Ocellar Input to the Flight Motor System of the Locust: Structure and Function

Hesse (1908) and Wilson (1978) proposed that one of the functions of the ocelli of flying insects is to mediate stability around the roll and pitch axes in flight. The retina of the ocellus does not receive a focussed image (Parry, 1947 ; Cornwell, 1955 ; Wilson, 1978) and cannot ‘see’ the horizon as such. If, however, as the consequence of roll or pitch, the centre of the field of a lateral or median ocellus dips below the horizon, the ocellus receives less total illumination; conversely, if the ocellus rotates skywards it receives more light. There are optical and neural reasons why such a system would be advantageous for flight stabilization (Wilson, 1978; Taylor, 1981a; C. P. Taylor & C. H. F. Rowell in preparation) relative to the information obtainable from the compound eye.

This hypothesis of ocellar function has been investigated experimentally by two groups of workers. Stange & Howard (1979) and Stange (1981) showed that when the lateral ocelli of dragonflies (Odonata, Anisoptera) are differentially illuminated, a tethered flying insect will roll its head so as to equalize the illumination to them. It is known in dragonflies and locusts that mismatch between the position of head and prothorax is detected by proprioceptors, and in flies that flight torque is then altered so as to realign the thorax and head (L. Goodman, 1959, 1965; Liske, 1977; Mittel-staedt, 1950) ; hence, head rolling is part of a flight correction system, and the occurrence of this response supports the Hesse/Wilson hypothesis. Taylor (198l a,b) investigated the response of locusts (Orthoptera, Acrididae), hoverflies (Diptera, Syr-phidae), and dragonflies (Odonata, Zygoptera) to simulated roll using an artificial horizon in a wind-tunnel. He too showed compensatory head rolling, even in nonflying insects. In flight, head rolling increased in amplitude, and was supplemented by changes in wing-beat pattern, rudderlike motions of the abdomen and legs, and flexion of the pterothoracic/prothoracic articulation. The changes in wing-beat pattern in locusts were reflected in positive and negative phase shifts in the timing of action potentials in the first basalar and first tergosternal flight muscles (the only two investigated) ; changes in the time of firing of these two muscles changes the down stroke of the wing (Möhl & Zamack, 1977) and would be expected to cause correcting torque around the roll axis.

Taylor demonstrated that either the compound eyes or the ocelli alone could mediate this behaviour. Steering responses to ocellar stimulation persisted at reduced amplitude when proprioceptive information was eliminated surgically, or when the head was prevented from rolling. This result indicated that the ocellar information is transmitted directly to the thoracic flight motor centres, as opposed to merely influencing them indirectly via the head-turning response. Taylor’s work establishes the Hesse/Wilson hypothesis; it does not, of course, exclude other further functions for ocellar input.

The anatomy of the ocellar pathway and the flight motor system has been best studied in the locust Schistocerca. The retinula cells synapse with 17 large second order neurones (L cells, C. Goodman, 1976) and are connected to 61 pairs of smaller neurones of unknown function (C. Goodman & Williams, 1976). The L cells interact with each other (L. Goodman, Mobbs & Guy, 1977; Taylor, 1981c; P. Simmons, 1982) and at least some synapse in the posterior protocerebrum with a number of interneurones which descend to the thoracic ganglia (Williams, 1975; C. Goodman, 1976; Guy, L. Goodman & Mobbs, 1977; Simmons, 1981). It has long been known (e.g. Parry, 1947) that action potentials can be recorded from the thoracic nerve cord in response to illumination changes at the ocelli (review, L. Goodman, 1970), and a number of the descending ocellar neurones (DONs) have been partially characterized, either anatomically (above) or physiologically (e.g. Guy et al. 1977). One DON (O₃ of Williams, 1975) has been studied in detail (Simmons, 1980) and shown to make direct synaptic connections with two of the flight motor neurones of both pterothoracic ganglia.

In this paper we report on studies of DONs and other neurones of the pterothoracic ganglia which receive ocellar information, as a further step towards elucidating the circuitry responsible for the flight correction behaviour described by Taylor. We have confined our attention to the flight motor system and have ignored the elements responsible for movements of the head, and for the rudder-like use of abdomen and legs (Gettrup & Wilson, 1964; Camhi, 1970; Taylor, 1981a). We have concentrated on the mesothoracic ganglion, as the forewings appear to play the larger part in flight correction (Wilson, 1963). We find that DONs synapse monosynaptically with most or all of the flight motor neurones, and also with a variety of previously undescribed interneurones (both intra- and interganglionic), of which at least some are also active during flight, and synapse with flight motor neurones.

