Neural Interactions Controlling Timing of Flight Muscle Activity in Drosophila

Interest in how a patterned output can be generated by a nervous system has led to a number of analyses of the firing pattern produced by the indirect flight muscles in various Diptera (e.g. Wyman, 1966, 1969a, b, 1970; Mulloney, 1970a). Two major observations have been made concerning this pattern in Drosophila: (1) frequency covariance among the various muscle fibres - which is thought to be the result of a common input and (2) non-random firing relationships between various pairs of ipsilateral fibres - which are thought to be the result of interactions between the motor neurones innervating the fibres (Levine, 1973; Levine & Wyman, 1973; Harcombe & Wyman, 1977, 1978).

The present paper, and the following one (Koenig & Ikeda, 1980), present further observations on the firing pattern of the dorsal longitudinal flight muscle (DLM) of Drosophila. This muscle is particularly useful for the analysis of firing patterns because it is composed of only six singly innervated fibres which represent five motor units (see Methods). In this study, all five units were recorded simultaneously while the animal was in stationary flight. This paper deals with the neural interactions that create the firing relationships between the various ipsilateral fibres, while the subsequent paper characterizes the common driving input which causes the high degree of frequency covariance observed among these fibres.

In the DLM of Drosophila, as well as some other species of fly, it has been observed that no ipsilateral fibre fires immediately after any other (Wyman, 1969 a; Levine, 1973; Harcombe & Wyman, 1977, 1978). It has been proposed that this effect is due to mutual inhibition between the motor neurones innervating these fibres (Wyman, 1969a). The results of the present analysis do not fit with this hypothesis, however, but suggest instead that an ‘excitatory effect’ (causing the cell to fire earlier than it would have fired) plays the major role.

In Drosophila the bilateral pair of DLMs runs longitudinally along the midline of the thorax. Each muscle is made up of six fibres that are arranged one on top of another. These fibres attach anteriorly to the dorsal thoracic cuticle and posteriorly to the posterior phragma. The fibres are designated 1–6 anterior to posterior (Mihályi, 1936). The six fibres are innervated by five excitatory motor neurones (Ikeda, Tsuruhara & Hori, 1975; Ikeda & Tsuruhara, 1978; Coggshall, 1978). Fibres 1–4 are singly innervated, each by a different excitatory motor neurone, while fibres 5 and 6 are both innervated by the same (5th) motor neurone (Ikeda et al. 1975; Ikeda et al. 1980). Therefore, intracellular recordings will subsequently be referred to in this paper as representing motor units 1–4 and 5/6, rather than fibres 1–6.

The flies used in this study were adult (3–5 days old), wild-type Drosophila (Oregon R strain). The flies were lightly anaesthetized with ether. The tip of a tungsten needle was then shallowly inserted through the cuticle at the dorsal midline of the scutum slightly anterior to the scutellum and glued in place with a cyanoacrylate glue. This needle, which was held in a manipulator, served as the reference electrode and was also used to suspend the fly in a stationary flight position. When the fly recovered from the ether it began to fly.

Intracellular recordings were made with glass micropipettes (tip diameter about 1 μm filled with a 1 % solution of Chicago Sky Blue dye in distilled water and of 5–15 MΩ resistance. The electrodes were inserted through the cuticle into the anterior end of each of the six ipsilateral DLM fibres where they attach to the scutum. With proper illumination the boundaries of tergal attachment of the individual DLM fibres were visible through the cuticle. Electrode placement could be confirmed by observing the recorded output pattern and by dye injection into the fibres after recording.

DC recordings were made with conventional high-input impedance DC amplifiers and a seven-track, FM tape-recorder and were monitored and photographed on an oscilloscope. Recordings were usually for a period of 2–5 min, although flight could continue for hours under the experimental conditions. Simultaneous intracellular recordings from all six ipsilateral DLM fibres were made and analysed in about 50 individuals. The experiments were conducted at room temperature (23–25°C). The various terms used in this paper are defined below:

The time (ms) between two consecutive spikes in one fibre. The time of occurrence of this interval is taken to be the time of occurrence of the second spike.

That interspike interval which overlaps (in time) the most with a given interspike interval.

The percentage of an interval that has elapsed when a spike in another fibre occurs.

A motor neurone and the muscle fibre or fibres which it innervates.

