Responses of compass neurons in the locust brain to visual motion and leg motor activity

When moving through their environment, many animals rely on visual information to orient towards specific targets, to avoid collision with obstacles, or to make course corrections during long-distance migration. In insects, as in vertebrates, specific brain regions are concerned with spatial orientation and navigation tasks. By integrating internal and external sensory information, navigation centers in the brain infer the animal's current location and orientation in space and adjust its motor actions accordingly to achieve a desired translocation while also taking into account the animal's current internal and behavioral state.

In insects, the central complex (CX) is a key player in sensory–motor integration for spatial orientation (Pfeiffer and Homberg, 2014; Turner-Evans and Jayaraman, 2016; Varga et al., 2017; Heinze, 2017). The CX is a group of neuropils that span the brain midline. It consists of the protocerebral bridge (PB), the upper and lower divisions of the central body (CBU, CBL), and the paired noduli (Homberg, 1987; Hanesch et al., 1989; Pfeiffer and Homberg, 2014; Wolff et al., 2014). In many insects, sensory input to the CX is largely visual, and response properties often have a clear navigational context. Strong responsiveness has been found to the angle of polarized light from dorsal directions in field crickets, desert locusts, monarch butterflies, dung beetles and sweet bees (Vitzthum et al., 2002; Heinze et al., 2009; Sakura et al., 2008; Heinze and Reppert, 2011; el Jundi et al., 2015; Stone et al., 2017), approaching dark objects and small moving targets in locusts (Rosner and Homberg, 2013; Bockhorst and Homberg, 2015b, 2017), wide-field motion in cockroaches, flies and bees (Kathman et al., 2014; Weir et al., 2014; Weir and Dickinson, 2015; Stone et al., 2017), and azimuth-dependent presentation of light spots or vertical bars in flies, beetles, cockroaches, monarch butterflies and desert locusts (Heinze and Reppert, 2011; el Jundi et al., 2015; Seelig and Jayaraman, 2013, 2015; Turner-Evans et al., 2017; Varga and Ritzmann, 2016; Weir and Dickinson, 2015; Pegel et al., 2018, 2019). In flies, visual responsiveness of certain CX neurons depends on the animal's behavioral state (Seelig and Jayaraman, 2013; Weir and Dickinson, 2015; Turner-Evans et al., 2017).

The CX harbors numerous cell types (Müller et al., 1997; Heinze and Homberg, 2008; Hanesch et al., 1989; Wolff et al., 2014). However, a picture of which population of neurons codes for which kind of stimuli and to what extent responsiveness depends on a particular behavioral context is just beginning to emerge. One of the best understood networks of neurons within the locust CX encodes heading direction relative to sky compass signals (Pfeiffer and Homberg, 2014; el Jundi et al., 2015; Bockhorst and Homberg, 2015a; Pegel et al., 2018, 2019). Blue, cloudless sky shows a pattern of polarized light, which is used by many insects for spatial orientation (Horváth and Varjú, 2004; Horváth, 2014). Compass cells of the CX are responsive to the angle of polarization, indicated by the orientation of its electric field vector (E-vector), and mainly ramify in the PB and the CBL of the CX (Heinze et al., 2009; Homberg et al., 2011; Pegel et al., 2018). In desert locusts, the PB represents E-vector orientations in a compass-like manner, presumably determining the orientation of the animal with regard to the sun (Heinze and Homberg, 2007). Compass cells also respond to the simulation of approaching and bypassing objects (Rosner and Homberg, 2013; Bockhorst and Homberg, 2015b), suggesting that they are also involved in collision avoidance. This raises the question of whether additional sensory cues relevant during locomotion as well as internally generated signals during behavior are integrated here to aid in navigation under various behavioral circumstances. Therefore, we tested neurons of the network of compass cells for responsiveness to large-field motion and, in addition, for responses to approaching objects while large-field motion occurred. As the animals were free to move their legs, we present the first evidence for the involvement of the network in integrating the locust's motor activity state when processing visual information.

Experiments were performed on 27 adult desert locusts (Schistocerca gregaria Forsskål 1775) of either sex. Animals were reared under crowded conditions at a constant temperature of 28°C and a 12 h:12 h light:dark cycle. Animals were mounted vertically on custom-made holders (Fig. 1A), their wings shortened and the hindlegs removed. The remaining stumps, the head and mouthparts were immobilized by embedding in wax. The forelegs and middle legs were free to move. They had tarsal contact with a vertical glass plate positioned below the animal and covered with a lubricant (vegetable or silicone oil).

