Neural Mechanism by Which Controlling Inputs Influence Motor Output in the Flying Locust

The wing motion during locust flight is caused by alternating contractions of elevator and depressor muscles. These muscle contractions are initiated by bursts of impulses which occur alternately in the elevator and depressor motor neurons (Wilson & Weis-Fogh, 1962) (see Fig. 1). The isolated thoracic ganglia can produce this pattern of impulse bursts in the absence of any patterned input, or even with abnormally patterned input (Wilson, 1961). Thus the thoracic central nervous system is relatively autonomous in the generation of the basic flight pattern.

On the other hand, inputs to the thoracic ganglia must be able to produce variation in the pattern of activity in the flight motor neurons so that the flight behaviour will change appropriately as circumstances differ. That is, there must be mechanisms for the regulation of the system generating the flight pattern. In this paper two different techniques are used to analyse these mechanisms. The first technique is to stimulate a flying locust with relevant inputs and then to study the resulting changes in motor output pattern. The second technique is to analyse other kinds of variation present in normal output patterns (especially the short-term variation) and then to infer from the characteristics of the variation what factors may have caused it.

The first technique has been used by previous workers. When the pitch, or body angle, of the locust is changed, sensory structures produce altered input to the thoracic ganglia (Gettrup, 1966). As a consequence, the amount of activity in particular motor neurons changes so that the wing motion is adjusted appropriately (Wilson & Weis-Fogh, 1962). These changes develop slowly, over 4–6 sec. (about 70–100 wingbeat cycles) (Gettrup, 1966). Similarly slow is the response to stimulation of nerves from the stretch receptors of the wing hinges; wingbeat frequency increases with a time constant of about 2 sec. (Wilson & Wyman, 1965). Both of these responses occur with little change in the relative timing of activity in different units. Basically similar observations are presented in this paper for the motor output patterns during the production of roll torques in response to asymmetric lighting and for the responses to stimulation of the wing nerve. Thus there is a variety of controlling inputs which produce slow changes in the average parameters of the flight motor pattern.

In addition to these slow changes in average parameters there is more rapid variation of some parameters. There are no known inputs which produce the cycle-to cycle changes in the flight pattern. Hence the first technique of analysis (the manipulation of inputs) is not applicable and it was necessary to turn to an analysis of the variation itself in an attempt to identify its sources. Such an analysis is presented in the second part of this paper. It is concluded that the rapid variation in output is apparently largely due to varying responsiveness of the individual motor neurons rather than to variation in controlling inputs.

These findings suggest a model in which all the observations are explained on the hypothesis that controlling inputs act by a single mechanism : the production of slow changes in the ‘average excitation ‘of individual motor neurons.

In all experiments adult male Schistocerca gregaria performed tethered flight in front of an open-jet wind tunnel. Muscle potentials were recorded, using the techniques described by Wilson & Weis-Fogh (1962). There is one muscle potential for each action potential in a motor axon, at least in the range of impulse frequencies which occurs during flight. With care it is possible to identify those muscle potentials which are caused by activity in a single motor neuron. Thus, though muscle potentials were recorded, the results also describe the behaviour of individual motor neurons.

In the experiments on roll turns, the animal was fixed so that it could not actually turn. Instead, the torque (i.e. the tendency to turn) was measured, using a torque meter which has been described by Thorson (1964) and was kindly lent to me by him. The roll torques were elicited by asymmetric visual stimulation, as shown in Fig. 2. In other experiments I studied the responses during flight to stimulation of the femoral nerve of one prothoracic leg or to stimulation of the nerves innervating the campaniform sensilla of both hind wings. In the former case the stimulus electrodes were inserted through the exoskeleton into the middle of the femur. In the latter case each cathode was a 100 μ copper wire, inserted into the subcostal vein of each wing. The nerve from the campaniform sensilla runs through this vein (Gettrup, 1966). Distal portions of the nerve were destroyed so that normal input from the campaniform sensilla of the hind wings was eliminated, for the most part. An anode was inserted in the intercostal vein of each hind wing.

