The Neuronal Basis of A Sensory Analyser, The Acridid Movement Detector System: IV. The Preference for Small Field Stimuli

In this paper we deal with one of the most striking features of the movement detector (MD) system (Rowell, 1971) - its preference for stimuli which affect only small areas of the retina. Stimuli which affect all or much of the retina simultaneously evoke only very small responses in the interneurones of the MD system ; stimuli of intermediate size, and those which affect large areas of the retina, but not absolutely synchronously, evoke intermediate levels of response (Palka, 1967 a, 1969, 1972; Hom & Rowell, 1968; Rowell, 1971). The most obvious biological result of this arrangement is that the responses of the MD system increases when the size of the moving target is enlarged to a critical dimension but is suppressed by further enlargement. Changes in the retina brought about by the animal’s own movements (Palka, 1967; Rowell & Horn, 1968), or alterations in ambient illumination, therefore, produce little or no response (Palka, 1967a; Rowell & O’Shea, 1976).

The mechanism of this discrimination has been considered by Palka (1967 a, 1969) and by O’Shea & Rowell (1975 a) with slightly different results, which we now summarize. The most obvious way in which an array of similar afferent channels can discriminate against large-field stimuli is by lateral inhibition between the different channels. The hypothesis of lateral inhibition was therefore the first selected for investigation by Palka (1967 a). He rejected it after demonstrating that the inhibitory effect between two stimuli on the retina was independent of the distance separating them. This is contrary to the effect predicted by a classic lateral inhibition circuit, in which the inhibitory effect of a nearby stimulus is much greater than that of a distant one. Intracellular records have subsequently shown (O’Shea & Rowell 1975 a) that this result was misleading. The light sources used by Palka to produce the stimuli emit some non-directional light, and the eye was in the dark adapted state. Adaptation is so efficient in this eye (Rowell & O’Shea, 1976a) that the onset or offset of non-directional light is in effect a large-field stimulus. More recently we found that such stimuli produce inhibition of the MD neurones mediated by IPSPs in the lobula giant movement detector (LGMD) (O’Shea & Rowell, 1975 a; and this paper), which masks any other effect (lateral inhibition, for example). Palka, working with extracellular techniques, correctly deduced the presence of this second type of inhibition, and discarded the lateral inhibition hypothesis.

Our intracellular recordings from the input fan-like arborization (dendritic subfield A, O’Shea & Williams, 1974) of the LGMD revealed that large-field stimuli elicited IPSPs, which appeared to be the basis of the inhibition described by Palka. As, however, the locus of response decrement to a repeated stimulus precedes the formation of the IPSPs (see O’Shea & Rowell, 1975 a; O’Shea & Rowell, 1976), although the IPSP might prevent the LGMD from spiking in response to whole-field stimuli, it cannot explain all the properties of the preference for small-field stimuli. This is because with the IPSP alone, large-field stimuli might not cause spikes butj would continue to excite the decrementing sites and therefore depress the responsiveness of the system to subsequent small-field stimuli. In fact, large-field stimuli do depress responsiveness slightly, but much less than even a few small-field stimuli would do. We therefore reconsidered lateral inhibition and showed (O’Shea & Rowell, 1975 a) that such a network does precede the labile afferent synapses on to the LGMD, and that it is primarily this which discriminates against large-field stimuli.

The available evidence indicates that the lateral inhibitory network seems to be fed by phasic, ON/OFF units and is not activated at all during tonic illumination. This characteristic distinguishes it from the other, more peripheral, lateral inhibitory network which we have already described from this system (Rowell & O’Shea, 1976b). Given that the area is greater than the optimal size for that particular ΔI (Palka, 1967 a), increasing the area progressively inhibits response up to a maximum of between 30 and 40°. Stimuli affecting areas larger than 40°are all rejected maximally. The units responsible for mediating this second, deeper, lateral inhibition have not yet been identified.

