A motion-sensitive neurone responds to signals from the two visual systems of the blowfly, the compound eyes and ocelli

Most flying insects have a rather limited range of behavioural goals during flight. Besides `higher order' behaviours such as courtship and territorial maintenance, the basic pre-requisites for successful flight are simply stability, and collision avoidance. Of these two, stability is the more interesting in terms of multi-sensory integration, because it requires contributions from many of the insect's wide range of sensory systems(Hengstenberg, 1993; Sherman and Dickinson,2003).

Each modality has inherent, unavoidable physical or neural limitations that restrict its effectiveness as the sole sensor for a particular type of in-flight stimulus (Hengstenberg,1993; Keil, 1997; Krapp and Hengstenberg, 1996; Nalbach, 1993). Complementary sensory modalities can be very important, to cover a large dynamic range and therefore provide adequate information on flight perturbations. For instance,in the fly visual system, self-rotations are estimated by integrating the outputs of many elementary motion detectors (EMDs). EMDs receive their inputs from the compound eye and locally analyse directional motion. However, because an EMD, or Reichardt Correlator (Haag et al., 2004; Reichardt,1987), is a delaybased directionally selective mechanism, it requires extra processing time, and this limits self-motion estimation to a range of comparatively low angular velocities. To maintain flight and gaze stability in the higher angular velocity range, flies make use of information from other sensory systems including mechanoreceptors on the halteres(Fraenkel and Pringle, 1938; Nalbach, 1993; Nalbach and Hengstenberg,1994) and the wings(Fayyazuddin and Dickinson,1999; Heide, 1974),and the photoreceptor-based ocelli.

The ocelli (simple eyes) of the fly are a set of three small single-lens eyes which, by their construction, are well adapted to the task of rotation detection in the manner of an artificial horizon(Schuppe and Hengstenberg,1993; Stange et al.,2002; Taylor,1981a; Taylor,1981b; Wilson, 1978a). By having a retina that is attached to the rear surface of the lens, the images at the level of the photoreceptors are highly blurred, i.e. they are composed only of very low spatial frequencies. It is believed that the ocelli do not make use of the remaining image structure and are only concerned with integrating light over their entire visual fields. Only for the dragonfly median ocellus is there evidence that the ocelli utilise image structure(Stange et al., 2002), based partially on anatomical adaptations of the ocellus - an enlarged, astigmatic lens and an effective `lens hood', which are not present in flies. The ocelli are very effective rotation detectors, crucial to proper gaze and flight stabilisation in dragonflies (Stange,1981; Stange and Howard,1979) as well as in locusts(Taylor, 1981a; Taylor, 1981b). Ocellar contributions to gaze stabilisation have also been shown in the fly(Schuppe and Hengstenberg,1993).

The neural processing required for an ocellar-based rotation detector is relatively simple, at least compared to the processing required for the directional motion analysis of compound eye inputs. Ocellar second-order interneurones (L-neurones) integrate the outputs of photoreceptors and code light intensity information as graded potentials(Simmons, 1999; Simmons et al., 1994; Wilson, 1978b). Although the neural mechanism by which ocellar rotation detection is implemented is unknown, one possible means of investigating this would be to directly compare the outputs of L-neurones. L-neurones also have some of the largest diameter axons in the fly nervous system(Nãssel and Hagberg,1985; Simmons et al.,1994), which allow for fast signal propagation. It is these functional adaptations of the ocellar system that facilitate fast motor responses; in the locust, the ocelli are known to elicit compensatory head movements twice as fast as the compound eyes(Taylor, 1981a). However,although a rapid response is desirable, motor responses elicited by the ocelli may in general be quite jerky in comparison to those mediated by the compound eyes, as has been observed in dragonfly(Taylor, 1981a). The integration of compound eye and ocellar outputs can realize a visual detection of rotations that combines high speed with high precision. How and where does such integration occur in the nervous system of the fly?

