Introduction

Animal escape behaviours are popular subjects for neuroethological investigation. To be successful, such emergency behaviours must be triggered quickly and reliably, and so relatively simple networks of large neurons have sometimes evolved for this function. Most escape movements are triggered as soon as a threat is detected, which sometimes allows activity in identified neurons to be associated with the behavioural response (e.g. Edwards et al. 1999; Korn and Faber 2005). However, in some cases an animal must prepare to escape before the escape movement itself can begin, meaning that escape can only be successful if preparations begin early.

On the ground, locusts evade predatory threats by jumping using their powerful hindlegs. In order to produce sufficient force for a jump, the hindleg tibia is flexed against the femur (cocking) and the flexor and extensor tibiae muscles are co-contracted for a period of up to 500 ms (mean co-contraction duration 110 ms in loosely tethered locusts) (e.g. Bennet-Clark 1975; Heitler and Burrows 1977; Pflüger and Burrows 1978; Burrows 1995; Santer et al. 2005b). During co-contraction, structural components of the femur are deformed, storing energy for the jump (Heitler 1974; Burrows and Morris 2001). Following co-contraction, a jump is triggered by relaxation of the flexor muscle (Heitler 1974; Heitler and Burrows 1977). Thus, in order to successfully evade a predatory threat by jumping, a locust must prepare to do so some 100 or so ms beforehand.

Locusts possess a pair of large and uniquely identifiable visual neurons which are one way that they can detect an approaching predator. These are the descending contralateral movement detectors, or DCMDs (Rowell 1971; O’Shea et al. 1974), which respond to a looming (approaching) visual stimulus with a train of spikes that increase in frequency as the object gets nearer (Schlotterer 1977; Rind and Simmons 1992). The DCMDs descend from the brain to the thoracic ganglia where they excite some of the motor and interneurons that mediate movements of the hindlegs (Burrows and Rowell 1973; Pearson et al. 1980; Pearson and Robertson 1981). This has prompted speculation that the DCMDs play a significant role in mediating a jump. Because the DCMDs are said not to respond to a visual stimulus presented during co-contraction (Heitler 1983), and because no correlation between DCMD spikes and jumping movements was found (Rowell 1971), one possibility is that the DCMDs are involved in a locust’s preparations to jump (i.e. the initial flexion and co-contraction phases), rather than in jump triggering itself (Burrows 1996), although the DCMDs do not make direct connections with hindleg flexor motor neurons. Recent experiments in which video and EMG recordings were correlated with DCMD activity recorded later in the same locusts, indicate that none of the stages of the escape jump are triggered with a fixed delay relative to the timing of the peak in DCMD spike frequency (Fotowat and Gabbiani 2007). Instead, the rising, peak and decaying spike frequency phases of the DCMD response are proposed to underlie jump production (Fotowat and Gabbiani 2007), but recordings of DCMD activity during the production of escape jumps are needed to support or refute such a supposition.

In this study we investigate the possibility that the DCMD mediates hindleg cocking by studying tethered locusts responding to looming stimuli. For the first time, we present recordings of DCMD spiking activity while locusts experienced looming stimuli and were free to move normally on a treadmill. We find that flexion of the hindlegs reliably occurs with a precise timing that correlates with high-frequency spikes in the DCMD. However, challenging these locusts with stimuli that expand at a constant rate causes no burst of high frequency DCMD spikes but still causes hindleg flexion to occur, although with increased variability in timing. Sectioning the connective containing a DCMD axon affects the timing of hindleg cocking in response to a looming stimulus but does not prevent it from occurring. Thus we propose that the DCMD may play a contributory role in a locust’s preparations to jump, but that it is not solely responsible for triggering this behaviour.

Methods

Experiments were conducted on adult Locusta migratoria L. from a crowded colony at Newcastle University, UK, that was periodically augmented with locusts from a colony at Durham University, UK, and from an outside supplier (Blades Biological, UK).