Experiments were performed on Schistocerca gregaria (Forskå l) and Schistocerca americana (Drury) of either sex. Both came from crowded laboratory culture. The culture of gregaria (at the Department of Zoology, University of British Columbia) has been inbred for many generations; the culture of americana (Department of Zoology, University of California at Berkeley) was derived from insects caught in the wild in Texas in the summer of 1979, and at the time of our experiments was in only its third captive generation. The results reported were obtained from 207 penetrations of units responding to ocellar stimulation in 46 animals. A larger number of animals yielded extracellular data.

Insects were restrained by the extended legs with pins and Plasticene. The head was stabilized by two pins in the genae. The thoracic pleura were pinned out horizontally after cutting through the terga in the midline. The inner layer of flight muscles was transected dorsally so as to expose the underlying layer; in this way all the flight muscles could be seen and checked for contraction in response to stimulation of motor neurones. The gut and the muscles attaching to the mesothoracic spina were removed and the mesothoracic sternal apophyses cut out, exposing the meso- and metathoracic ganglia and their nerve roots. The ganglia were stabilized by a steel platform which also served as electrical ground. The tracheae running from the ganglia to the longitudinal tracheae were left intact, and the preparation was continuously perfused with saline at room temperature (24°C). A 10 mm diameter tube was positioned close to the insect so that air could be blown frontally over the head. A slightly modified procedure (Robertson & Pearson, 1982) allows the insect to produce the motor output corresponding to flight in response to air on the head (‘simulated flight’).

The anterior border of each compound eye was painted with several coats of thick red nail-varnish, to prevent accidental stimulation of the compound eye with stray light from the light pipes which were used to stimulate the ocelli. In most preparations the middle and posterior parts of the eye remained uncovered; they saw an invariant scene of intensity 1–10cd/m². In some preparations the antennae were cut short, to prevent them from moving the light pipes. Plastic light pipes (Dupont ‘Crofon’) 50 mm long, of a diameter slightly less than that of the ocellus lens, were placed in contact with each of the three ocelli. Each light pipe was cemented to a green lightemittingdiode (Hewlett-Packard 5082-4190, peak emission 565 nm). The L cells, and presumably therefore the ocellar retinula cells of Schistocerca are maximally sensitive in the u.v. (λmax = 366 nm) and have a broad secondary peak in the yellow green (λmax = 516nm) (Wilson, 1978). A current of 24mA in the LEDs was used to provide background illumination and allowed a dynamic stimulus range of ± 1 log₁₀ unit, over which light emission was linearly related to current. Stimuli used were either 100 ms, 1 log₁₀ unit square-wave pulses of increased illumination (ON stimuli) or decreased illumination (OFF stimuli) relative to normal intensity, or a sinusoidal modulation (2 log₁₀ units peak-peak) about background intensity at 1–100 Hz. Stimuli could be applied to any combination of ocelli in any phase relation. Stimuli affecting only the median ocellus approximate to the input the flying animal would experience as the result of movement around the pitch axis; stimuli which affect both lateral ocelli in antiphase (or a single lateral ocellus, see Results), correspond to the effect of movement around the roll axis. The waveforms which produced these ocellar stimuli were derived from a triggered generator. Stimuli were monitored as the voltage output of a miniature silicon photocell which was cemented to the rear face of each LED. Over the range of intensities used, this voltage varied linearly with LED light output. The complex of lightguide, LED and photocell was given several coats of opaque varnish after cementing together, so that stimulating light came only from the end of the light pipe.

Activity in the pro-mesothoracic connectives was recorded conventionally with silver hook electrodes and AC amplification. Some animals showed little extracellular response in the connective to ocellar stimulation, and experience showed that in such animals little ocellar response could be found intracellularly in the thorax. Subsequently, only animals with a good extracellular response were accepted for experimentation. The experimental population was thus a selected one. Intracellular records were made from the neuropil of the mesothoracic ganglion with microelectrodes with tip resistances of 70–100 MΩ when filled with 3M-potassium acetate. The tips of the electrodes were routinely filled with 3 % Lucifer yellow and the shank with 3M-lithium chloride ; they had high initial impedances but were usually broken to around 100 MΩ when penetrating the sheath of the ganglion. Activity was recorded either directly or via an on-line digital signal averager. When the average as used, between 10 and 50 sweeps were usually averaged; occasional exceptions are noted in the figure legends. After recording, penetrated neurones were filled with Lucifer Yellow using a direct current of 1–10 nA for up to 40 min. The ganglion was then dissected out, fixed for 30 min in buffered (pH 7·2) 4% paraformaldehyde solution, dehydrated rapidly in ethyl alcohol, cleared in methyl salicylate, and examined by fluorescence microscopy. Drawings and photographs were completed within 2h of filling the cell. This procedure produced reliable fills of structures in both the meso- and metathoracic ganglia, but the dye did not diffuse further.