As mentioned above, the fly visual system contains a class of interneurones that are matched filters for particular selfmotions(Krapp, 2000; Krapp et al., 1998). In the third visual neuropile of each half of the brain there are about 50 lobula plate tangential cells (LPTCs). Many of them make direct synaptic connections with descending neurones, which then convey sensory information to the various motor systems. The neuroanatomy of ocellar L-neurones is also suggestive of synapses with both LPTCs and descending neurones(Strausfeld, 1976; Strausfeld and Bassemir,1985). Intracellular staining has shown that the terminal arborisations of ocellar L-neurones and some VS neurones overlap with the dendritic fields of descending neurones. It is known that descending neurones respond to stimuli of different modalities(Strausfeld and Gronenberg,1990) but that the responses are quite complex, and this complicates the electrophysiological analysis of multi-modal convergence. To fully characterise the interactions of multiple modalities at a neuronal level, it is necessary to be able to deconvolve the mixed responses of individual cells.

Here we show that the activity of the well-studied tangential cell V1 (e.g. Hausen, 1993) is driven by illumination of the lateral ocelli. V1 was previously thought to be dedicated solely to processing visual information from the compound eyes. Our results suggest that V1 encodes a single, self-motion-related variable that is independently sensed by two separate sensory systems, the compound eyes and the ocelli. This discovery raises the possibility of a quantitative study of the interactions between signals from the ocelli and the compound eyes by monitoring the response of the identified V1 cell.

Specimens of Calliphora vicina Robinson-Desvoidy were taken from a culture in the Department of Zoology, Cambridge, UK. Only female blowflies were used, because the larger gap between the compound eyes on the dorsal side of the head capsule allows easier access to the ocellar nerve.

The legs, wings and proboscis of the fly were removed, and the wounds waxed shut. The fly was mounted on a custom-built copper/plastic holder. The head of the fly was aligned with the horizontal and vertical planes by using the deep pseudopupil (Franceschini,1975). The thorax and abdomen were bent ventrally by approximately 20° to gain access to the rear of the head capsule. The cuticle of the right rear head capsule was cut open and the air sacs were moved aside to expose the lobula plate (Fig. 1). A silver wire reference electrode was inserted into a small hole in the left rear head capsule.

In experiments where the ocellar nerve was to be cauterised, the cuticle of the medial rear head capsule between the neck and the ocelli was cut away to expose the ocellar nerve. Cauterisation was performed by inserting a fine loop of tungsten wire into the exposed air cavity containing the ocellar nerve, and passing enough current through the wire to irreversibly dehydrate, but not burn, the nerve. A photograph and a schematic of the preparation are shown in Fig. 1.

Stimuli to the lateral ocelli were delivered via a pair of 62.5μm diameter glass optical fibres, such that the light output of each fibre covered only the lens of one ocellus (ocellar diameter ∼100 μm) plus a minimum area of cuticle surrounding it(Fig. 1A). The fibres were connected to a pair of blue LEDs, the intensity of which was controlled by the analogue outputs of a data acquisition card (NI-6025E) at an update-rate of 5 kHz. The wavelength of the LEDs (λ_max=470 nm) was sufficiently well matched to the absorption spectrum of the ocellar photoreceptor pigment (Kirschfeld et al.,1988) and was identical to that used in a previous study of fly L-neurone responses to light stimulus(Simmons et al., 1994). The output intensities of the fibres were calibrated and normalised with each other using a photoresistor.

The maximum intensity deliverable by our optical fibres was defined as I_max, and we chose our stimulus parameter to be intensity difference, where LED intensity difference, DI=(Left LED output-Right LED output). Our ocellar stimuli could therefore vary from I_max (maximum intensity at the left ocellus, zero at the right) to -I_max (maximum intensity at the right ocellus,zero at the left).