Behavioural experiments

In order to explore the role of the DCMD in a locust’s jumping behaviour it was important to restrain locusts in such a way that controlled visual stimuli could be delivered and escape jumps could still be triggered. To achieve this, locusts were tethered onto a brass rod via the dorsal pronotum and stood on a treadmill consisting of a 10 cm diameter polystyrene ball mounted on bicycle bearings, which allowed the locusts to walk in any direction (Fig. 1). Visual stimuli were displayed on a Kikusui COS1611 X–Y monitor using a green P31 phosphor placed perpendicular to the locust’s long axis and elevated slightly so that stimuli appeared to approach the locust from just above where the horizon would be. Under these conditions, locusts would reliably respond to a looming stimulus with cocking and sometimes jumping behaviours. During a jump, the locust’s leg movements would cause the ball it walked on to spin, allowing the jump to be completed normally.

Fig. 1
figure 1

Experimental set up. a Locusts were tethered via the dorsal pronotum to a brass bar and stood on a polystyrene ball mounted on bicycle bearings. In this situation, movements of the locust’s legs would rotate the ball, allowing it to “walk” freely. The tether ensured that locusts were optimally placed for the delivery of controlled visual stimuli from a monitor screen mounted perpendicular to a locust’s long axis so that objects appeared to loom towards its right eye. b In this situation locusts would reliably respond to looming stimuli by flexing their hindlegs in preparation for a jump (cocking). Arrow indicates flexion of the hindleg from the typical standing position shown in a

Full size image

Looming stimuli were generated using a Cambridge Research Systems VSG2/3 image synthesiser and RG2 raster generator. These stimuli were displayed on the monitor screen, which was refreshed at 200 Hz. Stimuli were 80 mm diameter circles displayed in the centre of the monitor screen that appeared to approach the locust at a range of constant speeds from 0.2 to 8 m/s and finished their approach level with the monitor screen, 7 cm from the locust’s eye. Looming objects are characterised by an increasing rate of angular subtense expansion on the viewer’s eye (Wheatstone 1852), and because our stimuli were presented monocularly, the time course of their angular subtense expansion depends on the ratio of the stimulus object half size, l, to its approach speed, |v| (e.g. Hatsopoulos et al. 1995; Gabbiani et al. 1999). For example, an object of 8 cm diameter approaching at 5 m/s (l/|v| = 8 ms) has the same profile of angular expansion over time as a 16 cm diameter object approaching at 10 m/s (l/|v| = 8 ms). For this reason, we describe our stimuli both in terms of their simulated sizes and speeds, and their l/|v| ratios. In their final position level with the monitor screen, stimuli subtended 60° on the locusts’ eyes. Each stimulus was presented to an individual locust ten times with a 2.5-min inter-stimulus interval.

Locusts’ responses to looming stimuli were filmed with a Redlake Motionscope PCI high-speed digital video camera at 125 fps. These recordings were synchronised with the stimulus display software (using a voltage pulse controlled by the software), and were used to measure the timing of a locust’s responses to the presented visual stimuli. Hindleg flexion timing was defined as the first frame from the recorded films that flexion of the hindleg femur–tibia joint was observed.

DCMD recordings and nerve cord section

In some experiments we made DCMD recordings during behavioural experiments conducted as outlined above. Recordings were made as in Santer et al. (2005a). Briefly, a small window was cut in the ventral cuticle of the mesothorax and fat deposits were carefully removed to expose the right and left ventral nerve cords. Silver wire hook electrodes were hooked around the nerve cord contralateral to stimulus delivery and the cuticle was re-sealed with a melted mixture of beeswax and rosin. Recordings were differentially amplified and captured to disc using Spike2 v5 software for Windows and a micro 1401 analogue to digital converter (Cambridge Electronic Design, UK). Extracellular waveforms were sampled at 16.67 kHz. Recordings were high-pass filtered offline to remove lower frequency muscle potentials from higher frequency neuronal recordings (Santer et al. 2005a). DCMD spikes were then easily identified as the largest spikes in the recordings.

