Quantitative predictions orchestrate visual signaling in Drosophila

SUMMARY

Vision influences behavior, but ongoing behavior also modulates vision in animals ranging from insects to primates. The function and biophysical mechanisms of most such modulations remain unresolved. Here we combine behavioral genetics, electrophysiology, and high-speed videography to advance a function for behavioral modulations of visual processing in Drosophila. We argue that a set of motion-sensitive visual neurons regulate gaze-stabilizing head movements. We describe how during flight turns Drosophila perform a set of head movements that require silencing their gaze-stability reflexes along the primary rotation axis of the turn. Consistent with this behavioral requirement, we find pervasive motor-related inputs to the visual neurons, which quantitatively silence their predicted visual responses to rotations around the relevant axis while preserving sensitivity around other axes. This work defines the function of a behavioral modulation of visual processing and illustrates how the brain can remove one sensory signal from a circuit carrying multiple related signals.

Graphical abstract

INTRODUCTION

When animals walk, run, or fly, their sensory systems process information differently than when the same animals are quiescent (Chiappe et al., 2010; Maimon et al., 2010; Niell and Stryker, 2010). Alongside locomotion-related modulations, generalized arousal states also drive changes in sensory processing (Vinck et al., 2015). Such state dependent changes have been described from nematodes to humans (Durgin and Gigone, 2007; Gordus et al., 2015), and they likely serve to promote adaptive behavioral responses to sensory input, depending on the animal’s current behavioral goals.

Locomotor and arousal states typically last many seconds or minutes. Within each state, however, more rapid modulations of sensory processing are also common. In the case of vision, for example, multiple times per second we perform rapid eye movements called saccades. During saccades, the image sweeps quickly over the retina due to our own behavior, making the read-out of visual information about the environment much more challenging. As a result, visual processing is altered in the primate brain during saccades, with some neurons suppressed or enhanced (Burr et al., 1994; Reppas et al., 2002), and others showing even more remarkable alterations to their receptive-field properties (Duhamel et al., 1992; Sommer and Wurtz, 2006). While many saccade-related modulations have been documented, defining their precise function is generally more difficult. Cell-type-specific perturbations and electrophysiology in behaving animals are vital for ascribing function within neurobiological systems. Given that these approaches are mature in Drosophila, this species offers the possibility of defining precise roles for saccadic alterations to visual signaling.

Like primates, flying Drosophila change their point of gaze every second or so (Heisenberg and Wolf, 1979; Tammero and Dickinson, 2002). Because flies have eyes that are rigidly attached to their heads, they perform these rapid gaze changes, also called saccades, largely through head and body movements rather than eye movements. We recently demonstrated that the visual responses of horizontal system north (HSN) cells in Drosophila are modulated during flight saccades (Kim et al., 2015). However, HSN cells are just one cell class in a network of horizontal system (HS) and vertical system (VS) optic flow-processing neurons in flies. What is the normal behavioral role of HS and VS cells? Does this role, once identified, clarify the functional architecture of saccade-related modulations across the entire HS/VS network? These are important questions if for no other reason than the fact that HS and VS cells are probably the most studied visual interneurons in invertebrates. Furthermore, firm answers to these questions in Drosophila may illuminate the function of saccadic modulations in larger brains.

We found that hyperpolarizing, and muting visual responses in, HS/VS cells reduces the size of visually driven stability movements of the head, also known as head optomotor responses. These new data, alongside prior work on the visual tuning properties of HS/VS cells (Hausen, 1982; Krapp et al., 1998), anatomical studies showing that HS/VS cells are gap-junction-coupled to neck motor neurons (Haag et al., 2010; Strausfeld and Seyan, 1985; Strausfeld and Bassemir, 1985; Wertz et al., 2012), and optogenetic activation experiments (Haikala et al., 2013), argue that HS/VS neurons regulate head optomotor responses. We further describe the pattern of head movements Drosophila make during flight saccades, and these movements call for antagonizing the head’s optomotor response along the yaw rotational axis. By performing exhaustive patch-clamp recordings from HS and VS cells in tethered, flying Drosophila, we find that HS and VS visual responses are antagonized by motor-related inputs during saccades, and the stronger an HS/VS cell responds to yaw optic flow, the stronger is the motor-related input it receives during saccades. We further suggest biophysical mechanisms for how the motor-related inputs modify visual signaling, which differ systematically across cell types. In sum, we argue that motor-related modulations of optic flow processing in Drosophila represent quantitatively tailored efference copies that function to mute yaw-axis visual responses–in a system that carries yaw, pitch and roll signals–so as to transiently suppress the head’s yaw optomotor response during each saccade. This functional model is one of the clearest put forward for any rapid modulation of visual signaling across the animal kingdom.

RESULTS

Classic optic flow-processing neurons participate in controlling head optomotor responses

There are three uniquely identifiable horizontal system (HS) cells and six uniquely identifiable vertical system (VS) cells on each side of the Drosophila brain (Figure 1A-B). These large cells form part of a visual-motion processing network that is particularly sensitive to the widefield patterns of visual motion, or optic flow, caused by the animal’s own movements through the world (Hausen, 1982; Krapp and Hengstenberg, 1996; Schnell et al., 2010). It has long been hypothesized that HS and VS cells underlie optomotor responses, wherein they drive corrective movements of the head or body in response to optic flow signals (Dvorak et al., 1975; Geiger and Nässel, 1981; Hausen and Wehrhahn, 1983) (Figure 1C). Corrective optomotor head movements would function to stabilize the visual image on the retina, whereas corrective optomotor body (or wing) movements would help the fly maintain a straight flight trajectory. Optomotor stability systems, of both the head and body, should be briefly silenced during intended flight turns, perhaps via an efference copy (von Holst and Mittelstaedt, 1950): a neural signal that counteracts the predicted responses of optomotor-mediating visual neurons to the optic flow arising from these intended turns (Figure 1D).

Figure 1. Horizontal system (HS) and vertical system (VS) cells in the fly visual system may mediate optomotor responses and thus might need to be silenced with an efference copy during intended turns.

(A) A schematic of the fly visual lobe with HS and VS cells in the lobula plate.

(B) Immuno-amplified GFP signal of HS and VS cells, using the DB331-GAL4 driver, with nc82 anti-Bruchpilot neuropil counterstain in magenta.

(C) Schematic illustrating head and body optomotor responses. A gust of wind causes the animal to turn, by accident. Visual neurons, V, (like HS and VS cells) respond to the optic-flow stimulus induced by this erroneous turn. These visual neurons activate motor neurons, M, which drive the body or head to rotate in the direction of experienced visual motion. A rotation of the body would stabilize straight flight (i.e., course stability), whereas a rotation of the head would stabilize the visual image on the retina (i.e., gaze stability).

(D) Schematic of a voluntary turn with an efference copy to suppress the optomotor response. During voluntary turns, an internal decision center, D, sends an efference (motor command) to motor neurons, M, to cause the animal to turn, and a copy of the motor command, an efference copy, to the visual neurons, V. This motor-related input counteracts the expected self-generated visual input caused by the active motor command to turn the animal, thus minimizing or eliminating optomotor responses in the context of voluntary turns.

While attractive, this functional model is based on the assumption that HS and VS cells normally drive optomotor responses. Optogenetic activation of HS cells has been shown to yield movements of the head and wings (Haikala et al., 2013), but no inactivation phenotype has been described for HS and VS cells using cell-type targeted transgenics. Is normal electrical signaling in HS/VS cells in fact necessary for generating wildtype optomotor responses?

We silenced HS and VS cells by overexpressing Kir2.1, a potassium channel, in their membranes, using two different driver lines to target transgene expression to these neurons (VT058487-GAL4 and R24E09-GAL4) (Movie S1). On average, HS/VS cells expressing Kir2.1 showed depolarizing visual responses that were reduced to ~25% of wildtype responses and their hyperpolarizing responses were completely abolished and replaced with tiny, slowly rising, depolarizations (Figure 2A and Figure S1; Movie S2 shows stimuli used). We measured head and wing movements in silenced and control flies in response to widefield motion stimuli during tethered flight (Figure 2B). Specifically, left/right (yaw) movements of the fly’s head were extracted from video images collected at 350 Hz, taken from above the fly. We also simultaneously measured left/right (yaw) wing steering responses from the same flies, on a wingstroke by wingstroke basis, as the amplitude of the right wing subtracted from that of the left wing (left minus right wingbeat amplitude, L–R WBA) using a wingbeat analyzer (Götz, 1987; Maimon et al., 2008).

Figure 2. Expressing a K⁺ channel, Kir2.1, in HS and VS cells impairs optomotor head movements.

(A) Average Vm traces of a wild-type HS cell, recorded in the right lobula plate (black) show direction-selective responses to leftward and rightward moving gratings. HS cells expressing Kir2.1 under the control of two GAL4 driver lines show strongly muted visual responses (orange). Average Vm traces for an example Kir2.1 expressing cell and the average of four Kir2.1 expressing cells are shown for each driver line (top traces: +/+;tsh-GAL80/UAS-Kir2.1::EGFP;VT058487-GAL4/+, bottom traces: w¹¹¹⁸/+;UAS-Kir2.1::EGFP/+; R24E09-GAL4/+).

(B) Experimental apparatus. The fly’s thorax is tethered to a tungsten pin while its head is free to move during tethered flight. We track the yaw angle of the head using a camera capturing frames at 350 Hz (left image) and a wingbeat analyzer to simultaneously measure wing-steering responses (right image). The visual stimulus consisted of a wide-field stimulus with a naturalistic (1/f) intensity profile along the horizontal dimension (STAR Methods) moving at four different speeds. A sample time series of the right and left wing beat amplitudes together with the concomitant head yaw angles are shown.