We have filled and characterized physiologically five descending units; these are summarized in Table 1. None of our fills extended into the brain, so we cannot yet homologize our units with the nomenclature of Williams (1975) which is based on cerebral morphology. O₃ and O₄, described by Simmons (1980, 1981) were not found by us (see Discussion), but represent two further known DONs. There are certainly a number of other DONs which we have seen in extracellular recordings which we have not characterized intracellularly. We estimate a total of at least 10 DONs in each connective. As we never filled more than one DON per animal, it is not excluded that some units we describe as different entities are in fact variants of the same neurone, but we do not think this to be so. Extracellular recordings under favourable conditions clearly show that a considerable number of units are present simultaneously.

Dye fills of DONs invariably showed an axon running through the mesothoracic ganglion and usually into the metathoracic ganglion, but with no soma in these ganglia. Other workers have found DON somata in the brain, and the close correspondence in physiology and thoracic anatomy between these and our units suggests that the latter too originate in the brain. Usually the axon terminates in the metathorax, but on one occasion was seen to enter the abdominal connectives [a similar pattern has been seen in DCMD (Pearson & C. Goodman, 1979) and O3 (Simmons, 1980)]. The majority of DONs, even when clearly different in their physiology, have similar morphologies (Fig. 1A-E). They have numerous branches medially and laterally within the ipsilateral hemiganglion, especially in the area of the neuropil of the flight motor neurones (FMNs) (Tyrer & Altman, 1974). The medial branches do not cross the midline. Axon and processes allliewithinthe dorsal 100 μm of the ganglion. Two of the six known DONs have different morphologies. The Slowl/M/(C) unit (for terminology see Table 1), which is also exceptional physiologically, fits the above description but terminates in the mesothoracic ganglion. The fast M unit has only a single branch in the mesothoracic ganglion, and this extends across the midline to branch in the contralateral FMN area: there were no branches within the ipsilateral hemiganglion.

Intracellular records of DONs in the ganglia showed action potentials rising direct-from the baseline; no synaptic activity was seen. The AP invariably correlated with Slightly preceding extracellular spike (latency ⩽ 1 ms) in the pro/mesothoracic connective (Fig. 2A, B). We have no evidence suggesting that any of the DONs are too small to be recorded extracellularly, though some gave very small spikes (approx. 50 μV). This is true not only for units we penetrated, but also for those known to exist only on the basis of short-latency postsynaptic potentials in FMNs or thoracic interneurones.

All of the five different DONs we filled gave a spike or burst of spikes at light OFF to one or more ocelli. There was never any response to light ON. Responses became erratic with sinewave modulation of light intensity at 0·2 Hz or slower, but reliability was restored if the stimulus risetime was shortened (e.g. square wave modulation instead of sine wave). No OFF response was ever influenced by a simultaneous ON stimulus to another ocellus. Consequently, the experimental abstraction of light OFF to one lateral ocellus appears to be a valid simplification of the antiphase OFF/ON which actually affects the two lateral ocelli during roll by a flying animal (Fig. 2C). The pattern of ocellar connection varied (Table 1); DONs responded to either one of the lateral ocelli, or to the median ocellus, or to the combination of one lateral and the medial. When two ocelli were effective, usually one was much more so than the other. The second ocellus evoked a spike only with low probability and with a variable latency, presumably indicating a longer synaptic pathway than between the DON and the main activating ocellus (Fig. 2D).

None of our DONs responded to wind on the head or body, or to light stimuli to the compound eyes. This distinguishes them from O₃ and O₄ (Simmons, 1980, 1981). The Fast Ipsi and the Fast Contra units (Table 1, Figs 1A, C; 2A, B) are the most conspicuous ocellar units in extracellular cord records, and can usually be recognized from one preparation to another by their large amplitude, short latency and response to particular ocelli. In extracellular recordings it sometimes seemed as if continuous frontal wind on the head increased the probability of doublets and triplets in the OFF response of these two units, but we were unable to confirm this in intracellular records where the possibility of confusing different units was excluded.