We used a CRT monitor to present pattern motion stimuli to the compound eyes. The monitor was driven by a dedicated visual stimulus card (Cambridge Research Systems 2/5), which operated at a refresh rate of 180 Hz. Compound eye stimuli consisted of a horizontal grating moving vertically downwards. The spatial intensity distribution of the grating was sinusoidal. To adjust the strength of the stimuli, we adjusted the contrast of the grating, rather than the velocity. In this way, the ocelli would see no overall change in brightness during compound eye stimulation. The grating was moved at a fixed contrast frequency of 5 Hz, with a spatial wavelength of 8°, across a display window of approximately 90° azimuth by 70° elevation.

We recorded extracellularly from V1 using tungsten electrodes(manufacturer: FHC) with an input resistance of 3±0.6 MΩ. The extracellular neuronal signal was band-pass filtered (0.3-5 kHz), amplified,and then sampled by a data acquisition card (AM Systems 1700 differential AC and National Instruments 6025E, respectively) at a sample rate of 40 kHz onto the hard-drive of a computer running Windows XP.

In order to identify V1 recordings, the centre of the CRT was positioned at an azimuth of -45° and an elevation of 0° (grey area in Fig. 2) relative to the reference frame of the fly head capsule, where 0° azimuth, 0°elevation means the point directly in front of the fly. In this region of its receptive field, V1 is most strongly excited/inhibited by downward/upward motion (Krapp et al., 2001)(Fig. 2). V1 can be easily distinguished from the other spiking LPTCs sensitive to vertical motion in the equatorial visual field, i.e. V2 and Vx. V2 is excited by pattern motion in the opposite direction (upwards), and Vx in this area responds, but is much less sensitive to, vertical downward motion(Krapp, 1995)

R(t) is referred to in the rest of the text as instantaneous firing rate. The FWHM of our smoothing function was always set as 5 ms. Errors on averaged data are always given as standard error of the mean (±1 s.e.m.) and are shown as white or shaded areas around the averaged data. Fits to the data in the last figure were obtained using the Matlab Data Fitting Toolbox, using a smoothing spline.

Once a stable extracellular recording was established in V1, we presented stimuli to the lateral ocelli that mimicked roll rotations of the head. V1 responded consistently to step changes in illumination of the left and right lateral ocelli. Transient excitation of V1 was seen subsequent to a switch in illumination from the right to left ocellus - a fictive clockwise roll. An extended period of inhibition of V1 followed a switch from the left to right ocellus - a fictive anticlockwise roll(Fig. 3A,B).

To confirm that this effect is ocellar-mediated, we performed two control experiments. It was possible, though unlikely, that the unshielded compound eyes could see LED light reflected from surrounding objects, and that this was changing the spike rate in V1. We eliminated this possibility by moving the light guides from their positions above the lateral ocelli to a nearby piece of opaque cuticle, and then repeating the experiment. This extinguished the response of V1 to the LED stimuli. We were also able to discount the other possible explanation for the effect, i.e. that light was transmitted through the ocelli, the transparent tissue of the brain, and into the compound eye photoreceptors. We left the light guides in their positions above the lateral ocelli, and then cauterised the ocellar nerve by passing current through a small wire loop (see Materials and methods). The response of V1 to changes in illumination of the lateral ocelli was again extinguished(Fig. 3C). These controls demonstrate that V1 is driven by the ocelli.

Finally, to make sure that V1 had not been damaged during cauterisation, we tested the cell's response to directional motion stimuli displayed to the compound eyes before and after cauterisation. We found that the response of V1 to a moving grating (Fig. 3E,time course of grating contrast) after cauterisation(Fig. 3F) was very similar to the response before (Fig. 3D);the datasets prior to and after cauterisation lie within each other's error bounds (cf. coloured lines in Fig. 3D,F). These results show that cauterising the ocellar nerve did not impair the spiking activity of V1.