A hole was cut into the anterior head capsule to allow access to the central brain. Tracheal sacs and fat surrounding the brain were removed. The neural sheath was stripped away at the region where the recording electrode was inserted. The gut was removed to prevent pumping movements and, to further stabilize the brain, a wire platform supported the brain from the posterior. During recording of neural activity, the brain was submerged in locust saline (Clements and May, 1974) containing 0.09 mol l⁻¹ saccharose.

We provided visual stimuli via two laterally positioned LCD screens (Faytech, Shenzhen, China) and via a rotating polarizer (polarization filter HN38S, Polaroid, Cambridge, MA, USA; LED light source-ELJ-465-617-EPIGAP Optoelektronik, Berlin, Germany; rotation speed 30 deg s⁻¹, visual stimulation angle 3.2 deg, intensity 10¹⁴ photons cm⁻² s⁻¹) positioned dorsally to the animal (Fig. 1A). Screens had a resolution of 800×600 pixels and a frame rate of 75 Hz. On each side, the monitors covered 113 deg anterior–posteriorly and 127 deg dorso-ventrally. The Michelson contrast of monitor stimuli was 0.97 calculated with luminances of 70 cd m⁻² (bright monitor) and 1 cd m⁻² (dark monitor) measured with an OptiCAL photometer (Cambridge Research Systems, Cambridge, UK). Stimuli were written with MATLAB (MathWorks) using the Psychophysics Toolbox (Brainard, 1997; Pelli, 1997). Light from the displays was depolarized by diffuser sheets mounted in front of the displays. We presented sinusoidal stationary and progressively (backward) moving gratings with a spatial frequency of 1.2 cm cycle⁻¹ corresponding to 34 deg cycle⁻¹ in the screen center. Moving gratings had a temporal frequency of 7.5 cycles s⁻¹. We used different parameters only for the test of TL2 neuron responses to gratings moving in different directions (horizontal and vertical extent of grating 108 deg, spatial frequency 53 deg cycle⁻¹, temporal frequency 6 cycles s⁻¹). In several experiments, we also presented a dark looming disc stimulus in the center of one of the screens, simulating an approaching object on a collision course. The disc was presented either in front of the stationary grating background or in front of the progressively moving grating background (Fig. 1A). The simulated approaching object had a diameter of 6 cm, started from a virtual distance of 2 m and approached with a speed of 1 m s⁻¹. The disc stopped expanding 27 ms before the time of theoretical collision when covering a visual angle of 97 deg and stayed stationary for 500 ms before it disappeared.

We recorded intracellularly from 27 CX neurons while monitoring leg movements from below with a near-infrared sensitive DMK22BUC03 video camera with IC Capture software (The Imaging Source Europe, Bremen, Germany) at 10 frames s⁻¹ (Fig. 1A). In some experiments, we employed a Dalsa high-speed camera system (Stemmer Imaging, Puchheim, Germany) for evaluation of walking bouts at 100 frames s⁻¹. Each cell was recorded in an individual animal; 24 of these neurons were tested with unpolarized light stimuli and 10 neurons were investigated for responsiveness to polarized light while walking activity occurred. Walking behavior often was not coordinated sufficiently to extract information about directional behavior. We thus restricted our analysis to the presence or absence of leg motor activity during intracellular recordings. In some experiments, we mechanically deflected individual legs with a pin or a little brush to test for mechanosensory input to the recorded neurons or elicit walking activity. In order to exclude that visual cues associated with the stimulus were responsible for neural activity changes, we moved the brush in the vicinity of the legs without touching them (sham mechanical stimulation).

Microelectrodes were drawn from borosilicate capillaries (1.5 mm outer diameter, Hilgenberg, Malsfeld, Germany) on a microelectrode puller (P-97, Sutter Instrument, Novato, CA, USA). The tips of the electrodes were filled with 4% Neurobiotin (Vector Laboratories) in 1 mol l⁻¹ KCl and their shanks with 1 mol l⁻¹ KCl. The electrodes had a tip resistance of 100–180 MΩ. Signals were amplified (SEC5-LX amplifier, npi electronic GmbH, Tamm, Germany), digitized (CED1401 micro, Cambridge Electronic Design, Cambridge, UK), and stored using a PC with Spike2 software (Cambridge Electronic Design). About 0.25–1 nA of depolarizing current was applied for several minutes to iontophoretically inject Neurobiotin immediately after recording and, in some recordings, in between the stimulus sequences.