Four of the five recorded DONs gave 1–2–(3) spikes at OFF, and showed no response decrement in response to prolonged stimulation at up to 80 Hz. All these units were ‘fast’ in terms of Table 1 - latency of the spike in the mesothoracic ganglion to light OFF at the ocellus was 20–30 ms. This value was markedly temperature dependent. The remaining DON we filled had quite different properties. This unit (the Slow I/M/(C) unit) responded to either the ipsi or the median ocellus; its reliability was increased when both ipsi and median ocelli were stimulated simultaneously. There was a long latency, low probability response to the contra ocellus too. It gave an initial burst of 5–6 spikes to ipsi or median OFF, immediately decrementing to one spike for the second and subsequent stimuli at an ISI of 0·7 Hz in an unaroused animal. Recovery required 1–2min rest. Responsiveness was restored without a rest interval (dishabituation) by touching either metathoracic tarsus, and decrement to repseated stimuli proceeded less slowly than usual while tarsal stimulus was maintained (Fig. 3). Other mechanical stimuli, including wind on the head, were ineffective in dishabituating the unit.

We have recorded from motor neurones innervating all but two (nos 103, Tergo-trochantal, and 85, Pleuroaxillary) of the mesothoracic flight muscles. In addition, we have a few records from metathoracic flight motor neurones. Simple or compound EPSPs and IPSPs are seen in the FMNs in response to OFF stimuli to the various ocelli. No PSPs are seen after ON stimuli to the ocelli. Often, but not always, the various components of compound PSPs can be correlated with the spikes of successive DONs of differing latency in the cord. Whenever EPSPs have been causally associated with particular spikes recorded in the pro-mesothoracic connectives their latency has been approx. 1 ms. They therefore appear to be monosynaptically derived from DONs. The fastest IPSPs have longer latencies, 6·5–10 ms longer than EPSPs recorded from the same cells. We presume that they are derived from DONs via thoracic interneurones (below). The IPSPs can be reversed by hyperpolarization: on some occasions hyperpolarization reveals two components, one with a reversal potential near resting potential (presumably Cl⁻ mediated). The amplitude of these fast PSPs typically remains stable even when stimulated at high frequency. In no FMN did the EPSP derived from ocellar input ever cause the cell to spike. Many of these characteristics are illustrated in Fig. 4A-D. The FMNs receive a great deal of synaptic input from unknown sources, and the majority of the records in Fig. 4 have been averaged minimize this ‘noise’ signal over 10–50 sweeps.

For an understanding of the effect of ocellar input on flight behaviour, it is important to see what pattern of inputs is produced in the different functional groups of flight muscles (elevators, depressors, and the wing-rotating supinators and pronators) by the visual consequences of roll and pitch. As explained above, roll simplifies down to an OFF signal to the ocellus on the downward side, and downward pitch to OFF to the medial ocellus, as ON signals to the ocelli have no effect at the level of the DONs. (The mechanical operation of the different muscle groups is summarized in the Discussion.) With this background, the nature of the synaptic input to the mesothoracic flight motor neurones in response to OFF signals to the three ocelli is shown in Table 2. For the sake of subsequent discussion, the data of Simmons (1980) are also included. The results can be summarized as follows.

• All recorded elevator MNs receive a rather reliable EPSP in response to an OFF to the contralateral ocellus. An OFF to the ipsilateral ocellus produces usually no effect, sometimes an IPSP or a very small EPSP. Consequently, the flight muscles contralateral to the stimulated ocellus receive nett excitation relative to their (ipsilaters homologues. Almost all these elevator MNs also receive an EPSP in response to OFF to the medial ocellus.

• The simple depressor MNs (nos 81 and 98) respond to stimulation of the lateral ocelli in the same way as the elevator MNs: that is, the units contralateral to the simulated ocellus receive nett excitation relative to their (ipsilateral) homologes. To an OFF to the medial ocellus, the MN of 98 sometimes receives an IPSP, and Simmons (1980) reports that those of 81 receive an EPSP.