In other experiments we carried out behavioural observations on locusts in which one or other of their ventral nerve cords had been cut through. To do this, a window of ventral cuticle was cut as for hook electrode implantation. The nerve cords were exposed and one or the other was sectioned with fine scissors between the pro- and mesothoracic ganglia. The cuticle was then resealed with insect wax and the removed window of cuticle.

Results

Initially, we presented locusts with a range of computer-generated, 8 cm diameter looming stimuli approaching at a range of speeds. The locusts stood on a simple treadmill, allowing them to walk but keeping them in position for delivery of a visual stimulus (Fig. 1). Under these conditions, we classified a locust’s initial response to a loom as one of: (1) “cocking”, if the hind tibia was fully flexed against the hind femur; (2) “partial flexion”, if the tibia flexed but not fully against the femur; (3) “startle”, if the locust responded with a startle behaviour similar to that described by Friedel (1999); or (4) no response. In response to most stimuli, the dominant reaction was cocking (∼70% occurrence) or partial flexion. Jumping was not included as a behavioural category since it only occurred following the preparatory “cocking” behaviour. Jumps occurred in only a small percentage of trials (Fig. 2a). In response to these stimuli, we found a linear relationship between the time that the hindleg tibia first began to flex in response to a loom, and the size to speed (l/|v|) ratio of the looming stimulus, which describes the time course of stimulus angular expansion on the locust’s eye (Fig. 2b). We also found that the standard deviation of the timing of hindleg flexion increased with stimulus l/|v|, with greater variability in hindleg flexion timing in response to slower objects (Fig. 2c).

Fig. 2
figure 2

a The responses to computer-generated looming stimuli of locusts tethered on a treadmill. Bars show the cumulative frequency of responses by six locusts that each received ten presentations of each stimulus. b Where hindleg flexions occurred from the stimulus presentations in a, the timing of these movements correlated with stimulus size to speed (l/|v|) ratio (RSq 0.57). c The variability (indicated here by standard deviation) in flexion timing also correlated with stimulus l/|v| (RSq 0.86)

Full size image

Using a wider range of looming velocities, and a range of objects which expanded at a constant rate rather than looming, we found that the mean timing of hindleg flexion between stimuli often occurred at a given time after the object reached a fixed angular size on the monitor screen. This was evident in four locusts from a total of nine that regularly flexed their hindlegs in response to these stimuli. Figure 3 illustrates this relationship for one locust in which it was particularly apparent. Multiple flexions were recorded in this locust and mean flexion timing for each type of stimulus was used to align the plots of stimulus angular subtense. Stimulus subtense at t = 0 shows the stimulus subtense at which flexion was observed and any coincidence of the plots to the left of this indicates a common subtense between stimuli reached at a similar time prior to flexion. The angular size and the strength of the relationship differed between individual locusts.

Fig. 3
figure 3

For four of the nine locusts that flexed their hindlegs regularly in response to looming and constantly expanding stimuli, hindleg flexion (cocking) appeared to occur at a fixed time after the stimulus reached a certain angular subtense. This subtense varied between individuals. Here, this relationship is illustrated for a single locust presented with several speeds of looming (a) and constantly expanding (b) visual stimuli. For each stimulus, several hindleg flexions were recorded and the mean timing of hindleg flexion calculated. Stimulus subtense plots were then aligned using this mean time of flexion in response to each stimulus type. When the traces are aligned in this way a common triggering stimulus subtense should be revealed as a co-incidence of the angular subtense plots some time before flexion. Data in a are from loom speeds 0.2 m/s—six flexions, 0.4 m/s—four flexions, 0.8 m/s—four flexions, and 4 m/s—two flexions. Data in b are from edge movement speeds 14.3° per s—six flexions, 28.6° per s—five flexions, 57.2° per s—five flexions, and 114.4° per s—five flexions