After completing the physiological recordings, brains were dissected out of the head capsule and fixed overnight in a mixture of 4% paraformaldehyde, 0.25% glutaraldehyde and 0.2% saturated picric acid in 0.1 mol l⁻¹ phosphate buffer. The labeled neurons were made visible for confocal laser scanning microscopy (Leica TCS-SP5, Leica Microsystems, Wetzlar, Germany) by treatment of the brains with Cy3-conjugated streptavidin (Dianova, Hamburg, Germany) as described previously (Rosner and Homberg, 2013).

Video recordings were analyzed by custom-written software in MATLAB, with Winanalyze tracking software (Mikromak, Berlin, Germany) or Kinovea tracking software (http://www.kinovea.org). In most cases, we determined only brightness changes in the videos without actually tracking individual leg movements, because this procedure proved to be much faster and gave unequivocal information about occurring leg movements. However, for all figures with leg motion data, we tracked individual legs, and calculated and summed the speed of the four freely movable legs. We also used the sum of the leg speed for cross-correlation of the leg motion data (recorded at 100 frames s⁻¹) with spiking data (see below). Before the cross-correlation was calculated, we convolved the spike train with a symmetrical triangular window with 10 ms length to estimate the spiking rate. Spiking rate as well as the leg speed trace were normalized by their standard deviations and then fed into the MATLAB xcorr-function. The cross-correlation was normalized by dividing it by the square root of the product of the auto-correlations (xcorr-function option ‘coeff’). This has the effect that the auto-correlations at zero lag equal 1.

Stained neurons were visualized with Amira (version 5.4.5, FEI Visualization Science Group, Mérignac Cedex, France), reconstructed using Adobe Photoshop CC (Adobe Systems, San José, CA, USA), and projected onto a standard CX (el Jundi et al., 2010).

As noted above, 24 of the 27 neurons of the study were tested with unpolarized light stimuli. Electrophysiological data were analyzed in MATLAB and Spike2. We employed built-in Spike2 functions for spike detection and visualizing recording traces (DC removal and simple thresholding) and also for visualizing spiking rates. All remaining analysis was done in MATLAB after exporting spike and stimulus times from Spike2.

Statistical significance of responses to grating motion was determined with the Wilcoxon signed rank test (MATLAB signrank function). We averaged spike trains without leg motion across all trials for each cell. One of the 24 animals that were tested with moving gratings showed leg movements in all trials and was thus excluded from data evaluation. The differences in spike number before grating motion onset and during grating motion were compared in the following way. We calculated the number of spikes within three 1 s time windows: time window 1 started 3 s before grating motion onset, window 2 started 1 s before motion onset and time window 3 started 1 s after motion onset. Thus, time windows 1 and 3 were symmetrical to time window 2. We calculated the difference of spike numbers between time windows 2 and 1 and, additionally, between time windows 2 and 3. Then, the absolute values for both differences were calculated for each cell and the resulting values were used in the statistical test. Because this analysis only determines whether the average compass cell responds to grating motion without determining whether a particular cell was inhibited or excited, we employed another procedure to determine inhibitory and excitatory responses. For this, we Gauss filtered (s.d. 200 ms, window size 1 s) average spike trains for each cell and determined the mean (±s.d.) spiking rate within a 2 s time window starting 2.5 s before grating motion onset. A cell was determined to be inhibited (or excited) if its average spiking rate was 3 (background) s.d. below (or above) its average background response continuously for at least 750 ms after grating motion onset. This allowed assessment of the responses of neurons with different response latencies and dynamics. We refer to cells that were neither inhibited nor excited as non-responsive.

For determining average responses across cells, we accounted for different background spiking rates in different neurons by subtracting the average background activities from the Gauss filtered traces. Because inhibitory and excitatory responses occurred in different recordings even from the same cell type, we inverted inhibitory responses by calculating the absolute value of the Gauss filtered traces. Finally, the average across cells was calculated.

