Motor Neurone Coupling in Locust Flight

Locust flight is of interest to neurophysiologists as an example of simple repetitive behaviour easily observable in the laboratory and susceptible to single-unit analysis. Flight aerodynamics (Weis-Fogh, 1956), muscle activity patterns (Wilson, 1961; Wilson & Weis-Fogh, 1962) and sensory regulation (Wilson & Gettrup, 1963; Gettrup & Wilson, 1964; Gettrup, 1966) have been intensively examined. The flight system has about eighty motor units active in a pattern of alternating bursts of action potentials in elevator and depressor muscles (Fig. 1a). The thoracic ganglia may produce the motor output pattern without temporally patterned input (Wilson, 1961). Among many possible models for the flight-pattern generator (e.g. Wilson, 1965) the most economical employ excitatory and inhibitory connexions among the motor neurones themselves to account for synchronous bursts of synergists and elevator-depressor alternation (Fig. 1b; Wilson, 1965).

Motor neurone coupling has been shown to play a part in the co-ordination of activity in the lobster cardiac ganglion (Hagiwara, 1961) and in the crab stomato-gastric ganglion (Maynard & Burke, 1966), both of which govern forms of repetitive activity similar in some respects to locust flight. In these examples evidence for motor neurone coupling was obtained by microelectrode recording from the motor neurones. Application of this technique to the flight system of the locust has not seemed promising because of the difficulty of ascertaining that units encountered in the large and complex thoracic ganglia are in fact concerned with flight.

The present paper describes methods for recording in the locust thoracic ganglia from single units identifiable as the motor neurones of particular flight muscles, presents evidence for several kinds of interactions between them, and discusses the importance of motor neurone interactions in the generation of the patterned output of the locust flight system.

Male locusts (Schistocerca gregaria Forskål) were obtained from the Anti-locust Research Centre, or from a biological supply company. For single-unit recording in the thoracic ganglia the animals were mounted ventral side up, the legs and wings were cut off, and the meso-or the metathoracic ganglion were exposed by removing the corresponding basisternum and the underlying air sacs. Small blocks of agar made by dissolving agar flakes in locust saline (Hoyle, 1953) were wedged around the ganglion to stabilize it mechanically and to keep it moist. The tracheae overlying the ganglion were pushed aside or cut. No special attempts were made to aerate the ganglion; experiments were terminated after a $12$ hr. of recording except in cases in which the preparation was deliberately allowed to deteriorate in order to obtain spontaneous motor neurone activity. The ganglionic connective tissue sheath was carefully opened and peeled back with sharpened watchmaker’s forceps, and a 3 M-KCl-filled glass capillary micropipette, with a tip resistance between 15 and 50 MΩ, was positioned on the surface of the ganglion and advanced with a modified Pfeiffer micromanipulator (Andrew Pfeiffer, Old Lyme, Conn.).

Recorded potentials were led through a capacitance-neutralized pre-amplifier (Amatniek, 1958), displayed on an oscilloscope and photographed or recorded on magnetic tape. Records from some experiments were analysed by a computer of average transients (Technical Measurement Corporation) kindly made available by Dr Paola S. Timiras of the University of California Physiology Department. The membrane potential of an impaled motor neurone could be raised or lowered by passing current through the recording electrode with the aid of a modified Wheatstone bridge (Araki & Otani, 1955).

In flight experiments the locusts were tethered in front of an open-jet wind tunnel. Copper wires, insulated except at the tip, were inserted into particular muscles to serve as stimulating or recording electrodes. In the main locust flight muscles action potentials correspond one-to-one with motor neurone impulses. Stimuli from intramuscular electrodes selectively excite the motor neurone terminals of the muscle in which they are embedded, and set up antidromic impulses in the motor axon. Stimuli were square pulses of 0·2 msec, duration from Grass S4 stimulators equipped with isolation units. Stimulation of an entire nerve or nerve branch was accomplished through bipolar electrodes of platinum or platinum-iridium wire hooked under the nerve. To prevent desiccation and current spread, electrode and nerve were coated with petroleum jelly which was made to flow by touching it with a warm needle.