(C) Baseline subtracted head yaw angle and left-minus-right-wingbeat-amplitude (L-R WBA) signals on single presentations of the fastest stimulus tested (thin lines) and the mean response across all trials (thick lines) for one control fly (black) and one silenced fly (orange). All single trials shown. L-R WBA traces were low-pass filtered using a Gaussian kernel (σ = 2 ms)

(D) Maximum z-projection of Kir2.1::EGFP expression in the VT058487-GAL4 line, used to silence HS and VS cells. The fly also carried a tsh-GAL80 transgene, minimizing GAL4 activity in the ventral nerve cord, which helped promote long flight bouts.

(E) Mean head and wing responses (+/− SEM) to visual motion at four speeds of flies expressing Kir2.1 (tsh-Gal80;VT058487-GAL4 > Kir2.1) and of parental controls. We presented both rightward and leftward visual motion. Traces represent average baseline subtracted responses to leftward motion and inverted responses to rightward motion at a given speed.

(F) Mean +/− SEM of the peak response and initial slope averaged across all conditions in panel E. Average responses of individual flies are shown as single dots. We calculated the peak of each signal by subtracting the mean L–R WBA and head-movement signals in a 100-ms baseline window immediately prior to visual motion onset from the peak signal in a window starting at visual motion onset and extending 200-ms after visual motion offset (peak). We calculated the initial slope of each signal by subtracting the mean signal in a 10-ms window immediately prior to visual motion onset from the mean signal in a 10-ms window starting 80 ms after visual motion onset, and dividing by the time difference between these windows (90 ms) (slope). Silenced flies show a head response that is both smaller and has a shallower slope than control flies (see Main Text for statistics).

(G) Same as D, but for R24E09-GAL4. No tsh-GAL80 transgene was used with this driver.

(H-I) Same as E-F, but for R24E09-GAL4. No tsh-GAL80 transgene was used with this driver.

A sample control fly turned her head and wings in the direction of a widefield visual motion stimulus, which is the classic optomotor response (Figure 2C, left column). A sample silenced fly showed smaller head movements than the control fly to the same stimulus, but her wing movements were grossly similar, or perhaps even stronger in these examples (Figure 2C, right column).

Across both GAL4 lines and all tested motion speeds, silenced flies showed weaker optomotor head movements in response to visual motion (Figure 2D-I). To quantify these effects, we measured the peak and initial slope of each head movement trace. Both measures, averaged over all speeds, were significantly smaller in VT058487>Kir2.1 flies compared to controls (Figure 2F) and in R24E09>Kir2.1 flies compared to controls (Figure 2I) (VT058487>Kir2.1: ‘peak’, p = 0.0028 and 0.0013 compared to the Kir2.1 and GAL4 controls; ‘slope’, p = 0.0028 and 0.0013; R24E09>Kir2.1: ‘peak’, p = 0.0059 and 0.0010; ‘slope’, p = 0.0031 and 0.0002; two sided Mann–Whitney U test, all values are significant after Bonferroni correction for two comparisons). (Identical p-values for peak and slope measures for VT058487>Kir2.1 are correct and indicate identical rank ordering of values for the relevant comparisons.)

Silencing HS and VS cells had a more modest and subtle effect on wing steering responses. At low visual motion speeds, L–R WBA signals seemed generally normal. At higher speeds, there appeared to be a trend toward shallower initial slopes in L–R WBA in both GAL4 lines (Figure 2E, H orange arrows). We averaged the peak and initial slope of L-R WBA traces across speeds. In the VT058487-GAL4 line, neither measure was significantly different between silenced and control flies, though there was a trend toward shallower initial slopes (‘peak’: p = 0.16 and 0.78 compared to the Kir2.1 and GAL4 controls; ‘slope’: p = 0.03 and 0.09, no comparison reaches significance after Bonferroni correction for two comparisons) (Figure 2F). In the R24E09-GAL4 line, the peak was not significantly different between silenced and control flies, but silenced flies had significantly shallower L–R WBA initial slopes than controls (‘peak’: p = 0.350 and 0.028; ‘slope’: p = 0.0147 and 0.0084, final two comparisons reach significance at p < 0.025 level, given a Bonferroni correction) (Figure 2I).

These results demonstrate that abnormal electrical signaling in HS/VS cells can impair head optomotor responses. These results also suggest that HS and VS cells may contribute to fast wing optomotor responses. The fact that wing steering responses to 180°/s visual motion, for example, were grossly similar in silenced and control flies, yet head movements were strongly dampened in silenced animals in response to the same stimulus, on the same trials, argues against the possibility that silenced animals had a very general visual perception or arousal problem. These flies were capable of perceiving optic flow and responding more-or-less appropriately with their wings, however, optomotor head movements were dampened by ~50%.

Drosophila perform an active head movement against gaze stability along the yaw axis during saccades

If HS and VS cells contribute to driving optomotor stabilization movements of the head, what are the head movements that flies make during saccades and can these inform the function of any putative motor-related inputs to HS/VS cells? Fly head movements, in general, occur as rotations along yaw, pitch or roll axes of the neck (yaw and roll rotations are depicted in Figure 3A-B), which typically act to keep the retinal image stable as the body rotates (Hengstenberg, 1991; van Hateren and Schilstra, 1999). Saccades are performed as banked turns of the body (Muijres et al., 2015; Schilstra and van Hateren, 1999), which places challenges on the gaze stabilizing apparatus of the head. Specifically, a fly performs a leftward saccade by rolling its body counterclockwise, yawing left, and rolling clockwise to recover the initial body roll orientation while heading off in a new direction (Figure 3C; roll and yaw body rotations are shown separated in time for clarity). Van Hateren and Schilstra were able to also measure head movements during these banked body maneuvers, in blowflies, and found that the head performs a near perfect, gaze-stabilizing, counter-roll to the body during the saccade, such that the eyes maintain an almost constant roll angle relative to the world during the turn (van Hateren and Schilstra, 1999) (Figure 3C, blue head movements). Along the yaw axis, however, blowflies perform an active head movement against gaze stability. This makes sense because unlike in roll, flies do not counter-yaw their bodies to conclude the saccade, and it would thus be futile for the fly to attempt to maintain gaze stability along this axis (Figure 3C, red head movements). The active yaw head movement, in blowflies, starts after the body is already turning and ends earlier than the body’s rotation, suggesting that it helps to minimize the time over which the image is blurred on the retina during flight turns (van Hateren and Schilstra, 1999).

Figure 3. During rapid flight turns, or saccades, Drosophila perform an active head rotation against gaze stability about the yaw axis.

(A) Schematic of head stabilization movements about the yaw axis.

(B) Schematic of head stabilization movements about the roll axis.

(C) Head movements expected during a banked turn saccade if all head stabilizing reflexes remain intact, emphasizing that the head yaw stability reflex (red) is maladaptive during saccades.

(D) Schematic of the magnetic tether apparatus. A fly, tethered to a ferromagnetic steel pin, is suspended within a magnetic field, and is thus free to perform saccadic rotations of its body around the yaw axis. A camera records the fly’s head and body movements from below.

(E) Sample images of the body and head of a magnetically tethered fly performing a leftward saccade. By 30 ms into the saccade (green arrow), the head has performed a leftward yaw rotation relative to the thorax (dotted line) and a clockwise roll rotation (yellow regions highlight the changing visible areas of the left and right compound eye, as viewed from below the fly, indicating a roll movement of the head).

(F) Sample traces of body and head angles during leftward saccades. The green arrowhead indicates the saccade illustrated in panel E.

(G) Head and body angles associated with saccades from a single fly. Individual saccades are shown in gray. Averages are shown in black. Rightward saccades were inverted and combined with leftward saccades.

(H) Head and body angles associated with saccades from all flies. Average traces from individual flies are shown in gray. Averages across all flies are shown in black.

In Drosophila, only body movements, not head movements, have been measured during free flight saccades, due to technical limitations associated with imaging both the head and body at sufficient resolution. Thus, we decided to measure Drosophila head movements during saccades by making use of the magnetic tethered-flight preparation (Figure 3D) (Bender and Dickinson, 2006). In this preparation, a fly is tethered to a ferromagnetic pin, which is placed inside a vertically oriented magnetic field that allows the fly’s body to rotate about the yaw axis. Tethered, flying flies in this device make spontaneous reorientations of their body that are likely to be analogs of free-flight and rigid-tether saccades (Bender and Dickinson, 2006). We developed algorithms to track the yaw orientation of the body as well as the yaw and roll orientation of the head relative to the body (Figure 3E-F, Figure S2) from video images captured at 350 Hz. Flies performed spontaneous saccades in the context of a uniformly lit panoramic (360°) LED display. We found that along the yaw axis, Drosophila rotate their heads in the same direction as the body (against image stability) during saccades, just like blowflies (Figure 3G-H). Whereas along the roll axis, their head rotates in the direction that would counteract the expected initial body roll at the onset of the saccade, assuming a banked turn. That is, during a left turn, the body is expected to perform a counterclockwise roll to start the saccade, and we measure a clockwise head movement in the magnetic tether (Figure 3G-H).