Looming responses for individual cells were determined similar to moving grating responses: Gauss filtered average spike trains for each cell were calculated and the mean (±s.d.) spiking rate within a 2 s time window starting 2.5 s before looming onset was determined. A cell was determined to be inhibited (or excited) by a looming stimulus if a nadir (or peak) response, that deviated by more than 4 (background) s.d. from the average, occurred within a time window from 1.5 to 2.5 s after looming onset. This time window corresponds to the typical occurrence of looming responses found previously (Rosner and Homberg, 2013).

We evaluated the responsiveness of neurons to the plane of polarized light using circular linear correlation analysis (Berens, 2009; Bockhorst and Homberg, 2015a). First, we pooled trials with identical stimulus configuration (same direction of polarization filter rotation during walking and without walking). We calculated stimulus–response curves by dividing the pooled spike trains recorded during a full rotation of the polarization filter into 36 bins (10 deg bin width). In each bin, the average spiking activity was calculated. The angular–linear correlation was determined by using the angles of the polarization filter/E-vector in the center of each bin as the angular variable and the average spiking activities within the bins as the linear variable (Zar, 1999; Berens, 2009). The criterion for responsiveness was the significance of the resulting correlation coefficient (α=0.05). The corresponding P-value was determined as described in detail in Berens (2009).

Neurons that are sensitive to polarized light typically show a sinusoidal oscillation of firing rate with a period of 180 deg of the rotation of the polarizer. Thus, to determine the strength of the response to the rotating E-vector, we calculated the amplitude of this frequency component by determining amplitude spectra of firing rates after the curves had been smoothed with our Gauss filter (s.d. 200 ms) and the average spiking rate had been subtracted. When comparing amplitudes of trials during walking activity of the animals with those without walking activity, we evaluated all trials with walking activity against the same number of trials without walking activity. For comparison, we selected trials without walking activity that were closest to those with walking activity.

We studied the responses of 27 neurons with arborizations in the CX of the locust brain. All neurons are part of a network of compass neurons that are sensitive to the oscillation plane of polarized light (Heinze and Homberg, 2009; Heinze et al., 2009). The responses of these neurons to large-field visual motion with and without concurrent presentation of a looming disc and/or to polarized light with and without concurrently occurring leg motion were analyzed. We studied cells at each level of information processing (Fig. 1B, Table 1), including inputs into the CX mediated by tangential neurons of the CBL (TL2 neurons), intermediate neurons, i.e. columnar neurons connecting the CBL to the PB (CL1 neurons) and tangential neurons of the PB (TB1 neurons), and output cells projecting from the PB to the lateral accessory lobes (CPU neurons).

Bilateral presentation of moving gratings resulted in activity changes in neurons at all levels of the polarization-vison network in the locust CX (Table 1). In two recordings from TL2 neurons, a range of translatory and rotatory moving grating stimuli were presented to both eyes (Fig. 2). Irrespective of movement direction, one of the two TL2 neurons showed inhibitory responses (Fig. 2C), while the second TL2 neuron was excited (Fig. 2D). In these and other experiments, some (Fig. 2C) or all stimulus presentations (Fig. 2D) were accompanied by leg movements elicited by the moving grating.

Except for the two TL2 neurons, all neurons were almost exclusively studied for effects of translational front to back (progressive) motion as experienced by a forward-moving animal. Only in a few cases did we also briefly test regressive large-field motion, which gave qualitatively the same results as progressive motion stimulation (not shown). This means neurons that were excited (or inhibited) by progressive motion were also excited (or inhibited) by regressive motion. To differentiate between visually elicited and motor activity-related neuronal responses, we excluded trials with leg movements from the analysis. As illustrated in Fig. 2, responses to the moving grating could differ substantially between different representatives of the same cell type (see also Table 1) but were consistent across trials in a given recording. In TL2 neurons, CL1 neurons and TB1 neurons, we found representatives that were excited or inhibited, while of the three CPU neurons tested, two were not responsive to the stimulus and the third was excited. In order to evaluate the responses across different recordings statistically, we therefore had to normalize the recordings to identical activity levels before the start of grating motion and, further, to invert inhibitory responses for pooling with excitatory responses (see Materials and Methods). Following these procedures, the average responses had a similar phasic–tonic dynamic profile in all cell types, with a strong response onset and a gradual decay of response amplitude over the following 4 s (Fig. 3). Statistical evaluation of spiking activities in symmetrical time windows before and after motion onset revealed the significance of responses across recordings (Fig. 3; Wilcoxon signed rank test, n=23, P=0.0008).