The subsequent part of this paper deals with the IPSPs generated by large-field stimuli in the LGMD, and which are the basis of the inhibition studied in detail by Palka (1967 a, 1969, 1972). We believe the occurrence of this inhibition is a necessary consequence of two conditions: the apparent design requirement that the system discriminate totally against large-field stimuli, and the inherent shortcomings of a classical lateral inhibitory network for accomplishing this. The problem is that lateral inhibition discriminates efficiently against sustained large-field stimuli (given appropriate time constants and extent of the inhibitory projections). In its simplest form (Fig. 1 A), however, it cannot discriminate against large-field transients, such as those which occur during a sudden change in ambient illumination, or at the start of movement of the animal. This is because transmission time to the inhibitory synapse is at least equal to that to the excitatory synapse. The situation is made worse where, as is usually the case, lateral inhibition is mediated not by collateral branches of the afferent neurones but by interneurones (Fig. 1B). To block transient excitation it is necessary to employ a feed-forward loop, by which the inhibitory signal can be made to arrive at the summing point simultaneously with, or even slightly before, the excitatory signal. In principle, this feed-forward arrangement could be made part of the lateral inhibitory network (Fig. 1C), but this requires that the inhibitory signal is derived at two synapses prior to the convergence of the retinotopic projection. An alternative arrangement, which we believe to be found in the MD system, is to leave the lateral inhibition network in its simpler form and to add a separate feed-forward loop which starts one synapse prior to the lateral inhibition circuit and causes IPSPs in the LGMD after the point of convergence of the afferent excitatory synapses (Fig. 1D).

In this paper we elucidate the circuit responsible for the IPSPs in the LGMD and show that it conforms to the model shown in Fig. 1D. There are actually two separate loops which produce feed-forward inhibition around the phasic lateral inhibitory network, and so effectively limit the response of the MD system to large-field transient excitation. Both ON or OFF stimuli produce IPSPs in the LGMD, and the channels mediating these two IPSPs are separate. The neurones mediating the OFF IPSPS have.been identified histologically and their role confirmed by selective lesion. They probably synapse with dendritic subfield C (see O’Shea & Williams, 1974) and their input dendritic fields are located on the inner face of the medulla. Together with the postulated phasic lateral inhibitory circuit, these two (ON and OFF) feed-forward inhibitory loops produce the selective discrimination of the MD neurones against both transient and sustained large-field movement stimuli.

The majority of experiments used Schistocerca nitens (Thunberg), from laboratory culture, and a few used S. americana gregaria (Forskå l), obtained from a commercial supplier. Intracellular recordings were made from the LGMD neurone, usually in dendritic subfield A (O’Shea & Williams, 1974), using electrodes filled with 3 M potassium acetate, as previously described (O’Shea & Rowell, 1976). Extracellular records of the DCMD neurone were obtained from the ventral nerve cord with hook electrodes, and the spike potentials discriminated and counted on the basis of amplitude by a variable window discriminator. Experiments were routinely tape recorded for further analysis.

ON and OFF stimuli were generated by a variety of light sources, including: (a) electronic flash (GEC strobelight) and (b) a light emitting diode (Hewlett Packard), emitting predominantly in the green (in both of these cases the change in light intensity was effectively instantaneous) ; (c) a compur-type shutter, actuating time a few milliseconds ; (d) tungsten filament indicator bulbs, controlled by a transistor switch in the supply (the thermal capacity of the filament slowed the rate of change to 20– 25 ms to 50% value on both ON and OFF) and (e) fluorescent room lighting; here the characteristics of the phosphor luminescence gave 18 ms to half light value at OFF, and about 100 ms on ON. All of these stimuli were highly effective, and the variation in rise time proved unimportant, except when precise latency measurements were required. Light intensity was controlled in various ways, including (a) crossed-polaroid filters followed by a diffuser, (b) distance of light source from diffuser, (c) neutral density filters, and (d) variation in filament voltage. The latter, which results in changes of spectral composition, produced the same results as the former, which do not. Changes in light intensity were monitored with a silicon photocell, and measured with a Tektronix digital photometer. In all experiments, the animal was light adapted; the surfaces viewed by the eye were generally of around 30 cd/m² in luminance (zero log units).