Particular stimulating and recording arrangements are described with the experiments in which they were used.

Many active units were usually encountered as the electrode was advanced through the ganglion. Motor neurones were identified as such either by the one-to-one relation between their action potentials and the spikes recorded from a particular muscle, or by their extremely short latency (under 1 msec.) all-or-nothing response to antidromic stimulation of a motor nerve. In most instances there was no clear shift in membrane potential to signify that the electrode had actually impaled a neurone, even when recorded impulses were 50−60 mV. in amplitude. The recordings may thus be ‘quasi-intracellular’ (McIlwain & Creutzfeldt, 1967). Units were held for up to 15 min., during which time the amplitudes of the recorded potentials did not diminish. Most units initially fired long trains of injury discharges, which decreased in frequency and eventually ceased.

Most units which were held for several minutes displayed large spikes and relatively small synaptic potentials, indicating that the electrode tip was distant from synaptic sites and probably in a large-diameter axonal region of the neurone. In other units, often more difficult to hold, large synaptic potentials were observed, indicating that the electrode tip was nearer synaptic regions. Some of these recordings may have been from neurone branches in the neuropil rather than from main axons. In no case was there evidence that the electrode tip was in a cell body; most recordings were made with the tip near the middle of the ganglion and rather dorsal, an area of neuropil and large fibres, not somata. Usually the electrode tip was placed close to the known paths of the motor neurones in the ganglia (Wilson, 1965).

In normal flight synergistic motor neurones fire nearly simultaneously. This tendency is even more pronounced when the motor neurones are firing in an uncoordinated fashion spontaneously or in response to electrical stimulation of the ventral cord (Wilson & Waldron, 1967). A short-latency motor neurone interaction is a probable explanation for the extremely close synchrony in the latter case.

In a number of preparations a spontaneously firing flight motor neurone was encountered and identified by the one-to-one relation between its impulses and muscle action potential spikes in one of the flight muscles. An example is shown in Fig. 2. In two preparations, in which several motor neurones were active, each spike in a certain muscle unit was preceded by a small depolarizing potential in the motor neurone of another unit (Fig. 2). The 4 msec, latency between small potential and corresponding muscle spike suggests that the small potential occurred simultaneously with the action potential in the other motor neurone, since there would be a delay of this order of magnitude due to conduction time from ganglion to muscle. The amplitude of the small potential did not appear to change when the membrane of the motor neurone was hyperpolarized. The shape of the potential was approximately triangular, with a fast falling phase. These characteristics are consistent with the suggestion that there is an electrical interaction between the motor neurones.

In the example shown in Fig. 2 the two interacting motor neurones innervate different units of the subalar muscle. During this experiment the larger unit did not fire alone, but always fired nearly simultaneously with the smaller; the spike in the motor neurone of the larger unit was observed to arise from the small depolarization. This small amplitude potential therefore appeared to contribute effectively to the excitatory state of the motor neurone in which it was recorded.

During extended recording in both cases in which this kind of synchronous potential was observed, and in many other experiments in which a motor neurone was encountered, still other motor units were active without an accompanying synchronous potential in the impaled motor neurone. It is possible that only certain flight-muscle motor neurones are coupled by such potentials. If all motor neurones are so coupled, it is possible that the sites of transmission are different and distant (in terms of the membrane space constant) from each other and, in the majority of recordings, from the electrode.

