To characterize visual descending neurons in the hoverfly we performed extracellular recordings (Fig. 1) in the cervical connective while stimulating the animal with visual stimuli. The data were spike sorted using the action potential amplitude and width of individual waveforms (Fig. 1). We found that some descending neurons were selective to the motion of small targets crossing a small part of the visual field (Fig. 1a). Such TSDNs have previously been described in dragonflies, robberflies and hoverflies (Olberg 1981; Gonzalez-Bellido et al. 2013; Nicholas et al. 2018a). Other neurons appeared to be more similar to looming neurons (Fig. 1b), previously described in, e.g. Drosophila and locusts (see, e.g. Santer et al. 2008; Peron and Gabbiani 2009; Fotowat et al. 2011; Yakubowski et al. 2016; Zacarias et al. 2018; Ache et al. 2019a, b). Other descending neurons (Fig. 1c, d) shared response properties with optic flow sensitive descending neurons previously described in Drosophila and larger flies (see, e.g. Wertz et al. 2008, 2009b; Suver et al. 2016). In the following sections, we will describe how these descending neurons can be identified based on their receptive fields, and how their responses to different visual stimuli vary from each other.
Descending neurons can be clustered based on their receptive field properties
We first mapped receptive fields using small, local sinusoidal gratings (modified from Krapp and Hengstenberg 1997; Straw et al. 2006). For this, a square patch with an average side of 38° was placed in a pseudo-randomly chosen location on the screen (pale blue, Fig. 2a-i) out of 48 overlapping possible positions (Fig. 2a-ii). For each location we also recorded the spontaneous rate preceding stimulation with the sinusoidal grating, which moved in eight different directions. By fitting a cosine function to the mean response for each direction, we could extract the local preferred direction (arrowhead, LPD, Fig. 2a-iii) and the local motion sensitivity (black vertical bar, LMS, Fig. 2a-iii) in each location (modified from Straw et al. 2006). We displayed this as vectors, with the preferred direction given by the vector angle and the sensitivity by its length (Fig. 2b-i). In addition, we calculated the average spiking frequency for each location, and after subtracting the spontaneous rate, we spatially interpolated this ten times (color coding, Fig. 2b-i).
As previously described (Nicholas et al. 2018a), TSDNs do not respond to sinusoidal gratings, so this technique could not be used for mapping their receptive fields. For all other descending neurons that we recorded from (N = 96), we extracted the center of each receptive field (blue circle, Fig. 2b-ii), the receptive field size (blue outline, Fig. 2b-ii), and the average preferred direction (gray arrows, Fig. 2b-ii). When plotted in three dimensions the receptive field data clearly form three main clusters, with no overlap (Fig. 2b-iii). Two neurons did not cluster with the others, and were, therefore, removed from further analysis (red crosses, Fig. 2b-iii). We are referring to the three clusters as looming sensitive (N = 28, magenta outline, Fig. 2b-iii), optic flow sensitive 1 (N = 24, green outline, Fig. 2b-iii) and optic flow sensitive 2 (N = 42, blue outline, Fig. 2b-iii), as justified below. All data from these neurons were included in subsequent analyses.
We next averaged the receptive fields of all neurons in each of the three clusters (Fig. 2b-iii). The neurons in the first of these clusters, ‘Looming sensitive’, have receptive fields with highest sensitivity in the ventral visual field, close to the visual midline, with local preferred direction away from the midline (N = 28, Fig. 2c-i). The neurons in the second cluster, ‘Optic flow sensitive 1’, have receptive fields in the dorsal visual field with preferred motion away from the visual midline (N = 24, Fig. 2c-ii). The local preferred direction pattern (Fig. 2c-ii) appears to follow the elevation lines, suggesting that it should respond optimally to yaw rotations in the dorsal visual field. The ‘Optic flow sensitive 2’s’ receptive field (N = 42, Fig. 2c-iii) responds to downward motion across a large part of the visual field. The local preferred direction pattern (Fig. 2c-iii) suggests that it should respond optimally to roll rotations.