Moving targets, moving and stationary stripes, and similar stimuli were presented by back projection on to a white perimeter screen formed by a portion of a tabletennis ball surrounding the eye. The other eye and the ocelli were covered with opaque black wax. Movement of the projected patterns was derived from a pen recorder mechanism, which moved opaque templates in the focal plane of the projector. The size of the projected field could be controlled by a diaphragm. In earlier experiments, the pen recorder mechanism was used to move larger patterns which the animal viewed directly by reflected light. The pen-recorder mechanism was driven by a triggered low-frequency waveform generator. Repetitive stimuli were controlled by a digital timing circuit.

(A) The responses to the first group of ten presentations of the single stripe, at an interstimulus interval of 10 s. (B) The response to the first replicate of this procedure, after a rest interval of 600 s in which there were no visual stimuli. Both curves show typical decremental responses, reading zero response around the 10th presentation. Previous work (e.g. Horn & Rowell, 1968 ; Rowell & Horn, 1968) shows that further stimulus presentations would maintain indefinitely a response level at or close to zero (mean level less than one spike/response), unless the animal’s arousal state changed and the response was dishabituated.

After the second group of stimuli the animal saw a moving stripe pattern continuously for 600 s, and immediately afterwards a second replicate of the small-field stimulus procedure was given. The response to this is shown in (C). The continuously moving stripes provide a very large number of light/dark transitions to the individual retinal cell (calculated to total 9600 over 600 s). If even a significant fraction of these stimuli had excited the labile synapses on the LGMD which mediate habituation in this system (O’Shea & Rowell, 1976), then their responsiveness would have been maintained at the minimal level it reached at the end of the second group of small-field stimuli. The third group would be expected to elicit zero response, as the synapses would be totally fatigued. Instead, (C) shows that significant, though not total, recovery of responsiveness has taken place during the 600 s of large-field stimulation. This shows that little excitation of the labile synapses occurred during this period.

The differences in response to these two categories of stimulus have been extensively described (Palka, 1967a; Horn & Rowell, 1968; Rowell, 1971; O’Shea & Rowell, 1975a, 1976; Rowell & O’Shea, 1976a) and are summarized here. Rapid change (either positive or negative) in illuminance of a small area of the retina produces EPSPs, in the dendritic fan of the LGMD (subfield A of O’Shea & Williams, 1974), and action potentials, both in its axon and that of the descending contralateral movement detector (DCMD) neurone to which it is connected by a spike-transmitting electrical synapse (O’Shea & Rowell, 1975 b) (Fig 2). Movement of small contrasting objects in the visual field produces a succession of such small stimuli, and the response of the neurone is accordingly enhanced and prolonged by summation. The response decreases with repetition, the size of the PSPs wanes and the number of spikes falls eventually to nil (O’Shea & Rowell, 1976). By contrast, changes in illumination affecting simultaneously all of the retina, or large parts of it, produce few or no spikes and movement of stripe patterns which excite all or most of the retina typically cause only a weak response at the onset of the movement. The same visual effect and neurophysiological outcome is produced by voluntary or passive movement of the animal within a structured visual environment (Palka, 1967 b, 1969, 1972; Rowell & Horn, 1968; Rowell, 1971 a).

After exposure to large-field moving stripes for periods of up to 0·5 h, the response of the LGMD to a test movement of a small contrasty target is relatively little diminished (Fig. 3), showing that the labile synapses on to the LGMD which mediate habituation in the system (O’Shea & Rowell, 1976) have been excited little by the moving stripes.

The above observations are consistent with the hypothesis that a lateral inhibition network between the retinotopic afferent pathways to the LGMD precedes the final synapse to the LGMD (see also Introduction). We tested this hypothesis by the following experiment (Fig. 4A). The animal saw in its visual field two horizontal bands of black and white vertical stripes, of period (λ) 15°. Each band subtended 22·5° vertically (i.e. the two together composed one quarter of the entire visual field). The separation between the two bands could be altered from 7·5° to 22·5°. When these bands were oscillated laterally at a speed corresponding to 7 transitions s^{− 1} at the individual receptor (i.e. contrast frequency 7 Hz), little or no spike activity was generated in the LGMD. Altering the separation of the bands had no effect on this result. When, however, a black square was added to the horizontal space between the two bands of stripes, and moved with them, the response was greatly increased. The size of the response increased with increasing separation of the stripe bands (Fig. 4A). This is to be expected from a classical lateral inhibitory network, where the inhibitory effect of activity elsewhere in the array is a function of its distance from the point under consideration.