• The pronators and supinators (nos 97 and 99, which also function incidentally as depressors) respond to the lateral ocelli in respectively opposite ways. An OFF to the ipsilateral ocellus causes an EPSP in the MN of 99, and either nothing or an IPSP in 97 : the nett effect will be towards supinating the wing. An OFF to the contraiated ocellus produces no effect in 99, and sometimes an EPSP in 97. Any nett effect will be towards pronating the wing. An OFF to the medial ocellus produces an IPSP in 97, and (according to Simmons, 1980) an EPSP in 99 : this will tend to supinate the wing.

These effects are summarized in Fig. 5, and their functional consequences are treated in the Discussion. In general, the observed pattern is compatible with the hypothesis that any alteration in wing movements produced by the ocellar inputs would tend towards correction of flight deviations.

It should be noted that the connections described above are not always found. In this there is extensive individual variation, even though the experimental population was preselected (see Methods) for animals with prominent extracellular DON discharges. This is especially true for the inputs to MNs from the ipsilateral ocellus; Table 2 shows that the described ipsilateral inputs were actually seen in respectively 1/3, 2/3, 2/7, 1/5, 1/8 and 2/2 penetrations (sum, 9/28). MNs of muscle 98 were penetrated eight times, but on only 3 of a possible 24 occasions was an ocellar PSP seen. Similarly, in our few penetrations of the relevant MNs we did not see the PSPs mediated by O3 described by Simmons (1980) (the PSPs that were recorded from these FMNs (e.g. Fig. 4A) were apparently derived from DONs with different properties from O3). This point is returned to in the Discussion.

Rarely, PSPs with longer latencies (20 ms or so after the fastest EPSPs) were seen (not included in Table 2). These may derive from the slow, rare responses of some DONs to auxiliary ocellar input (Table 1).

In addition to the DONs and FMNs, we have recorded from numerous other units in the mesothorax which receive synaptic input from the DONs. All which have been filled have been interneurones.

All these ‘thoracic ocellar’ neurones (TONs) are effectively planar, lying in the same dorsal sheet of neuropil as the FMNs. Their branching is concentrated in two areas, the lateral flight motor neuropil, and in a longitudinal strip lateral to the ganglion midline, approximately coinciding with the dorsal longitudinal axon tracts. Their morphology is illustrated on Figs 6 and 7,corresponding to two main groups.

(i) Mesothoracic intraganglionic interneurones. All of this category are unilateral only. Typically they extend to two or all of the anterior, intermediate or posterior areas of the lateral flight motor neuropil and also to the lateral edge of the dorsal longitudinal tract, which includes the DONs (Fig. 6).

(ii) Interganglionic interneurones. These fall into three types: unilateral projection in mesothorax, axon projecting to ipsilateral prothoracic ganglion; unilateral projection in mesothorax, axon projecting to ipsilateral metathoracic ganglion; bilateral projection in mesothorax, axon projecting unilaterally to metathoracic ganglion. The cell bodies of all these neurones lie in the mesothorax, and can be ipsilateral or contralateral to the axon running in the connective (Fig. 7). The interganglionic interneurones as a class differ from the intraganglionic in that their ramifications in the vicinity of the DONs can be either or both ipsi- and contralateral relative to the axon. They also have unilateral or bilateral projections to the FMN areas of the mesothoracic ganglion. Within the metathoracic ganglion, those cells that have been tilled all project unilaterally to the FMN area.

We do not know, as yet, how many of these TONs occur in the mesothoracic ganglion, but a total of twenty pairs seems a conservative estimate. We recognize as distinct at least the 14 neurones listed in Table 3. Some have been filled more than once, and then are recognizably of the same morphology and have the same physiological characteristics. Some morphologically similar cells are clearly different from one another in detail or show quite different physiological responses. Some physiologically distinctive cells have not yet been filled.

The physiological responses of the TONs are summarized in Table 3 and Fig. 8.

All penetrations were made in the neuropil, and none were from either cell body or from areas of spike-sustaining membrane, though spikes generated elsewhere in the cell were clearly visible. By definition, all TONs receive PSPs from OFF stimulation of one or more ocelli. EPSPs are normally of short latency (1–3 ms) with respect to the associated DON spike in the cord (e.g. Fig. 8A-C), and appear to derive monosynaptically from DONs. Compound EPSPs are usually derived from more than one DON (of different latencies) converging on the recorded intemeurone (Fig. 8B). IPSPs are 5–20 ms later, and presumably derive from DONs via TONs. Sometimes delayed EPSPs are also seen. EPSPs are much commoner than IPSPs (Fig. 8E), and 4 of the 13 neurones receiving EPSPs are driven to spike reliably by the ocellar input (e.g. Fig. 8A). Decrement in response to repeated stimulation occurs in some but by no means all of the TONs (Fig. 8D). Most of these TONs are not obviously multimodal. Only one (135) received input from cephalic wind-hairs, and one (315) received powerful mechanoreceptive excitatory input from the ipsilateral mesothoracic tarsus.