The records discussed above show the effect of a single spike in one motor neurone on another motor neurone. To determine if a number of motor neurones firing simultaneously could produce a different or a greater effect, the following experiment was carried out. The so-called recurrent nerve (the last nerve of the mesothoracic ganglion) contains four of the five motor axons to the dorsal longitudinal muscles, which act as wing depressors (Neville, 1963). There are no other large fibres in this nerve, and only a few much smaller ones. A short distance from the ganglion the nerve joins metathoracic nerve I. Cutting the latter central to the junction, leaving the recurrent nerve intact, transforms it into a functionally pure motor nerve of sufficient length to place on stimulating electrodes (Fig. 3). Both right and left nerves were prepared in this way, and the stimulus intensity was adjusted to a point just above the threshold for stimulation of all eight of the metathoracic dorsal longitudinal motor units innervated by these nerves. The nerves were then cut distal to the electrodes to eliminate the mechanical disturbance of the muscle contraction. A microelectrode was inserted into the ganglion near the exit of the recurrent nerve. When a unit was encountered which responded to a stimulus to the homolateral nerve with a very-short-latency, fast-rising action potential, it was considered to be an anti-dromically stimulated motor neurone. In such units stimulation of the heterolateral nerve singly or in short trains of four or five pulses at 200/sec. elicited a small (2 mV. at most) depolarizing potential with a latency of 5−6 msec, from the stimulus artifact (Fig. 3). There was no obvious facilitation of the potential when a train of stimuli was employed.

Similar experiments were carried out using mesothoracic nerve II, which innervates most of the mesothoracic vertical flight muscles. Electrodes were placed on the left and on the right mesothoracic second nerves, the stimulus intensity was adjusted to just over the threshold for stimulation of the motor neurones to several of the vertical flight muscles, and the nerves were cut distal to the electrodes. The same criterion—a short-latency, fast-rising action potential in response to stimulation of the homolateral nerve—was used to identify antidromically driven second-root motor neurones.

Impaled second-root motor neurones responded to stimulation of the heterolateral second nerve with a delayed potential (latency about 6 msec.) of small amplitude (2 mV. or less). In the case of the second nerve it was impossible to avoid the stimulation of sensory fibres. It is therefore not certain that the delayed potential observed in second-root motor neurones in response to stimulation of the heterolateral second nerve is a response to input from antidromically stimulated heterolateral motor neurones rather than from sensory neurones. There are, however, several reasons for thinking the former more likely: it is not known that there are sensory fibres in the second root which project to the flight system, nor is there any reason to expect them; the stimulus intensity was probably below threshold for most smaller fibres in the nerve and the larger axons are probably motor; and the delayed potential in second-root motor neurones very closely resembles the one which is seen in dorsal longitudinal muscle motor neurones when the heterolateral nerve is stimulated, and which almost certainly is not due to sensory input.

Because the delayed potential is small, variable and easily obscured in a noisy record, a taped record of the responses to 100 single stimuli, spaced 1 sec. apart, was processed by means of a computer of average transients. A record of fifty repetitions of the stimulus artifact alone, made just after the electrode had been withdrawn from the cell, was also averaged, and the appearance of the final trace was compared in the two records to detect the response of the cell (Fig. 4). The response of the motor neurone to stimulation of the heterolateral motor neurones appears to have two peaks. The apparent bimodality of the potential may be its actual shape, or may be a result of the summing process. It is possible that the neurone, which was firing spontaneously at a very low irregular frequency, was sufficiently excited by stimulation of the hetero-lateral nerve to fire an occasional spike in response to the stimulus. A somewhat greater than average number of spikes at a particular latency with respect to the stimulus artifact might appear on the averaged records as a small additional deflexion arising from the delayed potential.

That the delayed potential has an excitatory effect on the motor neurone is shown by records in which the neurone displayed an injury discharge of a regular frequency; the delayed potential summed with the injury depolarization to advance the position of the next spike in the series. A statistical dependence of firing time with a latency of 7 msec, often found between motor neurones in non-flying animals (Wilson & Waldron, 1967) may also be due to the delayed potential.

Models of the locust flight system usually employ inhibition to produce the appropriate phase relations between elevators and depressors. Although many hyperpolarizing potentials were observed in motor neurone recordings, none could be related to activity in other motor neurones, nor did altering the membrane potential of a motor neurone reveal any further sign of responses to other motor neurone activity. Antidromic stimulation of mesothoracic nerve II, which contains both elevator and depressor motor axons, produced no inhibitory potentials in heterolateral nerve II motor neurones.