If the stripes are stationary, and the small black square moved as before, there is no inhibitory effect. This result indicates that the inhibitory effect is derived from a phasic, not a tonic, signal as the stripes only exert their inhibitory effect when they move.

A similar result was obtained from an experiment in which the animals saw moving stripes (λ 20°) through a circular window of varying aperture. If the stripes moved slowly (CF = 6 Hz,) an excitatory response was seen in the LGMD at the onset of movement. This response increased as the diameter of the window decreased (Fig. 4B). Whole-field stimuli gave very little response. The first increase in response came when the window was reduced from 45° to 30° The response thereafter increased progressively down to a window diameter of 5° the smallest value tested. (At higher contrast frequencies other effects mask the response described here. This experiment is described in more detail in Section 6, iv.) This result shows that the responsiveness of an element of the retinal projection is inhibited by the activity of others up to 15– 22°away from it.

Intracellular records from the fan dendrites of the LGMD show only very weak EPSPs during stimulation with slowly moving large-field stripe patterns. By contrast, the response to a transient large-field stimulus involves major and very different intracellular events. Such stimuli include large-field ON and OFF, or movement of large areas of stripes at high contrast frequency. Such stimuli evoke an initial compound EPSP, often leading to a spike and, almost simultaneously a large compound IPSP, which abruptly curtails the excitatory potentials (Fig. 5). The latency of both components decreases with increasing stimulus (either ΔI or area), from around 50– 60 ms (at threshold) to around 25 ms (with powerful, whole-field flash stimuli). The latency of the IPSP is about 41– 42 ms at a stimulus intensity at which the EPSP would have a latency of about 36 ms. The IPSP amplitude is a function of stimulus amplitude (see below) and area. The EPSPs elicited by large-field stimuli will be considered in more detail elsewhere ; the IPSP which follows them ensures that they do not give rise to more than one to very few spikes (two in Fig. 5). The rest of this paper concentrates on the IPSP. The recorded size of the IPSP varies with electrode placement, being smallest when the electrode is most distal in the fan. In normal recordings near the base of the fan, the maximal IPSPs are about 15 mV in amplitude. They will stop a spike train generated by a small moving target (see Fig. 2, O’Shea & Rowell, 1975 a), or by a whole-field stimulus. When they are experimentally abolished the excitatory component of the large-field response generates many spikes (Fig. 12 and Section 7 below).

ON and OFF large-field stimuli both generate IPSPs, but these IPSPs differ from each other. The physiological differences between the ON and OFF IPSPs are as follows.

Except at very low stimulus amplitudes (see below), the ON IPSP is usually larger than the OFF, and its individual component IPSPs are usually less apparent. Basically, however, the two potentials are very similar in appearance. The latencies of both IPSPs vary with stimulus intensity and do not differ from each other significantly. In the ON IPSP, however, the initial hyperpolarization is followed by a long period in which synaptic noise is reduced (Fig. 6A) and spiking can be inhibited (Fig. 6B). This second phase is probably due to an increase in the permeability of the postsynaptic membrane to an ion with an equilibrium potential close to the resting potential (e.g. Cl⁻ ), or it might indicate presynaptic inhibition of the afferents. This second phase is absent from the OFF IPSP.