Full size image

In order to investigate the possibility that DCMD spiking activity might mediate these hindleg flexions, we first used implanted hook electrodes to record DCMD spikes of locusts that were standing on a treadmill and stimulated with looming visual stimuli as before. In response to looming stimuli of two different l/|v| ratios [an 8 cm diameter object approaching at either 8 m/s (l/|v| = 5 ms) or 1 m/s (l/|v| = 40 ms)], we found that the times of hindleg flexion were closely associated with the times of peak spike frequencies in the DCMD response (Fig. 4a). Furthermore, we noted that the breadth of the time period over which these high peak spike frequencies were maintained matched the breadth of the time period during which hindleg flexions were initiated, i.e. the longer high DCMD spike frequencies were maintained, the greater the variability in hindleg flexion timing (Fig. 4a). To test the importance of these high spike frequencies, we next compared DCMD and behavioural responses to a looming stimulus that reached a large final subtense (an 8 cm diameter object approaching at 8 m/s; l/|v| = 5 ms), and a looming stimulus that followed a similar expansion profile but stopped moving earlier and at a smaller final subtense (a 1 cm diameter object approaching at 1 m/s; l/|v| = 5 ms). As expected, the smaller stimulus prevented the DCMD from producing such high frequency spikes. This stimulus only elicited hindleg flexion in a single stimulus presentation from a total of between 10 and 15 to each of six locusts tested (Fig. 4b). We also compared DCMD and behavioural responses to a looming stimulus (8 cm, 1 m/s; l/|v| = 40 ms) with responses to a stimulus in which the edges expanded at a constant angular rate (14.3° per s), rather than an increasing rate characteristic of a loom. The two types of stimuli had the same duration, start and finish subtenses. The constant expansion rate stimulus elicited a profoundly different DCMD spiking response from the looming stimulus—an initially high, then slowly declining spike rate rather than a peak towards the end of stimulus movement (Fig. 4c). Whilst hindleg flexions were associated with high spike frequencies in the DCMD at the end of the looming stimulus, they occurred much earlier and with far greater variability in response to the constant expansion rate stimulus (Fig. 4c). The notably different association between DCMD spiking activity and behaviour between looming and constantly expanding stimuli was clearly evident when we plotted DCMD responses across all of these stimuli relative to the time of hindleg flexion in each trial (Fig. 4d). The mean DCMD spike frequency reached a peak prior to hindleg flexion in response to looming stimuli but not constantly expanding stimuli. From our electrophysiological recordings made during a locust’s preparations to escape a looming object we noted that the DCMD continued spiking following hindleg flexion (Fig. 5a, b).

Fig. 4
figure 4

The relationships between DCMD spiking activity and the timing of hindleg flexion in response to different visual stimuli. a–c Plots of stimulus subtense (bottom traces), mean DCMD spiking response (calculated as the mean of spike counts per 10 ms bin for the 60 stimulus presentations) (middle trace, circles) and flexion timing (top, box plots). a Looming stimuli of different l/|v| ratios elicited different DCMD response profiles. In response to the higher l/|v| stimulus, DCMD spike frequencies peaked prior to collision (l/|v| = 40 ms; open grey circles) and hindleg flexion behaviours correlated in time with this peak (open grey box plot). In response to the lower l/|v| stimulus, DCMD spikes reached high frequencies closer to the time of collision (l/|v| = 5 ms; filled black circles) and hindleg flexion occurred later, again correlated with the timing of these spikes (filled black box plot). High DCMD spike frequencies were maintained for longer, and flexion timing variability was greater in response to the higher l/|v| stimulus. b Looming stimuli whose maximum angular subtense was small and occurred relatively early before collision (l/|v| = 5 ms; open grey circles) evoked a weak DCMD response with lower peak DCMD spike frequencies and only a single hindleg flexion (open grey circle). Looming stimuli whose maximum angular subtense was larger and occurred close to the projected time of collision (l/|v| = 5 ms) evoked a stronger DCMD response with higher peak spike frequency and more frequent hindleg flexion (filled black circles and box plot). c Compared to a looming stimulus (8 cm diameter, 1 m/s, l/|v| = 40 ms; filled black circles and box), a stimulus that expanded at a constant rate (14.3° per s edge movement) induced initially high but slowly declining DCMD spike frequencies (open grey circles). This also induced earlier and more variable flexion timings (open grey box plot). d A peri-flexion time histogram showing DCMD spiking responses to the stimuli in a–c synchronised to the time that hindleg flexion (cocking) occurred in that trial. In response to looming stimuli, a DCMD spike frequency peak is evident prior to the time of flexion (circles and squares). In response to a constantly expanding stimulus, this is not the case (triangles). In a–c, DCMD responses are the mean of 10–15 stimulus presentations to each of six locusts. Bars are SEM and bin widths are 10 ms. Behaviour timings are for 1–6 flexions per locust during these stimulus presentations. In d, DCMD responses have been aligned with the time of flexion. Each curve is the mean of 1–6 DCMD responses in each of six locusts. Bars are SEM and bin widths are 10 ms (the 1 cm, 1 m/s curve is a single response by one locust)

