations on an LED arena in the preferred (blue) and null direction (red). (c,d) as in (a,b) for small horizontal bars (4.5˚ × 3˚) (n = 18 cells from 5 flies).

https://doi.org/10.7554/eLife.17421.006

What is the virtue of having null direction suppression in addition to the preferred direction enhancement? To address this question, we constructed a simple algorithmic model of a motion detector (Figure 5a). In this model, the motion detector receives 3 inputs. A central direct input (D) is flanked by two low-pass filtered inputs: one enhancing input (E) implementing a multiplicative and one suppressing input (S) implementing a divisive non-linearity. This way, the detector is designed to combine a preferred direction enhancement on one and a null direction suppression on the other side within a single motion-sensing unit. Testing the model with single pulses, responses were only detected for positions −1 and 0 (Figure 5b, left). Stimulating the model with 2-pulse-sequences at positions −1 and 0 led to pronounced preferred direction enhancement but no null direction suppression (Figure 5b, top 2 rows). When the 2-pulse-sequences were delivered at positions 0 and 1, no pronounced preferred direction enhancement occurred, but null direction suppression was substantial (Figure 5b, bottom 2 rows). Thus, these simulation results qualitatively match the respective experimental data (compare with Figure 3a and b). We next simulated an array of such units and calculated its responses to grating motion at different speeds and directions. We found that the model exhibits a high degree of directional selectivity over a broad range of velocities, with little responses to null direction motion (Figure 5c, middle left). In contrast, the model without null direction suppression revealed substantial null direction responses almost as large as its preferred direction responses (Figure 5c, middle center). The same was true for a model with no preferred direction enhancement (Figure 5c, middle right). Interestingly, no such qualitative differences were observed at the level of a model tangential cell, i.e. after subtraction of opponent units (Figure 5c, bottom). When we simulated the responses of the model to grating motion into different directions, we found the full model T4 cell to be more sharply tuned to its preferred direction than the one without preferred direction enhancement or the one without null direction suppression (Figure 5d). We conclude that combining preferred direction enhancement with null direction suppression leads to a strong direction selectivity of the primary motion-sensing unit.

Computer simulations.

(a) Model T4 cell combining preferred direction enhancement and null direction suppression (compare with Figure 1d). (b) Left: Responses to individual light pulses (‘Flicker’) delivered at the three different positions. The responses are shifted according to the stimulus protocol used for the subsequent apparent motion stimuli. Middle: Responses of the model unit to apparent motion stimuli in preferred and null direction (thick line = measured response, thin line = linear expectation, i.e. sum of responses to the single light pulses). Right: Nonlinear response component defined as the difference between measured response and linear expectation. Preferred direction enhancement occurs only between positions −1 and 0, null direction suppression only between positions 0 and 1. (c) Comparison of three models (top row): a full model, as in (a), implementing preferred direction enhancement and null direction suppression (left), one with preferred direction enhancement only (center) and one with null direction suppression only (right). Temporal frequency tuning of model T4 cells (middle row) and tangential cells (bottom row) using motion of a sine-grating (spatial wavelength = 50°, contrast = 1.0) along the preferred (PD) and null direction (ND), based on those 3 models. (d) Directional tuning of an array of model T4 cells using the motion of a sine-grating (spatial wavelength = 50°, contrast = 1.0, temporal frequency = 1.0) for the same 3 models.

https://doi.org/10.7554/eLife.17421.007

We next investigated the direction selectivity of T4 and T5 cells by recording their responses to moving gratings presented on an LED arena moving at various speeds and directions. Over a wide range of velocities spanning more than two orders of magnitude, T4/T5 cells responded almost exclusively to upward motion, i.e. along their preferred direction, with little or no responses at all to motion along their null direction (Figure 6a). When stimulating the cells with 1 Hz grating motion at various directions, we found a rather sharp directional tuning with about 90° half-width around its preferred direction (Figure 6b; see also Figure 3g in Maisak et al., 2013). To rule out that the highest degree of directional selectivity in T4/T5 cells is achieved by a hitherto unknown reciprocal inhibition between T4/T5 cells with opposite preferred direction, e.g. via inhibitory lobula plate interneurons (Mauss et al., 2015) we blocked the synaptic output from all T4 and T5 cells by expression of tetanus toxin light-chain (Sweeney et al., 1995). Having confirmed the effectiveness of the block (Supplementary file 1), we repeated the above experiments. T4/T5 cells revealed the same high degree of directional selectivity as before (Figure 6c,d). From this, we conclude that the mechanism underlying the high degree of directional selectivity in T4 cells does not include the output of oppositely tuned T4 cells but rather originates in the dendrite of the cells.

Response properties of T4/T5 neurons.