During flight the DLM fibres always responded with all-or-none type action potentials of about 110 mV in amplitude. Simultaneous recordings from all six DLM fibres showed that the firing times of the five motor units are spaced apart so that no unit fires right after another, as demonstrated with phase histograms in Fig. 1 (similar histograms appear in Harcombe & Wyman, 1978). Fig. 1A shows that the probability is high that unit 1 will fire in the middle of unit 2’s interval (15–80% of the interval) or synchronously (0%), and very low that it will fire at the beginning (1–5%) or end (80–99%)°f interval. The same relationship is seen for unit 3 relative to intervals of unit 4 (Fig. 1 B). Figs. 1C, D show that similar relationships exist between units 1 and 3, and 2 and 4, although the ‘periods of exclusion’ (1–5% 95–99%) are not so long as for pairs 1 and 2 or 3 and 4.

For the purpose of simplification, the following observations will deal with the firing pattern created by units 1 and 2. It will be shown subsequently, however, that units 3 and 4 demonstrate the same pattern, with 3 equivalent to 1, and 4 equivalent to 2, and that other pairs demonstrate a similar pattern as well. The firing pattern of unit 5/6 relative to the other 4 units is so close to a random pattern that it will not be considered in this paper.

A typical example of how the ‘exclusion periods’ of the above phase histograms are generated is shown by the phase plot of Fig. 2C. Unit 1 fires earlier and earlier in successive intervals of unit 2 until it reaches about the 20% level. Then, instead of firing soon after unit 2 (1–15%), unit 1 either (1) fires synchronously (within a ms) with unit 2 (plotted at o of the next interval) and then fires later in that interval; or less frequently (2), fires even earlier than unit 2, i.e. twice in the preceding interval. Examples of these two events are shown in Fig. 2A and B, respectively.

The phase progression of unit 1 in intervals of unit 2 suggests that when unit 1 should have fired soon after unit 2, a neuronal interaction occurs that either synchronizes the firing times of the two units or spaces them apart by causing unit 1 to fire much earlier than unit 2. The intervals involved in such interactions will be termed ‘interacting intervals’ and are defined as (1) any interval which ends or begins with a synchronous firing with another ipsilateral DLM unit (Fig 2 A, firing configuration 1); and (2) any interval which contains another ipsilateral interval or is contained by another ipsilateral interval (Fig. 2B, firing configuration 2). All other intervals will be referred to as ‘non-interacting’.

To investigate how these interactions change the firing time of a unit, the interspike intervals of a unit were compared with the non-interacting concurrent intervals of the other units. It has been previously observed that the interspike interval of a given DLM is usually similar in duration to such concurrent intervals, even though the duration of successive intervals may fluctuate considerably (Levine, 1973; Harcombe & Wyman, 1978). (This correlation is thought to be due to the influence of a common input to all of the units.) Thus, an estimate of what the length of an interval would be if it were not altered by a spacing interaction is given by the average length of the ‘non-interacting’ concurrent intervals. This average will be termed the ‘reference interval’.

The lengths of successive intervals of unit 1 and their reference intervals are compared in Fig. 3. As can be seen, the interacting intervals (arrows) show a reduced correspondence, especially those which were involved in an interaction with unit 2 (thicker arrows). This concurrent interval relationship can be observed for many successive intervals of a particular unit in a relative interval duration (RID) histogram, where each occurrence represents the percentage that an interval is of its reference interval. These histograms were always very similar in appearance (more than 50 flies), c.g. Figs. 4 and 5, which show data from 3 different flies.

The histograms for unit 1 (Fig. 4) are characteristically skewed with a long tail to the left. The occurrences making up this tail all represent interacting intervals, with the intervals which have been involved in an interacting configuration with unit 2 being the furthest to the left. The histograms for unit 2 (Fig. 5), demonstrate a fairly bell-shaped distribution, but a few occurrences are always found at the far right (120% or more). These are always intervals that have been involved in an interacting configuration with unit 1. Thus, the most poorly correlated intervals (below 80%, above 120%) are always those which have been defined as interacting, e.g. intervals where the probability was high that unit 1 should have fired soon after unit 2. These histograms, therefore, suggest that a neural interaction specifically related to the Base relationship between these units occurs to change the firing times of the units,

By observing how the interacting intervals correlate to their reference intervals, it can be determined in what direction (earlier or later) the firing times of the units were changed. Occurrences involving intervals of unit 1 which ended in a synchronous firing with unit 2 (◼, Fig. 4) average around 70% ; i.e. these intervals are considerably shorter than the concurrent intervals with which they are being compared. Since unit I’S non-interacting intervals (unshaded) normally are of similar duration to the concurrent reference intervals (averaging aroung 95%), these intervals which end in synchrony are much shorter than would be expected; that is, unit 1 has fired earlier than it normally would have fired. On the other hand, the companion intervals of unit 2, those which have ended in a synchronous firing with unit 1 (◼, Fig. 5), average around 100%, which means they are of about the same duration as their concurrent reference intervals. The non-interacting intervals of unit 2 also average around 100%. Thus, it appears that unit 2’s firing time is not changed.