Full size image
Fig. 5
figure 5

DCMD spikes continue throughout preparation for a jump. DCMD responses are shown for a looming (a) and a constantly expanding (b) stimulus. Stimulus subtense (bottom trace) and extracellularly recorded DCMD spikes (middle trace) are shown. The instantaneous frequency of DCMD spikes in the response is plotted in the top trace. Grey line indicates the timing of hindleg flexion (cocking) in each case. Following this, DCMD spikes continue

Full size image

In the previous paragraphs we described our finding that hindleg flexions were often associated with periods of high-frequency spikes in the DCMD neuron. To investigate whether DCMD spikes were necessary to trigger these flexions, we sectioned one or the other connective of a locust’s ventral nerve cord and investigated the locust’s responses to looming and constantly expanding stimuli. The DCMD descends in the contralateral nerve cord relative to the eye from which it receives input. So, if this cord is cut, the DCMD is ablated but all ipsilaterally descending visual neurons are left intact. In response to a looming stimulus, sectioning of the nerve cord ipsilateral to the stimulated eye (leaving the stimulated DCMD intact) did not prevent hindleg flexion from occurring (mean probability of occurrence was 0.56) or alter its timing (Fig. 6a). Sectioning the contralateral nerve cord (and therefore the stimulated DCMD) did not prevent hindleg flexion from occurring (mean probability of occurrence was 0.83) but did increase the variability in its timing (Fig. 6a). When we repeated this procedure using constantly expanding stimuli, we found that sectioning either nerve cord did not prevent hindleg flexion from occurring (mean probabilities of occurrence were: ipsilateral nerve cord cut—0.61; contralateral nerve cord cut—0.72), or alter its timing (Fig. 6b).

Fig. 6
figure 6

The effect of sectioning one or the other connective of the ventral nerve cord in the prothorax on the timing of hindleg flexion in response to a 4 m/s looming (a) and an equivalent constantly expanding stimulus (b; 57.2° per s edge movement speed). a The variability in flexion timing in response to a loom is increased by sectioning the contralateral, but not ipsilateral, nerve cord. b The variability in flexion timing in response to a constant expansion is unaffected by sectioning either nerve cord. Data in a are intact—19 flexions in six locusts, ipsilateral section—10 flexions in three locusts, contralateral section—15 flexions in three locusts. Data in b are intact—19 flexions in seven locusts, ipsilateral section—11 flexions in three locusts, contralateral section—13 flexions in three locusts

Full size image

Discussion

In this study we have investigated the possible role of DCMD neuron spikes in a locust’s preparation for escape by jumping. We found a close association between high DCMD spike frequencies and hindleg flexion (cocking) behaviour, in terms of both timing and response duration/flexion timing variability. However, constantly expanding stimuli that did not elicit high frequency spikes in the DCMD still triggered hindleg flexion, and severing the axon of the DCMD neuron between the pro and meso-thoracic ganglia did not prevent the occurrence of hindleg flexion but was accompanied by an increase in the variability of its timing. We therefore conclude that a DCMD cannot be necessary for triggering hindleg flexion during visual stimulation, but that it may contribute to evoking this movement.