Using the same stimulus procedure as described above(Fig. 3), we repetitively stimulated the ocelli with trains of step changes in light intensity, and calculated instantaneous firing rate (see Materials and methods) and spike time histograms. Repetitions of the stimulus trains were separated by 2.5 s intervals. In these experiments, V1 displayed marked phase-locking of spikes,along with clear periods of complete inhibition(Fig. 4A,B). The highest temporal coherence was only seen for the first spike following excitatory stimuli, hence the vertical lines seen in the spike raster plot(Fig. 4B). At the stimulus frequency shown here (10 Hz) the temporal coherence of the phase-locked spikes slightly decayed with each stimulus cycle (cf. Fig. 4A, peaks of the grey top trace). The standard deviation of the spike position, or `spike jitter', for this cell was 0.3 ms for the 1st cycle, decaying to 3 ms for the 15th cycle. The mean for the 1st and 15th stimulus cycle of 10 animals was ∼4 ms and∼11 ms, respectively. The decay in temporal coherence appears to halt after ∼500 ms of stimulation (Fig. 4A).

Ocellar L-neurones display rapid, transient changes in membrane potential in response to stimulation of the ocelli(Simmons et al., 1994), and this should be reflected in the responses of downstream neurones. We recorded the activity of V1 in response to light intensity steps and to sinusoidal changes in light intensity, for the purpose of qualitative comparison to L-neurone responses. As before, stimuli were applied in near darkness.

For excitatory stimuli, both step (Fig. 5B) and sinusoidal (Fig. 5D) excitatory stimuli elicited the characteristic phasic-tonic response shape of L-neurone responses to changes in light intensity(Fig. 5A,C). For inhibitory stimuli, any phasic inhibition of V1 was masked by the low spontaneous spike rate of the cell. The peak instantaneous firing rate during the sinusoidal stimulus (Fig. 5C) was considerably smaller than that elicited by the step stimulus. This observation suggests that, as in ocellar L-neurones, the peak response was related to the rate of change of light intensity at the ocelli(Simmons et al., 1994), which in turn was closely related to the velocity of head rotation.

All the characteristics of the results shown in Figs 4 and 5, such as phase-locked spiking and phasic-tonic responses were reproduced in nine more animals(Fig. 6A). In detail, however,we found considerable variability between individual responses. The amplitude and latency of the peaks (Fig. 6A) varied from 21-134 spikes s^-1 and 10.5-29 ms,respectively. Such high inter-individual response variability is well documented for the V1 cell (e.g. Karmeier et al., 2003). The latency of the first spike, given by the time difference between stimulus onset and the peak in the instantaneous firing rate, had a median of ∼16 ms. In two recordings, however, the latencies were markedly increased to ∼28 ms. The V1 cells studied in these two experiments also differed from the other V1 cells in that they showed particularly low spontaneous activities. On a graph of latency versusspontaneous rate (Fig. 6C)these two points appear at the top left. This graph suggests that the latency of ocellar drive to V1 is in some way dependent on the present activity of the cell. To assess the validity of this, we measured the peak latency again,except this time the rate of V1 was raised by means of compound eye stimulation (moving grating; see Fig. 3 and Materials and methods) during ocellar stimulation. By raising the spike rate of V1 to >250 spikes s^-1 in five cells,we found that the peak latency was reduced by an average of 4 ms, from 16 ms to 12 ms. These reduced latencies are shown in Fig. 6C as blue crosses.

where R_max is the excitatory peak of the instantaneous firing rate and R_min the inhibitory trough(Fig. 7A).

The relationship between R_mod and the rate of change of stimulus intensity is shown in Fig. 7B. We normalised the data by assuming that the largest response was generated using a step stimulus, and scaling the data between this and the zero response value of R_mod. The zero response point was set by calculating R_mod for data recorded where no stimulus was presented. The response rises in proportion to the rate of change of intensity in the range 0-50 I_max s^-1, and tends to saturate for higher values. The observation that a steeper slope elicits a stronger response modulation is to be expected, since a steeper slope implies a faster rotation of the head.