(a) Temporal frequency tuning of T4/T5 cells in control flies showing the normalized ΔF/F responses to square-wave gratings in the preferred (up, blue) and null direction (down, red) (n = 15 flies). (b) Directional tuning of T4/T5 cells to square-wave gratings moving at a temporal frequency of 1 Hz (n = 7). (c,d) Temporal (c, n = 9) and directional tuning (d, n = 9) of T4/T5 cells with their synaptic output blocked by expression of the tetanus toxin light chain. All data represent the mean ± SEM (shaded).

https://doi.org/10.7554/eLife.17421.008

Discussion

In the Hassenstein-Reichardt model (Hassenstein and Reichardt, 1956) as well as in the Barlow-Levick model (Barlow and Levick, 1965) direction selectivity emerges by a nonlinear interaction of asymmetrically filtered signals from adjacent image points: an enhancement along the preferred direction in the first, and a suppression along the null direction in the second model. Both models lead to weakly direction-selective signals in the first place, which, in the Hassenstein-Reichardt model, are improved downstream by subtraction of oppositely tuned components (for review, see Borst and Helmstaedter, 2015. Surprisingly, however, a high degree of directional selectivity is found already at the first stage where directional responses are observed, i.e. in T4 and T5 cells (Maisak et al., 2013) (Figure 6). This can now be explained by the fact that both preferred direction enhancement and null direction suppression are implemented in T4 cells. While the output of the full Hassenstein-Reichardt-correlator after the subtraction stage shows a high degree of directional selectivity in the absence of null-direction suppression, the relatively small differences between large, but poorly tuned signals (see Figure 5c, middle panel) would be highly prone to noise. Improving the direction-tuning already at the level of the half-detectors by the additional null-direction suppression increases robustness to noise and might in addition be energetically less costly.

A recent study addressing the mechanisms underlying the elementary motion detectors in Drosophila concluded preferred direction enhancement as the sole mechanism (Fisher et al. 2015). However, we find that their stimulation paradigm consisting of flashes at 2 positions simultaneously does not allow to discriminate between different mechanisms including lateral inhibition, adaptation and null direction suppression.

Since the operations of preferred direction enhancement and null direction suppression have offset receptive fields, it requires the cells to collect their input from more than two adjacent image points, in line with anatomical observations (Figure 1e; Takemura et al., 2013; Shinomiya et al., 2014) . In contrast to the cellular implementation of the motion detector proposed by Behnia et al, 2014 our findings also require that T4 cells receive input from more than two cell types. This again is in agreement with recent connectomic data (for T4 cells: Louis Scheffer, Janelia Research Campus, personal communication; for T5 cells: Shinomiya et al., 2014). Finally, functional studies where the synaptic input elements onto T4 and T5 neurons were blocked also suggested the involvement of more than two input elements in the ON- (Ammer et al., 2015) as well as in the OFF-pathway (Serbe et al., 2016).

Considering the columnar nature of the input elements and the structure of the T4 dendrites covering multiple columns, we predict, that the inputs representing the enhancing, direct and suppressing input segregate on different sub-regions of the T4 dendrite, with enhancing inputs on the ‘null-side’ and the suppressing inputs on the ‘preferred side’ of the dendrite. Future experiments will have to map the different cell types of the medulla onto the proposed model (Figure 5a) as well as to identify the transmitter receptors, responsible for enhancing, exciting and inhibiting the dendrites of T4 and T5 cells.

Materials and methods

Flies

(Drosophila melanogaster) were raised at 25°C and 60% humidity on a 12 hr light/12 hr dark cycle on standard cornmeal agar medium. For calcium imaging of T4/T5 cells, flies were used to express the genetically-encoded calcium indicator GCaMP6m (Chen et al., 2013) in T4/T5 neurons with axon terminals predominantly in layer 3 of the lobula plate (w-; Sp/cyo; VT50384-lexA, lexAop-GCaMP6m/TM6b). For the imaging experiments where synaptic output of T4/T5 activity was blocked, the above flies were crossed to flies driving expression of the tetanus toxin light chain in all T4/T5 cells (w-; R59E08-AD / UAS-TNT-E; R42F06-DBD / VT50384-lexA, lexAop-GCaMP6m). For imaging L2 cells, we used the 21D-Gal4 driver line (Joesch et al., 2010) to express GCaMP6m.

Calcium imaging

Fly surgeries were performed, and the neuronal activity was measured from the left optical lobe on a custom-built 2-photon microscope as previously described (Maisak et al., 2013). Images were acquired at resolutions between 64 × 64 and 256 × 256 pixels and frame rates between 1.88 and 15 Hz with the ScanImage software (Pologruto et al., 2003) in Matlab.

Visual stimulation with telescope

Antidromic illumination of the fly’s head through the objective used for two-photon-microscopy visualizes the hexagonal mosaic of the optical axes of the ommatidia of a living Drosophila (Franceschini, 1975; Schuling et al., 1989). The far field radiation pattern (FFRP) visible in the back focal plane of the objective (LD Epiplan 50x/0,50, Zeiss) is projected onto a CMOS camera (DCC1545M, Thorlabs) via several lenses, a beam splitter (CM1-BP145B5, Thorlabs) and a diaphragm, to reduce stray light. Visual stimuli are generated on the AMOLED display (800 × 600 pixels, pixel size 15 × 15 μm, maximal luminance > 1500 Cd/m2; lambda = 530 nm; refresh rate 85 Hz) (SVGA050SG, Olightek). Both stimulus pattern and FFRP can be visualized simultaneously by means of the beam splitter and a mirror with the CMOS camera. This allows to precisely position the stimuli onto the FFRP. In order to prevent stimulus light from entering the photomultiplier of the two-photon microscope, light generated by the AMOLED display was filtered with a long pass filter (514 LP, T: 529.4–900 nm, AHF). The AMOLED display was controlled with MATLAB and the psychophysics toolbox (V 3.0.11; Brainard, 1997).