The short-term motor variation analysed in the second part of the paper occurred under a wide variety of conditions. Records were taken at different wind speeds and angles, with varying degrees of wind turbulence, and with different kinds of room illumination. The described properties of the output pattern were similar under all these conditions.

Several terms have been used in the description of the motor output pattern (Fig. 1). The period is the duration of a wingbeat cycle. The burst length is the number of times a given unit fires in one wingbeat cycle. Following earlier conventions, bursts of length one, two, etc. are referred to as single, double, etc. firings. The latency is the interval between the first firing in the burst of one unit and the first firing in the next burst of another. The phase is the latency divided by the period. Another measure which is called latency is the time from the top of the wing stroke, or some other fixed point in the cycle of wing motion, to the first firing in the next burst of the motor unit (Fig. 5 b). These two measures of latency give equivalent results for the questions considered in this paper. Latency, as defined here, is a special case of ^‘ cross-interval ‘as defined by Moore, Perkel & Segundo (1966).

For some records, spike times were measured from film with a Gerber Scientific Instrument Co. GDDRS-38 digital data reduction system. These times were then used in computer calculations of the described parameters and of relevant statistics.

In the analysis of burst-length sequences I have calculated serial-correlation coefficients and cross-correlation coefficients. In this paper a first-order serial-correlation coefficient is a conventional correlation coefficient in which each burst length is compared to the burst length of the same unit in the next period. For the second-order serial-correlation coefficient each burst length is compared to the burst length occurring two periods later; etc. In cross-correlation coefficients analogous comparisons are made between burst lengths of different units. Wyman (1965) presents a more general usage of correlation coefficients for the analysis of nerve impulse data. Because burst length can assume only a few discrete values, χ² analysis can also be used to test for correlation between two sets of burst lengths. χ² tests have confirmed conclusions based on conventional tests of the statistical significance of the correlation coefficients. Approximately stationary records were used for these analyses ; each record had several hundred periods and was considered to be stationary if the serial-correlation coefficients of order 40 and 100 were not statistically significant at the 1 % level.

During the production of roll torques the activity in meso- and metathoracic depressor muscles usually shows one or the other of two characteristic patterns of change. In the simpler, but less common, pattern the average burst length in first basalar and subalar depressor muscles increases on the side which tends to be higher and decreases on the other. Since an increase in burst length results in stronger contractions (Wilson & Weis-Fogh, 1962), more power is produced by the wing stroke on the more active side. The observed torque is a result of the greater lift on that side. In the more common pattern, the activity of first basalars decreases relative to subalars on the upward side, and reciprocal changes occur on the other side (Fig. 3). These downstroke power muscles control the angle of attack of the wing during the downstroke, and opposite changes in this angle on the two sides may cause the reciprocal changes in lift which are responsible for the roll torques. Both types of motor pattern also occur during the production of yaw and roll torques which develop spontaneously or in response to asymmetric wind stimulation. Dugard (1967) has observed a similar pattern of response in the mesothoracic basalar muscles during turns, though she does not report changes in the activity of the metathoracic basalar or the subalar muscles.

There is little change in the relative timing of the activity in the various units during these responses. However, when the burst length of one unit increases relative to that of another the burst also begins slightly earlier relative to the burst in the other unit (as reported also by Dugard, 1967). The torque and the changes in muscle activity develop slowly over a period of at least 20 wingbeat cycles (about one second). This slowness may be due, partially or entirely, to slow development of the effective stimulus to the flight system. The visual input apparently stimulates the flight system only indirectly, by stimulating head turning, which results in asymmetric input from the neck hairs to the thoracic ganglia (Goodman, 1965).