The jumps of orthoptera are an effective escape strategy but one that takes considerable time to elicit due to the requirement for muscle co-contraction to build up sufficient force (e.g. Bennet-Clark 1975; Heitler and Burrows 1977; Pflüger and Burrows 1978; Burrows 1995; Santer et al. 2005b). In a previous study of the DCMD response to looming stimuli, Hatsopoulos et al (1995) noted that locusts often flexed their hindlegs at the approximate time of peak DCMD spike activity, but a more recent study reported that none of the phases of the escape jump are triggered with a fixed delay relative to the timing of the DCMD peak activity (Fotowat and Gabbiani 2007). Here, we report a correlation between flexion timing and the timing of high frequency DCMD spikes in locusts responding to visual stimuli. This correlation suggests that DCMD spikes may contribute to causing flexion of the hindlegs. Because flexor muscle activity likely precedes our measurements of the timing of joint flexion, rapidly increasing DCMD spike frequency would seem a plausible trigger for flexion (see also Fotowat and Gabbiani 2007). However, our experiments indicate in two different ways that the high frequency DCMD spikes cannot be solely responsible for triggering flexion. First, in experiments in which our nerve cord sections prevented DCMD spikes from reaching the meso- and metathoracic ganglia, flexion still occurred. Second, stimuli that expanded with a constant rate elicited no rapid increase in DCMD spike rate but did elicit hindleg flexion. Nevertheless, our data suggest that DCMD spikes may still contribute to causing flexion of the hindlegs.

The exact contribution of the DCMD to the production of hindleg flexion is unclear. DCMD activity has long been thought of as a potential influence on jumping behaviour due to the neuron’s large size and its output connections with leg motor- and interneurons involved in jump production (Burrows and Rowell 1973; Pearson et al. 1980; Pearson and Robertson 1981; Fotowat and Gabbiani 2007; Rogers et al. 2007). One potential route by which DCMD activity could elicit preparations for a jump is via the C (cocking) interneurons which have cell bodies in the mesothoracic ganglion and axons that descend to the contralateral side of the metathoracic ganglion. These cells receive excitation from a DCMD and from auditory and mechanosensory neurons (although none of these inputs alone can cause the cells to spike), and directly excite the fast extensor tibiae (FETi) and various flexor motor neurons powerfully enough to elicit a spike in them (Pearson and Robertson 1981). This lead to the proposal that the C interneurons could cause the cocking phase of jump preparation (Pearson and Robertson 1981). However, during bilateral kicks elicited spontaneously or in response to tactile or auditory stimuli, the C interneurons were active after co-contraction had started, indicating that they could not trigger cocking in these instances (Gynther and Pearson 1986). Furthermore, transections that should damage the C interneurons do not prevent jumps from occurring (Ronacher et al. 1988). Nevertheless, these cells are certainly a pathway which could mediate a DCMD’s influence on this phase of the locust jump, if only under the specific circumstance of escaping from a looming stimulus.