The above observations suggest that for the 1st firing configuration, when unit 1 should have fired soon after 2, 1 is caused to fire earlier than it normally would have fired, so that instead it fires synchronously with 2 (Figs. 6 A, 7).

Intertervals of unit 1 that were involved in the 2nd configuration (⧄, Fig. 4), that is, where unit 1 has fired earlier than unit 2, are also observed to be shorter relative to the concurrent reference intervals than they normallly would have been. In these examples there are relatively few such occurrences, averaging around 70%. However, in the histograms from other flies, a similar reduced correspondence is always seen.

The companion intervals of unit 2 (⧄, Fig. 5) are observed to be longer relative to their concurrent reference intervals than usual, averaging around 115%, as compared to 100% for the non-interacting intervals. Thus, it appears that 2’s firing time has been delayed at these times.

The above observations suggest that for the 2nd configuration the firing times of both units i and 2 are changed with 1 again firing earlier than expected and with 2 now firing later than expected (Fig. 6b, 7).

The intervals immediately following a synchronous firing (⊠) sometimes appear to be affected in a similar way to that for configuration 2 (unit 1 shorter, 2 longer) at other times do not. These intervals represent a somewhat different situation since the interval begins with units 1 and 2 firing synchronously. This means that phase progression will often put unit 1 firing ahead of unit 2 at their next firing. This situation results in a variety of possible interactions, which will be discussed in a later section.

The relationship between the RID value of a unit’s intervals and the phase at which another unit is firing within those intervals is demonstrated in the scattergram of Fig. 8. It can be seen that the RID for intervals of unit 1 averages about 96% whether unit 2 has fired at 30, 40, 50, 60 or 70% of the interval. However, the correspondence becomes significantly reduced when the interval is involved in one of the two interacting configurations, i.e. when unit 2 has fired synchronously with unit 1 (plotted at 100%) or when unit 2 has not fired within i’s interval (plotted at o). Thus, it was observed that except for the interacting configurations (those times when it was highly probable that one unit would have fired soon after another), the concurrent interval relationship is essentially the same no matter where one unit has fired in the other’s interval.

In the scattergram of Fig. 8 the points representing intervals that just preceded an grading interval are designated with an arrow. These are all intervals where unit 2 was at a phase close to the exclusion area (80–99%). However, it can be seen that unit 2 firing at this phase does not necessarily mean that an interaction will occur in the next interval. Thus, the interactions are not triggered by how close (at what phase) the two cells fired in the preceding interval. However, they are related to the probability that in the next interval, unit 2 should have fired in the ‘exclusion area’, since it is always at a phase close to the exlusion area in these preinteraction intervals. Thus, it appears that these interactions are triggerd by the event itself (1 about to fire soon after 2) rather than by some preceding event.

The phase progression of unit 1 in intervals of unit 2 predicts certain instances where unit 1 should have fired soon after unit 2. The RID histogram can also be used to predict approximately when a unit should have fired. For example, in Fig. 9, unit 1’S 2nd interval has ended in a synchronous firing with unit 2 (configuration 1). The concurrent reference interval for this interval is 125 ms. The RID histogram for this fly (fly no. 1 of Fig. 4) predicts that the non-interacting interval should be within 78–113% of its reference interval; that is 98–141 ms in duration. As can be seen, this range (designated by the bar) predicts that unit 1 should have fired well behind unit 2 with some of the shorter values falling soon after unit 2 in 2’s area of exclusion (designated by arrow) (the exclusion area for this fly was determined from the phase histogram as 0–15% of the interval). This suggests that unit i’s shorter intervals (relative to the concurrent reference interval) would be those which would become involved in an interaction. The effect of this on the distribution of the non-interacting (unshaded) intervals can be observed in Fig. 4. Since this distribution should be due to a random process (see next paper, Koenig & Ikeda, 1980), a bell-shaped distribution might be expected. However, there is a deficit of smaller values (left shoulder) in the distribution of non-interacting intervals. If the interacting intervals (shaded) were added to the left shoulder of the distribution of non-interacting intervals, the distribution would become more typically bell-shaped. Thus, it appears that the interacting intervals were displaced from the left shoulder of the non-interacting interval histogram, producing the tail on the left of these histograms.