Electrical stimulation of either of two different input nerves produces responses which support the generalization that controlling inputs to the thoracic ganglia produce slow changes in the quantitative features of the output pattern. In both cases, the response develops over tens of wingbeat cycles, and there is little change in the relative timing of the activity in different units. In neither case does the response depend much on the timing of the stimulation relative to the flight cycle. In the first case electrical stimulation of the nerves running through the femur of the prothoracic leg appears to crudely mimic tarsal contact, which usually slows or stops flight.* In response to electrical stimulation the wingbeat period increases, and the elevator muscles and some depressor muscles (the dorsal longitudinal muscles) generally show decreased burst length. The first basalar and subalar muscles, however, usually show an increase in burst length. In the second case stimulation of the nerves from the campaniform sensilla of the hind wings produces changes in the average burst length of various motor neurons. Which units respond, and whether they are excited or inhibited, differs in various preparations, probably because slightly different electrode placements result in varying degrees of stimulation of other nearby sensory nerves (from the wing hairs and from the tegulae). The time-course of the response varies considerably, but the typical response is half completed about 15 wingbeat cycles after a change in the stimulus (Fig. 4).

One exception has been observed to the generalization that responses to input develop slowly. Flight may start or stop abruptly (within 30 msec.) in response to wind on the head or loss of tarsal contact (Waldron, 1967). The special factors involved in abrupt stops and starts will not be discussed in this paper. The mechanism for one other kind of response is also beyond the scope of this discussion. Simultaneous increases in the burst length of many units (for instance, during spontaneous variation or in response to general stimuli such as wind) are usually accompanied by a gradual decrease in wingbeat period (Wilson & Weis-Fogh, 1962) (see Fig. 1). The mechanism of control of wingbeat frequency cannot be understood on the basis of the very limited assumptions about the mechanism of flight pattern generation which are made in this paper. (See Wilson & Waldron (1967) for a full discussion.)

It has been shown that changes in motor output during flight manoeuvres develop slowly, over tens of wingbeat cycles. Some parameters of the flight pattern show very little variation during shorter time spans. For instance, in a typical stationary flight, the period had a mean of 56·6 msec, with a standard deviation of only 1·4 msec. In this flight, for one motor unit the mean time between the two firings of a double firing was 11·1 msec, with a standard deviation of 1·1 msec., indicating that this parameter also does not vary much.

However, burst length does undergo rapid variation. Since burst length is a discrete variable, its minimum change from one cycle to the next is one impulse. Such changes in burst length may occur repeatedly within a few cycles (Fig. 3). But even for burst length the rapid variation is limited. A unit commonly is active for tens or hundreds of wingbeat cycles without any change in burst length. Furthermore, when burst length does vary, the change from one cycle to the next is rarely greater than one; for example, a single firing is seldom followed by a triple firing. Even when input changes are abrupt the burst length of a unit rarely changes by more than one impulse within ten flight cycles. Phase also shows limited, but significant, rapid variation.

Under all conditions phase and burst length are negatively correlated. That is, when a unit fires less, it fires relatively later in the wingbeat cycle (Fig. 5). This is true whether comparisons are made between consecutive cycles during rapid variation or between widely separated cycles during slow variation. The phase increase is such that the average single firing occurs roughly in the middle of the part of the wingbeat cycle spanned by an average double firing. Exceptions to this relationship occurred in individual cycles, but the average behaviour of a unit obeyed this relationship with only one exception out of hundreds of units studied.

Since burst length can assume only a few discrete values and phase is continuously variable, phase may vary within certain limits even when burst length is constant. Some of this variation is correlated to changes in burst length in previous or subsequent cycles. Thus, in addition to the negative correlation between phase and burst length in the same cycle, there is a negative correlation between phase in one cycle and burst length in the immediately preceding or following cycles (Fig. 5 a). The latter correlation is so weak that it never overrides the former correlation; for example, the first firing of a double firing comes earlier than a single firing no matter what the burst length in the neighbouring cycles may be.

Since phase and burst length are so consistently correlated, it seems that variation in the timing and amount of activity are controlled by a common mechanism. Thus, an analysis of the sources of short-term variation will be essentially complete once the causes of the burst-length variation are found, since these causes appear to be also responsible for the changes in phase, and no other parameters of the flight pattern vary much over the short term. The next two sections are devoted to an analysis of the rapid burst-length variation and its causes.