Experimental protocol

We determined the receptive field of T4 cells by stimulating single cartridges with light pulses of 472 ms duration at randomized positions. At each position, three stimulus presentations were delivered. The resulting responses were averaged and the peak of the averaged response was taken. We performed all experiments on T4 cells only. The cells were selected based on their response to light-on stimuli. While T4 cells respond to the onset of a light pulse (Figure 2b), the T5 cells respond to the light off. Apparent motion stimuli consisted either of consecutive light stimuli to two or three neighboring cartridges. The second stimulus was presented right after the first turned off, resulting in a delay from onset to onset of 472 ms.

Visual stimulation with LED arena

The LED arena subtended approximately 180° in azimuth and 90° in elevation with a resolution of 1.5°, based on a design modified from Reiser and Dickinson (2008) as previously described (Maisak et al., 2013). Stimuli were presented with 3–5 repetitions per experiment in a randomized fashion. All stimuli were presented in full contrast. To measure the directional and temporal frequency tuning, square-wave gratings with a spatial wavelength of 24° spanning the full extent of the stimulus arena were used. For the direction tuning those gratings were moved in 12 directions separated by 30° at a temporal frequency of 1 Hz. To determine the temporal frequency tuning, gratings were moved at temporal frequencies ranging 0.05 to 20 Hz moving in the preferred and null direction. For the apparent motion experiments, either large (180° wide × 4.5° high) or small (4.5 × 3°) bright horizontal bars were presented for 400 ms either in isolation (flicker) or in sequences of 2 pulses (apparent motion) offset by 4.5° in the preferred or null direction with a ∆t, onset to onset, of 400 ms.

Data analysis

Data analysis was performed offline using custom-written routines in Matlab. Regions of interests (ROIs) were selected by hand in layer 3 of the lobula plate. The time courses of relative fluorescence changes (ΔF/F) were calculated from the raw imaging sequence. Responses to the stimulus were baseline-subtracted, averaged across repetitions, and quantified as the peak responses over the stimulus epochs. For T4 cells, the baseline was determined during one second before the stimulus, for L2 cells (Figure 2a), it was determined during the last second before luminance off-set. Those responses were averaged across experiments. Where indicated, responses were normalized to the maximum average response before averaging. For the apparent motion experiments, non-linear response components were calculated as the differences of the time-courses of the responses to the apparent motion stimuli and the sum of the appropriately time-shifted responses to flicker stimuli at the corresponding positions.

Model simulations

Visual stimuli were represented as a two-dimensional array (200 x 300) at 1° spatial and 10 ms temporal resolution, mapped onto a linear array of 40 visual columns. The signal amplitude ranged from 0 to 1. The input to the ON pathway, as represented by lamina neuron L1, was modeled as previously described (Eichner et al., 2011). Briefly, the local luminance signal was high-pass-filtered (1st order, τ = 250 ms), and 10% of the DC value was added with subsequent rectification. The T4 cell was then modeled as receiving input from three adjacent columns (Figure 5a): an enhancing signal E at position −1, representing the low-pass filtered signal of L1 (1st order, τ = 250 ms), a direct signal D at position 0 which is identical to L1, and a suppressing signal S again representing the low-pass filtered signal of L1 (1st order, τ = 250 ms). The response was calculated as the product of E and D, divided by S. Signals had the following weights k: kE = 5, kD = 5, kS = 10. To avoid division by zero and to account for flicker responses, a DC term of 1.0 was added to each signal. To simulate flicker and apparent motion stimuli (Figure 5b), light pulses of 450 ms length and amplitude 1.0 were delivered to selected columns and the response of an individual model T4 cell was evaluated. To simulate the responses to sine-gratings (Figure 5c,d), responses were calculated either as the summed responses of an array of such units (‘T4 response’) or after subtracting the signals of oppositely tuned T4 cells (‘VS response’). Simulations without null direction suppression or preferred direction enhancement were performed by setting either kE or kS to zero. Software was written in the Python programming language and is available as source code.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
    Photoreceptor Optics
    1. N Franceschini
    (1975)
    98–125, Sampling of the visual environment by the compound eye of the fly: fundamentals and applications, Photoreceptor Optics, 10.1007/978-3-642-80934-7_6.
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30

Decision letter

  1. Fred Rieke
    Reviewing Editor; Howard Hughes Medical Institute, University of Washington, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Complementary Mechanisms Create Direction Selectivity in the Fly" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom, Fred Rieke, is a member of our Board of Reviewing Editors and the evaluation has been overseen by David Van Essen as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Tom Baden (Reviewer #2); Vivek Jayaraman (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

This is an interesting and timely paper investigating the mechanisms underlying motion detection in the fly visual system. The paper uses a powerful combination of two-photon calcium imaging and telescopic stimulus delivery, allowing greater spatial control of stimuli than more conventional approaches. All of the reviewers found the work of interest and the experiments generally well done. Several issues came up in the review process: (1) some cases in which the text stated results too strongly given a lack of direct experimental evidence; (2) a need to improve the clarity of the writing and figures. Four specific examples follow, agreed upon as most critical by all three reviewers; others examples are in the individual reviewers comments below.