In this study, sectioning the DCMD axon (and other contralaterally descending axons between the pro and meso-thoracic ganglia) did not prevent hindleg flexion behaviour from occurring, but did increase the variability in its timing. Potentially, this too may be explained by the DCMD’s connection with the C interneuron (Pearson and Robertson 1981). The DCMD has a characteristic response to a looming stimulus that contains peak spike frequencies at a predictable time relative to object approach (e.g. Gabbiani et al. 1999, 2001). Due to the DCMD’s connections with the C interneuron, these spikes would elicit a strong burst of EPSPs in the C interneuron that would occur at a particular time relative to object approach. In concert with other inputs, these could elicit activity in the C interneuron leading to activity in the extensor and flexor tibiae motor neurons, and this would be relatively precisely timed relative to object approach because the DCMD spike burst is. This would explain the relatively precisely timed flexion behaviours observed in response to looming stimuli. In contrast, under constantly expanding visual stimulation, or with the DCMD axon severed, there is no predictably timed burst of high frequency DCMD spikes impinging onto the C interneuron. As such, the C interneuron’s other inputs could still cause activity and cocking to occur, but with less precision in timing due to the lack of the strong and predictably timed DCMD spike burst.

It may also be possible that DCMD spikes influence the triggering phase of a jump. The DCMD excites the fast extensor tibiae (FETi) motor neuron directly—a neuron involved in jump triggering—with relatively small EPSPs in gregarious phase locusts and much larger EPSPs in solitarious phase locusts (Rogers et al. 2007). However, facilitation increases EPSP amplitude in gregarious locusts, contributing to a mechanism of homeostatic plasticity that ensures that the compound EPSPs elicited by a complete loom are of comparable amplitude in gregarious and solitarious locusts (Rogers et al. 2007). Furthermore, this process causes peak excitation of the FETi by the DCMD to occur much earlier for gregarious phase locusts (Rogers et al. 2007). However, the compound EPSPs elicited by a DCMD’s response to a loom are insufficient on their own to cause FETi to spike, and this connection is lacking entirely in some laboratory locust populations, making its possible role in triggering an escape jump difficult to interpret at present (Pearson and Goodman 1979; Rogers et al. 2007). The DCMD does, however, strongly excite the multimodal M neuron which inhibits flexor motor neurons (Steeves and Pearson 1982). Although this cell does not act alone in triggering a jump, this may be one way by which a DCMD could influence this process. Recently, timing associations between phases of the DCMD response in restrained locusts, and the timing of jumping movements in separate experiments where they were free to move, have been found (Fotowat and Gabbiani 2007). However, if these DCMD response phases are to underlie jump triggering, the DCMD must continue to respond throughout jump preparation, since it does not during preparations to kick (Heitler 1983). Here we show that the DCMD does indeed continue to spike following hindleg flexion and thus, it is at least possible that a DCMD neuron could also play a role in the triggering phase of a jump.

High frequency DCMD spikes have previously been implicated in triggering emergency gliding dives during flight that are suited to the evasion of flying bird predators (Santer et al. 2006). A glide is a last chance evasive response because it is triggered by high frequency DCMD activity which occurs at the very end of the approach of a small, bird-like looming object. This ensures that the glide is triggered at the last possible moment when course adjustment by a looming bird predator is impossible. In contrast, escape jumps take considerable time to trigger, due to the requirement for a period of co-contraction prior to the jump. However, preparatory hindleg flexion behaviour occurred too close to the end of stimulus motion in most of our experiments to elicit escape by jumping before collision would have occurred. When on the ground, predatory birds must land close to a locust rather than swooping to grab it. As a result, cocking late in a loom may prepare a locust to jump before a predatory bird lands, allowing it to jump after the bird lands and before it attacks. Larger looming stimuli have been reported to elicit a peak in DCMD spiking activity significantly before the end of stimulus movement (e.g. Hatsopoulos et al. 1995; Gabbiani et al. 1999, 2002). If such stimuli are better matches for terrestrial locust predators, this could indicate a role for the high frequency spikes comprising this peak in eliciting jump preparations so that sufficient time exists for escape. However, recent evidence argues against such a role for the DCMD spike frequency peak itself: Fotowat and Gabbiani (2007), using large, slow stimuli with l/|v| ratios of 40–120 ms, found that the timing of jump preparation or initiation in behavioural experiments was not correlated with the timing of DCMD peak activity measured subsequently to the same stimulus.