Evidence about the cause of the rapid changes in burst length can be obtained by analysing whether the variation in different units is independent or well-correlated. If variation in different units were strongly correlated, this would imply that strong interactions occur among the motor neurons and/or that there is variation in shared input, where input may come from interneurons as well as from sensory neurons. On the other hand, if the burst length variation is independent in different units, this variation must be due to unshared input, or to changing responsiveness within the individual motor neurons, or to both.

In many cases short-term variation in burst length occurs in one unit and not in another recorded unit. In records which are stationary (i.e. which do not show longterm trends) no tendency has been observed for different units to show variable burst length at the same time. For stationary records in which burst length does vary in two synergistic motor units there usually is a weak positive correlation of concurrent burst lengths. That is, in any cycle in which one unit has a larger than average burst length, the other tends also to have a large burst length (and similarly for small burst lengths) (see Fig. 6). This is observed as a positive zeroth-order cross-correlation coefficient, since the zeroth-order coefficient compares the burst lengths of one unit with the concurrent burst lengths of another (see Methods). For two motor neurons innervating the same muscle, the zeroth-order cross-correlation coefficient is about + 0·4. For all other pairs of synergistic motor neurons it is smaller, and generally similar for different pairs whether or not the units are closely related anatomically or functionally. There is no qualitative difference and little, if any, quantitative difference between relationships of motor neurons from the same ganglion and relationships of motor neurons from the two different (meso- and metathoracic) ganglia. The relationships between elevator motor neurons are somewhat stronger and more consistent than those between depressor motor neurons. In general, the zeroth-order coefficients for motor neurons innervating different synergistic muscles are about +0·2. (These coefficients, though small, are statistically significant at about the 0·001 level.) Since the fraction of variation shared in common equals the square of the correlation coefficient, only about 4% of the variation in burst length of a given unit is shared with variation in the concurrent burst length of a typical synergist, though about 20 % of the variation is shared in the case of units in the same muscle. In a few cases the zeroth-order correlation coefficient was negative. Higher-order correlation coefficients are even smaller and more variable than the zeroth-order coefficients.

The correlations between burst lengths of pairs of antagonistic motor neurons are even weaker and more variable than those between pairs of synergists. A positive correlation between elevator burst length and the length of the next depressor burst is the most consistent. The correlation between depressor burst length and the length of the next elevator burst is sometimes positive and sometimes negative. That is, the elevator burst length tends to be like the following depressor firing, and shows no consistent relation to the preceding one. Possibly, some cyclic sensory input stimulates depressor and elevator motor neurons just after depressor activity and before elevator activity in each cycle, thus tending to produce common variation in the elevator and subsequent depressor activity. Even the relatively stronger correlations of elevator with next depressor firing yielded correlation coefficients generally less than 0·1, indicating that less than 1 % of the variation was shared in common.

In conclusion, the low level of correlation of burst lengths for any pair of units indicates that only a small part of the short-term variation in burst length for a given unit is generated by changes in input shared with any particular other unit or by interactions with that other unit. Further, the interactions between motor neurons or the common input responsible for cycle-to-cycle correlations in burst length must be diffusely distributed to all synergists, since the strength of the correlations is so generally similar for closely or distantly related pairs of synergists. The only exception to this is the case of pairs of neurons innervating the same muscle, which show relatively stronger relationships. The conclusion that the chief cause of rapid variation in burst length is neither common input nor interaction between the motor neurons must be qualified by mentioning two possible sources of error. First, the net contribution from all of the other synergistic motor neurons to the variation in burst length of a particular unit cannot be estimated from this data on pairs of units. Secondly since burst length is a discrete variable which presumably does not change in response to every change in input, a change in shared input could result in a change in burst length in one unit and not in another, thus giving rise to uncorrelated variation in burst length. Nevertheless, the low correlations between burst lengths in different units imply that much of the short-term variation in burst length is generated by varying unshared input, or by varying responsiveness of the motor neurons to a constantly repeated input cycle. The importance of the latter factor is indicated in the next section.