1) The paper concludes that distinct regions of the T4 dendrites implement different motion detection mechanisms. But there are no direct measures of dendritic signaling in the paper; without such measures, this conclusion should be stated less strongly.

2) The paper suggests a role of inhibition in canceling signals for null direction motion. The evidence for this is based on sublinear summation of signals in the null direction – but this does not uniquely identify inhibition as the underlying mechanism. Was it possible to directly observe inhibition – e.g. to see a decrease in calcium when stimulating a single location in the T4 receptive field with flickering stimuli? Without a direct observation about inhibition this conclusion should also be stated less strongly.

3) The telescopic illumination should be described in more detail. Particularly how effective is it in restricting input spatially? If there is spread beyond the single column targeted, how much could such spread impact the conclusions of the paper? Part of the confusion on this point originates because Figure 1e is difficult to interpret in present form (see point 4 and individual reviewer comments).

4) The figures should be modified to specify exactly what is measured, particularly the optical configuration for each experiment. Specific examples are in the individual reviews below.

Reviewer #1:

This paper describes the origin of directional selectivity in the first cells in the fly visual system that exhibit directional responses. The combination of a telescopic stimulus delivery system to activate specific elements of the motion computation with functional calcium imaging is quite powerful. The results are clear and interesting. However, I found several aspects of the paper difficult to follow. Some specific suggestions to improve clarity follow.

1) Manipulations of stimuli to specific spatial regions are crucial to the paper. But it is often hard to discern exactly what stimulus is used in a given experiment. This could be improved greatly by adding simple schematics to the figures, showing which specific columns are being activated. When those are present (e.g. Figure 2d) they help quite a bit. Figure 2b, e and f are specific examples in which this could help.

2) The description of Figure 1e and f is quite minimal and hard to follow. Those figures made much more sense to me after I had read most of the paper, but when they first come up it is not clear what is important about the data, or even how the data was collected. The larger receptive field size of the T4 cells is important; that should get mentioned in the text where this data is described.

3) Figure 3 models. It was not clear to me what the modified models were. Specifically, do the models lacking null direction suppression or preferred direction enhancement still consist of three inputs (as in Figure 3a) just with one arm disabled? Or are they models with two inputs only (in a more classic Reichardt configuration)? A schematic of these models added to Figure 3 could be helpful.

4) Figure 3d. I am confused by the description of this figure. The legend does not have a panel d, but the legend for panel e suggests it is describing an array of model units, not a full T4 cell. That appears to make sense from comparing Figure 3c and d. But the main text describes a full model T4 cell. Because of these inconsistencies I could not evaluate that part of the paper.

Reviewer #2:

In their MS "Complemetary Mechanisms Create Direction Selectivity in the Fly", Haag et al. use a spatially precise optical setup capable of delivering single-ommatidial stimulation and with 2-photon calcium imaging to probe mechanisms driving selection selectivity in the visual system of the fruit fly. Focussing on T4 cells, the fly-visual system's "ON-DS cells", the authors show that both null-suppression as well as preferred direction enhancement appear to drive directionality in these neurons, computations that likely take place in different dendritic regions of T4s. They follow up their findings in a simple model, which in addition predicts that the 2 mechanisms in tandem yield a sharper spatial tuning than any one mechanism alone. They also demonstrate the "sharp" tuning in T4 cells experimentally (though they do no isolate the 2 mechansims in these experiments – something that I suspect would be quite difficult to do).

Overall, the MS is timely and reports important and to my knowledge novel findings about direction selective visual circuits of the fly. The data is of high quality and the writing is clear and succinct. I only have minor comments

Reviewer #3:

In this tantalizing manuscript, Haag, Borst and colleagues explore the underlying basis of visual motion direction computations in the Drosophila optic lobe. In both the vertebrate retina and in the early visual system of flies, this computation has been studied in the framework of two classic models of visual motion direction: the Hassenstein-Reichardt (H-R) model and the Barlow-Levick (B-L) model. Mechanistically, both rely on combining input from spatially offset pathways, with input from an adjacent location temporally delayed relative to the central pathway. One (H-R) produces an excitatory "preferred-direction" (PD) response by multiplication with temporally delayed input from a preceding location, while the other (B-L) uses division by input from a temporally delayed location in the "null direction" (ND). Direct evidence for the latter has been obtained in direction-selective retinal ganglion cells (DSRGCs), where starburst amacrine cells provide the appropriate ND inhibition. In the Drosophila optic lobe, the H-R model has thus far been the mechanism most often proposed for direction selectivity for large field motion, with information from "ON" and "OFF" motion pathways represented by so-called T4 and T5 cells respectively being combined in vertical system (VS) and horizontal system (HS) lobula plate tangential cells (LPTCs). Recent anatomical and electron microscopy (EM) evidence has suggested a more complex picture at the level of the input to T4 and T5 neurons – their inputs comprise more cell types than previously believed. In this study, the authors use precise optical stimulation that takes into account the fly's neural superposition eye (Franceschini, 1975) along with two-photon calcium imaging to suggest that B-L-like inhibition may combine with H-R like excitation to shape the responses of T4 neurons. The results are interesting but the evidence provided is indirect and the picture still somewhat murky. Detailed comments follow.