To investigate whether large-field motion has an effect on the response to an approaching object, we combined stimulation by bilaterally moving gratings with stimulation on one side of the animal by a looming disc stimulus. Average responses to the monocular looming stimulation occurred later in time when the looming disc was presented in front of a moving grating instead of a stationary one (Fig. 4). In some recordings, the responses to the looming stimulus were opposite to those to the moving grating (Fig. 5, Table 1). This means that grating motion was inhibitory while looming stimulation was excitatory (Fig. 5A–E) or vice versa (Fig. 5F). In some cases, looming responses only became obvious during concurrent large-field motion (Fig. 5C).

An influence of motor activity on neuronal activity of compass neurons could be observed at all levels from input (Fig. 2; TL2 neurons) over intermediate stages (Fig. 6; CL1 neuron) to output (Fig. 7A; CPU1 neuron) even though not every neuron was affected by motor activity. Leg motor activity had an inhibitory effect on background activity in a TL2 neuron and during visual motion stimulation (Fig. 2C). The neuron was also inhibited by grating motion, and inhibition was strongest, when grating motion and leg motion occurred concurrently. The CL1 neuron in Fig. 6 was excited when leg movements occurred spontaneously or were elicited mechanically. The excitations occurred on a short time scale, coinciding with individual leg movements (Fig. 6B). The average cross-correlogram (Fig. 6C), from three high-speed recordings of leg movements and the corresponding spike rate, peaked 30 ms after leg motion.

We tested eight labeled neurons for their responses to visual stimulation through a rotating polarizer during walking-like behavior and included two further neurons in our analysis whose morphological identity was not known owing to unsuccessful staining. The results were diverse and no clear picture across all cells emerged (Table 2). However, in some cells, motor activity of the animal was clearly correlated with an increase in response amplitude to polarized light (Fig. 7). A CPU1 neuron responded more strongly when the animal was walking, and the response was weaker again after walking had stopped (Fig. 7A). Similarly, for the two neurons of unknown type, the responses became stronger during walking, were reduced in amplitude when the animals stopped walking and increased again when the locust resumed walking (Fig. 7D,F). One of the four remaining responsive cells, a CL1 neuron, responded significantly only to the polarizer rotation when the animal was motionless but not when it was walking, and while the other three cells responded more strongly during walking, overlapping s.e.m. of the amplitude spectra indicate that the differences were not statistically significant.

We show here that locust compass cells comprising the neural network for processing polarized light information are also responsive to large-field motion. Responses to looming stimuli in front of moving gratings were often opposite to responses to the large-field motion alone, making looming stimulus responses more salient. Walking activity often influenced spiking rate at all stages of the network, and some neurons responded more strongly to polarized light stimulation during ongoing leg motor activity.

Sensitivity of CX neurons to large-field motion has been reported in three other species, the cockroach Blaberus (Kathman et al., 2014), the fruitfly Drosophila (Weir et al., 2014; Weir and Dickinson, 2015) and the sweat bee Megalopta (Stone et al., 2017). However, in the cockroach and fruitfly, the types of neurons were not determined or were different from those presented here. In our study, all cells investigated belong to what Stone et al. (2017) referred to as compass cells, because of their proposed function in determining heading direction relative to the outside world. Stone et al. (2017) did not find distinctive responses to large-field motion in compass neurons in the bee. Instead, they discovered a new class of neurons, characterized by directionally selective sensitivity to translatory large-field motion. Those neurons ramified in the noduli, a subcompartment of the CX. In contrast to the results in the bee many compass cells in the locust responded vigorously to moving gratings. This discrepancy might be due to species-specific differences but could also be based on differences in stimulation such as different temporal frequencies tested.

The types of neuron studied here are sensitive to the angle of zenithal polarized light (Heinze et al., 2009). They probably play a role in navigation during migratory flights under open sky where the scattering of sunlight gives rise to a pattern of polarized light. It is conceivable that the neurons are also involved in course stabilization by exploiting optic flow information. Alternatively, optic flow might be an indicator for ongoing motion without being evaluated for any directional information content. In this study, we presented large-field motion concurrently to both eyes but did not test monocular stimulation. If cells responded differently to large-field motion to the left and right eye, it might explain why neurons of the same type showed inhibitory and excitatory responses. Stimulating the left and right eye concurrently could yield an inhibitory or excitatory response, depending on which eye is the dominant one. Different contributions of the two eyes to the overall response could be useful in balancing the optic flow experienced by the locust. This balancing of optic flow on both sides is well known from bees (Srinivasan et al., 1991).