With increasing intensity (i.e. increasing ΔI, constant area) the OFF IPSP shows a rather sharp threshold, appearing abruptly at almost full amplitude and showing little change thereafter (Fig. 7 A, D). The ON IPSP is first seen at the higher stimulus intensities and rises in a graded manner with increasing stimulus amplitudes, eventually exceeding the OFF IPSP (Fig. 7B, D). Some gradation of response can be seen in the slow component of the ON IPSP. Thus for ΔI greater than about 0· 4 log₁₀ units, the ON IPSP is larger, whereas at lower stimulus intensities the reverse is true. This is best seen by examination of the ratios of paired ON and OFF IPSPS (Fig. 7C).

The larger maximal size of the ON IPSP could be due to (a) current of different ion species, or differing proportions of them, and thus basically a function of differing equilibrium potential; (b) to different conductance of the activated subsynaptic membrane to any one ion; or (c) because the OFF synapse was electrically more remote from the recording electrode. The last explanation is a priori unlikely because the relative size of the potentials is not obviously a function of electrode position, and is excluded by the next result reported below.

Injection of hyperpolarizing current through the microelectrode diminishes and eventually reverses both ON and OFF IPSPS. The slope for the OFF IPSP is linear, while that for the ON shows an inflexion. The amount of current differs for the two IPSPs (Fig. 8), the OFF IPSP reverses first. Once again this difference could be due to underlying differences in equilibrium potential, conductance or in geometry; however, the latter explanation would demand the reverse geometry from that required by the preceding observation (ii), so can be discarded in favour of the other two factors, either or both of which could be operating.

When trains of alternate large-field ON and OFF stimuli are given at various repetition rates, it is found that the two IPSPs differ. At low rates (interstimulus interval (I.S.I.) 2 s) both are stable indefinitely. At higher rates, the ON response decreases and eventually disappears. It recovers rapidly after a few seconds rest. The OFF response resists decrement at higher repetition rates, and shows Significant though slight incrementation (Table 1); it is stable at an interstimulus interval of 200 ms (Fig. 9).

In total, these observations suggest that the ON and OFF IPSP derive from different synaptic connexions with the LGMD and also differ in their afferent pathways from the retina. These interpretations are confirmed by anatomical studies and ablations described below in Section 7.

The response to complex stimuli is basically a summation of responses to the component ON and OFF stimuli which have been described above. We examined four different complex stimuli which could be readily reproduced and quantified.

A brief flash from an electronic xenon discharge tube elicits almost simultaneously ON and OFF responses, which summate. The high intensity and wholefield stimulation make this a maximal stimulus for the system. It produces the shortest latencies, densest initial spike response and the largest hyperpolarization of any stimulus (an example is seen in Fig. 5).

The field can be progressively darkened or lightened by the movement of a shadow edge across it. Slow movement of this edge produces little response of any sort in the LGMD. Faster movement increases the numbers of EPSPs (and sometimes spikes) and generates small IPSPs; and rapid movement approximates to a whole-field ON or OFF stimulus, depending on velocity.

The response to moving stripes is complex, involving not only the summation of large numbers of moving edge responses, but also increment and decrement (see Section 5, iv) due to repetitive stimulation. As would be expected from the above, IPSPs are elicited in proportion to both the speed of movement and the period of the stripe pattern - that is, to its contrast frequency (CF). The visual field of the eye is approximately 180°. When λ ≥ 180° (i.e. no more than one edge in the visual field), the stimulus at high speed of movement approximates to an alternation of whole-field ON and OFF stimuli. Under these conditions, the response is basically a series of IPSPs and EPSPs at the contrast frequency. When λ < 180° (i.e. more than one edge in the visual field at once), an increased contrast frequency increases the number of EPSPs and spikes to be elicited, but with higher contrast frequencies the summating IPSPs form an irregular but continuous envelope of hyperpolarization which suppresses most spikes as long as stripe movement is maintained. After such a stimulus there is a tendency for spikes to be produced by post-inhibitory rebound (Fig. 10).