Trains of single or paired stimuli at normal wingbeat frequency delivered to single motor neurones or to groups of motor neurones did not alter the timing of impulses in other motor neurones, which were firing at a regular frequency, in any way suggesting inhibitory interaction. The following combinations were tried in various experiments: a single elevator was stimulated while a depressor motor neurone was firing; a single depressor motor neurone was stimulated while an elevator muscle unit was active; the recurrent nerve, containing the motor axons of four depressor units, was stimulated during activity in a heterolateral elevator unit; nerve II, which contains axons of both elevator and depressor motor neurones, was stimulated while a heterolateral elevator unit was firing and also while a heterolateral depressor motor unit was firing. In no case did the stimulus appear to delay or prevent the next spike in the train fired by the active unit.

Negative results are particularly inconclusive in this case because of the large number of possible explanations for the apparent absence of inhibition. These will be discussed later.

To determine whether activity in one motor neurone could affect the excitatory state of another over a longer period than single wingbeat cycle, the following experiments were carried out. Stimulating and recording electrodes were inserted in the mesothoracic first basalar, the subalar, and the second tergosternal muscles on both sides and arranged in such a way that each of the six muscles could be separately stimulated and recorded from. The basalar and subalar muscles are vertical depressors; the former pronate, and the latter supinate, the wing. The tergosternals are wing elevators. Stimulating electrodes were placed on the ventral nerve cord anterior to the prothoracic ganglion and stimuli were delivered at high frequency to initiate ‘flight’, or alternatively the preparation was allowed to deteriorate until spontaneous activity began in one of the motor units. When a motor neurone innervating one of the six muscles was active, bursts of antidromic stimuli at 200/sec. were delivered to each of the other five muscles in turn and the effects of the stimulation on the discharge of the active motor neurone noted. The results of these experiments are difficult to quantify because the excitatory state of the active motor neurone was difficult to measure and impossible to control.

Table 1 summarizes the results. Prolonged stimulation of a silent unit could often elicit activity in that unit, activity which continued after the stimulus was turned off. Stimulation of the tergosternal motor neurones inhibited activity in homolateral depressor motor neurones; activity in the subalar motor neurones excited their heterolateral homologues; and stimulation of the first basalar inhibited activity in the homolateral subalar motor neurones and in the heterolateral first basalar motor neurone. In general it appeared that an effect was produced only if the motor neurone was just above or just below firing threshold. A unit firing at high frequency was not slowed or turned off by stimulating another, nor could a unit far below threshold be made to begin firing by stimulating a unit which excited it. The shortest observed latency of response was in a unit which began firing about 100 msec, after the start of stimulation of another unit; thus the effect seems to be a gradually accumulating one.

Since the muscle attachments to the skeleton were intact during these experiments, it is not certain that the observed effects were due to motor neurone interactions rather than to sensory reflexes.

An antidromic impulse in the motor neurones of a single muscle does not affect the timing of other units in a flying locust (Wilson, 1964). However, when a burst of antidromic impulses is set up in the motor neurones of eleven of the twelve vertical depressor muscles, the remaining depressor muscle motor neurone can be induced to fire up to 2 msec, earlier than expected (Fig. 5). The displacement of the muscle spike is small and difficult to demonstrate because of the normal variability in wing-beat frequency. The effect of the burst of antidromic impulses is confined to advancing the next burst in the unstimulated motor neurone; the second burst after the stimulus occurs at the expected time relative to the normal burst preceding the stimulus. Antidromic stimulation of a group of motor neurones therefore does not reset the wingbeat cycle.