Large-field motion-sensitive neurons in the locust optic lobe have been characterized by Rind (1990), Gewecke and Hou (1993) and Homberg and Würden (1997). The neurons studied by Rind (1990) and Gewecke and Hou (1993) ramified within the lobula complex. They were directionally selective and thus behaved like lobula plate tangential cells in flies (Hausen, 1981; Hengstenberg, 1982). Our recordings from TL neurons (Fig. 2) suggest that CX neurons of the polarization-vision network are not directionally sensitive. Therefore, the non-directionally sensitive neurons recorded by Homberg and Würden (1997) are more likely to be candidates for providing input to the neuronal network investigated here.

Several compass cells responded to looming and moving grating stimulation with opposite sign or became sensitive to looming stimuli only during moving grating stimulation. This behavior of the neurons could play a role during flight in a large swarm when the animals experience large-field motion and at the same time need to avoid collision with conspecifics or obstacles. The origin of the opposite sign of the looming disc and moving grating response is not clear but could be due to separate neuronal pathways feeding into the CX network or rebound effects due to the disc blocking out grating motion in part of the visual field and looming stop during the stationary disc phase. This would also explain why the average absolute responses to looming stimulation in front of a moving grating occurred later than responses to looming stimulation in front of a stationary grating.

Some neurons showed a remarkable increase in responsiveness to the angle of polarized light while walking-like behavior occurred, similar to an increased responsiveness to unpolarized light stimuli during flight within the fan-shaped body of flies (Weir and Dickinson, 2015). Responses of visual interneurons in the optic lobes of locusts and flies had earlier been found to depend on the animals' arousal or motor activity state (Rind et al., 2008; Chiappe et al., 2010; Maimon et al., 2010; Rosner et al., 2010). In locusts, an identified looming-sensitive neuron is less prone to habituation in an elevated arousal state (Rind et al., 2008) and thus presumably keeps the animal responsive to imminent collisions during locomotion. In flies, a rather complex picture of state-dependent visual responses has emerged. The amplitude of head optomotor responses to large-field motion is largely increased during locomotor activity, and the responsible gain modulation was concluded to be mediated downstream of the visual system (Rosner et al., 2009; Haag et al., 2010; Rosner and Warzecha, 2011). However, even at an earlier stage, at the level of visual interneurons mediating these optomotor responses, state-dependent modulation takes place (Chiappe et al., 2010; Maimon et al., 2010; Rosner et al., 2010; Jung et al., 2011; Rien et al., 2012). The effect of locomotor activity on visual interneurons is mediated by octopamine in flies (Longden and Krapp, 2009; Suver et al., 2012) and locusts (Bacon et al., 1995; Stern, 1999; Rind et al., 2008) as had been found earlier for bees (Kloppenburg and Erber, 1995). The response changes within and downstream of the visual system are mediated by a central signal rather than by proprioceptive feedback (Rosner et al., 2009, 2010; Haag et al., 2010; Fujiwara et al., 2017). Similar to flies, gain modulation of visual responses in locusts during flight occurs at the level of descending neurons (Reichert et al., 1985), and it can be expected that state-dependent modulation of compass cells is similarly complex to modulation in the fly optomotor system.

Similar to our data on walking behavior, Homberg (1994) showed that bouts of flight activity go along with spike rate changes in locust compass cells. Evidence was presented both for a central signal and for proprioceptive feedback signaling flight activity, as described for cockroach CX neurons during walking (Bender et al., 2010; Martin et al., 2015). Our experiments, likewise, suggest that changes in spike rate occur upon mechanical stimulation of single legs (Fig. 6A,B) and during spontaneous leg movements. To what extent these spike rate changes are caused by proprioceptive input from walking legs and central gain modulation during locomotor activity remains to be established. A multitude of mechanisms reporting the animal's current behavioral and internal state may, in fact, be expected from the large variety of neuronal inputs into the CX expressing a broad range of modulatory neurotransmitters (Pfeiffer and Homberg, 2014).

We thank Martina Kern and Jutta Seyfarth for technical assistance.