The major claim of the paper is that precise stimulation of single lamina columns reveals PD and ND responses that localize to different sub-regions of the T4 neuron dendrites. However, this claim, which is highlighted in the abstract, goes beyond the evidence presented, which is entirely based on recordings in T4 axons. Unless the authors can present direct anatomical and physiological evidence for such localized dendritic input, I would suggest that this claim be removed or significantly tempered and moved to Discussion.

Similarly, the B-L model proposes inhibition of one signal by another for ND motion, but the authors present no direct evidence for such inhibition. What is shown instead is supra- or sub-linear summation upon sequential (PD vs ND) excitation of successive lamina columns. Why does this have to necessarily result from inhibition in T4 dendrites rather than have presynaptic origins? While it is true that medulla neurons are not known to be sensitive to motion-direction, such neurons are known to have suppressed responses to bars wider than their receptive field (T5 input neurons for OFF pathway; Serbe et al., 2016 from the Borst lab), so might a skewed sampling of such inputs by T4 perhaps confer the observed ND suppression?

Related to the above point, the model (Figure 3b) predicts that we should be able to directly observe inhibition (not just suppression of a strong excitatory response), even if it is not large, when the 'wrong' lamina column is illuminated in a flicker experiment. This seems never to be observed in the experiments shown (Figure 1a,b). What is seen instead is sub-linearity (or supra-linearity), which the authors call a non-linear response component. Is this a calcium indicator limitation even with the higher-baseline GCaMP6m? If there is inhibition, might it be easier to pick up directly in the dendrites of single T4 neuron or do the authors find this experimentally infeasible?

In Figure 1e, the authors state that their telescopic illumination allows them to illuminate only a single lamina column, but the lamina responses suggest that neighboring columns are also stimulated. In the manuscript (introduction section), these responses are described as 'negligible', but they seem to reach 50% of peak responses. This requires further discussion, because the study's results rely heavily on the high precision of optical illumination. If the responses reflect the limits of the illumination system, it would be useful to discuss how that might affect the interpretation of results. Also, on the same note, it would be great to see what a 'standard' localized stimulus presentation (i.e., targeting single ommatidia and sequences of ommatidia without using the telescopic refinement) would evoke in T4 neurons. This would highlight the importance of telescopic stimulus presentation for visual motion studies, which would be useful to the field.

Although perhaps this is beyond the scope of this study, it would be very interesting to know what T5 neuron responses look like with telescopic stimulation. Do they also show B-L like ND suppression in the OFF pathway? Given that the authors have access to all the T5 inputs (Serbe et al., 2016), this could also provide an easier path to a more complete explanation in these neurons as opposed to T4, where the authors do not say much about what the potential inputs to these neurons might be and do not explore how they might act together to produce the observed response properties.

Lastly, it is still left unclear why T4 (and, perhaps T5) might need a potential B-L like mechanism in addition to H-R if there is no difference between the combination and pure H-R for LPTCs. Perhaps the authors could speculate on this in Discussion.

https://doi.org/10.7554/eLife.17421.011

Author response

1) The paper concludes that distinct regions of the T4 dendrites implement different motion detection mechanisms. But there are no direct measures of dendritic signaling in the paper; without such measures, this conclusion should be stated less strongly.

We recognize that we do not directly image calcium signals originating from the synaptic input within different regions of the dendrites of individual neurons. We rephrased more carefully in terms of “offset receptive fields for preferred-direction enhancement and null-direction suppression"

e.g.: "We conclude that the different mechanisms are offset in their receptive fields: preferred direction enhancement is centered towards the ‘null-side’, and null direction suppression towards the ‘preferred side’ with respect to each other. "

We predict that these offset receptive fields map onto different regions of the dendrite, but moved this argument to the discussion as suggested:

"Considering the columnar nature of the input elements and the structure of the T4 dendrites covering multiple columns, we predict, that the inputs representing the enhancing, direct and suppressing input segregate on different sub-regions of the T4 dendrite, with enhancing inputs on the ‘null-side’ and the suppressing inputs on the ‘preferred side’ of the dendrite. "

2) The paper suggests a role of inhibition in canceling signals for null direction motion. The evidence for this is based on sublinear summation of signals in the null direction – but this does not uniquely identify inhibition as the underlying mechanism. Was it possible to directly observe inhibition – e.g. to see a decrease in calcium when stimulating a single location in the T4 receptive field with flickering stimuli? Without a direct observation about inhibition this conclusion should also be stated less strongly.