Note that the magnetic tether setup restricts flies from actually rolling their bodies during saccades and thus the observed roll head movement is not actually counteracting a body roll in our experiments. Rather, we interpret the observed head roll as a reflection of a stereotyped motor command during saccades, in which the body is commanded to roll one way and the head is commanded, in a feedforward manner, to roll the other way. Only the head movement is observed in the magnetic tether. During a free flight saccade, vestibular feedback on the body’s roll movements from the halteres (which is lacking in the magnetic tether) and visual feedback would interact synergistically with this feedforward motor command to help maintain the head stable about the roll axis (Hengstenberg, 1991). Overall, the magnetic tether data argue that Drosophila perform very similar head movements to blowflies during saccades; they move their head in a direction consistent with preserving gaze stability about the roll axis in free flight–which would benefit from an active optomotor system about this axis–but in a direction violating gaze stability along the yaw axis.

Optic flow-processing neurons show broad tuning curves to roll, pitch and yaw visual stimuli

If Drosophila generate an active head movement against yaw gaze-stability during saccades (Figure 3H), then it would seem sensible for the fly brain to send an efference copy to the HS/VS system during these turns. This efference copy–a prediction of the yaw optic-flow input resulting from a saccade–would suppress yaw visual responses in the HS/VS system during saccades, so as to prevent these cells from driving deleterious gaze-stabilizing head optomotor responses (Figure 1D).

To test this hypothesis, we first needed to determine which HS and VS cells respond to yaw visual motion because those are the cells that would need to be suppressed during saccades. We aligned our visual display to the known motion-sensitive axes of the fly’s eye (Figure 4A and Figure S3) (Buchner, 1976; Hardie, 1985), and measured, using the whole-cell patch clamp technique, the membrane voltage (Vm) responses of 47 HS and VS cells to starfield optic flow stimuli that rotated along the yaw, pitch and roll axes (Weir and Dickinson, 2015) (Figure 4B-C). All cells were recorded on the fly’s right side and we focus on visuomotor interactions associated with contraversive (leftward) saccades; we return to ipsiversive saccades in the Discussion. Note also that our stimulus labels indicate the motion direction of the starfield stimuli, such that a “rightward yaw” stimulus means dots moved from left to right in the fly’s perspective; this stimulus would arise from the fly, behaviorally, yawing to the left.

Figure 4. Yaw optic flow elicits strong responses in HS cells, intermediate responses in VS4-6 cells and weak responses in VS1-3 cells.

(A) Electrophysiological setup. We angled our visual display by 66° as schematized, so that vertically and horizontally moving stimuli on the display were aligned to the vertical and horizontal motion-sensitive axes of the fly’s retina. We chose 66° based on the average, measured, eye angles of 11 flies placed in our setup (Figure S3).

(B) Sample starfield frame (top) and vector fields representing local motion associated with yaw, pitch and roll stimuli (i.e., optic flow fields).

(C) Sample Vm trace of one VS5 cell in response to four rotational optic flow stimuli: downward pitch, clockwise roll, counterclockwise roll, and rightward yaw. Stimulus intervals are indicated by gray rectangles. This cell responds strongest to roll, but also shows intermediate responses to yaw.

(D–F) Responses of HS, VS1–3, VS4-6 cells to the rotational stimuli experienced at the start of a leftward saccade: rightward yaw, downward pitch, and clockwise roll. Top row: Averaged Vm from a single cell. Bottom row: Population-averaged Vm (colored) +/− SEM (grey band), with the mean baseline Vm indicated (arrows).

(G–I) Mean visual responses to yaw, pitch and roll visual motion. We calculated these responses by subtracting the mean Vm in a 500-ms baseline window prior to stimulus onset from the mean Vm during the stimulus presentation (excluding the first 100 ms). Individual dots represent single cells; bars indicate mean +/− SEM.

(J) Visual responses to horizontally moving gratings (1 Hz temporal frequency), measured during non-flight. Responses are large in HS, small in VS1-3 and intermediate in VS4-6, similar to the pattern observed for yaw starfield stimuli.

(K) Visual responses to a wide-field stimulus mimicking natural scene statistics (1/f spatial-frequency weighting). The stimulus moved with a saccadic velocity profile over 130 ms with a peak velocity of 1000°/s. Like in panels G and J, responses are large in HS, small in VS1-3 and intermediate in VS4-6. All error bars indicate SEM.

We plotted the average responses of HS and VS cells to the yaw, pitch and roll rotation directions that a fly would experience at the start of a leftward saccade: rightward yaw, downward pitch, and clockwise roll (Muijres et al., 2015) (Figure 4D-I). A standard view is that HS cells are primarily sensitive to yaw whereas VS cells are primarily sensitive to pitch and roll (Krapp et al., 1998; Schnell et al., 2010). Consistent with this general view, HS cells responded strongest to yaw and weakly to pitch and roll (Figure 4D,G-I). VS cells, however, showed more nuanced responses. VS1-3 cells responded strongly to pitch, but also to roll, with weak responses to yaw (Figure 4E,G-I). VS4-6 cells responded strongest to roll with intermediate responses to yaw and negligible responses to pitch (Figure 4F,G-I).

Quantifying these data, we find that yaw visual responses were high in HS cells (5.6 ± 0.7 mV s.d.), intermediate in VS4-6 cells (2.5 ± 0.4 mV s.d.), and weak in VS1-3 cells (0.8 ± 0.4 mV s.d.) (Figure 4G). This gradient of yaw motion sensitivity–from strongest in HS to weakest in VS1-3–is not a quirk of HS/VS responses to tonically rotating starfields because the same gradient was evident when we used a squarewave grating (Figure 4J) or a panoramic stimulus with naturalistic spatial frequency statistics that rotated along the yaw axis very quickly, simulating the velocity profile of a natural saccade (Figure 4K). These data argue that HS cells as well as VS4-6 cells are sensitive to yaw optic flow and thus should receive efference copies of high and medium strength, respectively, during saccades. VS1-3 might be treated differently, perhaps spared from strong saccade-related modulations.

Optic flow-processing neurons show cell-type-tailored motor-related inputs during flight saccades

We previously described saccade-related potentials (SRPs) in horizontal system north (HSN) cells (Kim et al., 2015). SRPs in HSN cells persist in blind flies, demonstrating that they have an extraretinal origin and their sign and timing are consistent with them serving an efference copy function. Here, we measured the magnitude of SRPs in all HS and VS cells in the context of a uniformly lit (blank) screen and starfield stimuli moving at constant velocities (Figure 5A). Flies show tonic optomotor wing-steering responses to starfield stimuli, but they also generate spontaneous saccades riding on these optomotor responses–akin to optokinetic nystagmus saccades in vertebrates–which allow us to analyze associated SRPs. In Figure 5B, we show the Vm of a single HSN cell as the fly generated spontaneous leftward saccades in tethered flight. SRPs were evident during saccades made in the context of a blank screen or during the starfield stimulus, which tonically depolarized the cell. SRPs had larger magnitudes during the depolarizing stimulus than during the blank screen, as previously reported (Kim et al., 2015). Similarly, we observed SRPs in a sample VS5 cell, both in the context of a blank screen and with a starfield stimulus (Figure 5C).

Figure 5. All HS and VS cells show saccade-related potentials, but the magnitude of these potentials differs across cell classes.

(A) Setup for electrophysiology in tethered, flying Drosophila. We recorded from HS and VS cells in flying Drosophila and measured their wing steering behavior. Left and right wingbeat amplitudes (WBAs) were estimated on each frame (100 Hz frame rate) using image analysis, and these signals were used to calculate the left-minus-right-wingbeat-amplitude (L-R WBA).

(B) Sample Vm and L-R WBA traces from an HSN-cell recording session in the context of a uniformly lit screen and during presentation of yaw visual motion. L-R WBA traces were low-pass-filtered with a 10 Hz cut-off frequency. Stimulus interval is highlighted by a gray rectangle; dashed lines facilitate comparisons of baseline changes and SRP magnitudes over the entire time interval.

(C) Same as B, but data are shown for a VS5 cell with a clockwise roll visual stimulus.

(D) Average baseline-subtracted saccade-related potentials (SRPs) during leftward saccades for all stimulus conditions. Average SRPs from single cells are shown in gray. Average SRPs across all cells are shown in color. The average stimulus-driven (or blank screen) Vm immediately preceding each SRP is indicated in mV (arrows).

(E) Distribution of average SRP amplitudes across all seven visual conditions. The SRP amplitude was calculated as the mean Vm in a 50-ms window starting 75 ms after saccade onset from which we subtracted the mean Vm in a 150-ms window, starting 200-ms before saccade onset.

(F) Baseline subtracted saccadic L-R WBA signals, averaged across all flies, with all stimulus conditions overlaid.

Overall, in the 47 HS and VS cells recorded in the right lobula plate, we observed responses consistent with the sample traces. That is, HS and VS4-6 cells showed hyperpolarizing SRPs in the context of a blank screen and starfield stimuli, with slightly larger potentials in HS cells than VS4-6 cells (Figure 5D-E). By comparison, VS1-3 cells showed much smaller saccade-related modulations (Figure 5D-E). The average saccades we analyzed appeared similar across all recording sessions and visual conditions, as estimated from L–R WBA traces (Figure 5F).