How does V1 respond to excitation by both the ocelli and compound eyes, and is the ocellar input robust enough to be detected in the presence of powerful inputs from the compound eye? By stimulating the lateral ocelli while presenting a moving grating to the compound eye, we studied the time course and amplitude of the ocellar-mediated response over the full range of activity levels in V1. As before, the ocellar stimulus consisted of a train of step changes in intensity designed to mimic successive clockwise and anticlockwise roll rotations (cf. Fig. 8B,stimulus trace). For compound eye stimulation we presented a moving sinusoidal grating, whose contrast was varied to change the spike rate of V1(Fig. 8C). The combination of ocellar and compound eye stimuli elicited the responses shown in Fig. 8A. The small peaks of ocellarmediated responses are visible even at the highest spiking rates of V1(∼300 spikes s^-1). The response to ocellar stimuli is therefore robust enough to be observed at all levels of V1's response range. The spike raster plots (Fig. 8D) show that the temporal coherence of ocellar-induced spikes was increased by raising V1's spike rate, and as mentioned above, is accompanied by a reduction in response latency. In the experiment shown in Fig. 8 the peak-response latency to the ocellar stimuli was reduced from approximately 18 ms to 13 ms.

Fig. 8E,F shows, for two cells, the relationship between the absolute changes in spike rate due to ocellar stimuli and the spiking activity induced by the moving grating. This was found by subtracting the compound eye-only response from the combined stimuli response, at the time of each ocellarmediated excitatory peak. To highlight a particular feature of these data, we applied a set of smoothing spline fits (black lines: Fig. 8E,F, see Materials and methods). These show that the maximum ocellar-induced excitation and inhibition occur at similar spike rates, ∼70 spikes s^-1, and the magnitude of the ocellar response tails-off slowly as the spike rate is further increased.

We have shown that the lobula plate tangential cell V1 responds to signals originating in the ocellar visual system. Three properties of the response to ocellar signals strongly suggest that the connection is synaptic, rather than neuromodulatory: (i) the short latency of post-stimulus spiking in V1(Fig. 5), (ii) the high degree of temporal coherence of the first post-stimulus spike(Fig. 4), and (iii) the observation that the ocellar-mediated latency is reduced when V1's spike rate is increased as a result of compound eye input (Figs 6, 8). The response to ocellar input is dependent on the apparent direction of the stimulus; V1 was excited by changes in ocellar illumination that mimicked clockwise roll, and inhibited by changes that mimicked rotation in the opposite direction. Thus the ocellar component of V1's response appears to be tuned to rotation, similar to that shown for the compound eye component(Karmeier et al., 2003). The neurone's response increases with the rate at which light intensity changes at the ocelli (Fig. 7), implying that V1 might be able to code the angular velocity of the rotations sensed by the ocelli.

Since the primary objective was to test whether the ocelli have any impact on LPTC activity, we did not set up our ocellar stimuli for an exhaustive analysis. We only stimulated the lateral ocelli and this limited the rotations mimicked by our stimuli to the anterior-posterior axis. Under normal conditions during a rotation about this axis, the frontal ocellus would also be constantly illuminated. Nonetheless, the qualitative demonstration of the interactions between the ocellar and compound eye visual systems we present here is convincing. Our aim in future experiments is to stimulate all three ocelli and quantify the integration of more realistic inputs from the two visual systems.

Though the visual system of the fly is among the best studied of all insect model systems, there is little previous evidence for the integration of compound eye and ocellar-mediated information in the optic lobes. Milde recorded the responses to ocellar and visual stimuli of several types of neurones in the protocerebrum of the honeybee(Milde, 1986; Milde, 1988). Both third-order ocellar neurones and interneurones originating in the lobula or medulla of the bee were found to respond to ocellar and compound eye stimulation, but in a somewhat erratic way. Consequently, the responses of these neurones were not fully characterised. In the locust, Simmons found that the descending contralateral movement detector (DCMD), a well-characterised visual interneurone thought to be involved in looming detection, responds to changes in light intensity at the median ocellus(Simmons, 1981). Simmons concluded that the function of ocellar input to the DCMD is to transiently boost the response of the cell to compound eye input when ocellar light levels suddenly decrease.