In the four cases in which the experiment has been possible (Table 3) the TONs have been found to fire bursts of action potentials in phase with either depressor or elevator FMNs during simulated flight (Robertson & Pearson, 1982). One interneurone (302), which discharged in phase with the mesothoracic depressors, was shown (Fig. 9) to inhibit a metathoracic elevator FMN.

Despite specific search, we have so far found no non-spiking interneurones which receive ocellar input.

The present results, with those of Simmons (1980) on O3, corroborate at the single unit level what has been previously suggested about the population of DONs by earlier workers from extracellular recording. All DONs so far recorded by us or by others (e.g. Burtt & Catton, 1954) from the thorax are phasic OFF units. DONs with other properties (e.g. tonic response, or ON units) have been reported from the circumoesophageal connectives (Guy et al. 1977). Taylor’s (1981a) results show that tonic ocellar information is transmitted to the neck motor neurones. If these other DONs stop at the level of the suboesophageal or prothoracic ganglia (from which the neck muscles are driven), it would explain their absence from recordings in the pterothorax.

As both the medial ocellus and the cephalic wind-hairs signal pitch deviations, one might expect considerable convergence of these modalities in the DONs. Perhaps surprisingly, only two DONs to date are known to have wind-hair input (O₃ and O₄, Simmons, 1980, 1981). We found three other DONs with more or less important medial ocellus input, none of which is responsive to wind. Similarly, of the seven TONs found to respond to medial ocellar input, only one also responded to wind-hair input. The other properties of this TON, however, suggest that it is not driven by O₃, as it receives excitation from the contralateral and medial ocelli, whereas O₃ responds to the ipsilateral and medial.

The lateral ocelli signal roll, which leaves the wind-hairs largely unaffected (though they might respond to a rapid transient component, see Bacon & Möhl, 1979). Convergence of these two inputs in a single DON is consequently unexpected, and (except for the minor sensitivity of O₃ to ipsilateral ocellar input) is indeed not found. Another descending wind-hair unit, the tritocerebral commisure giant (TCG) has been subcribed by Bacon & Tyrer (1978). It receives additionally visual input from the compound eye. One would expect the visual response of the TCG to be specific yaw and pitch (the two rotations to which the wind-hairs are sensitive) but this information is not available; so far only ON/OFF stimuli have been tested.

We recorded no DON with the properties described by Simmons (1980) for O₃ and O₄, despite virtually identical stimulus regimes. The question arises as to whether any of our units could have been confused with O₃ and O_4. Both O₃ and O₄ responded strongly to wind on the head, and O₃ (which was characterized in detail) responded principally to the medial ocellus and weakly to the ipsilateral ocellus. Of our units the only one to respond principally to the medial ocellus (the Fast Medial unit, Table 1) had a totally different thoracic morphology. The Fast Ipsi (Medial) unit, with a weak, long latency (twice that of O₃) response to the medial ocellus, showed no response to wind and has a slightly different morphology. The thoracic morphology of what we call the Fast Ipsi unit is almost identical to that of O₃. We filled this unit twice and recorded it extracellularly in many preparations. It never responded to either wind on the hair or the medial ocellus. We conclude that we did not penetrate O₃. This in itself is unremarkable, as there are certainly several more DONs which we recorded extracellularly but failed to penetrate. We also failed to record responses to stimulation of the medial ocellus in the mesothoracic motor neurones postsynaptic to O₃,99 and 81 (Table 2). Probably this simply reflects the variability between individuals seen in the strength of the synaptic inputs to the FMNs in general (see Results), combined with the effects of a very small sample of indubitable 99 and 81 MNs.

The thoracic anatomy of all the known descending neurones in the grasshopper which respond to ocellar and wind-hair inputs (and to combinations of these with other modalities) seems to be extremely similar, as would be expected if they are all doing basically the same job (mediating flight corrections) and having the same thoracic targets. The characteristic unilateral projection of the locust DONs in the pterothorax is similar to that described for analogous units in the bee (Pan & L. Goodman, 1977).