We quantified the magnitude of visual responses (Figure 6A, inset black arrow) and associated SRPs (Figure 6A, inset maroon arrow) across the HS and VS population (Figure 6). We quantified SRPs by isolating individual saccades and measuring the mean Vm in a 50-ms window starting 75-ms after saccade onset minus the Vm in a 150-ms baseline window, starting 200 ms prior to the saccade. We quantified visual responses as the mean Vm during this same 150-ms pre-saccadic window minus the mean pre-saccadic Vm during a blank screen. We found that visual responses to the yaw starfield stimulus in these experiments again showed the expected trend of being strongest in HS, intermediate in VS4-6, and weakest in VS1-3 (Figure 6A left, black). Remarkably, SRPs measured during this yaw stimulus showed the same trend–strongest in HS, intermediate in VS4-6, and weakest in VS1-3–but with the opposite sign to the visual responses (Figure 6A left, maroon). We had sufficient data to compare yaw visual responses and SRPs for all nine HS and VS cell types independently (Figure 6A right) and, overall, there was a strong, statistically significant, correlation between the magnitude of yaw visual responses and SRP magnitudes across individual classes of HS and VS cells (r = −0.90, p = 0.003) (Figure 6B). In other words, the more a VS or HS cell depolarized (or hyperpolarized) to rightward yaw, the more it hyperpolarized (or depolarized) for spontaneous leftward saccades made during that stimulus.

Figure 6. Amplitudes of saccade-related potentials in HS/VS cells show a strong negative correlation with yaw visual responses, but not with pitch and roll responses, across the population.

(A) Mean visual-response amplitudes (in mV) to the yaw-right optic-flow stimulus (black) and mean SRP amplitudes (in mV) for leftward saccades generated during presentation of this stimulus (maroon), for HS, VS1-3 and VS4-6 cells (left) and all cell classes, individually (right). SRP amplitudes measured in the context of a blank screen are shown in orange. Inset: representation of visual response (black arrow) and SRP magnitudes (orange and magenta arrows) (see Main Text for details).

(B) The data in panel A, right, are shown here plotted one against the other. Magenta points show amplitudes of yaw-right visual responses plotted against amplitudes of SRPs during the yaw-right starfield stimulus. Orange points show amplitudes of yaw-right visual responses plotted against amplitudes of SRPs during a blank screen. Linear fits are shown for plots with statistically significant, negative correlations. Data points indicate mean +/− SEM.

(C–D) Same as panels A-B, but for downward pitch. Positive correlations in panel D are not statistically significant.

(E–F) Same as panels A-B, but for clockwise roll. Positive correlations in panel F are not statistically significant.

The strong negative correlation of saccadic and visual potentials during yaw starfield stimuli (Figure 6A-B) supports the idea that HS, VS4-6 and VS1-3 cells receive strong, intermediate and weak efference copies, respectively, during saccades. However, an important consideration is that the correlation in Figure 6B could have arisen, in principle, from an underlying mechanism in which all HS and VS cells actually express the identical motor-related conductance during saccades. In this model, saccade-related potentials would be larger in HS cells compared to VS1-3 cells simply because HS cells are more depolarized by yaw visual motion than are VS1-3 cells. The further a cell’s Vm is from the reversal potential of a conductance–i.e. the larger the electrical driving force–the larger is the change in Vm in association with the opening of that conductance.

Our first clue that this driving-force model could not account for the different SRP magnitudes across cell types came from analyzing blank screen saccades. Even when HS and VS cells were not depolarized by any visual stimulus, saccade-related potentials showed varying magnitudes across the HS/VS population, which still correlated with the expected sensitivity of each cell to yaw visual motion (r = −0.90, p = 0.002) (Figure 6A-B, orange). Furthermore, when we analyzed saccades made during pitch or roll stimuli, there was no significant anti-correlation between the magnitude of visual and motor-related potentials across HS and VS cells (Figure 6C-F). With pitch stimuli in particular, VS1-3 cells were strongly depolarized and yet we never observed large SRPs. L–R WBA signals were very similar, on average, across all analyzed saccades (Figure 5F), arguing that differences in SRP magnitudes across cells and stimulus conditions were not due to differences in the motor command to the wings. Rather, the data suggest that VS1-3 cells receive genuinely weaker saccade-related inputs than HS and VS4-6 cells during the same saccade.

The biophysical mechanism for differential motor-related inputs across optic flow-processing neurons

To directly measure the strength of saccade-related inputs across HS, VS4-6 and VS1-3 cells, we would ideally characterize the current-voltage (I-V) relation of the saccade-related conductance in each cell. However, current-injection experiments (Figure S4), and voltage-clamp experiments (data not shown), argued that we could not achieve proper voltage control over the relevant neurites via our patch pipette located at the soma, precluding accurate I-V measurements. We noticed, however, that in our standard current clamp experiments, within each cell class there was a strong linear relationship between the magnitude of SRPs and the stimulus-induced Vm immediately preceding the SRP (Figure 7A), a relationship that we could quantify with an SRP-vs-pre-Vm plot (Figure 7B and Figure S5). Specifically, the SRP-vs-pre-Vm plots had a linear fit with a slope and zero crossing of −0.29 and −60 mV (r = −0.97, p = 0.003) for HS cells, −0.27 and −58 mV (r = −0.84, p = 0.023) for VS4-6 cells, and −0.18 and −47 mV (r = −0.98, p = 0.0001) for VS1-3 cells (Figure 7B).

Figure 7. SRP amplitudes are anti-correlated with pre-saccade Vm for contraversive saccades and different HS/VS cell classes show a different linear relationship between these variables, suggesting different underlying conductance mechanisms.

(A) Population-averaged SRPs for contraversive saccades (same data as in Figure 5D), ordered by the stimulus-induced pre-saccadic Vm, so as to highlight its influence on SRP amplitude. Horizontal dashed lines indicate the resting membrane voltage, estimated as the pre-saccade membrane voltage in the context of a blank screen. Solid lines indicate the zero-crossing Vm calculated from the linear fits in panel B.

(B) SRP amplitude plotted against the pre-saccadic Vm. Lines indicate linear fits. SRP amplitude was calculated as in Figure 5E. Horizontal and vertical lines indicate +/− SEM of pre-saccadic Vm and SRP amplitude, respectively.

(C) A graphical representation of the distance between the in-flight resting potential of each cell class (square) in comparison to the zero crossing point of the linear fits in panel B (arrows). The relative distances of these voltages suggest strong hyperpolarizing inhibition in HS cells, intermediate-strength hyperpolarizing inhibition in VS4-6 and weak shunting/dampening inhibition in VS1-3.

(D) Summary diagram. Visual responses of HS and VS cells contribute to driving head stability movements. Cell-type tailored saccade-related inputs to these cells are tuned to match their yaw optic-flow sensitivity. Our working model is that these inputs serve as efference copies to silence the head’s yaw optomotor stability response, while leaving head stability responses around other axes intact.

In Figure 6, we argued that varying SRP magnitudes across HS/VS cell classes could not be explained with a common conductance/varying driving force model. By contrast, one interpretation of the linear fits in Figure 7B is that SRP magnitudes within single cell classes do conform to such a model. In this interpretation, on all saccades of a given magnitude, HSN cells, for example, activate the same saccade-related conductance, but the saccade-related potential varies in size as a simple linear function of how far away the HSN cell’s Vm is from the reversal potential of that conductance. Because HS and VS cells signal largely through graded changes in Vm, and do not typically fire large action potentials, it is conceivable that their membrane potentials may operate in a roughly linear regime, which would make our voltage measurements linearly proportional to the currents elicited by changes in an underlying conductance.

In this interpretation, HS and VS4-6 cells express a relatively strong conductance during saccades, because the slope of their SRP-vs-pre-Vm plot is steeper than in VS1-3 cells, which express a gentler conductance. The exact reversal potential of each putative conductance is hard to estimate given the likely distortion of voltage measurements between the neurites and soma, among other factors. However, if one makes the assumption that, for each cell class, these distortions are not qualitatively different, such that the zero-crossings of the linear fits in Figure 7B reflect the relative relationship between reversal potentials across cell classes, then one can reach the following conclusions (Figure 7C). The strong saccade-related conductance in HS cells drives these cells to a relatively hyperpolarized potential (8 mV hyperpolarized from rest, as measured at the soma). The strong saccade-related conductance in VS4-6 cells drives these cells to a similar, but perhaps slightly less hyperpolarized, potential as HS cells (6 mV hyperpolarized from rest, as measured at the soma). The gentle saccade-related conductance in VS1-3 cells drives these cells to a potential that seems closer to the cell’s resting potential (0 mV from rest, as measured at the soma). If the reversal potential of the gentle saccade-related conductance in VS1-3 cells were indeed close to the resting potential at the neurites, then this input would act to gently dampen any sensory drive away from rest in VS1-3. The functional role of such a putative input is not fully clear–the system may seek to generally dampen pitch- or roll-related optomotor stability during saccades–but what is clear is that the gentle saccade-related input to VS1-3 cells is qualitatively different from the stronger, hyperpolarizing, inputs to HS and VS4-6 cells. (Note that HS and VS cells are part of a gap-junction coupled network, which hinders making any standard assumptions about the nature of synaptic conductances, like all excitatory inputs having reversal potentials near 0 mV and all inhibitory inputs having reversal potentials between −60 and −90 mV.)

An important caveat is that in an SRP-vs-pre-Vm plot, a visual stimulus–rather than current injection through a pipette–induces Vm changes in the recorded cell. Visual stimuli influence not only the recorded cell but also thousands of other neurons in the visual lobe and beyond. It is thus possible that the magnitude of saccade-related conductances varies across visual conditions, not only the driving force that those conductances interact with on our recorded cell’s membrane. Note, however, that SRPs reversed polarity at the zero-crossing potential for VS1-3 cells, which is easier to understand with a driving force mechanism than a conductance-strength changing mechanism. Furthermore, for each cell class a different visual stimulus is associated with the largest SRP (Figure 7A)–the stimulus that most depolarizes that cell class–and thus with a conductance-strength changing mechanism one would have to imagine that each cell class is tuned to activate its strongest saccade-related conductance under different stimulus conditions from other classes. We see it as more parsimonious, at this stage, to imagine for each cell class that a common input arrives for all saccades of a given size, independent of the visual stimulus. However, we cannot rule out– and in many ways it would be even more interesting–if the system somehow tuned saccade-related conductance strength to each cell based on the background tenor of optic flow. Regardless of whether a simple driving-force mechanism, a more sophisticated conductance-strength-based mechanism, or some combination of the two, underlies SRP sizes in single cells, our data demonstrate that the HS/VS system receives quantitatively tuned saccade-related inputs that are poised to silence yaw visual responses in the population, thus functionally dampening yaw optomotor head movements during rapid flight turns (Figure 7D).