The situation in the blowfly lobula plate is more clear-cut. We already know that V1 is tuned to respond to the optic flow that is generated in the compound eye by rotation about a specific horizontal axis(Krapp et al., 2001), and we have discovered here that it also responds to ocellar stimuli that mimic similar rotations. We can now build on the knowledge that V1 encodes visual information in a simple, explicable way to investigate two related sets of problems. We can investigate ocellar function by monitoring V1 activity induced by the ocelli, and we can study multisensory integration by measuring how this identified neurone integrates rotation-related signals from two different sets of sensors.

How important are the ocelli for gaze and flight stabilisation? Arguably the most complete picture so far is evident in locusts. Locust L-neurones were shown to have a fairly direct route, via descending neurones, to the flight motor system (Simmons,2002) and are demonstrably important for flight steering(Taylor, 1981a; Taylor, 1981b). In dragonflies, there is strong behavioural evidence that similar pathways exist between ocellar neurones and neck muscles(Stange, 1981; Stange and Howard, 1979). The morphology of third-order ocellar neurones in the cockroach is similar to those in locusts, suggesting an analogous function(Mizunami, 1995a; Mizunami, 1995b).

In the blowfly, behavioural experiments investigating the phasic dorsal light response, which is mediated by the ocelli, showed only small amplitude compensatory head roll movements (Schuppe and Hengstenberg, 1993) in response to differential illumination of the lateral ocelli. The light level in these experiments, however, was only changed in one ocellus, while the other two were occluded. The same experimental paradigm was used in experiments to obtain the angular velocity dependence of compensatory head roll movements(Hengstenberg, 1988) for compound eye-mediated reflexes. It was found that the response to directional motion mediated by the compound eye was maximal at ∼70°s^-1, but that the dorsal light response, which also involves the ocelli, extended with high gain up to angular velocities of ∼500°s^-1. The conclusion of this behavioural work was that the ocelli act in conjunction with the compound eyes and mechanosensory halteres to detect and monitor self-rotations. The sensitivity of the ocellar system to changes in attitude during flight lies within an angular velocity range in between that of the compound eyes and the mechanosensory halteres(Hengstenberg, 1993). Our recordings from V1 show that the LPTCs could play an important role in combining information from ocelli and compound eyes, supplying descending neurones and motor neurones with appropriate signals for attitude control.

It is likely that a fast neural circuit connects the ocelli to V1, because the short response delays (10-20 ms) seen in our experiments exclude neuromodulatory effects. Because of the limitations of extracellular recordings, we were not able to identify synaptic inputs to V1, but there is circumstantial evidence for an interesting pattern of neuronal connectivity. During the course of the ocellar-V1 experiments, several recordings were also obtained from V2, another heterolateral LPTC(Hausen, 1993). Like V1, this cell also codes for selfrotation about a horizontal axis but, unlike V1, it did not respond to stimulation of the lateral ocelli. This difference may be explained by examining the function and morphology of the two cells. V1 and V2 are similar in that both cells send an axon from their dendritic input regions to the lobula plate on the opposite side of the brain, where telodendritic output arborisations are formed (Hausen,1993; Krapp et al.,2001; Kurtz et al.,2001; Warzecha et al.,2003). V2, however, most likely receives direct retinotopic input from the compound eyes (Krapp,1995), whereas V1 integrates the outputs of another set of LPTCs,the VS cells (primarily VS1-3) (Kurtz et al., 2001). It is possible then, that the reason we see ocellar-mediated activity changes in V1 but not in V2, is that the ocelli drive VS cells 1-3, which are presynaptic to V1. The connection between the ocellar system and the VS cells could be either directly from ocellar L-neurones, or via descending neurones, which, through electrical synapses, influence the membrane potential of VS1-VS3(Strausfeld and Bassemir,1985). Recent work on the VS-V1 synapse would seem to support the L-neurone-VS-V1 route. The relationship between presynaptic (VS-cell) membrane potential and postsynaptic (V1) spike rate is almost linear(Kurtz et al., 2001; Warzecha et al., 2003), but only above a certain level of pre-synaptic depolarisation. This fits with the observation (Fig. 8B,C) that the largest responses to ocellar inputs were observed when the activity level of V1 was raised slightly above baseline by compound eye stimulation. In any case, to identify the pathway mediating ocellar-induced activity changes in V1 further experiments are required, ideally monitoring the ocellar nerve activity while recording intracellularly and extracellularly from VS cells and V1, respectively.