A second method that also predicts that unit 1 should have fired soon after unit 2 at these times had the interaction not occurred is discussed in the subsequent paper (Koenig & Ikeda, 1980). This is related to how a particular interval fits into the pattern produced by the concurrent intervals. Both methods always predict that the firintr configurations made up of intervals defined as interacting were created as shown Fig. 6.

The interactions between units 3 and 4 appear to be the same as those described for units 1 and 2, with 3 being equivalent to 1, and 4 being equivalent to 2. A pair of relative interval duration histograms for units 3 and 4 are presented in Fig. 10. As can be seen, the interacting intervals (◼, ⊠ and ⧄) again exhibit reduced correlations of the same direction and magnitude as those observed for units 1 and 2. Again, the histogram for unit 3 is typically skewed to the left, while that of unit 4 is skewed to the right.

Interactions between other pairs of these units (1–3, 2–4, 1–4, 2–3) appear to be similar to those observed between 1–2 and 3–4, but they appear considerably weaker. The phase histograms for these pairs also exhibit areas of exclusion, but the areas are reduced in size (Fig. 1C, D). In Figs. 4 and 5 the instances where units 1 or 2 interacted with units 3 or 4 are represented by a dot. It can be seen that these intervals are somewhat less correlated than the non-interacting intervals, although they are more correlated than the intervals involved in interactions between units 1 and 2. This is what would be expected if these interactions (between 1–3, 2–4, etc.) have a smaller effect on interval size than interactions between 1–2 and 3–4. It cannot be conclusively shown by the method of analysis used here that the interaction between these pairs is same as for units 1–2 and 3–4. However, the analysis presented in the following ‘paper (Koenig & Ikeda, 1980) also suggests this is the case.

In this study, recordings from more than 50 flies were analysed. Although the vas majority showed the frequency relationships described above (unit 1 firing slightly faster than 2; unit 3 slightly faster than 4), a few cases were observed where 2 fired faster than 1, or 4 fired faster than 3. In these instances it was observed that when 2 fired progressively earlier in i’s intervals until it would have fired soon after 1, 2’s interval was observed to be shortened so that it fired synchronously with 1 or before 1 ; that is, the reciprocal interaction appeared to occur. This was also observed to occur occasionally with units 3 and 4 as well as with pairs 1–3, 2–4, etc.

These observations suggest that the interactions described above are all reciprocal and depend on which cell would have fired soon after the other. The reason that unit 1 ‘s intervals are almost always shortened while 2’s remain unchanged, or are sometimes lengthened, is simply because unit 1 is almost always firing faster than 2 and would therefore tend to fire progressively earlier in 2’s interval until it fired right after 2. At this time, an interaction occurs which alters the relative firing times of units 1 and 2 so that i is again firing in the middle of 2’s interval. Thus, there is never an opportunity for 2 to fire soon after 1.

In the case where units 1 and 2 have fired synchronously, unit 1 will be expected to next fire sometimes earlier than unit 2, and sometimes later. Relative interval duration analysis suggests that various interactions occur at these times. Sometimes, it appears that unit 1 has fired earlier than expected (configuration 2), suggesting that 1 should have fired soon after 2 if the interaction had not occurred. However, occasionally unit 1 ‘s firing time appears unchanged while 2’s firing time appears delayed, or both 1 and 2’s firing times appear unchanged. Also, it is occasionally observed that units 1 and 2 fire synchronously a second time (in this case, either 1 or 2 could have been shortened to produce this). Thus, the reciprocal nature of the interaction appears to produce a variety of configurations at this time.

In 1969, Wyman proposed a mutual-inhibition hypothesis to explain the phase relationships observed between various pairs of DLM units in another species of fly to that used here. This hypothesis proposes that when one motor neurone fires, it causes an inhibitory effect in the other motor neurones, thereby delaying the firing time of any motor neurone which was about to fire. To test this, one unit can be stimulated antidromically while observing the effect of the antidromic spike on the firing of the other units. This experiment was performed by Mulloney (1970,a) on Calliphora and Eristalis;Levine (1973), Levine & Wyman (1973), and Harcombe & Wyman (1977) on Drosophila.

Mulloney (1970,a) reported that the effect of the antidromic stimulation in Calliphora was to delay the next firing time of both stimulated and unstimulated units, so that their orthodromic interval (orthodromic spike to the next orthodromic spike) was increased by over 50%. It has been suggested by Harcombe & Wyman (1977) that this result might have been due to accidental stimulation of the supposedly unstimulated units, since Mulloney’s experimental design did not allow observations whether or not the units other than the one stimulated also exhibited an antidromic Spike.