DISCUSSION

What is the function of HS and VS cells?

Here, we report a weakening of optomotor head movements when one strongly expresses Kir2.1 in HS/VS cells (Figure 2), supporting a head-stability function for the HS/VS system. Because this perturbation might also directly hyperpolarize gap-junction-coupled neck motor neurons, the reduction in head opotomotor responses could be due to muted visual signaling in HS/VS cells (Figure S1), a general impairment of neck motor neurons, or a combination of the two. In any of these scenarios, our data argue that HS and VS cells are functionally coupled to the neck motor system.

Past experiments have inactivated HS/VS cells through genetic mutations that reduced the size of the entire lobula plate (Heisenberg et al., 1978), laser ablation of precursors to HS/VS cells in larvae (Geiger and Nässel, 1981), or through mechanical slicing of the neuropil region through which the HS axons course (Hausen and Wehrhahn, 1983). These perturbations affected the wings’ optomotor response. Here we report a strong reduction in the magnitude and speed of head optomotor responses with silenced HS/VS cells and a modest effect on wing movements (Figure 2). HS and VS cells may functionally couple to both head and wing motor systems (Haag et al., 2007; Haikala et al., 2013; Strausfeld and Seyan, 1985; Strausfeld and Bassemir, 1985; Strausfeld et al., 1987; Wertz et al., 2012; 2009), but the circuit underlying wing responses might have more redundancy such that this system can compensate for reduced HS/VS functionality. Note that past inactivation methods lacked the specificity afforded by the GAL4-UAS system and may have targeted not just HS/VS cells, but also other putative lobula plate neurons that contribute more to wing responses. Overall, the simplest interpretation of our data is that a different set of visual projection neurons drives the lion’s share of the wings’ optomotor response.

Contraversive versus ipsiversive saccades

Our SRP analysis in this paper focused on visuomotor signals in the right lobula plate during leftward, or contraversive, saccades. When analyzing signals for rightward, or ipsiversive, saccades, we found that HS cells show small, but clear, depolarizing SRPs (Figures S6 and S7). The fact that HS cells show large hyperpolarizing SRPs for contraversive saccades, but small, depolarizing SRPs for ipsiversive saccades, is consistent with these potentials acting to silence the asymmetrical visual responses in HS cells; HS cells show a depolarization to ipsiversive/front-to-back motion that is about twice the magnitude of the hyperpolarization they show to contraversive/back-to-front motion (Hausen, 1982; Schnell et al., 2010). In contrast to HS cells, VS cells did not exhibit clear depolarizing SRPs during ipsiversive saccades (Figures S6 and S7). These data suggest that there may be a functional dissociation in the role of VS cells in ipsiversive versus contraversive turns. For example, if we speculate that the gap junctions between VS4-6 cells and neck motor neurons are rectifying, such that only depolarizations in the VS4-6 system influenced the downstream cells, then there would be no need to send a detailed efference copy to the VS4-6 system during ipsilateral turns, when a yaw-associated visual hyperpolarization is expected in these cells. Regardless, during all saccades, the fly brain sends predictive, motor-related, signals to at least the VS network on the contralateral side of the brain.

Membrane conductance mechanisms for visuomotor efference copies

In their initial description of the concept of efference copy, von Holst and Mittlestaedt postulated a signal that would go to the fly visual system to silence the optomotor response during intended turns (von Holst and Mittelstaedt, 1950) (Figure 1D). The fact that HS and VS cells participate in driving visual stability movements of the fly’s head (Figure 2) and that motor-related inputs abrogate the visual activation of these cells during intended turns, means that the motor-related modulations of HS and VS cells accord well with von Holst and Mittlestaedt’s concept of an efference copy.

Whereas flies would do well to ignore the expected visual consequence of saccades, they ideally would not go completely blind during turns; unexpected visual input, of the wrong sign or magnitude, should still be perceived (Heisenberg and Wolf, 1979; von Holst and Mittelstaedt, 1950). HS and VS4-6 cells receive hyperpolarizing saccade-related inputs during contraversive saccades. If the strength of these inputs were tuned (by experience or evolution) to cancel out the expected yaw visual drive during saccades–which appears to be the case (Figures 4-6)–then the Vm of HS and VS4-6 cells would remain close to the resting potential during turns. However, importantly, these cells could still respond, in principle, if the fly were to experience an unexpected visual stimulus during the saccade. An alternative approach for silencing HS/VS cells could have been to activate an extremely strong, shunting inhibitory conductance with a reversal potential near rest rather than below rest. Such an input would also keep the cell near the resting potential during saccades but at the cost of inhibiting all changes in Vm during the saccade, rather than acting as a tuned prediction for the expected input. Note that VS1-3 cells seem to receive a weak version of this shunting type input, but it is sufficiently modest in strength that responses are gently muted during saccades rather than gated off.

A model for the visuomotor interactions in the fly visual system during flight turns

We put forward the following proposition for the sensorimotor interactions during a flight saccade in Drosophila. The fly’s body starts a leftward saccade by rolling counterclockwise and yawing left (Muijres et al., 2015). Concomitantly, the head performs a clockwise counter-roll in an attempt to maintain gaze stability along the roll axis (Figure 3). Active optomotor and vestibular reflex systems along the roll axis, working together with a feedforward head-roll command (Figure 3), all contribute toward stabilizing gaze along the roll axis during the saccade. In yaw, by contrast, an active motor command is given to turn the head left, against gaze stability (Figure 3). This anti-stability yaw-axis head movement likely helps to shorten the time during which the fly’s retina experiences unavoidable yaw-axis image blur during each saccade (van Hateren and Schilstra, 1999). In coordination with the yaw head movement, HS and VS cells sensitive to yaw visual motion receive an efference copy that antagonizes their visual responses to the saccade which would, if left unaltered, act to uselessly drive the head in aberrant directions. The fly finishes the leftward saccade by rolling its body clockwise, counter-rolling the head, and flying off in the new direction.

Behavioral modulation of vision across animals

Altogether, this work argues that HS and VS cells function to drive stabilizing head movements in flying flies and this head stabilization function is suppressed during saccades. This functional proposal is an important step forward for fly vision. Beyond flies, a spate of recent reports has described modulations of visual processing in rodents that can be linked to behavioral state changes, like locomotion or arousal (Keller et al., 2012; Niell and Stryker, 2010; Vinck et al., 2015). Analogously, visual neurons in flies are strongly modulated during flight or walking (Chiappe et al., 2010; Maimon et al., 2010). Here we show that within a broad behavioral state–flight–one observes dynamic motor-related inputs that ride upon the tonic state modulation and these dynamic inputs, we argue, can be understood to serve a predictive, efference copy function. Specifically, the data support a model in which the fly brain, during saccades, engages a set of precisely tuned conductances to nullify one sensory signal (yaw optic flow) in a population of broadly tuned sensory neurons that carry multiple related signals (yaw, pitch and roll optic flow). Similar computational processes are likely to exist in mammals and beyond. This work should thus serve as an important template for understanding rapid modulations of visual signaling across species.

STAR METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to the Lead Contact Gaby Maimon (maimon@rockefeller.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

In all experiments, we studied female Drosophila melanogaster. In electrophysiology experiments, flies were up to 3 days old (post-eclosion). In behavioral experiments, flies were 3-to-4 days post-eclosion. All flies were reared in 25°C incubators with a 12 hour light/dark cycle, in bottles (Applied Scientific; 57 mm × 57 mm × 103 mm with a square bottom) containing standard cornmeal agar, with ~5-25 flies per bottle. For electrophysiology experiments, we visualized horizontal system (HS) cells by crossing the w¹¹¹⁸;+;R81G07-GAL4 driver line to a +;+;UAS-2xEGFP responder line, and we visualized vertical system (VS) cells by crossing w¹¹¹⁸;+;R24E12-GAL4 driver line to the same responder. In behavioral genetics experiments (Figure 2), we expressed Kir2.1 in HS and VS cells by crossing two different driver lines (+;tsh-GAL80;VT058487-GAL4 and w¹¹¹⁸;+;R24E09-GAL4) to a +;UAS-Kir2.1::EGFP;+ responder line. The tsh-GAL80 transgene leads to expression of a Gal4 inhibitor, Gal80, in the thoracic ganglion, which, empirically, helped to promote long flight bouts. We used wild-type Canton-S flies for the magnetic tether experiments (Figure 3). Specific genotypes and sources of transgenic animals used are listed in the Key Resources Table. In all experiments, flies were anesthetized on a Peltier stage at ~4°C, and attached by the dorsal part of their prothorax either to a custom patch-clamp stage or to a steel pin using blue-light activated glue (Bondic, Canada). In behavioral experiments, the head remained free and we tracked its movements. In electrophysiology experiments, the head was affixed to the recording stage, to permit stable electrophysiological recordings.