Because the VS tangential cells are matched filters for selfrotation, they are ideal targets for inputs from ocellar L-neurones. By integrating the appropriate combination of ocellar L-neurone outputs, each tangential cell could include ocellar-mediated information on rotations in a way that matches its preferred rotation axis. Our data suggest that this may be the case for the V1 cell, and thus for the inputting VS-cells, because the ocellarmediated input to V1 conveys information about turns that is compatible with the neurone's preferred axis of rotation.

The compound eye stimulus that most strongly activates V1 corresponds to the optic flow generated by a rotation of the fly about a horizontal axis intermediate between pitch and roll (Fig. 2). For the roll component this is a clockwise rotation that decreases the light level in the right ocellus and increases it in the left ocellus. Mimicking such an illumination pattern increased the V1 spiking activity whereas the opposite pattern of illumination resulted in a reduced spike rate. The integration of ocellar inputs at the level of LPTCs could very well be advantageous in the context of sensorimotor transformation: it reduces the need for ocellar-specific interneurones to transform information obtained in sensory coordinates into the signals required by the motor system for attitude control. A similar simplification of the sensorimotor transformation has recently been suggested in the gaze stabilization system of the blowfly(Huston, 2005; S. Huston and H.G.K., unpublished data).

Any functional integration of ocellar information at the level of the LPTCs would be meaningless if a major advantage of the ocellar system - its speed -was negated. Our measurements of the latency of ocellar-mediated V1 modulations suggest that this advantage is maintained. We observed a median(over ten cells) time-to-peak latency of ∼16 ms, with a minimum latency in one cell of 10.5 ms, whereas the response latency in tangential neurones to compound eye stimulation is about 25 ms(Warzecha and Egelhaaf, 2000). Our estimates of the latency will be slightly conservative in comparison with the study mentioned in the previous sentence. This is because we looked at peak latency, which is essentially the mean position of the first post-stimulus spike, whereas Warzecha and Egelhaaf(Warzecha and Egelhaaf, 2000)based their estimates around discrimination of the neuronal signal from background noise. The precise latency advantage of ocellar responses must be confirmed using a single stimulus that drives both the ocelli and the compound eye simultaneously with equivalent light intensities.

The arguments presented above suggest how ocellar signals could be integrated in the lobula plate to produce faster motor responses, by using the VS cells as ready-tuned conduits to the motor system. There is preliminary evidence that neurones in the fly lobula plate integrate other forms of sensory input that are useful for controlling flight. Spiking in LPTCs can be modulated by wind stimuli applied to the antennae, and by mechanical stimulation of the abdomen (S. L. Maddess and S. Huston, unpublished observations). Further investigation of these interactions, along with a quantitative description of the integration of ocellar and compound eye-mediated signals in V1 and VS neurones, promise considerable insight into the neural principles of multisensory integration. Because the visual response properties of LPTCs are exceptionally well characterised, such studies may be able to identify some of the synergies that occur when different sensory systems independently encode the same parameter to control a specific behaviour and also show how these synergies are implemented in circuits composed of identified neurones.

This work was supported by a PhD studentship from the BBSRC to M.M.P. and a USAF grant to H.G.K.