When the area of the moving stripes is progressively limited, by a window of varying diameter, a number of changes are seen relative to the whole-field response. These depend on either (a) the decreasing area of the retina stimulated, causing a decrease in the IPSPs generated and an increase (by removal of lateral inhibition) of the EPSPs, or (6) geometrical interaction between the window and the stripes. To emphasize the latter, it is necessary to choose a stripe pattern where λ is comparable to the diameter of the window. In the experiments reported here (Fig. 11), the diameter of the window was varied between whole-field and 5°, and the stripes had a periodicity of λ = 20°, with bright and dark stripes of equal width (see Fig. 4B). The interaction of stripe size with window diameter is clearly indicated in Fig. 11A and shows the generation of synchronous IPSPs. The effect of reduction of field size on lateral inhibition is apparent in Fig. 11B ; the small window effectively converts the stripe pattern into a small-field stimulus at high repetition rate, and the neurone responds accordingly. Large EPSPs are generated at the start of the movement but they decrement rapidly. At high contrast frequency, however, even with a 5° field, appreciable IPSPs are generated. The EPSP is very rapidly fatigued and spikes are greatly reduced (Table 2). This dependence of IPSP generation on the temporal frequency of the stimulus suggests temporal summation in the pathway, whereas the dependence on area indicates spatial summation.

The distal portion of the LGMD consists of three separate dendritic subfields, A, B and C (see Fig. 14), which connect to the axon with C being most proximal (closest to the zone of spike initiation, see O’Shea, 1975; O’Shea & Rowell, 1976) and subfield A most distal (O’Shea & Williams, 1974). Spike initiation in the visual, mode takes place where the axon thickens, a considerable distance (∼ 150μm) proximal to the junction of subfield C. Most of our intracellular recordings are made from the basal (proximal) portion of subfield A which forms a fan curving around the outer face of the lobula, and which receives excitatory afferents from the second optic chiasma (see fig. 8 of paper II of the series, O’Shea & Rowell, 1976). In this area the dendritic branches are thick and present a relatively easy target. IPSPs recorded there are of large amplitude (15 mV) and can be reversed by the injection of hyperpolarizing current (see Section 5, iii); we have been unable, however, to demonstrate an increase of permeability during the IPSP by the injection of constant current pulses during the IPSP, which implies that the active sites are some distance from the electrode (or that the synapses are not chemical, which we think improbable). The synapses are unlikely to be more distal than the electrode because placements in the outer branches of the fan dendrites record a reduced amplitude IPSP. One would expect an inhibitory input to be close to the site of spike initiation (i.e. proximally), and we therefore adopted the hypothesis that the ON and OFF IPSPS were generated by synapses on the most proximal subfield, subfield C, or possibly on both C and B. Recordings from electrode placements near the junction of subfield C show large IPSPs and reduced EPSP amplitude, which is consistent with this hypothesis. The results of lesion experiments, reported below, provide direct evidence that synapses responsible for the OFF IPSP are indeed located on subfield C.

Subfield C branches in the dorsal lobe of the lobula, where its extent is larger than that described in our earlier account (Rowell & Bacher, unpubl.). A number of fibre tracts from various parts of the optic lobe run dorsally above the second chiasma; they do not show the lateral reversal of the chiasma projections and are therefore called ‘uncrossed bundles’. One of these arises from the proximal face of the medulla, runs dorsally and then proximally to the dorsal lobe of the lobula, and ramifies in the neuropile there. This dorsal uncrossed bundle (DUB) was selected as a candidate for the inhibitory input to the LGMD.

Lesions which cut the DUB between medulla and lobula produced a wholly atypical response to whole-field stimulation (Fig. 12). OFF stimuli elicited a large burst of spikes in the LGMD, as opposed to the 2– 3 which are the normal maximum. The ON response, however, was unchanged. Intracellular recordings from these animals showed that the OFF IPSP was largely abolished, whereas the ON IPSP was normal.

We attempted unsuccessfully to find a lesion which would selectively abolish the ON IPSP. We could not apply this technique to the chiasma, as lesions there abolish all excitatory visual input to the LGMD, and this in turn makes it impossible to dentify the neurone. We therefore conclude tentatively that the ON inhibitory pathway runs in the chiasma.