Looming neurons respond to the rapid growth of looming stimuli
We found that the descending neurons that we have referred to as looming sensitive (magenta outline, Fig. 2b-iii) respond weakly, or not at all, to the appearance of a large circular disc (magenta data, Fig. 3a-i), with a similarly weak response to a stationary disc whose luminance changes over time (magenta data, Fig. 3b-i). In contrast, the optic flow sensitive neurons respond strongly to both of these stimuli (green and blue data, Fig. 3a, b-ii, iii). When stimulated with a looming stimulus with an l/|v| of 10 ms both looming and optic flow sensitive neurons respond strongly (Fig. 3c). By comparing the response to the looming stimulus and the luminance-matched control (Fig. 3d) or to the appearance control (Fig. 3e), it is clear that the looming neurons prefer the looming stimulus (magenta data, Fig. 3d, e), similar to looming sensitive neurons described in, e.g. Drosophila and locusts (see, e.g. Klapoetke et al. 2017; Dewell and Gabbiani 2018). However, the neurons found in the second and third clusters (blue and green outlines, Fig. 2b-iii) are not responding to the looming as such, but rather to the rapid luminance change associated with the rapidly growing disk (green and blue data, Fig. 3d, e).
We next looked at the timing of the response to the looming stimulus, where we defined zero as the time when the stimulus reached its maximum size (‘Max’, Fig. 3g). We found that 9 out of the 12 looming neurons reached the peak response earlier than the optic flow sensitive neurons and this difference was significant (p < 0.001, 2-way ANOVA, Fig. 3f, g). In addition, the looming neurons reached their peak response significantly before the stimulus reached its maximum size (p = 0.0024, Wilcoxon signed rank test, Fig. 3f, g), similar to looming neurons described in locusts (Fotowat and Gabbiani 2007), crabs (Oliva and Tomsic 2014), and Drosophila (de Vries and Clandinin 2012; von Reyn et al. 2014). Indeed, the looming neurons start firing when the looming disc has a diameter of only 12°, and at the time the disc reaches its maximum diameter of 117°, the firing rate has saturated (magenta data, Fig. 3g-i). In contrast, the optic flow sensitive neurons reached their peak response around the time of maximum size (ns, Wilcoxon signed rank test, green and blue data, Fig. 3f, g), consistent with the hypothesis that they respond to luminance changes rather than looming per se. Note that when the looming stimulus reached its maximal size it covered the majority of the receptive fields of all three neurons (Fig. 3g). TSDNs do not respond to looming stimuli, or to any of the controls.
Both target-selective and looming neurons respond to small targets
We next investigated the size tuning of the descending neurons. For this purpose we first determined the preferred direction of each neuron, in response to small targets (3° square) scanning the screen vertically and horizontally along 20 evenly spaced trajectories, respectively (modified from Nordström et al. 2006). The example data show a TSDN that responds preferentially to leftward motion (Fig. 4a-i), and less to the other three directions (Fig. 4a-ii–v). In addition, we determined the elevation which gave the strongest response to the small target (Fig. 4a-vi).
We next quantified the size tuning of the descending neurons by scanning a 3°-wide bar through the area of peak sensitivity (Fig. 4a-vi), in each neuron’s preferred direction (Fig. 4a-v). Between scans we varied the height of the bar (the side perpendicular to the direction of travel) in a random order. We found that the size tuning for TSDNs was similar to that previously reported (Nicholas et al. 2018a) with a peak response to bars subtending a few degrees of the visual field and no response to larger bars (N = 7, Fig. 4b-i). In contrast, looming sensitive neurons (N = 12, Fig. 4b-ii) show a bimodal size selectivity, with one response peak to bars subtending a few degrees of the visual field, similar to the TSDN size tuning (Fig. 4b-i), followed by a dip to larger bars, and then an increased response as the bar was extended to cover the height of the screen. This result does not depend on the analysis window used (data not shown). The optic flow sensitive descending neurons give a stronger response to larger bars with a similar size dependence for optic flow sensitive 1 (N = 6, Fig. 4b-iii) and optic flow sensitive 2 (N = 7, Fig. 4b-iv). Note that the optic flow sensitive descending neurons did not respond as strongly to bars as the looming neurons did.