Levine (1973) observed that a contralateral unit did not fire immediately (within a few ms) after his stimulus and interpreted this to mean that the antidromic spike had delayed the firing of that unit. However, the probability that a spike should have occurred during this short time period is very low, so that it cannot be claimed that the observation is a result of the antidromic spike. Furthermore, contralateral units normally show little or no inhibition between them (as observed by their phase relationships).

Harcombe & Wyman (1977) reported that no ipsilateral unit fired soon after their stimulus and also interpreted this to mean that the antidromic spike had delayed the firing of the other units. However, in order to claim that antidromic stimulation has caused the observation that no other unit fires soon after the stimulus, one must establish that the units would have fired soon after the stimulus in the first place. This was not established by Harcombe & Wyman. On the contrary, the data which are presented suggest that the effective stimuli occurred primarily or totally when no other unit would have been firing. Although the stimuli were presented in a random manner relative to the muscle spikes, not all stimuli were effective in producing an antidromic spike. From the distribution of points in Fig. 3B of the Harcombe & Wyman paper it appears that it was more probable to get an effective stimulation soon after an orthodromic spike (thus, a preponderance of effective stimuli between o and 50 ms after an orthodromic spike) or right before one might have occurred (another preponderance of effective stimuli between 150 and 200 ms after the orthodromic spike). Indeed, it would be expected that a stimulus should be more effective after a muscle spike, since the depolarizing tail of the muscle fibre junction potential lasts for many milliseconds. This would effectively increase the amount of current stimulating the axon at this time. Thus, it appears that there was an increased probability that the effective stimuli would occur at a time when no other unit would be firing (right before and right after an orthodromic spike in the stimulated unit), and the ineffective stimuli would occur at a time when the other units would be firing. Under these conditions it cannot be claimed that the antidromic spike has caused the observation that no unit has fired right after it. Thus, the evidence for the mutualinhibition hypothesis is inconclusive. As will be discussed below, the observations presented in this paper are inconsistent with the predictions of such a hypothesis and suggest an alternative interpretation.

An inhibition hypothesis predicts that as phase progression causes one neurone to fire soon after another, the earlier firing neurone will inhibit (delay) the 2nd neurone’s firing time, i.e. it will lengthen its interval. This would continue to occur until the later firing neurone either skips ahead of (fires earlier than) the 1st neurone, or until it fires so closely behind the 1st neurone that there is no time for inhibition to occur. An example of this prediction is shown in Fig. 8 of Wyman (1969a).

The observations presented in this paper do not fit these predictions. For example, .unit’s interspike interval did not become longer than expected as phase progression caused it to fire closer and closer behind another unit, but rather, was independent of where (at what phase) the other units were firing within that interval, over a range of about 20-80% of the interval. This is demonstrated in the scattergram of Fig. 8, which shows that unit 1’s relative interval size remains constant no matter where unit 2 is firing within that interval (the effect of the other units should be constant for unenclosed points).

A second observation that does not fit the predictions of the mutual-inhibition hypothesis is that when a unit has fired synchronously or has skipped ahead of another unit, its interval is not of expected size (it is now receiving no inhibition) as the inhibition hypothesis would predict, but rather is much shorter than expected. This observation could only be explained by the mutual-inhibition hypothesis if it were assumed that the inhibitory effect of unit 2 on unit 1 (or unit 4 on unit 3) always caused unit 1’s intervals to be lengthened by about 20%, no matter where unit 2 was firing within the interval. Then, one might explain these shorter-than-expected intervals as being disinhibited. However, the magnitude of unit 2’s delaying effect on the firing of unit 1 should be greatest right before unit 1 would have fired and very slight right after unit 1 had fired. Thus, this interpretation is very unlikely.

Another observation that is inconsistent involves the synchronous firings. The mutual-inhibition hypothesis would explain these synchronous firings as instances where phase progression caused the 2nd unit to fire so closely behind the 1st that there was no time for inhibition to take effect. However, units 1–2 and 3–4 fire within 1 ms of each other much more frequently than would be expected due to phase progression of two independently firing units. The observations presented here show that for pairs 1–2 and 3–4 one unit never fires so closely behind the other that phase progression alone would bring it within range of firing synchronously in the next interval. Instead, it suggests that phase progression causes one unit to fire in another unit’s exclusion area (soon after the other unit), and that an additional input causes the unit to fire earlier so that it fires synchronously. This causes the preponderance of synchronous firings reported here. Thus, the observations presented in this paper suggest that the DLM units are firing independently of each other, except when one unit would have fired soon after another. At these times, a phase-related input alters their firing times as shown in Fig. 6.