METHOD DETAILS

Visual stimuli

In Figures 4-7, we used a cylindrical green LED (570 nm) visual display that extended 216° in azimuth and 76° in elevation (IORodeo, CA) (Reiser and Dickinson, 2008). Each pixel subtended 2.25°, which is well below the ~5-6° inter-ommatidial angle in Drosophila (Heisenberg and Wolf, 1984). Rather than each pixel being fully on or off during experiments, we used 4 or 8 gray-scale intensity levels, which allowed for edges in visual patterns to appear to move more smoothly across frames (rather than jump by 1 full pixel, or 2.25°). Such pixel ‘dithering’ increases the apparent resolution of the display fourfold or eightfold and also helps to minimize non-motion-related (flash) responses of HS and VS cells to each frame update. In the magnetic tether experiments (Figure 3), we used a cylindrical LED display covering 360° in azimuth and 94° in elevation with each pixel subtending ~3.75° (570 nm). The stimulus was a uniformly lit screen in these experiments. In measuring visual responses of HS and VS cells to optic flow stimuli (Figure 5) we angled our LED display such that the axis of horizontal motion of starfield stimuli (or gratings) was aligned with the horizontal motion detection axis of the fly’s eye, called the h-row, and vertical starfield motion was aligned with the v-row (Buchner, 1976; Hardie, 1985). To find the proper angle for the LED display, we took photos, from the side, of 11 flies tethered to our patch-clamp platform and identified the three major ommatidial axes of the eye–v-row, x-row, and y-row–in each fly (Figure S3). The h-row was noted as a line between x-row and y-row axes that links the closest, non-adjacent, ommatidia in the hexagonal grid of the compound eye. We measured the angular tilt of this axis with respect to the horizontal axis of their air table, or, equivalently, the roof of the plate to which the fly was tethered and found it to be, on average, 65.6° ± 5.4° (n = 11 flies). We thus tilted the visual arena by 66° for electrophysiology experiments (Figure 4A).

To generate starfield stimuli (Weir and Dickinson, 2015), we populated a virtual 3-D volume with identically sized spheres at random, uniformly distributed, positions, and projected the image of each sphere onto the fly’s head. We illuminated those pixels on the display that overlapped with the projection cone onto the fly’s head. Consequently, the image of each projected sphere on our visual display increased nonlinearly as a function of how close the sphere was to the fly. To prevent the size of each sphere from growing infinitely large as it approached the fly, we limited the diameter to have a maximum of 9°. Grating stimuli used in Figure 4J were square waves with an 18° wavelength, moving at a temporal frequency of 1 cycle/s. All stimuli had a nominal contrast of 100%.

In Figures 2 and 4K, we presented a wide-field stimulus with a random intensity profile along the horizontal axis that simulated the spatial frequency statistics of natural scenes, as described previously (Kim et al., 2015). Specifically, the intensity profile was generated by linearly superimposing sinewaves at a random phase after weighing the amplitudes by the reciprocal of their spatial frequency, thus approximating the known 1/f (f = spatial frequency) statistics of natural scenes. In Figure 2, we moved this stimulus by 60° to the right or left at four different speeds (90, 180, 360, or 720°/s). To quantify yaw visual sensitivity of HS and VS cells (Figure 4K), we moved a 1/f pattern for 130 ms with a peak speed of 1000°/s (rightward or leftward), simulating the velocity profile of a natural saccade, as described previously (Kim et al., 2015).

Electrophysiology

We performed whole-cell patch clamp recordings as described previously (Maimon et al., 2010). In brief, after tethering flies to our electrophysiology platform, we fed flies with ~100 nl of 500 mM sucrose solution from a pipette tip to help promote long flight bouts. The proboscis of each fly was glued immediately before the cuticle above the brain was dissected. During experiments, we perfused the preparation with oxygenated 275 mOsm extracellular saline that contained (in mM): 103 NaCl, 3 KCl, 5 N-Tris(hydroxymethyl) methyl-2-aminoethanesulfonic acid (TES), 10 Trehalose, 10 Glucose, 2 Sucrose, 26 NaHCO₃, 1 NaH₂PO₄, 1.5 CaCl_2, 4 MgCl₂. This yields a pH of 7.3 when equilibrated with 95% O₂/5% CO₂. The intra-cellular solution (pH 7.3, 265 mOsm) in patch-clamp electrodes (4-8 MOhm) contained (in mM): 140 K-Aspartate, 1 KCl, 10 HEPES, 1 EGTA, 0.5 Na₃GTP, and 4 MgATP, 0.02 Alexa-568-hydrazide-Na and 13 Biocytin hydrazide. The membrane voltage was amplified (A-M Systems Model 2400), digitized at 10 kHz (PCIe-6351, National Instruments; Digidata 1440a, Molecular Devices), and saved to a computer (WinEDR, University of Strathclyde; pClamp 10, Molecular Devices). Voltage measurements have been corrected for a 13 mV junction potential. We injected −10 pA of hyperpolarizing current into neurons to neutralize the depolarizing effects of the seal conductance. The membrane resistances of recorded cells in Figures 4-7, measured at the beginning of each recording, were as follows: 150 ± 51 MOhm for 13 HS cells, 140 ± 83 MOhm for 21 VS1–3 cells, and 151 ± 31 MOhm for 13 VS4–6 cells. In order to identify the cell type, we applied weak positive pressure (3–5 mmHg) to the recording electrodes, which, empirically, seemed to facilitate the filling of each cell with Alexa-Fluor-568. Immediately after each recording, we saved a widefield epifluorescence z-stack of the Alexa-568-filled cell and the GFP signal from all HS or VS cells (excitation wavelength: 470 nm for GFP and 565 nm for the Alexa 568). The Alexa-568 signal in 47 out of 55 recorded cells was sufficiently strong so as to allow us to unequivocally assign their identity to one out of the nine known HS and VS cells, and the recordings of these cells were further analyzed.

To quantify visual responses to starfield motion stimuli for Figure 4 (non-flight) we calculated a baseline Vm from each cell’s average trace as the mean Vm in the 500-ms interval prior to stimulus onset. We then calculated a stimulus Vm from each cell’s average trace as the mean Vm during the 2 s stimulus period (excluding the first 100 ms after stimulus onset to exclude flash responses). The visual response was then calculated as: stimulus Vm – baseline Vm.

To quantify saccade-related potentials (SRPs), we first isolated saccades, using an algorithm described below and calculated the average saccade-triggered Vm for each cell and stimulus condition (Figures 5-7, S5-S7). We calculated a pre-saccade Vm as the mean value of this trace in a 150 ms window, starting 200 ms before the saccade. The 50-ms time interval immediately preceding a saccade onset was not included for the calculation of the baseline Vm because SRPs can precede saccade onsets by ~30–40 ms (Kim et al., 2015). We calculated SRP amplitudes as the mean of the SRP Vm trace in a 50 ms window starting 75 ms after saccade onset, after subtracting the pre-saccade Vm from each sample point of the SRP trace. These analysis time windows were selected by eye to best capture the baseline time period and peak-response time period of SRPs; these windows were kept the same for all saccade-related analyses. For Figure 6, when flies were flying, we quantified visual responses differently than in non-flight conditions because during flight, SRPs, if they occur at a high enough rate, artifactually reduce the apparent visual response of an HS or VS cell. Thus, to estimate the visual response of an HS or VS cell in flight, we subtracted the average pre-saccade Vm during blank screen conditions from the average pre-saccade Vm for each stimulus condition. In the pre-saccadic time window, SRP influences on visual responses are minimized.

Immunohistochemistry and Imaging

We dissected 2-to-3-day-old female flies to visualize the expression of Kir2.1::EGFP (Figure 2). Brains and ventral nerve cords (VNCs) were dissected together, subsequently fixed in 1% PFA in S2 medium and nutated overnight at 4°C. We next blocked specimens in 3% normal goat serum in PAT3 for 90 minutes and incubated overnight in a primary antibody solution containing 1/25 chicken nc82 and 1/1000 rabbit anti-GFP. Finally, samples were incubated in a secondary antibody solution containing 1/400 anti-chicken-Alexa488 and 1/800 anti-rabbit Alexa594, for 3-4 days. Samples were washed in PAT3 three times after each step. We mounted the brains and VNCs on slides with their posterior surfaces on top, immersed in Vectashield, and took z-stacks with a two-photon microscope (Bruker, MA), using an excitation wavelength of 800-850 nm. The brain and the VNC were imaged separately, but with the same optical settings (Figure 2).

We registered z-stacks of Kir2.1::EGFP expression in the R24E09-GAL4 and VT058487-GAL4 driver lines to the Janelia fly brain template (see Key Resources Table) and visualized brain regions with transgenic expression by generating a multi-color z-stack image of Kir2.1::EGFP expression in the two lines (Movie S1). The only cells in the brain with obvious, strong, shared expression across these two driver lines are the HS and VS cells. Both these lines drove very high levels of Kir2.1::EGFP in the HS/VS system, which likely helped to promote driving electrophysiological and behavioral phenotypes.

QUANTIFICATION AND STATISTICAL ANALYSIS

Behavioral measurements and analyses

Videos of flies were taken with an infrared-sensitive camera (AVT-GE680), triggered externally (Maimon et al., 2010). For electrophysiology experiments (Figures 4-7), we recorded frames (640×480 pixels) at 100 frames/s. For behavioral experiments (Figures 2-3), we recorded frames (320×240 pixels) at 350 frames/s.