Two kinds of sensory input to the flight system have been studied: input from the stretch receptors and from the campaniform sensilla of the wing veins. Stretch-receptor activity increases wingbeat frequency (Wilson & Gettrup, 1963). The cam-paniform sensilla, which responds to lift forces on the wing, participates in the control of lift (Gettrup & Wilson, 1964) and probably other control reflexes; input from them also increases wingbeat frequency. These are slow reflexes which change the central state of excitation over many wingbeat cycles.

Stimulation of the nerve which contains the axons of the stretch-receptor sensory neurones (nerve IBb) or of the nerve which innervates the campaniform sensilla (la) produced small depolarizing excitatory potentials in flight motor neurones (Fig. 6; Plate 1).

In another experiment the tegula was destroyed and the stretch-receptor nerve was cut, leaving the wing nerve, I A, intact. Of the remaining wing sense organs innervated by I A, only the campaniform sensilla are known to project to the flight system. Bending the wing of this preparation produced a burst of activity in nerve I a and a concomitant slow depolarization and spike-frequency increase in motor neurones (Fig. 6c). The latency between a burst of electrical stimuli to either nerve and the response in a motor neurone was variable, at its shortest 7−8 msec, from the onset of the burst of stimuli. The depolarization in the motor neurone was 3−4 mV. in amplitude and lasted 10 msec, or more. The slow depolarization of the motor neurone in response to twisting the wing began about 45 msec, after the increase of activity in the wing sensory nerve, and declined slowly after the wing was released. There was a post-stimulus depression of motor neurone spike activity.

The characteristics of the response of a motor neurone to sensory input accord well with data on flight behaviour. Single stimuli to a sensory nerve produced a relatively ineffective depolarization with variable latency and amplitude, while sustained natural stimulation of wing receptors elicited a slow smooth response and a tonic increase in spike frequency. Large synaptic potentials in response to electrical stimulation of the sensory nerves were never seen in motor neurones. Such large responses have, however, been observed in unidentified units in locust thoracic ganglia (Iwasaki & Wilson, 1966). It seems likely therefore that the units in the latter case were not motor neurones but interneurones. Their responses to stimulation of wing sensory nerves may be evidence for an interneuronal network in flight motor control.

In the introduction to this paper a model for the flight-pattern generator was discussed in which interactions between motor neurones were responsible for the patterned features of the flight motor output. Evidence has been presented for the existence of motor neurone interactions. A series of experiments was performed to test whether activity in motor neurones is sufficient to produce the flight output pattern. In each of these experiments the procedure was to attempt to phase-lock activity in one or more flight-muscle motor neurones to a train of antidromic impulses, at about wingbeat frequency, in another group of motor neurones. The experiments and their results are described below.

In the first experiment stimulating electrodes were placed in the right metathoracic vertical depressor muscles and recording electrodes in their lateral homologues. Stimulation of the anterior nerve cord was used to elicit uncoordinated motor neurone activity. Antidromic stimulation of the right depressor motor neurones had no apparent effect on the timing of orthodromic impulses in those on the left.

The second type of experiment made use of the fact that only 8 of the flight-system motor units (all in the mesothoracic dorsal longitudinal muscles) are innervated from the prothoracic ganglion. Entrainment of units on the left side by antidromic impulses in those on the right might be easier to demonstrate if they were isolated from the rest of the flight system and activity in all eight motor neurones were monitored. It was also of interest to see whether the isolated prothoracic ganglion, with so few of the motor elements of the flight system, could independently produce the flight pattern. The prothoracic ganglion was isolated from the more posterior thoracic ganglia by cutting the ventral cord posterior to the ganglion and severing mesothoracic nerve I central to its anastomosis with the prothoracic recurrent nerve. The normal flightinitiating stimulus of wind on the head elicited activity in the prothoracic flight-muscle motor neurones in the form of a train of single pulses of decreasing frequency (Fig. 7a). Although the most common frequency within the train was close to normal wingbeat frequency, and synchronous impulses in different units were often present, the relationship of this activity to the flight pattern of bursts of impulses alternating with silent periods is questionable. The train of impulses in prothoracic flight-muscle motor neurones isolated from the rest of the flight system more nearly resembles the well-synchronized trains of impulses in the fast motor neurones of the tibial extensors during a jump (Fig. 7 b). In butterflies also, the isolated prothoracic ganglion fails to produce the flight pattern; in this case the train of spikes is thought to resemble normal warm-up behaviour (Moran & Ewer, 1966).