The interactions suggested in Fig. 6 would not require a complicated circuitry to produce them. For example, by assuming a weak electrical coupling between the motor neurones, one can account for most of the observations presented here. As suggested by Harcombe & Wyman (1977), and also by the data presented here (see next section), the firing time of these motor neurones seems to be dependent on an endogenous integrative process that is reset whenever the cell fires. This suggests that the cell’s cycle (spike to next spike) might consist of a slowly increasing depolarization of the cell’s membrane until it reaches threshold. Under conditions of weak electrical coupling, as motor neurone 2 depolarized toward threshold, it would also tend to depolarize 1. If the coupling were weak, this depolarization would have a significant effect on i’s firing time only if 1 were itself close to threshold (i.e. about to fire son after 2), and then the effect would be to cause 1 to fire earlier than it normally would have fired. This would increase the probability that the two cells would fire synchronously (configuration 1) or could even cause 1 to fire earlier than 2 (configuration 2) if I had a lower threshold (this is consistent with the observation that 1 fires slightly faster than 2).

The delayed firing of configuration 2 could be explained by the effect of a hyperpolarizing afterpotential of 1 on 2. The above interactions would be reciprocal if phase progression caused 2 to fire soon after 1 (then 2 would have a lower threshold than 1).

Conduction velocity might also affect the firing pattern. Because of the unusually close association (along almost their entire length) of the axons of the motor neurones innervating these muscle fibres (Ikeda, Koenig & Tsuruhara, 1980), the possibility exists that activity in one axon might affect the conduction velocity, and thus arrival time, of a spike in a neighbouring axon. An example of a direct electrical influence of one axon on another was shown by Katz & Schmitt (1940) while working on nonmyelinated crab nerve fibres. They showed that if a spike in one axon slightly preceded a spike in an adjacent axon, the earlier spike could cause the lagging spike to catch up and become synchronized with it (similar to configuration 1). It is interesting to note that the degree of ‘interaction’ between various pairs of these units correlates very well with the degree of association of the axons. Thus, the axons of fibres 1 and 2 stay coupled in the nerve as they branch out over muscle fibre 1 before innervating their respective fibres. The axons for fibres 3 and 4 also remain coupled as they branch out before innervating (Ikeda et al. 1980). These pairs also show a higher degree of ‘interaction’.

The authors do not wish to propose a detailed hypothesis concerning the neural circuitry which might produce the interactions suggested by these data, since various factors might be involved, and no direct evidence exists at this time. The above possibility is suggested primarily to demonstrate that the interactions described here would not require a complicated circuitry to produce them.

An analysis of the concurrent interval relationship between these units (see following paper, Koenig & Ikeda, 1980) suggests that the major factor determining interval length is a common input. The observations presented in this paper suggest that in addition, spacing interactions between certain pairs of neurones (probably motor neurones) occur intermittently to either shorten or lengthen the interval. The interval following one of these shortened or lengthened (interacting) intervals is observed to be once again of expected length. This suggests that while the input driving the neurone determines the basic firing rate, the actual firing time of the cell is determined by when it last fired, rather than by the driving interneurone. Thus, it appears that the next firing time of the neurone is determined by an endogenous process in the cell which is reset whenever it fires. This has also been suggested by Harcombe & Wyman (1977) since they observed that an antidromic spike in these motor neurones causes a resetting of the cell’s intrinsic rhythm in the same way that an interaction appears to do.

Phase relationships similar to those of Drosophila have been reported in the DLM of certain other species of fly (Wyman, 1965, 19696, 1970). These relationships are probably the result of interactions similar to those reported here for Drosophila. In Wyman’s 1965 paper a histogram of interspike intervals from an unidentified DLM motor unit of Calliphora is presented. The distribution of intervals in this histogram is skewed, showing a tail of shorter intervals to the left. The reason for this skewness was not known. However, in light of the present observations, it is probable that the short intervals making up the tail in this histogram were intervals that had been involved in a ‘spacing’ interaction with another unit. Thus, the observations presented here describe a mechanism common to the flight system of various species of fly.

We wish to thank Ms Lois Worth for her careful work in the preparation of the manuscript. This research was supported by the USPHS NIH grant NS-07442, USA.