We now included additional data to show that apparent motion stimulation in the null direction leads to significantly lower responses than when only the central column was stimulated, and that this suppression builds up with longer stimulation sequences covering more columns in the ND direction (Figure 3d). As far as we can see, this can only be explained by inhibition, not by adaptation or similar phenomena. The nature of the inhibition can be shunting or pre-synaptic on the inputs and does not have to be hyperpolarizing. In the former case, it would also not manifest itself as a decrease in the calcium signal.

3) The telescopic illumination should be described in more detail. Particularly how effective is it in restricting input spatially? If there is spread beyond the single column targeted, how much could such spread impact the conclusions of the paper? Part of the confusion on this point originates because Figure 1e is difficult to interpret in present form (see point 4 and individual reviewer comments).

The imaging experiment from L2 neurons illustrate the specificity of the stimulation and the restriction to single columns with minimal overlap into neighboring columns at the optical limit of the fly’s eye. In the meantime, we also increased the number of L2 neurons recorded. We additionally discuss this point now in more detail in the text:

"The slight activation of neighboring columns results directly from the optics of the fly’s eye where the visual fields of single ommatidia have Gaussian sensitivity profiles with an acceptance angle roughly matching the inter-ommatidial separation (Götz, 1965). These experiments illustrate the specificity of the telescopic stimulation to single lamina cartridges and thus optic lobe column with minimal cross-stimulation of neighboring columns."

The corresponding figure was labeled more clearly, including axis labels and a 3D bar plot to immediately visualize the results.

4) The figures should be modified to specify exactly what is measured, particularly the optical configuration for each experiment. Specific examples are in the individual reviews below.

We thought a lot about introducing xt-plots throughout, but found that it would add too much of space to the figures. As an example, just the experiment shown in Figure 3b would require the symbols below for visualizing the apparent motion sequences (Author response image 1.). We, therefore, decided to introduce the apparent motion stimulus in principle in Figure 1c as an xt-plot, and to be as clear as possible in the text otherwise.

Reviewer #1:

This paper describes the origin of directional selectivity in the first cells in the fly visual system that exhibit directional responses. The combination of a telescopic stimulus delivery system to activate specific elements of the motion computation with functional calcium imaging is quite powerful. The results are clear and interesting. However, I found several aspects of the paper difficult to follow. Some specific suggestions to improve clarity follow.

1) Manipulations of stimuli to specific spatial regions are crucial to the paper. But it is often hard to discern exactly what stimulus is used in a given experiment. This could be improved greatly by adding simple schematics to the figures, showing which specific columns are being activated. When those are present (e.g. Figure 2b) they help quite a bit. Figure 2b, e and f are specific examples in which this could help.

We changed the scheme in Figure 2b (now Figure 3b) to better illustrate that light pulses were stepped between adjacent neuro-ommatidia.

2) The description of Figure 1e and f is quite minimal and hard to follow. Those figures made much more sense to me after I had read most of the paper, but when they first come up it is not clear what is important about the data, or even how the data was collected. The larger receptive field size of the T4 cells is important; that should get mentioned in the text where this data is described.

First of all, we enlarged the figures, added axis labels to the bar histograms and also visualized the entire receptive field as an additional 3D bar plot. This should help quite a bit. We also added the following explaining sentence to the description of the receptive field of T4:

"This indicates that T4 neurons pool excitatory synaptic input from more than one column, consistent with their morphology (Figure 1d) and expected from a motion detector required to integrate information from spatially offset input. "

3) Figure 3 models. It was not clear to me what the modified models were. Specifically, do the models lacking null direction suppression or preferred direction enhancement still consist of three inputs (as in Figure 3a) just with one arm disabled? Or are they models with two inputs only (in a more classic Reichardt configuration)? A schematic of these models added to Figure 3 could be helpful.

We now included schemata to illustrate the configuration of the 3 models as suggested.

4) Figure 3d. I am confused by the description of this figure. The legend does not have a panel d, but the legend for panel e suggests it is describing an array of model units, not a full T4 cell. That appears to make sense from comparing Figure 3c and d. But the main text describes a full model T4 cell. Because of these inconsistencies I could not evaluate that part of the paper.

We use the term model units as a general term for any of the 3 models that we compare. Figure 3d (now 5d) compares the directional tuning of all 3 models, illustrating the sharpest tuning for the full model incorporating both PD enhancement and ND suppression. For clarity, we also included schemata of all 3 models. The typo in the legend was corrected from e to d.

Reviewer #3:

In this tantalizing manuscript, Haag, Borst and colleagues explore the underlying basis of visual motion direction computations in the Drosophila optic lobe. In both the vertebrate retina and in the early visual system of flies, this computation has been studied in the framework of two classic models of visual motion direction: the Hassenstein-Reichardt (H-R) model and the Barlow-Levick (B-L) model. Mechanistically, both rely on combining input from spatially offset pathways, with input from an adjacent location temporally delayed relative to the central pathway. One (H-R) produces an excitatory "preferred-direction" (PD) response by multiplication with temporally delayed input from a preceding location, while the other (B-L) uses division by input from a temporally delayed location in the "null direction" (ND). Direct evidence for the latter has been obtained in direction-selective retinal ganglion cells (DSRGCs), where starburst amacrine cells provide the appropriate ND inhibition. In the Drosophila optic lobe, the H-R model has thus far been the mechanism most often proposed for direction selectivity for large field motion, with information from "ON" and "OFF" motion pathways represented by so-called T4 and T5 cells respectively being combined in vertical system (VS) and horizontal system (HS) lobula plate tangential cells (LPTCs). Recent anatomical and electron microscopy (EM) evidence has suggested a more complex picture at the level of the input to T4 and T5 neurons – their inputs comprise more cell types than previously believed. In this study, the authors use precise optical stimulation that takes into account the fly's neural superposition eye (Franceschini, 1975) along with two-photon calcium imaging to suggest that B-L-like inhibition may combine with H-R like excitation to shape the responses of T4 neurons. The results are interesting but the evidence provided is indirect and the picture still somewhat murky. Detailed comments follow.