The apparent inability of the prothoracic ganglion to produce a clear flight pattern in isolation is probably not due simply to the absence of antagonist elevator motor neurones; in the intact flying locust depressor muscles can fire the normal pattern of bursts without known elevator activity (Waldron, 1967). If the flight pattern is produced by an inter-neuronal network, then the prothoracic ganglion lacks its own complete network and is dependent on the other thoracic ganglia; if the flight pattern is the result of motor neurone interactions, perhaps there are simply not enough motor neurones in the prothoracic ganglion to produce a stable pattern of synchronized bursts alternating with silent periods.

In an attempt to entrain activity in the motor neurones on one side to an imposed pattern of impulses in those on the other, the ganglion was isolated posteriorly as described above. Stimulating electrodes were inserted in the right mesothoracic dorsal longitudinal muscle, and recording electrodes were placed in both right and left dorsal longitudinal muscles. Wind was blown on the head while the right dorsal longitudinal motor neurones were stimulated at a frequency of 20/sec. Orthodromic impulses in both stimulated and non-stimulated units occurred at all phases with respect to the antidromic impulses, with no apparent phase preference. Antidromic impulses in four motor neurones are apparently not sufficient to pattern the output of the other four prothoracic flight motor neurones.

Experiments similar to the ones described above under long-term effects were also used in the present series. Separate stimulating and recording electrodes were placed in the mesothoracic first basalar, subalar and second tergosternal muscles on both sides and arranged to allow each muscle to be individually stimulated and recorded from. Antidromic impulses at a frequency of 20/sec. in the motor neurones of single muscles had no phasic effect on activity in either synergists or antagonists. In order to see if pure depressor activity could affect activity in elevator motor neurones, the right mesothoracic recurrent nerve, containing four motor axons to the metathoracic dorsal longitudinal muscles (depressors), was prepared as in Fig. 3 a. Stimulating electrodes were placed on it, recording electrodes were inserted in the left mesothoracic tergosternal muscles (elevators), and the preparation allowed to age until spontaneous elevator activity began. Stimulation of the depressor motor axons with paired impulses 5 msec, apart at a rate of 20 pairs/sec. had no effect on the timing of impulses in the elevator discharge.

In the final set of experiments stimulating electrodes were placed on the right mesothoracic nerve II and the nerve was cut distal to the electrodes. Nerve II innervates all the mesothoracic elevator muscles and two of the three mesothoracic vertical depressors. Recording electrodes were placed in the left tergosternal muscles (elevators) and in the left vertical depressors. The effect of stimulating nerve II on a heterolateral motor neurone was described above under latent positive coupling. Trains of spontaneous impulses in both elevator and depressor motor neurones were observed to lock in phase with the stimulus at latencies which varied from 5−10 msec. from the first of the pair of stimulus artifacts (Fig. 8 a). As predicted from experiments with neural analogues (Wilson, 1965), stable phase-locking occurred only when the unit was active at a frequency close to that of the stimulus (Fig. 8).

The synchronous potential in flight-muscle motor neurones is almost certainly electrotonic. Its timing, shape, and behaviour with respect to imposed alterations of the membrane potential are consistent with the assumption that it is the electrically transmitted, much attenuated action potential of another motor neurone. The other excitatory potential, the delayed potential, is probably a chemically transmitted EPSP. Its latency of 5−6 msec, might allow an intercalated interneurone between the motor neurones, but there is no reason to suppose there is one. As for the effects of these potentials and their part in locust flight both have been shown to have an excitatory effect but neither is large enough to elicit a spike except when the motor neurone is already very near threshold as a result of some other excitatory process. The synchronous potential can thus certainly act to synchronize impulses within a burst of synergists. Burst lengths (number of spikes per burst) tend to fluctuate together among synergist motor neurones (Waldron, 1967). The delayed potential, with its 5−6 msec, latency, may act with other input to spread an increase in burst length among synergist motor neurones.