Since the cycle-to-cycle changes in burst length occur largely independently in different units, the source of these changes must be some factor that affects individual units independently. Evidence that short-term burst-length variation may be due to varying refractoriness of the individual motor neurons comes from an analysis of the sequences of burst lengths. Again, records were used which were stationary over several hundred cycles. In such records burst length rarely varies by more than one, and thus each unit produces only two kinds of burst length.

For most units there is a tendency for burst lengths of the two different kinds to alternate (Fig. 6). This gives rise to a negative first order serial-correlation coefficient (Fig. 7). Alternation of long and short bursts may be due to differences in refractoriness following long and short bursts. Refractoriness accumulates with repeated activity (Wilson, 1964) so that there is greater refractoriness following the last firing of a longer burst Also the activity in a longer burst ends relatively later than the activity in a short burst (Fig. 6) so there is a later development of refractoriness in long bursts. For both these reasons there will be a greater refractoriness at the beginning of the next cycle, so a burst following a long burst will tend to start later and have fewer firings. In these stationary records the only statistically significant serial correlation coefficient was often a negative first-order one, indicating that changes in refractoriness may be the main source of non-random variation.

Records in which burst length varies were examined for evidence of regular patterns in the sequence of burst lengths, to see whether such patterns occurred and whether they might reveal new features of the mechanism which produces rapid variation in burst length. Recognizable, regular patterns were found, as illustrated by the record of Fig. 8 a in which the typical pattern, double-double-single, is repeated six times. The probability of occurrence of this repeated pattern, if burst lengths followed each other in random sequence, is less than 0·0002. Patterns such as this constitute about one-fifth to one-quarter of any record in which burst length is variable. These patterns are generated independently in the different units (Fig. 86). This implies that the pattern generation is taking place in the motor neurons, or at least not in neurons which provide input to them in common. It is shown in the next section that these burst-length patterns are probably due to the variation in the unit’s refractoriness.

Controlling inputs have been shown to result in slow changes in the amount of activity in the individual motor neurons. More rapid changes in the output pattern also occur, and have been attributed partly to changing refractoriness in the individual motor neurons. These two findings taken together suggest the following model for the mechanism of production of the variation in the flight motor output. The only assumption made about controlling inputs is that their sole effect is to produce slow changes in the average excitation of individual motor neurones. All of the observed properties of the variation in output emerge as predicted features of the response of motor neurons to the slowly changing average excitation, if a few very general assumptions about the flight pattern generator are made. Basically, it is assumed that the net excitation reaching each flight motor neuron varies cyclically. This postulate is almost certainly true with periodically varying excitation coming from the central nervous flight pattern generator, including the other motor neurons, and from some kinds of sensory input (Wilson & Waldron, 1967). Firing occurs in a motor neuron when the net excitation reaching it exceeds its threshold (Fig. 9 a). Threshold recovery after an action potential is assumed to be exponential with a time-constant of about 5 msec., which is the approximate value for locust motor axons (Wilson, 1964).

If a controlling input increases the average level of the cyclically varying excitation relative to the asymptotic threshold, then the first impulse in each cycle will occur earlier and there will be more impulses in each cycle (Fig. 9). Thus phase and burst length will be negatively correlated, as observed in the locust. Some intermediate levels of excitation will be just sufficient to result in a double firing in a cycle when the unit is rested, but will only result in a single firing when the threshold is elevated, as in a cycle following a double firing (Fig. 96). This will result in a tendency for alternation of bursts of different lengths, as observed in the locust. Since this burst-length variation is due to events occurring within individual units, it will be uncorrelated in different units, thus conforming to the observed low cross-correlation coefficients for burst length. Furthermore, if the relative levels of excitation and threshold are appropriate and if there is little noisiness either in the excitation or in the responsiveness of the unit, then the changes in refractoriness and the consequent changes in timing and burst length will repeat regularly, resulting in a regular pattern similar to the observed burst-length patterns. Such patterns have been previously observed for model and neural systems in which burst length varied between o and 1 only. As in this model, the mechanisms proposed have depended on variation in threshold recovery due to changes in timing (Harmon, 1961) and to accumulation of refractoriness (Wilson, 1964).