Lift and thrust forces arise during flight because the plane of the wing lies at an angle (the angle of attack) to the plane in which the wing is moved by the elevators and depressors. The angle of attack is controlled by the rotation of the wing about its long axis: rotation decreasing the angle of attack is called pronation, and rotation increasing the angle of attack is called supination. Supination of the wing during upstroke has been thought to be entirely passive, while downstroke supination is controlled by the balance of forces exerted by the first basalar (pronator-depressor) and the subalar (supinator-depressor) (Wilson & Weis-Fogh, 1962). (It has recently been shown, Peau & Nachtigall, 1981, that the pleuroaxillary musclé 85 also influences the angle of attack, acting as a tonic supinator; so far, no theories of flight take account of it, and no measurements of its activity in flight are available. Its inputs from the ocellar system, if any, are also unknown.) The amount of downstroke supination/pronation is critical, and is always modulated during flight corrections. In general, more downstroke supination generates more lift. The remaining motor units are also modulated during flight corrections, but only provide complement changes in power; thus when more lift is generated, there is increased general depressor activity, e.g. recruitment of the normally silent second basalar (Wilson & Weis-Fogh, 1962). Further details are to be found in Wilson, 1961, 1963; Gettrup & Wilson, 1964; Dugard, 1967; Waldron 1967a,b;,Koch, 1967; Zarnack & Möhl, 1977 and Baker, 1979a,b. The FMNs receive from the DONs monsynaptic EPSPs and longer latency IPSPs. The data of Table 2 indicate what synaptic inputs will operate on virtually all the mesothoracic flight MNs in response to a given roll or pitch. This pattern can then be compared with what would be predicted from our current understanding of the mechanics of the system. Simmons (1980) has recently done this for the effects of one DON (O₃) on two FMNs (subalar and dorsal longitudinal). He concluded that downward pitch could generate increased lift (through an EPSP to the supinating subalars) and that upward pitch would cause ON-induced inhibition of the continuous wind-hair response which he assumed to be present in O₃ during flight. Our results extend Simmons’ argument to all of the mesothoracic FMNs and DONs. Table 2 can be represented graphically by Fig. 5. This shows that the pattern of PSPs in the FMNs in response to roll could generate more lift (via greater wing supination) on the DOWN side, less lift (via greater pronation) on the UP side; the increased elevator and depressor activation on the UP side should generate more muscular power, the need for which is not obvious, but it could also cause the reduced wing stroke seen by Taylor via increased antagonism. When the animal pitches downwards the EPSP pattern would tend to supinate both forewings. Simultaneously, increased activation of all depressors and elevators should generate more power; depending on timing, a greater or lesser reduction in wingstroke might also be expected, but it is not clear to us from the literature what is actually observed under these circumstances. The overall result of pitch-down should be increased lift on both sides of the body. Hence, for both roll and pitch-down the ocellar signal would tend to produce correcting behaviour in the flight output, assuming that the subthreshold EPSPs would have some effect upon the spikes produced in the relevant MNs. A pitch-up deviation will, however, produce no OFF response from any DONs in the thorax, and unless these units are simultaneously carrying another signal (as Simmons suggested) they will not be susceptible to inhibition. Simmons’ suggestion is plausible for O3 under his conditions of a steady air stream on the head, but has some problems. The wind-hairs are actually activated phasically at wing beat frequency, not tonically (Horsmann, Heinzel & Wendler, in the press), and the TOG neurone, a typical wind-hair descending neurone which responds tonically under steady state conditions, similarly fires only once per wing beat cycle in real flight (Bacon & Möhl, 1979). O3 might be expected to behave similarly as a wind-hair neurone; as an ocellar neurone, it will probably also fire once per wing beat cycle, due to the 5 ° nodding movement of the head (Taylor, 1981 ; Horsmann, Heinzel & Wendler, in the press). This presents a smaller substrate for inhibition than a continuous discharge. However, the phasic ocellar inputs to the thorax are only a part of a much larger sensory system, and need not be expected to explain all functions. It is quite possible that pitch-up is corrected mainly by e.g. the wind-hair input, and that the phasic ocelli units play only a minor role.

The designation ‘thoracic ocellar neurones’ follows from the fact that the units were recognized from their response to ocellar input, but should be received with some caution. It does not imply that this is the only or even the most important input to these cells, which in several cases are known to have others. Further work on the flight oscillator and its coupling to the MNs is required before the ‘TONs’ can be seen in proper functional perspective. If, as seems likely, they are involved in mediating ocellar flight correction behaviour, they probably participate when this behaviour is driven by other sense organs and voluntary commands.