For tethered-flight experiments in Figure 2, we measured the yaw angle of the head by means of custom-written image-analysis code. The wing stroke amplitude was measured with a wingbeat analyzer (JFI Electronics Laboratory, University of Chicago, Chicago, IL) (Götz, 1987; Maimon et al., 2008). With a wingbeat analyzer, flies were illuminated with an 880 nm infrared diode from above so that the two flapping wings cast oscillating shadows onto two photodiodes below. The photodiodes yielded oscillating signals on each wing stroke, whose maximum amplitude provides a measure of the maximum wing-stroke angle achieved on that stroke (which we term wingbeat amplitude or WBA). The difference in bilateral stroke amplitude (left-minus-right wingbeat amplitude, L–R WBA) is correlated with yaw torque (Tammero 2004), and thus provides a good measure of the intended turning direction. In addition to the wingbeat analyzer LED, we also placed a 850-nm LED above the fly to illuminate the animal’s head for video analysis of its yaw angle. We placed an optical bandpass-filter (880 nm center wavelength, Edmund Optics, NJ) above the wingbeat analyzer’s photodiodes to prevent the 850 nm light source from interfering with the wingbeat analyzer’s signals. Left and right wingbeat amplitudes, camera triggers, and voltages encoding the visual stimuli were sampled at 10 KHz (Digidata 1440a, Molecular Devices, CA). We used a 2-s pause in the camera triggers at the start of each experiment to synchronize video frames with other time series data, like the output of the wingbeat analyzer and frame updates of the visual stimulus. We only analyzed behavioral data collected while the fly was flying. For each fly, we calculated baseline-subtracted averages of their head and wing responses to each stimulus. We used the mean signal in a 200-ms window immediately prior to stimulus onset as the baseline window. Standard errors of the mean shown in Figure 2 are of baseline-subtracted traces.

For the magnetic-tether experiments in Figure 3, we estimated the orientation of the body, and the head angle relative to the body, by analyzing video images captured by a camera viewing the fly from below (Figure S2). To measure the body’s yaw orientation, we first used a predefined threshold to create a binary version of the image and extracted the largest contiguous object, which corresponded to the fly’s body. We then estimated the yaw orientation and the center of mass of this object, and we used these parameters to rotate and translate the image so that the fly’s body was always oriented upward and centered at the same position on each frame. To compute the head’s yaw angle relative to the body from this standardized image, we detected the posterior edge pixels of the head and found the best-fit line to these pixels using a least-squares regression. This same algorithm was used to compute the head’s yaw angle relative to the body in the rigid pin tether experiments in Figure 2. We estimated the roll angle of the head relative to the body by calculating how much of the right and left eyes were visible on each frame. During a clockwise roll of the head, the fly’s left eye becomes less visible and right eye more visible as viewed from the bottom, for example. We drew a series of ~10 lines that intersected the head in parallel to the estimated head yaw angle (Figure S2B) and calculated the mean pixel intensities across these lines. The left and right eyes appear as intensity peaks (bright pixels) on the left and right edge of each pixel-intensity function (Figure S2B). We computed the average width of the left and right peak in each pixel intensity function, and calculated the head roll index based upon the following equation:

$$\mathrm{Head}\mathrm{roll}\mathrm{index}=(\mathrm{Left}\mathrm{eye}\mathrm{width}-\mathrm{Right}\mathrm{eye}\mathrm{width})/(\mathrm{Left}\mathrm{eye}\mathrm{width}+\mathrm{Right}\mathrm{eye}\mathrm{width})$$

To convert this index to a calibrated head roll, in degrees, we took a series of images of fly heads, from both the bottom and the front. We measured the exact roll angle from the front image and calculated the roll index from the bottom image (n = 8 flies, 7–8 samples per fly, Figure S2C). The resultant relationship between the head roll angle and the roll index was linear (Head roll angle = 36.3 × Head roll index), with the correlation coefficient of 0.95 (Figure S2). We performed a geometric analysis of a simple fly head model to show that when the roll angle is not too large (< ~40°), the roll index should indeed be linearly related to the head’s angle (Figure S2D).

Saccade detection

For electrophysiology experiments, we detected individual saccades and computed saccade-triggered averages of Vm and L-R WBA as described previously (Kim et al., 2015). Briefly, putative saccades were isolated from the derivative of the L-R WBA signal by scanning this derivative trace for crossings of a predefined threshold. Candidate saccades that immediately followed previous saccades were excluded because their onsets were confounded with the end of the previous event. We computed saccade-triggered averages from the remaining saccades.

For magnetic-tether experiments, we first smoothed the body orientation signal by using a constrained-least-squares linear-phase lowpass filter with a 10 Hz cutoff and computed its first order derivative. We then applied a threshold of ±140°/s to detect putative saccades. The threshold was determined as the angular velocity four standard deviations away from zero, as described previously (Bender and Dickinson, 2006). We excluded events with a relatively short duration (< 10 ms) or a low amplitude (< 20°). While many such small or brief events are likely to be genuine saccades, we wanted to limit our analysis to unequivocal, large, saccadic turns, in which head movements could be easily measured. To compute saccade-triggered averages, we aligned all saccades to a computationally determined onset time. To estimate onsets, we moved backward in time for each event from the detected high derivative point and identified a tentative saccade onset time as the point at which the derivative signal returned to zero. Because the derivative signal is low-pass-filtered, without further correction this onset time was early relative to the actual onset, as assessed by eye. To correct for this error, we noted the velocity at the tentative onset time (as found in the low-pass filtered trace), which we considered as a threshold; we then looked forward in time in the unfiltered orientation trace for the last sample before this threshold was crossed and the time point associated with this sample was considered the actual onset time.

Statistics

Details about statistical tests in Figure 2 are provided in the Main Text. R-values for correlations for Figures 5-7 are Pearson correlation coefficients. To test the statistical significance of these values, we used a permutation test method. In particular, we computed correlation coefficients of 5000 random permutations of each original data set and calculated the probability of observing a coefficient as large as the one actually observed in the distribution of coefficients from the permuted data.

DATA AND SOFTWARE AVAILABILITY

Raw data and analysis/stimulus code will be provided upon request by the Lead Contact Gaby Maimon (maimon@rockefeller.edu).

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-GFP (chicken)	Rockland Immunochemicals	Cat#600-901-215S
nc82 (mouse)	Developmental Studies Hybridoma Bank	N/A
Goat anti-chicken, Alexa488	Thermo Fisher Scientific	Cat#A-11039
Goat anti-mouse, Alexa594	Thermo Fisher Scientific	Cat#A-11032
Chemicals, Peptides, and Recombinant Proteins
PFA	Electron microscopy sciences	Cat#15713-S
S2 medium	Sigma Aldrich	Cat#S01416
Normal goat serum	Thermo Fisher Scientific	Cat#16210064
Deposited Data
Janelia fly brain template	Virtual fly brain project	https://github.com/VirtualFlyBrain/DrosAdultBRAINdomains
Experimental Models: Organisms/Strains
Drosophila: w1118 ;+;R24E09-GAL4	Bloomington Drosophila stock center (BDSC)	Stk#49083
Drosophila: w1118 ;+;R81G07-GAL4	BDSC	Stk#40122
Drosophila: w1118 ;+;R24E12-GAL4	BDSC	Stk#49084
Drosophila: +;tsh-GAL80/CyO;VT058487-GAL4/TM3,Ser	Andrew Straw	N/A
Drosophila: +;UAS-Kir2.1::EGFP;+	Andrew Straw	N/A
Drosophila: w1118/DB331-GAL4;UAS-2xEGFP/+;+	this paper	N/A
Drosophila: +;UAS-Kir2.1::EGFP/tsh-GAL80;VT058487-GAL4/+	this paper	N/A
Drosophila: +;UAS-Kir2.1::EGFP;+	this paper	N/A
Drosophila: +;tsh-GAL80/+;VT058487-GAL4/+	this paper	N/A
Drosophila: w1118/+;UAS-Kir2.1::EGFP/+;R24E09-GAL4/+	this paper	N/A
Drosophila: w1118/+;UAS-Kir2.1::EGFP;+	this paper	N/A
Drosophila: w1118/+;+;R24E09-GAL4/+	this paper	N/A
Drosophila: Wild-type Canton S	Martin Heisenberg	N/A
Drosophila: w1118/+;UAS-2xEGFP/+;R81G07-GAL4/+	this paper	N/A
Drosophila: w1118/+;UAS-2xEGFP/+;R24E12-GAL4/+	this paper	N/A
Software and Algorithms
Real-time wingbeat analysis for tethered, flying flies	Andrew Straw	https://github.com/motmot/strokelitude
Post hoc head angle analysis for tethered flies	this paper	N/A

Supplementary Material

1

Figure S1. HS cells show strong, direction-selective responses to fast visual motion, whereas HS cells expressing Kir2.1 show small, residual responses that are depolarizing for both motion directions. Related to Figure 2.

(A) Population-averaged, baseline-subtracted Vm (color) +/− SEM (gray) for 7 HS cells (1 HSE, 6 HSN) in response to the visual motion stimuli used in Figure 2. Six cells were recorded during flight, and one cell in non-flight. The pre-stimulus Vm is indicated (arrows).

(B) Same as panel A but for three VT058487>Kir2.1 flies (+/+;tsh-GAL80/UAS-Kir2.1::EGFP;VT058487-GAL4/+). One cell was recorded during flight, the others in non-flight.

(C) Same as panel A but for four R24E09>Kir2.1 flies (w¹¹¹⁸/+;UAS-Kir2.1::EGFP/+; R24E09-GAL4/+). All recordings in non-flight.

(D) Data from panels A-C shown overlaid on the same axes.