The unpatterned variation of burst length is presumably produced by unpattemed variation in the excitation or the threshold or both. I consider such unpatterned variation noisy, because I have no evidence for particular mechanisms causing it and it seems likely that factors like turbulence in the wind and biological noise could produce this variation. Since the burst length of a unit is very often constant, such factors presumably result in changes in burst length only when the average excitation is at a transition level, as in Fig. 9b, where a small change in the threshold or excitation can change the number of impulses produced in a cycle. More explicitly, it is assumed that the patterned and unpatterned perturbations present in the system are small relative to the difference between these transition levels, so that in between transition levels there are considerable ranges of average excitation level within which bursts of only one length are produced.

Two previously made assumptions are sufficient to explain the remaining observations on variation in motor neuron activity. First, the assumption that controlling inputs can produce independent changes in the average excitation of the individual units explains how average burst length can vary independently in different units during various control manoeuvres. Secondly, the assumption that changes in average excitation are slow relative to the wingbeat cycle explains why burst length very rarely changes by more than one from one cycle to the next. This assumption, that the average excitation tends to be similar in neighbouring cycles, is also the simplest explanation of the observation that the phase in one cycle tends to be like the phase in neighbouring cycles (e.g., a single firing has an earlier phase when it occurs between double firings) (see Figs, 5a and 9b).

In conclusion, the hypothesis that controlling inputs act by producing slow shifts in the average excitation of individual motor neurons was suggested by the observation that during flight manoeuvres the main changes in output pattern are slow shifts in the average burst length of various units. The description of the model has shown that this same hypothesis can account for all of the additional data about rapid variation in output patterns. Thus both the observed changes in timing of bursts and the variation in burst length can be explained without postulating any patterned change in excitation from one cycle to the next. In fact, attempts to explain these relationships in terms of cycle-to-cycle changes in the excitation reaching the motor neurons necessitate a variety of additional assumptions which seem either improbable or unnecessarily complicated.

Analysis of the variation in flight motor activity indicates that controlling inputs have their main effects on average activity over many cycles, rather than on activity in a particular cycle. This does not imply that controlling inputs may not change abruptly. The flight system output changes only slowly even when the input changes to the thoracic ganglia are known to be abrupt, such as when they are initiated by controlled electrical stimulation (Wilson & Wyman, 1965 ; and this paper). Thus the slow control seems to be at least partly a property of processes within the thoracic ganglia. The slow responsiveness does not seem to be a property of the first stage of input processing within the thoracic ganglia. Synaptic potentials produced in the thoracic ganglia by the wing sensory nerves are of ordinary duration, with a rise time of less than a fifth of a flight cycle, and show no marked facilitation (Iwasaki & Wilson, 1966). Presumably the later stages of input processing are responsible for the slowness of the changes in the motor neurons.

Though the neural mechanisms are not yet understood, it makes sense from a behavioural point of view that control of the flight system should be slow control over average output of particular muscles. This kind of control is consistent with locomotion through a relatively homogeneous medium (air), when there appears to be no need for regulation on a cycle-to-cycle basis, and where inertia will tend to prevent short-term variation in motion anyway. It should be noted that, though the changes are slow relative to the wingbeat cycle, the cycle is only 50 msec, long, so that major behavioural effects are evident within half a second and most manoeuvres are completed within a few seconds. Differential changes in the activity of particular units cause changes in the wing motion which result in turning, lift-control and other types of flight regulation. Thus, for the locust flight system differential control of the average output of individual motor units would presumably be behaviourally significant, and it is just this kind of control which is observed.

I am grateful to Dr Donald Wilson for stimulating discussions and the helpful advice he gave me throughout this work. Dr Horace Barlow and Dr Peter Marler made useful criticisms of the manuscript. Locusts were kindly supplied by the Anti-Locust Research Centre, London. Financial assistance is acknowledged from the University of California (Berkeley) Computer Center, from an NSF predoctoral fellowship to the author, and from research grants from NIH (NB-03927) and the Air Force (AFOSR-1246) to Dr Wilson in the Department of Molecular Biology (University of California, Berkeley).