We know only a few facts about the connectivity of the TONs. (i). TONs are excited or (rarely) inhibited by ocellar input (Table 3; Fig. 9). (ii). Several TONs discharge rhythmically during simulated flight in phase with either elevator or depressor FMNs (Table 3; Fig. 10 of Robertson & Pearson, 1982). (iii). At least one such TON has inhibitory connections with a FMN of the opposite type (Fig. 9), and the anatomical similarity of this cell with others suggests that this will turn out to be general.

If (ii) and (iii) are indeed typical, then, during flight, FMNs will receive regular IPSP volleys between their own bursts. Additional ocellar modulation of the sources of these IPSPs could readily produce the observed phase shifts (Taylor, 1981b) in FMN activity. This model would produce the observed behaviour, but further work is required to establish it.

At least some interneurones must act as ‘inverters’ to provide the delayed IPSPs seen in the FMNs in response to DON spikes. The simple unilateral intraganglionic interneurones appear to have a suitable morphology, but only some of them spike reliably in response to DON input, which the hypothesis requires.

The fast phasic information from the ocelli reaching the thorax goes to the FMNs by at least two routes. First, there is a simple ‘direct’ connection, consisting of monosynaptic EPSPs and slightly delayed (probably disynaptic) IPSPs of the sort figured by Simmons (1980) and in this paper. Secondly, the ocellar signal produces PSPs or spikes in a population of thoracic interneurones, of which an as yet unknown proportion spike rhythmically during flight (Table 2) and make synaptic connection with the motor neurones (as in Fig. 9). Obviously both routes must play some role in behaviour.

It may be premature to evaluate the relative significance of these channels, but some obvious factors need to be spelt out. The ‘direct’ PSPs at all our recording sites are very small (approx. 1 mV) relative to the 15–25 mV oscillation seen in the FMNs during simulated flight (Robertson & Pearson, 1982). In order for them to have any reliable effect on the timing of the FMN spike they would require precise synchronization with the wing beat cycle. It is not excluded that such a synchronization exists, either by neural feedback to the ocellus (Kondo, 1978) or by mechanical feedback via turbulence affecting wind-hairs and via head nodding (see above), but this needs to be demonstrated. We also have the impression that these synapses are unusually variable, in the sense of not being findable in many individuals (see Results) ; although realistic sample sizes do not allow this to be demonstrated with statistical conviction, we think this a real characteristic of the direct ocellar FMN connections. On the other hand, these connections are fast, which seems to be one of the advantages of the phasic ocellar and wind hair neurones over other parts of the stabilization machinery, such as the descending inputs from the compound eyes or those that work via head rotation and proprioception

The ‘indirect’ connection affects the FMNs only during flight, when at least a sizeable minority of the TONs (here assumed to be generally presynaptic to the FMNs) are firing in bursts at the flight frequency. The ocellar PSPs seen in these interneurones are relatively large, they sometimes alone bring the cell to spike or are competent to shut off a previously spiking cell (Fig. 8A, B). These considerations suggest that the interneurones probably play an important role in correction behaviour; possibly the small amplitude, direct connections serve to accelerate the ultimate response in the FMN. An analogous situation seems to be found in the wing stretch receptor. Activity in this organ probably influences flight motor neurone output (though this is much less clear than in the case of the ocelli) and there are subthreshold monosynaptic connections with the FMNs (Burrows, 1975). The stretch receptor, however, also makes powerful connections with thoracic interneurones rhythmically active during flight (R. M. Robertson & K. G. Pearson in preparation) and the relative importance of the two pathways in producing behaviour is not yet clear. Parallel sensory pathways to the motor output are commonly seen in arthropods (e.g. the crayfish tail-flip, the visually-induced jump of the locust) and their relative role in refining the behavioural output is only beginning to be investigated. The flight stabilization circuitry of the locust, with its multiple sensory inputs, each with several routes to the thoracic centres, may present good material for such an investigation.

This work was supported by NSF Grant BNS 78-26785 to C. H. F. Rowell and by a grant from the Canadian Medical Research Council to K. G. Pearson. Part of this work was carried out at the Zoology Department, University of California at Berkeley. We thank Dr J. Bacon for reading and criticizing the manuscript.