The DUB is composed of approximately 500 similar neurones which have a dendritic arborization superficially on the proximal face of the medulla (Fig. 13). There is a differentiation of this dendritic field into two subfields, the smaller of the two projects slightly deeper into the medulla than the other. The dendritic field of each fibre is a vertical ellipsoid which covers approximately 8 × 12° of the visual field, assuming a linear representation of space on the medulla surface. They are stacked Together such that the whole visual field appears to be covered with virtually no overlap. The fibres running in the DUB appear to be of one type. The axons have a diameter of about 2 μ m as they run between medulla and lobula and enter the dorsal lobe of the lobula just anterior to the posterior optic bundle, which runs above the lobula at this point. Silver impregnation (Golgi-Colonnier) shows that each fibre has extensive ramification of its terminals within the dorsal lobe. This region also contains the dendrites of subfield C of the LGMD, with which we assume they make synaptic connexion.

The results presented in this paper support the hypothesis that the synapses between the retinotopic projection of ON/OFF excitatory afferents and the LGMD fan are preceded by a phasic lateral inhibition network between the different afferent channels. There are two biological consequences of this circuitry. First, it provides discrimination against large-field stimuli, and secondly, it acts before the labile synapses and avoids the possibility that they will be appreciably fatigued by ‘unsuitable’ (i.e. whole-field) stimulation. This latter aspect, termed ‘protection from habituation’ by Krasne & Bryan (1973), was examined at more length in an earlier publication (O’Shea & Rowell, 1975 a).

What can be said of the nature of the lateral inhibition circuit, and where is it located? The results show it to be phasic in nature. As it works equally well for ON and OFF stimuli, it must be fed either by phasic ON/OFF units or by separate ON and OFF phasic units. This would seem to eliminate the possibility that inter-cartridge inhibition in the lamina is reponsible (Laughlin, 1974), and although spiking neurones of more complex properties than the large monopolar cells (LMCs) have been recorded in the lamina (Arnett, 1972; Mimura, 1974), they have not been shown to be centripetal and the lamina therefore is an unlikely site for the phasic lateral inhibition described here. The hypothesis that lateral inhibition is fed by phasic ON/OFF units, which is to be preferred on the basis of economy in the number of connexions required, would most likely locate it between the chiasma afferents to the LGMD (models of the input circuitry to the LGMD fan from the chiasma are given by O’Shea & Rowell, 1976). There appear to be no synaptic structures on the second chiasma itself, so the lateral interaction between these afferents must take place either in the proximal medulla or the distal lobula, before the synapses on to the LGMD. The other and more complex hypothesis (i.e. separate ON and OFF inputs) would almost certainly locate the lateral inhibitory connexions in the medulla, where the existence of DUB shows that a retinotopic projection of OFF units, suitable for this purpose, is available near the proximal face. On balance we favour the hypothesis that the lateral inhibitory circuit is situated between ON/OFF chiasma afferents and is probably mediated by amacrine neurones (Strausfeld, 1976; see especially his plates 7.12D and 7.12E) in the most proximal layers of the medulla.

The OFF IPSPS in the LGMD derive from the neurones of the DUB. These units have 8× 12° dendritic fields in the proximal layers of the medulla, where they must receive input from a similar portion of a retinotopic projection of OFF cells. As general properties are very similar for both ON and OFF IPSPS, we assume a similar anatomical substrate exists for the ON IPSPS. Both require either spatial or temporal summation of their input in order to function and this explains the equivalence of area or intensity of visual stimulation in producing inhibition. The DUB neurones are clearly derived from the retinal projection before the latter give rise to ON/OFF units which project to the DCMD. They thus form a feed-forward inhibitory loop around the lateral inhibition circuit, as shown in Fig. 14. It will be noticed that this arrangement agrees with the hypothetical scheme (Fig. 1D) developed in the Introduction as a way in which lateral inhibition can be supplemented by feed-forward inhibition to provide protection against transient whole-field stimuli.

We assume that the IPSPs generated in the LGMD by the two feed-forward loops are normal and chemically mediated. They can be diminished and reversed by injected current, and seem conventional in all ways. The hyperpolarization they generate at peak (at least 15 mV below resting potential) presumably indicates a K⁺ conductance.