(E-F) Mean amplitudes of visual responses. Amplitudes were calculated, for each fly and each motion speed (one dot), as a mean Vm in the stimulus interval relative to a 100-ms baseline window immediately preceding stimulus onset. Error bars indicate SEM.

Figure S2. Estimating head yaw and roll angles, relative to the body, from a bottom view of the fly in the magnetic tether. Related to Figure 3.

(A) Head yaw angle estimation. We created a binary mask of each fly image and reorientated the image such that the fly was oriented vertically. We computationally found the posterior edge of the head and computed its angle (STAR Methods).

(B) Estimating eye width. We drew lines parallel to the estimated head yaw angle (colored lines), and computed a pixel intensity profile along each line. Lines that intersected both eyes showed two separate peaks. We averaged the pixel intensity profile across lines and identified the lateral peaks on the left and right side in this averaged trace. The computationally estimated width of the left-eye and right-eye peaks were used to estimate the head’s roll angle.

(C) Estimation of the roll angle from eye widths. We computed a roll index for each image from the estimated left-eye width (S_L) and right-eye width (S_R). We converted the roll index to the actual roll angle of the head, in degrees, as follows. We first took a series of images of dead flies (n = 8 flies, 8–13 images per fly, each fly’s data are represented with points of a different color), simultaneously from the bottom and from the front, in rigid tether. The images taken from the front gave us the actual roll angle of the head (which we varied systematically) and the images from the bottom gave us the roll index. The relationship between the roll index and roll angle was linear with a slope of 36.33 degrees per roll index unit (r = 0.98).

(D) The relationship between the head roll angle and the roll index of a geometrically simplified fly head. When the roll angle is not too large (< ~40°), the roll index is expected to be related linearly to the head’s actual roll angle.

Figure S3. Estimating the angle of the ommatidial axis for horizontal motion detection. Related to Figures 4-7.

(A) We first saved a zoomed-out picture of a fly attached to the eletrophysiology plate (shown), from which we could visualize the angle of the plate’s precision-machined roof (green line). The angle of the plate’s roof is the same as the horizontal plane of our air table on the electrophysiology rig and can be used as a reference for measuring the angle of the eye. After estimating the roof’s angle, without moving the fly or the dissection microscope (which we used for grabbing images), we zoomed in to the fly’s head and took a high resolution image of the visible (right) eye. From the high-resolution image of the eye, we estimated the angle of the ommatidial axes relative to the roof of the plate.

(B) Images of 11 fly eyes and their assessed ommatidial axes. Green lines indicate angle of the recording platform (images were rotated to make this line horizontal on this figure). Because of the hexagonal arrangement of ommatidia, there are three principal axes that connect neighboring ommatidia: the x-row, y-row and v-row. The midline between x-row and y-row (indicated in yellow) is the behaviorally relevant angle of horizontal motion detection based on classical experiments (Buchner, 1976).

Figure S4. Injecting current into HS/VS cell somas influences visual response amplitudes only weakly, suggesting that we have poor voltage control over synaptic sites via our patch pipette. Related to Figure 7.

(A) A schematic of the experimental apparatus for performing current clamp experiments with current injection. We used the bridge balance circuit on the A-M Systems 2400 amplifier to compensate for our measured series resistance at the start of each experiment (not schematized).

(B) Sample traces of injected current (top) and Vm (bottom). Grey rectangles indicate presentations of visual stimuli. Injected current ranged from −100 pA to 100 pA with 20 pA steps, which led to modulations of the soma Vm from −95 to −25 mV.

(C) Vm responses of a single HSN cell to preferred-direction motion stimuli. Mean responses from this cell are shown in black and all 10 individual trials are shown in gray. The pre-stimulus Vm is indicated (arrows and black dashed line). We drew a red dashed line to aid in comparing the magnitude visual responses across different current injections.

(D) Population-averaged Vm traces (4 HS cells and 2 VS cells) in response to preferred-direction motion stimuli. Population averages are in black and individual cell averages in gray.

(E) Same cell as in C, but data are shown for null direction motion.

(F) Same cells as in D, but data are shown for null direction motion.

(G) Visual response amplitudes, to preferred-direction motion stimuli, decreased only weakly with varying the soma’s Vm: ~2 mV change in the magnitude of visual responses with > 60 mV change in soma Vm. For the null direction, visual response amplitudes remained almost completely unchanged with > 60 mV change in soma Vm. Since HS cells are likely to experience the same visually driven conductance with each presentation of an identical visual stimulus, the fact that we could not strongly alter the magnitude of the visually driven potential with large voltage swings at the soma suggests that we have poor voltage control of the dendrites where the visual conductances activate. Error bars indicate mean +/− SEM.

(H) In a sample HSN cell, we recorded spontaneous saccade-related potentials (black arrows), in flight, at different injected currents. The magnitude of these sample saccade-related potentials were also not strongly changed by varying the soma’s Vm from −90 mV to −28 mV.

Figure S5. The relationship between pre-saccade Vm and SRP amplitude for contraversive saccades, shown for individual cell types. Related to Figure 7.

We plot the average relationship between pre-saccade Vm and SRP amplitude for contraversive saccades, exactly as in Figure 7B, but for each individual cell class separately. Each dot represents cell-class-averaged values for different stimulus conditions. Note that all cell classes show a negative relationship between pre-saccade Vm and SRP amplitude. HSN cells show the steepest linear relationship and most negative zero-crossing of the three HS cell classes, suggesting that they receive the strongest inhibitory input (top row). SRPs in VS1, VS2 and VS3 (middle row) all invert sign, consistent with a driving force mechanism to explain SRP amplitudes (see Main Text). VS4 and VS5 seem to receive similar inputs (bottom row). More data points are needed to properly interpret VS6 (and HSS) data.

Figure S6. Ipsiversive saccades are associated with depolarizing SRPs in HS cells and with very weak SRPs in VS cells. Related to Figure 5.

(A) Sample Vm and L-R WBA traces from an HSN cell in the context of a uniformly lit screen and during presentation of yaw visual motion. L-R WBA traces were low-pass-filtered with a 10 Hz cutoff frequency. Stimulus presentation is highlighted by a gray rectangle; dashed lines facilitate comparisons of baseline changes and SRP magnitudes over the entire time interval.

(B) Same as B, but data are shown for an HSS cell with a counter-clockwise roll visual stimulus.

(C) Average baseline-subtracted saccade-related potentials (SRPs) during ipsiversive (rightward) saccades for all stimulus conditions. Average SRPs from single cells are shown in gray. Average SRPs across all cells are shown in color. The average stimulus-driven (or blank screen) Vm immediately preceding each SRP is indicated in mV (arrows).

(D) Distribution of average SRP amplitudes across all seven visual conditions. The SRP amplitude was calculated as the mean Vm in a 50-ms window starting 75 ms after saccade onset from which we subtracted the mean Vm in a 150-ms window, starting 200-ms before saccade onset.

(E) Baseline subtracted saccadic L-R WBA signals, averaged across all flies, with all stimulus conditions overlaid.

Figure S7. Amplitude of saccade-related potentials during ipsiversive/rightward saccades in comparison to visual responses to optic flow associated with ipsiversive/rightward saccades. Related to Figure 6.

(A) Mean visual-response amplitudes (in mV) to the yaw-left optic-flow stimulus (black) and mean SRP amplitudes (in mV) for ipsiversive/rightward saccades generated during presentation of this stimulus (maroon), for HS, VS1-3 and VS4-6 cells (left) and all cell classes, individually (right). SRP amplitudes measured in the context of a blank screen are drawn in orange.

(B) The data in panel A, right, are shown here plotted one against the other. Magenta points show amplitudes of yaw-left visual responses plotted against amplitudes of SRPs during the yaw-left starfield stimulus. Orange points show amplitudes of yaw-left visual responses plotted against amplitudes of SRPs during a blank screen. Data points indicate mean +/− SEM.

(C–D) Same as panels A-B, but for downward pitch.

(E–F) Same as panels A-B, but for counter-clockwise roll. None of the scatter plots in panels B,D, and F show a significant correlation.

Movie S1. Immuno-amplified z-stacks for the two GAL4 driver lines used to genetically silence HS/VS cells. Related to Figure 2. We took a z-stack showing GFP signal in a R24E09>Kir2.1::EGFP fly. We took a second z-stack showing GFP signal in a VT058487>Kir2.1::EGFP fly, separately. In both flies the neuropil was also imaged with the nc82 antibody and this neuropil stain was used to computationally register both brains to the Janelia fly brain template (JFRCtemplate2010.nrrd; see Key Resources Table). We show the overlay of the two computationally aligned z-stacks in the movie. Interval between frames is 1 μm. The z-projection of the same z-stacks are shown in Figure 2D,G (although those panels also show expression in the thoracic ganglion). See STAR Methods for details regarding immunohistochemistry and imaging protocols.

Movie S2. Movies of the visual motion stimuli used in the present study. Related to Figures 2-7.

Our LED display surrounded the fly, ±36° in elevation and ±108° in azimuth, on a cylindrical display (see Fig. 4A). The stimuli are shown in this movie on a rectangular, flat-screen representation of the display.

Highlights.

• Optic flow-processing neurons participate in controlling head stability responses

• During flight turns, these neurons receive precisely tuned motor-related inputs

• These inputs suppress specific visual responses, while preserving other responses

• These modulations of visual signaling mute maladaptive head movements during turns

In a flying Drosophila, quantitatively tailored motor input modulates visual processing via sophisticated substractive computations, allowing removal of a specific sensory signal from a complex circuit carrying multiple related signals.