The major claim of the paper is that precise stimulation of single lamina columns reveals PD and ND responses that localize to different sub-regions of the T4 neuron dendrites. However, this claim, which is highlighted in the abstract, goes beyond the evidence presented, which is entirely based on recordings in T4 axons. Unless the authors can present direct anatomical and physiological evidence for such localized dendritic input, I would suggest that this claim be removed or significantly tempered and moved to Discussion.

We rephrased this statement in terms of offset receptive fields that are shown in Figure 3b. Given the columnar organization of the input elements and the anatomy of the dendrite, we still find it highly likely that the inputs representing the enhancing and suppressing arm will be separated on the dendrite. This argument, however, is now moved to the Discussion.

Similarly, the B-L model proposes inhibition of one signal by another for ND motion, but the authors present no direct evidence for such inhibition. What is shown instead is supra- or sub-linear summation upon sequential (PD vs ND) excitation of successive lamina columns. Why does this have to necessarily result from inhibition in T4 dendrites rather than have presynaptic origins? While it is true that medulla neurons are not known to be sensitive to motion-direction, such neurons are known to have suppressed responses to bars wider than their receptive field (T5 input neurons for OFF pathway; Serbe et al., 2016 from the Borst lab), so might a skewed sampling of such inputs by T4 perhaps confer the observed ND suppression?

We precisely constructed this telescopic stimulation device to ensure accurate activation of single columns with minimal stimulation of the surround (see L2 imaging experiments in Figure 2) in perfect register with the columnar raster. This point is also illustrated in the recordings using the LED arena. Stimulation with larger bars or not in perfect register (see sharpness of the central peak for small stimuli) leads to a strongly reduced flicker response. However, surround inhibition of presynaptic elements would affect all stimuli regardless of their direction (single flicker, PD and ND) and thus cannot account for the ND suppression seen that is calculated with respect to the individual flicker responses.

Related to the above point, the model (Figure 3b) predicts that we should be able to directly observe inhibition (not just suppression of a strong excitatory response), even if it is not large, when the 'wrong' lamina column is illuminated in a flicker experiment. This seems never to be observed in the experiments shown (Figure 1a,b). What is seen instead is sub-linearity (or supra-linearity), which the authors call a non-linear response component. Is this a calcium indicator limitation even with the higher-baseline GCaMP6m? If there is inhibition, might it be easier to pick up directly in the dendrites of single T4 neuron or do the authors find this experimentally infeasible?

We did not observe an inhibition in terms of a decrease in the calcium signal reflecting a hyperpolarization. Instead, activation of the suppressive input strongly suppresses subsequent stimulation of the direct input, most likely via a shunting inhibition or a presynaptic inhibition of the suppressing input onto the direct input. We included an additional figure (Figure 3d) illustrating that the suppression not just leads to smaller-than-the-sum-responses, but reduces the responses below the level of isolated center stimulation. The figure also shows that this suppression builds up along sequences of steps in the null direction. We do not think it likely that it would be possible to directly observe an inhibition/decrease in the calcium signal in specific subregions of the dendrite that is not visible in the axon terminal or in the neuron close to the lobula plate. Those of the small neurons tested so far in the optic lobe tend to be electrically compact (Yang et al., 2016).

In Figure 1e, the authors state that their telescopic illumination allows them to illuminate only a single lamina column, but the lamina responses suggest that neighboring columns are also stimulated. In the manuscript (introduction section), these responses are described as 'negligible', but they seem to reach 50% of peak responses. This requires further discussion, because the study's results rely heavily on the high precision of optical illumination. If the responses reflect the limits of the illumination system, it would be useful to discuss how that might affect the interpretation of results. Also, on the same note, it would be great to see what a 'standard' localized stimulus presentation (i.e., targeting single ommatidia and sequences of ommatidia without using the telescopic refinement) would evoke in T4 neurons. This would highlight the importance of telescopic stimulus presentation for visual motion studies, which would be useful to the field.

To validate the alignment of the telescopic stimulation device we used a micrometer scale slide. When the first lens after the stimulation objective is taken out, the device changes from a telescope to a microscope. By putting a scale slide in the focal plane of the stimulation objective, we can image the micrometer scale with the CMOS-camera and position light at different positions on the scale. By using the objective used for 2-photon microscopy the positioning of the stimuli on the scale can be visualized with the CCD-camera included in the 2-photon setup. We then aligned the beam path of the stimulation device so that the position of the light stimuli on the CMOS-camera and the CCD-camera coincide.