In the course of this work we have repeated and substantiated virtually all the experiments reported by Palka in his work on inhibition in this system (1967a, b, 1969, 1972). His first two papers use the DCMD and the same animal as we do. The two later papers, however, use crickets and the DIMD neurone. The correspondence between the two sets of results is therefore of interest.

The neural mechanisms which determine the complex responses of the LGMD may underlie similar responses in other systems. This is because combining many complex response characteristics may set rather specific demands and limits on the possible underlying mechanisms and there may therefore be relatively few neural solutions to complex integrative problems. For example, it is essential in the MD system to protect decrement-prone neurones from the effects of repeated large-field stimulation. The most likely means of achieving this (according to parsimony) is to combine the mechanism of protection with the one which determines the optimal excitatory field size. We showed that this is the dual role of lateral inhibition in the MD system (ibid;O’Shea & Rowell, 1976,a), and suggest that preservation of sensitivity in movement detecting neurones is perhaps a general and hitherto little considered role for lateral inhibition. It is noteworthy, therefore, that a protective role for lateral inhibition has been demonstrated recently in a study on retinal ganglion cells in the cat (Barlow & Levick, 1976). These authors suggest that lateral inhibition may be the nervous system’s tool for controlling decrement in responses of retinal ganglion cells to maintained stimuli.

We have suggested that phasic lateral inhibition in combination with phasic feedforward inhibition underlies the preference in the MD system for small-field stimuli and insensitivity to large-field transients. This phasic inhibitory system is preceded in the locust by a tonic lateral inhibitory network (Rowell & O’Shea, 1976b). We do not know how or where the inhibitory tonic-to-phasic transformation is made but suggest that such a transformation is essential in preventing activation of movement detectors during rapid displacements of the visual field caused by the animal’s own movements.

A very similar organization exists in the vertebrate retina. All retinal ganglion cells are subject to tonic lateral inhibition, and some have a second, change-sensitive, inhibitory surround. Feed-forward inhibition is restricted to the change or movement sensitive retinal ganglion cells and is derived, via amacrine cells, from change in the surround of the ganglion cell’s receptive field (Werblin, 1972). Werblin (1973) has suggested that phasic inhibition provides for detection of movement of small objects while avoiding the overwhelming effect of the vast changes in contrast that result from blinking or eye movements. The parallel with the locust is striking! A major difference in organization does, however, exist.

The LGMD’s receptive field does not have the centre/surround organization of the retinal ganglion cell. Any part of the LGMD’s receptive field can be inhibitory, and change in any small area can be excitatory. Phasic inhibition in the vertebrate retina is fed laterally from bipolar cells and forward to the change-sensitive ganglion cells. In the vertebrates therefore, lateral and feed-forward phasic inhibition occurs at the same time and via the same intemeurones (the amacrine cells of the inner plexiform layer). In the locust we have shown that the need to protect labile sites from decrement means that feed-forward inhibition alone cannot determine the insensitivity of the LGMD to large area stimuli. Phasic lateral and feed-forward inhibition are therefore separated and have different functions in the MD system. It may be that inputs to neurones in vertebrates with characteristics similar to the LGMD are organized in the same way (i.e. arrays of On/Off ganglion cells which are sensitive to change and with phasic lateral inhibitory interactions at a pre-decremental site for protection from habituation and a separate feed-forward inhibition for protection against transient excitation). This organization could be the basis in vertebrates and insects for protecting labile sites and for suppressing responses during visual displacements of the kind which occur in saccadic movements.

This work was supported by grant NS 09404 from the U.S.N.I.H., The British Science Research Council and a travel grant to J.L.D.W. from the Max Planck Gesellschaft. C. H. F. R. was a research Professor of the Miller Institute for Basic Research in the Sciences for one year of the study. Ms Bea Bacher and Fr. H. Bamberg provided histological and photographic assistance. Dr K. G. Pearson, Professor Adrian Horridge and Ms Barbara Shotland critically reviewed the MS. We thank all these persons and institutions for their valued help.