Looming and optic flow sensitive neurons respond to widefield sinusoidal gratings
As described above both optic flow sensitive and looming sensitive descending neurons respond to small sinusoidal gratings (Fig. 2). We next investigated this sensitivity in more detail, using full-screen stimulation. We first determined the direction selectivity by using a sinusoidal grating with a temporal frequency of 5 Hz and a wavelength of 7° and found that looming neurons responded best to leftward motion (magenta data, Fig. 5a), consistent with their receptive fields (Fig. 2c-i). The direction sensitivity of the optic flow sensitive neurons was also consistent with the receptive fields (compare Fig. 5a and Fig. 2c-ii, iii), with a peak response to downward motion for optic flow sensitive 2 (blue data, Fig. 5a), whereas the optic flow sensitive 1 neurons preferred motion at an upwards angle (green data, Fig. 5a). The direction tuning was significantly different between the three neurons (p < 0.0001, 2-way ANOVA, Fig. 5a).
When stimulated with gratings with different temporal frequencies, with a wavelength of 7°, the looming neurons did not respond strongly to low temporal frequencies but the response increased vigorously to faster moving sinusoidal gratings (magenta data, Fig. 5b). The temporal frequency tuning of the looming neurons was significantly different from the response of the optic flow sensitive 2 neurons, but not of the optic flow sensitive 1 neurons (2-way ANOVA). The optic flow sensitive 1 neurons’ responses were significantly smaller than the optic flow sensitive type 2 neurons (compare green and blue data, p = 0.005, 2-way ANOVA, Fig. 5b).
In response to gratings with different wavelengths, moving at 5 Hz, the looming neurons gave a peak response at a wavelength around 40° (magenta data, Fig. 5c), whereas the optic flow sensitive 2 neurons gave a peak response at a wavelength of 30° (p < 0.0001, 2-way ANOVA, blue data, Fig. 5c). The response of the optic flow sensitive 1 neurons appeared to plateau at 20° (green data, Fig. 5c), and the response was significantly different from that of the optic flow sensitive 2 neurons (p = 0.0013, 2-way ANOVA). The three clusters of neurons were identified based on their receptive fields (Fig. 2c-iii); thus all respond to full-screen sinusoidal gratings (Fig. 5), whereas TSDNs do not (Nicholas et al. 2018a).
Looming and optic flow sensitive neurons respond to 3D optic flow
To investigate responses to widefield stimulation in more detail, we developed a perspective distorted, 3-dimensional starfield stimulus. For this purpose, we simulated a 4-m cubic space with the hoverfly placed in its center (Fig. 6a). The cube was filled with 2 cm spheres at a density of 100 per m3 (Fig. 6a), in which we simulated different types of optic flow. For example, during leftwards sideslip the entire space slides to the left (Fig. 6a). When this is projected onto the screen in front of the fly, dots that are simulated to be closer to the fly will move faster across the screen (bottom example, Fig. 6b) than dots that are simulated to be further away (central example, Fig. 6b).
We found that the 3-dimensional sideslip (50 cm/s) and yaw (50 °/s) stimuli excited both optic flow sensitive (blue and green, Fig. 6c) and looming sensitive neurons (magenta, Fig. 6c, note that the data points overlap each other). We also found that both looming and optic flow sensitive descending neurons responded well to thrust (Fig. 6c). However, the looming neurons’ responses were larger to a single looming object (magenta, Fig. 3) than to thrust motion (magenta, Fig. 6c). The optic flow sensitive 2 neurons responded strongly to lift and pitch motion (blue, Fig. 6c), whereas the optic flow sensitive 1 neurons responded to pitch in the opposite direction (green, Fig. 6c). Three-dimensional roll motion gave the most clearly differentiated responses, where both optic flow sensitive neurons responded, but looming neurons did not (Fig. 6c). The TSDNs did not respond to any of the optic flow stimuli.