We now also increased the number of recordings from L2 neurons and show that the stimulation is restricted to single columns, in the limits of the optics of the fly’s eye (below 30%). The responses to the outer ring of neuro-ommatidia around the central column are below 12% and indeed negligible, as stated in the manuscript. A comparison to classical stimulation is given by the LED arena experiments (former Figure 2e & f, now Figure 4). Here we used large and small bars, at different positions limited by the resolution of the LED arena. Presumably inhibitory surrounds of the input elements lead to a strong reduction in the flicker responses of large bars as compared to small bars. With small bars, both a preferred direction enhancement and a null direction suppression could be revealed (Figure 4d). However, in contrast to the telescopic stimulation, the register of the stimulus to the columnar raster has to be indirectly inferred from the flicker amplitude.

Although perhaps this is beyond the scope of this study, it would be very interesting to know what T5 neuron responses look like with telescopic stimulation. Do they also show B-L like ND suppression in the OFF pathway? Given that the authors have access to all the T5 inputs (Serbe et al., 2016), this could also provide an easier path to a more complete explanation in these neurons as opposed to T4, where the authors do not say much about what the potential inputs to these neurons might be and do not explore how they might act together to produce the observed response properties.

We are indeed planning to repeat this study on T5 cells and block their known input neurons in order to find out which neuron is doing what. This, however, will for sure take quite a long time to be completed.

Lastly, it is still left unclear why T4 (and, perhaps T5) might need a potential B-L like mechanism in addition to H-R if there is no difference between the combination and pure H-R for LPTCs. Perhaps the authors could speculate on this in Discussion.

This is an excellent point. We added the following text in the Discussion to address the potential functional consequences: "While the output of the full Hassenstein-Reichardt-correlator after the subtraction stage shows a high degree of direction-selectivity in the absence of null-direction suppression, the relatively small differences between poorly tuned signals would be very sensitive to noise. Improving the direction-tuning already at the level of the half-detectors by the additional null-direction suppression increases robustness to noise and might in addition be energetically less costly. “

https://doi.org/10.7554/eLife.17421.012

Article and author information

Author details

  1. Juergen Haag

    Max Planck Institute of Neurobiology, Martinsried, Germany
    Contribution
    JH, Designed the experiments and evaluations, Performed and analyzed the experiments using the telescopic stimulation, Conception and design
    Contributed equally with
    Alexander Arenz
    Competing interests
    No competing interests declared.
  2. Alexander Arenz

    Max Planck Institute of Neurobiology, Martinsried, Germany
    Contribution
    AA, Performed and analyzed the experiments using the LED arena, Designed the experiments and evaluations, Conception and design
    Contributed equally with
    Juergen Haag
    Competing interests
    No competing interests declared.
  3. Etienne Serbe

    Max Planck Institute of Neurobiology, Martinsried, Germany
    Contribution
    ES, Recorded from tangential cells in T4/T5 block flies
    Competing interests
    No competing interests declared.
  4. Fabrizio Gabbiani

    Baylor College of Medicine, Houston, United States
    Contribution
    FG, Supported by a fellowship of the Alexander-von-Humboldt Foundation and participated in the early phase by constructing the telescopic stimulus device
    Competing interests
    No competing interests declared.
  5. Alexander Borst

    Max Planck Institute of Neurobiology, Martinsried, Germany
    Contribution
    AB, Designed the experiments and evaluations, Wrote the manuscript with the help of all authors, Conception and design, Drafting or revising the article
    For correspondence
    borst@neuro.mpg.de
    Competing interests
    AB: Reviewing editor, elife

Funding

Max-Planck-Gesellschaft

  • Juergen Haag
  • Alexander Arenz
  • Etienne Serbe
  • Alexander Borst

Alexander von Humboldt-Stiftung

  • Fabrizio Gabbiani

Deutsche Forschungsgemeinschaft (SFB 870)

  • Alexander Borst

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We are grateful to Wolfgang Essbauer and Romina Kutlesa for fly work, to Stefan Prech for the 3D illustration in Figure 2, to Jesus Pujol-Marti for providing us with the picture of an individual T4 cell (Figure 1e), to Nicolas Franceschini for advice, to Alex Mauss and Aljoscha Leonhardt for carefully reading the manuscript, and to Winfried Denk for many discussions. This study was funded by the Max-Planck-Society, the Deutsche Forschungsgemeinschaft (SFB 870) and by a fellowship of the Alexander-von-Humboldt Foundation to FG.

Reviewing Editor

  1. Fred Rieke, Reviewing Editor, Howard Hughes Medical Institute, University of Washington, United States

Publication history

  1. Received: April 29, 2016
  2. Accepted: July 11, 2016
  3. Version of Record published: August 9, 2016 (version 1)
  4. Version of Record updated: April 21, 2017 (version 2)

Copyright

© 2016, Haag et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,766
    Page views
  • 378
    Downloads
  • 9
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Comments

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Neuroscience
    Nicolas Michalski et al.
    Research Article Updated
    1. Microbiology and Infectious Disease
    Longhui Zeng et al.
    Research Article Updated