Of the observed long-term effects the excitatory interactions between synergists, whether they represent motor neurone interactions or sensory feedback, may also participate in maintaining similar excitatory states among synergists, and may contribute to the stability and continuation of flight in the face of fluctuations in other excitatory input by providing a long-lasting positive feedback. The long-term inhibitory effects might help to achieve the proper ratios between the burst lengths of antagonists under conditions when asymmetry of excitatory states contributes to effective flight; in turning, for instance, and in the lift-control response, in both of which manoeuvres the basalar and subalar of the same side are antagonists in wing rotation as well as synergists in wing depression. In a turn, right and left basalars also change burst length in opposite directions.

The observed motor neurone interactions can therefore modify certain features of the flight pattern: the timing of individual impulses within a burst, and the burst length; the long-term effects can modify overall wingbeat frequency and buret-length ratios between certain muscles. The role that motor neurone interactions play in the actual generation of the flight pattern is more difficult to determine. A relatively large group of motor neurones firing antidromic impulses at wingbeat frequency can entrain other flight-muscle motor neurones if the latter are firing at nearly the same frequency as the stimulus. Under the experimental conditions in which this was demonstrated the phase-locking was not stable. Perhaps in normal flight, with a larger number of motor neurones active and maintained in similar excitatory states by natural sensory input to the flight system, motor neurone interactions could produce synchronous bursting among synergists.

The apparent absence of inhibitory interactions between the flight-muscle motor neurones is troublesome. Inhibition between motor neurones has been clearly demonstrated as a factor in the rhythmic output of the crayfish stomatogastric ganglion D. Maynard (personal communication), and there is evidence for it in the patterning of cricket song (D. Bentley, personal communication), a form of behaviour very closely related to flight.

It is possible that there are interactions between motor neurones, including inhibitory interactions, which cannot be demonstrated by the methods described in this paper. Antidromic impulses may not invade all the branches of a motor neurone that orthodromic impulses invade, and thus may not depolarize some important presynaptic endings. Intemeurones interposed between the motor neurones, such as those which would probably subserve inhibitory interactions, might have sufficient excitatory input to reach threshold only in the flying animal; they would then not fire in response to motor neurone input alone in experiments on non-flying animals. A further possibility that the present experiments would not reveal is a network of low-resistance connexions between motor neurones similar to those described for the crustacean cardiac ganglion (Hagiwara, 1961), which transmit slow changes in membrane potential but not spikes.

If motor neurone interactions are not sufficient to pattern flight output, the alternative is a network of interneurones projecting onto the flight-muscle motor neurones in a pattern which is at least partly independent of feedback from the motor neurones. The responses of units which are probably interneurones to stimulation of the wing sensory nerves may be evidence for such a network; there is little additional evidence. Although there are intemeurones in locust ganglia spontaneously active at approximately wingbeat frequency (Svidersky, 1967), they have not been identified as part of the flight system.

For the present the role of motor neurone interactions in the production of the flight-motor pattern is still unclear. The present results show that such interactions exist and that they can modify the timing of impulses. As demonstrated, however, they are not sufficient to account for the pattern itself. At this point, therefore, it seems necessary to postulate another kind of pattern generator in addition to motor neurone interactions to account for the rhythmic bursts and the alternation between elevator and depressor activity characteristic of locust flight.

The author is indebted to Dr Donald M. Wilson for his generous provision of facilities, suggestions and critical discussion. The work was supported by grants to Dr Wilson from the National Institutes of Health (NB 03927) and from the Air Force Office of Scientific Research (AFOSR-1246-67), and by a National Science Foundation Postdoctoral Fellowship to the author.